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;
183 struct netdev_monitor *monitor; /* Tracks carrier. NULL if miimon. */
184 long long int miimon_interval; /* Miimon status refresh interval. */
185 long long int miimon_next_update; /* Time of next miimon update. */
187 /* An ordinary bridge port has 1 interface.
188 * A bridge port for bonding has at least 2 interfaces. */
189 struct iface **ifaces;
190 size_t n_ifaces, allocated_ifaces;
193 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
194 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
195 tag_type active_iface_tag; /* Tag for bcast flows. */
196 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
197 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
198 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
199 long long int bond_next_fake_iface_update; /* Time of next update. */
201 /* LACP information. */
202 int lacp; /* LACP status flags. 0 if LACP is off. */
203 uint16_t lacp_key; /* LACP aggregation key. */
204 uint16_t lacp_priority; /* LACP system priority. */
205 bool lacp_need_update; /* Need to update attached interfaces? */
207 /* SLB specific bonding info. */
208 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
209 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
210 long long int bond_next_rebalance; /* Next rebalancing time. */
212 /* Port mirroring info. */
213 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
214 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
215 bool is_mirror_output_port; /* Does port mirroring send frames here? */
219 struct list node; /* Node in global list of bridges. */
220 char *name; /* User-specified arbitrary name. */
221 struct mac_learning *ml; /* MAC learning table. */
222 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
223 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
224 const struct ovsrec_bridge *cfg;
226 /* OpenFlow switch processing. */
227 struct ofproto *ofproto; /* OpenFlow switch. */
229 /* Kernel datapath information. */
230 struct dpif *dpif; /* Datapath. */
231 struct hmap ifaces; /* Contains "struct iface"s. */
235 size_t n_ports, allocated_ports;
236 struct shash iface_by_name; /* "struct iface"s indexed by name. */
237 struct shash port_by_name; /* "struct port"s indexed by name. */
240 bool has_bonded_ports;
245 /* Port mirroring. */
246 struct mirror *mirrors[MAX_MIRRORS];
249 /* List of all bridges. */
250 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
252 /* OVSDB IDL used to obtain configuration. */
253 static struct ovsdb_idl *idl;
255 /* Each time this timer expires, the bridge fetches systems and interface
256 * statistics and pushes them into the database. */
257 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
258 static long long int stats_timer = LLONG_MIN;
260 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
261 static void bridge_destroy(struct bridge *);
262 static struct bridge *bridge_lookup(const char *name);
263 static unixctl_cb_func bridge_unixctl_dump_flows;
264 static unixctl_cb_func bridge_unixctl_reconnect;
265 static int bridge_run_one(struct bridge *);
266 static size_t bridge_get_controllers(const struct bridge *br,
267 struct ovsrec_controller ***controllersp);
268 static void bridge_reconfigure_one(struct bridge *);
269 static void bridge_reconfigure_remotes(struct bridge *,
270 const struct sockaddr_in *managers,
272 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
273 static void bridge_fetch_dp_ifaces(struct bridge *);
274 static void bridge_flush(struct bridge *);
275 static void bridge_pick_local_hw_addr(struct bridge *,
276 uint8_t ea[ETH_ADDR_LEN],
277 struct iface **hw_addr_iface);
278 static uint64_t bridge_pick_datapath_id(struct bridge *,
279 const uint8_t bridge_ea[ETH_ADDR_LEN],
280 struct iface *hw_addr_iface);
281 static struct iface *bridge_get_local_iface(struct bridge *);
282 static uint64_t dpid_from_hash(const void *, size_t nbytes);
284 static unixctl_cb_func bridge_unixctl_fdb_show;
285 static unixctl_cb_func qos_unixctl_show;
287 static void lacp_run(struct port *);
288 static void lacp_wait(struct port *);
289 static void lacp_process_packet(const struct ofpbuf *, struct iface *);
291 static void bond_init(void);
292 static void bond_run(struct port *);
293 static void bond_wait(struct port *);
294 static void bond_rebalance_port(struct port *);
295 static void bond_send_learning_packets(struct port *);
296 static void bond_enable_slave(struct iface *iface, bool enable);
298 static void port_run(struct port *);
299 static void port_wait(struct port *);
300 static struct port *port_create(struct bridge *, const char *name);
301 static void port_reconfigure(struct port *, const struct ovsrec_port *);
302 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
303 static void port_destroy(struct port *);
304 static struct port *port_lookup(const struct bridge *, const char *name);
305 static struct iface *port_lookup_iface(const struct port *, const char *name);
306 static struct port *port_from_dp_ifidx(const struct bridge *,
308 static void port_update_bonding(struct port *);
309 static void port_update_lacp(struct port *);
311 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
312 static void mirror_destroy(struct mirror *);
313 static void mirror_reconfigure(struct bridge *);
314 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
315 static bool vlan_is_mirrored(const struct mirror *, int vlan);
317 static struct iface *iface_create(struct port *port,
318 const struct ovsrec_interface *if_cfg);
319 static void iface_destroy(struct iface *);
320 static struct iface *iface_lookup(const struct bridge *, const char *name);
321 static struct iface *iface_find(const char *name);
322 static struct iface *iface_from_dp_ifidx(const struct bridge *,
324 static void iface_set_mac(struct iface *);
325 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
326 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
327 static void iface_update_cfm(struct iface *);
328 static void iface_refresh_cfm_stats(struct iface *iface);
329 static void iface_send_packet(struct iface *, struct ofpbuf *packet);
330 static uint8_t iface_get_lacp_state(const struct iface *);
331 static void iface_get_lacp_priority(struct iface *, struct lacp_info *);
332 static void iface_set_lacp_defaulted(struct iface *);
333 static void iface_set_lacp_expired(struct iface *);
334 static void iface_update_carrier(struct iface *, bool carrier);
336 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
338 static void shash_to_ovs_idl_map(struct shash *,
339 char ***keys, char ***values, size_t *n);
342 /* Hooks into ofproto processing. */
343 static struct ofhooks bridge_ofhooks;
345 /* Public functions. */
347 /* Initializes the bridge module, configuring it to obtain its configuration
348 * from an OVSDB server accessed over 'remote', which should be a string in a
349 * form acceptable to ovsdb_idl_create(). */
351 bridge_init(const char *remote)
353 /* Create connection to database. */
354 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
356 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
357 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
358 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
360 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
362 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
363 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
365 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
366 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
367 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
369 /* Register unixctl commands. */
370 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
371 unixctl_command_register("qos/show", qos_unixctl_show, NULL);
372 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
374 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
382 struct bridge *br, *next_br;
384 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
387 ovsdb_idl_destroy(idl);
390 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
391 * but for which the ovs-vswitchd configuration 'cfg' is required. */
393 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
395 static bool already_configured_once;
396 struct svec bridge_names;
397 struct svec dpif_names, dpif_types;
400 /* Only do this once per ovs-vswitchd run. */
401 if (already_configured_once) {
404 already_configured_once = true;
406 stats_timer = time_msec() + STATS_INTERVAL;
408 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
409 svec_init(&bridge_names);
410 for (i = 0; i < cfg->n_bridges; i++) {
411 svec_add(&bridge_names, cfg->bridges[i]->name);
413 svec_sort(&bridge_names);
415 /* Iterate over all system dpifs and delete any of them that do not appear
417 svec_init(&dpif_names);
418 svec_init(&dpif_types);
419 dp_enumerate_types(&dpif_types);
420 for (i = 0; i < dpif_types.n; i++) {
423 dp_enumerate_names(dpif_types.names[i], &dpif_names);
425 /* Delete each dpif whose name is not in 'bridge_names'. */
426 for (j = 0; j < dpif_names.n; j++) {
427 if (!svec_contains(&bridge_names, dpif_names.names[j])) {
431 retval = dpif_open(dpif_names.names[j], dpif_types.names[i],
440 svec_destroy(&bridge_names);
441 svec_destroy(&dpif_names);
442 svec_destroy(&dpif_types);
445 /* Callback for iterate_and_prune_ifaces(). */
447 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
449 if (!iface->netdev) {
450 /* We already reported a related error, don't bother duplicating it. */
454 if (iface->dp_ifidx < 0) {
455 VLOG_ERR("%s interface not in %s, dropping",
456 iface->name, dpif_name(br->dpif));
460 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
461 iface->name, iface->dp_ifidx);
465 /* Callback for iterate_and_prune_ifaces(). */
467 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
468 void *aux OVS_UNUSED)
470 /* Set policing attributes. */
471 netdev_set_policing(iface->netdev,
472 iface->cfg->ingress_policing_rate,
473 iface->cfg->ingress_policing_burst);
475 /* Set MAC address of internal interfaces other than the local
477 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
478 iface_set_mac(iface);
484 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
485 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
486 * deletes from 'br' any ports that no longer have any interfaces. */
488 iterate_and_prune_ifaces(struct bridge *br,
489 bool (*cb)(struct bridge *, struct iface *,
495 for (i = 0; i < br->n_ports; ) {
496 struct port *port = br->ports[i];
497 for (j = 0; j < port->n_ifaces; ) {
498 struct iface *iface = port->ifaces[j];
499 if (cb(br, iface, aux)) {
502 iface_set_ofport(iface->cfg, -1);
503 iface_destroy(iface);
507 if (port->n_ifaces) {
510 VLOG_WARN("%s port has no interfaces, dropping", port->name);
516 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
517 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
518 * responsible for freeing '*managersp' (with free()).
520 * You may be asking yourself "why does ovs-vswitchd care?", because
521 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
522 * should not be and in fact is not directly involved in that. But
523 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
524 * it has to tell in-band control where the managers are to enable that.
525 * (Thus, only managers connected in-band are collected.)
528 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
529 struct sockaddr_in **managersp, size_t *n_managersp)
531 struct sockaddr_in *managers = NULL;
532 size_t n_managers = 0;
533 struct shash targets;
536 /* Collect all of the potential targets from the "targets" columns of the
537 * rows pointed to by "manager_options", excluding any that are
539 shash_init(&targets);
540 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
541 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
543 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
544 shash_find_and_delete(&targets, m->target);
546 shash_add_once(&targets, m->target, NULL);
550 /* Now extract the targets' IP addresses. */
551 if (!shash_is_empty(&targets)) {
552 struct shash_node *node;
554 managers = xmalloc(shash_count(&targets) * sizeof *managers);
555 SHASH_FOR_EACH (node, &targets) {
556 const char *target = node->name;
557 struct sockaddr_in *sin = &managers[n_managers];
559 if ((!strncmp(target, "tcp:", 4)
560 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
561 (!strncmp(target, "ssl:", 4)
562 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
567 shash_destroy(&targets);
569 *managersp = managers;
570 *n_managersp = n_managers;
574 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
576 struct shash old_br, new_br;
577 struct shash_node *node;
578 struct bridge *br, *next;
579 struct sockaddr_in *managers;
582 int sflow_bridge_number;
584 COVERAGE_INC(bridge_reconfigure);
586 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
588 /* Collect old and new bridges. */
591 LIST_FOR_EACH (br, node, &all_bridges) {
592 shash_add(&old_br, br->name, br);
594 for (i = 0; i < ovs_cfg->n_bridges; i++) {
595 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
596 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
597 VLOG_WARN("more than one bridge named %s", br_cfg->name);
601 /* Get rid of deleted bridges and add new bridges. */
602 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
603 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
610 SHASH_FOR_EACH (node, &new_br) {
611 const char *br_name = node->name;
612 const struct ovsrec_bridge *br_cfg = node->data;
613 br = shash_find_data(&old_br, br_name);
615 /* If the bridge datapath type has changed, we need to tear it
616 * down and recreate. */
617 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
619 bridge_create(br_cfg);
622 bridge_create(br_cfg);
625 shash_destroy(&old_br);
626 shash_destroy(&new_br);
628 /* Reconfigure all bridges. */
629 LIST_FOR_EACH (br, node, &all_bridges) {
630 bridge_reconfigure_one(br);
633 /* Add and delete ports on all datapaths.
635 * The kernel will reject any attempt to add a given port to a datapath if
636 * that port already belongs to a different datapath, so we must do all
637 * port deletions before any port additions. */
638 LIST_FOR_EACH (br, node, &all_bridges) {
639 struct dpif_port_dump dump;
640 struct shash want_ifaces;
641 struct dpif_port dpif_port;
643 bridge_get_all_ifaces(br, &want_ifaces);
644 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
645 if (!shash_find(&want_ifaces, dpif_port.name)
646 && strcmp(dpif_port.name, br->name)) {
647 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
649 VLOG_WARN("failed to remove %s interface from %s: %s",
650 dpif_port.name, dpif_name(br->dpif),
655 shash_destroy(&want_ifaces);
657 LIST_FOR_EACH (br, node, &all_bridges) {
658 struct shash cur_ifaces, want_ifaces;
659 struct dpif_port_dump dump;
660 struct dpif_port dpif_port;
662 /* Get the set of interfaces currently in this datapath. */
663 shash_init(&cur_ifaces);
664 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
665 struct dpif_port *port_info = xmalloc(sizeof *port_info);
666 dpif_port_clone(port_info, &dpif_port);
667 shash_add(&cur_ifaces, dpif_port.name, port_info);
670 /* Get the set of interfaces we want on this datapath. */
671 bridge_get_all_ifaces(br, &want_ifaces);
673 hmap_clear(&br->ifaces);
674 SHASH_FOR_EACH (node, &want_ifaces) {
675 const char *if_name = node->name;
676 struct iface *iface = node->data;
677 struct dpif_port *dpif_port;
681 type = iface ? iface->type : "internal";
682 dpif_port = shash_find_data(&cur_ifaces, if_name);
684 /* If we have a port or a netdev already, and it's not the type we
685 * want, then delete the port (if any) and close the netdev (if
687 if ((dpif_port && strcmp(dpif_port->type, type))
688 || (iface && iface->netdev
689 && strcmp(type, netdev_get_type(iface->netdev)))) {
691 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
698 netdev_close(iface->netdev);
699 iface->netdev = NULL;
703 /* If the port doesn't exist or we don't have the netdev open,
704 * we need to do more work. */
705 if (!dpif_port || (iface && !iface->netdev)) {
706 struct netdev_options options;
707 struct netdev *netdev;
710 /* First open the network device. */
711 options.name = if_name;
713 options.args = &args;
714 options.ethertype = NETDEV_ETH_TYPE_NONE;
718 shash_from_ovs_idl_map(iface->cfg->key_options,
719 iface->cfg->value_options,
720 iface->cfg->n_options, &args);
722 error = netdev_open(&options, &netdev);
723 shash_destroy(&args);
726 VLOG_WARN("could not open network device %s (%s)",
727 if_name, strerror(error));
731 /* Then add the port if we haven't already. */
733 error = dpif_port_add(br->dpif, netdev, NULL);
735 netdev_close(netdev);
736 if (error == EFBIG) {
737 VLOG_ERR("ran out of valid port numbers on %s",
738 dpif_name(br->dpif));
741 VLOG_WARN("failed to add %s interface to %s: %s",
742 if_name, dpif_name(br->dpif),
749 /* Update 'iface'. */
751 iface->netdev = netdev;
752 iface->enabled = netdev_get_carrier(iface->netdev);
753 iface->up = iface->enabled;
755 } else if (iface && iface->netdev) {
759 shash_from_ovs_idl_map(iface->cfg->key_options,
760 iface->cfg->value_options,
761 iface->cfg->n_options, &args);
762 netdev_set_config(iface->netdev, &args);
763 shash_destroy(&args);
766 shash_destroy(&want_ifaces);
768 SHASH_FOR_EACH (node, &cur_ifaces) {
769 struct dpif_port *port_info = node->data;
770 dpif_port_destroy(port_info);
773 shash_destroy(&cur_ifaces);
775 sflow_bridge_number = 0;
776 LIST_FOR_EACH (br, node, &all_bridges) {
779 struct iface *local_iface;
780 struct iface *hw_addr_iface;
783 bridge_fetch_dp_ifaces(br);
785 iterate_and_prune_ifaces(br, check_iface, NULL);
787 /* Pick local port hardware address, datapath ID. */
788 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
789 local_iface = bridge_get_local_iface(br);
791 int error = netdev_set_etheraddr(local_iface->netdev, ea);
793 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
794 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
795 "Ethernet address: %s",
796 br->name, strerror(error));
799 memcpy(br->ea, ea, ETH_ADDR_LEN);
801 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
802 ofproto_set_datapath_id(br->ofproto, dpid);
804 dpid_string = xasprintf("%016"PRIx64, dpid);
805 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
808 /* Set NetFlow configuration on this bridge. */
809 if (br->cfg->netflow) {
810 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
811 struct netflow_options opts;
813 memset(&opts, 0, sizeof opts);
815 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
816 if (nf_cfg->engine_type) {
817 opts.engine_type = *nf_cfg->engine_type;
819 if (nf_cfg->engine_id) {
820 opts.engine_id = *nf_cfg->engine_id;
823 opts.active_timeout = nf_cfg->active_timeout;
824 if (!opts.active_timeout) {
825 opts.active_timeout = -1;
826 } else if (opts.active_timeout < 0) {
827 VLOG_WARN("bridge %s: active timeout interval set to negative "
828 "value, using default instead (%d seconds)", br->name,
829 NF_ACTIVE_TIMEOUT_DEFAULT);
830 opts.active_timeout = -1;
833 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
834 if (opts.add_id_to_iface) {
835 if (opts.engine_id > 0x7f) {
836 VLOG_WARN("bridge %s: netflow port mangling may conflict "
837 "with another vswitch, choose an engine id less "
838 "than 128", br->name);
840 if (br->n_ports > 508) {
841 VLOG_WARN("bridge %s: netflow port mangling will conflict "
842 "with another port when more than 508 ports are "
847 opts.collectors.n = nf_cfg->n_targets;
848 opts.collectors.names = nf_cfg->targets;
849 if (ofproto_set_netflow(br->ofproto, &opts)) {
850 VLOG_ERR("bridge %s: problem setting netflow collectors",
854 ofproto_set_netflow(br->ofproto, NULL);
857 /* Set sFlow configuration on this bridge. */
858 if (br->cfg->sflow) {
859 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
860 struct ovsrec_controller **controllers;
861 struct ofproto_sflow_options oso;
862 size_t n_controllers;
864 memset(&oso, 0, sizeof oso);
866 oso.targets.n = sflow_cfg->n_targets;
867 oso.targets.names = sflow_cfg->targets;
869 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
870 if (sflow_cfg->sampling) {
871 oso.sampling_rate = *sflow_cfg->sampling;
874 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
875 if (sflow_cfg->polling) {
876 oso.polling_interval = *sflow_cfg->polling;
879 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
880 if (sflow_cfg->header) {
881 oso.header_len = *sflow_cfg->header;
884 oso.sub_id = sflow_bridge_number++;
885 oso.agent_device = sflow_cfg->agent;
887 oso.control_ip = NULL;
888 n_controllers = bridge_get_controllers(br, &controllers);
889 for (i = 0; i < n_controllers; i++) {
890 if (controllers[i]->local_ip) {
891 oso.control_ip = controllers[i]->local_ip;
895 ofproto_set_sflow(br->ofproto, &oso);
897 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
899 ofproto_set_sflow(br->ofproto, NULL);
902 /* Update the controller and related settings. It would be more
903 * straightforward to call this from bridge_reconfigure_one(), but we
904 * can't do it there for two reasons. First, and most importantly, at
905 * that point we don't know the dp_ifidx of any interfaces that have
906 * been added to the bridge (because we haven't actually added them to
907 * the datapath). Second, at that point we haven't set the datapath ID
908 * yet; when a controller is configured, resetting the datapath ID will
909 * immediately disconnect from the controller, so it's better to set
910 * the datapath ID before the controller. */
911 bridge_reconfigure_remotes(br, managers, n_managers);
913 LIST_FOR_EACH (br, node, &all_bridges) {
914 for (i = 0; i < br->n_ports; i++) {
915 struct port *port = br->ports[i];
919 for (j = 0; j < port->n_ifaces; j++) {
920 netdev_monitor_add(port->monitor, port->ifaces[j]->netdev);
923 port->miimon_next_update = 0;
926 port_update_lacp(port);
927 port_update_bonding(port);
929 for (j = 0; j < port->n_ifaces; j++) {
930 iface_update_qos(port->ifaces[j], port->cfg->qos);
934 LIST_FOR_EACH (br, node, &all_bridges) {
935 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
938 LIST_FOR_EACH (br, node, &all_bridges) {
940 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
941 iface_update_cfm(iface);
947 /* ovs-vswitchd has completed initialization, so allow the process that
948 * forked us to exit successfully. */
949 daemonize_complete();
953 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
954 const struct ovsdb_idl_column *column,
957 const struct ovsdb_datum *datum;
958 union ovsdb_atom atom;
961 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
962 atom.string = (char *) key;
963 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
964 return idx == UINT_MAX ? NULL : datum->values[idx].string;
968 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
970 return get_ovsrec_key_value(&br_cfg->header_,
971 &ovsrec_bridge_col_other_config, key);
975 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
976 struct iface **hw_addr_iface)
982 *hw_addr_iface = NULL;
984 /* Did the user request a particular MAC? */
985 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
986 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
987 if (eth_addr_is_multicast(ea)) {
988 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
989 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
990 } else if (eth_addr_is_zero(ea)) {
991 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
997 /* Otherwise choose the minimum non-local MAC address among all of the
999 memset(ea, 0xff, ETH_ADDR_LEN);
1000 for (i = 0; i < br->n_ports; i++) {
1001 struct port *port = br->ports[i];
1002 uint8_t iface_ea[ETH_ADDR_LEN];
1003 struct iface *iface;
1005 /* Mirror output ports don't participate. */
1006 if (port->is_mirror_output_port) {
1010 /* Choose the MAC address to represent the port. */
1011 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
1012 /* Find the interface with this Ethernet address (if any) so that
1013 * we can provide the correct devname to the caller. */
1015 for (j = 0; j < port->n_ifaces; j++) {
1016 struct iface *candidate = port->ifaces[j];
1017 uint8_t candidate_ea[ETH_ADDR_LEN];
1018 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
1019 && eth_addr_equals(iface_ea, candidate_ea)) {
1024 /* Choose the interface whose MAC address will represent the port.
1025 * The Linux kernel bonding code always chooses the MAC address of
1026 * the first slave added to a bond, and the Fedora networking
1027 * scripts always add slaves to a bond in alphabetical order, so
1028 * for compatibility we choose the interface with the name that is
1029 * first in alphabetical order. */
1030 iface = port->ifaces[0];
1031 for (j = 1; j < port->n_ifaces; j++) {
1032 struct iface *candidate = port->ifaces[j];
1033 if (strcmp(candidate->name, iface->name) < 0) {
1038 /* The local port doesn't count (since we're trying to choose its
1039 * MAC address anyway). */
1040 if (iface->dp_ifidx == ODPP_LOCAL) {
1045 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1047 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1048 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1049 iface->name, strerror(error));
1054 /* Compare against our current choice. */
1055 if (!eth_addr_is_multicast(iface_ea) &&
1056 !eth_addr_is_local(iface_ea) &&
1057 !eth_addr_is_reserved(iface_ea) &&
1058 !eth_addr_is_zero(iface_ea) &&
1059 eth_addr_compare_3way(iface_ea, ea) < 0)
1061 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1062 *hw_addr_iface = iface;
1065 if (eth_addr_is_multicast(ea)) {
1066 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1067 *hw_addr_iface = NULL;
1068 VLOG_WARN("bridge %s: using default bridge Ethernet "
1069 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1071 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1072 br->name, ETH_ADDR_ARGS(ea));
1076 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1077 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1078 * an interface on 'br', then that interface must be passed in as
1079 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1080 * 'hw_addr_iface' must be passed in as a null pointer. */
1082 bridge_pick_datapath_id(struct bridge *br,
1083 const uint8_t bridge_ea[ETH_ADDR_LEN],
1084 struct iface *hw_addr_iface)
1087 * The procedure for choosing a bridge MAC address will, in the most
1088 * ordinary case, also choose a unique MAC that we can use as a datapath
1089 * ID. In some special cases, though, multiple bridges will end up with
1090 * the same MAC address. This is OK for the bridges, but it will confuse
1091 * the OpenFlow controller, because each datapath needs a unique datapath
1094 * Datapath IDs must be unique. It is also very desirable that they be
1095 * stable from one run to the next, so that policy set on a datapath
1098 const char *datapath_id;
1101 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1102 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1106 if (hw_addr_iface) {
1108 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1110 * A bridge whose MAC address is taken from a VLAN network device
1111 * (that is, a network device created with vconfig(8) or similar
1112 * tool) will have the same MAC address as a bridge on the VLAN
1113 * device's physical network device.
1115 * Handle this case by hashing the physical network device MAC
1116 * along with the VLAN identifier.
1118 uint8_t buf[ETH_ADDR_LEN + 2];
1119 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1120 buf[ETH_ADDR_LEN] = vlan >> 8;
1121 buf[ETH_ADDR_LEN + 1] = vlan;
1122 return dpid_from_hash(buf, sizeof buf);
1125 * Assume that this bridge's MAC address is unique, since it
1126 * doesn't fit any of the cases we handle specially.
1131 * A purely internal bridge, that is, one that has no non-virtual
1132 * network devices on it at all, is more difficult because it has no
1133 * natural unique identifier at all.
1135 * When the host is a XenServer, we handle this case by hashing the
1136 * host's UUID with the name of the bridge. Names of bridges are
1137 * persistent across XenServer reboots, although they can be reused if
1138 * an internal network is destroyed and then a new one is later
1139 * created, so this is fairly effective.
1141 * When the host is not a XenServer, we punt by using a random MAC
1142 * address on each run.
1144 const char *host_uuid = xenserver_get_host_uuid();
1146 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1147 dpid = dpid_from_hash(combined, strlen(combined));
1153 return eth_addr_to_uint64(bridge_ea);
1157 dpid_from_hash(const void *data, size_t n)
1159 uint8_t hash[SHA1_DIGEST_SIZE];
1161 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1162 sha1_bytes(data, n, hash);
1163 eth_addr_mark_random(hash);
1164 return eth_addr_to_uint64(hash);
1168 iface_refresh_status(struct iface *iface)
1172 enum netdev_flags flags;
1181 if (!netdev_get_status(iface->netdev, &sh)) {
1183 char **keys, **values;
1185 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1186 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1191 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1194 shash_destroy_free_data(&sh);
1196 error = netdev_get_flags(iface->netdev, &flags);
1198 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1201 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1204 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1206 ovsrec_interface_set_duplex(iface->cfg,
1207 netdev_features_is_full_duplex(current)
1209 /* warning: uint64_t -> int64_t conversion */
1210 bps = netdev_features_to_bps(current);
1211 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1214 ovsrec_interface_set_duplex(iface->cfg, NULL);
1215 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1219 ovsrec_interface_set_link_state(iface->cfg,
1220 netdev_get_carrier(iface->netdev)
1223 error = netdev_get_mtu(iface->netdev, &mtu);
1224 if (!error && mtu != INT_MAX) {
1226 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1229 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1234 iface_refresh_cfm_stats(struct iface *iface)
1238 const struct ovsrec_monitor *mon;
1240 mon = iface->cfg->monitor;
1247 for (i = 0; i < mon->n_remote_mps; i++) {
1248 const struct ovsrec_maintenance_point *mp;
1249 const struct remote_mp *rmp;
1251 mp = mon->remote_mps[i];
1252 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1254 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1257 if (hmap_is_empty(&cfm->x_remote_mps)) {
1258 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1261 struct remote_mp *rmp;
1262 int64_t *x_remote_mps;
1264 length = hmap_count(&cfm->x_remote_mps);
1265 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1268 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1269 x_remote_mps[i++] = rmp->mpid;
1272 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1276 if (hmap_is_empty(&cfm->x_remote_maids)) {
1277 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1280 char **x_remote_maids;
1281 struct remote_maid *rmaid;
1283 length = hmap_count(&cfm->x_remote_maids);
1284 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1287 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1290 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1292 for (j = 0; j < CCM_MAID_LEN; j++) {
1293 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1298 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1300 for (i = 0; i < length; i++) {
1301 free(x_remote_maids[i]);
1303 free(x_remote_maids);
1306 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1310 iface_refresh_stats(struct iface *iface)
1316 static const struct iface_stat iface_stats[] = {
1317 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1318 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1319 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1320 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1321 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1322 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1323 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1324 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1325 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1326 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1327 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1328 { "collisions", offsetof(struct netdev_stats, collisions) },
1330 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1331 const struct iface_stat *s;
1333 char *keys[N_STATS];
1334 int64_t values[N_STATS];
1337 struct netdev_stats stats;
1339 /* Intentionally ignore return value, since errors will set 'stats' to
1340 * all-1s, and we will deal with that correctly below. */
1341 netdev_get_stats(iface->netdev, &stats);
1344 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1345 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1346 if (value != UINT64_MAX) {
1353 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1357 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1359 struct ovsdb_datum datum;
1363 get_system_stats(&stats);
1365 ovsdb_datum_from_shash(&datum, &stats);
1366 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1370 static inline const char *
1371 nx_role_to_str(enum nx_role role)
1376 case NX_ROLE_MASTER:
1381 return "*** INVALID ROLE ***";
1386 bridge_refresh_controller_status(const struct bridge *br)
1389 const struct ovsrec_controller *cfg;
1391 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1393 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1394 struct ofproto_controller_info *cinfo =
1395 shash_find_data(&info, cfg->target);
1398 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1399 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1400 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1401 (char **) cinfo->pairs.values,
1404 ovsrec_controller_set_is_connected(cfg, false);
1405 ovsrec_controller_set_role(cfg, NULL);
1406 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1410 ofproto_free_ofproto_controller_info(&info);
1416 const struct ovsrec_open_vswitch *cfg;
1418 bool datapath_destroyed;
1419 bool database_changed;
1422 /* Let each bridge do the work that it needs to do. */
1423 datapath_destroyed = false;
1424 LIST_FOR_EACH (br, node, &all_bridges) {
1425 int error = bridge_run_one(br);
1427 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1428 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1429 "forcing reconfiguration", br->name);
1430 datapath_destroyed = true;
1434 /* (Re)configure if necessary. */
1435 database_changed = ovsdb_idl_run(idl);
1436 cfg = ovsrec_open_vswitch_first(idl);
1438 /* Re-configure SSL. We do this on every trip through the main loop,
1439 * instead of just when the database changes, because the contents of the
1440 * key and certificate files can change without the database changing.
1442 * We do this before bridge_reconfigure() because that function might
1443 * initiate SSL connections and thus requires SSL to be configured. */
1444 if (cfg && cfg->ssl) {
1445 const struct ovsrec_ssl *ssl = cfg->ssl;
1447 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1448 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1451 if (database_changed || datapath_destroyed) {
1453 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1455 bridge_configure_once(cfg);
1456 bridge_reconfigure(cfg);
1458 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1459 ovsdb_idl_txn_commit(txn);
1460 ovsdb_idl_txn_destroy(txn); /* XXX */
1462 /* We still need to reconfigure to avoid dangling pointers to
1463 * now-destroyed ovsrec structures inside bridge data. */
1464 static const struct ovsrec_open_vswitch null_cfg;
1466 bridge_reconfigure(&null_cfg);
1470 /* Refresh system and interface stats if necessary. */
1471 if (time_msec() >= stats_timer) {
1473 struct ovsdb_idl_txn *txn;
1475 txn = ovsdb_idl_txn_create(idl);
1476 LIST_FOR_EACH (br, node, &all_bridges) {
1479 for (i = 0; i < br->n_ports; i++) {
1480 struct port *port = br->ports[i];
1483 for (j = 0; j < port->n_ifaces; j++) {
1484 struct iface *iface = port->ifaces[j];
1485 iface_refresh_stats(iface);
1486 iface_refresh_cfm_stats(iface);
1487 iface_refresh_status(iface);
1490 bridge_refresh_controller_status(br);
1492 refresh_system_stats(cfg);
1493 ovsdb_idl_txn_commit(txn);
1494 ovsdb_idl_txn_destroy(txn); /* XXX */
1497 stats_timer = time_msec() + STATS_INTERVAL;
1506 LIST_FOR_EACH (br, node, &all_bridges) {
1509 ofproto_wait(br->ofproto);
1510 if (ofproto_has_primary_controller(br->ofproto)) {
1514 mac_learning_wait(br->ml);
1516 for (i = 0; i < br->n_ports; i++) {
1517 port_wait(br->ports[i]);
1520 ovsdb_idl_wait(idl);
1521 poll_timer_wait_until(stats_timer);
1524 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1525 * configuration changes. */
1527 bridge_flush(struct bridge *br)
1529 COVERAGE_INC(bridge_flush);
1531 mac_learning_flush(br->ml);
1534 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1535 * such interface. */
1536 static struct iface *
1537 bridge_get_local_iface(struct bridge *br)
1541 for (i = 0; i < br->n_ports; i++) {
1542 struct port *port = br->ports[i];
1543 for (j = 0; j < port->n_ifaces; j++) {
1544 struct iface *iface = port->ifaces[j];
1545 if (iface->dp_ifidx == ODPP_LOCAL) {
1554 /* Bridge unixctl user interface functions. */
1556 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1557 const char *args, void *aux OVS_UNUSED)
1559 struct ds ds = DS_EMPTY_INITIALIZER;
1560 const struct bridge *br;
1561 const struct mac_entry *e;
1563 br = bridge_lookup(args);
1565 unixctl_command_reply(conn, 501, "no such bridge");
1569 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1570 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1571 if (e->port < 0 || e->port >= br->n_ports) {
1574 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1575 br->ports[e->port]->ifaces[0]->dp_ifidx,
1576 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1578 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1582 /* QoS unixctl user interface functions. */
1584 struct qos_unixctl_show_cbdata {
1586 struct iface *iface;
1590 qos_unixctl_show_cb(unsigned int queue_id,
1591 const struct shash *details,
1594 struct qos_unixctl_show_cbdata *data = aux;
1595 struct ds *ds = data->ds;
1596 struct iface *iface = data->iface;
1597 struct netdev_queue_stats stats;
1598 struct shash_node *node;
1601 ds_put_cstr(ds, "\n");
1603 ds_put_format(ds, "Queue %u:\n", queue_id);
1605 ds_put_cstr(ds, "Default:\n");
1608 SHASH_FOR_EACH (node, details) {
1609 ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data);
1612 error = netdev_get_queue_stats(iface->netdev, queue_id, &stats);
1614 if (stats.tx_packets != UINT64_MAX) {
1615 ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets);
1618 if (stats.tx_bytes != UINT64_MAX) {
1619 ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes);
1622 if (stats.tx_errors != UINT64_MAX) {
1623 ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors);
1626 ds_put_format(ds, "\tFailed to get statistics for queue %u: %s",
1627 queue_id, strerror(error));
1632 qos_unixctl_show(struct unixctl_conn *conn,
1633 const char *args, void *aux OVS_UNUSED)
1635 struct ds ds = DS_EMPTY_INITIALIZER;
1636 struct shash sh = SHASH_INITIALIZER(&sh);
1637 struct iface *iface;
1639 struct shash_node *node;
1640 struct qos_unixctl_show_cbdata data;
1643 iface = iface_find(args);
1645 unixctl_command_reply(conn, 501, "no such interface");
1649 netdev_get_qos(iface->netdev, &type, &sh);
1651 if (*type != '\0') {
1652 ds_put_format(&ds, "QoS: %s %s\n", iface->name, type);
1654 SHASH_FOR_EACH (node, &sh) {
1655 ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data);
1660 error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data);
1663 ds_put_format(&ds, "failed to dump queues: %s", strerror(error));
1665 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1667 ds_put_format(&ds, "QoS not configured on %s\n", iface->name);
1668 unixctl_command_reply(conn, 501, ds_cstr(&ds));
1671 shash_destroy_free_data(&sh);
1675 /* Bridge reconfiguration functions. */
1676 static struct bridge *
1677 bridge_create(const struct ovsrec_bridge *br_cfg)
1682 assert(!bridge_lookup(br_cfg->name));
1683 br = xzalloc(sizeof *br);
1685 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1691 dpif_flow_flush(br->dpif);
1693 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1696 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1698 dpif_delete(br->dpif);
1699 dpif_close(br->dpif);
1704 br->name = xstrdup(br_cfg->name);
1706 br->ml = mac_learning_create();
1707 eth_addr_nicira_random(br->default_ea);
1709 hmap_init(&br->ifaces);
1711 shash_init(&br->port_by_name);
1712 shash_init(&br->iface_by_name);
1716 list_push_back(&all_bridges, &br->node);
1718 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1724 bridge_destroy(struct bridge *br)
1729 while (br->n_ports > 0) {
1730 port_destroy(br->ports[br->n_ports - 1]);
1732 list_remove(&br->node);
1733 error = dpif_delete(br->dpif);
1734 if (error && error != ENOENT) {
1735 VLOG_ERR("failed to delete %s: %s",
1736 dpif_name(br->dpif), strerror(error));
1738 dpif_close(br->dpif);
1739 ofproto_destroy(br->ofproto);
1740 mac_learning_destroy(br->ml);
1741 hmap_destroy(&br->ifaces);
1742 shash_destroy(&br->port_by_name);
1743 shash_destroy(&br->iface_by_name);
1750 static struct bridge *
1751 bridge_lookup(const char *name)
1755 LIST_FOR_EACH (br, node, &all_bridges) {
1756 if (!strcmp(br->name, name)) {
1763 /* Handle requests for a listing of all flows known by the OpenFlow
1764 * stack, including those normally hidden. */
1766 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1767 const char *args, void *aux OVS_UNUSED)
1772 br = bridge_lookup(args);
1774 unixctl_command_reply(conn, 501, "Unknown bridge");
1779 ofproto_get_all_flows(br->ofproto, &results);
1781 unixctl_command_reply(conn, 200, ds_cstr(&results));
1782 ds_destroy(&results);
1785 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1786 * connections and reconnect. If BRIDGE is not specified, then all bridges
1787 * drop their controller connections and reconnect. */
1789 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1790 const char *args, void *aux OVS_UNUSED)
1793 if (args[0] != '\0') {
1794 br = bridge_lookup(args);
1796 unixctl_command_reply(conn, 501, "Unknown bridge");
1799 ofproto_reconnect_controllers(br->ofproto);
1801 LIST_FOR_EACH (br, node, &all_bridges) {
1802 ofproto_reconnect_controllers(br->ofproto);
1805 unixctl_command_reply(conn, 200, NULL);
1809 bridge_run_one(struct bridge *br)
1814 error = ofproto_run1(br->ofproto);
1819 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1821 for (i = 0; i < br->n_ports; i++) {
1822 port_run(br->ports[i]);
1825 error = ofproto_run2(br->ofproto, br->flush);
1832 bridge_get_controllers(const struct bridge *br,
1833 struct ovsrec_controller ***controllersp)
1835 struct ovsrec_controller **controllers;
1836 size_t n_controllers;
1838 controllers = br->cfg->controller;
1839 n_controllers = br->cfg->n_controller;
1841 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1847 *controllersp = controllers;
1849 return n_controllers;
1853 bridge_reconfigure_one(struct bridge *br)
1855 struct shash old_ports, new_ports;
1856 struct svec snoops, old_snoops;
1857 struct shash_node *node;
1858 enum ofproto_fail_mode fail_mode;
1861 /* Collect old ports. */
1862 shash_init(&old_ports);
1863 for (i = 0; i < br->n_ports; i++) {
1864 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1867 /* Collect new ports. */
1868 shash_init(&new_ports);
1869 for (i = 0; i < br->cfg->n_ports; i++) {
1870 const char *name = br->cfg->ports[i]->name;
1871 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1872 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1877 /* If we have a controller, then we need a local port. Complain if the
1878 * user didn't specify one.
1880 * XXX perhaps we should synthesize a port ourselves in this case. */
1881 if (bridge_get_controllers(br, NULL)) {
1882 char local_name[IF_NAMESIZE];
1885 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1886 local_name, sizeof local_name);
1887 if (!error && !shash_find(&new_ports, local_name)) {
1888 VLOG_WARN("bridge %s: controller specified but no local port "
1889 "(port named %s) defined",
1890 br->name, local_name);
1894 /* Get rid of deleted ports.
1895 * Get rid of deleted interfaces on ports that still exist. */
1896 SHASH_FOR_EACH (node, &old_ports) {
1897 struct port *port = node->data;
1898 const struct ovsrec_port *port_cfg;
1900 port_cfg = shash_find_data(&new_ports, node->name);
1904 port_del_ifaces(port, port_cfg);
1908 /* Create new ports.
1909 * Add new interfaces to existing ports.
1910 * Reconfigure existing ports. */
1911 SHASH_FOR_EACH (node, &new_ports) {
1912 struct port *port = shash_find_data(&old_ports, node->name);
1914 port = port_create(br, node->name);
1917 port_reconfigure(port, node->data);
1918 if (!port->n_ifaces) {
1919 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1920 br->name, port->name);
1924 shash_destroy(&old_ports);
1925 shash_destroy(&new_ports);
1927 /* Set the fail-mode */
1928 fail_mode = !br->cfg->fail_mode
1929 || !strcmp(br->cfg->fail_mode, "standalone")
1930 ? OFPROTO_FAIL_STANDALONE
1931 : OFPROTO_FAIL_SECURE;
1932 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1933 && !ofproto_has_primary_controller(br->ofproto)) {
1934 ofproto_flush_flows(br->ofproto);
1936 ofproto_set_fail_mode(br->ofproto, fail_mode);
1938 /* Delete all flows if we're switching from connected to standalone or vice
1939 * versa. (XXX Should we delete all flows if we are switching from one
1940 * controller to another?) */
1942 /* Configure OpenFlow controller connection snooping. */
1944 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1945 ovs_rundir(), br->name));
1946 svec_init(&old_snoops);
1947 ofproto_get_snoops(br->ofproto, &old_snoops);
1948 if (!svec_equal(&snoops, &old_snoops)) {
1949 ofproto_set_snoops(br->ofproto, &snoops);
1951 svec_destroy(&snoops);
1952 svec_destroy(&old_snoops);
1954 mirror_reconfigure(br);
1957 /* Initializes 'oc' appropriately as a management service controller for
1960 * The caller must free oc->target when it is no longer needed. */
1962 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1963 struct ofproto_controller *oc)
1965 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1966 oc->max_backoff = 0;
1967 oc->probe_interval = 60;
1968 oc->band = OFPROTO_OUT_OF_BAND;
1970 oc->burst_limit = 0;
1973 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1975 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1976 struct ofproto_controller *oc)
1978 oc->target = c->target;
1979 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1980 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1981 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1982 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1983 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1984 oc->burst_limit = (c->controller_burst_limit
1985 ? *c->controller_burst_limit : 0);
1988 /* Configures the IP stack for 'br''s local interface properly according to the
1989 * configuration in 'c'. */
1991 bridge_configure_local_iface_netdev(struct bridge *br,
1992 struct ovsrec_controller *c)
1994 struct netdev *netdev;
1995 struct in_addr mask, gateway;
1997 struct iface *local_iface;
2000 /* If there's no local interface or no IP address, give up. */
2001 local_iface = bridge_get_local_iface(br);
2002 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
2006 /* Bring up the local interface. */
2007 netdev = local_iface->netdev;
2008 netdev_turn_flags_on(netdev, NETDEV_UP, true);
2010 /* Configure the IP address and netmask. */
2011 if (!c->local_netmask
2012 || !inet_aton(c->local_netmask, &mask)
2014 mask.s_addr = guess_netmask(ip.s_addr);
2016 if (!netdev_set_in4(netdev, ip, mask)) {
2017 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
2018 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
2021 /* Configure the default gateway. */
2022 if (c->local_gateway
2023 && inet_aton(c->local_gateway, &gateway)
2024 && gateway.s_addr) {
2025 if (!netdev_add_router(netdev, gateway)) {
2026 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
2027 br->name, IP_ARGS(&gateway.s_addr));
2033 bridge_reconfigure_remotes(struct bridge *br,
2034 const struct sockaddr_in *managers,
2037 const char *disable_ib_str, *queue_id_str;
2038 bool disable_in_band = false;
2041 struct ovsrec_controller **controllers;
2042 size_t n_controllers;
2045 struct ofproto_controller *ocs;
2049 /* Check if we should disable in-band control on this bridge. */
2050 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
2051 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
2052 disable_in_band = true;
2055 /* Set OpenFlow queue ID for in-band control. */
2056 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
2057 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
2058 ofproto_set_in_band_queue(br->ofproto, queue_id);
2060 if (disable_in_band) {
2061 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
2063 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
2065 had_primary = ofproto_has_primary_controller(br->ofproto);
2067 n_controllers = bridge_get_controllers(br, &controllers);
2069 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
2072 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
2073 for (i = 0; i < n_controllers; i++) {
2074 struct ovsrec_controller *c = controllers[i];
2076 if (!strncmp(c->target, "punix:", 6)
2077 || !strncmp(c->target, "unix:", 5)) {
2078 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2080 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
2081 * domain sockets and overwriting arbitrary local files. */
2082 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2083 "\"%s\" due to possibility for remote exploit",
2084 dpif_name(br->dpif), c->target);
2088 bridge_configure_local_iface_netdev(br, c);
2089 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2090 if (disable_in_band) {
2091 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2096 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2097 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2100 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2101 ofproto_flush_flows(br->ofproto);
2104 /* If there are no controllers and the bridge is in standalone
2105 * mode, set up a flow that matches every packet and directs
2106 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2107 * switch is in secure mode and we won't pass any traffic until
2108 * a controller has been defined and it tells us to do so. */
2110 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2111 union ofp_action action;
2112 struct cls_rule rule;
2114 memset(&action, 0, sizeof action);
2115 action.type = htons(OFPAT_OUTPUT);
2116 action.output.len = htons(sizeof action);
2117 action.output.port = htons(OFPP_NORMAL);
2118 cls_rule_init_catchall(&rule, 0);
2119 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2124 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2129 for (i = 0; i < br->n_ports; i++) {
2130 struct port *port = br->ports[i];
2131 for (j = 0; j < port->n_ifaces; j++) {
2132 struct iface *iface = port->ifaces[j];
2133 shash_add_once(ifaces, iface->name, iface);
2135 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2136 shash_add_once(ifaces, port->name, NULL);
2141 /* For robustness, in case the administrator moves around datapath ports behind
2142 * our back, we re-check all the datapath port numbers here.
2144 * This function will set the 'dp_ifidx' members of interfaces that have
2145 * disappeared to -1, so only call this function from a context where those
2146 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2147 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2148 * datapath, which doesn't support UINT16_MAX+1 ports. */
2150 bridge_fetch_dp_ifaces(struct bridge *br)
2152 struct dpif_port_dump dump;
2153 struct dpif_port dpif_port;
2156 /* Reset all interface numbers. */
2157 for (i = 0; i < br->n_ports; i++) {
2158 struct port *port = br->ports[i];
2159 for (j = 0; j < port->n_ifaces; j++) {
2160 struct iface *iface = port->ifaces[j];
2161 iface->dp_ifidx = -1;
2164 hmap_clear(&br->ifaces);
2166 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2167 struct iface *iface = iface_lookup(br, dpif_port.name);
2169 if (iface->dp_ifidx >= 0) {
2170 VLOG_WARN("%s reported interface %s twice",
2171 dpif_name(br->dpif), dpif_port.name);
2172 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2173 VLOG_WARN("%s reported interface %"PRIu16" twice",
2174 dpif_name(br->dpif), dpif_port.port_no);
2176 iface->dp_ifidx = dpif_port.port_no;
2177 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2178 hash_int(iface->dp_ifidx, 0));
2181 iface_set_ofport(iface->cfg,
2182 (iface->dp_ifidx >= 0
2183 ? odp_port_to_ofp_port(iface->dp_ifidx)
2189 /* Bridge packet processing functions. */
2192 bond_is_tcp_hash(const struct port *port)
2194 return port->bond_mode == BM_TCP && port->lacp & LACP_NEGOTIATED;
2198 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2200 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2203 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2205 struct flow hash_flow;
2207 memcpy(&hash_flow, flow, sizeof hash_flow);
2208 hash_flow.vlan_tci = 0;
2210 /* The symmetric quality of this hash function is not required, but
2211 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2212 * purposes, so we use it out of convenience. */
2213 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2216 static struct bond_entry *
2217 lookup_bond_entry(const struct port *port, const struct flow *flow,
2220 assert(port->bond_mode != BM_AB);
2222 if (bond_is_tcp_hash(port)) {
2223 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2225 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2230 bond_choose_iface(const struct port *port)
2232 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2233 size_t i, best_down_slave = -1;
2234 long long next_delay_expiration = LLONG_MAX;
2236 for (i = 0; i < port->n_ifaces; i++) {
2237 struct iface *iface = port->ifaces[i];
2239 if (iface->enabled) {
2241 } else if (iface->delay_expires < next_delay_expiration
2242 && (iface->lacp_status & LACP_ATTACHED
2243 || !(port->lacp & LACP_NEGOTIATED))) {
2244 best_down_slave = i;
2245 next_delay_expiration = iface->delay_expires;
2249 if (best_down_slave != -1) {
2250 struct iface *iface = port->ifaces[best_down_slave];
2252 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2253 "since no other interface is up", iface->name,
2254 iface->delay_expires - time_msec());
2255 bond_enable_slave(iface, true);
2258 return best_down_slave;
2262 choose_output_iface(const struct port *port, const struct flow *flow,
2263 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2265 struct iface *iface;
2267 assert(port->n_ifaces);
2268 if (port->n_ifaces == 1) {
2269 iface = port->ifaces[0];
2270 } else if (port->bond_mode == BM_AB) {
2271 if (port->active_iface < 0) {
2272 *tags |= port->no_ifaces_tag;
2275 iface = port->ifaces[port->active_iface];
2277 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2278 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2279 || !port->ifaces[e->iface_idx]->enabled) {
2280 /* XXX select interface properly. The current interface selection
2281 * is only good for testing the rebalancing code. */
2282 e->iface_idx = bond_choose_iface(port);
2283 if (e->iface_idx < 0) {
2284 *tags |= port->no_ifaces_tag;
2287 e->iface_tag = tag_create_random();
2289 *tags |= e->iface_tag;
2290 iface = port->ifaces[e->iface_idx];
2292 *dp_ifidx = iface->dp_ifidx;
2293 *tags |= iface->tag; /* Currently only used for bonding. */
2298 bond_link_status_update(struct iface *iface)
2300 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2301 struct port *port = iface->port;
2302 bool up = iface->up;
2303 int updelay, downdelay;
2305 updelay = port->updelay;
2306 downdelay = port->downdelay;
2308 if (iface->port->lacp & LACP_NEGOTIATED) {
2313 if (iface->port->lacp && up) {
2314 /* The interface is up if it's attached to an aggregator and its
2315 * partner is synchronized. The only exception is defaulted links.
2316 * They are not required to have synchronized partners because they
2317 * have no partners at all. However, they will only be attached if
2318 * negotiations failed on all interfaces in the bond. */
2319 up = iface->lacp_status & LACP_ATTACHED
2320 && (iface->lacp_partner.state & LACP_STATE_SYNC
2321 || iface->lacp_status & LACP_DEFAULTED);
2325 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2326 /* Nothing to do. */
2329 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2330 iface->name, up ? "up" : "down");
2331 if (up == iface->enabled) {
2332 iface->delay_expires = LLONG_MAX;
2333 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2334 iface->name, up ? "disabled" : "enabled");
2335 } else if (up && port->active_iface < 0) {
2336 bond_enable_slave(iface, true);
2338 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2339 "other interface is up", iface->name, updelay);
2342 int delay = up ? updelay : downdelay;
2343 iface->delay_expires = time_msec() + delay;
2346 "interface %s: will be %s if it stays %s for %d ms",
2348 up ? "enabled" : "disabled",
2356 bond_choose_active_iface(struct port *port)
2358 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2360 port->active_iface = bond_choose_iface(port);
2361 port->active_iface_tag = tag_create_random();
2362 if (port->active_iface >= 0) {
2363 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2364 port->name, port->ifaces[port->active_iface]->name);
2366 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2372 bond_enable_slave(struct iface *iface, bool enable)
2374 struct port *port = iface->port;
2375 struct bridge *br = port->bridge;
2377 /* This acts as a recursion check. If the act of disabling a slave
2378 * causes a different slave to be enabled, the flag will allow us to
2379 * skip redundant work when we reenter this function. It must be
2380 * cleared on exit to keep things safe with multiple bonds. */
2381 static bool moving_active_iface = false;
2383 iface->delay_expires = LLONG_MAX;
2384 if (enable == iface->enabled) {
2388 iface->enabled = enable;
2389 if (!iface->enabled) {
2390 VLOG_WARN("interface %s: disabled", iface->name);
2391 ofproto_revalidate(br->ofproto, iface->tag);
2392 if (iface->port_ifidx == port->active_iface) {
2393 ofproto_revalidate(br->ofproto,
2394 port->active_iface_tag);
2396 /* Disabling a slave can lead to another slave being immediately
2397 * enabled if there will be no active slaves but one is waiting
2398 * on an updelay. In this case we do not need to run most of the
2399 * code for the newly enabled slave since there was no period
2400 * without an active slave and it is redundant with the disabling
2402 moving_active_iface = true;
2403 bond_choose_active_iface(port);
2405 bond_send_learning_packets(port);
2407 VLOG_WARN("interface %s: enabled", iface->name);
2408 if (port->active_iface < 0 && !moving_active_iface) {
2409 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2410 bond_choose_active_iface(port);
2411 bond_send_learning_packets(port);
2413 iface->tag = tag_create_random();
2416 moving_active_iface = false;
2419 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2420 * bond interface. */
2422 bond_update_fake_iface_stats(struct port *port)
2424 struct netdev_stats bond_stats;
2425 struct netdev *bond_dev;
2428 memset(&bond_stats, 0, sizeof bond_stats);
2430 for (i = 0; i < port->n_ifaces; i++) {
2431 struct netdev_stats slave_stats;
2433 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2434 /* XXX: We swap the stats here because they are swapped back when
2435 * reported by the internal device. The reason for this is
2436 * internal devices normally represent packets going into the system
2437 * but when used as fake bond device they represent packets leaving
2438 * the system. We really should do this in the internal device
2439 * itself because changing it here reverses the counts from the
2440 * perspective of the switch. However, the internal device doesn't
2441 * know what type of device it represents so we have to do it here
2443 bond_stats.tx_packets += slave_stats.rx_packets;
2444 bond_stats.tx_bytes += slave_stats.rx_bytes;
2445 bond_stats.rx_packets += slave_stats.tx_packets;
2446 bond_stats.rx_bytes += slave_stats.tx_bytes;
2450 if (!netdev_open_default(port->name, &bond_dev)) {
2451 netdev_set_stats(bond_dev, &bond_stats);
2452 netdev_close(bond_dev);
2457 bond_run(struct port *port)
2461 if (port->n_ifaces < 2) {
2465 for (i = 0; i < port->n_ifaces; i++) {
2466 bond_link_status_update(port->ifaces[i]);
2469 for (i = 0; i < port->n_ifaces; i++) {
2470 struct iface *iface = port->ifaces[i];
2471 if (time_msec() >= iface->delay_expires) {
2472 bond_enable_slave(iface, !iface->enabled);
2476 if (port->bond_fake_iface
2477 && time_msec() >= port->bond_next_fake_iface_update) {
2478 bond_update_fake_iface_stats(port);
2479 port->bond_next_fake_iface_update = time_msec() + 1000;
2484 bond_wait(struct port *port)
2488 if (port->n_ifaces < 2) {
2492 for (i = 0; i < port->n_ifaces; i++) {
2493 struct iface *iface = port->ifaces[i];
2494 if (iface->delay_expires != LLONG_MAX) {
2495 poll_timer_wait_until(iface->delay_expires);
2499 if (port->bond_fake_iface) {
2500 poll_timer_wait_until(port->bond_next_fake_iface_update);
2505 set_dst(struct dst *dst, const struct flow *flow,
2506 const struct port *in_port, const struct port *out_port,
2509 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2510 : in_port->vlan >= 0 ? in_port->vlan
2511 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2512 : vlan_tci_to_vid(flow->vlan_tci));
2513 return choose_output_iface(out_port, flow, dst->vlan,
2514 &dst->dp_ifidx, tags);
2518 swap_dst(struct dst *p, struct dst *q)
2520 struct dst tmp = *p;
2525 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2526 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2527 * that we push to the datapath. We could in fact fully sort the array by
2528 * vlan, but in most cases there are at most two different vlan tags so that's
2529 * possibly overkill.) */
2531 partition_dsts(struct dst_set *set, int vlan)
2533 struct dst *first = set->dsts;
2534 struct dst *last = set->dsts + set->n;
2536 while (first != last) {
2538 * - All dsts < first have vlan == 'vlan'.
2539 * - All dsts >= last have vlan != 'vlan'.
2540 * - first < last. */
2541 while (first->vlan == vlan) {
2542 if (++first == last) {
2547 /* Same invariants, plus one additional:
2548 * - first->vlan != vlan.
2550 while (last[-1].vlan != vlan) {
2551 if (--last == first) {
2556 /* Same invariants, plus one additional:
2557 * - last[-1].vlan == vlan.*/
2558 swap_dst(first++, --last);
2563 mirror_mask_ffs(mirror_mask_t mask)
2565 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2570 dst_set_init(struct dst_set *set)
2572 set->dsts = set->builtin;
2574 set->allocated = ARRAY_SIZE(set->builtin);
2578 dst_set_add(struct dst_set *set, const struct dst *dst)
2580 if (set->n >= set->allocated) {
2581 size_t new_allocated;
2582 struct dst *new_dsts;
2584 new_allocated = set->allocated * 2;
2585 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2586 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2590 set->dsts = new_dsts;
2591 set->allocated = new_allocated;
2593 set->dsts[set->n++] = *dst;
2597 dst_set_free(struct dst_set *set)
2599 if (set->dsts != set->builtin) {
2605 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2608 for (i = 0; i < set->n; i++) {
2609 if (set->dsts[i].vlan == test->vlan
2610 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2618 port_trunks_vlan(const struct port *port, uint16_t vlan)
2620 return (port->vlan < 0
2621 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2625 port_includes_vlan(const struct port *port, uint16_t vlan)
2627 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2631 port_is_floodable(const struct port *port)
2635 for (i = 0; i < port->n_ifaces; i++) {
2636 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2637 port->ifaces[i]->dp_ifidx)) {
2645 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2646 const struct port *in_port, const struct port *out_port,
2647 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2649 mirror_mask_t mirrors = in_port->src_mirrors;
2654 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2655 if (flow_vlan == 0) {
2656 flow_vlan = OFP_VLAN_NONE;
2659 if (out_port == FLOOD_PORT) {
2660 for (i = 0; i < br->n_ports; i++) {
2661 struct port *port = br->ports[i];
2663 && port_is_floodable(port)
2664 && port_includes_vlan(port, vlan)
2665 && !port->is_mirror_output_port
2666 && set_dst(&dst, flow, in_port, port, tags)) {
2667 mirrors |= port->dst_mirrors;
2668 dst_set_add(set, &dst);
2671 *nf_output_iface = NF_OUT_FLOOD;
2672 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2673 dst_set_add(set, &dst);
2674 *nf_output_iface = dst.dp_ifidx;
2675 mirrors |= out_port->dst_mirrors;
2679 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2680 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2682 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2683 && !dst_is_duplicate(set, &dst)) {
2684 dst_set_add(set, &dst);
2687 for (i = 0; i < br->n_ports; i++) {
2688 struct port *port = br->ports[i];
2689 if (port_includes_vlan(port, m->out_vlan)
2690 && set_dst(&dst, flow, in_port, port, tags))
2692 if (port->vlan < 0) {
2693 dst.vlan = m->out_vlan;
2695 if (dst_is_duplicate(set, &dst)) {
2699 /* Use the vlan tag on the original flow instead of
2700 * the one passed in the vlan parameter. This ensures
2701 * that we compare the vlan from before any implicit
2702 * tagging tags place. This is necessary because
2703 * dst->vlan is the final vlan, after removing implicit
2705 if (port == in_port && dst.vlan == flow_vlan) {
2706 /* Don't send out input port on same VLAN. */
2709 dst_set_add(set, &dst);
2714 mirrors &= mirrors - 1;
2717 partition_dsts(set, flow_vlan);
2720 static void OVS_UNUSED
2721 print_dsts(const struct dst_set *set)
2725 for (i = 0; i < set->n; i++) {
2726 const struct dst *dst = &set->dsts[i];
2728 printf(">p%"PRIu16, dst->dp_ifidx);
2729 if (dst->vlan != OFP_VLAN_NONE) {
2730 printf("v%"PRIu16, dst->vlan);
2736 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2737 const struct port *in_port, const struct port *out_port,
2738 tag_type *tags, struct ofpbuf *actions,
2739 uint16_t *nf_output_iface)
2746 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2749 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2750 if (cur_vlan == 0) {
2751 cur_vlan = OFP_VLAN_NONE;
2753 for (i = 0; i < set.n; i++) {
2754 const struct dst *dst = &set.dsts[i];
2755 if (dst->vlan != cur_vlan) {
2756 if (dst->vlan == OFP_VLAN_NONE) {
2757 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2760 tci = htons(dst->vlan & VLAN_VID_MASK);
2761 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2762 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2764 cur_vlan = dst->vlan;
2766 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2771 /* Returns the effective vlan of a packet, taking into account both the
2772 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2773 * the packet is untagged and -1 indicates it has an invalid header and
2774 * should be dropped. */
2775 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2776 struct port *in_port, bool have_packet)
2778 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2779 if (in_port->vlan >= 0) {
2781 /* XXX support double tagging? */
2783 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2784 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2785 "packet received on port %s configured with "
2786 "implicit VLAN %"PRIu16,
2787 br->name, vlan, in_port->name, in_port->vlan);
2791 vlan = in_port->vlan;
2793 if (!port_includes_vlan(in_port, vlan)) {
2795 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2796 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2797 "packet received on port %s not configured for "
2799 br->name, vlan, in_port->name, vlan);
2808 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2809 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2810 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2812 is_gratuitous_arp(const struct flow *flow)
2814 return (flow->dl_type == htons(ETH_TYPE_ARP)
2815 && eth_addr_is_broadcast(flow->dl_dst)
2816 && (flow->nw_proto == ARP_OP_REPLY
2817 || (flow->nw_proto == ARP_OP_REQUEST
2818 && flow->nw_src == flow->nw_dst)));
2822 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2823 struct port *in_port)
2825 enum grat_arp_lock_type lock_type;
2828 /* We don't want to learn from gratuitous ARP packets that are reflected
2829 * back over bond slaves so we lock the learning table. */
2830 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2831 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2832 GRAT_ARP_LOCK_CHECK;
2834 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2837 /* The log messages here could actually be useful in debugging,
2838 * so keep the rate limit relatively high. */
2839 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2841 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2842 "on port %s in VLAN %d",
2843 br->name, ETH_ADDR_ARGS(flow->dl_src),
2844 in_port->name, vlan);
2845 ofproto_revalidate(br->ofproto, rev_tag);
2849 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2850 * dropped. Returns true if they may be forwarded, false if they should be
2853 * If 'have_packet' is true, it indicates that the caller is processing a
2854 * received packet. If 'have_packet' is false, then the caller is just
2855 * revalidating an existing flow because configuration has changed. Either
2856 * way, 'have_packet' only affects logging (there is no point in logging errors
2857 * during revalidation).
2859 * Sets '*in_portp' to the input port. This will be a null pointer if
2860 * flow->in_port does not designate a known input port (in which case
2861 * is_admissible() returns false).
2863 * When returning true, sets '*vlanp' to the effective VLAN of the input
2864 * packet, as returned by flow_get_vlan().
2866 * May also add tags to '*tags', although the current implementation only does
2867 * so in one special case.
2870 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2871 tag_type *tags, int *vlanp, struct port **in_portp)
2873 struct iface *in_iface;
2874 struct port *in_port;
2877 /* Find the interface and port structure for the received packet. */
2878 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2880 /* No interface? Something fishy... */
2882 /* Odd. A few possible reasons here:
2884 * - We deleted an interface but there are still a few packets
2885 * queued up from it.
2887 * - Someone externally added an interface (e.g. with "ovs-dpctl
2888 * add-if") that we don't know about.
2890 * - Packet arrived on the local port but the local port is not
2891 * one of our bridge ports.
2893 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2895 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2896 "interface %"PRIu16, br->name, flow->in_port);
2902 *in_portp = in_port = in_iface->port;
2903 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2908 /* Drop frames for reserved multicast addresses. */
2909 if (eth_addr_is_reserved(flow->dl_dst)) {
2913 /* Drop frames on ports reserved for mirroring. */
2914 if (in_port->is_mirror_output_port) {
2916 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2917 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2918 "%s, which is reserved exclusively for mirroring",
2919 br->name, in_port->name);
2924 /* When using LACP, do not accept packets from disabled interfaces. */
2925 if (in_port->lacp & LACP_NEGOTIATED && !in_iface->enabled) {
2929 /* Packets received on non-LACP bonds need special attention to avoid
2931 if (in_port->n_ifaces > 1 && !(in_port->lacp & LACP_NEGOTIATED)) {
2933 bool is_grat_arp_locked;
2935 if (eth_addr_is_multicast(flow->dl_dst)) {
2936 *tags |= in_port->active_iface_tag;
2937 if (in_port->active_iface != in_iface->port_ifidx) {
2938 /* Drop all multicast packets on inactive slaves. */
2943 /* Drop all packets for which we have learned a different input
2944 * port, because we probably sent the packet on one slave and got
2945 * it back on the other. Gratuitous ARP packets are an exception
2946 * to this rule: the host has moved to another switch. The exception
2947 * to the exception is if we locked the learning table to avoid
2948 * reflections on bond slaves. If this is the case, just drop the
2950 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2951 &is_grat_arp_locked);
2952 if (src_idx != -1 && src_idx != in_port->port_idx &&
2953 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2961 /* If the composed actions may be applied to any packet in the given 'flow',
2962 * returns true. Otherwise, the actions should only be applied to 'packet', or
2963 * not at all, if 'packet' was NULL. */
2965 process_flow(struct bridge *br, const struct flow *flow,
2966 const struct ofpbuf *packet, struct ofpbuf *actions,
2967 tag_type *tags, uint16_t *nf_output_iface)
2969 struct port *in_port;
2970 struct port *out_port;
2974 /* Check whether we should drop packets in this flow. */
2975 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2980 /* Learn source MAC (but don't try to learn from revalidation). */
2982 update_learning_table(br, flow, vlan, in_port);
2985 /* Determine output port. */
2986 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2988 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2989 out_port = br->ports[out_port_idx];
2990 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2991 /* If we are revalidating but don't have a learning entry then
2992 * eject the flow. Installing a flow that floods packets opens
2993 * up a window of time where we could learn from a packet reflected
2994 * on a bond and blackhole packets before the learning table is
2995 * updated to reflect the correct port. */
2998 out_port = FLOOD_PORT;
3001 /* Don't send packets out their input ports. */
3002 if (in_port == out_port) {
3008 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
3016 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
3017 struct ofpbuf *actions, tag_type *tags,
3018 uint16_t *nf_output_iface, void *br_)
3020 struct bridge *br = br_;
3022 COVERAGE_INC(bridge_process_flow);
3023 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
3027 bridge_special_ofhook_cb(const struct flow *flow,
3028 const struct ofpbuf *packet, void *br_)
3030 struct iface *iface;
3031 struct bridge *br = br_;
3033 iface = iface_from_dp_ifidx(br, flow->in_port);
3035 if (cfm_should_process_flow(flow)) {
3037 if (iface && packet && iface->cfm) {
3038 COVERAGE_INC(bridge_process_cfm);
3039 cfm_process_heartbeat(iface->cfm, packet);
3042 } else if (flow->dl_type == htons(ETH_TYPE_LACP)) {
3044 if (iface && packet) {
3045 COVERAGE_INC(bridge_process_lacp);
3046 lacp_process_packet(packet, iface);
3055 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
3056 const struct nlattr *actions,
3058 uint64_t n_bytes, void *br_)
3060 struct bridge *br = br_;
3061 const struct nlattr *a;
3062 struct port *in_port;
3067 /* Feed information from the active flows back into the learning table to
3068 * ensure that table is always in sync with what is actually flowing
3069 * through the datapath.
3071 * We test that 'tags' is nonzero to ensure that only flows that include an
3072 * OFPP_NORMAL action are used for learning. This works because
3073 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3074 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3075 update_learning_table(br, flow, vlan, in_port);
3078 /* Account for bond slave utilization. */
3079 if (!br->has_bonded_ports) {
3082 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3083 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3084 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3085 if (out_port && out_port->n_ifaces >= 2 &&
3086 out_port->bond_mode != BM_AB) {
3087 uint16_t vlan = (flow->vlan_tci
3088 ? vlan_tci_to_vid(flow->vlan_tci)
3090 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
3091 e->tx_bytes += n_bytes;
3098 bridge_account_checkpoint_ofhook_cb(void *br_)
3100 struct bridge *br = br_;
3104 if (!br->has_bonded_ports) {
3109 for (i = 0; i < br->n_ports; i++) {
3110 struct port *port = br->ports[i];
3111 if (port->n_ifaces > 1 && port->bond_mode != BM_AB
3112 && now >= port->bond_next_rebalance) {
3113 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3114 bond_rebalance_port(port);
3119 static struct ofhooks bridge_ofhooks = {
3120 bridge_normal_ofhook_cb,
3121 bridge_special_ofhook_cb,
3122 bridge_account_flow_ofhook_cb,
3123 bridge_account_checkpoint_ofhook_cb,
3126 /* LACP functions. */
3129 lacp_process_packet(const struct ofpbuf *packet, struct iface *iface)
3131 const struct lacp_pdu *pdu;
3133 if (!iface->port->lacp) {
3137 pdu = parse_lacp_packet(packet);
3142 iface->lacp_status |= LACP_CURRENT;
3143 iface->lacp_status &= ~(LACP_EXPIRED | LACP_DEFAULTED);
3144 iface->lacp_rx = time_msec() + LACP_SLOW_TIME_RX;
3146 iface->lacp_actor.state = iface_get_lacp_state(iface);
3147 if (memcmp(&iface->lacp_actor, &pdu->partner, sizeof pdu->partner)) {
3151 if (memcmp(&iface->lacp_partner, &pdu->actor, sizeof pdu->actor)) {
3152 iface->port->lacp_need_update = true;
3153 iface->lacp_partner = pdu->actor;
3158 lacp_update_ifaces(struct port *port)
3162 struct lacp_info lead_pri;
3163 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3165 port->lacp_need_update = false;
3166 COVERAGE_INC(bridge_lacp_update);
3172 VLOG_DBG_RL(&rl, "port %s: re-evaluating LACP link status", port->name);
3175 for (i = 0; i < port->n_ifaces; i++) {
3176 struct iface *iface = port->ifaces[i];
3177 struct lacp_info pri;
3179 iface->lacp_status |= LACP_ATTACHED;
3180 ofproto_revalidate(port->bridge->ofproto, iface->tag);
3182 /* Don't allow loopback interfaces to send traffic or lead. */
3183 if (eth_addr_equals(iface->lacp_partner.sys_id,
3184 iface->lacp_actor.sys_id)) {
3185 VLOG_WARN_RL(&rl, "iface %s: Loopback detected. Interface is "
3186 "connected to its own bridge", iface->name);
3187 iface->lacp_status &= ~LACP_ATTACHED;
3191 if (iface->lacp_status & LACP_DEFAULTED) {
3195 iface_get_lacp_priority(iface, &pri);
3197 if (!lead || memcmp(&pri, &lead_pri, sizeof pri) < 0) {
3204 port->lacp &= ~LACP_NEGOTIATED;
3208 port->lacp |= LACP_NEGOTIATED;
3210 for (i = 0; i < port->n_ifaces; i++) {
3211 struct iface *iface = port->ifaces[i];
3213 if (iface->lacp_status & LACP_DEFAULTED
3214 || lead->lacp_partner.key != iface->lacp_partner.key
3215 || !eth_addr_equals(lead->lacp_partner.sys_id,
3216 iface->lacp_partner.sys_id)) {
3217 iface->lacp_status &= ~LACP_ATTACHED;
3223 lacp_iface_may_tx(const struct iface *iface)
3225 return iface->port->lacp & LACP_ACTIVE
3226 || iface->lacp_status & (LACP_CURRENT | LACP_EXPIRED);
3230 lacp_run(struct port *port)
3233 struct ofpbuf packet;
3239 ofpbuf_init(&packet, ETH_HEADER_LEN + LACP_PDU_LEN);
3241 for (i = 0; i < port->n_ifaces; i++) {
3242 struct iface *iface = port->ifaces[i];
3244 if (time_msec() > iface->lacp_rx) {
3245 if (iface->lacp_status & LACP_CURRENT) {
3246 iface_set_lacp_expired(iface);
3247 } else if (iface->lacp_status & LACP_EXPIRED) {
3248 iface_set_lacp_defaulted(iface);
3253 if (port->lacp_need_update) {
3254 lacp_update_ifaces(port);
3257 for (i = 0; i < port->n_ifaces; i++) {
3258 struct iface *iface = port->ifaces[i];
3259 uint8_t ea[ETH_ADDR_LEN];
3262 if (time_msec() < iface->lacp_tx || !lacp_iface_may_tx(iface)) {
3266 error = netdev_get_etheraddr(iface->netdev, ea);
3268 struct lacp_pdu pdu;
3270 iface->lacp_actor.state = iface_get_lacp_state(iface);
3271 compose_lacp_pdu(&iface->lacp_actor, &iface->lacp_partner, &pdu);
3272 compose_lacp_packet(&packet, ea, &pdu);
3273 iface_send_packet(iface, &packet);
3275 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3276 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3277 "(%s)", iface->name, strerror(error));
3280 iface->lacp_tx = time_msec() +
3281 (iface->lacp_partner.state & LACP_STATE_TIME
3283 : LACP_SLOW_TIME_TX);
3285 ofpbuf_uninit(&packet);
3289 lacp_wait(struct port *port)
3297 for (i = 0; i < port->n_ifaces; i++) {
3298 struct iface *iface = port->ifaces[i];
3300 if (lacp_iface_may_tx(iface)) {
3301 poll_timer_wait_until(iface->lacp_tx);
3304 if (iface->lacp_status & (LACP_CURRENT | LACP_EXPIRED)) {
3305 poll_timer_wait_until(iface->lacp_rx);
3310 /* Bonding functions. */
3312 /* Statistics for a single interface on a bonded port, used for load-based
3313 * bond rebalancing. */
3314 struct slave_balance {
3315 struct iface *iface; /* The interface. */
3316 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3318 /* All the "bond_entry"s that are assigned to this interface, in order of
3319 * increasing tx_bytes. */
3320 struct bond_entry **hashes;
3325 bond_mode_to_string(enum bond_mode bm) {
3326 static char *bm_slb = "balance-slb";
3327 static char *bm_ab = "active-backup";
3328 static char *bm_tcp = "balance-tcp";
3331 case BM_SLB: return bm_slb;
3332 case BM_AB: return bm_ab;
3333 case BM_TCP: return bm_tcp;
3340 /* Sorts pointers to pointers to bond_entries in ascending order by the
3341 * interface to which they are assigned, and within a single interface in
3342 * ascending order of bytes transmitted. */
3344 compare_bond_entries(const void *a_, const void *b_)
3346 const struct bond_entry *const *ap = a_;
3347 const struct bond_entry *const *bp = b_;
3348 const struct bond_entry *a = *ap;
3349 const struct bond_entry *b = *bp;
3350 if (a->iface_idx != b->iface_idx) {
3351 return a->iface_idx > b->iface_idx ? 1 : -1;
3352 } else if (a->tx_bytes != b->tx_bytes) {
3353 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3359 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3360 * *descending* order by number of bytes transmitted. */
3362 compare_slave_balance(const void *a_, const void *b_)
3364 const struct slave_balance *a = a_;
3365 const struct slave_balance *b = b_;
3366 if (a->iface->enabled != b->iface->enabled) {
3367 return a->iface->enabled ? -1 : 1;
3368 } else if (a->tx_bytes != b->tx_bytes) {
3369 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3376 swap_bals(struct slave_balance *a, struct slave_balance *b)
3378 struct slave_balance tmp = *a;
3383 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3384 * given that 'p' (and only 'p') might be in the wrong location.
3386 * This function invalidates 'p', since it might now be in a different memory
3389 resort_bals(struct slave_balance *p,
3390 struct slave_balance bals[], size_t n_bals)
3393 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3394 swap_bals(p, p - 1);
3396 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3397 swap_bals(p, p + 1);
3403 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3405 if (VLOG_IS_DBG_ENABLED()) {
3406 struct ds ds = DS_EMPTY_INITIALIZER;
3407 const struct slave_balance *b;
3409 for (b = bals; b < bals + n_bals; b++) {
3413 ds_put_char(&ds, ',');
3415 ds_put_format(&ds, " %s %"PRIu64"kB",
3416 b->iface->name, b->tx_bytes / 1024);
3418 if (!b->iface->enabled) {
3419 ds_put_cstr(&ds, " (disabled)");
3421 if (b->n_hashes > 0) {
3422 ds_put_cstr(&ds, " (");
3423 for (i = 0; i < b->n_hashes; i++) {
3424 const struct bond_entry *e = b->hashes[i];
3426 ds_put_cstr(&ds, " + ");
3428 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3429 e - port->bond_hash, e->tx_bytes / 1024);
3431 ds_put_cstr(&ds, ")");
3434 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3439 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3441 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3444 struct bond_entry *hash = from->hashes[hash_idx];
3445 struct port *port = from->iface->port;
3446 uint64_t delta = hash->tx_bytes;
3448 assert(port->bond_mode != BM_AB);
3450 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3451 "from %s to %s (now carrying %"PRIu64"kB and "
3452 "%"PRIu64"kB load, respectively)",
3453 port->name, delta / 1024, hash - port->bond_hash,
3454 from->iface->name, to->iface->name,
3455 (from->tx_bytes - delta) / 1024,
3456 (to->tx_bytes + delta) / 1024);
3458 /* Delete element from from->hashes.
3460 * We don't bother to add the element to to->hashes because not only would
3461 * it require more work, the only purpose it would be to allow that hash to
3462 * be migrated to another slave in this rebalancing run, and there is no
3463 * point in doing that. */
3464 if (hash_idx == 0) {
3467 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3468 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3472 /* Shift load away from 'from' to 'to'. */
3473 from->tx_bytes -= delta;
3474 to->tx_bytes += delta;
3476 /* Arrange for flows to be revalidated. */
3477 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
3478 hash->iface_idx = to->iface->port_ifidx;
3479 hash->iface_tag = tag_create_random();
3483 bond_rebalance_port(struct port *port)
3485 struct slave_balance *bals;
3487 struct bond_entry *hashes[BOND_MASK + 1];
3488 struct slave_balance *b, *from, *to;
3489 struct bond_entry *e;
3492 assert(port->bond_mode != BM_AB);
3494 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3495 * descending order of tx_bytes, so that bals[0] represents the most
3496 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3499 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3500 * array for each slave_balance structure, we sort our local array of
3501 * hashes in order by slave, so that all of the hashes for a given slave
3502 * become contiguous in memory, and then we point each 'hashes' members of
3503 * a slave_balance structure to the start of a contiguous group. */
3504 n_bals = port->n_ifaces;
3505 bals = xmalloc(n_bals * sizeof *bals);
3506 for (b = bals; b < &bals[n_bals]; b++) {
3507 b->iface = port->ifaces[b - bals];
3512 for (i = 0; i <= BOND_MASK; i++) {
3513 hashes[i] = &port->bond_hash[i];
3515 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3516 for (i = 0; i <= BOND_MASK; i++) {
3518 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3519 b = &bals[e->iface_idx];
3520 b->tx_bytes += e->tx_bytes;
3522 b->hashes = &hashes[i];
3527 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3528 log_bals(bals, n_bals, port);
3530 /* Discard slaves that aren't enabled (which were sorted to the back of the
3531 * array earlier). */
3532 while (!bals[n_bals - 1].iface->enabled) {
3539 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3540 to = &bals[n_bals - 1];
3541 for (from = bals; from < to; ) {
3542 uint64_t overload = from->tx_bytes - to->tx_bytes;
3543 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3544 /* The extra load on 'from' (and all less-loaded slaves), compared
3545 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3546 * it is less than ~1Mbps. No point in rebalancing. */
3548 } else if (from->n_hashes == 1) {
3549 /* 'from' only carries a single MAC hash, so we can't shift any
3550 * load away from it, even though we want to. */
3553 /* 'from' is carrying significantly more load than 'to', and that
3554 * load is split across at least two different hashes. Pick a hash
3555 * to migrate to 'to' (the least-loaded slave), given that doing so
3556 * must decrease the ratio of the load on the two slaves by at
3559 * The sort order we use means that we prefer to shift away the
3560 * smallest hashes instead of the biggest ones. There is little
3561 * reason behind this decision; we could use the opposite sort
3562 * order to shift away big hashes ahead of small ones. */
3565 for (i = 0; i < from->n_hashes; i++) {
3566 double old_ratio, new_ratio;
3567 uint64_t delta = from->hashes[i]->tx_bytes;
3569 if (delta == 0 || from->tx_bytes - delta == 0) {
3570 /* Pointless move. */
3574 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3576 if (to->tx_bytes == 0) {
3577 /* Nothing on the new slave, move it. */
3581 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3582 new_ratio = (double)(from->tx_bytes - delta) /
3583 (to->tx_bytes + delta);
3585 if (new_ratio == 0) {
3586 /* Should already be covered but check to prevent division
3591 if (new_ratio < 1) {
3592 new_ratio = 1 / new_ratio;
3595 if (old_ratio - new_ratio > 0.1) {
3596 /* Would decrease the ratio, move it. */
3600 if (i < from->n_hashes) {
3601 bond_shift_load(from, to, i);
3603 /* If the result of the migration changed the relative order of
3604 * 'from' and 'to' swap them back to maintain invariants. */
3605 if (order_swapped) {
3606 swap_bals(from, to);
3609 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3610 * point to different slave_balance structures. It is only
3611 * valid to do these two operations in a row at all because we
3612 * know that 'from' will not move past 'to' and vice versa. */
3613 resort_bals(from, bals, n_bals);
3614 resort_bals(to, bals, n_bals);
3621 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3622 * historical data to decay to <1% in 7 rebalancing runs. */
3623 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3632 bond_send_learning_packets(struct port *port)
3634 struct bridge *br = port->bridge;
3635 struct mac_entry *e;
3636 struct ofpbuf packet;
3637 int error, n_packets, n_errors;
3639 if (!port->n_ifaces || port->active_iface < 0 || bond_is_tcp_hash(port)) {
3643 ofpbuf_init(&packet, 128);
3644 error = n_packets = n_errors = 0;
3645 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3646 union ofp_action actions[2], *a;
3652 if (e->port == port->port_idx) {
3656 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3658 flow_extract(&packet, 0, ODPP_NONE, &flow);
3660 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3664 /* Compose actions. */
3665 memset(actions, 0, sizeof actions);
3668 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3669 a->vlan_vid.len = htons(sizeof *a);
3670 a->vlan_vid.vlan_vid = htons(e->vlan);
3673 a->output.type = htons(OFPAT_OUTPUT);
3674 a->output.len = htons(sizeof *a);
3675 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3680 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3687 ofpbuf_uninit(&packet);
3690 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3691 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3692 "packets, last error was: %s",
3693 port->name, n_errors, n_packets, strerror(error));
3695 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3696 port->name, n_packets);
3700 /* Bonding unixctl user interface functions. */
3703 bond_unixctl_list(struct unixctl_conn *conn,
3704 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3706 struct ds ds = DS_EMPTY_INITIALIZER;
3707 const struct bridge *br;
3709 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3711 LIST_FOR_EACH (br, node, &all_bridges) {
3714 for (i = 0; i < br->n_ports; i++) {
3715 const struct port *port = br->ports[i];
3716 if (port->n_ifaces > 1) {
3719 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3720 bond_mode_to_string(port->bond_mode));
3721 for (j = 0; j < port->n_ifaces; j++) {
3722 const struct iface *iface = port->ifaces[j];
3724 ds_put_cstr(&ds, ", ");
3726 ds_put_cstr(&ds, iface->name);
3728 ds_put_char(&ds, '\n');
3732 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3736 static struct port *
3737 bond_find(const char *name)
3739 const struct bridge *br;
3741 LIST_FOR_EACH (br, node, &all_bridges) {
3744 for (i = 0; i < br->n_ports; i++) {
3745 struct port *port = br->ports[i];
3746 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3755 ds_put_lacp_state(struct ds *ds, uint8_t state)
3757 if (state & LACP_STATE_ACT) {
3758 ds_put_cstr(ds, "activity ");
3761 if (state & LACP_STATE_TIME) {
3762 ds_put_cstr(ds, "timeout ");
3765 if (state & LACP_STATE_AGG) {
3766 ds_put_cstr(ds, "aggregation ");
3769 if (state & LACP_STATE_SYNC) {
3770 ds_put_cstr(ds, "synchronized ");
3773 if (state & LACP_STATE_COL) {
3774 ds_put_cstr(ds, "collecting ");
3777 if (state & LACP_STATE_DIST) {
3778 ds_put_cstr(ds, "distributing ");
3781 if (state & LACP_STATE_DEF) {
3782 ds_put_cstr(ds, "defaulted ");
3785 if (state & LACP_STATE_EXP) {
3786 ds_put_cstr(ds, "expired ");
3791 bond_unixctl_show(struct unixctl_conn *conn,
3792 const char *args, void *aux OVS_UNUSED)
3794 struct ds ds = DS_EMPTY_INITIALIZER;
3795 const struct port *port;
3798 port = bond_find(args);
3800 unixctl_command_reply(conn, 501, "no such bond");
3804 ds_put_format(&ds, "bond_mode: %s\n",
3805 bond_mode_to_string(port->bond_mode));
3808 ds_put_format(&ds, "lacp: %s\n",
3809 port->lacp & LACP_ACTIVE ? "active" : "passive");
3811 ds_put_cstr(&ds, "lacp: off\n");
3814 if (port->bond_mode != BM_AB) {
3815 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3816 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3820 ds_put_format(&ds, "bond-detect-mode: %s\n",
3821 port->monitor ? "carrier" : "miimon");
3823 if (!port->monitor) {
3824 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3825 port->miimon_interval);
3828 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3829 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3831 if (port->bond_mode != BM_AB) {
3832 ds_put_format(&ds, "next rebalance: %lld ms\n",
3833 port->bond_next_rebalance - time_msec());
3836 for (j = 0; j < port->n_ifaces; j++) {
3837 const struct iface *iface = port->ifaces[j];
3838 struct bond_entry *be;
3842 ds_put_format(&ds, "\nslave %s: %s\n",
3843 iface->name, iface->enabled ? "enabled" : "disabled");
3844 if (j == port->active_iface) {
3845 ds_put_cstr(&ds, "\tactive slave\n");
3847 if (iface->delay_expires != LLONG_MAX) {
3848 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3849 iface->enabled ? "downdelay" : "updelay",
3850 iface->delay_expires - time_msec());
3854 ds_put_cstr(&ds, "\tstatus: ");
3856 if (iface->lacp_status & LACP_CURRENT) {
3857 ds_put_cstr(&ds, "current ");
3860 if (iface->lacp_status & LACP_EXPIRED) {
3861 ds_put_cstr(&ds, "expired ");
3864 if (iface->lacp_status & LACP_DEFAULTED) {
3865 ds_put_cstr(&ds, "defaulted ");
3868 if (iface->lacp_status & LACP_ATTACHED) {
3869 ds_put_cstr(&ds, "attached ");
3872 ds_put_cstr(&ds, "\n");
3874 ds_put_cstr(&ds, "\n\tactor sys_id: ");
3875 ds_put_format(&ds, ETH_ADDR_FMT,
3876 ETH_ADDR_ARGS(iface->lacp_actor.sys_id));
3877 ds_put_cstr(&ds, "\n");
3879 ds_put_format(&ds, "\tactor sys_priority: %u\n",
3880 ntohs(iface->lacp_actor.sys_priority));
3882 ds_put_format(&ds, "\tactor port_id: %u\n",
3883 ntohs(iface->lacp_actor.port_id));
3885 ds_put_format(&ds, "\tactor port_priority: %u\n",
3886 ntohs(iface->lacp_actor.port_priority));
3888 ds_put_format(&ds, "\tactor key: %u\n",
3889 ntohs(iface->lacp_actor.key));
3891 ds_put_cstr(&ds, "\tactor state: ");
3892 ds_put_lacp_state(&ds, iface_get_lacp_state(iface));
3893 ds_put_cstr(&ds, "\n\n");
3895 ds_put_cstr(&ds, "\tpartner sys_id: ");
3896 ds_put_format(&ds, ETH_ADDR_FMT,
3897 ETH_ADDR_ARGS(iface->lacp_partner.sys_id));
3898 ds_put_cstr(&ds, "\n");
3900 ds_put_format(&ds, "\tpartner sys_priority: %u\n",
3901 ntohs(iface->lacp_partner.sys_priority));
3903 ds_put_format(&ds, "\tpartner port_id: %u\n",
3904 ntohs(iface->lacp_partner.port_id));
3906 ds_put_format(&ds, "\tpartner port_priority: %u\n",
3907 ntohs(iface->lacp_partner.port_priority));
3909 ds_put_format(&ds, "\tpartner key: %u\n",
3910 ntohs(iface->lacp_partner.key));
3912 ds_put_cstr(&ds, "\tpartner state: ");
3913 ds_put_lacp_state(&ds, iface->lacp_partner.state);
3914 ds_put_cstr(&ds, "\n");
3917 if (port->bond_mode == BM_AB) {
3922 memset(&flow, 0, sizeof flow);
3923 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3924 int hash = be - port->bond_hash;
3925 struct mac_entry *me;
3927 if (be->iface_idx != j) {
3931 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3932 hash, be->tx_bytes / 1024);
3934 if (port->bond_mode != BM_SLB) {
3939 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3943 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
3944 if (bond_hash_src(me->mac, me->vlan) == hash
3945 && me->port != port->port_idx
3946 && choose_output_iface(port, &flow, me->vlan,
3948 && dp_ifidx == iface->dp_ifidx)
3950 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3951 ETH_ADDR_ARGS(me->mac));
3956 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3961 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3962 void *aux OVS_UNUSED)
3964 char *args = (char *) args_;
3965 char *save_ptr = NULL;
3966 char *bond_s, *hash_s, *slave_s;
3968 struct iface *iface;
3969 struct bond_entry *entry;
3972 bond_s = strtok_r(args, " ", &save_ptr);
3973 hash_s = strtok_r(NULL, " ", &save_ptr);
3974 slave_s = strtok_r(NULL, " ", &save_ptr);
3976 unixctl_command_reply(conn, 501,
3977 "usage: bond/migrate BOND HASH SLAVE");
3981 port = bond_find(bond_s);
3983 unixctl_command_reply(conn, 501, "no such bond");
3987 if (port->bond_mode != BM_SLB) {
3988 unixctl_command_reply(conn, 501, "not an SLB bond");
3992 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3993 hash = atoi(hash_s) & BOND_MASK;
3995 unixctl_command_reply(conn, 501, "bad hash");
3999 iface = port_lookup_iface(port, slave_s);
4001 unixctl_command_reply(conn, 501, "no such slave");
4005 if (!iface->enabled) {
4006 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
4010 entry = &port->bond_hash[hash];
4011 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
4012 entry->iface_idx = iface->port_ifidx;
4013 entry->iface_tag = tag_create_random();
4014 unixctl_command_reply(conn, 200, "migrated");
4018 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
4019 void *aux OVS_UNUSED)
4021 char *args = (char *) args_;
4022 char *save_ptr = NULL;
4023 char *bond_s, *slave_s;
4025 struct iface *iface;
4027 bond_s = strtok_r(args, " ", &save_ptr);
4028 slave_s = strtok_r(NULL, " ", &save_ptr);
4030 unixctl_command_reply(conn, 501,
4031 "usage: bond/set-active-slave BOND SLAVE");
4035 port = bond_find(bond_s);
4037 unixctl_command_reply(conn, 501, "no such bond");
4041 iface = port_lookup_iface(port, slave_s);
4043 unixctl_command_reply(conn, 501, "no such slave");
4047 if (!iface->enabled) {
4048 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
4052 if (port->active_iface != iface->port_ifidx) {
4053 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4054 port->active_iface = iface->port_ifidx;
4055 port->active_iface_tag = tag_create_random();
4056 VLOG_INFO("port %s: active interface is now %s",
4057 port->name, iface->name);
4058 bond_send_learning_packets(port);
4059 unixctl_command_reply(conn, 200, "done");
4061 unixctl_command_reply(conn, 200, "no change");
4066 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
4068 char *args = (char *) args_;
4069 char *save_ptr = NULL;
4070 char *bond_s, *slave_s;
4072 struct iface *iface;
4074 bond_s = strtok_r(args, " ", &save_ptr);
4075 slave_s = strtok_r(NULL, " ", &save_ptr);
4077 unixctl_command_reply(conn, 501,
4078 "usage: bond/enable/disable-slave BOND SLAVE");
4082 port = bond_find(bond_s);
4084 unixctl_command_reply(conn, 501, "no such bond");
4088 iface = port_lookup_iface(port, slave_s);
4090 unixctl_command_reply(conn, 501, "no such slave");
4094 bond_enable_slave(iface, enable);
4095 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
4099 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
4100 void *aux OVS_UNUSED)
4102 enable_slave(conn, args, true);
4106 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
4107 void *aux OVS_UNUSED)
4109 enable_slave(conn, args, false);
4113 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
4114 void *aux OVS_UNUSED)
4116 char *args = (char *) args_;
4117 uint8_t mac[ETH_ADDR_LEN];
4121 char *mac_s, *vlan_s;
4122 char *save_ptr = NULL;
4124 mac_s = strtok_r(args, " ", &save_ptr);
4125 vlan_s = strtok_r(NULL, " ", &save_ptr);
4128 if (sscanf(vlan_s, "%u", &vlan) != 1) {
4129 unixctl_command_reply(conn, 501, "invalid vlan");
4133 vlan = OFP_VLAN_NONE;
4136 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
4137 == ETH_ADDR_SCAN_COUNT) {
4138 hash = bond_hash_src(mac, vlan);
4140 hash_cstr = xasprintf("%u", hash);
4141 unixctl_command_reply(conn, 200, hash_cstr);
4144 unixctl_command_reply(conn, 501, "invalid mac");
4151 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
4152 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
4153 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
4154 unixctl_command_register("bond/set-active-slave",
4155 bond_unixctl_set_active_slave, NULL);
4156 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
4158 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
4160 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
4163 /* Port functions. */
4166 port_run(struct port *port)
4170 if (port->monitor) {
4173 /* Track carrier going up and down on interfaces. */
4174 while (!netdev_monitor_poll(port->monitor, &devname)) {
4175 struct iface *iface;
4177 iface = port_lookup_iface(port, devname);
4179 iface_update_carrier(iface, netdev_get_carrier(iface->netdev));
4183 } else if (time_msec() >= port->miimon_next_update) {
4185 for (i = 0; i < port->n_ifaces; i++) {
4186 struct iface *iface = port->ifaces[i];
4187 iface_update_carrier(iface, netdev_get_miimon(iface->netdev));
4189 port->miimon_next_update = time_msec() + port->miimon_interval;
4195 for (i = 0; i < port->n_ifaces; i++) {
4196 struct iface *iface = port->ifaces[i];
4199 struct ofpbuf *packet = cfm_run(iface->cfm);
4201 iface_send_packet(iface, packet);
4202 ofpbuf_uninit(packet);
4210 port_wait(struct port *port)
4214 if (port->monitor) {
4215 netdev_monitor_poll_wait(port->monitor);
4217 poll_timer_wait_until(port->miimon_next_update);
4223 for (i = 0; i < port->n_ifaces; i++) {
4224 struct iface *iface = port->ifaces[i];
4226 cfm_wait(iface->cfm);
4231 static struct port *
4232 port_create(struct bridge *br, const char *name)
4236 port = xzalloc(sizeof *port);
4238 port->port_idx = br->n_ports;
4240 port->trunks = NULL;
4241 port->name = xstrdup(name);
4242 port->active_iface = -1;
4244 if (br->n_ports >= br->allocated_ports) {
4245 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
4248 br->ports[br->n_ports++] = port;
4249 shash_add_assert(&br->port_by_name, port->name, port);
4251 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
4258 get_port_other_config(const struct ovsrec_port *port, const char *key,
4259 const char *default_value)
4263 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
4265 return value ? value : default_value;
4269 get_interface_other_config(const struct ovsrec_interface *iface,
4270 const char *key, const char *default_value)
4274 value = get_ovsrec_key_value(&iface->header_,
4275 &ovsrec_interface_col_other_config, key);
4276 return value ? value : default_value;
4280 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
4282 struct shash new_ifaces;
4285 /* Collect list of new interfaces. */
4286 shash_init(&new_ifaces);
4287 for (i = 0; i < cfg->n_interfaces; i++) {
4288 const char *name = cfg->interfaces[i]->name;
4289 shash_add_once(&new_ifaces, name, NULL);
4292 /* Get rid of deleted interfaces. */
4293 for (i = 0; i < port->n_ifaces; ) {
4294 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
4295 iface_destroy(port->ifaces[i]);
4301 shash_destroy(&new_ifaces);
4305 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
4307 const char *detect_mode;
4308 struct shash new_ifaces;
4309 long long int next_rebalance, miimon_next_update, lacp_priority;
4310 unsigned long *trunks;
4316 /* Update settings. */
4317 port->updelay = cfg->bond_updelay;
4318 if (port->updelay < 0) {
4321 port->downdelay = cfg->bond_downdelay;
4322 if (port->downdelay < 0) {
4323 port->downdelay = 0;
4325 port->bond_rebalance_interval = atoi(
4326 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
4327 if (port->bond_rebalance_interval < 1000) {
4328 port->bond_rebalance_interval = 1000;
4330 next_rebalance = time_msec() + port->bond_rebalance_interval;
4331 if (port->bond_next_rebalance > next_rebalance) {
4332 port->bond_next_rebalance = next_rebalance;
4335 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
4338 netdev_monitor_destroy(port->monitor);
4339 port->monitor = NULL;
4341 if (strcmp(detect_mode, "miimon")) {
4342 port->monitor = netdev_monitor_create();
4344 if (strcmp(detect_mode, "carrier")) {
4345 VLOG_WARN("port %s: unsupported bond-detect-mode %s, "
4346 "defaulting to carrier", port->name, detect_mode);
4350 port->miimon_interval = atoi(
4351 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4352 if (port->miimon_interval < 100) {
4353 port->miimon_interval = 100;
4355 miimon_next_update = time_msec() + port->miimon_interval;
4356 if (port->miimon_next_update > miimon_next_update) {
4357 port->miimon_next_update = miimon_next_update;
4360 if (!port->cfg->bond_mode ||
4361 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4362 port->bond_mode = BM_SLB;
4363 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4364 port->bond_mode = BM_AB;
4365 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
4366 port->bond_mode = BM_TCP;
4368 port->bond_mode = BM_SLB;
4369 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4370 port->name, port->cfg->bond_mode,
4371 bond_mode_to_string(port->bond_mode));
4374 /* Add new interfaces and update 'cfg' member of existing ones. */
4375 shash_init(&new_ifaces);
4376 for (i = 0; i < cfg->n_interfaces; i++) {
4377 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4378 struct iface *iface;
4380 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
4381 VLOG_WARN("port %s: %s specified twice as port interface",
4382 port->name, if_cfg->name);
4383 iface_set_ofport(if_cfg, -1);
4387 iface = iface_lookup(port->bridge, if_cfg->name);
4389 if (iface->port != port) {
4390 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4392 port->bridge->name, if_cfg->name, iface->port->name);
4395 iface->cfg = if_cfg;
4397 iface = iface_create(port, if_cfg);
4400 /* Determine interface type. The local port always has type
4401 * "internal". Other ports take their type from the database and
4402 * default to "system" if none is specified. */
4403 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4404 : if_cfg->type[0] ? if_cfg->type
4408 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4411 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4412 iface->lacp_priority = UINT16_MAX;
4414 iface->lacp_priority = lacp_priority;
4417 shash_destroy(&new_ifaces);
4420 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4422 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4423 /* Prefer bondable links if unspecified. */
4424 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4426 port->lacp_priority = lacp_priority;
4429 if (!port->cfg->lacp) {
4430 /* XXX when LACP implementation has been sufficiently tested, enable by
4431 * default and make active on bonded ports. */
4433 } else if (!strcmp(port->cfg->lacp, "off")) {
4435 } else if (!strcmp(port->cfg->lacp, "active")) {
4436 port->lacp = LACP_ACTIVE;
4437 } else if (!strcmp(port->cfg->lacp, "passive")) {
4438 port->lacp = LACP_PASSIVE;
4440 VLOG_WARN("port %s: unknown LACP mode %s",
4441 port->name, port->cfg->lacp);
4448 if (port->n_ifaces < 2) {
4450 if (vlan >= 0 && vlan <= 4095) {
4451 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4456 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4457 * they even work as-is. But they have not been tested. */
4458 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4462 if (port->vlan != vlan) {
4464 bridge_flush(port->bridge);
4467 /* Get trunked VLANs. */
4469 if (vlan < 0 && cfg->n_trunks) {
4472 trunks = bitmap_allocate(4096);
4474 for (i = 0; i < cfg->n_trunks; i++) {
4475 int trunk = cfg->trunks[i];
4477 bitmap_set1(trunks, trunk);
4483 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4484 port->name, cfg->n_trunks);
4486 if (n_errors == cfg->n_trunks) {
4487 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4489 bitmap_free(trunks);
4492 } else if (vlan >= 0 && cfg->n_trunks) {
4493 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4497 ? port->trunks != NULL
4498 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4499 bridge_flush(port->bridge);
4501 bitmap_free(port->trunks);
4502 port->trunks = trunks;
4506 port_destroy(struct port *port)
4509 struct bridge *br = port->bridge;
4513 for (i = 0; i < MAX_MIRRORS; i++) {
4514 struct mirror *m = br->mirrors[i];
4515 if (m && m->out_port == port) {
4520 while (port->n_ifaces > 0) {
4521 iface_destroy(port->ifaces[port->n_ifaces - 1]);
4524 shash_find_and_delete_assert(&br->port_by_name, port->name);
4526 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
4527 del->port_idx = port->port_idx;
4529 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4531 netdev_monitor_destroy(port->monitor);
4533 bitmap_free(port->trunks);
4540 static struct port *
4541 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4543 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4544 return iface ? iface->port : NULL;
4547 static struct port *
4548 port_lookup(const struct bridge *br, const char *name)
4550 return shash_find_data(&br->port_by_name, name);
4553 static struct iface *
4554 port_lookup_iface(const struct port *port, const char *name)
4556 struct iface *iface = iface_lookup(port->bridge, name);
4557 return iface && iface->port == port ? iface : NULL;
4561 port_update_lacp(struct port *port)
4566 if (!port->lacp || port->n_ifaces < 1) {
4567 for (i = 0; i < port->n_ifaces; i++) {
4568 iface_set_lacp_defaulted(port->ifaces[i]);
4574 for (i = 0; i < port->n_ifaces; i++) {
4575 struct iface *iface = port->ifaces[i];
4577 if (iface->dp_ifidx <= 0 || iface->dp_ifidx > UINT16_MAX) {
4582 if (iface->dp_ifidx == port->lacp_key) {
4583 key_changed = false;
4588 port->lacp_key = port->ifaces[0]->dp_ifidx;
4591 for (i = 0; i < port->n_ifaces; i++) {
4592 struct iface *iface = port->ifaces[i];
4594 iface->lacp_actor.sys_priority = htons(port->lacp_priority);
4595 memcpy(&iface->lacp_actor.sys_id, port->bridge->ea, ETH_ADDR_LEN);
4597 iface->lacp_actor.port_priority = htons(iface->lacp_priority);
4598 iface->lacp_actor.port_id = htons(iface->dp_ifidx);
4599 iface->lacp_actor.key = htons(port->lacp_key);
4603 port->lacp_need_update = true;
4607 port_update_bonding(struct port *port)
4609 if (port->n_ifaces < 2) {
4610 /* Not a bonded port. */
4611 free(port->bond_hash);
4612 port->bond_hash = NULL;
4613 port->bond_fake_iface = false;
4614 port->active_iface = -1;
4615 port->no_ifaces_tag = 0;
4619 if (port->bond_mode != BM_AB && !port->bond_hash) {
4620 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4621 for (i = 0; i <= BOND_MASK; i++) {
4622 struct bond_entry *e = &port->bond_hash[i];
4626 port->bond_next_rebalance
4627 = time_msec() + port->bond_rebalance_interval;
4628 } else if (port->bond_mode == BM_AB) {
4629 free(port->bond_hash);
4630 port->bond_hash = NULL;
4633 if (!port->no_ifaces_tag) {
4634 port->no_ifaces_tag = tag_create_random();
4637 if (port->active_iface < 0) {
4638 bond_choose_active_iface(port);
4641 port->bond_fake_iface = port->cfg->bond_fake_iface;
4642 if (port->bond_fake_iface) {
4643 port->bond_next_fake_iface_update = time_msec();
4649 /* Interface functions. */
4652 iface_set_lacp_defaulted(struct iface *iface)
4654 memset(&iface->lacp_partner, 0, sizeof iface->lacp_partner);
4656 iface->lacp_status |= LACP_DEFAULTED;
4657 iface->lacp_status &= ~(LACP_CURRENT | LACP_EXPIRED);
4659 iface->port->lacp_need_update = true;
4663 iface_set_lacp_expired(struct iface *iface)
4665 iface->lacp_status &= ~LACP_CURRENT;
4666 iface->lacp_status |= LACP_EXPIRED;
4667 iface->lacp_partner.state |= LACP_STATE_TIME;
4668 iface->lacp_partner.state &= ~LACP_STATE_SYNC;
4670 iface->lacp_rx = time_msec() + LACP_FAST_TIME_RX;
4675 iface_get_lacp_state(const struct iface *iface)
4679 if (iface->port->lacp & LACP_ACTIVE) {
4680 state |= LACP_STATE_ACT;
4683 if (iface->lacp_status & LACP_ATTACHED) {
4684 state |= LACP_STATE_SYNC;
4687 if (iface->lacp_status & LACP_DEFAULTED) {
4688 state |= LACP_STATE_DEF;
4691 if (iface->lacp_status & LACP_EXPIRED) {
4692 state |= LACP_STATE_EXP;
4695 if (iface->port->n_ifaces > 1) {
4696 state |= LACP_STATE_AGG;
4699 if (iface->enabled) {
4700 state |= LACP_STATE_COL | LACP_STATE_DIST;
4706 /* Given 'iface', populates 'priority' with data representing its LACP link
4707 * priority. If two priority objects populated by this function are compared
4708 * using memcmp, the higher priority link will be less than the lower priority
4711 iface_get_lacp_priority(struct iface *iface, struct lacp_info *priority)
4713 uint16_t partner_priority, actor_priority;
4715 /* Choose the lacp_info of the higher priority system by comparing their
4716 * system priorities and mac addresses. */
4717 actor_priority = ntohs(iface->lacp_actor.sys_priority);
4718 partner_priority = ntohs(iface->lacp_partner.sys_priority);
4719 if (actor_priority < partner_priority) {
4720 *priority = iface->lacp_actor;
4721 } else if (partner_priority < actor_priority) {
4722 *priority = iface->lacp_partner;
4723 } else if (eth_addr_compare_3way(iface->lacp_actor.sys_id,
4724 iface->lacp_partner.sys_id) < 0) {
4725 *priority = iface->lacp_actor;
4727 *priority = iface->lacp_partner;
4730 /* Key and state are not used in priority comparisons. */
4732 priority->state = 0;
4736 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
4739 union ofp_action action;
4741 memset(&action, 0, sizeof action);
4742 action.output.type = htons(OFPAT_OUTPUT);
4743 action.output.len = htons(sizeof action);
4744 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
4746 flow_extract(packet, 0, ODPP_NONE, &flow);
4748 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
4750 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4751 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
4755 static struct iface *
4756 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4758 struct bridge *br = port->bridge;
4759 struct iface *iface;
4760 char *name = if_cfg->name;
4762 iface = xzalloc(sizeof *iface);
4764 iface->port_ifidx = port->n_ifaces;
4765 iface->name = xstrdup(name);
4766 iface->dp_ifidx = -1;
4767 iface->tag = tag_create_random();
4768 iface->delay_expires = LLONG_MAX;
4769 iface->netdev = NULL;
4770 iface->cfg = if_cfg;
4771 iface_set_lacp_defaulted(iface);
4773 if (port->lacp & LACP_ACTIVE) {
4774 iface_set_lacp_expired(iface);
4777 shash_add_assert(&br->iface_by_name, iface->name, iface);
4779 if (port->n_ifaces >= port->allocated_ifaces) {
4780 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
4781 sizeof *port->ifaces);
4783 port->ifaces[port->n_ifaces++] = iface;
4784 if (port->n_ifaces > 1) {
4785 br->has_bonded_ports = true;
4788 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4796 iface_destroy(struct iface *iface)
4799 struct port *port = iface->port;
4800 struct bridge *br = port->bridge;
4801 bool del_active = port->active_iface == iface->port_ifidx;
4804 if (port->monitor && iface->netdev) {
4805 netdev_monitor_remove(port->monitor, iface->netdev);
4808 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4810 if (iface->dp_ifidx >= 0) {
4811 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4814 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
4815 del->port_ifidx = iface->port_ifidx;
4817 netdev_close(iface->netdev);
4820 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4821 bond_choose_active_iface(port);
4822 bond_send_learning_packets(port);
4825 cfm_destroy(iface->cfm);
4830 bridge_flush(port->bridge);
4834 static struct iface *
4835 iface_lookup(const struct bridge *br, const char *name)
4837 return shash_find_data(&br->iface_by_name, name);
4840 static struct iface *
4841 iface_find(const char *name)
4843 const struct bridge *br;
4845 LIST_FOR_EACH (br, node, &all_bridges) {
4846 struct iface *iface = iface_lookup(br, name);
4855 static struct iface *
4856 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4858 struct iface *iface;
4860 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4861 hash_int(dp_ifidx, 0), &br->ifaces) {
4862 if (iface->dp_ifidx == dp_ifidx) {
4869 /* Set Ethernet address of 'iface', if one is specified in the configuration
4872 iface_set_mac(struct iface *iface)
4874 uint8_t ea[ETH_ADDR_LEN];
4876 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4877 if (eth_addr_is_multicast(ea)) {
4878 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4880 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4881 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4882 iface->name, iface->name);
4884 int error = netdev_set_etheraddr(iface->netdev, ea);
4886 VLOG_ERR("interface %s: setting MAC failed (%s)",
4887 iface->name, strerror(error));
4893 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4895 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4898 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4902 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4904 * The value strings in '*shash' are taken directly from values[], not copied,
4905 * so the caller should not modify or free them. */
4907 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4908 struct shash *shash)
4913 for (i = 0; i < n; i++) {
4914 shash_add(shash, keys[i], values[i]);
4918 /* Creates 'keys' and 'values' arrays from 'shash'.
4920 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4921 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4922 * are populated with with strings taken directly from 'shash' and thus have
4923 * the same ownership of the key-value pairs in shash.
4926 shash_to_ovs_idl_map(struct shash *shash,
4927 char ***keys, char ***values, size_t *n)
4931 struct shash_node *sn;
4933 count = shash_count(shash);
4935 k = xmalloc(count * sizeof *k);
4936 v = xmalloc(count * sizeof *v);
4939 SHASH_FOR_EACH(sn, shash) {
4950 struct iface_delete_queues_cbdata {
4951 struct netdev *netdev;
4952 const struct ovsdb_datum *queues;
4956 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4958 union ovsdb_atom atom;
4960 atom.integer = target;
4961 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4965 iface_delete_queues(unsigned int queue_id,
4966 const struct shash *details OVS_UNUSED, void *cbdata_)
4968 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4970 if (!queue_ids_include(cbdata->queues, queue_id)) {
4971 netdev_delete_queue(cbdata->netdev, queue_id);
4976 iface_update_carrier(struct iface *iface, bool carrier)
4978 if (carrier == iface->up) {
4982 iface->up = carrier;
4985 if (iface->lacp_status & LACP_CURRENT) {
4986 iface_set_lacp_expired(iface);
4991 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4993 if (!qos || qos->type[0] == '\0') {
4994 netdev_set_qos(iface->netdev, NULL, NULL);
4996 struct iface_delete_queues_cbdata cbdata;
4997 struct shash details;
5000 /* Configure top-level Qos for 'iface'. */
5001 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
5002 qos->n_other_config, &details);
5003 netdev_set_qos(iface->netdev, qos->type, &details);
5004 shash_destroy(&details);
5006 /* Deconfigure queues that were deleted. */
5007 cbdata.netdev = iface->netdev;
5008 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
5010 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
5012 /* Configure queues for 'iface'. */
5013 for (i = 0; i < qos->n_queues; i++) {
5014 const struct ovsrec_queue *queue = qos->value_queues[i];
5015 unsigned int queue_id = qos->key_queues[i];
5017 shash_from_ovs_idl_map(queue->key_other_config,
5018 queue->value_other_config,
5019 queue->n_other_config, &details);
5020 netdev_set_queue(iface->netdev, queue_id, &details);
5021 shash_destroy(&details);
5027 iface_update_cfm(struct iface *iface)
5031 uint16_t *remote_mps;
5032 struct ovsrec_monitor *mon;
5033 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
5035 mon = iface->cfg->monitor;
5038 cfm_destroy(iface->cfm);
5043 if (netdev_get_etheraddr(iface->netdev, ea)) {
5044 VLOG_WARN("interface %s: Failed to get ethernet address. "
5045 "Skipping Monitor.", iface->name);
5049 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
5050 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
5055 iface->cfm = cfm_create();
5059 cfm->mpid = mon->mpid;
5060 cfm->interval = mon->interval ? *mon->interval : 1000;
5062 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
5063 memcpy(cfm->maid, maid, sizeof cfm->maid);
5065 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
5066 for(i = 0; i < mon->n_remote_mps; i++) {
5067 remote_mps[i] = mon->remote_mps[i]->mpid;
5069 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
5072 if (!cfm_configure(iface->cfm)) {
5073 cfm_destroy(iface->cfm);
5078 /* Port mirroring. */
5080 static struct mirror *
5081 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
5085 for (i = 0; i < MAX_MIRRORS; i++) {
5086 struct mirror *m = br->mirrors[i];
5087 if (m && uuid_equals(uuid, &m->uuid)) {
5095 mirror_reconfigure(struct bridge *br)
5097 unsigned long *rspan_vlans;
5100 /* Get rid of deleted mirrors. */
5101 for (i = 0; i < MAX_MIRRORS; i++) {
5102 struct mirror *m = br->mirrors[i];
5104 const struct ovsdb_datum *mc;
5105 union ovsdb_atom atom;
5107 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
5108 atom.uuid = br->mirrors[i]->uuid;
5109 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
5115 /* Add new mirrors and reconfigure existing ones. */
5116 for (i = 0; i < br->cfg->n_mirrors; i++) {
5117 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
5118 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
5120 mirror_reconfigure_one(m, cfg);
5122 mirror_create(br, cfg);
5126 /* Update port reserved status. */
5127 for (i = 0; i < br->n_ports; i++) {
5128 br->ports[i]->is_mirror_output_port = false;
5130 for (i = 0; i < MAX_MIRRORS; i++) {
5131 struct mirror *m = br->mirrors[i];
5132 if (m && m->out_port) {
5133 m->out_port->is_mirror_output_port = true;
5137 /* Update flooded vlans (for RSPAN). */
5139 if (br->cfg->n_flood_vlans) {
5140 rspan_vlans = bitmap_allocate(4096);
5142 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
5143 int64_t vlan = br->cfg->flood_vlans[i];
5144 if (vlan >= 0 && vlan < 4096) {
5145 bitmap_set1(rspan_vlans, vlan);
5146 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
5149 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
5154 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
5160 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
5165 for (i = 0; ; i++) {
5166 if (i >= MAX_MIRRORS) {
5167 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
5168 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
5171 if (!br->mirrors[i]) {
5176 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
5179 br->mirrors[i] = m = xzalloc(sizeof *m);
5182 m->name = xstrdup(cfg->name);
5183 shash_init(&m->src_ports);
5184 shash_init(&m->dst_ports);
5190 mirror_reconfigure_one(m, cfg);
5194 mirror_destroy(struct mirror *m)
5197 struct bridge *br = m->bridge;
5200 for (i = 0; i < br->n_ports; i++) {
5201 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
5202 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
5205 shash_destroy(&m->src_ports);
5206 shash_destroy(&m->dst_ports);
5209 m->bridge->mirrors[m->idx] = NULL;
5218 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
5219 struct shash *names)
5223 for (i = 0; i < n_ports; i++) {
5224 const char *name = ports[i]->name;
5225 if (port_lookup(m->bridge, name)) {
5226 shash_add_once(names, name, NULL);
5228 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
5229 "port %s", m->bridge->name, m->name, name);
5235 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
5241 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
5243 for (i = 0; i < cfg->n_select_vlan; i++) {
5244 int64_t vlan = cfg->select_vlan[i];
5245 if (vlan < 0 || vlan > 4095) {
5246 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
5247 m->bridge->name, m->name, vlan);
5249 (*vlans)[n_vlans++] = vlan;
5256 vlan_is_mirrored(const struct mirror *m, int vlan)
5260 for (i = 0; i < m->n_vlans; i++) {
5261 if (m->vlans[i] == vlan) {
5269 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
5273 for (i = 0; i < m->n_vlans; i++) {
5274 if (port_trunks_vlan(p, m->vlans[i])) {
5282 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
5284 struct shash src_ports, dst_ports;
5285 mirror_mask_t mirror_bit;
5286 struct port *out_port;
5293 if (strcmp(cfg->name, m->name)) {
5295 m->name = xstrdup(cfg->name);
5298 /* Get output port. */
5299 if (cfg->output_port) {
5300 out_port = port_lookup(m->bridge, cfg->output_port->name);
5302 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
5303 m->bridge->name, m->name);
5309 if (cfg->output_vlan) {
5310 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
5311 "output vlan; ignoring output vlan",
5312 m->bridge->name, m->name);
5314 } else if (cfg->output_vlan) {
5316 out_vlan = *cfg->output_vlan;
5318 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
5319 m->bridge->name, m->name);
5324 shash_init(&src_ports);
5325 shash_init(&dst_ports);
5326 if (cfg->select_all) {
5327 for (i = 0; i < m->bridge->n_ports; i++) {
5328 const char *name = m->bridge->ports[i]->name;
5329 shash_add_once(&src_ports, name, NULL);
5330 shash_add_once(&dst_ports, name, NULL);
5335 /* Get ports, and drop duplicates and ports that don't exist. */
5336 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
5338 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
5341 /* Get all the vlans, and drop duplicate and invalid vlans. */
5342 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
5345 /* Update mirror data. */
5346 if (!shash_equal_keys(&m->src_ports, &src_ports)
5347 || !shash_equal_keys(&m->dst_ports, &dst_ports)
5348 || m->n_vlans != n_vlans
5349 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
5350 || m->out_port != out_port
5351 || m->out_vlan != out_vlan) {
5352 bridge_flush(m->bridge);
5354 shash_swap(&m->src_ports, &src_ports);
5355 shash_swap(&m->dst_ports, &dst_ports);
5358 m->n_vlans = n_vlans;
5359 m->out_port = out_port;
5360 m->out_vlan = out_vlan;
5363 mirror_bit = MIRROR_MASK_C(1) << m->idx;
5364 for (i = 0; i < m->bridge->n_ports; i++) {
5365 struct port *port = m->bridge->ports[i];
5367 if (shash_find(&m->src_ports, port->name)
5370 ? port_trunks_any_mirrored_vlan(m, port)
5371 : vlan_is_mirrored(m, port->vlan)))) {
5372 port->src_mirrors |= mirror_bit;
5374 port->src_mirrors &= ~mirror_bit;
5377 if (shash_find(&m->dst_ports, port->name)) {
5378 port->dst_mirrors |= mirror_bit;
5380 port->dst_mirrors &= ~mirror_bit;
5385 shash_destroy(&src_ports);
5386 shash_destroy(&dst_ports);