1 /* Copyright (c) 2008, 2009, 2010, 2011, 2012 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"
41 #include "dynamic-string.h"
47 #include "mac-learning.h"
51 #include "ofp-print.h"
53 #include "ofproto/netflow.h"
54 #include "ofproto/ofproto.h"
55 #include "ovsdb-data.h"
57 #include "poll-loop.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_process_cfm);
79 COVERAGE_DEFINE(bridge_process_lacp);
80 COVERAGE_DEFINE(bridge_reconfigure);
81 COVERAGE_DEFINE(bridge_lacp_update);
89 struct dst builtin[32];
94 static void dst_set_init(struct dst_set *);
95 static void dst_set_add(struct dst_set *, const struct dst *);
96 static void dst_set_free(struct dst_set *);
99 LACP_CURRENT = 0x01, /* Current State. */
100 LACP_EXPIRED = 0x02, /* Expired State. */
101 LACP_DEFAULTED = 0x04, /* Partner is defaulted. */
102 LACP_ATTACHED = 0x08, /* Attached. Interface may be choosen for flows. */
106 /* These members are always valid. */
107 struct port *port; /* Containing port. */
108 size_t port_ifidx; /* Index within containing port. */
109 char *name; /* Host network device name. */
110 tag_type tag; /* Tag associated with this interface. */
111 long long delay_expires; /* Time after which 'enabled' may change. */
113 /* These members are valid only after bridge_reconfigure() causes them to
115 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
116 int dp_ifidx; /* Index within kernel datapath. */
117 struct netdev *netdev; /* Network device. */
118 bool enabled; /* May be chosen for flows? */
119 bool up; /* Is the interface up? */
120 const char *type; /* Usually same as cfg->type. */
121 struct cfm *cfm; /* Connectivity Fault Management */
122 const struct ovsrec_interface *cfg;
124 /* LACP information. */
125 enum lacp_status lacp_status; /* LACP status. */
126 uint16_t lacp_priority; /* LACP port priority. */
127 struct lacp_info lacp_actor; /* LACP actor information. */
128 struct lacp_info lacp_partner; /* LACP partner information. */
129 long long int lacp_tx; /* Next LACP message transmission time. */
130 long long int lacp_rx; /* Next LACP message receive time. */
133 #define BOND_MASK 0xff
135 int iface_idx; /* Index of assigned iface, or -1 if none. */
136 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
137 tag_type iface_tag; /* Tag associated with iface_idx. */
141 BM_TCP, /* Transport Layer Load Balance. */
142 BM_SLB, /* Source Load Balance. */
143 BM_AB /* Active Backup. */
146 #define MAX_MIRRORS 32
147 typedef uint32_t mirror_mask_t;
148 #define MIRROR_MASK_C(X) UINT32_C(X)
149 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
151 struct bridge *bridge;
154 struct uuid uuid; /* UUID of this "mirror" record in database. */
156 /* Selection criteria. */
157 struct shash src_ports; /* Name is port name; data is always NULL. */
158 struct shash dst_ports; /* Name is port name; data is always NULL. */
163 struct port *out_port;
167 /* Flags for a port's lacp member. */
168 #define LACP_ACTIVE 0x01 /* LACP is in active mode. */
169 #define LACP_PASSIVE 0x02 /* LACP is in passive mode. */
170 #define LACP_NEGOTIATED 0x04 /* LACP has successfully negotiated. */
172 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
174 struct bridge *bridge;
176 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
177 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
178 * NULL if all VLANs are trunked. */
179 const struct ovsrec_port *cfg;
182 /* An ordinary bridge port has 1 interface.
183 * A bridge port for bonding has at least 2 interfaces. */
184 struct iface **ifaces;
185 size_t n_ifaces, allocated_ifaces;
188 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
189 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
190 tag_type active_iface_tag; /* Tag for bcast flows. */
191 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
192 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
193 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
194 bool miimon; /* Use miimon instead of carrier? */
195 long long int bond_miimon_interval; /* Miimon status refresh interval. */
196 long long int bond_miimon_next_update; /* Time of next miimon update. */
197 long long int bond_next_fake_iface_update; /* Time of next update. */
198 struct netdev_monitor *monitor; /* Tracks carrier up/down status. */
200 /* LACP information. */
201 int lacp; /* LACP status flags. 0 if LACP is off. */
202 uint16_t lacp_key; /* LACP aggregation key. */
203 uint16_t lacp_priority; /* LACP system priority. */
204 bool lacp_need_update; /* Need to update attached interfaces? */
206 /* SLB specific bonding info. */
207 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
208 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
209 long long int bond_next_rebalance; /* Next rebalancing time. */
211 /* Port mirroring info. */
212 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
213 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
214 bool is_mirror_output_port; /* Does port mirroring send frames here? */
218 struct list node; /* Node in global list of bridges. */
219 char *name; /* User-specified arbitrary name. */
220 struct mac_learning *ml; /* MAC learning table. */
221 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
222 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
223 const struct ovsrec_bridge *cfg;
225 /* OpenFlow switch processing. */
226 struct ofproto *ofproto; /* OpenFlow switch. */
228 /* Kernel datapath information. */
229 struct dpif *dpif; /* Datapath. */
230 struct hmap ifaces; /* Contains "struct iface"s. */
234 size_t n_ports, allocated_ports;
235 struct shash iface_by_name; /* "struct iface"s indexed by name. */
236 struct shash port_by_name; /* "struct port"s indexed by name. */
239 bool has_bonded_ports;
244 /* Port mirroring. */
245 struct mirror *mirrors[MAX_MIRRORS];
248 /* List of all bridges. */
249 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
251 /* OVSDB IDL used to obtain configuration. */
252 static struct ovsdb_idl *idl;
254 /* Each time this timer expires, the bridge fetches systems and interface
255 * statistics and pushes them into the database. */
256 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
257 static long long int stats_timer = LLONG_MIN;
259 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
260 static void bridge_destroy(struct bridge *);
261 static struct bridge *bridge_lookup(const char *name);
262 static unixctl_cb_func bridge_unixctl_dump_flows;
263 static unixctl_cb_func bridge_unixctl_reconnect;
264 static int bridge_run_one(struct bridge *);
265 static size_t bridge_get_controllers(const struct bridge *br,
266 struct ovsrec_controller ***controllersp);
267 static void bridge_reconfigure_one(struct bridge *);
268 static void bridge_reconfigure_remotes(struct bridge *,
269 const struct sockaddr_in *managers,
271 static void bridge_reconfigure_remotes_late(struct bridge *);
272 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
273 static void bridge_fetch_dp_ifaces(struct bridge *);
274 static void bridge_flush(struct bridge *);
275 static void bridge_pick_local_hw_addr(struct bridge *,
276 uint8_t ea[ETH_ADDR_LEN],
277 struct iface **hw_addr_iface);
278 static uint64_t bridge_pick_datapath_id(struct bridge *,
279 const uint8_t bridge_ea[ETH_ADDR_LEN],
280 struct iface *hw_addr_iface);
281 static struct iface *bridge_get_local_iface(struct bridge *);
282 static uint64_t dpid_from_hash(const void *, size_t nbytes);
284 static unixctl_cb_func bridge_unixctl_fdb_show;
285 static unixctl_cb_func qos_unixctl_show;
287 static void lacp_run(struct port *);
288 static void lacp_wait(struct port *);
289 static void lacp_process_packet(const struct ofpbuf *, struct iface *);
291 static void bond_init(void);
292 static void bond_run(struct port *);
293 static void bond_wait(struct port *);
294 static void bond_rebalance_port(struct port *);
295 static void bond_send_learning_packets(struct port *);
296 static void bond_enable_slave(struct iface *iface, bool enable);
298 static void port_run(struct port *);
299 static void port_wait(struct port *);
300 static struct port *port_create(struct bridge *, const char *name);
301 static void port_reconfigure(struct port *, const struct ovsrec_port *);
302 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
303 static void port_destroy(struct port *);
304 static struct port *port_lookup(const struct bridge *, const char *name);
305 static struct iface *port_lookup_iface(const struct port *, const char *name);
306 static struct port *port_from_dp_ifidx(const struct bridge *,
308 static void port_update_bonding(struct port *);
309 static void port_update_lacp(struct port *);
311 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
312 static void mirror_destroy(struct mirror *);
313 static void mirror_reconfigure(struct bridge *);
314 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
315 static bool vlan_is_mirrored(const struct mirror *, int vlan);
317 static struct iface *iface_create(struct port *port,
318 const struct ovsrec_interface *if_cfg);
319 static void iface_destroy(struct iface *);
320 static struct iface *iface_lookup(const struct bridge *, const char *name);
321 static struct iface *iface_find(const char *name);
322 static struct iface *iface_from_dp_ifidx(const struct bridge *,
324 static void iface_set_mac(struct iface *);
325 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
326 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
327 static void iface_update_cfm(struct iface *);
328 static void iface_refresh_cfm_stats(struct iface *iface);
329 static void iface_send_packet(struct iface *, struct ofpbuf *packet);
330 static uint8_t iface_get_lacp_state(const struct iface *);
331 static void iface_get_lacp_priority(struct iface *, struct lacp_info *);
332 static void iface_set_lacp_defaulted(struct iface *);
333 static void iface_set_lacp_expired(struct iface *);
335 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
337 static void shash_to_ovs_idl_map(struct shash *,
338 char ***keys, char ***values, size_t *n);
341 /* Hooks into ofproto processing. */
342 static struct ofhooks bridge_ofhooks;
344 /* Public functions. */
346 /* Initializes the bridge module, configuring it to obtain its configuration
347 * from an OVSDB server accessed over 'remote', which should be a string in a
348 * form acceptable to ovsdb_idl_create(). */
350 bridge_init(const char *remote)
352 /* Create connection to database. */
353 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
355 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
356 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
357 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
359 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
361 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
362 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
364 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
365 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
366 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
368 /* Register unixctl commands. */
369 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
370 unixctl_command_register("qos/show", qos_unixctl_show, NULL);
371 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
373 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
381 struct bridge *br, *next_br;
383 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
386 ovsdb_idl_destroy(idl);
389 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
390 * but for which the ovs-vswitchd configuration 'cfg' is required. */
392 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
394 static bool already_configured_once;
395 struct svec bridge_names;
396 struct svec dpif_names, dpif_types;
399 /* Only do this once per ovs-vswitchd run. */
400 if (already_configured_once) {
403 already_configured_once = true;
405 stats_timer = time_msec() + STATS_INTERVAL;
407 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
408 svec_init(&bridge_names);
409 for (i = 0; i < cfg->n_bridges; i++) {
410 svec_add(&bridge_names, cfg->bridges[i]->name);
412 svec_sort(&bridge_names);
414 /* Iterate over all system dpifs and delete any of them that do not appear
416 svec_init(&dpif_names);
417 svec_init(&dpif_types);
418 dp_enumerate_types(&dpif_types);
419 for (i = 0; i < dpif_types.n; i++) {
422 dp_enumerate_names(dpif_types.names[i], &dpif_names);
424 /* Delete each dpif whose name is not in 'bridge_names'. */
425 for (j = 0; j < dpif_names.n; j++) {
426 if (!svec_contains(&bridge_names, dpif_names.names[j])) {
430 retval = dpif_open(dpif_names.names[j], dpif_types.names[i],
439 svec_destroy(&bridge_names);
440 svec_destroy(&dpif_names);
441 svec_destroy(&dpif_types);
444 /* Callback for iterate_and_prune_ifaces(). */
446 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
448 if (!iface->netdev) {
449 /* We already reported a related error, don't bother duplicating it. */
453 if (iface->dp_ifidx < 0) {
454 VLOG_ERR("%s interface not in %s, dropping",
455 iface->name, dpif_name(br->dpif));
459 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
460 iface->name, iface->dp_ifidx);
464 /* Callback for iterate_and_prune_ifaces(). */
466 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
467 void *aux OVS_UNUSED)
469 /* Set policing attributes. */
470 netdev_set_policing(iface->netdev,
471 iface->cfg->ingress_policing_rate,
472 iface->cfg->ingress_policing_burst);
474 /* Set MAC address of internal interfaces other than the local
476 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
477 iface_set_mac(iface);
483 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
484 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
485 * deletes from 'br' any ports that no longer have any interfaces. */
487 iterate_and_prune_ifaces(struct bridge *br,
488 bool (*cb)(struct bridge *, struct iface *,
494 for (i = 0; i < br->n_ports; ) {
495 struct port *port = br->ports[i];
496 for (j = 0; j < port->n_ifaces; ) {
497 struct iface *iface = port->ifaces[j];
498 if (cb(br, iface, aux)) {
501 iface_set_ofport(iface->cfg, -1);
502 iface_destroy(iface);
506 if (port->n_ifaces) {
509 VLOG_WARN("%s port has no interfaces, dropping", port->name);
515 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
516 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
517 * responsible for freeing '*managersp' (with free()).
519 * You may be asking yourself "why does ovs-vswitchd care?", because
520 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
521 * should not be and in fact is not directly involved in that. But
522 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
523 * it has to tell in-band control where the managers are to enable that.
524 * (Thus, only managers connected in-band are collected.)
527 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
528 struct sockaddr_in **managersp, size_t *n_managersp)
530 struct sockaddr_in *managers = NULL;
531 size_t n_managers = 0;
532 struct shash targets;
535 /* Collect all of the potential targets from the "targets" columns of the
536 * rows pointed to by "manager_options", excluding any that are
538 shash_init(&targets);
539 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
540 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
542 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
543 shash_find_and_delete(&targets, m->target);
545 shash_add_once(&targets, m->target, NULL);
549 /* Now extract the targets' IP addresses. */
550 if (!shash_is_empty(&targets)) {
551 struct shash_node *node;
553 managers = xmalloc(shash_count(&targets) * sizeof *managers);
554 SHASH_FOR_EACH (node, &targets) {
555 const char *target = node->name;
556 struct sockaddr_in *sin = &managers[n_managers];
558 if ((!strncmp(target, "tcp:", 4)
559 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
560 (!strncmp(target, "ssl:", 4)
561 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
566 shash_destroy(&targets);
568 *managersp = managers;
569 *n_managersp = n_managers;
573 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
575 struct shash old_br, new_br;
576 struct shash_node *node;
577 struct bridge *br, *next;
578 struct sockaddr_in *managers;
581 int sflow_bridge_number;
583 COVERAGE_INC(bridge_reconfigure);
585 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
587 /* Collect old and new bridges. */
590 LIST_FOR_EACH (br, node, &all_bridges) {
591 shash_add(&old_br, br->name, br);
593 for (i = 0; i < ovs_cfg->n_bridges; i++) {
594 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
595 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
596 VLOG_WARN("more than one bridge named %s", br_cfg->name);
600 /* Get rid of deleted bridges and add new bridges. */
601 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
602 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
609 SHASH_FOR_EACH (node, &new_br) {
610 const char *br_name = node->name;
611 const struct ovsrec_bridge *br_cfg = node->data;
612 br = shash_find_data(&old_br, br_name);
614 /* If the bridge datapath type has changed, we need to tear it
615 * down and recreate. */
616 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
618 bridge_create(br_cfg);
621 bridge_create(br_cfg);
624 shash_destroy(&old_br);
625 shash_destroy(&new_br);
627 /* Reconfigure all bridges. */
628 LIST_FOR_EACH (br, node, &all_bridges) {
629 bridge_reconfigure_one(br);
632 /* Add and delete ports on all datapaths.
634 * The kernel will reject any attempt to add a given port to a datapath if
635 * that port already belongs to a different datapath, so we must do all
636 * port deletions before any port additions. */
637 LIST_FOR_EACH (br, node, &all_bridges) {
638 struct dpif_port_dump dump;
639 struct shash want_ifaces;
640 struct dpif_port dpif_port;
642 bridge_get_all_ifaces(br, &want_ifaces);
643 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
644 if (!shash_find(&want_ifaces, dpif_port.name)
645 && strcmp(dpif_port.name, br->name)) {
646 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
648 VLOG_WARN("failed to remove %s interface from %s: %s",
649 dpif_port.name, dpif_name(br->dpif),
654 shash_destroy(&want_ifaces);
656 LIST_FOR_EACH (br, node, &all_bridges) {
657 struct shash cur_ifaces, want_ifaces;
658 struct dpif_port_dump dump;
659 struct dpif_port dpif_port;
661 /* Get the set of interfaces currently in this datapath. */
662 shash_init(&cur_ifaces);
663 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
664 struct dpif_port *port_info = xmalloc(sizeof *port_info);
665 dpif_port_clone(port_info, &dpif_port);
666 shash_add(&cur_ifaces, dpif_port.name, port_info);
669 /* Get the set of interfaces we want on this datapath. */
670 bridge_get_all_ifaces(br, &want_ifaces);
672 hmap_clear(&br->ifaces);
673 SHASH_FOR_EACH (node, &want_ifaces) {
674 const char *if_name = node->name;
675 struct iface *iface = node->data;
676 struct dpif_port *dpif_port;
680 type = iface ? iface->type : "internal";
681 dpif_port = shash_find_data(&cur_ifaces, if_name);
683 /* If we have a port or a netdev already, and it's not the type we
684 * want, then delete the port (if any) and close the netdev (if
686 if ((dpif_port && strcmp(dpif_port->type, type))
687 || (iface && iface->netdev
688 && strcmp(type, netdev_get_type(iface->netdev)))) {
690 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
697 netdev_close(iface->netdev);
698 iface->netdev = NULL;
702 /* If the port doesn't exist or we don't have the netdev open,
703 * we need to do more work. */
704 if (!dpif_port || (iface && !iface->netdev)) {
705 struct netdev_options options;
706 struct netdev *netdev;
709 /* First open the network device. */
710 options.name = if_name;
712 options.args = &args;
713 options.ethertype = NETDEV_ETH_TYPE_NONE;
717 shash_from_ovs_idl_map(iface->cfg->key_options,
718 iface->cfg->value_options,
719 iface->cfg->n_options, &args);
721 error = netdev_open(&options, &netdev);
722 shash_destroy(&args);
725 VLOG_WARN("could not open network device %s (%s)",
726 if_name, strerror(error));
730 /* Then add the port if we haven't already. */
732 error = dpif_port_add(br->dpif, netdev, NULL);
734 netdev_close(netdev);
735 if (error == EFBIG) {
736 VLOG_ERR("ran out of valid port numbers on %s",
737 dpif_name(br->dpif));
740 VLOG_WARN("failed to add %s interface to %s: %s",
741 if_name, dpif_name(br->dpif),
748 /* Update 'iface'. */
750 iface->netdev = netdev;
751 iface->enabled = netdev_get_carrier(iface->netdev);
752 iface->up = iface->enabled;
754 } else if (iface && iface->netdev) {
758 shash_from_ovs_idl_map(iface->cfg->key_options,
759 iface->cfg->value_options,
760 iface->cfg->n_options, &args);
761 netdev_set_config(iface->netdev, &args);
762 shash_destroy(&args);
765 shash_destroy(&want_ifaces);
767 SHASH_FOR_EACH (node, &cur_ifaces) {
768 struct dpif_port *port_info = node->data;
769 dpif_port_destroy(port_info);
772 shash_destroy(&cur_ifaces);
774 sflow_bridge_number = 0;
775 LIST_FOR_EACH (br, node, &all_bridges) {
778 struct iface *local_iface;
779 struct iface *hw_addr_iface;
782 bridge_fetch_dp_ifaces(br);
784 iterate_and_prune_ifaces(br, check_iface, NULL);
786 /* Pick local port hardware address, datapath ID. */
787 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
788 local_iface = bridge_get_local_iface(br);
790 int error = netdev_set_etheraddr(local_iface->netdev, ea);
792 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
793 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
794 "Ethernet address: %s",
795 br->name, strerror(error));
798 memcpy(br->ea, ea, ETH_ADDR_LEN);
800 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
801 ofproto_set_datapath_id(br->ofproto, dpid);
803 dpid_string = xasprintf("%016"PRIx64, dpid);
804 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
807 /* Set NetFlow configuration on this bridge. */
808 if (br->cfg->netflow) {
809 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
810 struct netflow_options opts;
812 memset(&opts, 0, sizeof opts);
814 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
815 if (nf_cfg->engine_type) {
816 opts.engine_type = *nf_cfg->engine_type;
818 if (nf_cfg->engine_id) {
819 opts.engine_id = *nf_cfg->engine_id;
822 opts.active_timeout = nf_cfg->active_timeout;
823 if (!opts.active_timeout) {
824 opts.active_timeout = -1;
825 } else if (opts.active_timeout < 0) {
826 VLOG_WARN("bridge %s: active timeout interval set to negative "
827 "value, using default instead (%d seconds)", br->name,
828 NF_ACTIVE_TIMEOUT_DEFAULT);
829 opts.active_timeout = -1;
832 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
833 if (opts.add_id_to_iface) {
834 if (opts.engine_id > 0x7f) {
835 VLOG_WARN("bridge %s: netflow port mangling may conflict "
836 "with another vswitch, choose an engine id less "
837 "than 128", br->name);
839 if (br->n_ports > 508) {
840 VLOG_WARN("bridge %s: netflow port mangling will conflict "
841 "with another port when more than 508 ports are "
846 opts.collectors.n = nf_cfg->n_targets;
847 opts.collectors.names = nf_cfg->targets;
848 if (ofproto_set_netflow(br->ofproto, &opts)) {
849 VLOG_ERR("bridge %s: problem setting netflow collectors",
853 ofproto_set_netflow(br->ofproto, NULL);
856 /* Set sFlow configuration on this bridge. */
857 if (br->cfg->sflow) {
858 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
859 struct ovsrec_controller **controllers;
860 struct ofproto_sflow_options oso;
861 size_t n_controllers;
863 memset(&oso, 0, sizeof oso);
865 oso.targets.n = sflow_cfg->n_targets;
866 oso.targets.names = sflow_cfg->targets;
868 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
869 if (sflow_cfg->sampling) {
870 oso.sampling_rate = *sflow_cfg->sampling;
873 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
874 if (sflow_cfg->polling) {
875 oso.polling_interval = *sflow_cfg->polling;
878 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
879 if (sflow_cfg->header) {
880 oso.header_len = *sflow_cfg->header;
883 oso.sub_id = sflow_bridge_number++;
884 oso.agent_device = sflow_cfg->agent;
886 oso.control_ip = NULL;
887 n_controllers = bridge_get_controllers(br, &controllers);
888 for (i = 0; i < n_controllers; i++) {
889 if (controllers[i]->local_ip) {
890 oso.control_ip = controllers[i]->local_ip;
894 ofproto_set_sflow(br->ofproto, &oso);
896 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
898 ofproto_set_sflow(br->ofproto, NULL);
901 /* Update the controller and related settings. It would be more
902 * straightforward to call this from bridge_reconfigure_one(), but we
903 * can't do it there for two reasons. First, and most importantly, at
904 * that point we don't know the dp_ifidx of any interfaces that have
905 * been added to the bridge (because we haven't actually added them to
906 * the datapath). Second, at that point we haven't set the datapath ID
907 * yet; when a controller is configured, resetting the datapath ID will
908 * immediately disconnect from the controller, so it's better to set
909 * the datapath ID before the controller. */
910 bridge_reconfigure_remotes(br, managers, n_managers);
912 LIST_FOR_EACH (br, node, &all_bridges) {
913 for (i = 0; i < br->n_ports; i++) {
914 struct port *port = br->ports[i];
917 port_update_bonding(port);
918 port_update_lacp(port);
920 for (j = 0; j < port->n_ifaces; j++) {
921 iface_update_qos(port->ifaces[j], port->cfg->qos);
925 LIST_FOR_EACH (br, node, &all_bridges) {
926 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
929 LIST_FOR_EACH (br, node, &all_bridges) {
931 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
932 iface_update_cfm(iface);
934 bridge_reconfigure_remotes_late(br);
939 /* ovs-vswitchd has completed initialization, so allow the process that
940 * forked us to exit successfully. */
941 daemonize_complete();
945 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
946 const struct ovsdb_idl_column *column,
949 const struct ovsdb_datum *datum;
950 union ovsdb_atom atom;
953 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
954 atom.string = (char *) key;
955 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
956 return idx == UINT_MAX ? NULL : datum->values[idx].string;
960 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
962 return get_ovsrec_key_value(&br_cfg->header_,
963 &ovsrec_bridge_col_other_config, key);
967 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
968 struct iface **hw_addr_iface)
974 *hw_addr_iface = NULL;
976 /* Did the user request a particular MAC? */
977 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
978 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
979 if (eth_addr_is_multicast(ea)) {
980 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
981 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
982 } else if (eth_addr_is_zero(ea)) {
983 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
989 /* Otherwise choose the minimum non-local MAC address among all of the
991 memset(ea, 0xff, ETH_ADDR_LEN);
992 for (i = 0; i < br->n_ports; i++) {
993 struct port *port = br->ports[i];
994 uint8_t iface_ea[ETH_ADDR_LEN];
997 /* Mirror output ports don't participate. */
998 if (port->is_mirror_output_port) {
1002 /* Choose the MAC address to represent the port. */
1003 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
1004 /* Find the interface with this Ethernet address (if any) so that
1005 * we can provide the correct devname to the caller. */
1007 for (j = 0; j < port->n_ifaces; j++) {
1008 struct iface *candidate = port->ifaces[j];
1009 uint8_t candidate_ea[ETH_ADDR_LEN];
1010 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
1011 && eth_addr_equals(iface_ea, candidate_ea)) {
1016 /* Choose the interface whose MAC address will represent the port.
1017 * The Linux kernel bonding code always chooses the MAC address of
1018 * the first slave added to a bond, and the Fedora networking
1019 * scripts always add slaves to a bond in alphabetical order, so
1020 * for compatibility we choose the interface with the name that is
1021 * first in alphabetical order. */
1022 iface = port->ifaces[0];
1023 for (j = 1; j < port->n_ifaces; j++) {
1024 struct iface *candidate = port->ifaces[j];
1025 if (strcmp(candidate->name, iface->name) < 0) {
1030 /* The local port doesn't count (since we're trying to choose its
1031 * MAC address anyway). */
1032 if (iface->dp_ifidx == ODPP_LOCAL) {
1037 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1039 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1040 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1041 iface->name, strerror(error));
1046 /* Compare against our current choice. */
1047 if (!eth_addr_is_multicast(iface_ea) &&
1048 !eth_addr_is_local(iface_ea) &&
1049 !eth_addr_is_reserved(iface_ea) &&
1050 !eth_addr_is_zero(iface_ea) &&
1051 eth_addr_compare_3way(iface_ea, ea) < 0)
1053 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1054 *hw_addr_iface = iface;
1057 if (eth_addr_is_multicast(ea)) {
1058 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1059 *hw_addr_iface = NULL;
1060 VLOG_WARN("bridge %s: using default bridge Ethernet "
1061 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1063 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1064 br->name, ETH_ADDR_ARGS(ea));
1068 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1069 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1070 * an interface on 'br', then that interface must be passed in as
1071 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1072 * 'hw_addr_iface' must be passed in as a null pointer. */
1074 bridge_pick_datapath_id(struct bridge *br,
1075 const uint8_t bridge_ea[ETH_ADDR_LEN],
1076 struct iface *hw_addr_iface)
1079 * The procedure for choosing a bridge MAC address will, in the most
1080 * ordinary case, also choose a unique MAC that we can use as a datapath
1081 * ID. In some special cases, though, multiple bridges will end up with
1082 * the same MAC address. This is OK for the bridges, but it will confuse
1083 * the OpenFlow controller, because each datapath needs a unique datapath
1086 * Datapath IDs must be unique. It is also very desirable that they be
1087 * stable from one run to the next, so that policy set on a datapath
1090 const char *datapath_id;
1093 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1094 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1098 if (hw_addr_iface) {
1100 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1102 * A bridge whose MAC address is taken from a VLAN network device
1103 * (that is, a network device created with vconfig(8) or similar
1104 * tool) will have the same MAC address as a bridge on the VLAN
1105 * device's physical network device.
1107 * Handle this case by hashing the physical network device MAC
1108 * along with the VLAN identifier.
1110 uint8_t buf[ETH_ADDR_LEN + 2];
1111 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1112 buf[ETH_ADDR_LEN] = vlan >> 8;
1113 buf[ETH_ADDR_LEN + 1] = vlan;
1114 return dpid_from_hash(buf, sizeof buf);
1117 * Assume that this bridge's MAC address is unique, since it
1118 * doesn't fit any of the cases we handle specially.
1123 * A purely internal bridge, that is, one that has no non-virtual
1124 * network devices on it at all, is more difficult because it has no
1125 * natural unique identifier at all.
1127 * When the host is a XenServer, we handle this case by hashing the
1128 * host's UUID with the name of the bridge. Names of bridges are
1129 * persistent across XenServer reboots, although they can be reused if
1130 * an internal network is destroyed and then a new one is later
1131 * created, so this is fairly effective.
1133 * When the host is not a XenServer, we punt by using a random MAC
1134 * address on each run.
1136 const char *host_uuid = xenserver_get_host_uuid();
1138 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1139 dpid = dpid_from_hash(combined, strlen(combined));
1145 return eth_addr_to_uint64(bridge_ea);
1149 dpid_from_hash(const void *data, size_t n)
1151 uint8_t hash[SHA1_DIGEST_SIZE];
1153 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1154 sha1_bytes(data, n, hash);
1155 eth_addr_mark_random(hash);
1156 return eth_addr_to_uint64(hash);
1160 iface_refresh_status(struct iface *iface)
1164 enum netdev_flags flags;
1173 if (!netdev_get_status(iface->netdev, &sh)) {
1175 char **keys, **values;
1177 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1178 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1183 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1186 shash_destroy_free_data(&sh);
1188 error = netdev_get_flags(iface->netdev, &flags);
1190 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1193 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1196 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1198 ovsrec_interface_set_duplex(iface->cfg,
1199 netdev_features_is_full_duplex(current)
1201 /* warning: uint64_t -> int64_t conversion */
1202 bps = netdev_features_to_bps(current);
1203 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1206 ovsrec_interface_set_duplex(iface->cfg, NULL);
1207 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1211 ovsrec_interface_set_link_state(iface->cfg,
1212 netdev_get_carrier(iface->netdev)
1215 error = netdev_get_mtu(iface->netdev, &mtu);
1216 if (!error && mtu != INT_MAX) {
1218 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1221 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1226 iface_refresh_cfm_stats(struct iface *iface)
1230 const struct ovsrec_monitor *mon;
1232 mon = iface->cfg->monitor;
1239 for (i = 0; i < mon->n_remote_mps; i++) {
1240 const struct ovsrec_maintenance_point *mp;
1241 const struct remote_mp *rmp;
1243 mp = mon->remote_mps[i];
1244 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1246 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1249 if (hmap_is_empty(&cfm->x_remote_mps)) {
1250 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1253 struct remote_mp *rmp;
1254 int64_t *x_remote_mps;
1256 length = hmap_count(&cfm->x_remote_mps);
1257 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1260 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1261 x_remote_mps[i++] = rmp->mpid;
1264 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1268 if (hmap_is_empty(&cfm->x_remote_maids)) {
1269 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1272 char **x_remote_maids;
1273 struct remote_maid *rmaid;
1275 length = hmap_count(&cfm->x_remote_maids);
1276 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1279 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1282 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1284 for (j = 0; j < CCM_MAID_LEN; j++) {
1285 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1290 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1292 for (i = 0; i < length; i++) {
1293 free(x_remote_maids[i]);
1295 free(x_remote_maids);
1298 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1302 iface_refresh_stats(struct iface *iface)
1308 static const struct iface_stat iface_stats[] = {
1309 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1310 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1311 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1312 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1313 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1314 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1315 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1316 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1317 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1318 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1319 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1320 { "collisions", offsetof(struct netdev_stats, collisions) },
1322 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1323 const struct iface_stat *s;
1325 char *keys[N_STATS];
1326 int64_t values[N_STATS];
1329 struct netdev_stats stats;
1331 /* Intentionally ignore return value, since errors will set 'stats' to
1332 * all-1s, and we will deal with that correctly below. */
1333 netdev_get_stats(iface->netdev, &stats);
1336 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1337 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1338 if (value != UINT64_MAX) {
1345 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1349 enable_system_stats(const struct ovsrec_open_vswitch *cfg)
1353 /* Use other-config:enable-system-stats by preference. */
1354 enable = get_ovsrec_key_value(&cfg->header_,
1355 &ovsrec_open_vswitch_col_other_config,
1356 "enable-statistics");
1358 return !strcmp(enable, "true");
1361 /* Disable by default. */
1366 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1368 struct ovsdb_datum datum;
1372 if (enable_system_stats(cfg)) {
1373 get_system_stats(&stats);
1376 ovsdb_datum_from_shash(&datum, &stats);
1377 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1381 static inline const char *
1382 nx_role_to_str(enum nx_role role)
1387 case NX_ROLE_MASTER:
1392 return "*** INVALID ROLE ***";
1397 bridge_refresh_controller_status(const struct bridge *br)
1400 const struct ovsrec_controller *cfg;
1402 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1404 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1405 struct ofproto_controller_info *cinfo =
1406 shash_find_data(&info, cfg->target);
1409 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1410 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1411 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1412 (char **) cinfo->pairs.values,
1415 ovsrec_controller_set_is_connected(cfg, false);
1416 ovsrec_controller_set_role(cfg, NULL);
1417 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1421 ofproto_free_ofproto_controller_info(&info);
1427 const struct ovsrec_open_vswitch *cfg;
1429 bool datapath_destroyed;
1430 bool database_changed;
1433 /* Let each bridge do the work that it needs to do. */
1434 datapath_destroyed = false;
1435 LIST_FOR_EACH (br, node, &all_bridges) {
1436 int error = bridge_run_one(br);
1438 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1439 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1440 "forcing reconfiguration", br->name);
1441 datapath_destroyed = true;
1445 /* (Re)configure if necessary. */
1446 database_changed = ovsdb_idl_run(idl);
1447 cfg = ovsrec_open_vswitch_first(idl);
1449 /* Re-configure SSL. We do this on every trip through the main loop,
1450 * instead of just when the database changes, because the contents of the
1451 * key and certificate files can change without the database changing.
1453 * We do this before bridge_reconfigure() because that function might
1454 * initiate SSL connections and thus requires SSL to be configured. */
1455 if (cfg && cfg->ssl) {
1456 const struct ovsrec_ssl *ssl = cfg->ssl;
1458 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1459 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1462 if (database_changed || datapath_destroyed) {
1464 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1466 bridge_configure_once(cfg);
1467 bridge_reconfigure(cfg);
1469 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1470 ovsdb_idl_txn_commit(txn);
1471 ovsdb_idl_txn_destroy(txn); /* XXX */
1473 /* We still need to reconfigure to avoid dangling pointers to
1474 * now-destroyed ovsrec structures inside bridge data. */
1475 static const struct ovsrec_open_vswitch null_cfg;
1477 bridge_reconfigure(&null_cfg);
1481 /* Refresh system and interface stats if necessary. */
1482 if (time_msec() >= stats_timer) {
1484 struct ovsdb_idl_txn *txn;
1486 txn = ovsdb_idl_txn_create(idl);
1487 LIST_FOR_EACH (br, node, &all_bridges) {
1490 for (i = 0; i < br->n_ports; i++) {
1491 struct port *port = br->ports[i];
1494 for (j = 0; j < port->n_ifaces; j++) {
1495 struct iface *iface = port->ifaces[j];
1496 iface_refresh_stats(iface);
1497 iface_refresh_cfm_stats(iface);
1498 iface_refresh_status(iface);
1501 bridge_refresh_controller_status(br);
1503 refresh_system_stats(cfg);
1504 ovsdb_idl_txn_commit(txn);
1505 ovsdb_idl_txn_destroy(txn); /* XXX */
1508 stats_timer = time_msec() + STATS_INTERVAL;
1517 LIST_FOR_EACH (br, node, &all_bridges) {
1520 ofproto_wait(br->ofproto);
1521 if (ofproto_has_primary_controller(br->ofproto)) {
1525 mac_learning_wait(br->ml);
1527 for (i = 0; i < br->n_ports; i++) {
1528 port_wait(br->ports[i]);
1531 ovsdb_idl_wait(idl);
1532 poll_timer_wait_until(stats_timer);
1535 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1536 * configuration changes. */
1538 bridge_flush(struct bridge *br)
1540 COVERAGE_INC(bridge_flush);
1542 mac_learning_flush(br->ml);
1545 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1546 * such interface. */
1547 static struct iface *
1548 bridge_get_local_iface(struct bridge *br)
1552 for (i = 0; i < br->n_ports; i++) {
1553 struct port *port = br->ports[i];
1554 for (j = 0; j < port->n_ifaces; j++) {
1555 struct iface *iface = port->ifaces[j];
1556 if (iface->dp_ifidx == ODPP_LOCAL) {
1565 /* Bridge unixctl user interface functions. */
1567 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1568 const char *args, void *aux OVS_UNUSED)
1570 struct ds ds = DS_EMPTY_INITIALIZER;
1571 const struct bridge *br;
1572 const struct mac_entry *e;
1574 br = bridge_lookup(args);
1576 unixctl_command_reply(conn, 501, "no such bridge");
1580 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1581 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1582 if (e->port < 0 || e->port >= br->n_ports) {
1585 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1586 br->ports[e->port]->ifaces[0]->dp_ifidx,
1587 e->vlan, ETH_ADDR_ARGS(e->mac),
1588 mac_entry_age(br->ml, e));
1590 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1594 /* QoS unixctl user interface functions. */
1596 struct qos_unixctl_show_cbdata {
1598 struct iface *iface;
1602 qos_unixctl_show_cb(unsigned int queue_id,
1603 const struct shash *details,
1606 struct qos_unixctl_show_cbdata *data = aux;
1607 struct ds *ds = data->ds;
1608 struct iface *iface = data->iface;
1609 struct netdev_queue_stats stats;
1610 struct shash_node *node;
1613 ds_put_cstr(ds, "\n");
1615 ds_put_format(ds, "Queue %u:\n", queue_id);
1617 ds_put_cstr(ds, "Default:\n");
1620 SHASH_FOR_EACH (node, details) {
1621 ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data);
1624 error = netdev_get_queue_stats(iface->netdev, queue_id, &stats);
1626 if (stats.tx_packets != UINT64_MAX) {
1627 ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets);
1630 if (stats.tx_bytes != UINT64_MAX) {
1631 ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes);
1634 if (stats.tx_errors != UINT64_MAX) {
1635 ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors);
1638 ds_put_format(ds, "\tFailed to get statistics for queue %u: %s",
1639 queue_id, strerror(error));
1644 qos_unixctl_show(struct unixctl_conn *conn,
1645 const char *args, void *aux OVS_UNUSED)
1647 struct ds ds = DS_EMPTY_INITIALIZER;
1648 struct shash sh = SHASH_INITIALIZER(&sh);
1649 struct iface *iface;
1651 struct shash_node *node;
1652 struct qos_unixctl_show_cbdata data;
1655 iface = iface_find(args);
1657 unixctl_command_reply(conn, 501, "no such interface");
1661 netdev_get_qos(iface->netdev, &type, &sh);
1663 if (*type != '\0') {
1664 ds_put_format(&ds, "QoS: %s %s\n", iface->name, type);
1666 SHASH_FOR_EACH (node, &sh) {
1667 ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data);
1672 error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data);
1675 ds_put_format(&ds, "failed to dump queues: %s", strerror(error));
1677 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1679 ds_put_format(&ds, "QoS not configured on %s\n", iface->name);
1680 unixctl_command_reply(conn, 501, ds_cstr(&ds));
1683 shash_destroy_free_data(&sh);
1687 /* Bridge reconfiguration functions. */
1688 static struct bridge *
1689 bridge_create(const struct ovsrec_bridge *br_cfg)
1694 assert(!bridge_lookup(br_cfg->name));
1695 br = xzalloc(sizeof *br);
1697 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1703 dpif_flow_flush(br->dpif);
1705 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1708 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1710 dpif_delete(br->dpif);
1711 dpif_close(br->dpif);
1716 br->name = xstrdup(br_cfg->name);
1718 br->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1719 eth_addr_nicira_random(br->default_ea);
1721 hmap_init(&br->ifaces);
1723 shash_init(&br->port_by_name);
1724 shash_init(&br->iface_by_name);
1728 list_push_back(&all_bridges, &br->node);
1730 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1736 bridge_destroy(struct bridge *br)
1741 while (br->n_ports > 0) {
1742 port_destroy(br->ports[br->n_ports - 1]);
1744 list_remove(&br->node);
1745 error = dpif_delete(br->dpif);
1746 if (error && error != ENOENT) {
1747 VLOG_ERR("failed to delete %s: %s",
1748 dpif_name(br->dpif), strerror(error));
1750 dpif_close(br->dpif);
1751 ofproto_destroy(br->ofproto);
1752 mac_learning_destroy(br->ml);
1753 hmap_destroy(&br->ifaces);
1754 shash_destroy(&br->port_by_name);
1755 shash_destroy(&br->iface_by_name);
1762 static struct bridge *
1763 bridge_lookup(const char *name)
1767 LIST_FOR_EACH (br, node, &all_bridges) {
1768 if (!strcmp(br->name, name)) {
1775 /* Handle requests for a listing of all flows known by the OpenFlow
1776 * stack, including those normally hidden. */
1778 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1779 const char *args, void *aux OVS_UNUSED)
1784 br = bridge_lookup(args);
1786 unixctl_command_reply(conn, 501, "Unknown bridge");
1791 ofproto_get_all_flows(br->ofproto, &results);
1793 unixctl_command_reply(conn, 200, ds_cstr(&results));
1794 ds_destroy(&results);
1797 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1798 * connections and reconnect. If BRIDGE is not specified, then all bridges
1799 * drop their controller connections and reconnect. */
1801 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1802 const char *args, void *aux OVS_UNUSED)
1805 if (args[0] != '\0') {
1806 br = bridge_lookup(args);
1808 unixctl_command_reply(conn, 501, "Unknown bridge");
1811 ofproto_reconnect_controllers(br->ofproto);
1813 LIST_FOR_EACH (br, node, &all_bridges) {
1814 ofproto_reconnect_controllers(br->ofproto);
1817 unixctl_command_reply(conn, 200, NULL);
1821 bridge_run_one(struct bridge *br)
1826 error = ofproto_run1(br->ofproto);
1831 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1833 for (i = 0; i < br->n_ports; i++) {
1834 port_run(br->ports[i]);
1837 error = ofproto_run2(br->ofproto, br->flush);
1844 bridge_get_controllers(const struct bridge *br,
1845 struct ovsrec_controller ***controllersp)
1847 struct ovsrec_controller **controllers;
1848 size_t n_controllers;
1850 controllers = br->cfg->controller;
1851 n_controllers = br->cfg->n_controller;
1853 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1859 *controllersp = controllers;
1861 return n_controllers;
1865 bridge_reconfigure_one(struct bridge *br)
1867 struct shash old_ports, new_ports;
1868 struct svec snoops, old_snoops;
1869 struct shash_node *node;
1870 enum ofproto_fail_mode fail_mode;
1871 const char *idle_time_str;
1875 /* Collect old ports. */
1876 shash_init(&old_ports);
1877 for (i = 0; i < br->n_ports; i++) {
1878 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1881 /* Collect new ports. */
1882 shash_init(&new_ports);
1883 for (i = 0; i < br->cfg->n_ports; i++) {
1884 const char *name = br->cfg->ports[i]->name;
1885 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1886 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1891 /* If we have a controller, then we need a local port. Complain if the
1892 * user didn't specify one.
1894 * XXX perhaps we should synthesize a port ourselves in this case. */
1895 if (bridge_get_controllers(br, NULL)) {
1896 char local_name[IF_NAMESIZE];
1899 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1900 local_name, sizeof local_name);
1901 if (!error && !shash_find(&new_ports, local_name)) {
1902 VLOG_WARN("bridge %s: controller specified but no local port "
1903 "(port named %s) defined",
1904 br->name, local_name);
1908 /* Get rid of deleted ports.
1909 * Get rid of deleted interfaces on ports that still exist. */
1910 SHASH_FOR_EACH (node, &old_ports) {
1911 struct port *port = node->data;
1912 const struct ovsrec_port *port_cfg;
1914 port_cfg = shash_find_data(&new_ports, node->name);
1918 port_del_ifaces(port, port_cfg);
1922 /* Create new ports.
1923 * Add new interfaces to existing ports.
1924 * Reconfigure existing ports. */
1925 SHASH_FOR_EACH (node, &new_ports) {
1926 struct port *port = shash_find_data(&old_ports, node->name);
1928 port = port_create(br, node->name);
1931 port_reconfigure(port, node->data);
1932 if (!port->n_ifaces) {
1933 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1934 br->name, port->name);
1938 shash_destroy(&old_ports);
1939 shash_destroy(&new_ports);
1941 /* Set the fail-mode */
1942 fail_mode = !br->cfg->fail_mode
1943 || !strcmp(br->cfg->fail_mode, "standalone")
1944 ? OFPROTO_FAIL_STANDALONE
1945 : OFPROTO_FAIL_SECURE;
1946 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1947 && !ofproto_has_primary_controller(br->ofproto)) {
1948 ofproto_flush_flows(br->ofproto);
1950 ofproto_set_fail_mode(br->ofproto, fail_mode);
1952 /* Set the MAC learning aging timeout. */
1953 idle_time_str = bridge_get_other_config(br->cfg, "mac-aging-time");
1954 idle_time = (idle_time_str && atoi(idle_time_str)
1955 ? atoi(idle_time_str)
1956 : MAC_ENTRY_DEFAULT_IDLE_TIME);
1957 mac_learning_set_idle_time(br->ml, idle_time);
1959 /* Delete all flows if we're switching from connected to standalone or vice
1960 * versa. (XXX Should we delete all flows if we are switching from one
1961 * controller to another?) */
1963 /* Configure OpenFlow controller connection snooping. */
1965 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1966 ovs_rundir(), br->name));
1967 svec_init(&old_snoops);
1968 ofproto_get_snoops(br->ofproto, &old_snoops);
1969 if (!svec_equal(&snoops, &old_snoops)) {
1970 ofproto_set_snoops(br->ofproto, &snoops);
1972 svec_destroy(&snoops);
1973 svec_destroy(&old_snoops);
1975 mirror_reconfigure(br);
1978 /* Initializes 'oc' appropriately as a management service controller for
1981 * The caller must free oc->target when it is no longer needed. */
1983 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1984 struct ofproto_controller *oc)
1986 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1987 oc->max_backoff = 0;
1988 oc->probe_interval = 60;
1989 oc->band = OFPROTO_OUT_OF_BAND;
1991 oc->burst_limit = 0;
1994 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1996 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1997 struct ofproto_controller *oc)
1999 oc->target = c->target;
2000 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
2001 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
2002 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
2003 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
2004 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
2005 oc->burst_limit = (c->controller_burst_limit
2006 ? *c->controller_burst_limit : 0);
2009 /* Configures the IP stack for 'br''s local interface properly according to the
2010 * configuration in 'c'. */
2012 bridge_configure_local_iface_netdev(struct bridge *br,
2013 struct ovsrec_controller *c)
2015 struct netdev *netdev;
2016 struct in_addr mask, gateway;
2018 struct iface *local_iface;
2021 /* If there's no local interface or no IP address, give up. */
2022 local_iface = bridge_get_local_iface(br);
2023 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
2027 /* Bring up the local interface. */
2028 netdev = local_iface->netdev;
2029 netdev_turn_flags_on(netdev, NETDEV_UP, true);
2031 /* Configure the IP address and netmask. */
2032 if (!c->local_netmask
2033 || !inet_aton(c->local_netmask, &mask)
2035 mask.s_addr = guess_netmask(ip.s_addr);
2037 if (!netdev_set_in4(netdev, ip, mask)) {
2038 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
2039 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
2042 /* Configure the default gateway. */
2043 if (c->local_gateway
2044 && inet_aton(c->local_gateway, &gateway)
2045 && gateway.s_addr) {
2046 if (!netdev_add_router(netdev, gateway)) {
2047 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
2048 br->name, IP_ARGS(&gateway.s_addr));
2054 bridge_reconfigure_remotes(struct bridge *br,
2055 const struct sockaddr_in *managers,
2058 const char *disable_ib_str, *queue_id_str;
2059 bool disable_in_band = false;
2062 struct ovsrec_controller **controllers;
2063 size_t n_controllers;
2066 struct ofproto_controller *ocs;
2070 /* Check if we should disable in-band control on this bridge. */
2071 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
2072 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
2073 disable_in_band = true;
2076 /* Set OpenFlow queue ID for in-band control. */
2077 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
2078 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
2079 ofproto_set_in_band_queue(br->ofproto, queue_id);
2081 if (disable_in_band) {
2082 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
2084 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
2086 had_primary = ofproto_has_primary_controller(br->ofproto);
2088 n_controllers = bridge_get_controllers(br, &controllers);
2090 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
2093 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
2094 for (i = 0; i < n_controllers; i++) {
2095 struct ovsrec_controller *c = controllers[i];
2097 if (!strncmp(c->target, "punix:", 6)
2098 || !strncmp(c->target, "unix:", 5)) {
2099 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2101 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
2102 * domain sockets and overwriting arbitrary local files. */
2103 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2104 "\"%s\" due to possibility for remote exploit",
2105 dpif_name(br->dpif), c->target);
2109 bridge_configure_local_iface_netdev(br, c);
2110 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2111 if (disable_in_band) {
2112 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2117 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2118 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2121 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2122 ofproto_flush_flows(br->ofproto);
2126 /* Does configuration of remotes that must happen after all of the ports and
2127 * interfaces are fully configured, that is, when flow translation can be
2128 * expected to succeed. (This is because ofproto_add_flow() immediately
2129 * re-translates any existing facets for the rule that it replaces, if any.)
2130 * In particular, it must be called after port_update_bonding(), to ensure that
2131 * 'bond_hash' is non-NULL for bonded ports. */
2133 bridge_reconfigure_remotes_late(struct bridge *br)
2135 /* If there are no controllers and the bridge is in standalone
2136 * mode, set up a flow that matches every packet and directs
2137 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2138 * switch is in secure mode and we won't pass any traffic until
2139 * a controller has been defined and it tells us to do so. */
2140 if (!bridge_get_controllers(br, NULL)
2141 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2142 union ofp_action action;
2143 struct cls_rule rule;
2145 memset(&action, 0, sizeof action);
2146 action.type = htons(OFPAT_OUTPUT);
2147 action.output.len = htons(sizeof action);
2148 action.output.port = htons(OFPP_NORMAL);
2149 cls_rule_init_catchall(&rule, 0);
2150 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2155 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2160 for (i = 0; i < br->n_ports; i++) {
2161 struct port *port = br->ports[i];
2162 for (j = 0; j < port->n_ifaces; j++) {
2163 struct iface *iface = port->ifaces[j];
2164 shash_add_once(ifaces, iface->name, iface);
2166 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2167 shash_add_once(ifaces, port->name, NULL);
2172 /* For robustness, in case the administrator moves around datapath ports behind
2173 * our back, we re-check all the datapath port numbers here.
2175 * This function will set the 'dp_ifidx' members of interfaces that have
2176 * disappeared to -1, so only call this function from a context where those
2177 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2178 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2179 * datapath, which doesn't support UINT16_MAX+1 ports. */
2181 bridge_fetch_dp_ifaces(struct bridge *br)
2183 struct dpif_port_dump dump;
2184 struct dpif_port dpif_port;
2187 /* Reset all interface numbers. */
2188 for (i = 0; i < br->n_ports; i++) {
2189 struct port *port = br->ports[i];
2190 for (j = 0; j < port->n_ifaces; j++) {
2191 struct iface *iface = port->ifaces[j];
2192 iface->dp_ifidx = -1;
2195 hmap_clear(&br->ifaces);
2197 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2198 struct iface *iface = iface_lookup(br, dpif_port.name);
2200 if (iface->dp_ifidx >= 0) {
2201 VLOG_WARN("%s reported interface %s twice",
2202 dpif_name(br->dpif), dpif_port.name);
2203 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2204 VLOG_WARN("%s reported interface %"PRIu16" twice",
2205 dpif_name(br->dpif), dpif_port.port_no);
2207 iface->dp_ifidx = dpif_port.port_no;
2208 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2209 hash_int(iface->dp_ifidx, 0));
2212 iface_set_ofport(iface->cfg,
2213 (iface->dp_ifidx >= 0
2214 ? odp_port_to_ofp_port(iface->dp_ifidx)
2220 /* Bridge packet processing functions. */
2223 bond_is_tcp_hash(const struct port *port)
2225 return port->bond_mode == BM_TCP && port->lacp & LACP_NEGOTIATED;
2229 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2231 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2234 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2236 struct flow hash_flow;
2238 memcpy(&hash_flow, flow, sizeof hash_flow);
2239 hash_flow.vlan_tci = 0;
2241 /* The symmetric quality of this hash function is not required, but
2242 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2243 * purposes, so we use it out of convenience. */
2244 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2247 static struct bond_entry *
2248 lookup_bond_entry(const struct port *port, const struct flow *flow,
2251 assert(port->bond_mode != BM_AB);
2253 if (bond_is_tcp_hash(port)) {
2254 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2256 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2261 bond_choose_iface(const struct port *port)
2263 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2264 size_t i, best_down_slave = -1;
2265 long long next_delay_expiration = LLONG_MAX;
2267 for (i = 0; i < port->n_ifaces; i++) {
2268 struct iface *iface = port->ifaces[i];
2270 if (iface->enabled) {
2272 } else if (iface->delay_expires < next_delay_expiration
2273 && (iface->lacp_status & LACP_ATTACHED
2274 || !(port->lacp & LACP_NEGOTIATED))) {
2275 best_down_slave = i;
2276 next_delay_expiration = iface->delay_expires;
2280 if (best_down_slave != -1) {
2281 struct iface *iface = port->ifaces[best_down_slave];
2283 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2284 "since no other interface is up", iface->name,
2285 iface->delay_expires - time_msec());
2286 bond_enable_slave(iface, true);
2289 return best_down_slave;
2293 choose_output_iface(const struct port *port, const struct flow *flow,
2294 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2296 struct iface *iface;
2298 assert(port->n_ifaces);
2299 if (port->n_ifaces == 1) {
2300 iface = port->ifaces[0];
2301 } else if (port->bond_mode == BM_AB) {
2302 if (port->active_iface < 0) {
2303 *tags |= port->no_ifaces_tag;
2306 iface = port->ifaces[port->active_iface];
2308 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2309 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2310 || !port->ifaces[e->iface_idx]->enabled) {
2311 /* XXX select interface properly. The current interface selection
2312 * is only good for testing the rebalancing code. */
2313 e->iface_idx = bond_choose_iface(port);
2314 if (e->iface_idx < 0) {
2315 *tags |= port->no_ifaces_tag;
2318 e->iface_tag = tag_create_random();
2320 *tags |= e->iface_tag;
2321 iface = port->ifaces[e->iface_idx];
2323 *dp_ifidx = iface->dp_ifidx;
2324 *tags |= iface->tag; /* Currently only used for bonding. */
2329 bond_link_status_update(struct iface *iface)
2331 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2332 struct port *port = iface->port;
2333 bool up = iface->up;
2334 int updelay, downdelay;
2336 updelay = port->updelay;
2337 downdelay = port->downdelay;
2339 if (iface->port->lacp & LACP_NEGOTIATED) {
2344 if (iface->port->lacp && up) {
2345 /* The interface is up if it's attached to an aggregator and its
2346 * partner is synchronized. The only exception is defaulted links.
2347 * They are not required to have synchronized partners because they
2348 * have no partners at all. However, they will only be attached if
2349 * negotiations failed on all interfaces in the bond. */
2350 up = iface->lacp_status & LACP_ATTACHED
2351 && (iface->lacp_partner.state & LACP_STATE_SYNC
2352 || iface->lacp_status & LACP_DEFAULTED);
2356 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2357 /* Nothing to do. */
2360 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2361 iface->name, up ? "up" : "down");
2362 if (up == iface->enabled) {
2363 iface->delay_expires = LLONG_MAX;
2364 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2365 iface->name, up ? "disabled" : "enabled");
2366 } else if (up && port->active_iface < 0) {
2367 bond_enable_slave(iface, true);
2369 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2370 "other interface is up", iface->name, updelay);
2373 int delay = up ? updelay : downdelay;
2374 iface->delay_expires = time_msec() + delay;
2377 "interface %s: will be %s if it stays %s for %d ms",
2379 up ? "enabled" : "disabled",
2387 bond_choose_active_iface(struct port *port)
2389 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2391 port->active_iface = bond_choose_iface(port);
2392 port->active_iface_tag = tag_create_random();
2393 if (port->active_iface >= 0) {
2394 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2395 port->name, port->ifaces[port->active_iface]->name);
2397 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2403 bond_enable_slave(struct iface *iface, bool enable)
2405 struct port *port = iface->port;
2406 struct bridge *br = port->bridge;
2408 /* This acts as a recursion check. If the act of disabling a slave
2409 * causes a different slave to be enabled, the flag will allow us to
2410 * skip redundant work when we reenter this function. It must be
2411 * cleared on exit to keep things safe with multiple bonds. */
2412 static bool moving_active_iface = false;
2414 iface->delay_expires = LLONG_MAX;
2415 if (enable == iface->enabled) {
2419 iface->enabled = enable;
2420 if (!iface->enabled) {
2421 VLOG_WARN("interface %s: disabled", iface->name);
2422 ofproto_revalidate(br->ofproto, iface->tag);
2423 if (iface->port_ifidx == port->active_iface) {
2424 ofproto_revalidate(br->ofproto,
2425 port->active_iface_tag);
2427 /* Disabling a slave can lead to another slave being immediately
2428 * enabled if there will be no active slaves but one is waiting
2429 * on an updelay. In this case we do not need to run most of the
2430 * code for the newly enabled slave since there was no period
2431 * without an active slave and it is redundant with the disabling
2433 moving_active_iface = true;
2434 bond_choose_active_iface(port);
2436 bond_send_learning_packets(port);
2438 VLOG_WARN("interface %s: enabled", iface->name);
2439 if (port->active_iface < 0 && !moving_active_iface) {
2440 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2441 bond_choose_active_iface(port);
2442 bond_send_learning_packets(port);
2444 iface->tag = tag_create_random();
2447 moving_active_iface = false;
2450 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2451 * bond interface. */
2453 bond_update_fake_iface_stats(struct port *port)
2455 struct netdev_stats bond_stats;
2456 struct netdev *bond_dev;
2459 memset(&bond_stats, 0, sizeof bond_stats);
2461 for (i = 0; i < port->n_ifaces; i++) {
2462 struct netdev_stats slave_stats;
2464 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2465 /* XXX: We swap the stats here because they are swapped back when
2466 * reported by the internal device. The reason for this is
2467 * internal devices normally represent packets going into the system
2468 * but when used as fake bond device they represent packets leaving
2469 * the system. We really should do this in the internal device
2470 * itself because changing it here reverses the counts from the
2471 * perspective of the switch. However, the internal device doesn't
2472 * know what type of device it represents so we have to do it here
2474 bond_stats.tx_packets += slave_stats.rx_packets;
2475 bond_stats.tx_bytes += slave_stats.rx_bytes;
2476 bond_stats.rx_packets += slave_stats.tx_packets;
2477 bond_stats.rx_bytes += slave_stats.tx_bytes;
2481 if (!netdev_open_default(port->name, &bond_dev)) {
2482 netdev_set_stats(bond_dev, &bond_stats);
2483 netdev_close(bond_dev);
2488 bond_link_carrier_update(struct iface *iface, bool carrier)
2490 if (carrier == iface->up) {
2494 if (iface->lacp_status & LACP_CURRENT) {
2495 iface_set_lacp_expired(iface);
2498 iface->up = carrier;
2503 bond_run(struct port *port)
2508 if (port->n_ifaces < 2) {
2512 if (port->monitor) {
2513 assert(!port->miimon);
2515 /* Track carrier going up and down on interfaces. */
2516 while (!netdev_monitor_poll(port->monitor, &devname)) {
2517 struct iface *iface;
2519 iface = port_lookup_iface(port, devname);
2521 bool up = netdev_get_carrier(iface->netdev);
2522 bond_link_carrier_update(iface, up);
2527 assert(port->miimon);
2529 if (time_msec() >= port->bond_miimon_next_update) {
2530 for (i = 0; i < port->n_ifaces; i++) {
2531 struct iface *iface = port->ifaces[i];
2532 bool up = netdev_get_miimon(iface->netdev);
2533 bond_link_carrier_update(iface, up);
2535 port->bond_miimon_next_update = time_msec() +
2536 port->bond_miimon_interval;
2540 for (i = 0; i < port->n_ifaces; i++) {
2541 bond_link_status_update(port->ifaces[i]);
2544 for (i = 0; i < port->n_ifaces; i++) {
2545 struct iface *iface = port->ifaces[i];
2546 if (time_msec() >= iface->delay_expires) {
2547 bond_enable_slave(iface, !iface->enabled);
2551 if (port->bond_fake_iface
2552 && time_msec() >= port->bond_next_fake_iface_update) {
2553 bond_update_fake_iface_stats(port);
2554 port->bond_next_fake_iface_update = time_msec() + 1000;
2559 bond_wait(struct port *port)
2563 if (port->n_ifaces < 2) {
2567 if (port->monitor) {
2568 netdev_monitor_poll_wait(port->monitor);
2572 poll_timer_wait_until(port->bond_miimon_next_update);
2575 for (i = 0; i < port->n_ifaces; i++) {
2576 struct iface *iface = port->ifaces[i];
2577 if (iface->delay_expires != LLONG_MAX) {
2578 poll_timer_wait_until(iface->delay_expires);
2582 if (port->bond_fake_iface) {
2583 poll_timer_wait_until(port->bond_next_fake_iface_update);
2588 set_dst(struct dst *dst, const struct flow *flow,
2589 const struct port *in_port, const struct port *out_port,
2592 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2593 : in_port->vlan >= 0 ? in_port->vlan
2594 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2595 : vlan_tci_to_vid(flow->vlan_tci));
2596 return choose_output_iface(out_port, flow, dst->vlan,
2597 &dst->dp_ifidx, tags);
2601 swap_dst(struct dst *p, struct dst *q)
2603 struct dst tmp = *p;
2608 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2609 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2610 * that we push to the datapath. We could in fact fully sort the array by
2611 * vlan, but in most cases there are at most two different vlan tags so that's
2612 * possibly overkill.) */
2614 partition_dsts(struct dst_set *set, int vlan)
2616 struct dst *first = set->dsts;
2617 struct dst *last = set->dsts + set->n;
2619 while (first != last) {
2621 * - All dsts < first have vlan == 'vlan'.
2622 * - All dsts >= last have vlan != 'vlan'.
2623 * - first < last. */
2624 while (first->vlan == vlan) {
2625 if (++first == last) {
2630 /* Same invariants, plus one additional:
2631 * - first->vlan != vlan.
2633 while (last[-1].vlan != vlan) {
2634 if (--last == first) {
2639 /* Same invariants, plus one additional:
2640 * - last[-1].vlan == vlan.*/
2641 swap_dst(first++, --last);
2646 mirror_mask_ffs(mirror_mask_t mask)
2648 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2653 dst_set_init(struct dst_set *set)
2655 set->dsts = set->builtin;
2657 set->allocated = ARRAY_SIZE(set->builtin);
2661 dst_set_add(struct dst_set *set, const struct dst *dst)
2663 if (set->n >= set->allocated) {
2664 size_t new_allocated;
2665 struct dst *new_dsts;
2667 new_allocated = set->allocated * 2;
2668 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2669 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2673 set->dsts = new_dsts;
2674 set->allocated = new_allocated;
2676 set->dsts[set->n++] = *dst;
2680 dst_set_free(struct dst_set *set)
2682 if (set->dsts != set->builtin) {
2688 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2691 for (i = 0; i < set->n; i++) {
2692 if (set->dsts[i].vlan == test->vlan
2693 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2701 port_trunks_vlan(const struct port *port, uint16_t vlan)
2703 return (port->vlan < 0
2704 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2708 port_includes_vlan(const struct port *port, uint16_t vlan)
2710 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2714 port_is_floodable(const struct port *port)
2718 for (i = 0; i < port->n_ifaces; i++) {
2719 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2720 port->ifaces[i]->dp_ifidx)) {
2727 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
2728 * to a VLAN. In general most packets may be mirrored but we want to drop
2729 * protocols that may confuse switches. */
2731 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
2733 /* If you change this function's behavior, please update corresponding
2734 * documentation in vswitch.xml at the same time. */
2735 if (dst[0] != 0x01) {
2736 /* All the currently banned MACs happen to start with 01 currently, so
2737 * this is a quick way to eliminate most of the good ones. */
2739 if (eth_addr_is_reserved(dst)) {
2740 /* Drop STP, IEEE pause frames, and other reserved protocols
2741 * (01-80-c2-00-00-0x). */
2745 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
2747 if ((dst[3] & 0xfe) == 0xcc &&
2748 (dst[4] & 0xfe) == 0xcc &&
2749 (dst[5] & 0xfe) == 0xcc) {
2750 /* Drop the following protocols plus others following the same
2753 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
2754 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
2755 STP Uplink Fast (01-00-0c-cd-cd-cd) */
2759 if (!(dst[3] | dst[4] | dst[5])) {
2760 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
2769 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2770 const struct port *in_port, const struct port *out_port,
2771 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2773 mirror_mask_t mirrors = in_port->src_mirrors;
2778 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2779 if (flow_vlan == 0) {
2780 flow_vlan = OFP_VLAN_NONE;
2783 if (out_port == FLOOD_PORT) {
2784 for (i = 0; i < br->n_ports; i++) {
2785 struct port *port = br->ports[i];
2787 && port_is_floodable(port)
2788 && port_includes_vlan(port, vlan)
2789 && !port->is_mirror_output_port
2790 && set_dst(&dst, flow, in_port, port, tags)) {
2791 mirrors |= port->dst_mirrors;
2792 dst_set_add(set, &dst);
2795 *nf_output_iface = NF_OUT_FLOOD;
2796 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2797 dst_set_add(set, &dst);
2798 *nf_output_iface = dst.dp_ifidx;
2799 mirrors |= out_port->dst_mirrors;
2803 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2804 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2806 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2807 && !dst_is_duplicate(set, &dst)) {
2808 dst_set_add(set, &dst);
2810 } else if (eth_dst_may_rspan(flow->dl_dst)) {
2811 for (i = 0; i < br->n_ports; i++) {
2812 struct port *port = br->ports[i];
2813 if (port_includes_vlan(port, m->out_vlan)
2814 && set_dst(&dst, flow, in_port, port, tags))
2816 if (port->vlan < 0) {
2817 dst.vlan = m->out_vlan;
2819 if (dst_is_duplicate(set, &dst)) {
2823 /* Use the vlan tag on the original flow instead of
2824 * the one passed in the vlan parameter. This ensures
2825 * that we compare the vlan from before any implicit
2826 * tagging tags place. This is necessary because
2827 * dst->vlan is the final vlan, after removing implicit
2829 if (port == in_port && dst.vlan == flow_vlan) {
2830 /* Don't send out input port on same VLAN. */
2833 dst_set_add(set, &dst);
2838 mirrors &= mirrors - 1;
2841 partition_dsts(set, flow_vlan);
2844 static void OVS_UNUSED
2845 print_dsts(const struct dst_set *set)
2849 for (i = 0; i < set->n; i++) {
2850 const struct dst *dst = &set->dsts[i];
2852 printf(">p%"PRIu16, dst->dp_ifidx);
2853 if (dst->vlan != OFP_VLAN_NONE) {
2854 printf("v%"PRIu16, dst->vlan);
2860 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2861 const struct port *in_port, const struct port *out_port,
2862 tag_type *tags, struct ofpbuf *actions,
2863 uint16_t *nf_output_iface)
2870 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2873 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2874 if (cur_vlan == 0) {
2875 cur_vlan = OFP_VLAN_NONE;
2877 for (i = 0; i < set.n; i++) {
2878 const struct dst *dst = &set.dsts[i];
2879 if (dst->vlan != cur_vlan) {
2880 if (dst->vlan == OFP_VLAN_NONE) {
2881 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2884 tci = htons(dst->vlan & VLAN_VID_MASK);
2885 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2886 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2888 cur_vlan = dst->vlan;
2890 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2895 /* Returns the effective vlan of a packet, taking into account both the
2896 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2897 * the packet is untagged and -1 indicates it has an invalid header and
2898 * should be dropped. */
2899 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2900 struct port *in_port, bool have_packet)
2902 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2903 if (in_port->vlan >= 0) {
2905 /* XXX support double tagging? */
2907 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2908 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2909 "packet received on port %s configured with "
2910 "implicit VLAN %"PRIu16,
2911 br->name, vlan, in_port->name, in_port->vlan);
2915 vlan = in_port->vlan;
2917 if (!port_includes_vlan(in_port, vlan)) {
2919 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2920 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2921 "packet received on port %s not configured for "
2923 br->name, vlan, in_port->name, vlan);
2932 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2933 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2934 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2936 is_gratuitous_arp(const struct flow *flow)
2938 return (flow->dl_type == htons(ETH_TYPE_ARP)
2939 && eth_addr_is_broadcast(flow->dl_dst)
2940 && (flow->nw_proto == ARP_OP_REPLY
2941 || (flow->nw_proto == ARP_OP_REQUEST
2942 && flow->nw_src == flow->nw_dst)));
2946 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2947 struct port *in_port)
2949 enum grat_arp_lock_type lock_type;
2952 /* We don't want to learn from gratuitous ARP packets that are reflected
2953 * back over bond slaves so we lock the learning table. */
2954 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2955 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2956 GRAT_ARP_LOCK_CHECK;
2958 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2961 /* The log messages here could actually be useful in debugging,
2962 * so keep the rate limit relatively high. */
2963 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2965 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2966 "on port %s in VLAN %d",
2967 br->name, ETH_ADDR_ARGS(flow->dl_src),
2968 in_port->name, vlan);
2969 ofproto_revalidate(br->ofproto, rev_tag);
2973 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2974 * dropped. Returns true if they may be forwarded, false if they should be
2977 * If 'have_packet' is true, it indicates that the caller is processing a
2978 * received packet. If 'have_packet' is false, then the caller is just
2979 * revalidating an existing flow because configuration has changed. Either
2980 * way, 'have_packet' only affects logging (there is no point in logging errors
2981 * during revalidation).
2983 * Sets '*in_portp' to the input port. This will be a null pointer if
2984 * flow->in_port does not designate a known input port (in which case
2985 * is_admissible() returns false).
2987 * When returning true, sets '*vlanp' to the effective VLAN of the input
2988 * packet, as returned by flow_get_vlan().
2990 * May also add tags to '*tags', although the current implementation only does
2991 * so in one special case.
2994 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2995 tag_type *tags, int *vlanp, struct port **in_portp)
2997 struct iface *in_iface;
2998 struct port *in_port;
3001 /* Find the interface and port structure for the received packet. */
3002 in_iface = iface_from_dp_ifidx(br, flow->in_port);
3004 /* No interface? Something fishy... */
3006 /* Odd. A few possible reasons here:
3008 * - We deleted an interface but there are still a few packets
3009 * queued up from it.
3011 * - Someone externally added an interface (e.g. with "ovs-dpctl
3012 * add-if") that we don't know about.
3014 * - Packet arrived on the local port but the local port is not
3015 * one of our bridge ports.
3017 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3019 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
3020 "interface %"PRIu16, br->name, flow->in_port);
3026 *in_portp = in_port = in_iface->port;
3027 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
3032 /* Drop frames for reserved multicast addresses. */
3033 if (eth_addr_is_reserved(flow->dl_dst)) {
3037 /* Drop frames on ports reserved for mirroring. */
3038 if (in_port->is_mirror_output_port) {
3040 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3041 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
3042 "%s, which is reserved exclusively for mirroring",
3043 br->name, in_port->name);
3048 /* When using LACP, do not accept packets from disabled interfaces. */
3049 if (in_port->lacp & LACP_NEGOTIATED && !in_iface->enabled) {
3053 /* Packets received on non-LACP bonds need special attention to avoid
3055 if (in_port->n_ifaces > 1 && !(in_port->lacp & LACP_NEGOTIATED)) {
3057 bool is_grat_arp_locked;
3059 if (eth_addr_is_multicast(flow->dl_dst)) {
3060 *tags |= in_port->active_iface_tag;
3061 if (in_port->active_iface != in_iface->port_ifidx) {
3062 /* Drop all multicast packets on inactive slaves. */
3067 /* Drop all packets for which we have learned a different input
3068 * port, because we probably sent the packet on one slave and got
3069 * it back on the other. Gratuitous ARP packets are an exception
3070 * to this rule: the host has moved to another switch. The exception
3071 * to the exception is if we locked the learning table to avoid
3072 * reflections on bond slaves. If this is the case, just drop the
3074 if (in_port->bond_mode != BM_AB) {
3075 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
3076 &is_grat_arp_locked);
3077 if (src_idx != -1 && src_idx != in_port->port_idx &&
3078 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
3084 /* Drop all packets which arrive on backup slaves. This is similar to how
3085 * Linux bonding handles active-backup bonds. */
3086 if (in_port->bond_mode == BM_AB) {
3088 *tags |= in_port->active_iface;
3089 if (in_port->active_iface != in_iface->port_ifidx) {
3090 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3092 VLOG_WARN_RL(&rl, "active-backup bond received packet on backup"
3093 " interface (%s) destined for " ETH_ADDR_FMT,
3094 in_iface->name, ETH_ADDR_ARGS(flow->dl_dst));
3102 /* If the composed actions may be applied to any packet in the given 'flow',
3103 * returns true. Otherwise, the actions should only be applied to 'packet', or
3104 * not at all, if 'packet' was NULL. */
3106 process_flow(struct bridge *br, const struct flow *flow,
3107 const struct ofpbuf *packet, struct ofpbuf *actions,
3108 tag_type *tags, uint16_t *nf_output_iface)
3110 struct port *in_port;
3111 struct port *out_port;
3115 /* Check whether we should drop packets in this flow. */
3116 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
3121 /* Learn source MAC (but don't try to learn from revalidation). */
3123 update_learning_table(br, flow, vlan, in_port);
3126 /* Determine output port. */
3127 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
3129 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
3130 out_port = br->ports[out_port_idx];
3131 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
3132 /* If we are revalidating but don't have a learning entry then
3133 * eject the flow. Installing a flow that floods packets opens
3134 * up a window of time where we could learn from a packet reflected
3135 * on a bond and blackhole packets before the learning table is
3136 * updated to reflect the correct port. */
3139 out_port = FLOOD_PORT;
3142 /* Don't send packets out their input ports. */
3143 if (in_port == out_port) {
3149 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
3157 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
3158 struct ofpbuf *actions, tag_type *tags,
3159 uint16_t *nf_output_iface, void *br_)
3161 struct bridge *br = br_;
3163 COVERAGE_INC(bridge_process_flow);
3164 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
3168 bridge_special_ofhook_cb(const struct flow *flow,
3169 const struct ofpbuf *packet, void *br_)
3171 struct iface *iface;
3172 struct bridge *br = br_;
3174 iface = iface_from_dp_ifidx(br, flow->in_port);
3176 if (cfm_should_process_flow(flow)) {
3178 if (iface && packet && iface->cfm) {
3179 COVERAGE_INC(bridge_process_cfm);
3180 cfm_process_heartbeat(iface->cfm, packet);
3183 } else if (flow->dl_type == htons(ETH_TYPE_LACP)) {
3185 if (iface && packet) {
3186 COVERAGE_INC(bridge_process_lacp);
3187 lacp_process_packet(packet, iface);
3196 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
3197 const struct nlattr *actions,
3199 uint64_t n_bytes, void *br_)
3201 struct bridge *br = br_;
3202 const struct nlattr *a;
3203 struct port *in_port;
3208 /* Feed information from the active flows back into the learning table to
3209 * ensure that table is always in sync with what is actually flowing
3210 * through the datapath.
3212 * We test that 'tags' is nonzero to ensure that only flows that include an
3213 * OFPP_NORMAL action are used for learning. This works because
3214 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3215 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3216 update_learning_table(br, flow, vlan, in_port);
3219 /* Account for bond slave utilization. */
3220 if (!br->has_bonded_ports) {
3223 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3224 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3225 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3226 if (out_port && out_port->n_ifaces >= 2 &&
3227 out_port->bond_mode != BM_AB) {
3228 uint16_t vlan = (flow->vlan_tci
3229 ? vlan_tci_to_vid(flow->vlan_tci)
3231 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
3232 e->tx_bytes += n_bytes;
3239 bridge_account_checkpoint_ofhook_cb(void *br_)
3241 struct bridge *br = br_;
3245 if (!br->has_bonded_ports) {
3250 for (i = 0; i < br->n_ports; i++) {
3251 struct port *port = br->ports[i];
3252 if (port->n_ifaces > 1 && port->bond_mode != BM_AB
3253 && now >= port->bond_next_rebalance) {
3254 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3255 bond_rebalance_port(port);
3260 static struct ofhooks bridge_ofhooks = {
3261 bridge_normal_ofhook_cb,
3262 bridge_special_ofhook_cb,
3263 bridge_account_flow_ofhook_cb,
3264 bridge_account_checkpoint_ofhook_cb,
3267 /* LACP functions. */
3270 lacp_process_packet(const struct ofpbuf *packet, struct iface *iface)
3272 const struct lacp_pdu *pdu;
3274 if (!iface->port->lacp) {
3278 pdu = parse_lacp_packet(packet);
3283 iface->lacp_status |= LACP_CURRENT;
3284 iface->lacp_status &= ~(LACP_EXPIRED | LACP_DEFAULTED);
3285 iface->lacp_rx = time_msec() + LACP_SLOW_TIME_RX;
3287 iface->lacp_actor.state = iface_get_lacp_state(iface);
3288 if (memcmp(&iface->lacp_actor, &pdu->partner, sizeof pdu->partner)) {
3292 if (memcmp(&iface->lacp_partner, &pdu->actor, sizeof pdu->actor)) {
3293 iface->port->lacp_need_update = true;
3294 iface->lacp_partner = pdu->actor;
3299 lacp_update_ifaces(struct port *port)
3303 struct lacp_info lead_pri;
3304 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3306 port->lacp_need_update = false;
3307 COVERAGE_INC(bridge_lacp_update);
3313 VLOG_DBG_RL(&rl, "port %s: re-evaluating LACP link status", port->name);
3316 for (i = 0; i < port->n_ifaces; i++) {
3317 struct iface *iface = port->ifaces[i];
3318 struct lacp_info pri;
3320 iface->lacp_status |= LACP_ATTACHED;
3321 ofproto_revalidate(port->bridge->ofproto, iface->tag);
3323 /* Don't allow loopback interfaces to send traffic or lead. */
3324 if (eth_addr_equals(iface->lacp_partner.sysid,
3325 iface->lacp_actor.sysid)) {
3326 VLOG_WARN_RL(&rl, "iface %s: Loopback detected. Interface is "
3327 "connected to its own bridge", iface->name);
3328 iface->lacp_status &= ~LACP_ATTACHED;
3332 if (iface->lacp_status & LACP_DEFAULTED) {
3336 iface_get_lacp_priority(iface, &pri);
3338 if (!lead || memcmp(&pri, &lead_pri, sizeof pri) < 0) {
3345 port->lacp &= ~LACP_NEGOTIATED;
3349 port->lacp |= LACP_NEGOTIATED;
3351 for (i = 0; i < port->n_ifaces; i++) {
3352 struct iface *iface = port->ifaces[i];
3354 if (iface->lacp_status & LACP_DEFAULTED
3355 || lead->lacp_partner.key != iface->lacp_partner.key
3356 || !eth_addr_equals(lead->lacp_partner.sysid,
3357 iface->lacp_partner.sysid)) {
3358 iface->lacp_status &= ~LACP_ATTACHED;
3364 lacp_iface_may_tx(const struct iface *iface)
3366 return iface->port->lacp & LACP_ACTIVE
3367 || iface->lacp_status & (LACP_CURRENT | LACP_EXPIRED);
3371 lacp_run(struct port *port)
3374 struct ofpbuf packet;
3380 ofpbuf_init(&packet, ETH_HEADER_LEN + LACP_PDU_LEN);
3382 for (i = 0; i < port->n_ifaces; i++) {
3383 struct iface *iface = port->ifaces[i];
3385 if (time_msec() > iface->lacp_rx) {
3386 if (iface->lacp_status & LACP_CURRENT) {
3387 iface_set_lacp_expired(iface);
3388 } else if (iface->lacp_status & LACP_EXPIRED) {
3389 iface_set_lacp_defaulted(iface);
3394 if (port->lacp_need_update) {
3395 lacp_update_ifaces(port);
3398 for (i = 0; i < port->n_ifaces; i++) {
3399 struct iface *iface = port->ifaces[i];
3400 uint8_t ea[ETH_ADDR_LEN];
3403 if (time_msec() < iface->lacp_tx || !lacp_iface_may_tx(iface)) {
3407 error = netdev_get_etheraddr(iface->netdev, ea);
3409 iface->lacp_actor.state = iface_get_lacp_state(iface);
3410 compose_lacp_packet(&packet, &iface->lacp_actor,
3411 &iface->lacp_partner, ea);
3412 iface_send_packet(iface, &packet);
3414 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3415 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3416 "(%s)", iface->name, strerror(error));
3419 iface->lacp_tx = time_msec() +
3420 (iface->lacp_partner.state & LACP_STATE_TIME
3422 : LACP_SLOW_TIME_TX);
3424 ofpbuf_uninit(&packet);
3428 lacp_wait(struct port *port)
3436 for (i = 0; i < port->n_ifaces; i++) {
3437 struct iface *iface = port->ifaces[i];
3439 if (lacp_iface_may_tx(iface)) {
3440 poll_timer_wait_until(iface->lacp_tx);
3443 if (iface->lacp_status & (LACP_CURRENT | LACP_EXPIRED)) {
3444 poll_timer_wait_until(iface->lacp_rx);
3449 /* Bonding functions. */
3451 /* Statistics for a single interface on a bonded port, used for load-based
3452 * bond rebalancing. */
3453 struct slave_balance {
3454 struct iface *iface; /* The interface. */
3455 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3457 /* All the "bond_entry"s that are assigned to this interface, in order of
3458 * increasing tx_bytes. */
3459 struct bond_entry **hashes;
3464 bond_mode_to_string(enum bond_mode bm) {
3465 static char *bm_slb = "balance-slb";
3466 static char *bm_ab = "active-backup";
3467 static char *bm_tcp = "balance-tcp";
3470 case BM_SLB: return bm_slb;
3471 case BM_AB: return bm_ab;
3472 case BM_TCP: return bm_tcp;
3479 /* Sorts pointers to pointers to bond_entries in ascending order by the
3480 * interface to which they are assigned, and within a single interface in
3481 * ascending order of bytes transmitted. */
3483 compare_bond_entries(const void *a_, const void *b_)
3485 const struct bond_entry *const *ap = a_;
3486 const struct bond_entry *const *bp = b_;
3487 const struct bond_entry *a = *ap;
3488 const struct bond_entry *b = *bp;
3489 if (a->iface_idx != b->iface_idx) {
3490 return a->iface_idx > b->iface_idx ? 1 : -1;
3491 } else if (a->tx_bytes != b->tx_bytes) {
3492 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3498 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3499 * *descending* order by number of bytes transmitted. */
3501 compare_slave_balance(const void *a_, const void *b_)
3503 const struct slave_balance *a = a_;
3504 const struct slave_balance *b = b_;
3505 if (a->iface->enabled != b->iface->enabled) {
3506 return a->iface->enabled ? -1 : 1;
3507 } else if (a->tx_bytes != b->tx_bytes) {
3508 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3515 swap_bals(struct slave_balance *a, struct slave_balance *b)
3517 struct slave_balance tmp = *a;
3522 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3523 * given that 'p' (and only 'p') might be in the wrong location.
3525 * This function invalidates 'p', since it might now be in a different memory
3528 resort_bals(struct slave_balance *p,
3529 struct slave_balance bals[], size_t n_bals)
3532 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3533 swap_bals(p, p - 1);
3535 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3536 swap_bals(p, p + 1);
3542 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3544 if (VLOG_IS_DBG_ENABLED()) {
3545 struct ds ds = DS_EMPTY_INITIALIZER;
3546 const struct slave_balance *b;
3548 for (b = bals; b < bals + n_bals; b++) {
3552 ds_put_char(&ds, ',');
3554 ds_put_format(&ds, " %s %"PRIu64"kB",
3555 b->iface->name, b->tx_bytes / 1024);
3557 if (!b->iface->enabled) {
3558 ds_put_cstr(&ds, " (disabled)");
3560 if (b->n_hashes > 0) {
3561 ds_put_cstr(&ds, " (");
3562 for (i = 0; i < b->n_hashes; i++) {
3563 const struct bond_entry *e = b->hashes[i];
3565 ds_put_cstr(&ds, " + ");
3567 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3568 e - port->bond_hash, e->tx_bytes / 1024);
3570 ds_put_cstr(&ds, ")");
3573 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3578 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3580 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3583 struct bond_entry *hash = from->hashes[hash_idx];
3584 struct port *port = from->iface->port;
3585 uint64_t delta = hash->tx_bytes;
3587 assert(port->bond_mode != BM_AB);
3589 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3590 "from %s to %s (now carrying %"PRIu64"kB and "
3591 "%"PRIu64"kB load, respectively)",
3592 port->name, delta / 1024, hash - port->bond_hash,
3593 from->iface->name, to->iface->name,
3594 (from->tx_bytes - delta) / 1024,
3595 (to->tx_bytes + delta) / 1024);
3597 /* Delete element from from->hashes.
3599 * We don't bother to add the element to to->hashes because not only would
3600 * it require more work, the only purpose it would be to allow that hash to
3601 * be migrated to another slave in this rebalancing run, and there is no
3602 * point in doing that. */
3603 if (hash_idx == 0) {
3606 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3607 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3611 /* Shift load away from 'from' to 'to'. */
3612 from->tx_bytes -= delta;
3613 to->tx_bytes += delta;
3615 /* Arrange for flows to be revalidated. */
3616 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
3617 hash->iface_idx = to->iface->port_ifidx;
3618 hash->iface_tag = tag_create_random();
3622 bond_rebalance_port(struct port *port)
3624 struct slave_balance *bals;
3626 struct bond_entry *hashes[BOND_MASK + 1];
3627 struct slave_balance *b, *from, *to;
3628 struct bond_entry *e;
3631 assert(port->bond_mode != BM_AB);
3633 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3634 * descending order of tx_bytes, so that bals[0] represents the most
3635 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3638 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3639 * array for each slave_balance structure, we sort our local array of
3640 * hashes in order by slave, so that all of the hashes for a given slave
3641 * become contiguous in memory, and then we point each 'hashes' members of
3642 * a slave_balance structure to the start of a contiguous group. */
3643 n_bals = port->n_ifaces;
3644 bals = xmalloc(n_bals * sizeof *bals);
3645 for (b = bals; b < &bals[n_bals]; b++) {
3646 b->iface = port->ifaces[b - bals];
3651 for (i = 0; i <= BOND_MASK; i++) {
3652 hashes[i] = &port->bond_hash[i];
3654 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3655 for (i = 0; i <= BOND_MASK; i++) {
3657 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3658 b = &bals[e->iface_idx];
3659 b->tx_bytes += e->tx_bytes;
3661 b->hashes = &hashes[i];
3666 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3667 log_bals(bals, n_bals, port);
3669 /* Discard slaves that aren't enabled (which were sorted to the back of the
3670 * array earlier). */
3671 while (!bals[n_bals - 1].iface->enabled) {
3678 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3679 to = &bals[n_bals - 1];
3680 for (from = bals; from < to; ) {
3681 uint64_t overload = from->tx_bytes - to->tx_bytes;
3682 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3683 /* The extra load on 'from' (and all less-loaded slaves), compared
3684 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3685 * it is less than ~1Mbps. No point in rebalancing. */
3687 } else if (from->n_hashes == 1) {
3688 /* 'from' only carries a single MAC hash, so we can't shift any
3689 * load away from it, even though we want to. */
3692 /* 'from' is carrying significantly more load than 'to', and that
3693 * load is split across at least two different hashes. Pick a hash
3694 * to migrate to 'to' (the least-loaded slave), given that doing so
3695 * must decrease the ratio of the load on the two slaves by at
3698 * The sort order we use means that we prefer to shift away the
3699 * smallest hashes instead of the biggest ones. There is little
3700 * reason behind this decision; we could use the opposite sort
3701 * order to shift away big hashes ahead of small ones. */
3704 for (i = 0; i < from->n_hashes; i++) {
3705 double old_ratio, new_ratio;
3706 uint64_t delta = from->hashes[i]->tx_bytes;
3708 if (delta == 0 || from->tx_bytes - delta == 0) {
3709 /* Pointless move. */
3713 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3715 if (to->tx_bytes == 0) {
3716 /* Nothing on the new slave, move it. */
3720 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3721 new_ratio = (double)(from->tx_bytes - delta) /
3722 (to->tx_bytes + delta);
3724 if (new_ratio == 0) {
3725 /* Should already be covered but check to prevent division
3730 if (new_ratio < 1) {
3731 new_ratio = 1 / new_ratio;
3734 if (old_ratio - new_ratio > 0.1) {
3735 /* Would decrease the ratio, move it. */
3739 if (i < from->n_hashes) {
3740 bond_shift_load(from, to, i);
3742 /* If the result of the migration changed the relative order of
3743 * 'from' and 'to' swap them back to maintain invariants. */
3744 if (order_swapped) {
3745 swap_bals(from, to);
3748 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3749 * point to different slave_balance structures. It is only
3750 * valid to do these two operations in a row at all because we
3751 * know that 'from' will not move past 'to' and vice versa. */
3752 resort_bals(from, bals, n_bals);
3753 resort_bals(to, bals, n_bals);
3760 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3761 * historical data to decay to <1% in 7 rebalancing runs. */
3762 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3771 bond_send_learning_packets(struct port *port)
3773 struct bridge *br = port->bridge;
3774 struct mac_entry *e;
3775 struct ofpbuf packet;
3776 int error, n_packets, n_errors;
3778 if (!port->n_ifaces || port->active_iface < 0 || bond_is_tcp_hash(port)) {
3782 ofpbuf_init(&packet, 128);
3783 error = n_packets = n_errors = 0;
3784 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3785 union ofp_action actions[2], *a;
3791 if (e->port == port->port_idx) {
3795 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3797 flow_extract(&packet, 0, ODPP_NONE, &flow);
3799 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3803 /* Compose actions. */
3804 memset(actions, 0, sizeof actions);
3807 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3808 a->vlan_vid.len = htons(sizeof *a);
3809 a->vlan_vid.vlan_vid = htons(e->vlan);
3812 a->output.type = htons(OFPAT_OUTPUT);
3813 a->output.len = htons(sizeof *a);
3814 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3819 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3826 ofpbuf_uninit(&packet);
3829 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3830 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3831 "packets, last error was: %s",
3832 port->name, n_errors, n_packets, strerror(error));
3834 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3835 port->name, n_packets);
3839 /* Bonding unixctl user interface functions. */
3842 bond_unixctl_list(struct unixctl_conn *conn,
3843 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3845 struct ds ds = DS_EMPTY_INITIALIZER;
3846 const struct bridge *br;
3848 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3850 LIST_FOR_EACH (br, node, &all_bridges) {
3853 for (i = 0; i < br->n_ports; i++) {
3854 const struct port *port = br->ports[i];
3855 if (port->n_ifaces > 1) {
3858 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3859 bond_mode_to_string(port->bond_mode));
3860 for (j = 0; j < port->n_ifaces; j++) {
3861 const struct iface *iface = port->ifaces[j];
3863 ds_put_cstr(&ds, ", ");
3865 ds_put_cstr(&ds, iface->name);
3867 ds_put_char(&ds, '\n');
3871 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3875 static struct port *
3876 bond_find(const char *name)
3878 const struct bridge *br;
3880 LIST_FOR_EACH (br, node, &all_bridges) {
3883 for (i = 0; i < br->n_ports; i++) {
3884 struct port *port = br->ports[i];
3885 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3894 ds_put_lacp_state(struct ds *ds, uint8_t state)
3896 if (state & LACP_STATE_ACT) {
3897 ds_put_cstr(ds, "activity ");
3900 if (state & LACP_STATE_TIME) {
3901 ds_put_cstr(ds, "timeout ");
3904 if (state & LACP_STATE_AGG) {
3905 ds_put_cstr(ds, "aggregation ");
3908 if (state & LACP_STATE_SYNC) {
3909 ds_put_cstr(ds, "synchronized ");
3912 if (state & LACP_STATE_COL) {
3913 ds_put_cstr(ds, "collecting ");
3916 if (state & LACP_STATE_DIST) {
3917 ds_put_cstr(ds, "distributing ");
3920 if (state & LACP_STATE_DEF) {
3921 ds_put_cstr(ds, "defaulted ");
3924 if (state & LACP_STATE_EXP) {
3925 ds_put_cstr(ds, "expired ");
3930 bond_unixctl_show(struct unixctl_conn *conn,
3931 const char *args, void *aux OVS_UNUSED)
3933 struct ds ds = DS_EMPTY_INITIALIZER;
3934 const struct port *port;
3937 port = bond_find(args);
3939 unixctl_command_reply(conn, 501, "no such bond");
3943 ds_put_format(&ds, "bond_mode: %s\n",
3944 bond_mode_to_string(port->bond_mode));
3947 ds_put_format(&ds, "lacp: %s\n",
3948 port->lacp & LACP_ACTIVE ? "active" : "passive");
3950 ds_put_cstr(&ds, "lacp: off\n");
3953 if (port->bond_mode != BM_AB) {
3954 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3955 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3959 ds_put_format(&ds, "bond-detect-mode: %s\n",
3960 port->miimon ? "miimon" : "carrier");
3963 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3964 port->bond_miimon_interval);
3967 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3968 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3970 if (port->bond_mode != BM_AB) {
3971 ds_put_format(&ds, "next rebalance: %lld ms\n",
3972 port->bond_next_rebalance - time_msec());
3975 for (j = 0; j < port->n_ifaces; j++) {
3976 const struct iface *iface = port->ifaces[j];
3977 struct bond_entry *be;
3981 ds_put_format(&ds, "\nslave %s: %s\n",
3982 iface->name, iface->enabled ? "enabled" : "disabled");
3983 if (j == port->active_iface) {
3984 ds_put_cstr(&ds, "\tactive slave\n");
3986 if (iface->delay_expires != LLONG_MAX) {
3987 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3988 iface->enabled ? "downdelay" : "updelay",
3989 iface->delay_expires - time_msec());
3993 ds_put_cstr(&ds, "\tstatus: ");
3995 if (iface->lacp_status & LACP_CURRENT) {
3996 ds_put_cstr(&ds, "current ");
3999 if (iface->lacp_status & LACP_EXPIRED) {
4000 ds_put_cstr(&ds, "expired ");
4003 if (iface->lacp_status & LACP_DEFAULTED) {
4004 ds_put_cstr(&ds, "defaulted ");
4007 if (iface->lacp_status & LACP_ATTACHED) {
4008 ds_put_cstr(&ds, "attached ");
4011 ds_put_cstr(&ds, "\n");
4013 ds_put_cstr(&ds, "\n\tactor sysid: ");
4014 ds_put_format(&ds, ETH_ADDR_FMT,
4015 ETH_ADDR_ARGS(iface->lacp_actor.sysid));
4016 ds_put_cstr(&ds, "\n");
4018 ds_put_format(&ds, "\tactor sys_priority: %u\n",
4019 ntohs(iface->lacp_actor.sys_priority));
4021 ds_put_format(&ds, "\tactor portid: %u\n",
4022 ntohs(iface->lacp_actor.portid));
4024 ds_put_format(&ds, "\tactor port_priority: %u\n",
4025 ntohs(iface->lacp_actor.port_priority));
4027 ds_put_format(&ds, "\tactor key: %u\n",
4028 ntohs(iface->lacp_actor.key));
4030 ds_put_cstr(&ds, "\tactor state: ");
4031 ds_put_lacp_state(&ds, iface_get_lacp_state(iface));
4032 ds_put_cstr(&ds, "\n\n");
4034 ds_put_cstr(&ds, "\tpartner sysid: ");
4035 ds_put_format(&ds, ETH_ADDR_FMT,
4036 ETH_ADDR_ARGS(iface->lacp_partner.sysid));
4037 ds_put_cstr(&ds, "\n");
4039 ds_put_format(&ds, "\tpartner sys_priority: %u\n",
4040 ntohs(iface->lacp_partner.sys_priority));
4042 ds_put_format(&ds, "\tpartner portid: %u\n",
4043 ntohs(iface->lacp_partner.portid));
4045 ds_put_format(&ds, "\tpartner port_priority: %u\n",
4046 ntohs(iface->lacp_partner.port_priority));
4048 ds_put_format(&ds, "\tpartner key: %u\n",
4049 ntohs(iface->lacp_partner.key));
4051 ds_put_cstr(&ds, "\tpartner state: ");
4052 ds_put_lacp_state(&ds, iface->lacp_partner.state);
4053 ds_put_cstr(&ds, "\n");
4056 if (port->bond_mode == BM_AB) {
4061 memset(&flow, 0, sizeof flow);
4062 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
4063 int hash = be - port->bond_hash;
4064 struct mac_entry *me;
4066 if (be->iface_idx != j) {
4070 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
4071 hash, be->tx_bytes / 1024);
4073 if (port->bond_mode != BM_SLB) {
4078 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
4082 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
4083 if (bond_hash_src(me->mac, me->vlan) == hash
4084 && me->port != port->port_idx
4085 && choose_output_iface(port, &flow, me->vlan,
4087 && dp_ifidx == iface->dp_ifidx)
4089 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
4090 ETH_ADDR_ARGS(me->mac));
4095 unixctl_command_reply(conn, 200, ds_cstr(&ds));
4100 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
4101 void *aux OVS_UNUSED)
4103 char *args = (char *) args_;
4104 char *save_ptr = NULL;
4105 char *bond_s, *hash_s, *slave_s;
4107 struct iface *iface;
4108 struct bond_entry *entry;
4111 bond_s = strtok_r(args, " ", &save_ptr);
4112 hash_s = strtok_r(NULL, " ", &save_ptr);
4113 slave_s = strtok_r(NULL, " ", &save_ptr);
4115 unixctl_command_reply(conn, 501,
4116 "usage: bond/migrate BOND HASH SLAVE");
4120 port = bond_find(bond_s);
4122 unixctl_command_reply(conn, 501, "no such bond");
4126 if (port->bond_mode != BM_SLB) {
4127 unixctl_command_reply(conn, 501, "not an SLB bond");
4131 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
4132 hash = atoi(hash_s) & BOND_MASK;
4134 unixctl_command_reply(conn, 501, "bad hash");
4138 iface = port_lookup_iface(port, slave_s);
4140 unixctl_command_reply(conn, 501, "no such slave");
4144 if (!iface->enabled) {
4145 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
4149 entry = &port->bond_hash[hash];
4150 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
4151 entry->iface_idx = iface->port_ifidx;
4152 entry->iface_tag = tag_create_random();
4153 unixctl_command_reply(conn, 200, "migrated");
4157 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
4158 void *aux OVS_UNUSED)
4160 char *args = (char *) args_;
4161 char *save_ptr = NULL;
4162 char *bond_s, *slave_s;
4164 struct iface *iface;
4166 bond_s = strtok_r(args, " ", &save_ptr);
4167 slave_s = strtok_r(NULL, " ", &save_ptr);
4169 unixctl_command_reply(conn, 501,
4170 "usage: bond/set-active-slave BOND SLAVE");
4174 port = bond_find(bond_s);
4176 unixctl_command_reply(conn, 501, "no such bond");
4180 iface = port_lookup_iface(port, slave_s);
4182 unixctl_command_reply(conn, 501, "no such slave");
4186 if (!iface->enabled) {
4187 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
4191 if (port->active_iface != iface->port_ifidx) {
4192 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4193 port->active_iface = iface->port_ifidx;
4194 port->active_iface_tag = tag_create_random();
4195 VLOG_INFO("port %s: active interface is now %s",
4196 port->name, iface->name);
4197 bond_send_learning_packets(port);
4198 unixctl_command_reply(conn, 200, "done");
4200 unixctl_command_reply(conn, 200, "no change");
4205 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
4207 char *args = (char *) args_;
4208 char *save_ptr = NULL;
4209 char *bond_s, *slave_s;
4211 struct iface *iface;
4213 bond_s = strtok_r(args, " ", &save_ptr);
4214 slave_s = strtok_r(NULL, " ", &save_ptr);
4216 unixctl_command_reply(conn, 501,
4217 "usage: bond/enable/disable-slave BOND SLAVE");
4221 port = bond_find(bond_s);
4223 unixctl_command_reply(conn, 501, "no such bond");
4227 iface = port_lookup_iface(port, slave_s);
4229 unixctl_command_reply(conn, 501, "no such slave");
4233 bond_enable_slave(iface, enable);
4234 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
4238 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
4239 void *aux OVS_UNUSED)
4241 enable_slave(conn, args, true);
4245 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
4246 void *aux OVS_UNUSED)
4248 enable_slave(conn, args, false);
4252 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
4253 void *aux OVS_UNUSED)
4255 char *args = (char *) args_;
4256 uint8_t mac[ETH_ADDR_LEN];
4260 char *mac_s, *vlan_s;
4261 char *save_ptr = NULL;
4263 mac_s = strtok_r(args, " ", &save_ptr);
4264 vlan_s = strtok_r(NULL, " ", &save_ptr);
4267 if (sscanf(vlan_s, "%u", &vlan) != 1) {
4268 unixctl_command_reply(conn, 501, "invalid vlan");
4272 vlan = OFP_VLAN_NONE;
4275 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
4276 == ETH_ADDR_SCAN_COUNT) {
4277 hash = bond_hash_src(mac, vlan);
4279 hash_cstr = xasprintf("%u", hash);
4280 unixctl_command_reply(conn, 200, hash_cstr);
4283 unixctl_command_reply(conn, 501, "invalid mac");
4290 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
4291 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
4292 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
4293 unixctl_command_register("bond/set-active-slave",
4294 bond_unixctl_set_active_slave, NULL);
4295 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
4297 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
4299 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
4302 /* Port functions. */
4305 port_run(struct port *port)
4312 for (i = 0; i < port->n_ifaces; i++) {
4313 struct iface *iface = port->ifaces[i];
4316 struct ofpbuf *packet = cfm_run(iface->cfm);
4318 iface_send_packet(iface, packet);
4319 ofpbuf_uninit(packet);
4327 port_wait(struct port *port)
4334 for (i = 0; i < port->n_ifaces; i++) {
4335 struct iface *iface = port->ifaces[i];
4337 cfm_wait(iface->cfm);
4342 static struct port *
4343 port_create(struct bridge *br, const char *name)
4347 port = xzalloc(sizeof *port);
4349 port->port_idx = br->n_ports;
4351 port->trunks = NULL;
4352 port->name = xstrdup(name);
4353 port->active_iface = -1;
4355 if (br->n_ports >= br->allocated_ports) {
4356 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
4359 br->ports[br->n_ports++] = port;
4360 shash_add_assert(&br->port_by_name, port->name, port);
4362 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
4369 get_port_other_config(const struct ovsrec_port *port, const char *key,
4370 const char *default_value)
4374 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
4376 return value ? value : default_value;
4380 get_interface_other_config(const struct ovsrec_interface *iface,
4381 const char *key, const char *default_value)
4385 value = get_ovsrec_key_value(&iface->header_,
4386 &ovsrec_interface_col_other_config, key);
4387 return value ? value : default_value;
4391 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
4393 struct shash new_ifaces;
4396 /* Collect list of new interfaces. */
4397 shash_init(&new_ifaces);
4398 for (i = 0; i < cfg->n_interfaces; i++) {
4399 const char *name = cfg->interfaces[i]->name;
4400 shash_add_once(&new_ifaces, name, NULL);
4403 /* Get rid of deleted interfaces. */
4404 for (i = 0; i < port->n_ifaces; ) {
4405 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
4406 iface_destroy(port->ifaces[i]);
4412 shash_destroy(&new_ifaces);
4416 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
4418 const char *detect_mode;
4419 struct shash new_ifaces;
4420 long long int next_rebalance, miimon_next_update, lacp_priority;
4421 unsigned long *trunks;
4427 /* Update settings. */
4428 port->updelay = cfg->bond_updelay;
4429 if (port->updelay < 0) {
4432 port->downdelay = cfg->bond_downdelay;
4433 if (port->downdelay < 0) {
4434 port->downdelay = 0;
4436 port->bond_rebalance_interval = atoi(
4437 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
4438 if (port->bond_rebalance_interval < 1000) {
4439 port->bond_rebalance_interval = 1000;
4441 next_rebalance = time_msec() + port->bond_rebalance_interval;
4442 if (port->bond_next_rebalance > next_rebalance) {
4443 port->bond_next_rebalance = next_rebalance;
4446 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
4449 if (!strcmp(detect_mode, "carrier")) {
4450 port->miimon = false;
4451 } else if (!strcmp(detect_mode, "miimon")) {
4452 port->miimon = true;
4454 port->miimon = false;
4455 VLOG_WARN("port %s: unsupported bond-detect-mode %s, defaulting to "
4456 "carrier", port->name, detect_mode);
4459 port->bond_miimon_interval = atoi(
4460 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4461 if (port->bond_miimon_interval < 100) {
4462 port->bond_miimon_interval = 100;
4464 miimon_next_update = time_msec() + port->bond_miimon_interval;
4465 if (port->bond_miimon_next_update > miimon_next_update) {
4466 port->bond_miimon_next_update = miimon_next_update;
4469 if (!port->cfg->bond_mode ||
4470 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4471 port->bond_mode = BM_SLB;
4472 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4473 port->bond_mode = BM_AB;
4474 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
4475 port->bond_mode = BM_TCP;
4477 port->bond_mode = BM_SLB;
4478 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4479 port->name, port->cfg->bond_mode,
4480 bond_mode_to_string(port->bond_mode));
4483 /* Add new interfaces and update 'cfg' member of existing ones. */
4484 shash_init(&new_ifaces);
4485 for (i = 0; i < cfg->n_interfaces; i++) {
4486 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4487 struct iface *iface;
4489 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
4490 VLOG_WARN("port %s: %s specified twice as port interface",
4491 port->name, if_cfg->name);
4492 iface_set_ofport(if_cfg, -1);
4496 iface = iface_lookup(port->bridge, if_cfg->name);
4498 if (iface->port != port) {
4499 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4501 port->bridge->name, if_cfg->name, iface->port->name);
4504 iface->cfg = if_cfg;
4506 iface = iface_create(port, if_cfg);
4509 /* Determine interface type. The local port always has type
4510 * "internal". Other ports take their type from the database and
4511 * default to "system" if none is specified. */
4512 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4513 : if_cfg->type[0] ? if_cfg->type
4517 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4520 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4521 iface->lacp_priority = UINT16_MAX;
4523 iface->lacp_priority = lacp_priority;
4526 shash_destroy(&new_ifaces);
4529 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4531 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4532 /* Prefer bondable links if unspecified. */
4533 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4535 port->lacp_priority = lacp_priority;
4538 if (!port->cfg->lacp) {
4539 /* XXX when LACP implementation has been sufficiently tested, enable by
4540 * default and make active on bonded ports. */
4542 } else if (!strcmp(port->cfg->lacp, "off")) {
4544 } else if (!strcmp(port->cfg->lacp, "active")) {
4545 port->lacp = LACP_ACTIVE;
4546 } else if (!strcmp(port->cfg->lacp, "passive")) {
4547 port->lacp = LACP_PASSIVE;
4549 VLOG_WARN("port %s: unknown LACP mode %s",
4550 port->name, port->cfg->lacp);
4557 if (port->n_ifaces < 2) {
4559 if (vlan >= 0 && vlan <= 4095) {
4560 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4565 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4566 * they even work as-is. But they have not been tested. */
4567 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4571 if (port->vlan != vlan) {
4573 bridge_flush(port->bridge);
4576 /* Get trunked VLANs. */
4578 if (vlan < 0 && cfg->n_trunks) {
4581 trunks = bitmap_allocate(4096);
4583 for (i = 0; i < cfg->n_trunks; i++) {
4584 int trunk = cfg->trunks[i];
4586 bitmap_set1(trunks, trunk);
4592 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4593 port->name, cfg->n_trunks);
4595 if (n_errors == cfg->n_trunks) {
4596 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4598 bitmap_free(trunks);
4601 } else if (vlan >= 0 && cfg->n_trunks) {
4602 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4606 ? port->trunks != NULL
4607 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4608 bridge_flush(port->bridge);
4610 bitmap_free(port->trunks);
4611 port->trunks = trunks;
4615 port_destroy(struct port *port)
4618 struct bridge *br = port->bridge;
4622 for (i = 0; i < MAX_MIRRORS; i++) {
4623 struct mirror *m = br->mirrors[i];
4624 if (m && m->out_port == port) {
4629 while (port->n_ifaces > 0) {
4630 iface_destroy(port->ifaces[port->n_ifaces - 1]);
4633 shash_find_and_delete_assert(&br->port_by_name, port->name);
4635 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
4636 del->port_idx = port->port_idx;
4638 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4640 netdev_monitor_destroy(port->monitor);
4642 bitmap_free(port->trunks);
4643 free(port->bond_hash);
4650 static struct port *
4651 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4653 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4654 return iface ? iface->port : NULL;
4657 static struct port *
4658 port_lookup(const struct bridge *br, const char *name)
4660 return shash_find_data(&br->port_by_name, name);
4663 static struct iface *
4664 port_lookup_iface(const struct port *port, const char *name)
4666 struct iface *iface = iface_lookup(port->bridge, name);
4667 return iface && iface->port == port ? iface : NULL;
4671 port_update_lacp(struct port *port)
4676 if (!port->lacp || port->n_ifaces < 1) {
4677 for (i = 0; i < port->n_ifaces; i++) {
4678 iface_set_lacp_defaulted(port->ifaces[i]);
4684 for (i = 0; i < port->n_ifaces; i++) {
4685 struct iface *iface = port->ifaces[i];
4687 if (iface->dp_ifidx <= 0 || iface->dp_ifidx > UINT16_MAX) {
4692 if (iface->dp_ifidx == port->lacp_key) {
4693 key_changed = false;
4698 port->lacp_key = port->ifaces[0]->dp_ifidx;
4701 for (i = 0; i < port->n_ifaces; i++) {
4702 struct iface *iface = port->ifaces[i];
4704 iface->lacp_actor.sys_priority = htons(port->lacp_priority);
4705 memcpy(&iface->lacp_actor.sysid, port->bridge->ea, ETH_ADDR_LEN);
4707 iface->lacp_actor.port_priority = htons(iface->lacp_priority);
4708 iface->lacp_actor.portid = htons(iface->dp_ifidx);
4709 iface->lacp_actor.key = htons(port->lacp_key);
4713 port->lacp_need_update = true;
4717 port_update_bonding(struct port *port)
4719 if (port->monitor) {
4720 netdev_monitor_destroy(port->monitor);
4721 port->monitor = NULL;
4723 if (port->n_ifaces < 2) {
4724 /* Not a bonded port. */
4725 free(port->bond_hash);
4726 port->bond_hash = NULL;
4727 port->bond_fake_iface = false;
4728 port->active_iface = -1;
4729 port->no_ifaces_tag = 0;
4733 if (port->bond_mode != BM_AB && !port->bond_hash) {
4734 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4735 for (i = 0; i <= BOND_MASK; i++) {
4736 struct bond_entry *e = &port->bond_hash[i];
4740 port->bond_next_rebalance
4741 = time_msec() + port->bond_rebalance_interval;
4742 } else if (port->bond_mode == BM_AB) {
4743 free(port->bond_hash);
4744 port->bond_hash = NULL;
4747 if (!port->no_ifaces_tag) {
4748 port->no_ifaces_tag = tag_create_random();
4751 if (port->active_iface < 0) {
4752 bond_choose_active_iface(port);
4755 port->bond_fake_iface = port->cfg->bond_fake_iface;
4756 if (port->bond_fake_iface) {
4757 port->bond_next_fake_iface_update = time_msec();
4760 if (!port->miimon) {
4761 port->monitor = netdev_monitor_create();
4762 for (i = 0; i < port->n_ifaces; i++) {
4763 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
4769 /* Interface functions. */
4772 iface_set_lacp_defaulted(struct iface *iface)
4774 memset(&iface->lacp_partner, 0, sizeof iface->lacp_partner);
4776 iface->lacp_status |= LACP_DEFAULTED;
4777 iface->lacp_status &= ~(LACP_CURRENT | LACP_EXPIRED);
4779 iface->port->lacp_need_update = true;
4783 iface_set_lacp_expired(struct iface *iface)
4785 iface->lacp_status &= ~LACP_CURRENT;
4786 iface->lacp_status |= LACP_EXPIRED;
4787 iface->lacp_partner.state |= LACP_STATE_TIME;
4788 iface->lacp_partner.state &= ~LACP_STATE_SYNC;
4790 iface->lacp_rx = time_msec() + LACP_FAST_TIME_RX;
4795 iface_get_lacp_state(const struct iface *iface)
4799 if (iface->port->lacp & LACP_ACTIVE) {
4800 state |= LACP_STATE_ACT;
4803 if (iface->lacp_status & LACP_ATTACHED) {
4804 state |= LACP_STATE_SYNC;
4807 if (iface->lacp_status & LACP_DEFAULTED) {
4808 state |= LACP_STATE_DEF;
4811 if (iface->lacp_status & LACP_EXPIRED) {
4812 state |= LACP_STATE_EXP;
4815 if (iface->port->n_ifaces > 1) {
4816 state |= LACP_STATE_AGG;
4819 if (iface->enabled) {
4820 state |= LACP_STATE_COL | LACP_STATE_DIST;
4826 /* Given 'iface', populates 'priority' with data representing its LACP link
4827 * priority. If two priority objects populated by this function are compared
4828 * using memcmp, the higher priority link will be less than the lower priority
4831 iface_get_lacp_priority(struct iface *iface, struct lacp_info *priority)
4833 uint16_t partner_priority, actor_priority;
4835 /* Choose the lacp_info of the higher priority system by comparing their
4836 * system priorities and mac addresses. */
4837 actor_priority = ntohs(iface->lacp_actor.sys_priority);
4838 partner_priority = ntohs(iface->lacp_partner.sys_priority);
4839 if (actor_priority < partner_priority) {
4840 *priority = iface->lacp_actor;
4841 } else if (partner_priority < actor_priority) {
4842 *priority = iface->lacp_partner;
4843 } else if (eth_addr_compare_3way(iface->lacp_actor.sysid,
4844 iface->lacp_partner.sysid) < 0) {
4845 *priority = iface->lacp_actor;
4847 *priority = iface->lacp_partner;
4850 /* Key and state are not used in priority comparisons. */
4852 priority->state = 0;
4856 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
4859 union ofp_action action;
4861 memset(&action, 0, sizeof action);
4862 action.output.type = htons(OFPAT_OUTPUT);
4863 action.output.len = htons(sizeof action);
4864 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
4866 flow_extract(packet, 0, ODPP_NONE, &flow);
4868 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
4870 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4871 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
4875 static struct iface *
4876 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4878 struct bridge *br = port->bridge;
4879 struct iface *iface;
4880 char *name = if_cfg->name;
4882 iface = xzalloc(sizeof *iface);
4884 iface->port_ifidx = port->n_ifaces;
4885 iface->name = xstrdup(name);
4886 iface->dp_ifidx = -1;
4887 iface->tag = tag_create_random();
4888 iface->delay_expires = LLONG_MAX;
4889 iface->netdev = NULL;
4890 iface->cfg = if_cfg;
4891 iface_set_lacp_defaulted(iface);
4893 if (port->lacp & LACP_ACTIVE) {
4894 iface_set_lacp_expired(iface);
4897 shash_add_assert(&br->iface_by_name, iface->name, iface);
4899 if (port->n_ifaces >= port->allocated_ifaces) {
4900 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
4901 sizeof *port->ifaces);
4903 port->ifaces[port->n_ifaces++] = iface;
4904 if (port->n_ifaces > 1) {
4905 br->has_bonded_ports = true;
4908 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4916 iface_destroy(struct iface *iface)
4919 struct port *port = iface->port;
4920 struct bridge *br = port->bridge;
4921 bool del_active = port->active_iface == iface->port_ifidx;
4924 if (port->monitor) {
4925 netdev_monitor_remove(port->monitor, iface->netdev);
4928 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4930 if (iface->dp_ifidx >= 0) {
4931 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4934 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
4935 del->port_ifidx = iface->port_ifidx;
4937 netdev_close(iface->netdev);
4940 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4941 bond_choose_active_iface(port);
4942 bond_send_learning_packets(port);
4945 cfm_destroy(iface->cfm);
4950 bridge_flush(port->bridge);
4954 static struct iface *
4955 iface_lookup(const struct bridge *br, const char *name)
4957 return shash_find_data(&br->iface_by_name, name);
4960 static struct iface *
4961 iface_find(const char *name)
4963 const struct bridge *br;
4965 LIST_FOR_EACH (br, node, &all_bridges) {
4966 struct iface *iface = iface_lookup(br, name);
4975 static struct iface *
4976 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4978 struct iface *iface;
4980 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4981 hash_int(dp_ifidx, 0), &br->ifaces) {
4982 if (iface->dp_ifidx == dp_ifidx) {
4989 /* Set Ethernet address of 'iface', if one is specified in the configuration
4992 iface_set_mac(struct iface *iface)
4994 uint8_t ea[ETH_ADDR_LEN];
4996 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4997 if (eth_addr_is_multicast(ea)) {
4998 VLOG_ERR("interface %s: cannot set MAC to multicast address",
5000 } else if (iface->dp_ifidx == ODPP_LOCAL) {
5001 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
5002 iface->name, iface->name);
5004 int error = netdev_set_etheraddr(iface->netdev, ea);
5006 VLOG_ERR("interface %s: setting MAC failed (%s)",
5007 iface->name, strerror(error));
5013 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
5015 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
5018 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
5022 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
5024 * The value strings in '*shash' are taken directly from values[], not copied,
5025 * so the caller should not modify or free them. */
5027 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
5028 struct shash *shash)
5033 for (i = 0; i < n; i++) {
5034 shash_add(shash, keys[i], values[i]);
5038 /* Creates 'keys' and 'values' arrays from 'shash'.
5040 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
5041 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
5042 * are populated with with strings taken directly from 'shash' and thus have
5043 * the same ownership of the key-value pairs in shash.
5046 shash_to_ovs_idl_map(struct shash *shash,
5047 char ***keys, char ***values, size_t *n)
5051 struct shash_node *sn;
5053 count = shash_count(shash);
5055 k = xmalloc(count * sizeof *k);
5056 v = xmalloc(count * sizeof *v);
5059 SHASH_FOR_EACH(sn, shash) {
5070 struct iface_delete_queues_cbdata {
5071 struct netdev *netdev;
5072 const struct ovsdb_datum *queues;
5076 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
5078 union ovsdb_atom atom;
5080 atom.integer = target;
5081 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
5085 iface_delete_queues(unsigned int queue_id,
5086 const struct shash *details OVS_UNUSED, void *cbdata_)
5088 struct iface_delete_queues_cbdata *cbdata = cbdata_;
5090 if (!queue_ids_include(cbdata->queues, queue_id)) {
5091 netdev_delete_queue(cbdata->netdev, queue_id);
5096 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
5098 if (!qos || qos->type[0] == '\0') {
5099 netdev_set_qos(iface->netdev, NULL, NULL);
5101 struct iface_delete_queues_cbdata cbdata;
5102 struct shash details;
5105 /* Configure top-level Qos for 'iface'. */
5106 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
5107 qos->n_other_config, &details);
5108 netdev_set_qos(iface->netdev, qos->type, &details);
5109 shash_destroy(&details);
5111 /* Deconfigure queues that were deleted. */
5112 cbdata.netdev = iface->netdev;
5113 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
5115 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
5117 /* Configure queues for 'iface'. */
5118 for (i = 0; i < qos->n_queues; i++) {
5119 const struct ovsrec_queue *queue = qos->value_queues[i];
5120 unsigned int queue_id = qos->key_queues[i];
5122 shash_from_ovs_idl_map(queue->key_other_config,
5123 queue->value_other_config,
5124 queue->n_other_config, &details);
5125 netdev_set_queue(iface->netdev, queue_id, &details);
5126 shash_destroy(&details);
5132 iface_update_cfm(struct iface *iface)
5136 uint16_t *remote_mps;
5137 struct ovsrec_monitor *mon;
5138 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
5140 mon = iface->cfg->monitor;
5143 cfm_destroy(iface->cfm);
5148 if (netdev_get_etheraddr(iface->netdev, ea)) {
5149 VLOG_WARN("interface %s: Failed to get ethernet address. "
5150 "Skipping Monitor.", iface->name);
5154 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
5155 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
5160 iface->cfm = cfm_create();
5164 cfm->mpid = mon->mpid;
5165 cfm->interval = mon->interval ? *mon->interval : 1000;
5167 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
5168 memcpy(cfm->maid, maid, sizeof cfm->maid);
5170 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
5171 for(i = 0; i < mon->n_remote_mps; i++) {
5172 remote_mps[i] = mon->remote_mps[i]->mpid;
5174 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
5177 if (!cfm_configure(iface->cfm)) {
5178 cfm_destroy(iface->cfm);
5183 /* Port mirroring. */
5185 static struct mirror *
5186 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
5190 for (i = 0; i < MAX_MIRRORS; i++) {
5191 struct mirror *m = br->mirrors[i];
5192 if (m && uuid_equals(uuid, &m->uuid)) {
5200 mirror_reconfigure(struct bridge *br)
5202 unsigned long *rspan_vlans;
5205 /* Get rid of deleted mirrors. */
5206 for (i = 0; i < MAX_MIRRORS; i++) {
5207 struct mirror *m = br->mirrors[i];
5209 const struct ovsdb_datum *mc;
5210 union ovsdb_atom atom;
5212 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
5213 atom.uuid = br->mirrors[i]->uuid;
5214 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
5220 /* Add new mirrors and reconfigure existing ones. */
5221 for (i = 0; i < br->cfg->n_mirrors; i++) {
5222 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
5223 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
5225 mirror_reconfigure_one(m, cfg);
5227 mirror_create(br, cfg);
5231 /* Update port reserved status. */
5232 for (i = 0; i < br->n_ports; i++) {
5233 br->ports[i]->is_mirror_output_port = false;
5235 for (i = 0; i < MAX_MIRRORS; i++) {
5236 struct mirror *m = br->mirrors[i];
5237 if (m && m->out_port) {
5238 m->out_port->is_mirror_output_port = true;
5242 /* Update flooded vlans (for RSPAN). */
5244 if (br->cfg->n_flood_vlans) {
5245 rspan_vlans = bitmap_allocate(4096);
5247 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
5248 int64_t vlan = br->cfg->flood_vlans[i];
5249 if (vlan >= 0 && vlan < 4096) {
5250 bitmap_set1(rspan_vlans, vlan);
5251 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
5254 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
5259 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
5265 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
5270 for (i = 0; ; i++) {
5271 if (i >= MAX_MIRRORS) {
5272 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
5273 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
5276 if (!br->mirrors[i]) {
5281 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
5284 br->mirrors[i] = m = xzalloc(sizeof *m);
5287 m->name = xstrdup(cfg->name);
5288 shash_init(&m->src_ports);
5289 shash_init(&m->dst_ports);
5295 mirror_reconfigure_one(m, cfg);
5299 mirror_destroy(struct mirror *m)
5302 struct bridge *br = m->bridge;
5305 for (i = 0; i < br->n_ports; i++) {
5306 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
5307 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
5310 shash_destroy(&m->src_ports);
5311 shash_destroy(&m->dst_ports);
5314 m->bridge->mirrors[m->idx] = NULL;
5323 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
5324 struct shash *names)
5328 for (i = 0; i < n_ports; i++) {
5329 const char *name = ports[i]->name;
5330 if (port_lookup(m->bridge, name)) {
5331 shash_add_once(names, name, NULL);
5333 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
5334 "port %s", m->bridge->name, m->name, name);
5340 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
5346 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
5348 for (i = 0; i < cfg->n_select_vlan; i++) {
5349 int64_t vlan = cfg->select_vlan[i];
5350 if (vlan < 0 || vlan > 4095) {
5351 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
5352 m->bridge->name, m->name, vlan);
5354 (*vlans)[n_vlans++] = vlan;
5361 vlan_is_mirrored(const struct mirror *m, int vlan)
5365 for (i = 0; i < m->n_vlans; i++) {
5366 if (m->vlans[i] == vlan) {
5374 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
5378 for (i = 0; i < m->n_vlans; i++) {
5379 if (port_trunks_vlan(p, m->vlans[i])) {
5387 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
5389 struct shash src_ports, dst_ports;
5390 mirror_mask_t mirror_bit;
5391 struct port *out_port;
5398 if (strcmp(cfg->name, m->name)) {
5400 m->name = xstrdup(cfg->name);
5403 /* Get output port. */
5404 if (cfg->output_port) {
5405 out_port = port_lookup(m->bridge, cfg->output_port->name);
5407 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
5408 m->bridge->name, m->name);
5414 if (cfg->output_vlan) {
5415 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
5416 "output vlan; ignoring output vlan",
5417 m->bridge->name, m->name);
5419 } else if (cfg->output_vlan) {
5421 out_vlan = *cfg->output_vlan;
5423 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
5424 m->bridge->name, m->name);
5429 shash_init(&src_ports);
5430 shash_init(&dst_ports);
5431 if (cfg->select_all) {
5432 for (i = 0; i < m->bridge->n_ports; i++) {
5433 const char *name = m->bridge->ports[i]->name;
5434 shash_add_once(&src_ports, name, NULL);
5435 shash_add_once(&dst_ports, name, NULL);
5440 /* Get ports, and drop duplicates and ports that don't exist. */
5441 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
5443 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
5446 /* Get all the vlans, and drop duplicate and invalid vlans. */
5447 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
5450 /* Update mirror data. */
5451 if (!shash_equal_keys(&m->src_ports, &src_ports)
5452 || !shash_equal_keys(&m->dst_ports, &dst_ports)
5453 || m->n_vlans != n_vlans
5454 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
5455 || m->out_port != out_port
5456 || m->out_vlan != out_vlan) {
5457 bridge_flush(m->bridge);
5459 shash_swap(&m->src_ports, &src_ports);
5460 shash_swap(&m->dst_ports, &dst_ports);
5463 m->n_vlans = n_vlans;
5464 m->out_port = out_port;
5465 m->out_vlan = out_vlan;
5468 mirror_bit = MIRROR_MASK_C(1) << m->idx;
5469 for (i = 0; i < m->bridge->n_ports; i++) {
5470 struct port *port = m->bridge->ports[i];
5472 if (shash_find(&m->src_ports, port->name)
5475 ? port_trunks_any_mirrored_vlan(m, port)
5476 : vlan_is_mirrored(m, port->vlan)))) {
5477 port->src_mirrors |= mirror_bit;
5479 port->src_mirrors &= ~mirror_bit;
5482 if (shash_find(&m->dst_ports, port->name)) {
5483 port->dst_mirrors |= mirror_bit;
5485 port->dst_mirrors &= ~mirror_bit;
5490 shash_destroy(&src_ports);
5491 shash_destroy(&dst_ports);