1 /* Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
18 #include "byte-order.h"
21 #include <arpa/inet.h>
24 #include <sys/socket.h>
26 #include <openflow/openflow.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
36 #include "classifier.h"
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_get_all_ifaces(const struct bridge *, struct shash *ifaces);
272 static void bridge_fetch_dp_ifaces(struct bridge *);
273 static void bridge_flush(struct bridge *);
274 static void bridge_pick_local_hw_addr(struct bridge *,
275 uint8_t ea[ETH_ADDR_LEN],
276 struct iface **hw_addr_iface);
277 static uint64_t bridge_pick_datapath_id(struct bridge *,
278 const uint8_t bridge_ea[ETH_ADDR_LEN],
279 struct iface *hw_addr_iface);
280 static struct iface *bridge_get_local_iface(struct bridge *);
281 static uint64_t dpid_from_hash(const void *, size_t nbytes);
283 static unixctl_cb_func bridge_unixctl_fdb_show;
284 static unixctl_cb_func qos_unixctl_show;
286 static void lacp_run(struct port *);
287 static void lacp_wait(struct port *);
288 static void lacp_process_packet(const struct ofpbuf *, struct iface *);
290 static void bond_init(void);
291 static void bond_run(struct port *);
292 static void bond_wait(struct port *);
293 static void bond_rebalance_port(struct port *);
294 static void bond_send_learning_packets(struct port *);
295 static void bond_enable_slave(struct iface *iface, bool enable);
297 static void port_run(struct port *);
298 static void port_wait(struct port *);
299 static struct port *port_create(struct bridge *, const char *name);
300 static void port_reconfigure(struct port *, const struct ovsrec_port *);
301 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
302 static void port_destroy(struct port *);
303 static struct port *port_lookup(const struct bridge *, const char *name);
304 static struct iface *port_lookup_iface(const struct port *, const char *name);
305 static struct port *port_from_dp_ifidx(const struct bridge *,
307 static void port_update_bonding(struct port *);
308 static void port_update_lacp(struct port *);
310 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
311 static void mirror_destroy(struct mirror *);
312 static void mirror_reconfigure(struct bridge *);
313 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
314 static bool vlan_is_mirrored(const struct mirror *, int vlan);
316 static struct iface *iface_create(struct port *port,
317 const struct ovsrec_interface *if_cfg);
318 static void iface_destroy(struct iface *);
319 static struct iface *iface_lookup(const struct bridge *, const char *name);
320 static struct iface *iface_find(const char *name);
321 static struct iface *iface_from_dp_ifidx(const struct bridge *,
323 static void iface_set_mac(struct iface *);
324 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
325 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
326 static void iface_update_cfm(struct iface *);
327 static void iface_refresh_cfm_stats(struct iface *iface);
328 static void iface_send_packet(struct iface *, struct ofpbuf *packet);
329 static uint8_t iface_get_lacp_state(const struct iface *);
330 static void iface_get_lacp_priority(struct iface *, struct lacp_info *);
331 static void iface_set_lacp_defaulted(struct iface *);
332 static void iface_set_lacp_expired(struct iface *);
334 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
336 static void shash_to_ovs_idl_map(struct shash *,
337 char ***keys, char ***values, size_t *n);
340 /* Hooks into ofproto processing. */
341 static struct ofhooks bridge_ofhooks;
343 /* Public functions. */
345 /* Initializes the bridge module, configuring it to obtain its configuration
346 * from an OVSDB server accessed over 'remote', which should be a string in a
347 * form acceptable to ovsdb_idl_create(). */
349 bridge_init(const char *remote)
351 /* Create connection to database. */
352 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
354 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
355 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
356 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
358 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
360 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
361 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
363 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
364 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
365 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
367 /* Register unixctl commands. */
368 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
369 unixctl_command_register("qos/show", qos_unixctl_show, NULL);
370 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
372 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
380 struct bridge *br, *next_br;
382 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
385 ovsdb_idl_destroy(idl);
388 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
389 * but for which the ovs-vswitchd configuration 'cfg' is required. */
391 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
393 static bool already_configured_once;
394 struct svec bridge_names;
395 struct svec dpif_names, dpif_types;
398 /* Only do this once per ovs-vswitchd run. */
399 if (already_configured_once) {
402 already_configured_once = true;
404 stats_timer = time_msec() + STATS_INTERVAL;
406 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
407 svec_init(&bridge_names);
408 for (i = 0; i < cfg->n_bridges; i++) {
409 svec_add(&bridge_names, cfg->bridges[i]->name);
411 svec_sort(&bridge_names);
413 /* Iterate over all system dpifs and delete any of them that do not appear
415 svec_init(&dpif_names);
416 svec_init(&dpif_types);
417 dp_enumerate_types(&dpif_types);
418 for (i = 0; i < dpif_types.n; i++) {
421 dp_enumerate_names(dpif_types.names[i], &dpif_names);
423 /* Delete each dpif whose name is not in 'bridge_names'. */
424 for (j = 0; j < dpif_names.n; j++) {
425 if (!svec_contains(&bridge_names, dpif_names.names[j])) {
429 retval = dpif_open(dpif_names.names[j], dpif_types.names[i],
438 svec_destroy(&bridge_names);
439 svec_destroy(&dpif_names);
440 svec_destroy(&dpif_types);
443 /* Callback for iterate_and_prune_ifaces(). */
445 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
447 if (!iface->netdev) {
448 /* We already reported a related error, don't bother duplicating it. */
452 if (iface->dp_ifidx < 0) {
453 VLOG_ERR("%s interface not in %s, dropping",
454 iface->name, dpif_name(br->dpif));
458 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
459 iface->name, iface->dp_ifidx);
463 /* Callback for iterate_and_prune_ifaces(). */
465 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
466 void *aux OVS_UNUSED)
468 /* Set policing attributes. */
469 netdev_set_policing(iface->netdev,
470 iface->cfg->ingress_policing_rate,
471 iface->cfg->ingress_policing_burst);
473 /* Set MAC address of internal interfaces other than the local
475 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
476 iface_set_mac(iface);
482 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
483 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
484 * deletes from 'br' any ports that no longer have any interfaces. */
486 iterate_and_prune_ifaces(struct bridge *br,
487 bool (*cb)(struct bridge *, struct iface *,
493 for (i = 0; i < br->n_ports; ) {
494 struct port *port = br->ports[i];
495 for (j = 0; j < port->n_ifaces; ) {
496 struct iface *iface = port->ifaces[j];
497 if (cb(br, iface, aux)) {
500 iface_set_ofport(iface->cfg, -1);
501 iface_destroy(iface);
505 if (port->n_ifaces) {
508 VLOG_WARN("%s port has no interfaces, dropping", port->name);
514 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
515 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
516 * responsible for freeing '*managersp' (with free()).
518 * You may be asking yourself "why does ovs-vswitchd care?", because
519 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
520 * should not be and in fact is not directly involved in that. But
521 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
522 * it has to tell in-band control where the managers are to enable that.
523 * (Thus, only managers connected in-band are collected.)
526 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
527 struct sockaddr_in **managersp, size_t *n_managersp)
529 struct sockaddr_in *managers = NULL;
530 size_t n_managers = 0;
531 struct shash targets;
534 /* Collect all of the potential targets from the "targets" columns of the
535 * rows pointed to by "manager_options", excluding any that are
537 shash_init(&targets);
538 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
539 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
541 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
542 shash_find_and_delete(&targets, m->target);
544 shash_add_once(&targets, m->target, NULL);
548 /* Now extract the targets' IP addresses. */
549 if (!shash_is_empty(&targets)) {
550 struct shash_node *node;
552 managers = xmalloc(shash_count(&targets) * sizeof *managers);
553 SHASH_FOR_EACH (node, &targets) {
554 const char *target = node->name;
555 struct sockaddr_in *sin = &managers[n_managers];
557 if ((!strncmp(target, "tcp:", 4)
558 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
559 (!strncmp(target, "ssl:", 4)
560 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
565 shash_destroy(&targets);
567 *managersp = managers;
568 *n_managersp = n_managers;
572 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
574 struct shash old_br, new_br;
575 struct shash_node *node;
576 struct bridge *br, *next;
577 struct sockaddr_in *managers;
580 int sflow_bridge_number;
582 COVERAGE_INC(bridge_reconfigure);
584 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
586 /* Collect old and new bridges. */
589 LIST_FOR_EACH (br, node, &all_bridges) {
590 shash_add(&old_br, br->name, br);
592 for (i = 0; i < ovs_cfg->n_bridges; i++) {
593 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
594 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
595 VLOG_WARN("more than one bridge named %s", br_cfg->name);
599 /* Get rid of deleted bridges and add new bridges. */
600 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
601 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
608 SHASH_FOR_EACH (node, &new_br) {
609 const char *br_name = node->name;
610 const struct ovsrec_bridge *br_cfg = node->data;
611 br = shash_find_data(&old_br, br_name);
613 /* If the bridge datapath type has changed, we need to tear it
614 * down and recreate. */
615 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
617 bridge_create(br_cfg);
620 bridge_create(br_cfg);
623 shash_destroy(&old_br);
624 shash_destroy(&new_br);
626 /* Reconfigure all bridges. */
627 LIST_FOR_EACH (br, node, &all_bridges) {
628 bridge_reconfigure_one(br);
631 /* Add and delete ports on all datapaths.
633 * The kernel will reject any attempt to add a given port to a datapath if
634 * that port already belongs to a different datapath, so we must do all
635 * port deletions before any port additions. */
636 LIST_FOR_EACH (br, node, &all_bridges) {
637 struct dpif_port_dump dump;
638 struct shash want_ifaces;
639 struct dpif_port dpif_port;
641 bridge_get_all_ifaces(br, &want_ifaces);
642 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
643 if (!shash_find(&want_ifaces, dpif_port.name)
644 && strcmp(dpif_port.name, br->name)) {
645 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
647 VLOG_WARN("failed to remove %s interface from %s: %s",
648 dpif_port.name, dpif_name(br->dpif),
653 shash_destroy(&want_ifaces);
655 LIST_FOR_EACH (br, node, &all_bridges) {
656 struct shash cur_ifaces, want_ifaces;
657 struct dpif_port_dump dump;
658 struct dpif_port dpif_port;
660 /* Get the set of interfaces currently in this datapath. */
661 shash_init(&cur_ifaces);
662 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
663 struct dpif_port *port_info = xmalloc(sizeof *port_info);
664 dpif_port_clone(port_info, &dpif_port);
665 shash_add(&cur_ifaces, dpif_port.name, port_info);
668 /* Get the set of interfaces we want on this datapath. */
669 bridge_get_all_ifaces(br, &want_ifaces);
671 hmap_clear(&br->ifaces);
672 SHASH_FOR_EACH (node, &want_ifaces) {
673 const char *if_name = node->name;
674 struct iface *iface = node->data;
675 struct dpif_port *dpif_port;
679 type = iface ? iface->type : "internal";
680 dpif_port = shash_find_data(&cur_ifaces, if_name);
682 /* If we have a port or a netdev already, and it's not the type we
683 * want, then delete the port (if any) and close the netdev (if
685 if ((dpif_port && strcmp(dpif_port->type, type))
686 || (iface && iface->netdev
687 && strcmp(type, netdev_get_type(iface->netdev)))) {
689 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
696 netdev_close(iface->netdev);
697 iface->netdev = NULL;
701 /* If the port doesn't exist or we don't have the netdev open,
702 * we need to do more work. */
703 if (!dpif_port || (iface && !iface->netdev)) {
704 struct netdev_options options;
705 struct netdev *netdev;
708 /* First open the network device. */
709 options.name = if_name;
711 options.args = &args;
712 options.ethertype = NETDEV_ETH_TYPE_NONE;
716 shash_from_ovs_idl_map(iface->cfg->key_options,
717 iface->cfg->value_options,
718 iface->cfg->n_options, &args);
720 error = netdev_open(&options, &netdev);
721 shash_destroy(&args);
724 VLOG_WARN("could not open network device %s (%s)",
725 if_name, strerror(error));
729 /* Then add the port if we haven't already. */
731 error = dpif_port_add(br->dpif, netdev, NULL);
733 netdev_close(netdev);
734 if (error == EFBIG) {
735 VLOG_ERR("ran out of valid port numbers on %s",
736 dpif_name(br->dpif));
739 VLOG_WARN("failed to add %s interface to %s: %s",
740 if_name, dpif_name(br->dpif),
747 /* Update 'iface'. */
749 iface->netdev = netdev;
750 iface->enabled = netdev_get_carrier(iface->netdev);
751 iface->up = iface->enabled;
753 } else if (iface && iface->netdev) {
757 shash_from_ovs_idl_map(iface->cfg->key_options,
758 iface->cfg->value_options,
759 iface->cfg->n_options, &args);
760 netdev_set_config(iface->netdev, &args);
761 shash_destroy(&args);
764 shash_destroy(&want_ifaces);
766 SHASH_FOR_EACH (node, &cur_ifaces) {
767 struct dpif_port *port_info = node->data;
768 dpif_port_destroy(port_info);
771 shash_destroy(&cur_ifaces);
773 sflow_bridge_number = 0;
774 LIST_FOR_EACH (br, node, &all_bridges) {
777 struct iface *local_iface;
778 struct iface *hw_addr_iface;
781 bridge_fetch_dp_ifaces(br);
783 iterate_and_prune_ifaces(br, check_iface, NULL);
785 /* Pick local port hardware address, datapath ID. */
786 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
787 local_iface = bridge_get_local_iface(br);
789 int error = netdev_set_etheraddr(local_iface->netdev, ea);
791 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
792 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
793 "Ethernet address: %s",
794 br->name, strerror(error));
797 memcpy(br->ea, ea, ETH_ADDR_LEN);
799 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
800 ofproto_set_datapath_id(br->ofproto, dpid);
802 dpid_string = xasprintf("%016"PRIx64, dpid);
803 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
806 /* Set NetFlow configuration on this bridge. */
807 if (br->cfg->netflow) {
808 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
809 struct netflow_options opts;
811 memset(&opts, 0, sizeof opts);
813 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
814 if (nf_cfg->engine_type) {
815 opts.engine_type = *nf_cfg->engine_type;
817 if (nf_cfg->engine_id) {
818 opts.engine_id = *nf_cfg->engine_id;
821 opts.active_timeout = nf_cfg->active_timeout;
822 if (!opts.active_timeout) {
823 opts.active_timeout = -1;
824 } else if (opts.active_timeout < 0) {
825 VLOG_WARN("bridge %s: active timeout interval set to negative "
826 "value, using default instead (%d seconds)", br->name,
827 NF_ACTIVE_TIMEOUT_DEFAULT);
828 opts.active_timeout = -1;
831 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
832 if (opts.add_id_to_iface) {
833 if (opts.engine_id > 0x7f) {
834 VLOG_WARN("bridge %s: netflow port mangling may conflict "
835 "with another vswitch, choose an engine id less "
836 "than 128", br->name);
838 if (br->n_ports > 508) {
839 VLOG_WARN("bridge %s: netflow port mangling will conflict "
840 "with another port when more than 508 ports are "
845 opts.collectors.n = nf_cfg->n_targets;
846 opts.collectors.names = nf_cfg->targets;
847 if (ofproto_set_netflow(br->ofproto, &opts)) {
848 VLOG_ERR("bridge %s: problem setting netflow collectors",
852 ofproto_set_netflow(br->ofproto, NULL);
855 /* Set sFlow configuration on this bridge. */
856 if (br->cfg->sflow) {
857 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
858 struct ovsrec_controller **controllers;
859 struct ofproto_sflow_options oso;
860 size_t n_controllers;
862 memset(&oso, 0, sizeof oso);
864 oso.targets.n = sflow_cfg->n_targets;
865 oso.targets.names = sflow_cfg->targets;
867 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
868 if (sflow_cfg->sampling) {
869 oso.sampling_rate = *sflow_cfg->sampling;
872 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
873 if (sflow_cfg->polling) {
874 oso.polling_interval = *sflow_cfg->polling;
877 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
878 if (sflow_cfg->header) {
879 oso.header_len = *sflow_cfg->header;
882 oso.sub_id = sflow_bridge_number++;
883 oso.agent_device = sflow_cfg->agent;
885 oso.control_ip = NULL;
886 n_controllers = bridge_get_controllers(br, &controllers);
887 for (i = 0; i < n_controllers; i++) {
888 if (controllers[i]->local_ip) {
889 oso.control_ip = controllers[i]->local_ip;
893 ofproto_set_sflow(br->ofproto, &oso);
895 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
897 ofproto_set_sflow(br->ofproto, NULL);
900 /* Update the controller and related settings. It would be more
901 * straightforward to call this from bridge_reconfigure_one(), but we
902 * can't do it there for two reasons. First, and most importantly, at
903 * that point we don't know the dp_ifidx of any interfaces that have
904 * been added to the bridge (because we haven't actually added them to
905 * the datapath). Second, at that point we haven't set the datapath ID
906 * yet; when a controller is configured, resetting the datapath ID will
907 * immediately disconnect from the controller, so it's better to set
908 * the datapath ID before the controller. */
909 bridge_reconfigure_remotes(br, managers, n_managers);
911 LIST_FOR_EACH (br, node, &all_bridges) {
912 for (i = 0; i < br->n_ports; i++) {
913 struct port *port = br->ports[i];
916 port_update_bonding(port);
917 port_update_lacp(port);
919 for (j = 0; j < port->n_ifaces; j++) {
920 iface_update_qos(port->ifaces[j], port->cfg->qos);
924 LIST_FOR_EACH (br, node, &all_bridges) {
925 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
928 LIST_FOR_EACH (br, node, &all_bridges) {
930 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
931 iface_update_cfm(iface);
937 /* ovs-vswitchd has completed initialization, so allow the process that
938 * forked us to exit successfully. */
939 daemonize_complete();
943 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
944 const struct ovsdb_idl_column *column,
947 const struct ovsdb_datum *datum;
948 union ovsdb_atom atom;
951 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
952 atom.string = (char *) key;
953 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
954 return idx == UINT_MAX ? NULL : datum->values[idx].string;
958 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
960 return get_ovsrec_key_value(&br_cfg->header_,
961 &ovsrec_bridge_col_other_config, key);
965 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
966 struct iface **hw_addr_iface)
972 *hw_addr_iface = NULL;
974 /* Did the user request a particular MAC? */
975 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
976 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
977 if (eth_addr_is_multicast(ea)) {
978 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
979 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
980 } else if (eth_addr_is_zero(ea)) {
981 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
987 /* Otherwise choose the minimum non-local MAC address among all of the
989 memset(ea, 0xff, ETH_ADDR_LEN);
990 for (i = 0; i < br->n_ports; i++) {
991 struct port *port = br->ports[i];
992 uint8_t iface_ea[ETH_ADDR_LEN];
995 /* Mirror output ports don't participate. */
996 if (port->is_mirror_output_port) {
1000 /* Choose the MAC address to represent the port. */
1001 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
1002 /* Find the interface with this Ethernet address (if any) so that
1003 * we can provide the correct devname to the caller. */
1005 for (j = 0; j < port->n_ifaces; j++) {
1006 struct iface *candidate = port->ifaces[j];
1007 uint8_t candidate_ea[ETH_ADDR_LEN];
1008 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
1009 && eth_addr_equals(iface_ea, candidate_ea)) {
1014 /* Choose the interface whose MAC address will represent the port.
1015 * The Linux kernel bonding code always chooses the MAC address of
1016 * the first slave added to a bond, and the Fedora networking
1017 * scripts always add slaves to a bond in alphabetical order, so
1018 * for compatibility we choose the interface with the name that is
1019 * first in alphabetical order. */
1020 iface = port->ifaces[0];
1021 for (j = 1; j < port->n_ifaces; j++) {
1022 struct iface *candidate = port->ifaces[j];
1023 if (strcmp(candidate->name, iface->name) < 0) {
1028 /* The local port doesn't count (since we're trying to choose its
1029 * MAC address anyway). */
1030 if (iface->dp_ifidx == ODPP_LOCAL) {
1035 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1037 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1038 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1039 iface->name, strerror(error));
1044 /* Compare against our current choice. */
1045 if (!eth_addr_is_multicast(iface_ea) &&
1046 !eth_addr_is_local(iface_ea) &&
1047 !eth_addr_is_reserved(iface_ea) &&
1048 !eth_addr_is_zero(iface_ea) &&
1049 eth_addr_compare_3way(iface_ea, ea) < 0)
1051 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1052 *hw_addr_iface = iface;
1055 if (eth_addr_is_multicast(ea)) {
1056 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1057 *hw_addr_iface = NULL;
1058 VLOG_WARN("bridge %s: using default bridge Ethernet "
1059 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1061 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1062 br->name, ETH_ADDR_ARGS(ea));
1066 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1067 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1068 * an interface on 'br', then that interface must be passed in as
1069 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1070 * 'hw_addr_iface' must be passed in as a null pointer. */
1072 bridge_pick_datapath_id(struct bridge *br,
1073 const uint8_t bridge_ea[ETH_ADDR_LEN],
1074 struct iface *hw_addr_iface)
1077 * The procedure for choosing a bridge MAC address will, in the most
1078 * ordinary case, also choose a unique MAC that we can use as a datapath
1079 * ID. In some special cases, though, multiple bridges will end up with
1080 * the same MAC address. This is OK for the bridges, but it will confuse
1081 * the OpenFlow controller, because each datapath needs a unique datapath
1084 * Datapath IDs must be unique. It is also very desirable that they be
1085 * stable from one run to the next, so that policy set on a datapath
1088 const char *datapath_id;
1091 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1092 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1096 if (hw_addr_iface) {
1098 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1100 * A bridge whose MAC address is taken from a VLAN network device
1101 * (that is, a network device created with vconfig(8) or similar
1102 * tool) will have the same MAC address as a bridge on the VLAN
1103 * device's physical network device.
1105 * Handle this case by hashing the physical network device MAC
1106 * along with the VLAN identifier.
1108 uint8_t buf[ETH_ADDR_LEN + 2];
1109 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1110 buf[ETH_ADDR_LEN] = vlan >> 8;
1111 buf[ETH_ADDR_LEN + 1] = vlan;
1112 return dpid_from_hash(buf, sizeof buf);
1115 * Assume that this bridge's MAC address is unique, since it
1116 * doesn't fit any of the cases we handle specially.
1121 * A purely internal bridge, that is, one that has no non-virtual
1122 * network devices on it at all, is more difficult because it has no
1123 * natural unique identifier at all.
1125 * When the host is a XenServer, we handle this case by hashing the
1126 * host's UUID with the name of the bridge. Names of bridges are
1127 * persistent across XenServer reboots, although they can be reused if
1128 * an internal network is destroyed and then a new one is later
1129 * created, so this is fairly effective.
1131 * When the host is not a XenServer, we punt by using a random MAC
1132 * address on each run.
1134 const char *host_uuid = xenserver_get_host_uuid();
1136 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1137 dpid = dpid_from_hash(combined, strlen(combined));
1143 return eth_addr_to_uint64(bridge_ea);
1147 dpid_from_hash(const void *data, size_t n)
1149 uint8_t hash[SHA1_DIGEST_SIZE];
1151 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1152 sha1_bytes(data, n, hash);
1153 eth_addr_mark_random(hash);
1154 return eth_addr_to_uint64(hash);
1158 iface_refresh_status(struct iface *iface)
1162 enum netdev_flags flags;
1171 if (!netdev_get_status(iface->netdev, &sh)) {
1173 char **keys, **values;
1175 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1176 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1181 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1184 shash_destroy_free_data(&sh);
1186 error = netdev_get_flags(iface->netdev, &flags);
1188 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1191 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1194 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1196 ovsrec_interface_set_duplex(iface->cfg,
1197 netdev_features_is_full_duplex(current)
1199 /* warning: uint64_t -> int64_t conversion */
1200 bps = netdev_features_to_bps(current);
1201 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1204 ovsrec_interface_set_duplex(iface->cfg, NULL);
1205 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1209 ovsrec_interface_set_link_state(iface->cfg,
1210 netdev_get_carrier(iface->netdev)
1213 error = netdev_get_mtu(iface->netdev, &mtu);
1214 if (!error && mtu != INT_MAX) {
1216 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1219 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1224 iface_refresh_cfm_stats(struct iface *iface)
1228 const struct ovsrec_monitor *mon;
1230 mon = iface->cfg->monitor;
1237 for (i = 0; i < mon->n_remote_mps; i++) {
1238 const struct ovsrec_maintenance_point *mp;
1239 const struct remote_mp *rmp;
1241 mp = mon->remote_mps[i];
1242 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1244 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1247 if (hmap_is_empty(&cfm->x_remote_mps)) {
1248 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1251 struct remote_mp *rmp;
1252 int64_t *x_remote_mps;
1254 length = hmap_count(&cfm->x_remote_mps);
1255 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1258 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1259 x_remote_mps[i++] = rmp->mpid;
1262 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1266 if (hmap_is_empty(&cfm->x_remote_maids)) {
1267 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1270 char **x_remote_maids;
1271 struct remote_maid *rmaid;
1273 length = hmap_count(&cfm->x_remote_maids);
1274 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1277 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1280 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1282 for (j = 0; j < CCM_MAID_LEN; j++) {
1283 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1288 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1290 for (i = 0; i < length; i++) {
1291 free(x_remote_maids[i]);
1293 free(x_remote_maids);
1296 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1300 iface_refresh_stats(struct iface *iface)
1306 static const struct iface_stat iface_stats[] = {
1307 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1308 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1309 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1310 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1311 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1312 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1313 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1314 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1315 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1316 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1317 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1318 { "collisions", offsetof(struct netdev_stats, collisions) },
1320 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1321 const struct iface_stat *s;
1323 char *keys[N_STATS];
1324 int64_t values[N_STATS];
1327 struct netdev_stats stats;
1329 /* Intentionally ignore return value, since errors will set 'stats' to
1330 * all-1s, and we will deal with that correctly below. */
1331 netdev_get_stats(iface->netdev, &stats);
1334 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1335 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1336 if (value != UINT64_MAX) {
1343 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1347 enable_system_stats(const struct ovsrec_open_vswitch *cfg)
1351 /* Use other-config:enable-system-stats by preference. */
1352 enable = get_ovsrec_key_value(&cfg->header_,
1353 &ovsrec_open_vswitch_col_other_config,
1354 "enable-statistics");
1356 return !strcmp(enable, "true");
1359 /* Disable by default. */
1364 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1366 struct ovsdb_datum datum;
1370 if (enable_system_stats(cfg)) {
1371 get_system_stats(&stats);
1374 ovsdb_datum_from_shash(&datum, &stats);
1375 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1379 static inline const char *
1380 nx_role_to_str(enum nx_role role)
1385 case NX_ROLE_MASTER:
1390 return "*** INVALID ROLE ***";
1395 bridge_refresh_controller_status(const struct bridge *br)
1398 const struct ovsrec_controller *cfg;
1400 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1402 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1403 struct ofproto_controller_info *cinfo =
1404 shash_find_data(&info, cfg->target);
1407 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1408 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1409 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1410 (char **) cinfo->pairs.values,
1413 ovsrec_controller_set_is_connected(cfg, false);
1414 ovsrec_controller_set_role(cfg, NULL);
1415 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1419 ofproto_free_ofproto_controller_info(&info);
1425 const struct ovsrec_open_vswitch *cfg;
1427 bool datapath_destroyed;
1428 bool database_changed;
1431 /* Let each bridge do the work that it needs to do. */
1432 datapath_destroyed = false;
1433 LIST_FOR_EACH (br, node, &all_bridges) {
1434 int error = bridge_run_one(br);
1436 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1437 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1438 "forcing reconfiguration", br->name);
1439 datapath_destroyed = true;
1443 /* (Re)configure if necessary. */
1444 database_changed = ovsdb_idl_run(idl);
1445 cfg = ovsrec_open_vswitch_first(idl);
1447 /* Re-configure SSL. We do this on every trip through the main loop,
1448 * instead of just when the database changes, because the contents of the
1449 * key and certificate files can change without the database changing.
1451 * We do this before bridge_reconfigure() because that function might
1452 * initiate SSL connections and thus requires SSL to be configured. */
1453 if (cfg && cfg->ssl) {
1454 const struct ovsrec_ssl *ssl = cfg->ssl;
1456 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1457 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1460 if (database_changed || datapath_destroyed) {
1462 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1464 bridge_configure_once(cfg);
1465 bridge_reconfigure(cfg);
1467 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1468 ovsdb_idl_txn_commit(txn);
1469 ovsdb_idl_txn_destroy(txn); /* XXX */
1471 /* We still need to reconfigure to avoid dangling pointers to
1472 * now-destroyed ovsrec structures inside bridge data. */
1473 static const struct ovsrec_open_vswitch null_cfg;
1475 bridge_reconfigure(&null_cfg);
1479 /* Refresh system and interface stats if necessary. */
1480 if (time_msec() >= stats_timer) {
1482 struct ovsdb_idl_txn *txn;
1484 txn = ovsdb_idl_txn_create(idl);
1485 LIST_FOR_EACH (br, node, &all_bridges) {
1488 for (i = 0; i < br->n_ports; i++) {
1489 struct port *port = br->ports[i];
1492 for (j = 0; j < port->n_ifaces; j++) {
1493 struct iface *iface = port->ifaces[j];
1494 iface_refresh_stats(iface);
1495 iface_refresh_cfm_stats(iface);
1496 iface_refresh_status(iface);
1499 bridge_refresh_controller_status(br);
1501 refresh_system_stats(cfg);
1502 ovsdb_idl_txn_commit(txn);
1503 ovsdb_idl_txn_destroy(txn); /* XXX */
1506 stats_timer = time_msec() + STATS_INTERVAL;
1515 LIST_FOR_EACH (br, node, &all_bridges) {
1518 ofproto_wait(br->ofproto);
1519 if (ofproto_has_primary_controller(br->ofproto)) {
1523 mac_learning_wait(br->ml);
1525 for (i = 0; i < br->n_ports; i++) {
1526 port_wait(br->ports[i]);
1529 ovsdb_idl_wait(idl);
1530 poll_timer_wait_until(stats_timer);
1533 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1534 * configuration changes. */
1536 bridge_flush(struct bridge *br)
1538 COVERAGE_INC(bridge_flush);
1540 mac_learning_flush(br->ml);
1543 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1544 * such interface. */
1545 static struct iface *
1546 bridge_get_local_iface(struct bridge *br)
1550 for (i = 0; i < br->n_ports; i++) {
1551 struct port *port = br->ports[i];
1552 for (j = 0; j < port->n_ifaces; j++) {
1553 struct iface *iface = port->ifaces[j];
1554 if (iface->dp_ifidx == ODPP_LOCAL) {
1563 /* Bridge unixctl user interface functions. */
1565 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1566 const char *args, void *aux OVS_UNUSED)
1568 struct ds ds = DS_EMPTY_INITIALIZER;
1569 const struct bridge *br;
1570 const struct mac_entry *e;
1572 br = bridge_lookup(args);
1574 unixctl_command_reply(conn, 501, "no such bridge");
1578 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1579 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1580 if (e->port < 0 || e->port >= br->n_ports) {
1583 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1584 br->ports[e->port]->ifaces[0]->dp_ifidx,
1585 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1587 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1591 /* QoS unixctl user interface functions. */
1593 struct qos_unixctl_show_cbdata {
1595 struct iface *iface;
1599 qos_unixctl_show_cb(unsigned int queue_id,
1600 const struct shash *details,
1603 struct qos_unixctl_show_cbdata *data = aux;
1604 struct ds *ds = data->ds;
1605 struct iface *iface = data->iface;
1606 struct netdev_queue_stats stats;
1607 struct shash_node *node;
1610 ds_put_cstr(ds, "\n");
1612 ds_put_format(ds, "Queue %u:\n", queue_id);
1614 ds_put_cstr(ds, "Default:\n");
1617 SHASH_FOR_EACH (node, details) {
1618 ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data);
1621 error = netdev_get_queue_stats(iface->netdev, queue_id, &stats);
1623 if (stats.tx_packets != UINT64_MAX) {
1624 ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets);
1627 if (stats.tx_bytes != UINT64_MAX) {
1628 ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes);
1631 if (stats.tx_errors != UINT64_MAX) {
1632 ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors);
1635 ds_put_format(ds, "\tFailed to get statistics for queue %u: %s",
1636 queue_id, strerror(error));
1641 qos_unixctl_show(struct unixctl_conn *conn,
1642 const char *args, void *aux OVS_UNUSED)
1644 struct ds ds = DS_EMPTY_INITIALIZER;
1645 struct shash sh = SHASH_INITIALIZER(&sh);
1646 struct iface *iface;
1648 struct shash_node *node;
1649 struct qos_unixctl_show_cbdata data;
1652 iface = iface_find(args);
1654 unixctl_command_reply(conn, 501, "no such interface");
1658 netdev_get_qos(iface->netdev, &type, &sh);
1660 if (*type != '\0') {
1661 ds_put_format(&ds, "QoS: %s %s\n", iface->name, type);
1663 SHASH_FOR_EACH (node, &sh) {
1664 ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data);
1669 error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data);
1672 ds_put_format(&ds, "failed to dump queues: %s", strerror(error));
1674 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1676 ds_put_format(&ds, "QoS not configured on %s\n", iface->name);
1677 unixctl_command_reply(conn, 501, ds_cstr(&ds));
1680 shash_destroy_free_data(&sh);
1684 /* Bridge reconfiguration functions. */
1685 static struct bridge *
1686 bridge_create(const struct ovsrec_bridge *br_cfg)
1691 assert(!bridge_lookup(br_cfg->name));
1692 br = xzalloc(sizeof *br);
1694 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1700 dpif_flow_flush(br->dpif);
1702 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1705 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1707 dpif_delete(br->dpif);
1708 dpif_close(br->dpif);
1713 br->name = xstrdup(br_cfg->name);
1715 br->ml = mac_learning_create();
1716 eth_addr_nicira_random(br->default_ea);
1718 hmap_init(&br->ifaces);
1720 shash_init(&br->port_by_name);
1721 shash_init(&br->iface_by_name);
1725 list_push_back(&all_bridges, &br->node);
1727 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1733 bridge_destroy(struct bridge *br)
1738 while (br->n_ports > 0) {
1739 port_destroy(br->ports[br->n_ports - 1]);
1741 list_remove(&br->node);
1742 error = dpif_delete(br->dpif);
1743 if (error && error != ENOENT) {
1744 VLOG_ERR("failed to delete %s: %s",
1745 dpif_name(br->dpif), strerror(error));
1747 dpif_close(br->dpif);
1748 ofproto_destroy(br->ofproto);
1749 mac_learning_destroy(br->ml);
1750 hmap_destroy(&br->ifaces);
1751 shash_destroy(&br->port_by_name);
1752 shash_destroy(&br->iface_by_name);
1759 static struct bridge *
1760 bridge_lookup(const char *name)
1764 LIST_FOR_EACH (br, node, &all_bridges) {
1765 if (!strcmp(br->name, name)) {
1772 /* Handle requests for a listing of all flows known by the OpenFlow
1773 * stack, including those normally hidden. */
1775 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1776 const char *args, void *aux OVS_UNUSED)
1781 br = bridge_lookup(args);
1783 unixctl_command_reply(conn, 501, "Unknown bridge");
1788 ofproto_get_all_flows(br->ofproto, &results);
1790 unixctl_command_reply(conn, 200, ds_cstr(&results));
1791 ds_destroy(&results);
1794 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1795 * connections and reconnect. If BRIDGE is not specified, then all bridges
1796 * drop their controller connections and reconnect. */
1798 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1799 const char *args, void *aux OVS_UNUSED)
1802 if (args[0] != '\0') {
1803 br = bridge_lookup(args);
1805 unixctl_command_reply(conn, 501, "Unknown bridge");
1808 ofproto_reconnect_controllers(br->ofproto);
1810 LIST_FOR_EACH (br, node, &all_bridges) {
1811 ofproto_reconnect_controllers(br->ofproto);
1814 unixctl_command_reply(conn, 200, NULL);
1818 bridge_run_one(struct bridge *br)
1823 error = ofproto_run1(br->ofproto);
1828 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1830 for (i = 0; i < br->n_ports; i++) {
1831 port_run(br->ports[i]);
1834 error = ofproto_run2(br->ofproto, br->flush);
1841 bridge_get_controllers(const struct bridge *br,
1842 struct ovsrec_controller ***controllersp)
1844 struct ovsrec_controller **controllers;
1845 size_t n_controllers;
1847 controllers = br->cfg->controller;
1848 n_controllers = br->cfg->n_controller;
1850 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1856 *controllersp = controllers;
1858 return n_controllers;
1862 bridge_reconfigure_one(struct bridge *br)
1864 struct shash old_ports, new_ports;
1865 struct svec snoops, old_snoops;
1866 struct shash_node *node;
1867 enum ofproto_fail_mode fail_mode;
1870 /* Collect old ports. */
1871 shash_init(&old_ports);
1872 for (i = 0; i < br->n_ports; i++) {
1873 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1876 /* Collect new ports. */
1877 shash_init(&new_ports);
1878 for (i = 0; i < br->cfg->n_ports; i++) {
1879 const char *name = br->cfg->ports[i]->name;
1880 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1881 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1886 /* If we have a controller, then we need a local port. Complain if the
1887 * user didn't specify one.
1889 * XXX perhaps we should synthesize a port ourselves in this case. */
1890 if (bridge_get_controllers(br, NULL)) {
1891 char local_name[IF_NAMESIZE];
1894 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1895 local_name, sizeof local_name);
1896 if (!error && !shash_find(&new_ports, local_name)) {
1897 VLOG_WARN("bridge %s: controller specified but no local port "
1898 "(port named %s) defined",
1899 br->name, local_name);
1903 /* Get rid of deleted ports.
1904 * Get rid of deleted interfaces on ports that still exist. */
1905 SHASH_FOR_EACH (node, &old_ports) {
1906 struct port *port = node->data;
1907 const struct ovsrec_port *port_cfg;
1909 port_cfg = shash_find_data(&new_ports, node->name);
1913 port_del_ifaces(port, port_cfg);
1917 /* Create new ports.
1918 * Add new interfaces to existing ports.
1919 * Reconfigure existing ports. */
1920 SHASH_FOR_EACH (node, &new_ports) {
1921 struct port *port = shash_find_data(&old_ports, node->name);
1923 port = port_create(br, node->name);
1926 port_reconfigure(port, node->data);
1927 if (!port->n_ifaces) {
1928 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1929 br->name, port->name);
1933 shash_destroy(&old_ports);
1934 shash_destroy(&new_ports);
1936 /* Set the fail-mode */
1937 fail_mode = !br->cfg->fail_mode
1938 || !strcmp(br->cfg->fail_mode, "standalone")
1939 ? OFPROTO_FAIL_STANDALONE
1940 : OFPROTO_FAIL_SECURE;
1941 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1942 && !ofproto_has_primary_controller(br->ofproto)) {
1943 ofproto_flush_flows(br->ofproto);
1945 ofproto_set_fail_mode(br->ofproto, fail_mode);
1947 /* Delete all flows if we're switching from connected to standalone or vice
1948 * versa. (XXX Should we delete all flows if we are switching from one
1949 * controller to another?) */
1951 /* Configure OpenFlow controller connection snooping. */
1953 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1954 ovs_rundir(), br->name));
1955 svec_init(&old_snoops);
1956 ofproto_get_snoops(br->ofproto, &old_snoops);
1957 if (!svec_equal(&snoops, &old_snoops)) {
1958 ofproto_set_snoops(br->ofproto, &snoops);
1960 svec_destroy(&snoops);
1961 svec_destroy(&old_snoops);
1963 mirror_reconfigure(br);
1966 /* Initializes 'oc' appropriately as a management service controller for
1969 * The caller must free oc->target when it is no longer needed. */
1971 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1972 struct ofproto_controller *oc)
1974 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1975 oc->max_backoff = 0;
1976 oc->probe_interval = 60;
1977 oc->band = OFPROTO_OUT_OF_BAND;
1979 oc->burst_limit = 0;
1982 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1984 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1985 struct ofproto_controller *oc)
1987 oc->target = c->target;
1988 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1989 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1990 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1991 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1992 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1993 oc->burst_limit = (c->controller_burst_limit
1994 ? *c->controller_burst_limit : 0);
1997 /* Configures the IP stack for 'br''s local interface properly according to the
1998 * configuration in 'c'. */
2000 bridge_configure_local_iface_netdev(struct bridge *br,
2001 struct ovsrec_controller *c)
2003 struct netdev *netdev;
2004 struct in_addr mask, gateway;
2006 struct iface *local_iface;
2009 /* If there's no local interface or no IP address, give up. */
2010 local_iface = bridge_get_local_iface(br);
2011 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
2015 /* Bring up the local interface. */
2016 netdev = local_iface->netdev;
2017 netdev_turn_flags_on(netdev, NETDEV_UP, true);
2019 /* Configure the IP address and netmask. */
2020 if (!c->local_netmask
2021 || !inet_aton(c->local_netmask, &mask)
2023 mask.s_addr = guess_netmask(ip.s_addr);
2025 if (!netdev_set_in4(netdev, ip, mask)) {
2026 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
2027 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
2030 /* Configure the default gateway. */
2031 if (c->local_gateway
2032 && inet_aton(c->local_gateway, &gateway)
2033 && gateway.s_addr) {
2034 if (!netdev_add_router(netdev, gateway)) {
2035 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
2036 br->name, IP_ARGS(&gateway.s_addr));
2042 bridge_reconfigure_remotes(struct bridge *br,
2043 const struct sockaddr_in *managers,
2046 const char *disable_ib_str, *queue_id_str;
2047 bool disable_in_band = false;
2050 struct ovsrec_controller **controllers;
2051 size_t n_controllers;
2054 struct ofproto_controller *ocs;
2058 /* Check if we should disable in-band control on this bridge. */
2059 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
2060 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
2061 disable_in_band = true;
2064 /* Set OpenFlow queue ID for in-band control. */
2065 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
2066 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
2067 ofproto_set_in_band_queue(br->ofproto, queue_id);
2069 if (disable_in_band) {
2070 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
2072 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
2074 had_primary = ofproto_has_primary_controller(br->ofproto);
2076 n_controllers = bridge_get_controllers(br, &controllers);
2078 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
2081 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
2082 for (i = 0; i < n_controllers; i++) {
2083 struct ovsrec_controller *c = controllers[i];
2085 if (!strncmp(c->target, "punix:", 6)
2086 || !strncmp(c->target, "unix:", 5)) {
2087 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2089 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
2090 * domain sockets and overwriting arbitrary local files. */
2091 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2092 "\"%s\" due to possibility for remote exploit",
2093 dpif_name(br->dpif), c->target);
2097 bridge_configure_local_iface_netdev(br, c);
2098 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2099 if (disable_in_band) {
2100 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2105 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2106 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2109 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2110 ofproto_flush_flows(br->ofproto);
2113 /* If there are no controllers and the bridge is in standalone
2114 * mode, set up a flow that matches every packet and directs
2115 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2116 * switch is in secure mode and we won't pass any traffic until
2117 * a controller has been defined and it tells us to do so. */
2119 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2120 union ofp_action action;
2121 struct cls_rule rule;
2123 memset(&action, 0, sizeof action);
2124 action.type = htons(OFPAT_OUTPUT);
2125 action.output.len = htons(sizeof action);
2126 action.output.port = htons(OFPP_NORMAL);
2127 cls_rule_init_catchall(&rule, 0);
2128 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2133 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2138 for (i = 0; i < br->n_ports; i++) {
2139 struct port *port = br->ports[i];
2140 for (j = 0; j < port->n_ifaces; j++) {
2141 struct iface *iface = port->ifaces[j];
2142 shash_add_once(ifaces, iface->name, iface);
2144 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2145 shash_add_once(ifaces, port->name, NULL);
2150 /* For robustness, in case the administrator moves around datapath ports behind
2151 * our back, we re-check all the datapath port numbers here.
2153 * This function will set the 'dp_ifidx' members of interfaces that have
2154 * disappeared to -1, so only call this function from a context where those
2155 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2156 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2157 * datapath, which doesn't support UINT16_MAX+1 ports. */
2159 bridge_fetch_dp_ifaces(struct bridge *br)
2161 struct dpif_port_dump dump;
2162 struct dpif_port dpif_port;
2165 /* Reset all interface numbers. */
2166 for (i = 0; i < br->n_ports; i++) {
2167 struct port *port = br->ports[i];
2168 for (j = 0; j < port->n_ifaces; j++) {
2169 struct iface *iface = port->ifaces[j];
2170 iface->dp_ifidx = -1;
2173 hmap_clear(&br->ifaces);
2175 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2176 struct iface *iface = iface_lookup(br, dpif_port.name);
2178 if (iface->dp_ifidx >= 0) {
2179 VLOG_WARN("%s reported interface %s twice",
2180 dpif_name(br->dpif), dpif_port.name);
2181 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2182 VLOG_WARN("%s reported interface %"PRIu16" twice",
2183 dpif_name(br->dpif), dpif_port.port_no);
2185 iface->dp_ifidx = dpif_port.port_no;
2186 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2187 hash_int(iface->dp_ifidx, 0));
2190 iface_set_ofport(iface->cfg,
2191 (iface->dp_ifidx >= 0
2192 ? odp_port_to_ofp_port(iface->dp_ifidx)
2198 /* Bridge packet processing functions. */
2201 bond_is_tcp_hash(const struct port *port)
2203 return port->bond_mode == BM_TCP && port->lacp & LACP_NEGOTIATED;
2207 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2209 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2212 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2214 struct flow hash_flow;
2216 memcpy(&hash_flow, flow, sizeof hash_flow);
2217 hash_flow.vlan_tci = 0;
2219 /* The symmetric quality of this hash function is not required, but
2220 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2221 * purposes, so we use it out of convenience. */
2222 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2225 static struct bond_entry *
2226 lookup_bond_entry(const struct port *port, const struct flow *flow,
2229 assert(port->bond_mode != BM_AB);
2231 if (bond_is_tcp_hash(port)) {
2232 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2234 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2239 bond_choose_iface(const struct port *port)
2241 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2242 size_t i, best_down_slave = -1;
2243 long long next_delay_expiration = LLONG_MAX;
2245 for (i = 0; i < port->n_ifaces; i++) {
2246 struct iface *iface = port->ifaces[i];
2248 if (iface->enabled) {
2250 } else if (iface->delay_expires < next_delay_expiration
2251 && (iface->lacp_status & LACP_ATTACHED
2252 || !(port->lacp & LACP_NEGOTIATED))) {
2253 best_down_slave = i;
2254 next_delay_expiration = iface->delay_expires;
2258 if (best_down_slave != -1) {
2259 struct iface *iface = port->ifaces[best_down_slave];
2261 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2262 "since no other interface is up", iface->name,
2263 iface->delay_expires - time_msec());
2264 bond_enable_slave(iface, true);
2267 return best_down_slave;
2271 choose_output_iface(const struct port *port, const struct flow *flow,
2272 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2274 struct iface *iface;
2276 assert(port->n_ifaces);
2277 if (port->n_ifaces == 1) {
2278 iface = port->ifaces[0];
2279 } else if (port->bond_mode == BM_AB) {
2280 if (port->active_iface < 0) {
2281 *tags |= port->no_ifaces_tag;
2284 iface = port->ifaces[port->active_iface];
2286 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2287 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2288 || !port->ifaces[e->iface_idx]->enabled) {
2289 /* XXX select interface properly. The current interface selection
2290 * is only good for testing the rebalancing code. */
2291 e->iface_idx = bond_choose_iface(port);
2292 if (e->iface_idx < 0) {
2293 *tags |= port->no_ifaces_tag;
2296 e->iface_tag = tag_create_random();
2298 *tags |= e->iface_tag;
2299 iface = port->ifaces[e->iface_idx];
2301 *dp_ifidx = iface->dp_ifidx;
2302 *tags |= iface->tag; /* Currently only used for bonding. */
2307 bond_link_status_update(struct iface *iface)
2309 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2310 struct port *port = iface->port;
2311 bool up = iface->up;
2312 int updelay, downdelay;
2314 updelay = port->updelay;
2315 downdelay = port->downdelay;
2317 if (iface->port->lacp & LACP_NEGOTIATED) {
2322 if (iface->port->lacp && up) {
2323 /* The interface is up if it's attached to an aggregator and its
2324 * partner is synchronized. The only exception is defaulted links.
2325 * They are not required to have synchronized partners because they
2326 * have no partners at all. However, they will only be attached if
2327 * negotiations failed on all interfaces in the bond. */
2328 up = iface->lacp_status & LACP_ATTACHED
2329 && (iface->lacp_partner.state & LACP_STATE_SYNC
2330 || iface->lacp_status & LACP_DEFAULTED);
2334 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2335 /* Nothing to do. */
2338 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2339 iface->name, up ? "up" : "down");
2340 if (up == iface->enabled) {
2341 iface->delay_expires = LLONG_MAX;
2342 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2343 iface->name, up ? "disabled" : "enabled");
2344 } else if (up && port->active_iface < 0) {
2345 bond_enable_slave(iface, true);
2347 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2348 "other interface is up", iface->name, updelay);
2351 int delay = up ? updelay : downdelay;
2352 iface->delay_expires = time_msec() + delay;
2355 "interface %s: will be %s if it stays %s for %d ms",
2357 up ? "enabled" : "disabled",
2365 bond_choose_active_iface(struct port *port)
2367 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2369 port->active_iface = bond_choose_iface(port);
2370 port->active_iface_tag = tag_create_random();
2371 if (port->active_iface >= 0) {
2372 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2373 port->name, port->ifaces[port->active_iface]->name);
2375 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2381 bond_enable_slave(struct iface *iface, bool enable)
2383 struct port *port = iface->port;
2384 struct bridge *br = port->bridge;
2386 /* This acts as a recursion check. If the act of disabling a slave
2387 * causes a different slave to be enabled, the flag will allow us to
2388 * skip redundant work when we reenter this function. It must be
2389 * cleared on exit to keep things safe with multiple bonds. */
2390 static bool moving_active_iface = false;
2392 iface->delay_expires = LLONG_MAX;
2393 if (enable == iface->enabled) {
2397 iface->enabled = enable;
2398 if (!iface->enabled) {
2399 VLOG_WARN("interface %s: disabled", iface->name);
2400 ofproto_revalidate(br->ofproto, iface->tag);
2401 if (iface->port_ifidx == port->active_iface) {
2402 ofproto_revalidate(br->ofproto,
2403 port->active_iface_tag);
2405 /* Disabling a slave can lead to another slave being immediately
2406 * enabled if there will be no active slaves but one is waiting
2407 * on an updelay. In this case we do not need to run most of the
2408 * code for the newly enabled slave since there was no period
2409 * without an active slave and it is redundant with the disabling
2411 moving_active_iface = true;
2412 bond_choose_active_iface(port);
2414 bond_send_learning_packets(port);
2416 VLOG_WARN("interface %s: enabled", iface->name);
2417 if (port->active_iface < 0 && !moving_active_iface) {
2418 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2419 bond_choose_active_iface(port);
2420 bond_send_learning_packets(port);
2422 iface->tag = tag_create_random();
2425 moving_active_iface = false;
2428 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2429 * bond interface. */
2431 bond_update_fake_iface_stats(struct port *port)
2433 struct netdev_stats bond_stats;
2434 struct netdev *bond_dev;
2437 memset(&bond_stats, 0, sizeof bond_stats);
2439 for (i = 0; i < port->n_ifaces; i++) {
2440 struct netdev_stats slave_stats;
2442 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2443 /* XXX: We swap the stats here because they are swapped back when
2444 * reported by the internal device. The reason for this is
2445 * internal devices normally represent packets going into the system
2446 * but when used as fake bond device they represent packets leaving
2447 * the system. We really should do this in the internal device
2448 * itself because changing it here reverses the counts from the
2449 * perspective of the switch. However, the internal device doesn't
2450 * know what type of device it represents so we have to do it here
2452 bond_stats.tx_packets += slave_stats.rx_packets;
2453 bond_stats.tx_bytes += slave_stats.rx_bytes;
2454 bond_stats.rx_packets += slave_stats.tx_packets;
2455 bond_stats.rx_bytes += slave_stats.tx_bytes;
2459 if (!netdev_open_default(port->name, &bond_dev)) {
2460 netdev_set_stats(bond_dev, &bond_stats);
2461 netdev_close(bond_dev);
2466 bond_link_carrier_update(struct iface *iface, bool carrier)
2468 if (carrier == iface->up) {
2472 if (iface->lacp_status & LACP_CURRENT) {
2473 iface_set_lacp_expired(iface);
2476 iface->up = carrier;
2481 bond_run(struct port *port)
2486 if (port->n_ifaces < 2) {
2490 if (port->monitor) {
2491 assert(!port->miimon);
2493 /* Track carrier going up and down on interfaces. */
2494 while (!netdev_monitor_poll(port->monitor, &devname)) {
2495 struct iface *iface;
2497 iface = port_lookup_iface(port, devname);
2499 bool up = netdev_get_carrier(iface->netdev);
2500 bond_link_carrier_update(iface, up);
2505 assert(port->miimon);
2507 if (time_msec() >= port->bond_miimon_next_update) {
2508 for (i = 0; i < port->n_ifaces; i++) {
2509 struct iface *iface = port->ifaces[i];
2510 bool up = netdev_get_miimon(iface->netdev);
2511 bond_link_carrier_update(iface, up);
2513 port->bond_miimon_next_update = time_msec() +
2514 port->bond_miimon_interval;
2518 for (i = 0; i < port->n_ifaces; i++) {
2519 bond_link_status_update(port->ifaces[i]);
2522 for (i = 0; i < port->n_ifaces; i++) {
2523 struct iface *iface = port->ifaces[i];
2524 if (time_msec() >= iface->delay_expires) {
2525 bond_enable_slave(iface, !iface->enabled);
2529 if (port->bond_fake_iface
2530 && time_msec() >= port->bond_next_fake_iface_update) {
2531 bond_update_fake_iface_stats(port);
2532 port->bond_next_fake_iface_update = time_msec() + 1000;
2537 bond_wait(struct port *port)
2541 if (port->n_ifaces < 2) {
2545 if (port->monitor) {
2546 netdev_monitor_poll_wait(port->monitor);
2550 poll_timer_wait_until(port->bond_miimon_next_update);
2553 for (i = 0; i < port->n_ifaces; i++) {
2554 struct iface *iface = port->ifaces[i];
2555 if (iface->delay_expires != LLONG_MAX) {
2556 poll_timer_wait_until(iface->delay_expires);
2560 if (port->bond_fake_iface) {
2561 poll_timer_wait_until(port->bond_next_fake_iface_update);
2566 set_dst(struct dst *dst, const struct flow *flow,
2567 const struct port *in_port, const struct port *out_port,
2570 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2571 : in_port->vlan >= 0 ? in_port->vlan
2572 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2573 : vlan_tci_to_vid(flow->vlan_tci));
2574 return choose_output_iface(out_port, flow, dst->vlan,
2575 &dst->dp_ifidx, tags);
2579 swap_dst(struct dst *p, struct dst *q)
2581 struct dst tmp = *p;
2586 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2587 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2588 * that we push to the datapath. We could in fact fully sort the array by
2589 * vlan, but in most cases there are at most two different vlan tags so that's
2590 * possibly overkill.) */
2592 partition_dsts(struct dst_set *set, int vlan)
2594 struct dst *first = set->dsts;
2595 struct dst *last = set->dsts + set->n;
2597 while (first != last) {
2599 * - All dsts < first have vlan == 'vlan'.
2600 * - All dsts >= last have vlan != 'vlan'.
2601 * - first < last. */
2602 while (first->vlan == vlan) {
2603 if (++first == last) {
2608 /* Same invariants, plus one additional:
2609 * - first->vlan != vlan.
2611 while (last[-1].vlan != vlan) {
2612 if (--last == first) {
2617 /* Same invariants, plus one additional:
2618 * - last[-1].vlan == vlan.*/
2619 swap_dst(first++, --last);
2624 mirror_mask_ffs(mirror_mask_t mask)
2626 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2631 dst_set_init(struct dst_set *set)
2633 set->dsts = set->builtin;
2635 set->allocated = ARRAY_SIZE(set->builtin);
2639 dst_set_add(struct dst_set *set, const struct dst *dst)
2641 if (set->n >= set->allocated) {
2642 size_t new_allocated;
2643 struct dst *new_dsts;
2645 new_allocated = set->allocated * 2;
2646 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2647 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2651 set->dsts = new_dsts;
2652 set->allocated = new_allocated;
2654 set->dsts[set->n++] = *dst;
2658 dst_set_free(struct dst_set *set)
2660 if (set->dsts != set->builtin) {
2666 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2669 for (i = 0; i < set->n; i++) {
2670 if (set->dsts[i].vlan == test->vlan
2671 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2679 port_trunks_vlan(const struct port *port, uint16_t vlan)
2681 return (port->vlan < 0
2682 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2686 port_includes_vlan(const struct port *port, uint16_t vlan)
2688 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2692 port_is_floodable(const struct port *port)
2696 for (i = 0; i < port->n_ifaces; i++) {
2697 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2698 port->ifaces[i]->dp_ifidx)) {
2706 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2707 const struct port *in_port, const struct port *out_port,
2708 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2710 mirror_mask_t mirrors = in_port->src_mirrors;
2715 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2716 if (flow_vlan == 0) {
2717 flow_vlan = OFP_VLAN_NONE;
2720 if (out_port == FLOOD_PORT) {
2721 for (i = 0; i < br->n_ports; i++) {
2722 struct port *port = br->ports[i];
2724 && port_is_floodable(port)
2725 && port_includes_vlan(port, vlan)
2726 && !port->is_mirror_output_port
2727 && set_dst(&dst, flow, in_port, port, tags)) {
2728 mirrors |= port->dst_mirrors;
2729 dst_set_add(set, &dst);
2732 *nf_output_iface = NF_OUT_FLOOD;
2733 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2734 dst_set_add(set, &dst);
2735 *nf_output_iface = dst.dp_ifidx;
2736 mirrors |= out_port->dst_mirrors;
2740 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2741 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2743 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2744 && !dst_is_duplicate(set, &dst)) {
2745 dst_set_add(set, &dst);
2748 for (i = 0; i < br->n_ports; i++) {
2749 struct port *port = br->ports[i];
2750 if (port_includes_vlan(port, m->out_vlan)
2751 && set_dst(&dst, flow, in_port, port, tags))
2753 if (port->vlan < 0) {
2754 dst.vlan = m->out_vlan;
2756 if (dst_is_duplicate(set, &dst)) {
2760 /* Use the vlan tag on the original flow instead of
2761 * the one passed in the vlan parameter. This ensures
2762 * that we compare the vlan from before any implicit
2763 * tagging tags place. This is necessary because
2764 * dst->vlan is the final vlan, after removing implicit
2766 if (port == in_port && dst.vlan == flow_vlan) {
2767 /* Don't send out input port on same VLAN. */
2770 dst_set_add(set, &dst);
2775 mirrors &= mirrors - 1;
2778 partition_dsts(set, flow_vlan);
2781 static void OVS_UNUSED
2782 print_dsts(const struct dst_set *set)
2786 for (i = 0; i < set->n; i++) {
2787 const struct dst *dst = &set->dsts[i];
2789 printf(">p%"PRIu16, dst->dp_ifidx);
2790 if (dst->vlan != OFP_VLAN_NONE) {
2791 printf("v%"PRIu16, dst->vlan);
2797 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2798 const struct port *in_port, const struct port *out_port,
2799 tag_type *tags, struct ofpbuf *actions,
2800 uint16_t *nf_output_iface)
2807 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2810 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2811 if (cur_vlan == 0) {
2812 cur_vlan = OFP_VLAN_NONE;
2814 for (i = 0; i < set.n; i++) {
2815 const struct dst *dst = &set.dsts[i];
2816 if (dst->vlan != cur_vlan) {
2817 if (dst->vlan == OFP_VLAN_NONE) {
2818 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2821 tci = htons(dst->vlan & VLAN_VID_MASK);
2822 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2823 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2825 cur_vlan = dst->vlan;
2827 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2832 /* Returns the effective vlan of a packet, taking into account both the
2833 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2834 * the packet is untagged and -1 indicates it has an invalid header and
2835 * should be dropped. */
2836 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2837 struct port *in_port, bool have_packet)
2839 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2840 if (in_port->vlan >= 0) {
2842 /* XXX support double tagging? */
2844 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2845 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2846 "packet received on port %s configured with "
2847 "implicit VLAN %"PRIu16,
2848 br->name, vlan, in_port->name, in_port->vlan);
2852 vlan = in_port->vlan;
2854 if (!port_includes_vlan(in_port, vlan)) {
2856 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2857 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2858 "packet received on port %s not configured for "
2860 br->name, vlan, in_port->name, vlan);
2869 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2870 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2871 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2873 is_gratuitous_arp(const struct flow *flow)
2875 return (flow->dl_type == htons(ETH_TYPE_ARP)
2876 && eth_addr_is_broadcast(flow->dl_dst)
2877 && (flow->nw_proto == ARP_OP_REPLY
2878 || (flow->nw_proto == ARP_OP_REQUEST
2879 && flow->nw_src == flow->nw_dst)));
2883 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2884 struct port *in_port)
2886 enum grat_arp_lock_type lock_type;
2889 /* We don't want to learn from gratuitous ARP packets that are reflected
2890 * back over bond slaves so we lock the learning table. */
2891 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2892 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2893 GRAT_ARP_LOCK_CHECK;
2895 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2898 /* The log messages here could actually be useful in debugging,
2899 * so keep the rate limit relatively high. */
2900 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2902 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2903 "on port %s in VLAN %d",
2904 br->name, ETH_ADDR_ARGS(flow->dl_src),
2905 in_port->name, vlan);
2906 ofproto_revalidate(br->ofproto, rev_tag);
2910 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2911 * dropped. Returns true if they may be forwarded, false if they should be
2914 * If 'have_packet' is true, it indicates that the caller is processing a
2915 * received packet. If 'have_packet' is false, then the caller is just
2916 * revalidating an existing flow because configuration has changed. Either
2917 * way, 'have_packet' only affects logging (there is no point in logging errors
2918 * during revalidation).
2920 * Sets '*in_portp' to the input port. This will be a null pointer if
2921 * flow->in_port does not designate a known input port (in which case
2922 * is_admissible() returns false).
2924 * When returning true, sets '*vlanp' to the effective VLAN of the input
2925 * packet, as returned by flow_get_vlan().
2927 * May also add tags to '*tags', although the current implementation only does
2928 * so in one special case.
2931 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2932 tag_type *tags, int *vlanp, struct port **in_portp)
2934 struct iface *in_iface;
2935 struct port *in_port;
2938 /* Find the interface and port structure for the received packet. */
2939 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2941 /* No interface? Something fishy... */
2943 /* Odd. A few possible reasons here:
2945 * - We deleted an interface but there are still a few packets
2946 * queued up from it.
2948 * - Someone externally added an interface (e.g. with "ovs-dpctl
2949 * add-if") that we don't know about.
2951 * - Packet arrived on the local port but the local port is not
2952 * one of our bridge ports.
2954 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2956 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2957 "interface %"PRIu16, br->name, flow->in_port);
2963 *in_portp = in_port = in_iface->port;
2964 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2969 /* Drop frames for reserved multicast addresses. */
2970 if (eth_addr_is_reserved(flow->dl_dst)) {
2974 /* Drop frames on ports reserved for mirroring. */
2975 if (in_port->is_mirror_output_port) {
2977 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2978 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2979 "%s, which is reserved exclusively for mirroring",
2980 br->name, in_port->name);
2985 /* When using LACP, do not accept packets from disabled interfaces. */
2986 if (in_port->lacp & LACP_NEGOTIATED && !in_iface->enabled) {
2990 /* Packets received on non-LACP bonds need special attention to avoid
2992 if (in_port->n_ifaces > 1 && !(in_port->lacp & LACP_NEGOTIATED)) {
2994 bool is_grat_arp_locked;
2996 if (eth_addr_is_multicast(flow->dl_dst)) {
2997 *tags |= in_port->active_iface_tag;
2998 if (in_port->active_iface != in_iface->port_ifidx) {
2999 /* Drop all multicast packets on inactive slaves. */
3004 /* Drop all packets for which we have learned a different input
3005 * port, because we probably sent the packet on one slave and got
3006 * it back on the other. Gratuitous ARP packets are an exception
3007 * to this rule: the host has moved to another switch. The exception
3008 * to the exception is if we locked the learning table to avoid
3009 * reflections on bond slaves. If this is the case, just drop the
3011 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
3012 &is_grat_arp_locked);
3013 if (src_idx != -1 && src_idx != in_port->port_idx &&
3014 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
3019 /* Drop all packets which arrive on backup slaves. This is similar to how
3020 * Linux bonding handles active-backup bonds. */
3021 if (in_port->bond_mode == BM_AB) {
3023 *tags |= in_port->active_iface;
3024 if (in_port->active_iface != in_iface->port_ifidx) {
3025 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3027 VLOG_WARN_RL(&rl, "active-backup bond received packet on backup"
3028 " interface (%s) destined for " ETH_ADDR_FMT,
3029 in_iface->name, ETH_ADDR_ARGS(flow->dl_dst));
3037 /* If the composed actions may be applied to any packet in the given 'flow',
3038 * returns true. Otherwise, the actions should only be applied to 'packet', or
3039 * not at all, if 'packet' was NULL. */
3041 process_flow(struct bridge *br, const struct flow *flow,
3042 const struct ofpbuf *packet, struct ofpbuf *actions,
3043 tag_type *tags, uint16_t *nf_output_iface)
3045 struct port *in_port;
3046 struct port *out_port;
3050 /* Check whether we should drop packets in this flow. */
3051 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
3056 /* Learn source MAC (but don't try to learn from revalidation). */
3058 update_learning_table(br, flow, vlan, in_port);
3061 /* Determine output port. */
3062 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
3064 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
3065 out_port = br->ports[out_port_idx];
3066 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
3067 /* If we are revalidating but don't have a learning entry then
3068 * eject the flow. Installing a flow that floods packets opens
3069 * up a window of time where we could learn from a packet reflected
3070 * on a bond and blackhole packets before the learning table is
3071 * updated to reflect the correct port. */
3074 out_port = FLOOD_PORT;
3077 /* Don't send packets out their input ports. */
3078 if (in_port == out_port) {
3084 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
3092 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
3093 struct ofpbuf *actions, tag_type *tags,
3094 uint16_t *nf_output_iface, void *br_)
3096 struct bridge *br = br_;
3098 COVERAGE_INC(bridge_process_flow);
3099 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
3103 bridge_special_ofhook_cb(const struct flow *flow,
3104 const struct ofpbuf *packet, void *br_)
3106 struct iface *iface;
3107 struct bridge *br = br_;
3109 iface = iface_from_dp_ifidx(br, flow->in_port);
3111 if (cfm_should_process_flow(flow)) {
3113 if (iface && packet && iface->cfm) {
3114 COVERAGE_INC(bridge_process_cfm);
3115 cfm_process_heartbeat(iface->cfm, packet);
3118 } else if (flow->dl_type == htons(ETH_TYPE_LACP)) {
3120 if (iface && packet) {
3121 COVERAGE_INC(bridge_process_lacp);
3122 lacp_process_packet(packet, iface);
3131 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
3132 const struct nlattr *actions,
3134 uint64_t n_bytes, void *br_)
3136 struct bridge *br = br_;
3137 const struct nlattr *a;
3138 struct port *in_port;
3143 /* Feed information from the active flows back into the learning table to
3144 * ensure that table is always in sync with what is actually flowing
3145 * through the datapath.
3147 * We test that 'tags' is nonzero to ensure that only flows that include an
3148 * OFPP_NORMAL action are used for learning. This works because
3149 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3150 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3151 update_learning_table(br, flow, vlan, in_port);
3154 /* Account for bond slave utilization. */
3155 if (!br->has_bonded_ports) {
3158 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3159 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3160 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3161 if (out_port && out_port->n_ifaces >= 2 &&
3162 out_port->bond_mode != BM_AB) {
3163 uint16_t vlan = (flow->vlan_tci
3164 ? vlan_tci_to_vid(flow->vlan_tci)
3166 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
3167 e->tx_bytes += n_bytes;
3174 bridge_account_checkpoint_ofhook_cb(void *br_)
3176 struct bridge *br = br_;
3180 if (!br->has_bonded_ports) {
3185 for (i = 0; i < br->n_ports; i++) {
3186 struct port *port = br->ports[i];
3187 if (port->n_ifaces > 1 && port->bond_mode != BM_AB
3188 && now >= port->bond_next_rebalance) {
3189 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3190 bond_rebalance_port(port);
3195 static struct ofhooks bridge_ofhooks = {
3196 bridge_normal_ofhook_cb,
3197 bridge_special_ofhook_cb,
3198 bridge_account_flow_ofhook_cb,
3199 bridge_account_checkpoint_ofhook_cb,
3202 /* LACP functions. */
3205 lacp_process_packet(const struct ofpbuf *packet, struct iface *iface)
3207 const struct lacp_pdu *pdu;
3209 if (!iface->port->lacp) {
3213 pdu = parse_lacp_packet(packet);
3218 iface->lacp_status |= LACP_CURRENT;
3219 iface->lacp_status &= ~(LACP_EXPIRED | LACP_DEFAULTED);
3220 iface->lacp_rx = time_msec() + LACP_SLOW_TIME_RX;
3222 iface->lacp_actor.state = iface_get_lacp_state(iface);
3223 if (memcmp(&iface->lacp_actor, &pdu->partner, sizeof pdu->partner)) {
3227 if (memcmp(&iface->lacp_partner, &pdu->actor, sizeof pdu->actor)) {
3228 iface->port->lacp_need_update = true;
3229 iface->lacp_partner = pdu->actor;
3234 lacp_update_ifaces(struct port *port)
3238 struct lacp_info lead_pri;
3239 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3241 port->lacp_need_update = false;
3242 COVERAGE_INC(bridge_lacp_update);
3248 VLOG_DBG_RL(&rl, "port %s: re-evaluating LACP link status", port->name);
3251 for (i = 0; i < port->n_ifaces; i++) {
3252 struct iface *iface = port->ifaces[i];
3253 struct lacp_info pri;
3255 iface->lacp_status |= LACP_ATTACHED;
3256 ofproto_revalidate(port->bridge->ofproto, iface->tag);
3258 /* Don't allow loopback interfaces to send traffic or lead. */
3259 if (eth_addr_equals(iface->lacp_partner.sysid,
3260 iface->lacp_actor.sysid)) {
3261 VLOG_WARN_RL(&rl, "iface %s: Loopback detected. Interface is "
3262 "connected to its own bridge", iface->name);
3263 iface->lacp_status &= ~LACP_ATTACHED;
3267 if (iface->lacp_status & LACP_DEFAULTED) {
3271 iface_get_lacp_priority(iface, &pri);
3273 if (!lead || memcmp(&pri, &lead_pri, sizeof pri) < 0) {
3280 port->lacp &= ~LACP_NEGOTIATED;
3284 port->lacp |= LACP_NEGOTIATED;
3286 for (i = 0; i < port->n_ifaces; i++) {
3287 struct iface *iface = port->ifaces[i];
3289 if (iface->lacp_status & LACP_DEFAULTED
3290 || lead->lacp_partner.key != iface->lacp_partner.key
3291 || !eth_addr_equals(lead->lacp_partner.sysid,
3292 iface->lacp_partner.sysid)) {
3293 iface->lacp_status &= ~LACP_ATTACHED;
3299 lacp_iface_may_tx(const struct iface *iface)
3301 return iface->port->lacp & LACP_ACTIVE
3302 || iface->lacp_status & (LACP_CURRENT | LACP_EXPIRED);
3306 lacp_run(struct port *port)
3309 struct ofpbuf packet;
3315 ofpbuf_init(&packet, ETH_HEADER_LEN + LACP_PDU_LEN);
3317 for (i = 0; i < port->n_ifaces; i++) {
3318 struct iface *iface = port->ifaces[i];
3320 if (time_msec() > iface->lacp_rx) {
3321 if (iface->lacp_status & LACP_CURRENT) {
3322 iface_set_lacp_expired(iface);
3323 } else if (iface->lacp_status & LACP_EXPIRED) {
3324 iface_set_lacp_defaulted(iface);
3329 if (port->lacp_need_update) {
3330 lacp_update_ifaces(port);
3333 for (i = 0; i < port->n_ifaces; i++) {
3334 struct iface *iface = port->ifaces[i];
3335 uint8_t ea[ETH_ADDR_LEN];
3338 if (time_msec() < iface->lacp_tx || !lacp_iface_may_tx(iface)) {
3342 error = netdev_get_etheraddr(iface->netdev, ea);
3344 iface->lacp_actor.state = iface_get_lacp_state(iface);
3345 compose_lacp_packet(&packet, &iface->lacp_actor,
3346 &iface->lacp_partner, ea);
3347 iface_send_packet(iface, &packet);
3349 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3350 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3351 "(%s)", iface->name, strerror(error));
3354 iface->lacp_tx = time_msec() +
3355 (iface->lacp_partner.state & LACP_STATE_TIME
3357 : LACP_SLOW_TIME_TX);
3359 ofpbuf_uninit(&packet);
3363 lacp_wait(struct port *port)
3371 for (i = 0; i < port->n_ifaces; i++) {
3372 struct iface *iface = port->ifaces[i];
3374 if (lacp_iface_may_tx(iface)) {
3375 poll_timer_wait_until(iface->lacp_tx);
3378 if (iface->lacp_status & (LACP_CURRENT | LACP_EXPIRED)) {
3379 poll_timer_wait_until(iface->lacp_rx);
3384 /* Bonding functions. */
3386 /* Statistics for a single interface on a bonded port, used for load-based
3387 * bond rebalancing. */
3388 struct slave_balance {
3389 struct iface *iface; /* The interface. */
3390 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3392 /* All the "bond_entry"s that are assigned to this interface, in order of
3393 * increasing tx_bytes. */
3394 struct bond_entry **hashes;
3399 bond_mode_to_string(enum bond_mode bm) {
3400 static char *bm_slb = "balance-slb";
3401 static char *bm_ab = "active-backup";
3402 static char *bm_tcp = "balance-tcp";
3405 case BM_SLB: return bm_slb;
3406 case BM_AB: return bm_ab;
3407 case BM_TCP: return bm_tcp;
3414 /* Sorts pointers to pointers to bond_entries in ascending order by the
3415 * interface to which they are assigned, and within a single interface in
3416 * ascending order of bytes transmitted. */
3418 compare_bond_entries(const void *a_, const void *b_)
3420 const struct bond_entry *const *ap = a_;
3421 const struct bond_entry *const *bp = b_;
3422 const struct bond_entry *a = *ap;
3423 const struct bond_entry *b = *bp;
3424 if (a->iface_idx != b->iface_idx) {
3425 return a->iface_idx > b->iface_idx ? 1 : -1;
3426 } else if (a->tx_bytes != b->tx_bytes) {
3427 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3433 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3434 * *descending* order by number of bytes transmitted. */
3436 compare_slave_balance(const void *a_, const void *b_)
3438 const struct slave_balance *a = a_;
3439 const struct slave_balance *b = b_;
3440 if (a->iface->enabled != b->iface->enabled) {
3441 return a->iface->enabled ? -1 : 1;
3442 } else if (a->tx_bytes != b->tx_bytes) {
3443 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3450 swap_bals(struct slave_balance *a, struct slave_balance *b)
3452 struct slave_balance tmp = *a;
3457 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3458 * given that 'p' (and only 'p') might be in the wrong location.
3460 * This function invalidates 'p', since it might now be in a different memory
3463 resort_bals(struct slave_balance *p,
3464 struct slave_balance bals[], size_t n_bals)
3467 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3468 swap_bals(p, p - 1);
3470 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3471 swap_bals(p, p + 1);
3477 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3479 if (VLOG_IS_DBG_ENABLED()) {
3480 struct ds ds = DS_EMPTY_INITIALIZER;
3481 const struct slave_balance *b;
3483 for (b = bals; b < bals + n_bals; b++) {
3487 ds_put_char(&ds, ',');
3489 ds_put_format(&ds, " %s %"PRIu64"kB",
3490 b->iface->name, b->tx_bytes / 1024);
3492 if (!b->iface->enabled) {
3493 ds_put_cstr(&ds, " (disabled)");
3495 if (b->n_hashes > 0) {
3496 ds_put_cstr(&ds, " (");
3497 for (i = 0; i < b->n_hashes; i++) {
3498 const struct bond_entry *e = b->hashes[i];
3500 ds_put_cstr(&ds, " + ");
3502 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3503 e - port->bond_hash, e->tx_bytes / 1024);
3505 ds_put_cstr(&ds, ")");
3508 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3513 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3515 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3518 struct bond_entry *hash = from->hashes[hash_idx];
3519 struct port *port = from->iface->port;
3520 uint64_t delta = hash->tx_bytes;
3522 assert(port->bond_mode != BM_AB);
3524 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3525 "from %s to %s (now carrying %"PRIu64"kB and "
3526 "%"PRIu64"kB load, respectively)",
3527 port->name, delta / 1024, hash - port->bond_hash,
3528 from->iface->name, to->iface->name,
3529 (from->tx_bytes - delta) / 1024,
3530 (to->tx_bytes + delta) / 1024);
3532 /* Delete element from from->hashes.
3534 * We don't bother to add the element to to->hashes because not only would
3535 * it require more work, the only purpose it would be to allow that hash to
3536 * be migrated to another slave in this rebalancing run, and there is no
3537 * point in doing that. */
3538 if (hash_idx == 0) {
3541 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3542 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3546 /* Shift load away from 'from' to 'to'. */
3547 from->tx_bytes -= delta;
3548 to->tx_bytes += delta;
3550 /* Arrange for flows to be revalidated. */
3551 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
3552 hash->iface_idx = to->iface->port_ifidx;
3553 hash->iface_tag = tag_create_random();
3557 bond_rebalance_port(struct port *port)
3559 struct slave_balance *bals;
3561 struct bond_entry *hashes[BOND_MASK + 1];
3562 struct slave_balance *b, *from, *to;
3563 struct bond_entry *e;
3566 assert(port->bond_mode != BM_AB);
3568 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3569 * descending order of tx_bytes, so that bals[0] represents the most
3570 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3573 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3574 * array for each slave_balance structure, we sort our local array of
3575 * hashes in order by slave, so that all of the hashes for a given slave
3576 * become contiguous in memory, and then we point each 'hashes' members of
3577 * a slave_balance structure to the start of a contiguous group. */
3578 n_bals = port->n_ifaces;
3579 bals = xmalloc(n_bals * sizeof *bals);
3580 for (b = bals; b < &bals[n_bals]; b++) {
3581 b->iface = port->ifaces[b - bals];
3586 for (i = 0; i <= BOND_MASK; i++) {
3587 hashes[i] = &port->bond_hash[i];
3589 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3590 for (i = 0; i <= BOND_MASK; i++) {
3592 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3593 b = &bals[e->iface_idx];
3594 b->tx_bytes += e->tx_bytes;
3596 b->hashes = &hashes[i];
3601 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3602 log_bals(bals, n_bals, port);
3604 /* Discard slaves that aren't enabled (which were sorted to the back of the
3605 * array earlier). */
3606 while (!bals[n_bals - 1].iface->enabled) {
3613 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3614 to = &bals[n_bals - 1];
3615 for (from = bals; from < to; ) {
3616 uint64_t overload = from->tx_bytes - to->tx_bytes;
3617 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3618 /* The extra load on 'from' (and all less-loaded slaves), compared
3619 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3620 * it is less than ~1Mbps. No point in rebalancing. */
3622 } else if (from->n_hashes == 1) {
3623 /* 'from' only carries a single MAC hash, so we can't shift any
3624 * load away from it, even though we want to. */
3627 /* 'from' is carrying significantly more load than 'to', and that
3628 * load is split across at least two different hashes. Pick a hash
3629 * to migrate to 'to' (the least-loaded slave), given that doing so
3630 * must decrease the ratio of the load on the two slaves by at
3633 * The sort order we use means that we prefer to shift away the
3634 * smallest hashes instead of the biggest ones. There is little
3635 * reason behind this decision; we could use the opposite sort
3636 * order to shift away big hashes ahead of small ones. */
3639 for (i = 0; i < from->n_hashes; i++) {
3640 double old_ratio, new_ratio;
3641 uint64_t delta = from->hashes[i]->tx_bytes;
3643 if (delta == 0 || from->tx_bytes - delta == 0) {
3644 /* Pointless move. */
3648 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3650 if (to->tx_bytes == 0) {
3651 /* Nothing on the new slave, move it. */
3655 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3656 new_ratio = (double)(from->tx_bytes - delta) /
3657 (to->tx_bytes + delta);
3659 if (new_ratio == 0) {
3660 /* Should already be covered but check to prevent division
3665 if (new_ratio < 1) {
3666 new_ratio = 1 / new_ratio;
3669 if (old_ratio - new_ratio > 0.1) {
3670 /* Would decrease the ratio, move it. */
3674 if (i < from->n_hashes) {
3675 bond_shift_load(from, to, i);
3677 /* If the result of the migration changed the relative order of
3678 * 'from' and 'to' swap them back to maintain invariants. */
3679 if (order_swapped) {
3680 swap_bals(from, to);
3683 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3684 * point to different slave_balance structures. It is only
3685 * valid to do these two operations in a row at all because we
3686 * know that 'from' will not move past 'to' and vice versa. */
3687 resort_bals(from, bals, n_bals);
3688 resort_bals(to, bals, n_bals);
3695 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3696 * historical data to decay to <1% in 7 rebalancing runs. */
3697 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3706 bond_send_learning_packets(struct port *port)
3708 struct bridge *br = port->bridge;
3709 struct mac_entry *e;
3710 struct ofpbuf packet;
3711 int error, n_packets, n_errors;
3713 if (!port->n_ifaces || port->active_iface < 0 || bond_is_tcp_hash(port)) {
3717 ofpbuf_init(&packet, 128);
3718 error = n_packets = n_errors = 0;
3719 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3720 union ofp_action actions[2], *a;
3726 if (e->port == port->port_idx) {
3730 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3732 flow_extract(&packet, 0, ODPP_NONE, &flow);
3734 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3738 /* Compose actions. */
3739 memset(actions, 0, sizeof actions);
3742 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3743 a->vlan_vid.len = htons(sizeof *a);
3744 a->vlan_vid.vlan_vid = htons(e->vlan);
3747 a->output.type = htons(OFPAT_OUTPUT);
3748 a->output.len = htons(sizeof *a);
3749 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3754 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3761 ofpbuf_uninit(&packet);
3764 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3765 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3766 "packets, last error was: %s",
3767 port->name, n_errors, n_packets, strerror(error));
3769 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3770 port->name, n_packets);
3774 /* Bonding unixctl user interface functions. */
3777 bond_unixctl_list(struct unixctl_conn *conn,
3778 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3780 struct ds ds = DS_EMPTY_INITIALIZER;
3781 const struct bridge *br;
3783 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3785 LIST_FOR_EACH (br, node, &all_bridges) {
3788 for (i = 0; i < br->n_ports; i++) {
3789 const struct port *port = br->ports[i];
3790 if (port->n_ifaces > 1) {
3793 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3794 bond_mode_to_string(port->bond_mode));
3795 for (j = 0; j < port->n_ifaces; j++) {
3796 const struct iface *iface = port->ifaces[j];
3798 ds_put_cstr(&ds, ", ");
3800 ds_put_cstr(&ds, iface->name);
3802 ds_put_char(&ds, '\n');
3806 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3810 static struct port *
3811 bond_find(const char *name)
3813 const struct bridge *br;
3815 LIST_FOR_EACH (br, node, &all_bridges) {
3818 for (i = 0; i < br->n_ports; i++) {
3819 struct port *port = br->ports[i];
3820 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3829 ds_put_lacp_state(struct ds *ds, uint8_t state)
3831 if (state & LACP_STATE_ACT) {
3832 ds_put_cstr(ds, "activity ");
3835 if (state & LACP_STATE_TIME) {
3836 ds_put_cstr(ds, "timeout ");
3839 if (state & LACP_STATE_AGG) {
3840 ds_put_cstr(ds, "aggregation ");
3843 if (state & LACP_STATE_SYNC) {
3844 ds_put_cstr(ds, "synchronized ");
3847 if (state & LACP_STATE_COL) {
3848 ds_put_cstr(ds, "collecting ");
3851 if (state & LACP_STATE_DIST) {
3852 ds_put_cstr(ds, "distributing ");
3855 if (state & LACP_STATE_DEF) {
3856 ds_put_cstr(ds, "defaulted ");
3859 if (state & LACP_STATE_EXP) {
3860 ds_put_cstr(ds, "expired ");
3865 bond_unixctl_show(struct unixctl_conn *conn,
3866 const char *args, void *aux OVS_UNUSED)
3868 struct ds ds = DS_EMPTY_INITIALIZER;
3869 const struct port *port;
3872 port = bond_find(args);
3874 unixctl_command_reply(conn, 501, "no such bond");
3878 ds_put_format(&ds, "bond_mode: %s\n",
3879 bond_mode_to_string(port->bond_mode));
3882 ds_put_format(&ds, "lacp: %s\n",
3883 port->lacp & LACP_ACTIVE ? "active" : "passive");
3885 ds_put_cstr(&ds, "lacp: off\n");
3888 if (port->bond_mode != BM_AB) {
3889 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3890 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3894 ds_put_format(&ds, "bond-detect-mode: %s\n",
3895 port->miimon ? "miimon" : "carrier");
3898 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3899 port->bond_miimon_interval);
3902 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3903 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3905 if (port->bond_mode != BM_AB) {
3906 ds_put_format(&ds, "next rebalance: %lld ms\n",
3907 port->bond_next_rebalance - time_msec());
3910 for (j = 0; j < port->n_ifaces; j++) {
3911 const struct iface *iface = port->ifaces[j];
3912 struct bond_entry *be;
3916 ds_put_format(&ds, "\nslave %s: %s\n",
3917 iface->name, iface->enabled ? "enabled" : "disabled");
3918 if (j == port->active_iface) {
3919 ds_put_cstr(&ds, "\tactive slave\n");
3921 if (iface->delay_expires != LLONG_MAX) {
3922 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3923 iface->enabled ? "downdelay" : "updelay",
3924 iface->delay_expires - time_msec());
3928 ds_put_cstr(&ds, "\tstatus: ");
3930 if (iface->lacp_status & LACP_CURRENT) {
3931 ds_put_cstr(&ds, "current ");
3934 if (iface->lacp_status & LACP_EXPIRED) {
3935 ds_put_cstr(&ds, "expired ");
3938 if (iface->lacp_status & LACP_DEFAULTED) {
3939 ds_put_cstr(&ds, "defaulted ");
3942 if (iface->lacp_status & LACP_ATTACHED) {
3943 ds_put_cstr(&ds, "attached ");
3946 ds_put_cstr(&ds, "\n");
3948 ds_put_cstr(&ds, "\n\tactor sysid: ");
3949 ds_put_format(&ds, ETH_ADDR_FMT,
3950 ETH_ADDR_ARGS(iface->lacp_actor.sysid));
3951 ds_put_cstr(&ds, "\n");
3953 ds_put_format(&ds, "\tactor sys_priority: %u\n",
3954 ntohs(iface->lacp_actor.sys_priority));
3956 ds_put_format(&ds, "\tactor portid: %u\n",
3957 ntohs(iface->lacp_actor.portid));
3959 ds_put_format(&ds, "\tactor port_priority: %u\n",
3960 ntohs(iface->lacp_actor.port_priority));
3962 ds_put_format(&ds, "\tactor key: %u\n",
3963 ntohs(iface->lacp_actor.key));
3965 ds_put_cstr(&ds, "\tactor state: ");
3966 ds_put_lacp_state(&ds, iface_get_lacp_state(iface));
3967 ds_put_cstr(&ds, "\n\n");
3969 ds_put_cstr(&ds, "\tpartner sysid: ");
3970 ds_put_format(&ds, ETH_ADDR_FMT,
3971 ETH_ADDR_ARGS(iface->lacp_partner.sysid));
3972 ds_put_cstr(&ds, "\n");
3974 ds_put_format(&ds, "\tpartner sys_priority: %u\n",
3975 ntohs(iface->lacp_partner.sys_priority));
3977 ds_put_format(&ds, "\tpartner portid: %u\n",
3978 ntohs(iface->lacp_partner.portid));
3980 ds_put_format(&ds, "\tpartner port_priority: %u\n",
3981 ntohs(iface->lacp_partner.port_priority));
3983 ds_put_format(&ds, "\tpartner key: %u\n",
3984 ntohs(iface->lacp_partner.key));
3986 ds_put_cstr(&ds, "\tpartner state: ");
3987 ds_put_lacp_state(&ds, iface->lacp_partner.state);
3988 ds_put_cstr(&ds, "\n");
3991 if (port->bond_mode == BM_AB) {
3996 memset(&flow, 0, sizeof flow);
3997 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3998 int hash = be - port->bond_hash;
3999 struct mac_entry *me;
4001 if (be->iface_idx != j) {
4005 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
4006 hash, be->tx_bytes / 1024);
4008 if (port->bond_mode != BM_SLB) {
4013 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
4017 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
4018 if (bond_hash_src(me->mac, me->vlan) == hash
4019 && me->port != port->port_idx
4020 && choose_output_iface(port, &flow, me->vlan,
4022 && dp_ifidx == iface->dp_ifidx)
4024 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
4025 ETH_ADDR_ARGS(me->mac));
4030 unixctl_command_reply(conn, 200, ds_cstr(&ds));
4035 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
4036 void *aux OVS_UNUSED)
4038 char *args = (char *) args_;
4039 char *save_ptr = NULL;
4040 char *bond_s, *hash_s, *slave_s;
4042 struct iface *iface;
4043 struct bond_entry *entry;
4046 bond_s = strtok_r(args, " ", &save_ptr);
4047 hash_s = strtok_r(NULL, " ", &save_ptr);
4048 slave_s = strtok_r(NULL, " ", &save_ptr);
4050 unixctl_command_reply(conn, 501,
4051 "usage: bond/migrate BOND HASH SLAVE");
4055 port = bond_find(bond_s);
4057 unixctl_command_reply(conn, 501, "no such bond");
4061 if (port->bond_mode != BM_SLB) {
4062 unixctl_command_reply(conn, 501, "not an SLB bond");
4066 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
4067 hash = atoi(hash_s) & BOND_MASK;
4069 unixctl_command_reply(conn, 501, "bad hash");
4073 iface = port_lookup_iface(port, slave_s);
4075 unixctl_command_reply(conn, 501, "no such slave");
4079 if (!iface->enabled) {
4080 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
4084 entry = &port->bond_hash[hash];
4085 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
4086 entry->iface_idx = iface->port_ifidx;
4087 entry->iface_tag = tag_create_random();
4088 unixctl_command_reply(conn, 200, "migrated");
4092 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
4093 void *aux OVS_UNUSED)
4095 char *args = (char *) args_;
4096 char *save_ptr = NULL;
4097 char *bond_s, *slave_s;
4099 struct iface *iface;
4101 bond_s = strtok_r(args, " ", &save_ptr);
4102 slave_s = strtok_r(NULL, " ", &save_ptr);
4104 unixctl_command_reply(conn, 501,
4105 "usage: bond/set-active-slave BOND SLAVE");
4109 port = bond_find(bond_s);
4111 unixctl_command_reply(conn, 501, "no such bond");
4115 iface = port_lookup_iface(port, slave_s);
4117 unixctl_command_reply(conn, 501, "no such slave");
4121 if (!iface->enabled) {
4122 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
4126 if (port->active_iface != iface->port_ifidx) {
4127 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4128 port->active_iface = iface->port_ifidx;
4129 port->active_iface_tag = tag_create_random();
4130 VLOG_INFO("port %s: active interface is now %s",
4131 port->name, iface->name);
4132 bond_send_learning_packets(port);
4133 unixctl_command_reply(conn, 200, "done");
4135 unixctl_command_reply(conn, 200, "no change");
4140 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
4142 char *args = (char *) args_;
4143 char *save_ptr = NULL;
4144 char *bond_s, *slave_s;
4146 struct iface *iface;
4148 bond_s = strtok_r(args, " ", &save_ptr);
4149 slave_s = strtok_r(NULL, " ", &save_ptr);
4151 unixctl_command_reply(conn, 501,
4152 "usage: bond/enable/disable-slave BOND SLAVE");
4156 port = bond_find(bond_s);
4158 unixctl_command_reply(conn, 501, "no such bond");
4162 iface = port_lookup_iface(port, slave_s);
4164 unixctl_command_reply(conn, 501, "no such slave");
4168 bond_enable_slave(iface, enable);
4169 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
4173 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
4174 void *aux OVS_UNUSED)
4176 enable_slave(conn, args, true);
4180 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
4181 void *aux OVS_UNUSED)
4183 enable_slave(conn, args, false);
4187 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
4188 void *aux OVS_UNUSED)
4190 char *args = (char *) args_;
4191 uint8_t mac[ETH_ADDR_LEN];
4195 char *mac_s, *vlan_s;
4196 char *save_ptr = NULL;
4198 mac_s = strtok_r(args, " ", &save_ptr);
4199 vlan_s = strtok_r(NULL, " ", &save_ptr);
4202 if (sscanf(vlan_s, "%u", &vlan) != 1) {
4203 unixctl_command_reply(conn, 501, "invalid vlan");
4207 vlan = OFP_VLAN_NONE;
4210 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
4211 == ETH_ADDR_SCAN_COUNT) {
4212 hash = bond_hash_src(mac, vlan);
4214 hash_cstr = xasprintf("%u", hash);
4215 unixctl_command_reply(conn, 200, hash_cstr);
4218 unixctl_command_reply(conn, 501, "invalid mac");
4225 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
4226 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
4227 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
4228 unixctl_command_register("bond/set-active-slave",
4229 bond_unixctl_set_active_slave, NULL);
4230 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
4232 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
4234 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
4237 /* Port functions. */
4240 port_run(struct port *port)
4247 for (i = 0; i < port->n_ifaces; i++) {
4248 struct iface *iface = port->ifaces[i];
4251 struct ofpbuf *packet = cfm_run(iface->cfm);
4253 iface_send_packet(iface, packet);
4254 ofpbuf_uninit(packet);
4262 port_wait(struct port *port)
4269 for (i = 0; i < port->n_ifaces; i++) {
4270 struct iface *iface = port->ifaces[i];
4272 cfm_wait(iface->cfm);
4277 static struct port *
4278 port_create(struct bridge *br, const char *name)
4282 port = xzalloc(sizeof *port);
4284 port->port_idx = br->n_ports;
4286 port->trunks = NULL;
4287 port->name = xstrdup(name);
4288 port->active_iface = -1;
4290 if (br->n_ports >= br->allocated_ports) {
4291 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
4294 br->ports[br->n_ports++] = port;
4295 shash_add_assert(&br->port_by_name, port->name, port);
4297 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
4304 get_port_other_config(const struct ovsrec_port *port, const char *key,
4305 const char *default_value)
4309 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
4311 return value ? value : default_value;
4315 get_interface_other_config(const struct ovsrec_interface *iface,
4316 const char *key, const char *default_value)
4320 value = get_ovsrec_key_value(&iface->header_,
4321 &ovsrec_interface_col_other_config, key);
4322 return value ? value : default_value;
4326 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
4328 struct shash new_ifaces;
4331 /* Collect list of new interfaces. */
4332 shash_init(&new_ifaces);
4333 for (i = 0; i < cfg->n_interfaces; i++) {
4334 const char *name = cfg->interfaces[i]->name;
4335 shash_add_once(&new_ifaces, name, NULL);
4338 /* Get rid of deleted interfaces. */
4339 for (i = 0; i < port->n_ifaces; ) {
4340 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
4341 iface_destroy(port->ifaces[i]);
4347 shash_destroy(&new_ifaces);
4351 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
4353 const char *detect_mode;
4354 struct shash new_ifaces;
4355 long long int next_rebalance, miimon_next_update, lacp_priority;
4356 unsigned long *trunks;
4362 /* Update settings. */
4363 port->updelay = cfg->bond_updelay;
4364 if (port->updelay < 0) {
4367 port->downdelay = cfg->bond_downdelay;
4368 if (port->downdelay < 0) {
4369 port->downdelay = 0;
4371 port->bond_rebalance_interval = atoi(
4372 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
4373 if (port->bond_rebalance_interval < 1000) {
4374 port->bond_rebalance_interval = 1000;
4376 next_rebalance = time_msec() + port->bond_rebalance_interval;
4377 if (port->bond_next_rebalance > next_rebalance) {
4378 port->bond_next_rebalance = next_rebalance;
4381 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
4384 if (!strcmp(detect_mode, "carrier")) {
4385 port->miimon = false;
4386 } else if (!strcmp(detect_mode, "miimon")) {
4387 port->miimon = true;
4389 port->miimon = false;
4390 VLOG_WARN("port %s: unsupported bond-detect-mode %s, defaulting to "
4391 "carrier", port->name, detect_mode);
4394 port->bond_miimon_interval = atoi(
4395 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4396 if (port->bond_miimon_interval < 100) {
4397 port->bond_miimon_interval = 100;
4399 miimon_next_update = time_msec() + port->bond_miimon_interval;
4400 if (port->bond_miimon_next_update > miimon_next_update) {
4401 port->bond_miimon_next_update = miimon_next_update;
4404 if (!port->cfg->bond_mode ||
4405 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4406 port->bond_mode = BM_SLB;
4407 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4408 port->bond_mode = BM_AB;
4409 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
4410 port->bond_mode = BM_TCP;
4412 port->bond_mode = BM_SLB;
4413 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4414 port->name, port->cfg->bond_mode,
4415 bond_mode_to_string(port->bond_mode));
4418 /* Add new interfaces and update 'cfg' member of existing ones. */
4419 shash_init(&new_ifaces);
4420 for (i = 0; i < cfg->n_interfaces; i++) {
4421 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4422 struct iface *iface;
4424 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
4425 VLOG_WARN("port %s: %s specified twice as port interface",
4426 port->name, if_cfg->name);
4427 iface_set_ofport(if_cfg, -1);
4431 iface = iface_lookup(port->bridge, if_cfg->name);
4433 if (iface->port != port) {
4434 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4436 port->bridge->name, if_cfg->name, iface->port->name);
4439 iface->cfg = if_cfg;
4441 iface = iface_create(port, if_cfg);
4444 /* Determine interface type. The local port always has type
4445 * "internal". Other ports take their type from the database and
4446 * default to "system" if none is specified. */
4447 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4448 : if_cfg->type[0] ? if_cfg->type
4452 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4455 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4456 iface->lacp_priority = UINT16_MAX;
4458 iface->lacp_priority = lacp_priority;
4461 shash_destroy(&new_ifaces);
4464 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4466 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4467 /* Prefer bondable links if unspecified. */
4468 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4470 port->lacp_priority = lacp_priority;
4473 if (!port->cfg->lacp) {
4474 /* XXX when LACP implementation has been sufficiently tested, enable by
4475 * default and make active on bonded ports. */
4477 } else if (!strcmp(port->cfg->lacp, "off")) {
4479 } else if (!strcmp(port->cfg->lacp, "active")) {
4480 port->lacp = LACP_ACTIVE;
4481 } else if (!strcmp(port->cfg->lacp, "passive")) {
4482 port->lacp = LACP_PASSIVE;
4484 VLOG_WARN("port %s: unknown LACP mode %s",
4485 port->name, port->cfg->lacp);
4492 if (port->n_ifaces < 2) {
4494 if (vlan >= 0 && vlan <= 4095) {
4495 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4500 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4501 * they even work as-is. But they have not been tested. */
4502 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4506 if (port->vlan != vlan) {
4508 bridge_flush(port->bridge);
4511 /* Get trunked VLANs. */
4513 if (vlan < 0 && cfg->n_trunks) {
4516 trunks = bitmap_allocate(4096);
4518 for (i = 0; i < cfg->n_trunks; i++) {
4519 int trunk = cfg->trunks[i];
4521 bitmap_set1(trunks, trunk);
4527 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4528 port->name, cfg->n_trunks);
4530 if (n_errors == cfg->n_trunks) {
4531 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4533 bitmap_free(trunks);
4536 } else if (vlan >= 0 && cfg->n_trunks) {
4537 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4541 ? port->trunks != NULL
4542 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4543 bridge_flush(port->bridge);
4545 bitmap_free(port->trunks);
4546 port->trunks = trunks;
4550 port_destroy(struct port *port)
4553 struct bridge *br = port->bridge;
4557 for (i = 0; i < MAX_MIRRORS; i++) {
4558 struct mirror *m = br->mirrors[i];
4559 if (m && m->out_port == port) {
4564 while (port->n_ifaces > 0) {
4565 iface_destroy(port->ifaces[port->n_ifaces - 1]);
4568 shash_find_and_delete_assert(&br->port_by_name, port->name);
4570 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
4571 del->port_idx = port->port_idx;
4573 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4575 netdev_monitor_destroy(port->monitor);
4577 bitmap_free(port->trunks);
4578 free(port->bond_hash);
4585 static struct port *
4586 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4588 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4589 return iface ? iface->port : NULL;
4592 static struct port *
4593 port_lookup(const struct bridge *br, const char *name)
4595 return shash_find_data(&br->port_by_name, name);
4598 static struct iface *
4599 port_lookup_iface(const struct port *port, const char *name)
4601 struct iface *iface = iface_lookup(port->bridge, name);
4602 return iface && iface->port == port ? iface : NULL;
4606 port_update_lacp(struct port *port)
4611 if (!port->lacp || port->n_ifaces < 1) {
4612 for (i = 0; i < port->n_ifaces; i++) {
4613 iface_set_lacp_defaulted(port->ifaces[i]);
4619 for (i = 0; i < port->n_ifaces; i++) {
4620 struct iface *iface = port->ifaces[i];
4622 if (iface->dp_ifidx <= 0 || iface->dp_ifidx > UINT16_MAX) {
4627 if (iface->dp_ifidx == port->lacp_key) {
4628 key_changed = false;
4633 port->lacp_key = port->ifaces[0]->dp_ifidx;
4636 for (i = 0; i < port->n_ifaces; i++) {
4637 struct iface *iface = port->ifaces[i];
4639 iface->lacp_actor.sys_priority = htons(port->lacp_priority);
4640 memcpy(&iface->lacp_actor.sysid, port->bridge->ea, ETH_ADDR_LEN);
4642 iface->lacp_actor.port_priority = htons(iface->lacp_priority);
4643 iface->lacp_actor.portid = htons(iface->dp_ifidx);
4644 iface->lacp_actor.key = htons(port->lacp_key);
4648 port->lacp_need_update = true;
4652 port_update_bonding(struct port *port)
4654 if (port->monitor) {
4655 netdev_monitor_destroy(port->monitor);
4656 port->monitor = NULL;
4658 if (port->n_ifaces < 2) {
4659 /* Not a bonded port. */
4660 free(port->bond_hash);
4661 port->bond_hash = NULL;
4662 port->bond_fake_iface = false;
4663 port->active_iface = -1;
4664 port->no_ifaces_tag = 0;
4668 if (port->bond_mode != BM_AB && !port->bond_hash) {
4669 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4670 for (i = 0; i <= BOND_MASK; i++) {
4671 struct bond_entry *e = &port->bond_hash[i];
4675 port->bond_next_rebalance
4676 = time_msec() + port->bond_rebalance_interval;
4677 } else if (port->bond_mode == BM_AB) {
4678 free(port->bond_hash);
4679 port->bond_hash = NULL;
4682 if (!port->no_ifaces_tag) {
4683 port->no_ifaces_tag = tag_create_random();
4686 if (port->active_iface < 0) {
4687 bond_choose_active_iface(port);
4690 port->bond_fake_iface = port->cfg->bond_fake_iface;
4691 if (port->bond_fake_iface) {
4692 port->bond_next_fake_iface_update = time_msec();
4695 if (!port->miimon) {
4696 port->monitor = netdev_monitor_create();
4697 for (i = 0; i < port->n_ifaces; i++) {
4698 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
4704 /* Interface functions. */
4707 iface_set_lacp_defaulted(struct iface *iface)
4709 memset(&iface->lacp_partner, 0, sizeof iface->lacp_partner);
4711 iface->lacp_status |= LACP_DEFAULTED;
4712 iface->lacp_status &= ~(LACP_CURRENT | LACP_EXPIRED);
4714 iface->port->lacp_need_update = true;
4718 iface_set_lacp_expired(struct iface *iface)
4720 iface->lacp_status &= ~LACP_CURRENT;
4721 iface->lacp_status |= LACP_EXPIRED;
4722 iface->lacp_partner.state |= LACP_STATE_TIME;
4723 iface->lacp_partner.state &= ~LACP_STATE_SYNC;
4725 iface->lacp_rx = time_msec() + LACP_FAST_TIME_RX;
4730 iface_get_lacp_state(const struct iface *iface)
4734 if (iface->port->lacp & LACP_ACTIVE) {
4735 state |= LACP_STATE_ACT;
4738 if (iface->lacp_status & LACP_ATTACHED) {
4739 state |= LACP_STATE_SYNC;
4742 if (iface->lacp_status & LACP_DEFAULTED) {
4743 state |= LACP_STATE_DEF;
4746 if (iface->lacp_status & LACP_EXPIRED) {
4747 state |= LACP_STATE_EXP;
4750 if (iface->port->n_ifaces > 1) {
4751 state |= LACP_STATE_AGG;
4754 if (iface->enabled) {
4755 state |= LACP_STATE_COL | LACP_STATE_DIST;
4761 /* Given 'iface', populates 'priority' with data representing its LACP link
4762 * priority. If two priority objects populated by this function are compared
4763 * using memcmp, the higher priority link will be less than the lower priority
4766 iface_get_lacp_priority(struct iface *iface, struct lacp_info *priority)
4768 uint16_t partner_priority, actor_priority;
4770 /* Choose the lacp_info of the higher priority system by comparing their
4771 * system priorities and mac addresses. */
4772 actor_priority = ntohs(iface->lacp_actor.sys_priority);
4773 partner_priority = ntohs(iface->lacp_partner.sys_priority);
4774 if (actor_priority < partner_priority) {
4775 *priority = iface->lacp_actor;
4776 } else if (partner_priority < actor_priority) {
4777 *priority = iface->lacp_partner;
4778 } else if (eth_addr_compare_3way(iface->lacp_actor.sysid,
4779 iface->lacp_partner.sysid) < 0) {
4780 *priority = iface->lacp_actor;
4782 *priority = iface->lacp_partner;
4785 /* Key and state are not used in priority comparisons. */
4787 priority->state = 0;
4791 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
4794 union ofp_action action;
4796 memset(&action, 0, sizeof action);
4797 action.output.type = htons(OFPAT_OUTPUT);
4798 action.output.len = htons(sizeof action);
4799 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
4801 flow_extract(packet, 0, ODPP_NONE, &flow);
4803 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
4805 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4806 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
4810 static struct iface *
4811 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4813 struct bridge *br = port->bridge;
4814 struct iface *iface;
4815 char *name = if_cfg->name;
4817 iface = xzalloc(sizeof *iface);
4819 iface->port_ifidx = port->n_ifaces;
4820 iface->name = xstrdup(name);
4821 iface->dp_ifidx = -1;
4822 iface->tag = tag_create_random();
4823 iface->delay_expires = LLONG_MAX;
4824 iface->netdev = NULL;
4825 iface->cfg = if_cfg;
4826 iface_set_lacp_defaulted(iface);
4828 if (port->lacp & LACP_ACTIVE) {
4829 iface_set_lacp_expired(iface);
4832 shash_add_assert(&br->iface_by_name, iface->name, iface);
4834 if (port->n_ifaces >= port->allocated_ifaces) {
4835 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
4836 sizeof *port->ifaces);
4838 port->ifaces[port->n_ifaces++] = iface;
4839 if (port->n_ifaces > 1) {
4840 br->has_bonded_ports = true;
4843 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4851 iface_destroy(struct iface *iface)
4854 struct port *port = iface->port;
4855 struct bridge *br = port->bridge;
4856 bool del_active = port->active_iface == iface->port_ifidx;
4859 if (port->monitor) {
4860 netdev_monitor_remove(port->monitor, iface->netdev);
4863 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4865 if (iface->dp_ifidx >= 0) {
4866 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4869 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
4870 del->port_ifidx = iface->port_ifidx;
4872 netdev_close(iface->netdev);
4875 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4876 bond_choose_active_iface(port);
4877 bond_send_learning_packets(port);
4880 cfm_destroy(iface->cfm);
4885 bridge_flush(port->bridge);
4889 static struct iface *
4890 iface_lookup(const struct bridge *br, const char *name)
4892 return shash_find_data(&br->iface_by_name, name);
4895 static struct iface *
4896 iface_find(const char *name)
4898 const struct bridge *br;
4900 LIST_FOR_EACH (br, node, &all_bridges) {
4901 struct iface *iface = iface_lookup(br, name);
4910 static struct iface *
4911 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4913 struct iface *iface;
4915 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4916 hash_int(dp_ifidx, 0), &br->ifaces) {
4917 if (iface->dp_ifidx == dp_ifidx) {
4924 /* Set Ethernet address of 'iface', if one is specified in the configuration
4927 iface_set_mac(struct iface *iface)
4929 uint8_t ea[ETH_ADDR_LEN];
4931 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4932 if (eth_addr_is_multicast(ea)) {
4933 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4935 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4936 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4937 iface->name, iface->name);
4939 int error = netdev_set_etheraddr(iface->netdev, ea);
4941 VLOG_ERR("interface %s: setting MAC failed (%s)",
4942 iface->name, strerror(error));
4948 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4950 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4953 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4957 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4959 * The value strings in '*shash' are taken directly from values[], not copied,
4960 * so the caller should not modify or free them. */
4962 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4963 struct shash *shash)
4968 for (i = 0; i < n; i++) {
4969 shash_add(shash, keys[i], values[i]);
4973 /* Creates 'keys' and 'values' arrays from 'shash'.
4975 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4976 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4977 * are populated with with strings taken directly from 'shash' and thus have
4978 * the same ownership of the key-value pairs in shash.
4981 shash_to_ovs_idl_map(struct shash *shash,
4982 char ***keys, char ***values, size_t *n)
4986 struct shash_node *sn;
4988 count = shash_count(shash);
4990 k = xmalloc(count * sizeof *k);
4991 v = xmalloc(count * sizeof *v);
4994 SHASH_FOR_EACH(sn, shash) {
5005 struct iface_delete_queues_cbdata {
5006 struct netdev *netdev;
5007 const struct ovsdb_datum *queues;
5011 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
5013 union ovsdb_atom atom;
5015 atom.integer = target;
5016 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
5020 iface_delete_queues(unsigned int queue_id,
5021 const struct shash *details OVS_UNUSED, void *cbdata_)
5023 struct iface_delete_queues_cbdata *cbdata = cbdata_;
5025 if (!queue_ids_include(cbdata->queues, queue_id)) {
5026 netdev_delete_queue(cbdata->netdev, queue_id);
5031 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
5033 if (!qos || qos->type[0] == '\0') {
5034 netdev_set_qos(iface->netdev, NULL, NULL);
5036 struct iface_delete_queues_cbdata cbdata;
5037 struct shash details;
5040 /* Configure top-level Qos for 'iface'. */
5041 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
5042 qos->n_other_config, &details);
5043 netdev_set_qos(iface->netdev, qos->type, &details);
5044 shash_destroy(&details);
5046 /* Deconfigure queues that were deleted. */
5047 cbdata.netdev = iface->netdev;
5048 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
5050 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
5052 /* Configure queues for 'iface'. */
5053 for (i = 0; i < qos->n_queues; i++) {
5054 const struct ovsrec_queue *queue = qos->value_queues[i];
5055 unsigned int queue_id = qos->key_queues[i];
5057 shash_from_ovs_idl_map(queue->key_other_config,
5058 queue->value_other_config,
5059 queue->n_other_config, &details);
5060 netdev_set_queue(iface->netdev, queue_id, &details);
5061 shash_destroy(&details);
5067 iface_update_cfm(struct iface *iface)
5071 uint16_t *remote_mps;
5072 struct ovsrec_monitor *mon;
5073 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
5075 mon = iface->cfg->monitor;
5078 cfm_destroy(iface->cfm);
5083 if (netdev_get_etheraddr(iface->netdev, ea)) {
5084 VLOG_WARN("interface %s: Failed to get ethernet address. "
5085 "Skipping Monitor.", iface->name);
5089 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
5090 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
5095 iface->cfm = cfm_create();
5099 cfm->mpid = mon->mpid;
5100 cfm->interval = mon->interval ? *mon->interval : 1000;
5102 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
5103 memcpy(cfm->maid, maid, sizeof cfm->maid);
5105 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
5106 for(i = 0; i < mon->n_remote_mps; i++) {
5107 remote_mps[i] = mon->remote_mps[i]->mpid;
5109 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
5112 if (!cfm_configure(iface->cfm)) {
5113 cfm_destroy(iface->cfm);
5118 /* Port mirroring. */
5120 static struct mirror *
5121 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
5125 for (i = 0; i < MAX_MIRRORS; i++) {
5126 struct mirror *m = br->mirrors[i];
5127 if (m && uuid_equals(uuid, &m->uuid)) {
5135 mirror_reconfigure(struct bridge *br)
5137 unsigned long *rspan_vlans;
5140 /* Get rid of deleted mirrors. */
5141 for (i = 0; i < MAX_MIRRORS; i++) {
5142 struct mirror *m = br->mirrors[i];
5144 const struct ovsdb_datum *mc;
5145 union ovsdb_atom atom;
5147 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
5148 atom.uuid = br->mirrors[i]->uuid;
5149 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
5155 /* Add new mirrors and reconfigure existing ones. */
5156 for (i = 0; i < br->cfg->n_mirrors; i++) {
5157 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
5158 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
5160 mirror_reconfigure_one(m, cfg);
5162 mirror_create(br, cfg);
5166 /* Update port reserved status. */
5167 for (i = 0; i < br->n_ports; i++) {
5168 br->ports[i]->is_mirror_output_port = false;
5170 for (i = 0; i < MAX_MIRRORS; i++) {
5171 struct mirror *m = br->mirrors[i];
5172 if (m && m->out_port) {
5173 m->out_port->is_mirror_output_port = true;
5177 /* Update flooded vlans (for RSPAN). */
5179 if (br->cfg->n_flood_vlans) {
5180 rspan_vlans = bitmap_allocate(4096);
5182 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
5183 int64_t vlan = br->cfg->flood_vlans[i];
5184 if (vlan >= 0 && vlan < 4096) {
5185 bitmap_set1(rspan_vlans, vlan);
5186 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
5189 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
5194 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
5200 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
5205 for (i = 0; ; i++) {
5206 if (i >= MAX_MIRRORS) {
5207 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
5208 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
5211 if (!br->mirrors[i]) {
5216 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
5219 br->mirrors[i] = m = xzalloc(sizeof *m);
5222 m->name = xstrdup(cfg->name);
5223 shash_init(&m->src_ports);
5224 shash_init(&m->dst_ports);
5230 mirror_reconfigure_one(m, cfg);
5234 mirror_destroy(struct mirror *m)
5237 struct bridge *br = m->bridge;
5240 for (i = 0; i < br->n_ports; i++) {
5241 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
5242 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
5245 shash_destroy(&m->src_ports);
5246 shash_destroy(&m->dst_ports);
5249 m->bridge->mirrors[m->idx] = NULL;
5258 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
5259 struct shash *names)
5263 for (i = 0; i < n_ports; i++) {
5264 const char *name = ports[i]->name;
5265 if (port_lookup(m->bridge, name)) {
5266 shash_add_once(names, name, NULL);
5268 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
5269 "port %s", m->bridge->name, m->name, name);
5275 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
5281 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
5283 for (i = 0; i < cfg->n_select_vlan; i++) {
5284 int64_t vlan = cfg->select_vlan[i];
5285 if (vlan < 0 || vlan > 4095) {
5286 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
5287 m->bridge->name, m->name, vlan);
5289 (*vlans)[n_vlans++] = vlan;
5296 vlan_is_mirrored(const struct mirror *m, int vlan)
5300 for (i = 0; i < m->n_vlans; i++) {
5301 if (m->vlans[i] == vlan) {
5309 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
5313 for (i = 0; i < m->n_vlans; i++) {
5314 if (port_trunks_vlan(p, m->vlans[i])) {
5322 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
5324 struct shash src_ports, dst_ports;
5325 mirror_mask_t mirror_bit;
5326 struct port *out_port;
5333 if (strcmp(cfg->name, m->name)) {
5335 m->name = xstrdup(cfg->name);
5338 /* Get output port. */
5339 if (cfg->output_port) {
5340 out_port = port_lookup(m->bridge, cfg->output_port->name);
5342 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
5343 m->bridge->name, m->name);
5349 if (cfg->output_vlan) {
5350 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
5351 "output vlan; ignoring output vlan",
5352 m->bridge->name, m->name);
5354 } else if (cfg->output_vlan) {
5356 out_vlan = *cfg->output_vlan;
5358 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
5359 m->bridge->name, m->name);
5364 shash_init(&src_ports);
5365 shash_init(&dst_ports);
5366 if (cfg->select_all) {
5367 for (i = 0; i < m->bridge->n_ports; i++) {
5368 const char *name = m->bridge->ports[i]->name;
5369 shash_add_once(&src_ports, name, NULL);
5370 shash_add_once(&dst_ports, name, NULL);
5375 /* Get ports, and drop duplicates and ports that don't exist. */
5376 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
5378 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
5381 /* Get all the vlans, and drop duplicate and invalid vlans. */
5382 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
5385 /* Update mirror data. */
5386 if (!shash_equal_keys(&m->src_ports, &src_ports)
5387 || !shash_equal_keys(&m->dst_ports, &dst_ports)
5388 || m->n_vlans != n_vlans
5389 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
5390 || m->out_port != out_port
5391 || m->out_vlan != out_vlan) {
5392 bridge_flush(m->bridge);
5394 shash_swap(&m->src_ports, &src_ports);
5395 shash_swap(&m->dst_ports, &dst_ports);
5398 m->n_vlans = n_vlans;
5399 m->out_port = out_port;
5400 m->out_vlan = out_vlan;
5403 mirror_bit = MIRROR_MASK_C(1) << m->idx;
5404 for (i = 0; i < m->bridge->n_ports; i++) {
5405 struct port *port = m->bridge->ports[i];
5407 if (shash_find(&m->src_ports, port->name)
5410 ? port_trunks_any_mirrored_vlan(m, port)
5411 : vlan_is_mirrored(m, port->vlan)))) {
5412 port->src_mirrors |= mirror_bit;
5414 port->src_mirrors &= ~mirror_bit;
5417 if (shash_find(&m->dst_ports, port->name)) {
5418 port->dst_mirrors |= mirror_bit;
5420 port->dst_mirrors &= ~mirror_bit;
5425 shash_destroy(&src_ports);
5426 shash_destroy(&dst_ports);