1 /* Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
18 #include "byte-order.h"
21 #include <arpa/inet.h>
24 #include <sys/socket.h>
26 #include <openflow/openflow.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
36 #include "classifier.h"
41 #include "dynamic-string.h"
47 #include "mac-learning.h"
51 #include "ofp-print.h"
53 #include "ofproto/netflow.h"
54 #include "ofproto/ofproto.h"
55 #include "ovsdb-data.h"
57 #include "poll-loop.h"
61 #include "socket-util.h"
62 #include "stream-ssl.h"
64 #include "system-stats.h"
69 #include "vswitchd/vswitch-idl.h"
70 #include "xenserver.h"
72 #include "sflow_api.h"
74 VLOG_DEFINE_THIS_MODULE(bridge);
76 COVERAGE_DEFINE(bridge_flush);
77 COVERAGE_DEFINE(bridge_process_flow);
78 COVERAGE_DEFINE(bridge_process_cfm);
79 COVERAGE_DEFINE(bridge_process_lacp);
80 COVERAGE_DEFINE(bridge_reconfigure);
81 COVERAGE_DEFINE(bridge_lacp_update);
89 struct dst builtin[32];
94 static void dst_set_init(struct dst_set *);
95 static void dst_set_add(struct dst_set *, const struct dst *);
96 static void dst_set_free(struct dst_set *);
99 LACP_CURRENT = 0x01, /* Current State. */
100 LACP_EXPIRED = 0x02, /* Expired State. */
101 LACP_DEFAULTED = 0x04, /* Partner is defaulted. */
102 LACP_ATTACHED = 0x08, /* Attached. Interface may be choosen for flows. */
106 /* These members are always valid. */
107 struct port *port; /* Containing port. */
108 size_t port_ifidx; /* Index within containing port. */
109 char *name; /* Host network device name. */
110 tag_type tag; /* Tag associated with this interface. */
111 long long delay_expires; /* Time after which 'enabled' may change. */
113 /* These members are valid only after bridge_reconfigure() causes them to
115 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
116 int dp_ifidx; /* Index within kernel datapath. */
117 struct netdev *netdev; /* Network device. */
118 bool enabled; /* May be chosen for flows? */
119 bool up; /* Is the interface up? */
120 const char *type; /* Usually same as cfg->type. */
121 struct cfm *cfm; /* Connectivity Fault Management */
122 const struct ovsrec_interface *cfg;
124 /* LACP information. */
125 enum lacp_status lacp_status; /* LACP status. */
126 uint16_t lacp_priority; /* LACP port priority. */
127 struct lacp_info lacp_actor; /* LACP actor information. */
128 struct lacp_info lacp_partner; /* LACP partner information. */
129 long long int lacp_tx; /* Next LACP message transmission time. */
130 long long int lacp_rx; /* Next LACP message receive time. */
133 #define BOND_MASK 0xff
135 int iface_idx; /* Index of assigned iface, or -1 if none. */
136 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
137 tag_type iface_tag; /* Tag associated with iface_idx. */
141 BM_TCP, /* Transport Layer Load Balance. */
142 BM_SLB, /* Source Load Balance. */
143 BM_AB /* Active Backup. */
146 #define MAX_MIRRORS 32
147 typedef uint32_t mirror_mask_t;
148 #define MIRROR_MASK_C(X) UINT32_C(X)
149 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
151 struct bridge *bridge;
154 struct uuid uuid; /* UUID of this "mirror" record in database. */
156 /* Selection criteria. */
157 struct shash src_ports; /* Name is port name; data is always NULL. */
158 struct shash dst_ports; /* Name is port name; data is always NULL. */
163 struct port *out_port;
167 /* Flags for a port's lacp member. */
168 #define LACP_ACTIVE 0x01 /* LACP is in active mode. */
169 #define LACP_PASSIVE 0x02 /* LACP is in passive mode. */
170 #define LACP_NEGOTIATED 0x04 /* LACP has successfully negotiated. */
172 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
174 struct bridge *bridge;
176 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
177 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
178 * NULL if all VLANs are trunked. */
179 const struct ovsrec_port *cfg;
182 /* An ordinary bridge port has 1 interface.
183 * A bridge port for bonding has at least 2 interfaces. */
184 struct iface **ifaces;
185 size_t n_ifaces, allocated_ifaces;
188 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
189 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
190 tag_type active_iface_tag; /* Tag for bcast flows. */
191 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
192 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
193 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
194 bool miimon; /* Use miimon instead of carrier? */
195 long long int bond_miimon_interval; /* Miimon status refresh interval. */
196 long long int bond_miimon_next_update; /* Time of next miimon update. */
197 long long int bond_next_fake_iface_update; /* Time of next update. */
198 struct netdev_monitor *monitor; /* Tracks carrier up/down status. */
200 /* LACP information. */
201 int lacp; /* LACP status flags. 0 if LACP is off. */
202 uint16_t lacp_key; /* LACP aggregation key. */
203 uint16_t lacp_priority; /* LACP system priority. */
204 bool lacp_need_update; /* Need to update attached interfaces? */
206 /* SLB specific bonding info. */
207 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
208 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
209 long long int bond_next_rebalance; /* Next rebalancing time. */
211 /* Port mirroring info. */
212 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
213 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
214 bool is_mirror_output_port; /* Does port mirroring send frames here? */
218 struct list node; /* Node in global list of bridges. */
219 char *name; /* User-specified arbitrary name. */
220 struct mac_learning *ml; /* MAC learning table. */
221 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
222 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
223 const struct ovsrec_bridge *cfg;
225 /* OpenFlow switch processing. */
226 struct ofproto *ofproto; /* OpenFlow switch. */
228 /* Kernel datapath information. */
229 struct dpif *dpif; /* Datapath. */
230 struct hmap ifaces; /* Contains "struct iface"s. */
234 size_t n_ports, allocated_ports;
235 struct shash iface_by_name; /* "struct iface"s indexed by name. */
236 struct shash port_by_name; /* "struct port"s indexed by name. */
239 bool has_bonded_ports;
244 /* Port mirroring. */
245 struct mirror *mirrors[MAX_MIRRORS];
248 /* List of all bridges. */
249 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
251 /* OVSDB IDL used to obtain configuration. */
252 static struct ovsdb_idl *idl;
254 /* Each time this timer expires, the bridge fetches systems and interface
255 * statistics and pushes them into the database. */
256 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
257 static long long int stats_timer = LLONG_MIN;
259 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
260 static void bridge_destroy(struct bridge *);
261 static struct bridge *bridge_lookup(const char *name);
262 static unixctl_cb_func bridge_unixctl_dump_flows;
263 static unixctl_cb_func bridge_unixctl_reconnect;
264 static int bridge_run_one(struct bridge *);
265 static size_t bridge_get_controllers(const struct bridge *br,
266 struct ovsrec_controller ***controllersp);
267 static void bridge_reconfigure_one(struct bridge *);
268 static void bridge_reconfigure_remotes(struct bridge *,
269 const struct sockaddr_in *managers,
271 static void bridge_reconfigure_remotes_late(struct bridge *);
272 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
273 static void bridge_fetch_dp_ifaces(struct bridge *);
274 static void bridge_flush(struct bridge *);
275 static void bridge_pick_local_hw_addr(struct bridge *,
276 uint8_t ea[ETH_ADDR_LEN],
277 struct iface **hw_addr_iface);
278 static uint64_t bridge_pick_datapath_id(struct bridge *,
279 const uint8_t bridge_ea[ETH_ADDR_LEN],
280 struct iface *hw_addr_iface);
281 static struct iface *bridge_get_local_iface(struct bridge *);
282 static uint64_t dpid_from_hash(const void *, size_t nbytes);
284 static unixctl_cb_func bridge_unixctl_fdb_show;
285 static unixctl_cb_func qos_unixctl_show;
287 static void lacp_run(struct port *);
288 static void lacp_wait(struct port *);
289 static void lacp_process_packet(const struct ofpbuf *, struct iface *);
291 static void bond_init(void);
292 static void bond_run(struct port *);
293 static void bond_wait(struct port *);
294 static void bond_rebalance_port(struct port *);
295 static void bond_send_learning_packets(struct port *);
296 static void bond_enable_slave(struct iface *iface, bool enable);
298 static void port_run(struct port *);
299 static void port_wait(struct port *);
300 static struct port *port_create(struct bridge *, const char *name);
301 static void port_reconfigure(struct port *, const struct ovsrec_port *);
302 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
303 static void port_destroy(struct port *);
304 static struct port *port_lookup(const struct bridge *, const char *name);
305 static struct iface *port_lookup_iface(const struct port *, const char *name);
306 static struct port *port_from_dp_ifidx(const struct bridge *,
308 static void port_update_bonding(struct port *);
309 static void port_update_lacp(struct port *);
311 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
312 static void mirror_destroy(struct mirror *);
313 static void mirror_reconfigure(struct bridge *);
314 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
315 static bool vlan_is_mirrored(const struct mirror *, int vlan);
317 static struct iface *iface_create(struct port *port,
318 const struct ovsrec_interface *if_cfg);
319 static void iface_destroy(struct iface *);
320 static struct iface *iface_lookup(const struct bridge *, const char *name);
321 static struct iface *iface_find(const char *name);
322 static struct iface *iface_from_dp_ifidx(const struct bridge *,
324 static void iface_set_mac(struct iface *);
325 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
326 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
327 static void iface_update_cfm(struct iface *);
328 static void iface_refresh_cfm_stats(struct iface *iface);
329 static void iface_send_packet(struct iface *, struct ofpbuf *packet);
330 static uint8_t iface_get_lacp_state(const struct iface *);
331 static void iface_get_lacp_priority(struct iface *, struct lacp_info *);
332 static void iface_set_lacp_defaulted(struct iface *);
333 static void iface_set_lacp_expired(struct iface *);
335 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
337 static void shash_to_ovs_idl_map(struct shash *,
338 char ***keys, char ***values, size_t *n);
341 /* Hooks into ofproto processing. */
342 static struct ofhooks bridge_ofhooks;
344 /* Public functions. */
346 /* Initializes the bridge module, configuring it to obtain its configuration
347 * from an OVSDB server accessed over 'remote', which should be a string in a
348 * form acceptable to ovsdb_idl_create(). */
350 bridge_init(const char *remote)
352 /* Create connection to database. */
353 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
355 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
356 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
357 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
359 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
361 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
362 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
364 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
365 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
366 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
368 /* Register unixctl commands. */
369 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
370 unixctl_command_register("qos/show", qos_unixctl_show, NULL);
371 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
373 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
381 struct bridge *br, *next_br;
383 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
386 ovsdb_idl_destroy(idl);
389 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
390 * but for which the ovs-vswitchd configuration 'cfg' is required. */
392 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
394 static bool already_configured_once;
395 struct svec bridge_names;
396 struct svec dpif_names, dpif_types;
399 /* Only do this once per ovs-vswitchd run. */
400 if (already_configured_once) {
403 already_configured_once = true;
405 stats_timer = time_msec() + STATS_INTERVAL;
407 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
408 svec_init(&bridge_names);
409 for (i = 0; i < cfg->n_bridges; i++) {
410 svec_add(&bridge_names, cfg->bridges[i]->name);
412 svec_sort(&bridge_names);
414 /* Iterate over all system dpifs and delete any of them that do not appear
416 svec_init(&dpif_names);
417 svec_init(&dpif_types);
418 dp_enumerate_types(&dpif_types);
419 for (i = 0; i < dpif_types.n; i++) {
422 dp_enumerate_names(dpif_types.names[i], &dpif_names);
424 /* Delete each dpif whose name is not in 'bridge_names'. */
425 for (j = 0; j < dpif_names.n; j++) {
426 if (!svec_contains(&bridge_names, dpif_names.names[j])) {
430 retval = dpif_open(dpif_names.names[j], dpif_types.names[i],
439 svec_destroy(&bridge_names);
440 svec_destroy(&dpif_names);
441 svec_destroy(&dpif_types);
444 /* Callback for iterate_and_prune_ifaces(). */
446 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
448 if (!iface->netdev) {
449 /* We already reported a related error, don't bother duplicating it. */
453 if (iface->dp_ifidx < 0) {
454 VLOG_ERR("%s interface not in %s, dropping",
455 iface->name, dpif_name(br->dpif));
459 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
460 iface->name, iface->dp_ifidx);
464 /* Callback for iterate_and_prune_ifaces(). */
466 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
467 void *aux OVS_UNUSED)
469 /* Set policing attributes. */
470 netdev_set_policing(iface->netdev,
471 iface->cfg->ingress_policing_rate,
472 iface->cfg->ingress_policing_burst);
474 /* Set MAC address of internal interfaces other than the local
476 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
477 iface_set_mac(iface);
483 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
484 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
485 * deletes from 'br' any ports that no longer have any interfaces. */
487 iterate_and_prune_ifaces(struct bridge *br,
488 bool (*cb)(struct bridge *, struct iface *,
494 for (i = 0; i < br->n_ports; ) {
495 struct port *port = br->ports[i];
496 for (j = 0; j < port->n_ifaces; ) {
497 struct iface *iface = port->ifaces[j];
498 if (cb(br, iface, aux)) {
501 iface_set_ofport(iface->cfg, -1);
502 iface_destroy(iface);
506 if (port->n_ifaces) {
509 VLOG_WARN("%s port has no interfaces, dropping", port->name);
515 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
516 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
517 * responsible for freeing '*managersp' (with free()).
519 * You may be asking yourself "why does ovs-vswitchd care?", because
520 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
521 * should not be and in fact is not directly involved in that. But
522 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
523 * it has to tell in-band control where the managers are to enable that.
524 * (Thus, only managers connected in-band are collected.)
527 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
528 struct sockaddr_in **managersp, size_t *n_managersp)
530 struct sockaddr_in *managers = NULL;
531 size_t n_managers = 0;
532 struct shash targets;
535 /* Collect all of the potential targets from the "targets" columns of the
536 * rows pointed to by "manager_options", excluding any that are
538 shash_init(&targets);
539 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
540 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
542 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
543 shash_find_and_delete(&targets, m->target);
545 shash_add_once(&targets, m->target, NULL);
549 /* Now extract the targets' IP addresses. */
550 if (!shash_is_empty(&targets)) {
551 struct shash_node *node;
553 managers = xmalloc(shash_count(&targets) * sizeof *managers);
554 SHASH_FOR_EACH (node, &targets) {
555 const char *target = node->name;
556 struct sockaddr_in *sin = &managers[n_managers];
558 if ((!strncmp(target, "tcp:", 4)
559 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
560 (!strncmp(target, "ssl:", 4)
561 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
566 shash_destroy(&targets);
568 *managersp = managers;
569 *n_managersp = n_managers;
573 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
575 struct shash old_br, new_br;
576 struct shash_node *node;
577 struct bridge *br, *next;
578 struct sockaddr_in *managers;
581 int sflow_bridge_number;
583 COVERAGE_INC(bridge_reconfigure);
585 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
587 /* Collect old and new bridges. */
590 LIST_FOR_EACH (br, node, &all_bridges) {
591 shash_add(&old_br, br->name, br);
593 for (i = 0; i < ovs_cfg->n_bridges; i++) {
594 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
595 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
596 VLOG_WARN("more than one bridge named %s", br_cfg->name);
600 /* Get rid of deleted bridges and add new bridges. */
601 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
602 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
609 SHASH_FOR_EACH (node, &new_br) {
610 const char *br_name = node->name;
611 const struct ovsrec_bridge *br_cfg = node->data;
612 br = shash_find_data(&old_br, br_name);
614 /* If the bridge datapath type has changed, we need to tear it
615 * down and recreate. */
616 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
618 bridge_create(br_cfg);
621 bridge_create(br_cfg);
624 shash_destroy(&old_br);
625 shash_destroy(&new_br);
627 /* Reconfigure all bridges. */
628 LIST_FOR_EACH (br, node, &all_bridges) {
629 bridge_reconfigure_one(br);
632 /* Add and delete ports on all datapaths.
634 * The kernel will reject any attempt to add a given port to a datapath if
635 * that port already belongs to a different datapath, so we must do all
636 * port deletions before any port additions. */
637 LIST_FOR_EACH (br, node, &all_bridges) {
638 struct dpif_port_dump dump;
639 struct shash want_ifaces;
640 struct dpif_port dpif_port;
642 bridge_get_all_ifaces(br, &want_ifaces);
643 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
644 if (!shash_find(&want_ifaces, dpif_port.name)
645 && strcmp(dpif_port.name, br->name)) {
646 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
648 VLOG_WARN("failed to remove %s interface from %s: %s",
649 dpif_port.name, dpif_name(br->dpif),
654 shash_destroy(&want_ifaces);
656 LIST_FOR_EACH (br, node, &all_bridges) {
657 struct shash cur_ifaces, want_ifaces;
658 struct dpif_port_dump dump;
659 struct dpif_port dpif_port;
661 /* Get the set of interfaces currently in this datapath. */
662 shash_init(&cur_ifaces);
663 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
664 struct dpif_port *port_info = xmalloc(sizeof *port_info);
665 dpif_port_clone(port_info, &dpif_port);
666 shash_add(&cur_ifaces, dpif_port.name, port_info);
669 /* Get the set of interfaces we want on this datapath. */
670 bridge_get_all_ifaces(br, &want_ifaces);
672 hmap_clear(&br->ifaces);
673 SHASH_FOR_EACH (node, &want_ifaces) {
674 const char *if_name = node->name;
675 struct iface *iface = node->data;
676 struct dpif_port *dpif_port;
680 type = iface ? iface->type : "internal";
681 dpif_port = shash_find_data(&cur_ifaces, if_name);
683 /* If we have a port or a netdev already, and it's not the type we
684 * want, then delete the port (if any) and close the netdev (if
686 if ((dpif_port && strcmp(dpif_port->type, type))
687 || (iface && iface->netdev
688 && strcmp(type, netdev_get_type(iface->netdev)))) {
690 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
697 netdev_close(iface->netdev);
698 iface->netdev = NULL;
702 /* If the port doesn't exist or we don't have the netdev open,
703 * we need to do more work. */
704 if (!dpif_port || (iface && !iface->netdev)) {
705 struct netdev_options options;
706 struct netdev *netdev;
709 /* First open the network device. */
710 options.name = if_name;
712 options.args = &args;
713 options.ethertype = NETDEV_ETH_TYPE_NONE;
717 shash_from_ovs_idl_map(iface->cfg->key_options,
718 iface->cfg->value_options,
719 iface->cfg->n_options, &args);
721 error = netdev_open(&options, &netdev);
722 shash_destroy(&args);
725 VLOG_WARN("could not open network device %s (%s)",
726 if_name, strerror(error));
730 /* Then add the port if we haven't already. */
732 error = dpif_port_add(br->dpif, netdev, NULL);
734 netdev_close(netdev);
735 if (error == EFBIG) {
736 VLOG_ERR("ran out of valid port numbers on %s",
737 dpif_name(br->dpif));
740 VLOG_WARN("failed to add %s interface to %s: %s",
741 if_name, dpif_name(br->dpif),
748 /* Update 'iface'. */
750 iface->netdev = netdev;
751 iface->enabled = netdev_get_carrier(iface->netdev);
752 iface->up = iface->enabled;
754 } else if (iface && iface->netdev) {
758 shash_from_ovs_idl_map(iface->cfg->key_options,
759 iface->cfg->value_options,
760 iface->cfg->n_options, &args);
761 netdev_set_config(iface->netdev, &args);
762 shash_destroy(&args);
765 shash_destroy(&want_ifaces);
767 SHASH_FOR_EACH (node, &cur_ifaces) {
768 struct dpif_port *port_info = node->data;
769 dpif_port_destroy(port_info);
772 shash_destroy(&cur_ifaces);
774 sflow_bridge_number = 0;
775 LIST_FOR_EACH (br, node, &all_bridges) {
778 struct iface *local_iface;
779 struct iface *hw_addr_iface;
782 bridge_fetch_dp_ifaces(br);
784 iterate_and_prune_ifaces(br, check_iface, NULL);
786 /* Pick local port hardware address, datapath ID. */
787 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
788 local_iface = bridge_get_local_iface(br);
790 int error = netdev_set_etheraddr(local_iface->netdev, ea);
792 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
793 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
794 "Ethernet address: %s",
795 br->name, strerror(error));
798 memcpy(br->ea, ea, ETH_ADDR_LEN);
800 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
801 ofproto_set_datapath_id(br->ofproto, dpid);
803 dpid_string = xasprintf("%016"PRIx64, dpid);
804 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
807 /* Set NetFlow configuration on this bridge. */
808 if (br->cfg->netflow) {
809 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
810 struct netflow_options opts;
812 memset(&opts, 0, sizeof opts);
814 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
815 if (nf_cfg->engine_type) {
816 opts.engine_type = *nf_cfg->engine_type;
818 if (nf_cfg->engine_id) {
819 opts.engine_id = *nf_cfg->engine_id;
822 opts.active_timeout = nf_cfg->active_timeout;
823 if (!opts.active_timeout) {
824 opts.active_timeout = -1;
825 } else if (opts.active_timeout < 0) {
826 VLOG_WARN("bridge %s: active timeout interval set to negative "
827 "value, using default instead (%d seconds)", br->name,
828 NF_ACTIVE_TIMEOUT_DEFAULT);
829 opts.active_timeout = -1;
832 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
833 if (opts.add_id_to_iface) {
834 if (opts.engine_id > 0x7f) {
835 VLOG_WARN("bridge %s: netflow port mangling may conflict "
836 "with another vswitch, choose an engine id less "
837 "than 128", br->name);
839 if (br->n_ports > 508) {
840 VLOG_WARN("bridge %s: netflow port mangling will conflict "
841 "with another port when more than 508 ports are "
846 opts.collectors.n = nf_cfg->n_targets;
847 opts.collectors.names = nf_cfg->targets;
848 if (ofproto_set_netflow(br->ofproto, &opts)) {
849 VLOG_ERR("bridge %s: problem setting netflow collectors",
853 ofproto_set_netflow(br->ofproto, NULL);
856 /* Set sFlow configuration on this bridge. */
857 if (br->cfg->sflow) {
858 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
859 struct ovsrec_controller **controllers;
860 struct ofproto_sflow_options oso;
861 size_t n_controllers;
863 memset(&oso, 0, sizeof oso);
865 oso.targets.n = sflow_cfg->n_targets;
866 oso.targets.names = sflow_cfg->targets;
868 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
869 if (sflow_cfg->sampling) {
870 oso.sampling_rate = *sflow_cfg->sampling;
873 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
874 if (sflow_cfg->polling) {
875 oso.polling_interval = *sflow_cfg->polling;
878 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
879 if (sflow_cfg->header) {
880 oso.header_len = *sflow_cfg->header;
883 oso.sub_id = sflow_bridge_number++;
884 oso.agent_device = sflow_cfg->agent;
886 oso.control_ip = NULL;
887 n_controllers = bridge_get_controllers(br, &controllers);
888 for (i = 0; i < n_controllers; i++) {
889 if (controllers[i]->local_ip) {
890 oso.control_ip = controllers[i]->local_ip;
894 ofproto_set_sflow(br->ofproto, &oso);
896 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
898 ofproto_set_sflow(br->ofproto, NULL);
901 /* Update the controller and related settings. It would be more
902 * straightforward to call this from bridge_reconfigure_one(), but we
903 * can't do it there for two reasons. First, and most importantly, at
904 * that point we don't know the dp_ifidx of any interfaces that have
905 * been added to the bridge (because we haven't actually added them to
906 * the datapath). Second, at that point we haven't set the datapath ID
907 * yet; when a controller is configured, resetting the datapath ID will
908 * immediately disconnect from the controller, so it's better to set
909 * the datapath ID before the controller. */
910 bridge_reconfigure_remotes(br, managers, n_managers);
912 LIST_FOR_EACH (br, node, &all_bridges) {
913 for (i = 0; i < br->n_ports; i++) {
914 struct port *port = br->ports[i];
917 port_update_bonding(port);
918 port_update_lacp(port);
920 for (j = 0; j < port->n_ifaces; j++) {
921 iface_update_qos(port->ifaces[j], port->cfg->qos);
925 LIST_FOR_EACH (br, node, &all_bridges) {
926 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
929 LIST_FOR_EACH (br, node, &all_bridges) {
931 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
932 iface_update_cfm(iface);
934 bridge_reconfigure_remotes_late(br);
939 /* ovs-vswitchd has completed initialization, so allow the process that
940 * forked us to exit successfully. */
941 daemonize_complete();
945 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
946 const struct ovsdb_idl_column *column,
949 const struct ovsdb_datum *datum;
950 union ovsdb_atom atom;
953 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
954 atom.string = (char *) key;
955 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
956 return idx == UINT_MAX ? NULL : datum->values[idx].string;
960 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
962 return get_ovsrec_key_value(&br_cfg->header_,
963 &ovsrec_bridge_col_other_config, key);
967 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
968 struct iface **hw_addr_iface)
974 *hw_addr_iface = NULL;
976 /* Did the user request a particular MAC? */
977 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
978 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
979 if (eth_addr_is_multicast(ea)) {
980 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
981 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
982 } else if (eth_addr_is_zero(ea)) {
983 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
989 /* Otherwise choose the minimum non-local MAC address among all of the
991 memset(ea, 0xff, ETH_ADDR_LEN);
992 for (i = 0; i < br->n_ports; i++) {
993 struct port *port = br->ports[i];
994 uint8_t iface_ea[ETH_ADDR_LEN];
997 /* Mirror output ports don't participate. */
998 if (port->is_mirror_output_port) {
1002 /* Choose the MAC address to represent the port. */
1003 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
1004 /* Find the interface with this Ethernet address (if any) so that
1005 * we can provide the correct devname to the caller. */
1007 for (j = 0; j < port->n_ifaces; j++) {
1008 struct iface *candidate = port->ifaces[j];
1009 uint8_t candidate_ea[ETH_ADDR_LEN];
1010 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
1011 && eth_addr_equals(iface_ea, candidate_ea)) {
1016 /* Choose the interface whose MAC address will represent the port.
1017 * The Linux kernel bonding code always chooses the MAC address of
1018 * the first slave added to a bond, and the Fedora networking
1019 * scripts always add slaves to a bond in alphabetical order, so
1020 * for compatibility we choose the interface with the name that is
1021 * first in alphabetical order. */
1022 iface = port->ifaces[0];
1023 for (j = 1; j < port->n_ifaces; j++) {
1024 struct iface *candidate = port->ifaces[j];
1025 if (strcmp(candidate->name, iface->name) < 0) {
1030 /* The local port doesn't count (since we're trying to choose its
1031 * MAC address anyway). */
1032 if (iface->dp_ifidx == ODPP_LOCAL) {
1037 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1039 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1040 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1041 iface->name, strerror(error));
1046 /* Compare against our current choice. */
1047 if (!eth_addr_is_multicast(iface_ea) &&
1048 !eth_addr_is_local(iface_ea) &&
1049 !eth_addr_is_reserved(iface_ea) &&
1050 !eth_addr_is_zero(iface_ea) &&
1051 eth_addr_compare_3way(iface_ea, ea) < 0)
1053 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1054 *hw_addr_iface = iface;
1057 if (eth_addr_is_multicast(ea)) {
1058 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1059 *hw_addr_iface = NULL;
1060 VLOG_WARN("bridge %s: using default bridge Ethernet "
1061 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1063 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1064 br->name, ETH_ADDR_ARGS(ea));
1068 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1069 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1070 * an interface on 'br', then that interface must be passed in as
1071 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1072 * 'hw_addr_iface' must be passed in as a null pointer. */
1074 bridge_pick_datapath_id(struct bridge *br,
1075 const uint8_t bridge_ea[ETH_ADDR_LEN],
1076 struct iface *hw_addr_iface)
1079 * The procedure for choosing a bridge MAC address will, in the most
1080 * ordinary case, also choose a unique MAC that we can use as a datapath
1081 * ID. In some special cases, though, multiple bridges will end up with
1082 * the same MAC address. This is OK for the bridges, but it will confuse
1083 * the OpenFlow controller, because each datapath needs a unique datapath
1086 * Datapath IDs must be unique. It is also very desirable that they be
1087 * stable from one run to the next, so that policy set on a datapath
1090 const char *datapath_id;
1093 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1094 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1098 if (hw_addr_iface) {
1100 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1102 * A bridge whose MAC address is taken from a VLAN network device
1103 * (that is, a network device created with vconfig(8) or similar
1104 * tool) will have the same MAC address as a bridge on the VLAN
1105 * device's physical network device.
1107 * Handle this case by hashing the physical network device MAC
1108 * along with the VLAN identifier.
1110 uint8_t buf[ETH_ADDR_LEN + 2];
1111 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1112 buf[ETH_ADDR_LEN] = vlan >> 8;
1113 buf[ETH_ADDR_LEN + 1] = vlan;
1114 return dpid_from_hash(buf, sizeof buf);
1117 * Assume that this bridge's MAC address is unique, since it
1118 * doesn't fit any of the cases we handle specially.
1123 * A purely internal bridge, that is, one that has no non-virtual
1124 * network devices on it at all, is more difficult because it has no
1125 * natural unique identifier at all.
1127 * When the host is a XenServer, we handle this case by hashing the
1128 * host's UUID with the name of the bridge. Names of bridges are
1129 * persistent across XenServer reboots, although they can be reused if
1130 * an internal network is destroyed and then a new one is later
1131 * created, so this is fairly effective.
1133 * When the host is not a XenServer, we punt by using a random MAC
1134 * address on each run.
1136 const char *host_uuid = xenserver_get_host_uuid();
1138 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1139 dpid = dpid_from_hash(combined, strlen(combined));
1145 return eth_addr_to_uint64(bridge_ea);
1149 dpid_from_hash(const void *data, size_t n)
1151 uint8_t hash[SHA1_DIGEST_SIZE];
1153 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1154 sha1_bytes(data, n, hash);
1155 eth_addr_mark_random(hash);
1156 return eth_addr_to_uint64(hash);
1160 iface_refresh_status(struct iface *iface)
1164 enum netdev_flags flags;
1173 if (!netdev_get_status(iface->netdev, &sh)) {
1175 char **keys, **values;
1177 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1178 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1183 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1186 shash_destroy_free_data(&sh);
1188 error = netdev_get_flags(iface->netdev, &flags);
1190 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1193 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1196 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1198 ovsrec_interface_set_duplex(iface->cfg,
1199 netdev_features_is_full_duplex(current)
1201 /* warning: uint64_t -> int64_t conversion */
1202 bps = netdev_features_to_bps(current);
1203 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1206 ovsrec_interface_set_duplex(iface->cfg, NULL);
1207 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1211 ovsrec_interface_set_link_state(iface->cfg,
1212 netdev_get_carrier(iface->netdev)
1215 error = netdev_get_mtu(iface->netdev, &mtu);
1216 if (!error && mtu != INT_MAX) {
1218 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1221 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1226 iface_refresh_cfm_stats(struct iface *iface)
1230 const struct ovsrec_monitor *mon;
1232 mon = iface->cfg->monitor;
1239 for (i = 0; i < mon->n_remote_mps; i++) {
1240 const struct ovsrec_maintenance_point *mp;
1241 const struct remote_mp *rmp;
1243 mp = mon->remote_mps[i];
1244 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1246 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1249 if (hmap_is_empty(&cfm->x_remote_mps)) {
1250 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1253 struct remote_mp *rmp;
1254 int64_t *x_remote_mps;
1256 length = hmap_count(&cfm->x_remote_mps);
1257 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1260 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1261 x_remote_mps[i++] = rmp->mpid;
1264 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1268 if (hmap_is_empty(&cfm->x_remote_maids)) {
1269 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1272 char **x_remote_maids;
1273 struct remote_maid *rmaid;
1275 length = hmap_count(&cfm->x_remote_maids);
1276 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1279 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1282 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1284 for (j = 0; j < CCM_MAID_LEN; j++) {
1285 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1290 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1292 for (i = 0; i < length; i++) {
1293 free(x_remote_maids[i]);
1295 free(x_remote_maids);
1298 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1302 iface_refresh_stats(struct iface *iface)
1308 static const struct iface_stat iface_stats[] = {
1309 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1310 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1311 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1312 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1313 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1314 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1315 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1316 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1317 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1318 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1319 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1320 { "collisions", offsetof(struct netdev_stats, collisions) },
1322 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1323 const struct iface_stat *s;
1325 char *keys[N_STATS];
1326 int64_t values[N_STATS];
1329 struct netdev_stats stats;
1331 /* Intentionally ignore return value, since errors will set 'stats' to
1332 * all-1s, and we will deal with that correctly below. */
1333 netdev_get_stats(iface->netdev, &stats);
1336 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1337 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1338 if (value != UINT64_MAX) {
1345 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1349 enable_system_stats(const struct ovsrec_open_vswitch *cfg)
1353 /* Use other-config:enable-system-stats by preference. */
1354 enable = get_ovsrec_key_value(&cfg->header_,
1355 &ovsrec_open_vswitch_col_other_config,
1356 "enable-statistics");
1358 return !strcmp(enable, "true");
1361 /* Disable by default. */
1366 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1368 struct ovsdb_datum datum;
1372 if (enable_system_stats(cfg)) {
1373 get_system_stats(&stats);
1376 ovsdb_datum_from_shash(&datum, &stats);
1377 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1381 static inline const char *
1382 nx_role_to_str(enum nx_role role)
1387 case NX_ROLE_MASTER:
1392 return "*** INVALID ROLE ***";
1397 bridge_refresh_controller_status(const struct bridge *br)
1400 const struct ovsrec_controller *cfg;
1402 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1404 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1405 struct ofproto_controller_info *cinfo =
1406 shash_find_data(&info, cfg->target);
1409 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1410 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1411 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1412 (char **) cinfo->pairs.values,
1415 ovsrec_controller_set_is_connected(cfg, false);
1416 ovsrec_controller_set_role(cfg, NULL);
1417 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1421 ofproto_free_ofproto_controller_info(&info);
1427 const struct ovsrec_open_vswitch *cfg;
1429 bool datapath_destroyed;
1430 bool database_changed;
1433 /* Let each bridge do the work that it needs to do. */
1434 datapath_destroyed = false;
1435 LIST_FOR_EACH (br, node, &all_bridges) {
1436 int error = bridge_run_one(br);
1438 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1439 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1440 "forcing reconfiguration", br->name);
1441 datapath_destroyed = true;
1445 /* (Re)configure if necessary. */
1446 database_changed = ovsdb_idl_run(idl);
1447 cfg = ovsrec_open_vswitch_first(idl);
1449 /* Re-configure SSL. We do this on every trip through the main loop,
1450 * instead of just when the database changes, because the contents of the
1451 * key and certificate files can change without the database changing.
1453 * We do this before bridge_reconfigure() because that function might
1454 * initiate SSL connections and thus requires SSL to be configured. */
1455 if (cfg && cfg->ssl) {
1456 const struct ovsrec_ssl *ssl = cfg->ssl;
1458 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1459 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1462 if (database_changed || datapath_destroyed) {
1464 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1466 bridge_configure_once(cfg);
1467 bridge_reconfigure(cfg);
1469 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1470 ovsdb_idl_txn_commit(txn);
1471 ovsdb_idl_txn_destroy(txn); /* XXX */
1473 /* We still need to reconfigure to avoid dangling pointers to
1474 * now-destroyed ovsrec structures inside bridge data. */
1475 static const struct ovsrec_open_vswitch null_cfg;
1477 bridge_reconfigure(&null_cfg);
1481 /* Refresh system and interface stats if necessary. */
1482 if (time_msec() >= stats_timer) {
1484 struct ovsdb_idl_txn *txn;
1486 txn = ovsdb_idl_txn_create(idl);
1487 LIST_FOR_EACH (br, node, &all_bridges) {
1490 for (i = 0; i < br->n_ports; i++) {
1491 struct port *port = br->ports[i];
1494 for (j = 0; j < port->n_ifaces; j++) {
1495 struct iface *iface = port->ifaces[j];
1496 iface_refresh_stats(iface);
1497 iface_refresh_cfm_stats(iface);
1498 iface_refresh_status(iface);
1501 bridge_refresh_controller_status(br);
1503 refresh_system_stats(cfg);
1504 ovsdb_idl_txn_commit(txn);
1505 ovsdb_idl_txn_destroy(txn); /* XXX */
1508 stats_timer = time_msec() + STATS_INTERVAL;
1517 LIST_FOR_EACH (br, node, &all_bridges) {
1520 ofproto_wait(br->ofproto);
1521 if (ofproto_has_primary_controller(br->ofproto)) {
1525 mac_learning_wait(br->ml);
1527 for (i = 0; i < br->n_ports; i++) {
1528 port_wait(br->ports[i]);
1531 ovsdb_idl_wait(idl);
1532 poll_timer_wait_until(stats_timer);
1535 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1536 * configuration changes. */
1538 bridge_flush(struct bridge *br)
1540 COVERAGE_INC(bridge_flush);
1542 mac_learning_flush(br->ml);
1545 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1546 * such interface. */
1547 static struct iface *
1548 bridge_get_local_iface(struct bridge *br)
1552 for (i = 0; i < br->n_ports; i++) {
1553 struct port *port = br->ports[i];
1554 for (j = 0; j < port->n_ifaces; j++) {
1555 struct iface *iface = port->ifaces[j];
1556 if (iface->dp_ifidx == ODPP_LOCAL) {
1565 /* Bridge unixctl user interface functions. */
1567 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1568 const char *args, void *aux OVS_UNUSED)
1570 struct ds ds = DS_EMPTY_INITIALIZER;
1571 const struct bridge *br;
1572 const struct mac_entry *e;
1574 br = bridge_lookup(args);
1576 unixctl_command_reply(conn, 501, "no such bridge");
1580 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1581 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1582 if (e->port < 0 || e->port >= br->n_ports) {
1585 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1586 br->ports[e->port]->ifaces[0]->dp_ifidx,
1587 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1589 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1593 /* QoS unixctl user interface functions. */
1595 struct qos_unixctl_show_cbdata {
1597 struct iface *iface;
1601 qos_unixctl_show_cb(unsigned int queue_id,
1602 const struct shash *details,
1605 struct qos_unixctl_show_cbdata *data = aux;
1606 struct ds *ds = data->ds;
1607 struct iface *iface = data->iface;
1608 struct netdev_queue_stats stats;
1609 struct shash_node *node;
1612 ds_put_cstr(ds, "\n");
1614 ds_put_format(ds, "Queue %u:\n", queue_id);
1616 ds_put_cstr(ds, "Default:\n");
1619 SHASH_FOR_EACH (node, details) {
1620 ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data);
1623 error = netdev_get_queue_stats(iface->netdev, queue_id, &stats);
1625 if (stats.tx_packets != UINT64_MAX) {
1626 ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets);
1629 if (stats.tx_bytes != UINT64_MAX) {
1630 ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes);
1633 if (stats.tx_errors != UINT64_MAX) {
1634 ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors);
1637 ds_put_format(ds, "\tFailed to get statistics for queue %u: %s",
1638 queue_id, strerror(error));
1643 qos_unixctl_show(struct unixctl_conn *conn,
1644 const char *args, void *aux OVS_UNUSED)
1646 struct ds ds = DS_EMPTY_INITIALIZER;
1647 struct shash sh = SHASH_INITIALIZER(&sh);
1648 struct iface *iface;
1650 struct shash_node *node;
1651 struct qos_unixctl_show_cbdata data;
1654 iface = iface_find(args);
1656 unixctl_command_reply(conn, 501, "no such interface");
1660 netdev_get_qos(iface->netdev, &type, &sh);
1662 if (*type != '\0') {
1663 ds_put_format(&ds, "QoS: %s %s\n", iface->name, type);
1665 SHASH_FOR_EACH (node, &sh) {
1666 ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data);
1671 error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data);
1674 ds_put_format(&ds, "failed to dump queues: %s", strerror(error));
1676 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1678 ds_put_format(&ds, "QoS not configured on %s\n", iface->name);
1679 unixctl_command_reply(conn, 501, ds_cstr(&ds));
1682 shash_destroy_free_data(&sh);
1686 /* Bridge reconfiguration functions. */
1687 static struct bridge *
1688 bridge_create(const struct ovsrec_bridge *br_cfg)
1693 assert(!bridge_lookup(br_cfg->name));
1694 br = xzalloc(sizeof *br);
1696 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1702 dpif_flow_flush(br->dpif);
1704 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1707 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1709 dpif_delete(br->dpif);
1710 dpif_close(br->dpif);
1715 br->name = xstrdup(br_cfg->name);
1717 br->ml = mac_learning_create();
1718 eth_addr_nicira_random(br->default_ea);
1720 hmap_init(&br->ifaces);
1722 shash_init(&br->port_by_name);
1723 shash_init(&br->iface_by_name);
1727 list_push_back(&all_bridges, &br->node);
1729 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1735 bridge_destroy(struct bridge *br)
1740 while (br->n_ports > 0) {
1741 port_destroy(br->ports[br->n_ports - 1]);
1743 list_remove(&br->node);
1744 error = dpif_delete(br->dpif);
1745 if (error && error != ENOENT) {
1746 VLOG_ERR("failed to delete %s: %s",
1747 dpif_name(br->dpif), strerror(error));
1749 dpif_close(br->dpif);
1750 ofproto_destroy(br->ofproto);
1751 mac_learning_destroy(br->ml);
1752 hmap_destroy(&br->ifaces);
1753 shash_destroy(&br->port_by_name);
1754 shash_destroy(&br->iface_by_name);
1761 static struct bridge *
1762 bridge_lookup(const char *name)
1766 LIST_FOR_EACH (br, node, &all_bridges) {
1767 if (!strcmp(br->name, name)) {
1774 /* Handle requests for a listing of all flows known by the OpenFlow
1775 * stack, including those normally hidden. */
1777 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1778 const char *args, void *aux OVS_UNUSED)
1783 br = bridge_lookup(args);
1785 unixctl_command_reply(conn, 501, "Unknown bridge");
1790 ofproto_get_all_flows(br->ofproto, &results);
1792 unixctl_command_reply(conn, 200, ds_cstr(&results));
1793 ds_destroy(&results);
1796 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1797 * connections and reconnect. If BRIDGE is not specified, then all bridges
1798 * drop their controller connections and reconnect. */
1800 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1801 const char *args, void *aux OVS_UNUSED)
1804 if (args[0] != '\0') {
1805 br = bridge_lookup(args);
1807 unixctl_command_reply(conn, 501, "Unknown bridge");
1810 ofproto_reconnect_controllers(br->ofproto);
1812 LIST_FOR_EACH (br, node, &all_bridges) {
1813 ofproto_reconnect_controllers(br->ofproto);
1816 unixctl_command_reply(conn, 200, NULL);
1820 bridge_run_one(struct bridge *br)
1825 error = ofproto_run1(br->ofproto);
1830 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1832 for (i = 0; i < br->n_ports; i++) {
1833 port_run(br->ports[i]);
1836 error = ofproto_run2(br->ofproto, br->flush);
1843 bridge_get_controllers(const struct bridge *br,
1844 struct ovsrec_controller ***controllersp)
1846 struct ovsrec_controller **controllers;
1847 size_t n_controllers;
1849 controllers = br->cfg->controller;
1850 n_controllers = br->cfg->n_controller;
1852 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1858 *controllersp = controllers;
1860 return n_controllers;
1864 bridge_reconfigure_one(struct bridge *br)
1866 struct shash old_ports, new_ports;
1867 struct svec snoops, old_snoops;
1868 struct shash_node *node;
1869 enum ofproto_fail_mode fail_mode;
1872 /* Collect old ports. */
1873 shash_init(&old_ports);
1874 for (i = 0; i < br->n_ports; i++) {
1875 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1878 /* Collect new ports. */
1879 shash_init(&new_ports);
1880 for (i = 0; i < br->cfg->n_ports; i++) {
1881 const char *name = br->cfg->ports[i]->name;
1882 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1883 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1888 /* If we have a controller, then we need a local port. Complain if the
1889 * user didn't specify one.
1891 * XXX perhaps we should synthesize a port ourselves in this case. */
1892 if (bridge_get_controllers(br, NULL)) {
1893 char local_name[IF_NAMESIZE];
1896 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1897 local_name, sizeof local_name);
1898 if (!error && !shash_find(&new_ports, local_name)) {
1899 VLOG_WARN("bridge %s: controller specified but no local port "
1900 "(port named %s) defined",
1901 br->name, local_name);
1905 /* Get rid of deleted ports.
1906 * Get rid of deleted interfaces on ports that still exist. */
1907 SHASH_FOR_EACH (node, &old_ports) {
1908 struct port *port = node->data;
1909 const struct ovsrec_port *port_cfg;
1911 port_cfg = shash_find_data(&new_ports, node->name);
1915 port_del_ifaces(port, port_cfg);
1919 /* Create new ports.
1920 * Add new interfaces to existing ports.
1921 * Reconfigure existing ports. */
1922 SHASH_FOR_EACH (node, &new_ports) {
1923 struct port *port = shash_find_data(&old_ports, node->name);
1925 port = port_create(br, node->name);
1928 port_reconfigure(port, node->data);
1929 if (!port->n_ifaces) {
1930 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1931 br->name, port->name);
1935 shash_destroy(&old_ports);
1936 shash_destroy(&new_ports);
1938 /* Set the fail-mode */
1939 fail_mode = !br->cfg->fail_mode
1940 || !strcmp(br->cfg->fail_mode, "standalone")
1941 ? OFPROTO_FAIL_STANDALONE
1942 : OFPROTO_FAIL_SECURE;
1943 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1944 && !ofproto_has_primary_controller(br->ofproto)) {
1945 ofproto_flush_flows(br->ofproto);
1947 ofproto_set_fail_mode(br->ofproto, fail_mode);
1949 /* Delete all flows if we're switching from connected to standalone or vice
1950 * versa. (XXX Should we delete all flows if we are switching from one
1951 * controller to another?) */
1953 /* Configure OpenFlow controller connection snooping. */
1955 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1956 ovs_rundir(), br->name));
1957 svec_init(&old_snoops);
1958 ofproto_get_snoops(br->ofproto, &old_snoops);
1959 if (!svec_equal(&snoops, &old_snoops)) {
1960 ofproto_set_snoops(br->ofproto, &snoops);
1962 svec_destroy(&snoops);
1963 svec_destroy(&old_snoops);
1965 mirror_reconfigure(br);
1968 /* Initializes 'oc' appropriately as a management service controller for
1971 * The caller must free oc->target when it is no longer needed. */
1973 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1974 struct ofproto_controller *oc)
1976 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1977 oc->max_backoff = 0;
1978 oc->probe_interval = 60;
1979 oc->band = OFPROTO_OUT_OF_BAND;
1981 oc->burst_limit = 0;
1984 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1986 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1987 struct ofproto_controller *oc)
1989 oc->target = c->target;
1990 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1991 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1992 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1993 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1994 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1995 oc->burst_limit = (c->controller_burst_limit
1996 ? *c->controller_burst_limit : 0);
1999 /* Configures the IP stack for 'br''s local interface properly according to the
2000 * configuration in 'c'. */
2002 bridge_configure_local_iface_netdev(struct bridge *br,
2003 struct ovsrec_controller *c)
2005 struct netdev *netdev;
2006 struct in_addr mask, gateway;
2008 struct iface *local_iface;
2011 /* If there's no local interface or no IP address, give up. */
2012 local_iface = bridge_get_local_iface(br);
2013 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
2017 /* Bring up the local interface. */
2018 netdev = local_iface->netdev;
2019 netdev_turn_flags_on(netdev, NETDEV_UP, true);
2021 /* Configure the IP address and netmask. */
2022 if (!c->local_netmask
2023 || !inet_aton(c->local_netmask, &mask)
2025 mask.s_addr = guess_netmask(ip.s_addr);
2027 if (!netdev_set_in4(netdev, ip, mask)) {
2028 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
2029 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
2032 /* Configure the default gateway. */
2033 if (c->local_gateway
2034 && inet_aton(c->local_gateway, &gateway)
2035 && gateway.s_addr) {
2036 if (!netdev_add_router(netdev, gateway)) {
2037 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
2038 br->name, IP_ARGS(&gateway.s_addr));
2044 bridge_reconfigure_remotes(struct bridge *br,
2045 const struct sockaddr_in *managers,
2048 const char *disable_ib_str, *queue_id_str;
2049 bool disable_in_band = false;
2052 struct ovsrec_controller **controllers;
2053 size_t n_controllers;
2056 struct ofproto_controller *ocs;
2060 /* Check if we should disable in-band control on this bridge. */
2061 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
2062 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
2063 disable_in_band = true;
2066 /* Set OpenFlow queue ID for in-band control. */
2067 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
2068 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
2069 ofproto_set_in_band_queue(br->ofproto, queue_id);
2071 if (disable_in_band) {
2072 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
2074 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
2076 had_primary = ofproto_has_primary_controller(br->ofproto);
2078 n_controllers = bridge_get_controllers(br, &controllers);
2080 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
2083 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
2084 for (i = 0; i < n_controllers; i++) {
2085 struct ovsrec_controller *c = controllers[i];
2087 if (!strncmp(c->target, "punix:", 6)
2088 || !strncmp(c->target, "unix:", 5)) {
2089 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2091 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
2092 * domain sockets and overwriting arbitrary local files. */
2093 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2094 "\"%s\" due to possibility for remote exploit",
2095 dpif_name(br->dpif), c->target);
2099 bridge_configure_local_iface_netdev(br, c);
2100 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2101 if (disable_in_band) {
2102 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2107 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2108 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2111 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2112 ofproto_flush_flows(br->ofproto);
2116 /* Does configuration of remotes that must happen after all of the ports and
2117 * interfaces are fully configured, that is, when flow translation can be
2118 * expected to succeed. (This is because ofproto_add_flow() immediately
2119 * re-translates any existing facets for the rule that it replaces, if any.)
2120 * In particular, it must be called after port_update_bonding(), to ensure that
2121 * 'bond_hash' is non-NULL for bonded ports. */
2123 bridge_reconfigure_remotes_late(struct bridge *br)
2125 /* If there are no controllers and the bridge is in standalone
2126 * mode, set up a flow that matches every packet and directs
2127 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2128 * switch is in secure mode and we won't pass any traffic until
2129 * a controller has been defined and it tells us to do so. */
2130 if (!bridge_get_controllers(br, NULL)
2131 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2132 union ofp_action action;
2133 struct cls_rule rule;
2135 memset(&action, 0, sizeof action);
2136 action.type = htons(OFPAT_OUTPUT);
2137 action.output.len = htons(sizeof action);
2138 action.output.port = htons(OFPP_NORMAL);
2139 cls_rule_init_catchall(&rule, 0);
2140 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2145 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2150 for (i = 0; i < br->n_ports; i++) {
2151 struct port *port = br->ports[i];
2152 for (j = 0; j < port->n_ifaces; j++) {
2153 struct iface *iface = port->ifaces[j];
2154 shash_add_once(ifaces, iface->name, iface);
2156 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2157 shash_add_once(ifaces, port->name, NULL);
2162 /* For robustness, in case the administrator moves around datapath ports behind
2163 * our back, we re-check all the datapath port numbers here.
2165 * This function will set the 'dp_ifidx' members of interfaces that have
2166 * disappeared to -1, so only call this function from a context where those
2167 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2168 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2169 * datapath, which doesn't support UINT16_MAX+1 ports. */
2171 bridge_fetch_dp_ifaces(struct bridge *br)
2173 struct dpif_port_dump dump;
2174 struct dpif_port dpif_port;
2177 /* Reset all interface numbers. */
2178 for (i = 0; i < br->n_ports; i++) {
2179 struct port *port = br->ports[i];
2180 for (j = 0; j < port->n_ifaces; j++) {
2181 struct iface *iface = port->ifaces[j];
2182 iface->dp_ifidx = -1;
2185 hmap_clear(&br->ifaces);
2187 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2188 struct iface *iface = iface_lookup(br, dpif_port.name);
2190 if (iface->dp_ifidx >= 0) {
2191 VLOG_WARN("%s reported interface %s twice",
2192 dpif_name(br->dpif), dpif_port.name);
2193 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2194 VLOG_WARN("%s reported interface %"PRIu16" twice",
2195 dpif_name(br->dpif), dpif_port.port_no);
2197 iface->dp_ifidx = dpif_port.port_no;
2198 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2199 hash_int(iface->dp_ifidx, 0));
2202 iface_set_ofport(iface->cfg,
2203 (iface->dp_ifidx >= 0
2204 ? odp_port_to_ofp_port(iface->dp_ifidx)
2210 /* Bridge packet processing functions. */
2213 bond_is_tcp_hash(const struct port *port)
2215 return port->bond_mode == BM_TCP && port->lacp & LACP_NEGOTIATED;
2219 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2221 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2224 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2226 struct flow hash_flow;
2228 memcpy(&hash_flow, flow, sizeof hash_flow);
2229 hash_flow.vlan_tci = 0;
2231 /* The symmetric quality of this hash function is not required, but
2232 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2233 * purposes, so we use it out of convenience. */
2234 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2237 static struct bond_entry *
2238 lookup_bond_entry(const struct port *port, const struct flow *flow,
2241 assert(port->bond_mode != BM_AB);
2243 if (bond_is_tcp_hash(port)) {
2244 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2246 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2251 bond_choose_iface(const struct port *port)
2253 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2254 size_t i, best_down_slave = -1;
2255 long long next_delay_expiration = LLONG_MAX;
2257 for (i = 0; i < port->n_ifaces; i++) {
2258 struct iface *iface = port->ifaces[i];
2260 if (iface->enabled) {
2262 } else if (iface->delay_expires < next_delay_expiration
2263 && (iface->lacp_status & LACP_ATTACHED
2264 || !(port->lacp & LACP_NEGOTIATED))) {
2265 best_down_slave = i;
2266 next_delay_expiration = iface->delay_expires;
2270 if (best_down_slave != -1) {
2271 struct iface *iface = port->ifaces[best_down_slave];
2273 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2274 "since no other interface is up", iface->name,
2275 iface->delay_expires - time_msec());
2276 bond_enable_slave(iface, true);
2279 return best_down_slave;
2283 choose_output_iface(const struct port *port, const struct flow *flow,
2284 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2286 struct iface *iface;
2288 assert(port->n_ifaces);
2289 if (port->n_ifaces == 1) {
2290 iface = port->ifaces[0];
2291 } else if (port->bond_mode == BM_AB) {
2292 if (port->active_iface < 0) {
2293 *tags |= port->no_ifaces_tag;
2296 iface = port->ifaces[port->active_iface];
2298 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2299 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2300 || !port->ifaces[e->iface_idx]->enabled) {
2301 /* XXX select interface properly. The current interface selection
2302 * is only good for testing the rebalancing code. */
2303 e->iface_idx = bond_choose_iface(port);
2304 if (e->iface_idx < 0) {
2305 *tags |= port->no_ifaces_tag;
2308 e->iface_tag = tag_create_random();
2310 *tags |= e->iface_tag;
2311 iface = port->ifaces[e->iface_idx];
2313 *dp_ifidx = iface->dp_ifidx;
2314 *tags |= iface->tag; /* Currently only used for bonding. */
2319 bond_link_status_update(struct iface *iface)
2321 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2322 struct port *port = iface->port;
2323 bool up = iface->up;
2324 int updelay, downdelay;
2326 updelay = port->updelay;
2327 downdelay = port->downdelay;
2329 if (iface->port->lacp & LACP_NEGOTIATED) {
2334 if (iface->port->lacp && up) {
2335 /* The interface is up if it's attached to an aggregator and its
2336 * partner is synchronized. The only exception is defaulted links.
2337 * They are not required to have synchronized partners because they
2338 * have no partners at all. However, they will only be attached if
2339 * negotiations failed on all interfaces in the bond. */
2340 up = iface->lacp_status & LACP_ATTACHED
2341 && (iface->lacp_partner.state & LACP_STATE_SYNC
2342 || iface->lacp_status & LACP_DEFAULTED);
2346 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2347 /* Nothing to do. */
2350 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2351 iface->name, up ? "up" : "down");
2352 if (up == iface->enabled) {
2353 iface->delay_expires = LLONG_MAX;
2354 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2355 iface->name, up ? "disabled" : "enabled");
2356 } else if (up && port->active_iface < 0) {
2357 bond_enable_slave(iface, true);
2359 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2360 "other interface is up", iface->name, updelay);
2363 int delay = up ? updelay : downdelay;
2364 iface->delay_expires = time_msec() + delay;
2367 "interface %s: will be %s if it stays %s for %d ms",
2369 up ? "enabled" : "disabled",
2377 bond_choose_active_iface(struct port *port)
2379 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2381 port->active_iface = bond_choose_iface(port);
2382 port->active_iface_tag = tag_create_random();
2383 if (port->active_iface >= 0) {
2384 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2385 port->name, port->ifaces[port->active_iface]->name);
2387 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2393 bond_enable_slave(struct iface *iface, bool enable)
2395 struct port *port = iface->port;
2396 struct bridge *br = port->bridge;
2398 /* This acts as a recursion check. If the act of disabling a slave
2399 * causes a different slave to be enabled, the flag will allow us to
2400 * skip redundant work when we reenter this function. It must be
2401 * cleared on exit to keep things safe with multiple bonds. */
2402 static bool moving_active_iface = false;
2404 iface->delay_expires = LLONG_MAX;
2405 if (enable == iface->enabled) {
2409 iface->enabled = enable;
2410 if (!iface->enabled) {
2411 VLOG_WARN("interface %s: disabled", iface->name);
2412 ofproto_revalidate(br->ofproto, iface->tag);
2413 if (iface->port_ifidx == port->active_iface) {
2414 ofproto_revalidate(br->ofproto,
2415 port->active_iface_tag);
2417 /* Disabling a slave can lead to another slave being immediately
2418 * enabled if there will be no active slaves but one is waiting
2419 * on an updelay. In this case we do not need to run most of the
2420 * code for the newly enabled slave since there was no period
2421 * without an active slave and it is redundant with the disabling
2423 moving_active_iface = true;
2424 bond_choose_active_iface(port);
2426 bond_send_learning_packets(port);
2428 VLOG_WARN("interface %s: enabled", iface->name);
2429 if (port->active_iface < 0 && !moving_active_iface) {
2430 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2431 bond_choose_active_iface(port);
2432 bond_send_learning_packets(port);
2434 iface->tag = tag_create_random();
2437 moving_active_iface = false;
2440 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2441 * bond interface. */
2443 bond_update_fake_iface_stats(struct port *port)
2445 struct netdev_stats bond_stats;
2446 struct netdev *bond_dev;
2449 memset(&bond_stats, 0, sizeof bond_stats);
2451 for (i = 0; i < port->n_ifaces; i++) {
2452 struct netdev_stats slave_stats;
2454 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2455 /* XXX: We swap the stats here because they are swapped back when
2456 * reported by the internal device. The reason for this is
2457 * internal devices normally represent packets going into the system
2458 * but when used as fake bond device they represent packets leaving
2459 * the system. We really should do this in the internal device
2460 * itself because changing it here reverses the counts from the
2461 * perspective of the switch. However, the internal device doesn't
2462 * know what type of device it represents so we have to do it here
2464 bond_stats.tx_packets += slave_stats.rx_packets;
2465 bond_stats.tx_bytes += slave_stats.rx_bytes;
2466 bond_stats.rx_packets += slave_stats.tx_packets;
2467 bond_stats.rx_bytes += slave_stats.tx_bytes;
2471 if (!netdev_open_default(port->name, &bond_dev)) {
2472 netdev_set_stats(bond_dev, &bond_stats);
2473 netdev_close(bond_dev);
2478 bond_link_carrier_update(struct iface *iface, bool carrier)
2480 if (carrier == iface->up) {
2484 if (iface->lacp_status & LACP_CURRENT) {
2485 iface_set_lacp_expired(iface);
2488 iface->up = carrier;
2493 bond_run(struct port *port)
2498 if (port->n_ifaces < 2) {
2502 if (port->monitor) {
2503 assert(!port->miimon);
2505 /* Track carrier going up and down on interfaces. */
2506 while (!netdev_monitor_poll(port->monitor, &devname)) {
2507 struct iface *iface;
2509 iface = port_lookup_iface(port, devname);
2511 bool up = netdev_get_carrier(iface->netdev);
2512 bond_link_carrier_update(iface, up);
2517 assert(port->miimon);
2519 if (time_msec() >= port->bond_miimon_next_update) {
2520 for (i = 0; i < port->n_ifaces; i++) {
2521 struct iface *iface = port->ifaces[i];
2522 bool up = netdev_get_miimon(iface->netdev);
2523 bond_link_carrier_update(iface, up);
2525 port->bond_miimon_next_update = time_msec() +
2526 port->bond_miimon_interval;
2530 for (i = 0; i < port->n_ifaces; i++) {
2531 bond_link_status_update(port->ifaces[i]);
2534 for (i = 0; i < port->n_ifaces; i++) {
2535 struct iface *iface = port->ifaces[i];
2536 if (time_msec() >= iface->delay_expires) {
2537 bond_enable_slave(iface, !iface->enabled);
2541 if (port->bond_fake_iface
2542 && time_msec() >= port->bond_next_fake_iface_update) {
2543 bond_update_fake_iface_stats(port);
2544 port->bond_next_fake_iface_update = time_msec() + 1000;
2549 bond_wait(struct port *port)
2553 if (port->n_ifaces < 2) {
2557 if (port->monitor) {
2558 netdev_monitor_poll_wait(port->monitor);
2562 poll_timer_wait_until(port->bond_miimon_next_update);
2565 for (i = 0; i < port->n_ifaces; i++) {
2566 struct iface *iface = port->ifaces[i];
2567 if (iface->delay_expires != LLONG_MAX) {
2568 poll_timer_wait_until(iface->delay_expires);
2572 if (port->bond_fake_iface) {
2573 poll_timer_wait_until(port->bond_next_fake_iface_update);
2578 set_dst(struct dst *dst, const struct flow *flow,
2579 const struct port *in_port, const struct port *out_port,
2582 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2583 : in_port->vlan >= 0 ? in_port->vlan
2584 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2585 : vlan_tci_to_vid(flow->vlan_tci));
2586 return choose_output_iface(out_port, flow, dst->vlan,
2587 &dst->dp_ifidx, tags);
2591 swap_dst(struct dst *p, struct dst *q)
2593 struct dst tmp = *p;
2598 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2599 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2600 * that we push to the datapath. We could in fact fully sort the array by
2601 * vlan, but in most cases there are at most two different vlan tags so that's
2602 * possibly overkill.) */
2604 partition_dsts(struct dst_set *set, int vlan)
2606 struct dst *first = set->dsts;
2607 struct dst *last = set->dsts + set->n;
2609 while (first != last) {
2611 * - All dsts < first have vlan == 'vlan'.
2612 * - All dsts >= last have vlan != 'vlan'.
2613 * - first < last. */
2614 while (first->vlan == vlan) {
2615 if (++first == last) {
2620 /* Same invariants, plus one additional:
2621 * - first->vlan != vlan.
2623 while (last[-1].vlan != vlan) {
2624 if (--last == first) {
2629 /* Same invariants, plus one additional:
2630 * - last[-1].vlan == vlan.*/
2631 swap_dst(first++, --last);
2636 mirror_mask_ffs(mirror_mask_t mask)
2638 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2643 dst_set_init(struct dst_set *set)
2645 set->dsts = set->builtin;
2647 set->allocated = ARRAY_SIZE(set->builtin);
2651 dst_set_add(struct dst_set *set, const struct dst *dst)
2653 if (set->n >= set->allocated) {
2654 size_t new_allocated;
2655 struct dst *new_dsts;
2657 new_allocated = set->allocated * 2;
2658 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2659 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2663 set->dsts = new_dsts;
2664 set->allocated = new_allocated;
2666 set->dsts[set->n++] = *dst;
2670 dst_set_free(struct dst_set *set)
2672 if (set->dsts != set->builtin) {
2678 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2681 for (i = 0; i < set->n; i++) {
2682 if (set->dsts[i].vlan == test->vlan
2683 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2691 port_trunks_vlan(const struct port *port, uint16_t vlan)
2693 return (port->vlan < 0
2694 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2698 port_includes_vlan(const struct port *port, uint16_t vlan)
2700 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2704 port_is_floodable(const struct port *port)
2708 for (i = 0; i < port->n_ifaces; i++) {
2709 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2710 port->ifaces[i]->dp_ifidx)) {
2717 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
2718 * to a VLAN. In general most packets may be mirrored but we want to drop
2719 * protocols that may confuse switches. */
2721 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
2723 /* If you change this function's behavior, please update corresponding
2724 * documentation in vswitch.xml at the same time. */
2725 if (dst[0] != 0x01) {
2726 /* All the currently banned MACs happen to start with 01 currently, so
2727 * this is a quick way to eliminate most of the good ones. */
2729 if (eth_addr_is_reserved(dst)) {
2730 /* Drop STP, IEEE pause frames, and other reserved protocols
2731 * (01-80-c2-00-00-0x). */
2735 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
2737 if ((dst[3] & 0xfe) == 0xcc &&
2738 (dst[4] & 0xfe) == 0xcc &&
2739 (dst[5] & 0xfe) == 0xcc) {
2740 /* Drop the following protocols plus others following the same
2743 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
2744 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
2745 STP Uplink Fast (01-00-0c-cd-cd-cd) */
2749 if (!(dst[3] | dst[4] | dst[5])) {
2750 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
2759 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2760 const struct port *in_port, const struct port *out_port,
2761 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2763 mirror_mask_t mirrors = in_port->src_mirrors;
2768 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2769 if (flow_vlan == 0) {
2770 flow_vlan = OFP_VLAN_NONE;
2773 if (out_port == FLOOD_PORT) {
2774 for (i = 0; i < br->n_ports; i++) {
2775 struct port *port = br->ports[i];
2777 && port_is_floodable(port)
2778 && port_includes_vlan(port, vlan)
2779 && !port->is_mirror_output_port
2780 && set_dst(&dst, flow, in_port, port, tags)) {
2781 mirrors |= port->dst_mirrors;
2782 dst_set_add(set, &dst);
2785 *nf_output_iface = NF_OUT_FLOOD;
2786 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2787 dst_set_add(set, &dst);
2788 *nf_output_iface = dst.dp_ifidx;
2789 mirrors |= out_port->dst_mirrors;
2793 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2794 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2796 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2797 && !dst_is_duplicate(set, &dst)) {
2798 dst_set_add(set, &dst);
2800 } else if (eth_dst_may_rspan(flow->dl_dst)) {
2801 for (i = 0; i < br->n_ports; i++) {
2802 struct port *port = br->ports[i];
2803 if (port_includes_vlan(port, m->out_vlan)
2804 && set_dst(&dst, flow, in_port, port, tags))
2806 if (port->vlan < 0) {
2807 dst.vlan = m->out_vlan;
2809 if (dst_is_duplicate(set, &dst)) {
2813 /* Use the vlan tag on the original flow instead of
2814 * the one passed in the vlan parameter. This ensures
2815 * that we compare the vlan from before any implicit
2816 * tagging tags place. This is necessary because
2817 * dst->vlan is the final vlan, after removing implicit
2819 if (port == in_port && dst.vlan == flow_vlan) {
2820 /* Don't send out input port on same VLAN. */
2823 dst_set_add(set, &dst);
2828 mirrors &= mirrors - 1;
2831 partition_dsts(set, flow_vlan);
2834 static void OVS_UNUSED
2835 print_dsts(const struct dst_set *set)
2839 for (i = 0; i < set->n; i++) {
2840 const struct dst *dst = &set->dsts[i];
2842 printf(">p%"PRIu16, dst->dp_ifidx);
2843 if (dst->vlan != OFP_VLAN_NONE) {
2844 printf("v%"PRIu16, dst->vlan);
2850 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2851 const struct port *in_port, const struct port *out_port,
2852 tag_type *tags, struct ofpbuf *actions,
2853 uint16_t *nf_output_iface)
2860 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2863 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2864 if (cur_vlan == 0) {
2865 cur_vlan = OFP_VLAN_NONE;
2867 for (i = 0; i < set.n; i++) {
2868 const struct dst *dst = &set.dsts[i];
2869 if (dst->vlan != cur_vlan) {
2870 if (dst->vlan == OFP_VLAN_NONE) {
2871 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2874 tci = htons(dst->vlan & VLAN_VID_MASK);
2875 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2876 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2878 cur_vlan = dst->vlan;
2880 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2885 /* Returns the effective vlan of a packet, taking into account both the
2886 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2887 * the packet is untagged and -1 indicates it has an invalid header and
2888 * should be dropped. */
2889 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2890 struct port *in_port, bool have_packet)
2892 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2893 if (in_port->vlan >= 0) {
2895 /* XXX support double tagging? */
2897 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2898 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2899 "packet received on port %s configured with "
2900 "implicit VLAN %"PRIu16,
2901 br->name, vlan, in_port->name, in_port->vlan);
2905 vlan = in_port->vlan;
2907 if (!port_includes_vlan(in_port, vlan)) {
2909 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2910 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2911 "packet received on port %s not configured for "
2913 br->name, vlan, in_port->name, vlan);
2922 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2923 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2924 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2926 is_gratuitous_arp(const struct flow *flow)
2928 return (flow->dl_type == htons(ETH_TYPE_ARP)
2929 && eth_addr_is_broadcast(flow->dl_dst)
2930 && (flow->nw_proto == ARP_OP_REPLY
2931 || (flow->nw_proto == ARP_OP_REQUEST
2932 && flow->nw_src == flow->nw_dst)));
2936 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2937 struct port *in_port)
2939 enum grat_arp_lock_type lock_type;
2942 /* We don't want to learn from gratuitous ARP packets that are reflected
2943 * back over bond slaves so we lock the learning table. */
2944 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2945 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2946 GRAT_ARP_LOCK_CHECK;
2948 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2951 /* The log messages here could actually be useful in debugging,
2952 * so keep the rate limit relatively high. */
2953 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2955 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2956 "on port %s in VLAN %d",
2957 br->name, ETH_ADDR_ARGS(flow->dl_src),
2958 in_port->name, vlan);
2959 ofproto_revalidate(br->ofproto, rev_tag);
2963 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2964 * dropped. Returns true if they may be forwarded, false if they should be
2967 * If 'have_packet' is true, it indicates that the caller is processing a
2968 * received packet. If 'have_packet' is false, then the caller is just
2969 * revalidating an existing flow because configuration has changed. Either
2970 * way, 'have_packet' only affects logging (there is no point in logging errors
2971 * during revalidation).
2973 * Sets '*in_portp' to the input port. This will be a null pointer if
2974 * flow->in_port does not designate a known input port (in which case
2975 * is_admissible() returns false).
2977 * When returning true, sets '*vlanp' to the effective VLAN of the input
2978 * packet, as returned by flow_get_vlan().
2980 * May also add tags to '*tags', although the current implementation only does
2981 * so in one special case.
2984 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2985 tag_type *tags, int *vlanp, struct port **in_portp)
2987 struct iface *in_iface;
2988 struct port *in_port;
2991 /* Find the interface and port structure for the received packet. */
2992 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2994 /* No interface? Something fishy... */
2996 /* Odd. A few possible reasons here:
2998 * - We deleted an interface but there are still a few packets
2999 * queued up from it.
3001 * - Someone externally added an interface (e.g. with "ovs-dpctl
3002 * add-if") that we don't know about.
3004 * - Packet arrived on the local port but the local port is not
3005 * one of our bridge ports.
3007 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3009 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
3010 "interface %"PRIu16, br->name, flow->in_port);
3016 *in_portp = in_port = in_iface->port;
3017 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
3022 /* Drop frames for reserved multicast addresses. */
3023 if (eth_addr_is_reserved(flow->dl_dst)) {
3027 /* Drop frames on ports reserved for mirroring. */
3028 if (in_port->is_mirror_output_port) {
3030 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3031 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
3032 "%s, which is reserved exclusively for mirroring",
3033 br->name, in_port->name);
3038 /* When using LACP, do not accept packets from disabled interfaces. */
3039 if (in_port->lacp & LACP_NEGOTIATED && !in_iface->enabled) {
3043 /* Packets received on non-LACP bonds need special attention to avoid
3045 if (in_port->n_ifaces > 1 && !(in_port->lacp & LACP_NEGOTIATED)) {
3047 bool is_grat_arp_locked;
3049 if (eth_addr_is_multicast(flow->dl_dst)) {
3050 *tags |= in_port->active_iface_tag;
3051 if (in_port->active_iface != in_iface->port_ifidx) {
3052 /* Drop all multicast packets on inactive slaves. */
3057 /* Drop all packets for which we have learned a different input
3058 * port, because we probably sent the packet on one slave and got
3059 * it back on the other. Gratuitous ARP packets are an exception
3060 * to this rule: the host has moved to another switch. The exception
3061 * to the exception is if we locked the learning table to avoid
3062 * reflections on bond slaves. If this is the case, just drop the
3064 if (in_port->bond_mode != BM_AB) {
3065 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
3066 &is_grat_arp_locked);
3067 if (src_idx != -1 && src_idx != in_port->port_idx &&
3068 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
3074 /* Drop all packets which arrive on backup slaves. This is similar to how
3075 * Linux bonding handles active-backup bonds. */
3076 if (in_port->bond_mode == BM_AB) {
3078 *tags |= in_port->active_iface;
3079 if (in_port->active_iface != in_iface->port_ifidx) {
3080 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3082 VLOG_WARN_RL(&rl, "active-backup bond received packet on backup"
3083 " interface (%s) destined for " ETH_ADDR_FMT,
3084 in_iface->name, ETH_ADDR_ARGS(flow->dl_dst));
3092 /* If the composed actions may be applied to any packet in the given 'flow',
3093 * returns true. Otherwise, the actions should only be applied to 'packet', or
3094 * not at all, if 'packet' was NULL. */
3096 process_flow(struct bridge *br, const struct flow *flow,
3097 const struct ofpbuf *packet, struct ofpbuf *actions,
3098 tag_type *tags, uint16_t *nf_output_iface)
3100 struct port *in_port;
3101 struct port *out_port;
3105 /* Check whether we should drop packets in this flow. */
3106 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
3111 /* Learn source MAC (but don't try to learn from revalidation). */
3113 update_learning_table(br, flow, vlan, in_port);
3116 /* Determine output port. */
3117 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
3119 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
3120 out_port = br->ports[out_port_idx];
3121 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
3122 /* If we are revalidating but don't have a learning entry then
3123 * eject the flow. Installing a flow that floods packets opens
3124 * up a window of time where we could learn from a packet reflected
3125 * on a bond and blackhole packets before the learning table is
3126 * updated to reflect the correct port. */
3129 out_port = FLOOD_PORT;
3132 /* Don't send packets out their input ports. */
3133 if (in_port == out_port) {
3139 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
3147 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
3148 struct ofpbuf *actions, tag_type *tags,
3149 uint16_t *nf_output_iface, void *br_)
3151 struct bridge *br = br_;
3153 COVERAGE_INC(bridge_process_flow);
3154 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
3158 bridge_special_ofhook_cb(const struct flow *flow,
3159 const struct ofpbuf *packet, void *br_)
3161 struct iface *iface;
3162 struct bridge *br = br_;
3164 iface = iface_from_dp_ifidx(br, flow->in_port);
3166 if (cfm_should_process_flow(flow)) {
3168 if (iface && packet && iface->cfm) {
3169 COVERAGE_INC(bridge_process_cfm);
3170 cfm_process_heartbeat(iface->cfm, packet);
3173 } else if (flow->dl_type == htons(ETH_TYPE_LACP)) {
3175 if (iface && packet) {
3176 COVERAGE_INC(bridge_process_lacp);
3177 lacp_process_packet(packet, iface);
3186 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
3187 const struct nlattr *actions,
3189 uint64_t n_bytes, void *br_)
3191 struct bridge *br = br_;
3192 const struct nlattr *a;
3193 struct port *in_port;
3198 /* Feed information from the active flows back into the learning table to
3199 * ensure that table is always in sync with what is actually flowing
3200 * through the datapath.
3202 * We test that 'tags' is nonzero to ensure that only flows that include an
3203 * OFPP_NORMAL action are used for learning. This works because
3204 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3205 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3206 update_learning_table(br, flow, vlan, in_port);
3209 /* Account for bond slave utilization. */
3210 if (!br->has_bonded_ports) {
3213 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3214 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3215 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3216 if (out_port && out_port->n_ifaces >= 2 &&
3217 out_port->bond_mode != BM_AB) {
3218 uint16_t vlan = (flow->vlan_tci
3219 ? vlan_tci_to_vid(flow->vlan_tci)
3221 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
3222 e->tx_bytes += n_bytes;
3229 bridge_account_checkpoint_ofhook_cb(void *br_)
3231 struct bridge *br = br_;
3235 if (!br->has_bonded_ports) {
3240 for (i = 0; i < br->n_ports; i++) {
3241 struct port *port = br->ports[i];
3242 if (port->n_ifaces > 1 && port->bond_mode != BM_AB
3243 && now >= port->bond_next_rebalance) {
3244 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3245 bond_rebalance_port(port);
3250 static struct ofhooks bridge_ofhooks = {
3251 bridge_normal_ofhook_cb,
3252 bridge_special_ofhook_cb,
3253 bridge_account_flow_ofhook_cb,
3254 bridge_account_checkpoint_ofhook_cb,
3257 /* LACP functions. */
3260 lacp_process_packet(const struct ofpbuf *packet, struct iface *iface)
3262 const struct lacp_pdu *pdu;
3264 if (!iface->port->lacp) {
3268 pdu = parse_lacp_packet(packet);
3273 iface->lacp_status |= LACP_CURRENT;
3274 iface->lacp_status &= ~(LACP_EXPIRED | LACP_DEFAULTED);
3275 iface->lacp_rx = time_msec() + LACP_SLOW_TIME_RX;
3277 iface->lacp_actor.state = iface_get_lacp_state(iface);
3278 if (memcmp(&iface->lacp_actor, &pdu->partner, sizeof pdu->partner)) {
3282 if (memcmp(&iface->lacp_partner, &pdu->actor, sizeof pdu->actor)) {
3283 iface->port->lacp_need_update = true;
3284 iface->lacp_partner = pdu->actor;
3289 lacp_update_ifaces(struct port *port)
3293 struct lacp_info lead_pri;
3294 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3296 port->lacp_need_update = false;
3297 COVERAGE_INC(bridge_lacp_update);
3303 VLOG_DBG_RL(&rl, "port %s: re-evaluating LACP link status", port->name);
3306 for (i = 0; i < port->n_ifaces; i++) {
3307 struct iface *iface = port->ifaces[i];
3308 struct lacp_info pri;
3310 iface->lacp_status |= LACP_ATTACHED;
3311 ofproto_revalidate(port->bridge->ofproto, iface->tag);
3313 /* Don't allow loopback interfaces to send traffic or lead. */
3314 if (eth_addr_equals(iface->lacp_partner.sysid,
3315 iface->lacp_actor.sysid)) {
3316 VLOG_WARN_RL(&rl, "iface %s: Loopback detected. Interface is "
3317 "connected to its own bridge", iface->name);
3318 iface->lacp_status &= ~LACP_ATTACHED;
3322 if (iface->lacp_status & LACP_DEFAULTED) {
3326 iface_get_lacp_priority(iface, &pri);
3328 if (!lead || memcmp(&pri, &lead_pri, sizeof pri) < 0) {
3335 port->lacp &= ~LACP_NEGOTIATED;
3339 port->lacp |= LACP_NEGOTIATED;
3341 for (i = 0; i < port->n_ifaces; i++) {
3342 struct iface *iface = port->ifaces[i];
3344 if (iface->lacp_status & LACP_DEFAULTED
3345 || lead->lacp_partner.key != iface->lacp_partner.key
3346 || !eth_addr_equals(lead->lacp_partner.sysid,
3347 iface->lacp_partner.sysid)) {
3348 iface->lacp_status &= ~LACP_ATTACHED;
3354 lacp_iface_may_tx(const struct iface *iface)
3356 return iface->port->lacp & LACP_ACTIVE
3357 || iface->lacp_status & (LACP_CURRENT | LACP_EXPIRED);
3361 lacp_run(struct port *port)
3364 struct ofpbuf packet;
3370 ofpbuf_init(&packet, ETH_HEADER_LEN + LACP_PDU_LEN);
3372 for (i = 0; i < port->n_ifaces; i++) {
3373 struct iface *iface = port->ifaces[i];
3375 if (time_msec() > iface->lacp_rx) {
3376 if (iface->lacp_status & LACP_CURRENT) {
3377 iface_set_lacp_expired(iface);
3378 } else if (iface->lacp_status & LACP_EXPIRED) {
3379 iface_set_lacp_defaulted(iface);
3384 if (port->lacp_need_update) {
3385 lacp_update_ifaces(port);
3388 for (i = 0; i < port->n_ifaces; i++) {
3389 struct iface *iface = port->ifaces[i];
3390 uint8_t ea[ETH_ADDR_LEN];
3393 if (time_msec() < iface->lacp_tx || !lacp_iface_may_tx(iface)) {
3397 error = netdev_get_etheraddr(iface->netdev, ea);
3399 iface->lacp_actor.state = iface_get_lacp_state(iface);
3400 compose_lacp_packet(&packet, &iface->lacp_actor,
3401 &iface->lacp_partner, ea);
3402 iface_send_packet(iface, &packet);
3404 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3405 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3406 "(%s)", iface->name, strerror(error));
3409 iface->lacp_tx = time_msec() +
3410 (iface->lacp_partner.state & LACP_STATE_TIME
3412 : LACP_SLOW_TIME_TX);
3414 ofpbuf_uninit(&packet);
3418 lacp_wait(struct port *port)
3426 for (i = 0; i < port->n_ifaces; i++) {
3427 struct iface *iface = port->ifaces[i];
3429 if (lacp_iface_may_tx(iface)) {
3430 poll_timer_wait_until(iface->lacp_tx);
3433 if (iface->lacp_status & (LACP_CURRENT | LACP_EXPIRED)) {
3434 poll_timer_wait_until(iface->lacp_rx);
3439 /* Bonding functions. */
3441 /* Statistics for a single interface on a bonded port, used for load-based
3442 * bond rebalancing. */
3443 struct slave_balance {
3444 struct iface *iface; /* The interface. */
3445 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3447 /* All the "bond_entry"s that are assigned to this interface, in order of
3448 * increasing tx_bytes. */
3449 struct bond_entry **hashes;
3454 bond_mode_to_string(enum bond_mode bm) {
3455 static char *bm_slb = "balance-slb";
3456 static char *bm_ab = "active-backup";
3457 static char *bm_tcp = "balance-tcp";
3460 case BM_SLB: return bm_slb;
3461 case BM_AB: return bm_ab;
3462 case BM_TCP: return bm_tcp;
3469 /* Sorts pointers to pointers to bond_entries in ascending order by the
3470 * interface to which they are assigned, and within a single interface in
3471 * ascending order of bytes transmitted. */
3473 compare_bond_entries(const void *a_, const void *b_)
3475 const struct bond_entry *const *ap = a_;
3476 const struct bond_entry *const *bp = b_;
3477 const struct bond_entry *a = *ap;
3478 const struct bond_entry *b = *bp;
3479 if (a->iface_idx != b->iface_idx) {
3480 return a->iface_idx > b->iface_idx ? 1 : -1;
3481 } else if (a->tx_bytes != b->tx_bytes) {
3482 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3488 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3489 * *descending* order by number of bytes transmitted. */
3491 compare_slave_balance(const void *a_, const void *b_)
3493 const struct slave_balance *a = a_;
3494 const struct slave_balance *b = b_;
3495 if (a->iface->enabled != b->iface->enabled) {
3496 return a->iface->enabled ? -1 : 1;
3497 } else if (a->tx_bytes != b->tx_bytes) {
3498 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3505 swap_bals(struct slave_balance *a, struct slave_balance *b)
3507 struct slave_balance tmp = *a;
3512 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3513 * given that 'p' (and only 'p') might be in the wrong location.
3515 * This function invalidates 'p', since it might now be in a different memory
3518 resort_bals(struct slave_balance *p,
3519 struct slave_balance bals[], size_t n_bals)
3522 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3523 swap_bals(p, p - 1);
3525 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3526 swap_bals(p, p + 1);
3532 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3534 if (VLOG_IS_DBG_ENABLED()) {
3535 struct ds ds = DS_EMPTY_INITIALIZER;
3536 const struct slave_balance *b;
3538 for (b = bals; b < bals + n_bals; b++) {
3542 ds_put_char(&ds, ',');
3544 ds_put_format(&ds, " %s %"PRIu64"kB",
3545 b->iface->name, b->tx_bytes / 1024);
3547 if (!b->iface->enabled) {
3548 ds_put_cstr(&ds, " (disabled)");
3550 if (b->n_hashes > 0) {
3551 ds_put_cstr(&ds, " (");
3552 for (i = 0; i < b->n_hashes; i++) {
3553 const struct bond_entry *e = b->hashes[i];
3555 ds_put_cstr(&ds, " + ");
3557 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3558 e - port->bond_hash, e->tx_bytes / 1024);
3560 ds_put_cstr(&ds, ")");
3563 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3568 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3570 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3573 struct bond_entry *hash = from->hashes[hash_idx];
3574 struct port *port = from->iface->port;
3575 uint64_t delta = hash->tx_bytes;
3577 assert(port->bond_mode != BM_AB);
3579 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3580 "from %s to %s (now carrying %"PRIu64"kB and "
3581 "%"PRIu64"kB load, respectively)",
3582 port->name, delta / 1024, hash - port->bond_hash,
3583 from->iface->name, to->iface->name,
3584 (from->tx_bytes - delta) / 1024,
3585 (to->tx_bytes + delta) / 1024);
3587 /* Delete element from from->hashes.
3589 * We don't bother to add the element to to->hashes because not only would
3590 * it require more work, the only purpose it would be to allow that hash to
3591 * be migrated to another slave in this rebalancing run, and there is no
3592 * point in doing that. */
3593 if (hash_idx == 0) {
3596 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3597 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3601 /* Shift load away from 'from' to 'to'. */
3602 from->tx_bytes -= delta;
3603 to->tx_bytes += delta;
3605 /* Arrange for flows to be revalidated. */
3606 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
3607 hash->iface_idx = to->iface->port_ifidx;
3608 hash->iface_tag = tag_create_random();
3612 bond_rebalance_port(struct port *port)
3614 struct slave_balance *bals;
3616 struct bond_entry *hashes[BOND_MASK + 1];
3617 struct slave_balance *b, *from, *to;
3618 struct bond_entry *e;
3621 assert(port->bond_mode != BM_AB);
3623 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3624 * descending order of tx_bytes, so that bals[0] represents the most
3625 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3628 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3629 * array for each slave_balance structure, we sort our local array of
3630 * hashes in order by slave, so that all of the hashes for a given slave
3631 * become contiguous in memory, and then we point each 'hashes' members of
3632 * a slave_balance structure to the start of a contiguous group. */
3633 n_bals = port->n_ifaces;
3634 bals = xmalloc(n_bals * sizeof *bals);
3635 for (b = bals; b < &bals[n_bals]; b++) {
3636 b->iface = port->ifaces[b - bals];
3641 for (i = 0; i <= BOND_MASK; i++) {
3642 hashes[i] = &port->bond_hash[i];
3644 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3645 for (i = 0; i <= BOND_MASK; i++) {
3647 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3648 b = &bals[e->iface_idx];
3649 b->tx_bytes += e->tx_bytes;
3651 b->hashes = &hashes[i];
3656 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3657 log_bals(bals, n_bals, port);
3659 /* Discard slaves that aren't enabled (which were sorted to the back of the
3660 * array earlier). */
3661 while (!bals[n_bals - 1].iface->enabled) {
3668 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3669 to = &bals[n_bals - 1];
3670 for (from = bals; from < to; ) {
3671 uint64_t overload = from->tx_bytes - to->tx_bytes;
3672 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3673 /* The extra load on 'from' (and all less-loaded slaves), compared
3674 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3675 * it is less than ~1Mbps. No point in rebalancing. */
3677 } else if (from->n_hashes == 1) {
3678 /* 'from' only carries a single MAC hash, so we can't shift any
3679 * load away from it, even though we want to. */
3682 /* 'from' is carrying significantly more load than 'to', and that
3683 * load is split across at least two different hashes. Pick a hash
3684 * to migrate to 'to' (the least-loaded slave), given that doing so
3685 * must decrease the ratio of the load on the two slaves by at
3688 * The sort order we use means that we prefer to shift away the
3689 * smallest hashes instead of the biggest ones. There is little
3690 * reason behind this decision; we could use the opposite sort
3691 * order to shift away big hashes ahead of small ones. */
3694 for (i = 0; i < from->n_hashes; i++) {
3695 double old_ratio, new_ratio;
3696 uint64_t delta = from->hashes[i]->tx_bytes;
3698 if (delta == 0 || from->tx_bytes - delta == 0) {
3699 /* Pointless move. */
3703 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3705 if (to->tx_bytes == 0) {
3706 /* Nothing on the new slave, move it. */
3710 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3711 new_ratio = (double)(from->tx_bytes - delta) /
3712 (to->tx_bytes + delta);
3714 if (new_ratio == 0) {
3715 /* Should already be covered but check to prevent division
3720 if (new_ratio < 1) {
3721 new_ratio = 1 / new_ratio;
3724 if (old_ratio - new_ratio > 0.1) {
3725 /* Would decrease the ratio, move it. */
3729 if (i < from->n_hashes) {
3730 bond_shift_load(from, to, i);
3732 /* If the result of the migration changed the relative order of
3733 * 'from' and 'to' swap them back to maintain invariants. */
3734 if (order_swapped) {
3735 swap_bals(from, to);
3738 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3739 * point to different slave_balance structures. It is only
3740 * valid to do these two operations in a row at all because we
3741 * know that 'from' will not move past 'to' and vice versa. */
3742 resort_bals(from, bals, n_bals);
3743 resort_bals(to, bals, n_bals);
3750 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3751 * historical data to decay to <1% in 7 rebalancing runs. */
3752 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3761 bond_send_learning_packets(struct port *port)
3763 struct bridge *br = port->bridge;
3764 struct mac_entry *e;
3765 struct ofpbuf packet;
3766 int error, n_packets, n_errors;
3768 if (!port->n_ifaces || port->active_iface < 0 || bond_is_tcp_hash(port)) {
3772 ofpbuf_init(&packet, 128);
3773 error = n_packets = n_errors = 0;
3774 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3775 union ofp_action actions[2], *a;
3781 if (e->port == port->port_idx) {
3785 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3787 flow_extract(&packet, 0, ODPP_NONE, &flow);
3789 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3793 /* Compose actions. */
3794 memset(actions, 0, sizeof actions);
3797 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3798 a->vlan_vid.len = htons(sizeof *a);
3799 a->vlan_vid.vlan_vid = htons(e->vlan);
3802 a->output.type = htons(OFPAT_OUTPUT);
3803 a->output.len = htons(sizeof *a);
3804 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3809 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3816 ofpbuf_uninit(&packet);
3819 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3820 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3821 "packets, last error was: %s",
3822 port->name, n_errors, n_packets, strerror(error));
3824 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3825 port->name, n_packets);
3829 /* Bonding unixctl user interface functions. */
3832 bond_unixctl_list(struct unixctl_conn *conn,
3833 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3835 struct ds ds = DS_EMPTY_INITIALIZER;
3836 const struct bridge *br;
3838 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3840 LIST_FOR_EACH (br, node, &all_bridges) {
3843 for (i = 0; i < br->n_ports; i++) {
3844 const struct port *port = br->ports[i];
3845 if (port->n_ifaces > 1) {
3848 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3849 bond_mode_to_string(port->bond_mode));
3850 for (j = 0; j < port->n_ifaces; j++) {
3851 const struct iface *iface = port->ifaces[j];
3853 ds_put_cstr(&ds, ", ");
3855 ds_put_cstr(&ds, iface->name);
3857 ds_put_char(&ds, '\n');
3861 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3865 static struct port *
3866 bond_find(const char *name)
3868 const struct bridge *br;
3870 LIST_FOR_EACH (br, node, &all_bridges) {
3873 for (i = 0; i < br->n_ports; i++) {
3874 struct port *port = br->ports[i];
3875 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3884 ds_put_lacp_state(struct ds *ds, uint8_t state)
3886 if (state & LACP_STATE_ACT) {
3887 ds_put_cstr(ds, "activity ");
3890 if (state & LACP_STATE_TIME) {
3891 ds_put_cstr(ds, "timeout ");
3894 if (state & LACP_STATE_AGG) {
3895 ds_put_cstr(ds, "aggregation ");
3898 if (state & LACP_STATE_SYNC) {
3899 ds_put_cstr(ds, "synchronized ");
3902 if (state & LACP_STATE_COL) {
3903 ds_put_cstr(ds, "collecting ");
3906 if (state & LACP_STATE_DIST) {
3907 ds_put_cstr(ds, "distributing ");
3910 if (state & LACP_STATE_DEF) {
3911 ds_put_cstr(ds, "defaulted ");
3914 if (state & LACP_STATE_EXP) {
3915 ds_put_cstr(ds, "expired ");
3920 bond_unixctl_show(struct unixctl_conn *conn,
3921 const char *args, void *aux OVS_UNUSED)
3923 struct ds ds = DS_EMPTY_INITIALIZER;
3924 const struct port *port;
3927 port = bond_find(args);
3929 unixctl_command_reply(conn, 501, "no such bond");
3933 ds_put_format(&ds, "bond_mode: %s\n",
3934 bond_mode_to_string(port->bond_mode));
3937 ds_put_format(&ds, "lacp: %s\n",
3938 port->lacp & LACP_ACTIVE ? "active" : "passive");
3940 ds_put_cstr(&ds, "lacp: off\n");
3943 if (port->bond_mode != BM_AB) {
3944 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3945 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3949 ds_put_format(&ds, "bond-detect-mode: %s\n",
3950 port->miimon ? "miimon" : "carrier");
3953 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3954 port->bond_miimon_interval);
3957 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3958 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3960 if (port->bond_mode != BM_AB) {
3961 ds_put_format(&ds, "next rebalance: %lld ms\n",
3962 port->bond_next_rebalance - time_msec());
3965 for (j = 0; j < port->n_ifaces; j++) {
3966 const struct iface *iface = port->ifaces[j];
3967 struct bond_entry *be;
3971 ds_put_format(&ds, "\nslave %s: %s\n",
3972 iface->name, iface->enabled ? "enabled" : "disabled");
3973 if (j == port->active_iface) {
3974 ds_put_cstr(&ds, "\tactive slave\n");
3976 if (iface->delay_expires != LLONG_MAX) {
3977 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3978 iface->enabled ? "downdelay" : "updelay",
3979 iface->delay_expires - time_msec());
3983 ds_put_cstr(&ds, "\tstatus: ");
3985 if (iface->lacp_status & LACP_CURRENT) {
3986 ds_put_cstr(&ds, "current ");
3989 if (iface->lacp_status & LACP_EXPIRED) {
3990 ds_put_cstr(&ds, "expired ");
3993 if (iface->lacp_status & LACP_DEFAULTED) {
3994 ds_put_cstr(&ds, "defaulted ");
3997 if (iface->lacp_status & LACP_ATTACHED) {
3998 ds_put_cstr(&ds, "attached ");
4001 ds_put_cstr(&ds, "\n");
4003 ds_put_cstr(&ds, "\n\tactor sysid: ");
4004 ds_put_format(&ds, ETH_ADDR_FMT,
4005 ETH_ADDR_ARGS(iface->lacp_actor.sysid));
4006 ds_put_cstr(&ds, "\n");
4008 ds_put_format(&ds, "\tactor sys_priority: %u\n",
4009 ntohs(iface->lacp_actor.sys_priority));
4011 ds_put_format(&ds, "\tactor portid: %u\n",
4012 ntohs(iface->lacp_actor.portid));
4014 ds_put_format(&ds, "\tactor port_priority: %u\n",
4015 ntohs(iface->lacp_actor.port_priority));
4017 ds_put_format(&ds, "\tactor key: %u\n",
4018 ntohs(iface->lacp_actor.key));
4020 ds_put_cstr(&ds, "\tactor state: ");
4021 ds_put_lacp_state(&ds, iface_get_lacp_state(iface));
4022 ds_put_cstr(&ds, "\n\n");
4024 ds_put_cstr(&ds, "\tpartner sysid: ");
4025 ds_put_format(&ds, ETH_ADDR_FMT,
4026 ETH_ADDR_ARGS(iface->lacp_partner.sysid));
4027 ds_put_cstr(&ds, "\n");
4029 ds_put_format(&ds, "\tpartner sys_priority: %u\n",
4030 ntohs(iface->lacp_partner.sys_priority));
4032 ds_put_format(&ds, "\tpartner portid: %u\n",
4033 ntohs(iface->lacp_partner.portid));
4035 ds_put_format(&ds, "\tpartner port_priority: %u\n",
4036 ntohs(iface->lacp_partner.port_priority));
4038 ds_put_format(&ds, "\tpartner key: %u\n",
4039 ntohs(iface->lacp_partner.key));
4041 ds_put_cstr(&ds, "\tpartner state: ");
4042 ds_put_lacp_state(&ds, iface->lacp_partner.state);
4043 ds_put_cstr(&ds, "\n");
4046 if (port->bond_mode == BM_AB) {
4051 memset(&flow, 0, sizeof flow);
4052 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
4053 int hash = be - port->bond_hash;
4054 struct mac_entry *me;
4056 if (be->iface_idx != j) {
4060 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
4061 hash, be->tx_bytes / 1024);
4063 if (port->bond_mode != BM_SLB) {
4068 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
4072 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
4073 if (bond_hash_src(me->mac, me->vlan) == hash
4074 && me->port != port->port_idx
4075 && choose_output_iface(port, &flow, me->vlan,
4077 && dp_ifidx == iface->dp_ifidx)
4079 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
4080 ETH_ADDR_ARGS(me->mac));
4085 unixctl_command_reply(conn, 200, ds_cstr(&ds));
4090 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
4091 void *aux OVS_UNUSED)
4093 char *args = (char *) args_;
4094 char *save_ptr = NULL;
4095 char *bond_s, *hash_s, *slave_s;
4097 struct iface *iface;
4098 struct bond_entry *entry;
4101 bond_s = strtok_r(args, " ", &save_ptr);
4102 hash_s = strtok_r(NULL, " ", &save_ptr);
4103 slave_s = strtok_r(NULL, " ", &save_ptr);
4105 unixctl_command_reply(conn, 501,
4106 "usage: bond/migrate BOND HASH SLAVE");
4110 port = bond_find(bond_s);
4112 unixctl_command_reply(conn, 501, "no such bond");
4116 if (port->bond_mode != BM_SLB) {
4117 unixctl_command_reply(conn, 501, "not an SLB bond");
4121 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
4122 hash = atoi(hash_s) & BOND_MASK;
4124 unixctl_command_reply(conn, 501, "bad hash");
4128 iface = port_lookup_iface(port, slave_s);
4130 unixctl_command_reply(conn, 501, "no such slave");
4134 if (!iface->enabled) {
4135 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
4139 entry = &port->bond_hash[hash];
4140 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
4141 entry->iface_idx = iface->port_ifidx;
4142 entry->iface_tag = tag_create_random();
4143 unixctl_command_reply(conn, 200, "migrated");
4147 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
4148 void *aux OVS_UNUSED)
4150 char *args = (char *) args_;
4151 char *save_ptr = NULL;
4152 char *bond_s, *slave_s;
4154 struct iface *iface;
4156 bond_s = strtok_r(args, " ", &save_ptr);
4157 slave_s = strtok_r(NULL, " ", &save_ptr);
4159 unixctl_command_reply(conn, 501,
4160 "usage: bond/set-active-slave BOND SLAVE");
4164 port = bond_find(bond_s);
4166 unixctl_command_reply(conn, 501, "no such bond");
4170 iface = port_lookup_iface(port, slave_s);
4172 unixctl_command_reply(conn, 501, "no such slave");
4176 if (!iface->enabled) {
4177 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
4181 if (port->active_iface != iface->port_ifidx) {
4182 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4183 port->active_iface = iface->port_ifidx;
4184 port->active_iface_tag = tag_create_random();
4185 VLOG_INFO("port %s: active interface is now %s",
4186 port->name, iface->name);
4187 bond_send_learning_packets(port);
4188 unixctl_command_reply(conn, 200, "done");
4190 unixctl_command_reply(conn, 200, "no change");
4195 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
4197 char *args = (char *) args_;
4198 char *save_ptr = NULL;
4199 char *bond_s, *slave_s;
4201 struct iface *iface;
4203 bond_s = strtok_r(args, " ", &save_ptr);
4204 slave_s = strtok_r(NULL, " ", &save_ptr);
4206 unixctl_command_reply(conn, 501,
4207 "usage: bond/enable/disable-slave BOND SLAVE");
4211 port = bond_find(bond_s);
4213 unixctl_command_reply(conn, 501, "no such bond");
4217 iface = port_lookup_iface(port, slave_s);
4219 unixctl_command_reply(conn, 501, "no such slave");
4223 bond_enable_slave(iface, enable);
4224 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
4228 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
4229 void *aux OVS_UNUSED)
4231 enable_slave(conn, args, true);
4235 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
4236 void *aux OVS_UNUSED)
4238 enable_slave(conn, args, false);
4242 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
4243 void *aux OVS_UNUSED)
4245 char *args = (char *) args_;
4246 uint8_t mac[ETH_ADDR_LEN];
4250 char *mac_s, *vlan_s;
4251 char *save_ptr = NULL;
4253 mac_s = strtok_r(args, " ", &save_ptr);
4254 vlan_s = strtok_r(NULL, " ", &save_ptr);
4257 if (sscanf(vlan_s, "%u", &vlan) != 1) {
4258 unixctl_command_reply(conn, 501, "invalid vlan");
4262 vlan = OFP_VLAN_NONE;
4265 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
4266 == ETH_ADDR_SCAN_COUNT) {
4267 hash = bond_hash_src(mac, vlan);
4269 hash_cstr = xasprintf("%u", hash);
4270 unixctl_command_reply(conn, 200, hash_cstr);
4273 unixctl_command_reply(conn, 501, "invalid mac");
4280 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
4281 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
4282 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
4283 unixctl_command_register("bond/set-active-slave",
4284 bond_unixctl_set_active_slave, NULL);
4285 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
4287 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
4289 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
4292 /* Port functions. */
4295 port_run(struct port *port)
4302 for (i = 0; i < port->n_ifaces; i++) {
4303 struct iface *iface = port->ifaces[i];
4306 struct ofpbuf *packet = cfm_run(iface->cfm);
4308 iface_send_packet(iface, packet);
4309 ofpbuf_uninit(packet);
4317 port_wait(struct port *port)
4324 for (i = 0; i < port->n_ifaces; i++) {
4325 struct iface *iface = port->ifaces[i];
4327 cfm_wait(iface->cfm);
4332 static struct port *
4333 port_create(struct bridge *br, const char *name)
4337 port = xzalloc(sizeof *port);
4339 port->port_idx = br->n_ports;
4341 port->trunks = NULL;
4342 port->name = xstrdup(name);
4343 port->active_iface = -1;
4345 if (br->n_ports >= br->allocated_ports) {
4346 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
4349 br->ports[br->n_ports++] = port;
4350 shash_add_assert(&br->port_by_name, port->name, port);
4352 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
4359 get_port_other_config(const struct ovsrec_port *port, const char *key,
4360 const char *default_value)
4364 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
4366 return value ? value : default_value;
4370 get_interface_other_config(const struct ovsrec_interface *iface,
4371 const char *key, const char *default_value)
4375 value = get_ovsrec_key_value(&iface->header_,
4376 &ovsrec_interface_col_other_config, key);
4377 return value ? value : default_value;
4381 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
4383 struct shash new_ifaces;
4386 /* Collect list of new interfaces. */
4387 shash_init(&new_ifaces);
4388 for (i = 0; i < cfg->n_interfaces; i++) {
4389 const char *name = cfg->interfaces[i]->name;
4390 shash_add_once(&new_ifaces, name, NULL);
4393 /* Get rid of deleted interfaces. */
4394 for (i = 0; i < port->n_ifaces; ) {
4395 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
4396 iface_destroy(port->ifaces[i]);
4402 shash_destroy(&new_ifaces);
4406 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
4408 const char *detect_mode;
4409 struct shash new_ifaces;
4410 long long int next_rebalance, miimon_next_update, lacp_priority;
4411 unsigned long *trunks;
4417 /* Update settings. */
4418 port->updelay = cfg->bond_updelay;
4419 if (port->updelay < 0) {
4422 port->downdelay = cfg->bond_downdelay;
4423 if (port->downdelay < 0) {
4424 port->downdelay = 0;
4426 port->bond_rebalance_interval = atoi(
4427 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
4428 if (port->bond_rebalance_interval < 1000) {
4429 port->bond_rebalance_interval = 1000;
4431 next_rebalance = time_msec() + port->bond_rebalance_interval;
4432 if (port->bond_next_rebalance > next_rebalance) {
4433 port->bond_next_rebalance = next_rebalance;
4436 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
4439 if (!strcmp(detect_mode, "carrier")) {
4440 port->miimon = false;
4441 } else if (!strcmp(detect_mode, "miimon")) {
4442 port->miimon = true;
4444 port->miimon = false;
4445 VLOG_WARN("port %s: unsupported bond-detect-mode %s, defaulting to "
4446 "carrier", port->name, detect_mode);
4449 port->bond_miimon_interval = atoi(
4450 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4451 if (port->bond_miimon_interval < 100) {
4452 port->bond_miimon_interval = 100;
4454 miimon_next_update = time_msec() + port->bond_miimon_interval;
4455 if (port->bond_miimon_next_update > miimon_next_update) {
4456 port->bond_miimon_next_update = miimon_next_update;
4459 if (!port->cfg->bond_mode ||
4460 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4461 port->bond_mode = BM_SLB;
4462 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4463 port->bond_mode = BM_AB;
4464 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
4465 port->bond_mode = BM_TCP;
4467 port->bond_mode = BM_SLB;
4468 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4469 port->name, port->cfg->bond_mode,
4470 bond_mode_to_string(port->bond_mode));
4473 /* Add new interfaces and update 'cfg' member of existing ones. */
4474 shash_init(&new_ifaces);
4475 for (i = 0; i < cfg->n_interfaces; i++) {
4476 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4477 struct iface *iface;
4479 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
4480 VLOG_WARN("port %s: %s specified twice as port interface",
4481 port->name, if_cfg->name);
4482 iface_set_ofport(if_cfg, -1);
4486 iface = iface_lookup(port->bridge, if_cfg->name);
4488 if (iface->port != port) {
4489 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4491 port->bridge->name, if_cfg->name, iface->port->name);
4494 iface->cfg = if_cfg;
4496 iface = iface_create(port, if_cfg);
4499 /* Determine interface type. The local port always has type
4500 * "internal". Other ports take their type from the database and
4501 * default to "system" if none is specified. */
4502 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4503 : if_cfg->type[0] ? if_cfg->type
4507 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4510 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4511 iface->lacp_priority = UINT16_MAX;
4513 iface->lacp_priority = lacp_priority;
4516 shash_destroy(&new_ifaces);
4519 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4521 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4522 /* Prefer bondable links if unspecified. */
4523 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4525 port->lacp_priority = lacp_priority;
4528 if (!port->cfg->lacp) {
4529 /* XXX when LACP implementation has been sufficiently tested, enable by
4530 * default and make active on bonded ports. */
4532 } else if (!strcmp(port->cfg->lacp, "off")) {
4534 } else if (!strcmp(port->cfg->lacp, "active")) {
4535 port->lacp = LACP_ACTIVE;
4536 } else if (!strcmp(port->cfg->lacp, "passive")) {
4537 port->lacp = LACP_PASSIVE;
4539 VLOG_WARN("port %s: unknown LACP mode %s",
4540 port->name, port->cfg->lacp);
4547 if (port->n_ifaces < 2) {
4549 if (vlan >= 0 && vlan <= 4095) {
4550 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4555 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4556 * they even work as-is. But they have not been tested. */
4557 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4561 if (port->vlan != vlan) {
4563 bridge_flush(port->bridge);
4566 /* Get trunked VLANs. */
4568 if (vlan < 0 && cfg->n_trunks) {
4571 trunks = bitmap_allocate(4096);
4573 for (i = 0; i < cfg->n_trunks; i++) {
4574 int trunk = cfg->trunks[i];
4576 bitmap_set1(trunks, trunk);
4582 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4583 port->name, cfg->n_trunks);
4585 if (n_errors == cfg->n_trunks) {
4586 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4588 bitmap_free(trunks);
4591 } else if (vlan >= 0 && cfg->n_trunks) {
4592 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4596 ? port->trunks != NULL
4597 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4598 bridge_flush(port->bridge);
4600 bitmap_free(port->trunks);
4601 port->trunks = trunks;
4605 port_destroy(struct port *port)
4608 struct bridge *br = port->bridge;
4612 for (i = 0; i < MAX_MIRRORS; i++) {
4613 struct mirror *m = br->mirrors[i];
4614 if (m && m->out_port == port) {
4619 while (port->n_ifaces > 0) {
4620 iface_destroy(port->ifaces[port->n_ifaces - 1]);
4623 shash_find_and_delete_assert(&br->port_by_name, port->name);
4625 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
4626 del->port_idx = port->port_idx;
4628 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4630 netdev_monitor_destroy(port->monitor);
4632 bitmap_free(port->trunks);
4633 free(port->bond_hash);
4640 static struct port *
4641 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4643 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4644 return iface ? iface->port : NULL;
4647 static struct port *
4648 port_lookup(const struct bridge *br, const char *name)
4650 return shash_find_data(&br->port_by_name, name);
4653 static struct iface *
4654 port_lookup_iface(const struct port *port, const char *name)
4656 struct iface *iface = iface_lookup(port->bridge, name);
4657 return iface && iface->port == port ? iface : NULL;
4661 port_update_lacp(struct port *port)
4666 if (!port->lacp || port->n_ifaces < 1) {
4667 for (i = 0; i < port->n_ifaces; i++) {
4668 iface_set_lacp_defaulted(port->ifaces[i]);
4674 for (i = 0; i < port->n_ifaces; i++) {
4675 struct iface *iface = port->ifaces[i];
4677 if (iface->dp_ifidx <= 0 || iface->dp_ifidx > UINT16_MAX) {
4682 if (iface->dp_ifidx == port->lacp_key) {
4683 key_changed = false;
4688 port->lacp_key = port->ifaces[0]->dp_ifidx;
4691 for (i = 0; i < port->n_ifaces; i++) {
4692 struct iface *iface = port->ifaces[i];
4694 iface->lacp_actor.sys_priority = htons(port->lacp_priority);
4695 memcpy(&iface->lacp_actor.sysid, port->bridge->ea, ETH_ADDR_LEN);
4697 iface->lacp_actor.port_priority = htons(iface->lacp_priority);
4698 iface->lacp_actor.portid = htons(iface->dp_ifidx);
4699 iface->lacp_actor.key = htons(port->lacp_key);
4703 port->lacp_need_update = true;
4707 port_update_bonding(struct port *port)
4709 if (port->monitor) {
4710 netdev_monitor_destroy(port->monitor);
4711 port->monitor = NULL;
4713 if (port->n_ifaces < 2) {
4714 /* Not a bonded port. */
4715 free(port->bond_hash);
4716 port->bond_hash = NULL;
4717 port->bond_fake_iface = false;
4718 port->active_iface = -1;
4719 port->no_ifaces_tag = 0;
4723 if (port->bond_mode != BM_AB && !port->bond_hash) {
4724 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4725 for (i = 0; i <= BOND_MASK; i++) {
4726 struct bond_entry *e = &port->bond_hash[i];
4730 port->bond_next_rebalance
4731 = time_msec() + port->bond_rebalance_interval;
4732 } else if (port->bond_mode == BM_AB) {
4733 free(port->bond_hash);
4734 port->bond_hash = NULL;
4737 if (!port->no_ifaces_tag) {
4738 port->no_ifaces_tag = tag_create_random();
4741 if (port->active_iface < 0) {
4742 bond_choose_active_iface(port);
4745 port->bond_fake_iface = port->cfg->bond_fake_iface;
4746 if (port->bond_fake_iface) {
4747 port->bond_next_fake_iface_update = time_msec();
4750 if (!port->miimon) {
4751 port->monitor = netdev_monitor_create();
4752 for (i = 0; i < port->n_ifaces; i++) {
4753 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
4759 /* Interface functions. */
4762 iface_set_lacp_defaulted(struct iface *iface)
4764 memset(&iface->lacp_partner, 0, sizeof iface->lacp_partner);
4766 iface->lacp_status |= LACP_DEFAULTED;
4767 iface->lacp_status &= ~(LACP_CURRENT | LACP_EXPIRED);
4769 iface->port->lacp_need_update = true;
4773 iface_set_lacp_expired(struct iface *iface)
4775 iface->lacp_status &= ~LACP_CURRENT;
4776 iface->lacp_status |= LACP_EXPIRED;
4777 iface->lacp_partner.state |= LACP_STATE_TIME;
4778 iface->lacp_partner.state &= ~LACP_STATE_SYNC;
4780 iface->lacp_rx = time_msec() + LACP_FAST_TIME_RX;
4785 iface_get_lacp_state(const struct iface *iface)
4789 if (iface->port->lacp & LACP_ACTIVE) {
4790 state |= LACP_STATE_ACT;
4793 if (iface->lacp_status & LACP_ATTACHED) {
4794 state |= LACP_STATE_SYNC;
4797 if (iface->lacp_status & LACP_DEFAULTED) {
4798 state |= LACP_STATE_DEF;
4801 if (iface->lacp_status & LACP_EXPIRED) {
4802 state |= LACP_STATE_EXP;
4805 if (iface->port->n_ifaces > 1) {
4806 state |= LACP_STATE_AGG;
4809 if (iface->enabled) {
4810 state |= LACP_STATE_COL | LACP_STATE_DIST;
4816 /* Given 'iface', populates 'priority' with data representing its LACP link
4817 * priority. If two priority objects populated by this function are compared
4818 * using memcmp, the higher priority link will be less than the lower priority
4821 iface_get_lacp_priority(struct iface *iface, struct lacp_info *priority)
4823 uint16_t partner_priority, actor_priority;
4825 /* Choose the lacp_info of the higher priority system by comparing their
4826 * system priorities and mac addresses. */
4827 actor_priority = ntohs(iface->lacp_actor.sys_priority);
4828 partner_priority = ntohs(iface->lacp_partner.sys_priority);
4829 if (actor_priority < partner_priority) {
4830 *priority = iface->lacp_actor;
4831 } else if (partner_priority < actor_priority) {
4832 *priority = iface->lacp_partner;
4833 } else if (eth_addr_compare_3way(iface->lacp_actor.sysid,
4834 iface->lacp_partner.sysid) < 0) {
4835 *priority = iface->lacp_actor;
4837 *priority = iface->lacp_partner;
4840 /* Key and state are not used in priority comparisons. */
4842 priority->state = 0;
4846 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
4849 union ofp_action action;
4851 memset(&action, 0, sizeof action);
4852 action.output.type = htons(OFPAT_OUTPUT);
4853 action.output.len = htons(sizeof action);
4854 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
4856 flow_extract(packet, 0, ODPP_NONE, &flow);
4858 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
4860 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4861 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
4865 static struct iface *
4866 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4868 struct bridge *br = port->bridge;
4869 struct iface *iface;
4870 char *name = if_cfg->name;
4872 iface = xzalloc(sizeof *iface);
4874 iface->port_ifidx = port->n_ifaces;
4875 iface->name = xstrdup(name);
4876 iface->dp_ifidx = -1;
4877 iface->tag = tag_create_random();
4878 iface->delay_expires = LLONG_MAX;
4879 iface->netdev = NULL;
4880 iface->cfg = if_cfg;
4881 iface_set_lacp_defaulted(iface);
4883 if (port->lacp & LACP_ACTIVE) {
4884 iface_set_lacp_expired(iface);
4887 shash_add_assert(&br->iface_by_name, iface->name, iface);
4889 if (port->n_ifaces >= port->allocated_ifaces) {
4890 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
4891 sizeof *port->ifaces);
4893 port->ifaces[port->n_ifaces++] = iface;
4894 if (port->n_ifaces > 1) {
4895 br->has_bonded_ports = true;
4898 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4906 iface_destroy(struct iface *iface)
4909 struct port *port = iface->port;
4910 struct bridge *br = port->bridge;
4911 bool del_active = port->active_iface == iface->port_ifidx;
4914 if (port->monitor) {
4915 netdev_monitor_remove(port->monitor, iface->netdev);
4918 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4920 if (iface->dp_ifidx >= 0) {
4921 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4924 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
4925 del->port_ifidx = iface->port_ifidx;
4927 netdev_close(iface->netdev);
4930 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4931 bond_choose_active_iface(port);
4932 bond_send_learning_packets(port);
4935 cfm_destroy(iface->cfm);
4940 bridge_flush(port->bridge);
4944 static struct iface *
4945 iface_lookup(const struct bridge *br, const char *name)
4947 return shash_find_data(&br->iface_by_name, name);
4950 static struct iface *
4951 iface_find(const char *name)
4953 const struct bridge *br;
4955 LIST_FOR_EACH (br, node, &all_bridges) {
4956 struct iface *iface = iface_lookup(br, name);
4965 static struct iface *
4966 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4968 struct iface *iface;
4970 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4971 hash_int(dp_ifidx, 0), &br->ifaces) {
4972 if (iface->dp_ifidx == dp_ifidx) {
4979 /* Set Ethernet address of 'iface', if one is specified in the configuration
4982 iface_set_mac(struct iface *iface)
4984 uint8_t ea[ETH_ADDR_LEN];
4986 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4987 if (eth_addr_is_multicast(ea)) {
4988 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4990 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4991 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4992 iface->name, iface->name);
4994 int error = netdev_set_etheraddr(iface->netdev, ea);
4996 VLOG_ERR("interface %s: setting MAC failed (%s)",
4997 iface->name, strerror(error));
5003 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
5005 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
5008 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
5012 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
5014 * The value strings in '*shash' are taken directly from values[], not copied,
5015 * so the caller should not modify or free them. */
5017 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
5018 struct shash *shash)
5023 for (i = 0; i < n; i++) {
5024 shash_add(shash, keys[i], values[i]);
5028 /* Creates 'keys' and 'values' arrays from 'shash'.
5030 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
5031 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
5032 * are populated with with strings taken directly from 'shash' and thus have
5033 * the same ownership of the key-value pairs in shash.
5036 shash_to_ovs_idl_map(struct shash *shash,
5037 char ***keys, char ***values, size_t *n)
5041 struct shash_node *sn;
5043 count = shash_count(shash);
5045 k = xmalloc(count * sizeof *k);
5046 v = xmalloc(count * sizeof *v);
5049 SHASH_FOR_EACH(sn, shash) {
5060 struct iface_delete_queues_cbdata {
5061 struct netdev *netdev;
5062 const struct ovsdb_datum *queues;
5066 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
5068 union ovsdb_atom atom;
5070 atom.integer = target;
5071 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
5075 iface_delete_queues(unsigned int queue_id,
5076 const struct shash *details OVS_UNUSED, void *cbdata_)
5078 struct iface_delete_queues_cbdata *cbdata = cbdata_;
5080 if (!queue_ids_include(cbdata->queues, queue_id)) {
5081 netdev_delete_queue(cbdata->netdev, queue_id);
5086 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
5088 if (!qos || qos->type[0] == '\0') {
5089 netdev_set_qos(iface->netdev, NULL, NULL);
5091 struct iface_delete_queues_cbdata cbdata;
5092 struct shash details;
5095 /* Configure top-level Qos for 'iface'. */
5096 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
5097 qos->n_other_config, &details);
5098 netdev_set_qos(iface->netdev, qos->type, &details);
5099 shash_destroy(&details);
5101 /* Deconfigure queues that were deleted. */
5102 cbdata.netdev = iface->netdev;
5103 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
5105 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
5107 /* Configure queues for 'iface'. */
5108 for (i = 0; i < qos->n_queues; i++) {
5109 const struct ovsrec_queue *queue = qos->value_queues[i];
5110 unsigned int queue_id = qos->key_queues[i];
5112 shash_from_ovs_idl_map(queue->key_other_config,
5113 queue->value_other_config,
5114 queue->n_other_config, &details);
5115 netdev_set_queue(iface->netdev, queue_id, &details);
5116 shash_destroy(&details);
5122 iface_update_cfm(struct iface *iface)
5126 uint16_t *remote_mps;
5127 struct ovsrec_monitor *mon;
5128 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
5130 mon = iface->cfg->monitor;
5133 cfm_destroy(iface->cfm);
5138 if (netdev_get_etheraddr(iface->netdev, ea)) {
5139 VLOG_WARN("interface %s: Failed to get ethernet address. "
5140 "Skipping Monitor.", iface->name);
5144 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
5145 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
5150 iface->cfm = cfm_create();
5154 cfm->mpid = mon->mpid;
5155 cfm->interval = mon->interval ? *mon->interval : 1000;
5157 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
5158 memcpy(cfm->maid, maid, sizeof cfm->maid);
5160 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
5161 for(i = 0; i < mon->n_remote_mps; i++) {
5162 remote_mps[i] = mon->remote_mps[i]->mpid;
5164 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
5167 if (!cfm_configure(iface->cfm)) {
5168 cfm_destroy(iface->cfm);
5173 /* Port mirroring. */
5175 static struct mirror *
5176 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
5180 for (i = 0; i < MAX_MIRRORS; i++) {
5181 struct mirror *m = br->mirrors[i];
5182 if (m && uuid_equals(uuid, &m->uuid)) {
5190 mirror_reconfigure(struct bridge *br)
5192 unsigned long *rspan_vlans;
5195 /* Get rid of deleted mirrors. */
5196 for (i = 0; i < MAX_MIRRORS; i++) {
5197 struct mirror *m = br->mirrors[i];
5199 const struct ovsdb_datum *mc;
5200 union ovsdb_atom atom;
5202 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
5203 atom.uuid = br->mirrors[i]->uuid;
5204 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
5210 /* Add new mirrors and reconfigure existing ones. */
5211 for (i = 0; i < br->cfg->n_mirrors; i++) {
5212 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
5213 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
5215 mirror_reconfigure_one(m, cfg);
5217 mirror_create(br, cfg);
5221 /* Update port reserved status. */
5222 for (i = 0; i < br->n_ports; i++) {
5223 br->ports[i]->is_mirror_output_port = false;
5225 for (i = 0; i < MAX_MIRRORS; i++) {
5226 struct mirror *m = br->mirrors[i];
5227 if (m && m->out_port) {
5228 m->out_port->is_mirror_output_port = true;
5232 /* Update flooded vlans (for RSPAN). */
5234 if (br->cfg->n_flood_vlans) {
5235 rspan_vlans = bitmap_allocate(4096);
5237 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
5238 int64_t vlan = br->cfg->flood_vlans[i];
5239 if (vlan >= 0 && vlan < 4096) {
5240 bitmap_set1(rspan_vlans, vlan);
5241 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
5244 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
5249 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
5255 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
5260 for (i = 0; ; i++) {
5261 if (i >= MAX_MIRRORS) {
5262 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
5263 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
5266 if (!br->mirrors[i]) {
5271 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
5274 br->mirrors[i] = m = xzalloc(sizeof *m);
5277 m->name = xstrdup(cfg->name);
5278 shash_init(&m->src_ports);
5279 shash_init(&m->dst_ports);
5285 mirror_reconfigure_one(m, cfg);
5289 mirror_destroy(struct mirror *m)
5292 struct bridge *br = m->bridge;
5295 for (i = 0; i < br->n_ports; i++) {
5296 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
5297 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
5300 shash_destroy(&m->src_ports);
5301 shash_destroy(&m->dst_ports);
5304 m->bridge->mirrors[m->idx] = NULL;
5313 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
5314 struct shash *names)
5318 for (i = 0; i < n_ports; i++) {
5319 const char *name = ports[i]->name;
5320 if (port_lookup(m->bridge, name)) {
5321 shash_add_once(names, name, NULL);
5323 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
5324 "port %s", m->bridge->name, m->name, name);
5330 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
5336 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
5338 for (i = 0; i < cfg->n_select_vlan; i++) {
5339 int64_t vlan = cfg->select_vlan[i];
5340 if (vlan < 0 || vlan > 4095) {
5341 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
5342 m->bridge->name, m->name, vlan);
5344 (*vlans)[n_vlans++] = vlan;
5351 vlan_is_mirrored(const struct mirror *m, int vlan)
5355 for (i = 0; i < m->n_vlans; i++) {
5356 if (m->vlans[i] == vlan) {
5364 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
5368 for (i = 0; i < m->n_vlans; i++) {
5369 if (port_trunks_vlan(p, m->vlans[i])) {
5377 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
5379 struct shash src_ports, dst_ports;
5380 mirror_mask_t mirror_bit;
5381 struct port *out_port;
5388 if (strcmp(cfg->name, m->name)) {
5390 m->name = xstrdup(cfg->name);
5393 /* Get output port. */
5394 if (cfg->output_port) {
5395 out_port = port_lookup(m->bridge, cfg->output_port->name);
5397 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
5398 m->bridge->name, m->name);
5404 if (cfg->output_vlan) {
5405 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
5406 "output vlan; ignoring output vlan",
5407 m->bridge->name, m->name);
5409 } else if (cfg->output_vlan) {
5411 out_vlan = *cfg->output_vlan;
5413 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
5414 m->bridge->name, m->name);
5419 shash_init(&src_ports);
5420 shash_init(&dst_ports);
5421 if (cfg->select_all) {
5422 for (i = 0; i < m->bridge->n_ports; i++) {
5423 const char *name = m->bridge->ports[i]->name;
5424 shash_add_once(&src_ports, name, NULL);
5425 shash_add_once(&dst_ports, name, NULL);
5430 /* Get ports, and drop duplicates and ports that don't exist. */
5431 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
5433 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
5436 /* Get all the vlans, and drop duplicate and invalid vlans. */
5437 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
5440 /* Update mirror data. */
5441 if (!shash_equal_keys(&m->src_ports, &src_ports)
5442 || !shash_equal_keys(&m->dst_ports, &dst_ports)
5443 || m->n_vlans != n_vlans
5444 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
5445 || m->out_port != out_port
5446 || m->out_vlan != out_vlan) {
5447 bridge_flush(m->bridge);
5449 shash_swap(&m->src_ports, &src_ports);
5450 shash_swap(&m->dst_ports, &dst_ports);
5453 m->n_vlans = n_vlans;
5454 m->out_port = out_port;
5455 m->out_vlan = out_vlan;
5458 mirror_bit = MIRROR_MASK_C(1) << m->idx;
5459 for (i = 0; i < m->bridge->n_ports; i++) {
5460 struct port *port = m->bridge->ports[i];
5462 if (shash_find(&m->src_ports, port->name)
5465 ? port_trunks_any_mirrored_vlan(m, port)
5466 : vlan_is_mirrored(m, port->vlan)))) {
5467 port->src_mirrors |= mirror_bit;
5469 port->src_mirrors &= ~mirror_bit;
5472 if (shash_find(&m->dst_ports, port->name)) {
5473 port->dst_mirrors |= mirror_bit;
5475 port->dst_mirrors &= ~mirror_bit;
5480 shash_destroy(&src_ports);
5481 shash_destroy(&dst_ports);