1 /* Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
18 #include "byte-order.h"
21 #include <arpa/inet.h>
24 #include <sys/socket.h>
26 #include <openflow/openflow.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
36 #include "classifier.h"
41 #include "dynamic-string.h"
48 #include "mac-learning.h"
52 #include "ofp-print.h"
54 #include "ofproto/netflow.h"
55 #include "ofproto/ofproto.h"
56 #include "ovsdb-data.h"
58 #include "poll-loop.h"
62 #include "socket-util.h"
63 #include "stream-ssl.h"
66 #include "system-stats.h"
71 #include "vswitchd/vswitch-idl.h"
72 #include "xenserver.h"
74 #include "sflow_api.h"
76 VLOG_DEFINE_THIS_MODULE(bridge);
78 COVERAGE_DEFINE(bridge_flush);
79 COVERAGE_DEFINE(bridge_process_flow);
80 COVERAGE_DEFINE(bridge_process_cfm);
81 COVERAGE_DEFINE(bridge_process_lacp);
82 COVERAGE_DEFINE(bridge_reconfigure);
83 COVERAGE_DEFINE(bridge_lacp_update);
91 struct dst builtin[32];
96 static void dst_set_init(struct dst_set *);
97 static void dst_set_add(struct dst_set *, const struct dst *);
98 static void dst_set_free(struct dst_set *);
101 /* These members are always valid. */
102 struct list port_elem; /* Element in struct port's "ifaces" list. */
103 struct port *port; /* Containing port. */
104 char *name; /* Host network device name. */
105 tag_type tag; /* Tag associated with this interface. */
106 long long delay_expires; /* Time after which 'enabled' may change. */
108 /* These members are valid only after bridge_reconfigure() causes them to
110 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
111 int dp_ifidx; /* Index within kernel datapath. */
112 struct netdev *netdev; /* Network device. */
113 bool enabled; /* May be chosen for flows? */
114 bool up; /* Is the interface up? */
115 const char *type; /* Usually same as cfg->type. */
116 const struct ovsrec_interface *cfg;
118 /* LACP information. */
119 uint16_t lacp_priority; /* LACP port priority. */
122 #define BOND_MASK 0xff
124 struct iface *iface; /* Assigned iface, or NULL if none. */
125 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
126 tag_type tag; /* Tag for bond_entry<->iface association. */
130 BM_TCP, /* Transport Layer Load Balance. */
131 BM_SLB, /* Source Load Balance. */
132 BM_AB /* Active Backup. */
135 #define MAX_MIRRORS 32
136 typedef uint32_t mirror_mask_t;
137 #define MIRROR_MASK_C(X) UINT32_C(X)
138 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
140 struct bridge *bridge;
143 struct uuid uuid; /* UUID of this "mirror" record in database. */
145 /* Selection criteria. */
146 struct sset src_ports; /* Source port names. */
147 struct sset dst_ports; /* Destination port names. */
152 struct port *out_port;
156 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
158 struct bridge *bridge;
159 struct hmap_node hmap_node; /* Element in struct bridge's "ports" hmap. */
162 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
163 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
164 * NULL if all VLANs are trunked. */
165 const struct ovsrec_port *cfg;
168 struct netdev_monitor *monitor; /* Tracks carrier. NULL if miimon. */
169 long long int miimon_interval; /* Miimon status refresh interval. */
170 long long int miimon_next_update; /* Time of next miimon update. */
172 /* An ordinary bridge port has 1 interface.
173 * A bridge port for bonding has at least 2 interfaces. */
174 struct list ifaces; /* List of "struct iface"s. */
175 size_t n_ifaces; /* list_size(ifaces). */
178 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
179 struct iface *active_iface; /* iface on which bcasts accepted, or NULL. */
180 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
181 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
182 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
183 long long int bond_next_fake_iface_update; /* Time of next update. */
185 /* LACP information. */
186 struct lacp *lacp; /* LACP object. NULL if LACP is disabled. */
187 bool lacp_active; /* True if LACP is active */
188 bool lacp_fast; /* True if LACP is in fast mode. */
189 uint16_t lacp_priority; /* LACP system priority. */
191 /* SLB specific bonding info. */
192 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
193 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
194 long long int bond_next_rebalance; /* Next rebalancing time. */
196 /* Port mirroring info. */
197 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
198 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
199 bool is_mirror_output_port; /* Does port mirroring send frames here? */
203 struct list node; /* Node in global list of bridges. */
204 char *name; /* User-specified arbitrary name. */
205 struct mac_learning *ml; /* MAC learning table. */
206 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
207 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
208 const struct ovsrec_bridge *cfg;
210 /* OpenFlow switch processing. */
211 struct ofproto *ofproto; /* OpenFlow switch. */
213 /* Kernel datapath information. */
214 struct dpif *dpif; /* Datapath. */
215 struct hmap ifaces; /* Contains "struct iface"s. */
218 struct hmap ports; /* "struct port"s indexed by name. */
219 struct shash iface_by_name; /* "struct iface"s indexed by name. */
222 bool has_bonded_ports;
227 /* Port mirroring. */
228 struct mirror *mirrors[MAX_MIRRORS];
230 /* Synthetic local port if necessary. */
231 struct ovsrec_port synth_local_port;
232 struct ovsrec_interface synth_local_iface;
233 struct ovsrec_interface *synth_local_ifacep;
236 /* List of all bridges. */
237 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
239 /* OVSDB IDL used to obtain configuration. */
240 static struct ovsdb_idl *idl;
242 /* Each time this timer expires, the bridge fetches systems and interface
243 * statistics and pushes them into the database. */
244 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
245 static long long int stats_timer = LLONG_MIN;
247 /* Stores the time after which CFM statistics may be written to the database.
248 * Only updated when changes to the database require rate limiting. */
249 #define CFM_LIMIT_INTERVAL (1 * 1000) /* In milliseconds. */
250 static long long int cfm_limiter = LLONG_MIN;
252 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
253 static void bridge_destroy(struct bridge *);
254 static struct bridge *bridge_lookup(const char *name);
255 static unixctl_cb_func bridge_unixctl_dump_flows;
256 static unixctl_cb_func bridge_unixctl_reconnect;
257 static int bridge_run_one(struct bridge *);
258 static size_t bridge_get_controllers(const struct bridge *br,
259 struct ovsrec_controller ***controllersp);
260 static void bridge_reconfigure_one(struct bridge *);
261 static void bridge_reconfigure_remotes(struct bridge *,
262 const struct sockaddr_in *managers,
264 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
265 static void bridge_fetch_dp_ifaces(struct bridge *);
266 static void bridge_flush(struct bridge *);
267 static void bridge_pick_local_hw_addr(struct bridge *,
268 uint8_t ea[ETH_ADDR_LEN],
269 struct iface **hw_addr_iface);
270 static uint64_t bridge_pick_datapath_id(struct bridge *,
271 const uint8_t bridge_ea[ETH_ADDR_LEN],
272 struct iface *hw_addr_iface);
273 static uint64_t dpid_from_hash(const void *, size_t nbytes);
275 static unixctl_cb_func bridge_unixctl_fdb_show;
276 static unixctl_cb_func cfm_unixctl_show;
277 static unixctl_cb_func qos_unixctl_show;
279 static void bond_init(void);
280 static void bond_run(struct port *);
281 static void bond_wait(struct port *);
282 static void bond_rebalance_port(struct port *);
283 static void bond_send_learning_packets(struct port *);
284 static void bond_enable_slave(struct iface *iface, bool enable);
286 static void port_run(struct port *);
287 static void port_wait(struct port *);
288 static struct port *port_create(struct bridge *, const char *name);
289 static void port_reconfigure(struct port *, const struct ovsrec_port *);
290 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
291 static void port_destroy(struct port *);
292 static struct port *port_lookup(const struct bridge *, const char *name);
293 static struct iface *port_lookup_iface(const struct port *, const char *name);
294 static struct iface *port_get_an_iface(const struct port *);
295 static struct port *port_from_dp_ifidx(const struct bridge *,
297 static void port_update_bonding(struct port *);
298 static void port_update_lacp(struct port *);
300 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
301 static void mirror_destroy(struct mirror *);
302 static void mirror_reconfigure(struct bridge *);
303 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
304 static bool vlan_is_mirrored(const struct mirror *, int vlan);
306 static struct iface *iface_create(struct port *port,
307 const struct ovsrec_interface *if_cfg);
308 static void iface_destroy(struct iface *);
309 static struct iface *iface_lookup(const struct bridge *, const char *name);
310 static struct iface *iface_find(const char *name);
311 static struct iface *iface_from_dp_ifidx(const struct bridge *,
313 static void iface_set_mac(struct iface *);
314 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
315 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
316 static void iface_update_cfm(struct iface *);
317 static bool iface_refresh_cfm_stats(struct iface *iface);
318 static void iface_update_carrier(struct iface *);
319 static bool iface_get_carrier(const struct iface *);
320 static bool iface_is_synthetic(const struct iface *);
322 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
324 static void shash_to_ovs_idl_map(struct shash *,
325 char ***keys, char ***values, size_t *n);
327 /* Hooks into ofproto processing. */
328 static struct ofhooks bridge_ofhooks;
330 /* Public functions. */
332 /* Initializes the bridge module, configuring it to obtain its configuration
333 * from an OVSDB server accessed over 'remote', which should be a string in a
334 * form acceptable to ovsdb_idl_create(). */
336 bridge_init(const char *remote)
338 /* Create connection to database. */
339 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
341 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
342 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
343 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
344 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_ovs_version);
345 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_db_version);
346 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_system_type);
347 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_system_version);
349 ovsdb_idl_omit_alert(idl, &ovsrec_bridge_col_datapath_id);
350 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
352 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
353 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
355 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_admin_state);
356 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_duplex);
357 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_link_speed);
358 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_link_state);
359 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_mtu);
360 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
361 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
362 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_status);
363 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
365 ovsdb_idl_omit_alert(idl, &ovsrec_controller_col_is_connected);
366 ovsdb_idl_omit_alert(idl, &ovsrec_controller_col_role);
367 ovsdb_idl_omit_alert(idl, &ovsrec_controller_col_status);
368 ovsdb_idl_omit(idl, &ovsrec_controller_col_external_ids);
370 ovsdb_idl_omit_alert(idl, &ovsrec_maintenance_point_col_fault);
372 ovsdb_idl_omit_alert(idl, &ovsrec_monitor_col_fault);
374 ovsdb_idl_omit(idl, &ovsrec_qos_col_external_ids);
376 ovsdb_idl_omit(idl, &ovsrec_queue_col_external_ids);
378 ovsdb_idl_omit(idl, &ovsrec_mirror_col_external_ids);
380 ovsdb_idl_omit(idl, &ovsrec_netflow_col_external_ids);
382 ovsdb_idl_omit(idl, &ovsrec_sflow_col_external_ids);
384 ovsdb_idl_omit(idl, &ovsrec_manager_col_external_ids);
385 ovsdb_idl_omit(idl, &ovsrec_manager_col_inactivity_probe);
386 ovsdb_idl_omit(idl, &ovsrec_manager_col_is_connected);
387 ovsdb_idl_omit(idl, &ovsrec_manager_col_max_backoff);
388 ovsdb_idl_omit(idl, &ovsrec_manager_col_status);
390 ovsdb_idl_omit(idl, &ovsrec_ssl_col_external_ids);
392 /* Register unixctl commands. */
393 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
394 unixctl_command_register("cfm/show", cfm_unixctl_show, NULL);
395 unixctl_command_register("qos/show", qos_unixctl_show, NULL);
396 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
398 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
407 struct bridge *br, *next_br;
409 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
412 ovsdb_idl_destroy(idl);
415 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
416 * but for which the ovs-vswitchd configuration 'cfg' is required. */
418 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
420 static bool already_configured_once;
421 struct sset bridge_names;
422 struct sset dpif_names, dpif_types;
426 /* Only do this once per ovs-vswitchd run. */
427 if (already_configured_once) {
430 already_configured_once = true;
432 stats_timer = time_msec() + STATS_INTERVAL;
434 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
435 sset_init(&bridge_names);
436 for (i = 0; i < cfg->n_bridges; i++) {
437 sset_add(&bridge_names, cfg->bridges[i]->name);
440 /* Iterate over all system dpifs and delete any of them that do not appear
442 sset_init(&dpif_names);
443 sset_init(&dpif_types);
444 dp_enumerate_types(&dpif_types);
445 SSET_FOR_EACH (type, &dpif_types) {
448 dp_enumerate_names(type, &dpif_names);
450 /* Delete each dpif whose name is not in 'bridge_names'. */
451 SSET_FOR_EACH (name, &dpif_names) {
452 if (!sset_contains(&bridge_names, name)) {
456 retval = dpif_open(name, type, &dpif);
464 sset_destroy(&bridge_names);
465 sset_destroy(&dpif_names);
466 sset_destroy(&dpif_types);
469 /* Callback for iterate_and_prune_ifaces(). */
471 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
473 if (!iface->netdev) {
474 /* We already reported a related error, don't bother duplicating it. */
478 if (iface->dp_ifidx < 0) {
479 VLOG_ERR("%s interface not in %s, dropping",
480 iface->name, dpif_name(br->dpif));
484 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
485 iface->name, iface->dp_ifidx);
489 /* Callback for iterate_and_prune_ifaces(). */
491 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
492 void *aux OVS_UNUSED)
494 /* Set policing attributes. */
495 netdev_set_policing(iface->netdev,
496 iface->cfg->ingress_policing_rate,
497 iface->cfg->ingress_policing_burst);
499 /* Set MAC address of internal interfaces other than the local
501 iface_set_mac(iface);
506 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
507 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
508 * deletes from 'br' any ports that no longer have any interfaces. */
510 iterate_and_prune_ifaces(struct bridge *br,
511 bool (*cb)(struct bridge *, struct iface *,
515 struct port *port, *next_port;
517 HMAP_FOR_EACH_SAFE (port, next_port, hmap_node, &br->ports) {
518 struct iface *iface, *next_iface;
520 LIST_FOR_EACH_SAFE (iface, next_iface, port_elem, &port->ifaces) {
521 if (!cb(br, iface, aux)) {
522 iface_set_ofport(iface->cfg, -1);
523 iface_destroy(iface);
527 if (!port->n_ifaces) {
528 VLOG_WARN("%s port has no interfaces, dropping", port->name);
534 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
535 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
536 * responsible for freeing '*managersp' (with free()).
538 * You may be asking yourself "why does ovs-vswitchd care?", because
539 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
540 * should not be and in fact is not directly involved in that. But
541 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
542 * it has to tell in-band control where the managers are to enable that.
543 * (Thus, only managers connected in-band are collected.)
546 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
547 struct sockaddr_in **managersp, size_t *n_managersp)
549 struct sockaddr_in *managers = NULL;
550 size_t n_managers = 0;
554 /* Collect all of the potential targets from the "targets" columns of the
555 * rows pointed to by "manager_options", excluding any that are
558 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
559 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
561 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
562 sset_find_and_delete(&targets, m->target);
564 sset_add(&targets, m->target);
568 /* Now extract the targets' IP addresses. */
569 if (!sset_is_empty(&targets)) {
572 managers = xmalloc(sset_count(&targets) * sizeof *managers);
573 SSET_FOR_EACH (target, &targets) {
574 struct sockaddr_in *sin = &managers[n_managers];
576 if ((!strncmp(target, "tcp:", 4)
577 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
578 (!strncmp(target, "ssl:", 4)
579 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
584 sset_destroy(&targets);
586 *managersp = managers;
587 *n_managersp = n_managers;
591 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
593 struct shash old_br, new_br;
594 struct shash_node *node;
595 struct bridge *br, *next;
596 struct sockaddr_in *managers;
599 int sflow_bridge_number;
601 COVERAGE_INC(bridge_reconfigure);
603 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
605 /* Collect old and new bridges. */
608 LIST_FOR_EACH (br, node, &all_bridges) {
609 shash_add(&old_br, br->name, br);
611 for (i = 0; i < ovs_cfg->n_bridges; i++) {
612 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
613 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
614 VLOG_WARN("more than one bridge named %s", br_cfg->name);
618 /* Get rid of deleted bridges and add new bridges. */
619 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
620 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
627 SHASH_FOR_EACH (node, &new_br) {
628 const char *br_name = node->name;
629 const struct ovsrec_bridge *br_cfg = node->data;
630 br = shash_find_data(&old_br, br_name);
632 /* If the bridge datapath type has changed, we need to tear it
633 * down and recreate. */
634 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
636 bridge_create(br_cfg);
639 bridge_create(br_cfg);
642 shash_destroy(&old_br);
643 shash_destroy(&new_br);
645 /* Reconfigure all bridges. */
646 LIST_FOR_EACH (br, node, &all_bridges) {
647 bridge_reconfigure_one(br);
650 /* Add and delete ports on all datapaths.
652 * The kernel will reject any attempt to add a given port to a datapath if
653 * that port already belongs to a different datapath, so we must do all
654 * port deletions before any port additions. */
655 LIST_FOR_EACH (br, node, &all_bridges) {
656 struct dpif_port_dump dump;
657 struct shash want_ifaces;
658 struct dpif_port dpif_port;
660 bridge_get_all_ifaces(br, &want_ifaces);
661 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
662 if (!shash_find(&want_ifaces, dpif_port.name)
663 && strcmp(dpif_port.name, br->name)) {
664 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
666 VLOG_WARN("failed to remove %s interface from %s: %s",
667 dpif_port.name, dpif_name(br->dpif),
672 shash_destroy(&want_ifaces);
674 LIST_FOR_EACH (br, node, &all_bridges) {
675 struct shash cur_ifaces, want_ifaces;
676 struct dpif_port_dump dump;
677 struct dpif_port dpif_port;
679 /* Get the set of interfaces currently in this datapath. */
680 shash_init(&cur_ifaces);
681 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
682 struct dpif_port *port_info = xmalloc(sizeof *port_info);
683 dpif_port_clone(port_info, &dpif_port);
684 shash_add(&cur_ifaces, dpif_port.name, port_info);
687 /* Get the set of interfaces we want on this datapath. */
688 bridge_get_all_ifaces(br, &want_ifaces);
690 hmap_clear(&br->ifaces);
691 SHASH_FOR_EACH (node, &want_ifaces) {
692 const char *if_name = node->name;
693 struct iface *iface = node->data;
694 struct dpif_port *dpif_port;
698 type = iface ? iface->type : "internal";
699 dpif_port = shash_find_data(&cur_ifaces, if_name);
701 /* If we have a port or a netdev already, and it's not the type we
702 * want, then delete the port (if any) and close the netdev (if
704 if ((dpif_port && strcmp(dpif_port->type, type))
705 || (iface && iface->netdev
706 && strcmp(type, netdev_get_type(iface->netdev)))) {
708 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
715 netdev_close(iface->netdev);
716 iface->netdev = NULL;
720 /* If the port doesn't exist or we don't have the netdev open,
721 * we need to do more work. */
722 if (!dpif_port || (iface && !iface->netdev)) {
723 struct netdev_options options;
724 struct netdev *netdev;
727 /* First open the network device. */
728 options.name = if_name;
730 options.args = &args;
731 options.ethertype = NETDEV_ETH_TYPE_NONE;
735 shash_from_ovs_idl_map(iface->cfg->key_options,
736 iface->cfg->value_options,
737 iface->cfg->n_options, &args);
739 error = netdev_open(&options, &netdev);
740 shash_destroy(&args);
743 VLOG_WARN("could not open network device %s (%s)",
744 if_name, strerror(error));
748 /* Then add the port if we haven't already. */
750 error = dpif_port_add(br->dpif, netdev, NULL);
752 netdev_close(netdev);
753 if (error == EFBIG) {
754 VLOG_ERR("ran out of valid port numbers on %s",
755 dpif_name(br->dpif));
758 VLOG_WARN("failed to add %s interface to %s: %s",
759 if_name, dpif_name(br->dpif),
766 /* Update 'iface'. */
768 iface->netdev = netdev;
769 iface->enabled = iface_get_carrier(iface);
770 iface->up = iface->enabled;
772 } else if (iface && iface->netdev) {
776 shash_from_ovs_idl_map(iface->cfg->key_options,
777 iface->cfg->value_options,
778 iface->cfg->n_options, &args);
779 netdev_set_config(iface->netdev, &args);
780 shash_destroy(&args);
783 shash_destroy(&want_ifaces);
785 SHASH_FOR_EACH (node, &cur_ifaces) {
786 struct dpif_port *port_info = node->data;
787 dpif_port_destroy(port_info);
790 shash_destroy(&cur_ifaces);
792 sflow_bridge_number = 0;
793 LIST_FOR_EACH (br, node, &all_bridges) {
796 struct iface *local_iface;
797 struct iface *hw_addr_iface;
800 bridge_fetch_dp_ifaces(br);
802 iterate_and_prune_ifaces(br, check_iface, NULL);
804 /* Pick local port hardware address, datapath ID. */
805 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
806 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
808 int error = netdev_set_etheraddr(local_iface->netdev, ea);
810 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
811 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
812 "Ethernet address: %s",
813 br->name, strerror(error));
816 memcpy(br->ea, ea, ETH_ADDR_LEN);
818 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
819 ofproto_set_datapath_id(br->ofproto, dpid);
821 dpid_string = xasprintf("%016"PRIx64, dpid);
822 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
825 /* Set NetFlow configuration on this bridge. */
826 if (br->cfg->netflow) {
827 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
828 struct netflow_options opts;
830 memset(&opts, 0, sizeof opts);
832 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
833 if (nf_cfg->engine_type) {
834 opts.engine_type = *nf_cfg->engine_type;
836 if (nf_cfg->engine_id) {
837 opts.engine_id = *nf_cfg->engine_id;
840 opts.active_timeout = nf_cfg->active_timeout;
841 if (!opts.active_timeout) {
842 opts.active_timeout = -1;
843 } else if (opts.active_timeout < 0) {
844 VLOG_WARN("bridge %s: active timeout interval set to negative "
845 "value, using default instead (%d seconds)", br->name,
846 NF_ACTIVE_TIMEOUT_DEFAULT);
847 opts.active_timeout = -1;
850 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
851 if (opts.add_id_to_iface) {
852 if (opts.engine_id > 0x7f) {
853 VLOG_WARN("bridge %s: netflow port mangling may conflict "
854 "with another vswitch, choose an engine id less "
855 "than 128", br->name);
857 if (hmap_count(&br->ports) > 508) {
858 VLOG_WARN("bridge %s: netflow port mangling will conflict "
859 "with another port when more than 508 ports are "
864 sset_init(&opts.collectors);
865 sset_add_array(&opts.collectors,
866 nf_cfg->targets, nf_cfg->n_targets);
867 if (ofproto_set_netflow(br->ofproto, &opts)) {
868 VLOG_ERR("bridge %s: problem setting netflow collectors",
871 sset_destroy(&opts.collectors);
873 ofproto_set_netflow(br->ofproto, NULL);
876 /* Set sFlow configuration on this bridge. */
877 if (br->cfg->sflow) {
878 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
879 struct ovsrec_controller **controllers;
880 struct ofproto_sflow_options oso;
881 size_t n_controllers;
883 memset(&oso, 0, sizeof oso);
885 sset_init(&oso.targets);
886 sset_add_array(&oso.targets,
887 sflow_cfg->targets, sflow_cfg->n_targets);
889 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
890 if (sflow_cfg->sampling) {
891 oso.sampling_rate = *sflow_cfg->sampling;
894 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
895 if (sflow_cfg->polling) {
896 oso.polling_interval = *sflow_cfg->polling;
899 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
900 if (sflow_cfg->header) {
901 oso.header_len = *sflow_cfg->header;
904 oso.sub_id = sflow_bridge_number++;
905 oso.agent_device = sflow_cfg->agent;
907 oso.control_ip = NULL;
908 n_controllers = bridge_get_controllers(br, &controllers);
909 for (i = 0; i < n_controllers; i++) {
910 if (controllers[i]->local_ip) {
911 oso.control_ip = controllers[i]->local_ip;
915 ofproto_set_sflow(br->ofproto, &oso);
917 sset_destroy(&oso.targets);
919 ofproto_set_sflow(br->ofproto, NULL);
922 /* Update the controller and related settings. It would be more
923 * straightforward to call this from bridge_reconfigure_one(), but we
924 * can't do it there for two reasons. First, and most importantly, at
925 * that point we don't know the dp_ifidx of any interfaces that have
926 * been added to the bridge (because we haven't actually added them to
927 * the datapath). Second, at that point we haven't set the datapath ID
928 * yet; when a controller is configured, resetting the datapath ID will
929 * immediately disconnect from the controller, so it's better to set
930 * the datapath ID before the controller. */
931 bridge_reconfigure_remotes(br, managers, n_managers);
933 LIST_FOR_EACH (br, node, &all_bridges) {
936 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
940 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
941 netdev_monitor_add(port->monitor, iface->netdev);
944 port->miimon_next_update = 0;
947 port_update_lacp(port);
948 port_update_bonding(port);
950 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
951 iface_update_qos(iface, port->cfg->qos);
955 LIST_FOR_EACH (br, node, &all_bridges) {
956 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
959 /* Some reconfiguration operations require the bridge to have been run at
961 LIST_FOR_EACH (br, node, &all_bridges) {
966 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
967 iface_update_cfm(iface);
973 /* ovs-vswitchd has completed initialization, so allow the process that
974 * forked us to exit successfully. */
975 daemonize_complete();
979 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
980 const struct ovsdb_idl_column *column,
983 const struct ovsdb_datum *datum;
984 union ovsdb_atom atom;
987 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
988 atom.string = (char *) key;
989 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
990 return idx == UINT_MAX ? NULL : datum->values[idx].string;
994 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
996 return get_ovsrec_key_value(&br_cfg->header_,
997 &ovsrec_bridge_col_other_config, key);
1001 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
1002 struct iface **hw_addr_iface)
1008 *hw_addr_iface = NULL;
1010 /* Did the user request a particular MAC? */
1011 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
1012 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
1013 if (eth_addr_is_multicast(ea)) {
1014 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
1015 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1016 } else if (eth_addr_is_zero(ea)) {
1017 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
1023 /* Otherwise choose the minimum non-local MAC address among all of the
1025 memset(ea, 0xff, ETH_ADDR_LEN);
1026 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1027 uint8_t iface_ea[ETH_ADDR_LEN];
1028 struct iface *candidate;
1029 struct iface *iface;
1031 /* Mirror output ports don't participate. */
1032 if (port->is_mirror_output_port) {
1036 /* Choose the MAC address to represent the port. */
1038 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
1039 /* Find the interface with this Ethernet address (if any) so that
1040 * we can provide the correct devname to the caller. */
1041 LIST_FOR_EACH (candidate, port_elem, &port->ifaces) {
1042 uint8_t candidate_ea[ETH_ADDR_LEN];
1043 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
1044 && eth_addr_equals(iface_ea, candidate_ea)) {
1049 /* Choose the interface whose MAC address will represent the port.
1050 * The Linux kernel bonding code always chooses the MAC address of
1051 * the first slave added to a bond, and the Fedora networking
1052 * scripts always add slaves to a bond in alphabetical order, so
1053 * for compatibility we choose the interface with the name that is
1054 * first in alphabetical order. */
1055 LIST_FOR_EACH (candidate, port_elem, &port->ifaces) {
1056 if (!iface || strcmp(candidate->name, iface->name) < 0) {
1061 /* The local port doesn't count (since we're trying to choose its
1062 * MAC address anyway). */
1063 if (iface->dp_ifidx == ODPP_LOCAL) {
1068 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1070 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1071 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1072 iface->name, strerror(error));
1077 /* Compare against our current choice. */
1078 if (!eth_addr_is_multicast(iface_ea) &&
1079 !eth_addr_is_local(iface_ea) &&
1080 !eth_addr_is_reserved(iface_ea) &&
1081 !eth_addr_is_zero(iface_ea) &&
1082 eth_addr_compare_3way(iface_ea, ea) < 0)
1084 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1085 *hw_addr_iface = iface;
1088 if (eth_addr_is_multicast(ea)) {
1089 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1090 *hw_addr_iface = NULL;
1091 VLOG_WARN("bridge %s: using default bridge Ethernet "
1092 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1094 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1095 br->name, ETH_ADDR_ARGS(ea));
1099 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1100 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1101 * an interface on 'br', then that interface must be passed in as
1102 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1103 * 'hw_addr_iface' must be passed in as a null pointer. */
1105 bridge_pick_datapath_id(struct bridge *br,
1106 const uint8_t bridge_ea[ETH_ADDR_LEN],
1107 struct iface *hw_addr_iface)
1110 * The procedure for choosing a bridge MAC address will, in the most
1111 * ordinary case, also choose a unique MAC that we can use as a datapath
1112 * ID. In some special cases, though, multiple bridges will end up with
1113 * the same MAC address. This is OK for the bridges, but it will confuse
1114 * the OpenFlow controller, because each datapath needs a unique datapath
1117 * Datapath IDs must be unique. It is also very desirable that they be
1118 * stable from one run to the next, so that policy set on a datapath
1121 const char *datapath_id;
1124 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1125 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1129 if (hw_addr_iface) {
1131 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1133 * A bridge whose MAC address is taken from a VLAN network device
1134 * (that is, a network device created with vconfig(8) or similar
1135 * tool) will have the same MAC address as a bridge on the VLAN
1136 * device's physical network device.
1138 * Handle this case by hashing the physical network device MAC
1139 * along with the VLAN identifier.
1141 uint8_t buf[ETH_ADDR_LEN + 2];
1142 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1143 buf[ETH_ADDR_LEN] = vlan >> 8;
1144 buf[ETH_ADDR_LEN + 1] = vlan;
1145 return dpid_from_hash(buf, sizeof buf);
1148 * Assume that this bridge's MAC address is unique, since it
1149 * doesn't fit any of the cases we handle specially.
1154 * A purely internal bridge, that is, one that has no non-virtual
1155 * network devices on it at all, is more difficult because it has no
1156 * natural unique identifier at all.
1158 * When the host is a XenServer, we handle this case by hashing the
1159 * host's UUID with the name of the bridge. Names of bridges are
1160 * persistent across XenServer reboots, although they can be reused if
1161 * an internal network is destroyed and then a new one is later
1162 * created, so this is fairly effective.
1164 * When the host is not a XenServer, we punt by using a random MAC
1165 * address on each run.
1167 const char *host_uuid = xenserver_get_host_uuid();
1169 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1170 dpid = dpid_from_hash(combined, strlen(combined));
1176 return eth_addr_to_uint64(bridge_ea);
1180 dpid_from_hash(const void *data, size_t n)
1182 uint8_t hash[SHA1_DIGEST_SIZE];
1184 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1185 sha1_bytes(data, n, hash);
1186 eth_addr_mark_random(hash);
1187 return eth_addr_to_uint64(hash);
1191 iface_refresh_status(struct iface *iface)
1195 enum netdev_flags flags;
1202 if (iface_is_synthetic(iface)) {
1208 if (!netdev_get_status(iface->netdev, &sh)) {
1210 char **keys, **values;
1212 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1213 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1218 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1221 shash_destroy_free_data(&sh);
1223 error = netdev_get_flags(iface->netdev, &flags);
1225 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1228 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1231 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1233 ovsrec_interface_set_duplex(iface->cfg,
1234 netdev_features_is_full_duplex(current)
1236 /* warning: uint64_t -> int64_t conversion */
1237 bps = netdev_features_to_bps(current);
1238 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1241 ovsrec_interface_set_duplex(iface->cfg, NULL);
1242 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1246 ovsrec_interface_set_link_state(iface->cfg,
1247 iface_get_carrier(iface) ? "up" : "down");
1249 error = netdev_get_mtu(iface->netdev, &mtu);
1250 if (!error && mtu != INT_MAX) {
1252 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1255 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1259 /* Writes 'iface''s CFM statistics to the database. Returns true if anything
1260 * changed, false otherwise. */
1262 iface_refresh_cfm_stats(struct iface *iface)
1264 const struct ovsrec_monitor *mon;
1265 const struct cfm *cfm;
1266 bool changed = false;
1269 mon = iface->cfg->monitor;
1270 cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
1276 for (i = 0; i < mon->n_remote_mps; i++) {
1277 const struct ovsrec_maintenance_point *mp;
1278 const struct remote_mp *rmp;
1280 mp = mon->remote_mps[i];
1281 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1283 if (mp->n_fault != 1 || mp->fault[0] != rmp->fault) {
1284 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1289 if (mon->n_fault != 1 || mon->fault[0] != cfm->fault) {
1290 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1298 iface_refresh_stats(struct iface *iface)
1304 static const struct iface_stat iface_stats[] = {
1305 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1306 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1307 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1308 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1309 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1310 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1311 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1312 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1313 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1314 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1315 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1316 { "collisions", offsetof(struct netdev_stats, collisions) },
1318 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1319 const struct iface_stat *s;
1321 char *keys[N_STATS];
1322 int64_t values[N_STATS];
1325 struct netdev_stats stats;
1327 if (iface_is_synthetic(iface)) {
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 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1351 struct ovsdb_datum datum;
1355 get_system_stats(&stats);
1357 ovsdb_datum_from_shash(&datum, &stats);
1358 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1362 static inline const char *
1363 nx_role_to_str(enum nx_role role)
1368 case NX_ROLE_MASTER:
1373 return "*** INVALID ROLE ***";
1378 bridge_refresh_controller_status(const struct bridge *br)
1381 const struct ovsrec_controller *cfg;
1383 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1385 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1386 struct ofproto_controller_info *cinfo =
1387 shash_find_data(&info, cfg->target);
1390 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1391 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1392 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1393 (char **) cinfo->pairs.values,
1396 ovsrec_controller_set_is_connected(cfg, false);
1397 ovsrec_controller_set_role(cfg, NULL);
1398 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1402 ofproto_free_ofproto_controller_info(&info);
1408 const struct ovsrec_open_vswitch *cfg;
1410 bool datapath_destroyed;
1411 bool database_changed;
1414 /* Let each bridge do the work that it needs to do. */
1415 datapath_destroyed = false;
1416 LIST_FOR_EACH (br, node, &all_bridges) {
1417 int error = bridge_run_one(br);
1419 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1420 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1421 "forcing reconfiguration", br->name);
1422 datapath_destroyed = true;
1426 /* (Re)configure if necessary. */
1427 database_changed = ovsdb_idl_run(idl);
1428 cfg = ovsrec_open_vswitch_first(idl);
1430 /* Re-configure SSL. We do this on every trip through the main loop,
1431 * instead of just when the database changes, because the contents of the
1432 * key and certificate files can change without the database changing.
1434 * We do this before bridge_reconfigure() because that function might
1435 * initiate SSL connections and thus requires SSL to be configured. */
1436 if (cfg && cfg->ssl) {
1437 const struct ovsrec_ssl *ssl = cfg->ssl;
1439 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1440 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1443 if (database_changed || datapath_destroyed) {
1445 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1447 bridge_configure_once(cfg);
1448 bridge_reconfigure(cfg);
1450 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1451 ovsdb_idl_txn_commit(txn);
1452 ovsdb_idl_txn_destroy(txn); /* XXX */
1454 /* We still need to reconfigure to avoid dangling pointers to
1455 * now-destroyed ovsrec structures inside bridge data. */
1456 static const struct ovsrec_open_vswitch null_cfg;
1458 bridge_reconfigure(&null_cfg);
1462 /* Refresh system and interface stats if necessary. */
1463 if (time_msec() >= stats_timer) {
1465 struct ovsdb_idl_txn *txn;
1467 txn = ovsdb_idl_txn_create(idl);
1468 LIST_FOR_EACH (br, node, &all_bridges) {
1471 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1472 struct iface *iface;
1474 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
1475 iface_refresh_stats(iface);
1476 iface_refresh_status(iface);
1479 bridge_refresh_controller_status(br);
1481 refresh_system_stats(cfg);
1482 ovsdb_idl_txn_commit(txn);
1483 ovsdb_idl_txn_destroy(txn); /* XXX */
1486 stats_timer = time_msec() + STATS_INTERVAL;
1489 if (time_msec() >= cfm_limiter) {
1490 struct ovsdb_idl_txn *txn;
1491 bool changed = false;
1493 txn = ovsdb_idl_txn_create(idl);
1494 LIST_FOR_EACH (br, node, &all_bridges) {
1497 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1498 struct iface *iface;
1500 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
1501 changed = iface_refresh_cfm_stats(iface) || changed;
1507 cfm_limiter = time_msec() + CFM_LIMIT_INTERVAL;
1510 ovsdb_idl_txn_commit(txn);
1511 ovsdb_idl_txn_destroy(txn);
1520 LIST_FOR_EACH (br, node, &all_bridges) {
1523 ofproto_wait(br->ofproto);
1524 mac_learning_wait(br->ml);
1525 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1529 ovsdb_idl_wait(idl);
1530 poll_timer_wait_until(stats_timer);
1532 if (cfm_limiter > time_msec()) {
1533 poll_timer_wait_until(cfm_limiter);
1537 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1538 * configuration changes. */
1540 bridge_flush(struct bridge *br)
1542 COVERAGE_INC(bridge_flush);
1546 /* Bridge unixctl user interface functions. */
1548 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1549 const char *args, void *aux OVS_UNUSED)
1551 struct ds ds = DS_EMPTY_INITIALIZER;
1552 const struct bridge *br;
1553 const struct mac_entry *e;
1555 br = bridge_lookup(args);
1557 unixctl_command_reply(conn, 501, "no such bridge");
1561 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1562 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1563 struct port *port = e->port.p;
1564 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1565 port_get_an_iface(port)->dp_ifidx,
1566 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1568 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1572 /* CFM unixctl user interface functions. */
1574 cfm_unixctl_show(struct unixctl_conn *conn,
1575 const char *args, void *aux OVS_UNUSED)
1577 struct ds ds = DS_EMPTY_INITIALIZER;
1578 struct iface *iface;
1579 const struct cfm *cfm;
1581 iface = iface_find(args);
1583 unixctl_command_reply(conn, 501, "no such interface");
1587 cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
1590 unixctl_command_reply(conn, 501, "CFM not enabled");
1594 cfm_dump_ds(cfm, &ds);
1595 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1599 /* QoS unixctl user interface functions. */
1601 struct qos_unixctl_show_cbdata {
1603 struct iface *iface;
1607 qos_unixctl_show_cb(unsigned int queue_id,
1608 const struct shash *details,
1611 struct qos_unixctl_show_cbdata *data = aux;
1612 struct ds *ds = data->ds;
1613 struct iface *iface = data->iface;
1614 struct netdev_queue_stats stats;
1615 struct shash_node *node;
1618 ds_put_cstr(ds, "\n");
1620 ds_put_format(ds, "Queue %u:\n", queue_id);
1622 ds_put_cstr(ds, "Default:\n");
1625 SHASH_FOR_EACH (node, details) {
1626 ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data);
1629 error = netdev_get_queue_stats(iface->netdev, queue_id, &stats);
1631 if (stats.tx_packets != UINT64_MAX) {
1632 ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets);
1635 if (stats.tx_bytes != UINT64_MAX) {
1636 ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes);
1639 if (stats.tx_errors != UINT64_MAX) {
1640 ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors);
1643 ds_put_format(ds, "\tFailed to get statistics for queue %u: %s",
1644 queue_id, strerror(error));
1649 qos_unixctl_show(struct unixctl_conn *conn,
1650 const char *args, void *aux OVS_UNUSED)
1652 struct ds ds = DS_EMPTY_INITIALIZER;
1653 struct shash sh = SHASH_INITIALIZER(&sh);
1654 struct iface *iface;
1656 struct shash_node *node;
1657 struct qos_unixctl_show_cbdata data;
1660 iface = iface_find(args);
1662 unixctl_command_reply(conn, 501, "no such interface");
1666 netdev_get_qos(iface->netdev, &type, &sh);
1668 if (*type != '\0') {
1669 ds_put_format(&ds, "QoS: %s %s\n", iface->name, type);
1671 SHASH_FOR_EACH (node, &sh) {
1672 ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data);
1677 error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data);
1680 ds_put_format(&ds, "failed to dump queues: %s", strerror(error));
1682 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1684 ds_put_format(&ds, "QoS not configured on %s\n", iface->name);
1685 unixctl_command_reply(conn, 501, ds_cstr(&ds));
1688 shash_destroy_free_data(&sh);
1692 /* Bridge reconfiguration functions. */
1693 static struct bridge *
1694 bridge_create(const struct ovsrec_bridge *br_cfg)
1699 assert(!bridge_lookup(br_cfg->name));
1700 br = xzalloc(sizeof *br);
1702 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1708 dpif_flow_flush(br->dpif);
1710 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1713 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1715 dpif_delete(br->dpif);
1716 dpif_close(br->dpif);
1721 br->name = xstrdup(br_cfg->name);
1723 br->ml = mac_learning_create();
1724 eth_addr_nicira_random(br->default_ea);
1726 hmap_init(&br->ports);
1727 hmap_init(&br->ifaces);
1728 shash_init(&br->iface_by_name);
1732 list_push_back(&all_bridges, &br->node);
1734 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1740 bridge_destroy(struct bridge *br)
1743 struct port *port, *next;
1747 HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) {
1750 for (i = 0; i < MAX_MIRRORS; i++) {
1751 mirror_destroy(br->mirrors[i]);
1753 list_remove(&br->node);
1754 ofproto_destroy(br->ofproto);
1755 error = dpif_delete(br->dpif);
1756 if (error && error != ENOENT) {
1757 VLOG_ERR("failed to delete %s: %s",
1758 dpif_name(br->dpif), strerror(error));
1760 dpif_close(br->dpif);
1761 mac_learning_destroy(br->ml);
1762 hmap_destroy(&br->ifaces);
1763 hmap_destroy(&br->ports);
1764 shash_destroy(&br->iface_by_name);
1765 free(br->synth_local_iface.type);
1771 static struct bridge *
1772 bridge_lookup(const char *name)
1776 LIST_FOR_EACH (br, node, &all_bridges) {
1777 if (!strcmp(br->name, name)) {
1784 /* Handle requests for a listing of all flows known by the OpenFlow
1785 * stack, including those normally hidden. */
1787 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1788 const char *args, void *aux OVS_UNUSED)
1793 br = bridge_lookup(args);
1795 unixctl_command_reply(conn, 501, "Unknown bridge");
1800 ofproto_get_all_flows(br->ofproto, &results);
1802 unixctl_command_reply(conn, 200, ds_cstr(&results));
1803 ds_destroy(&results);
1806 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1807 * connections and reconnect. If BRIDGE is not specified, then all bridges
1808 * drop their controller connections and reconnect. */
1810 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1811 const char *args, void *aux OVS_UNUSED)
1814 if (args[0] != '\0') {
1815 br = bridge_lookup(args);
1817 unixctl_command_reply(conn, 501, "Unknown bridge");
1820 ofproto_reconnect_controllers(br->ofproto);
1822 LIST_FOR_EACH (br, node, &all_bridges) {
1823 ofproto_reconnect_controllers(br->ofproto);
1826 unixctl_command_reply(conn, 200, NULL);
1830 bridge_run_one(struct bridge *br)
1835 error = ofproto_run1(br->ofproto);
1840 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1842 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1846 error = ofproto_run2(br->ofproto, br->flush);
1853 bridge_get_controllers(const struct bridge *br,
1854 struct ovsrec_controller ***controllersp)
1856 struct ovsrec_controller **controllers;
1857 size_t n_controllers;
1859 controllers = br->cfg->controller;
1860 n_controllers = br->cfg->n_controller;
1862 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1868 *controllersp = controllers;
1870 return n_controllers;
1874 bridge_reconfigure_one(struct bridge *br)
1876 enum ofproto_fail_mode fail_mode;
1877 struct port *port, *next;
1878 struct shash_node *node;
1879 struct shash new_ports;
1882 /* Collect new ports. */
1883 shash_init(&new_ports);
1884 for (i = 0; i < br->cfg->n_ports; i++) {
1885 const char *name = br->cfg->ports[i]->name;
1886 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1887 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1891 if (!shash_find(&new_ports, br->name)) {
1892 struct dpif_port dpif_port;
1895 VLOG_WARN("bridge %s: no port named %s, synthesizing one",
1896 br->name, br->name);
1898 dpif_port_query_by_number(br->dpif, ODPP_LOCAL, &dpif_port);
1899 type = xstrdup(dpif_port.type ? dpif_port.type : "internal");
1900 dpif_port_destroy(&dpif_port);
1902 br->synth_local_port.interfaces = &br->synth_local_ifacep;
1903 br->synth_local_port.n_interfaces = 1;
1904 br->synth_local_port.name = br->name;
1906 br->synth_local_iface.name = br->name;
1907 free(br->synth_local_iface.type);
1908 br->synth_local_iface.type = type;
1910 br->synth_local_ifacep = &br->synth_local_iface;
1912 shash_add(&new_ports, br->name, &br->synth_local_port);
1915 /* Get rid of deleted ports.
1916 * Get rid of deleted interfaces on ports that still exist. */
1917 HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) {
1918 const struct ovsrec_port *port_cfg;
1920 port_cfg = shash_find_data(&new_ports, port->name);
1924 port_del_ifaces(port, port_cfg);
1928 /* Create new ports.
1929 * Add new interfaces to existing ports.
1930 * Reconfigure existing ports. */
1931 SHASH_FOR_EACH (node, &new_ports) {
1932 struct port *port = port_lookup(br, node->name);
1934 port = port_create(br, node->name);
1937 port_reconfigure(port, node->data);
1938 if (!port->n_ifaces) {
1939 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1940 br->name, port->name);
1944 shash_destroy(&new_ports);
1946 /* Set the fail-mode */
1947 fail_mode = !br->cfg->fail_mode
1948 || !strcmp(br->cfg->fail_mode, "standalone")
1949 ? OFPROTO_FAIL_STANDALONE
1950 : OFPROTO_FAIL_SECURE;
1951 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1952 && !ofproto_has_primary_controller(br->ofproto)) {
1953 ofproto_flush_flows(br->ofproto);
1955 ofproto_set_fail_mode(br->ofproto, fail_mode);
1957 /* Delete all flows if we're switching from connected to standalone or vice
1958 * versa. (XXX Should we delete all flows if we are switching from one
1959 * controller to another?) */
1961 /* Configure OpenFlow controller connection snooping. */
1962 if (!ofproto_has_snoops(br->ofproto)) {
1966 sset_add_and_free(&snoops, xasprintf("punix:%s/%s.snoop",
1967 ovs_rundir(), br->name));
1968 ofproto_set_snoops(br->ofproto, &snoops);
1969 sset_destroy(&snoops);
1972 mirror_reconfigure(br);
1975 /* Initializes 'oc' appropriately as a management service controller for
1978 * The caller must free oc->target when it is no longer needed. */
1980 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1981 struct ofproto_controller *oc)
1983 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1984 oc->max_backoff = 0;
1985 oc->probe_interval = 60;
1986 oc->band = OFPROTO_OUT_OF_BAND;
1988 oc->burst_limit = 0;
1991 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1993 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1994 struct ofproto_controller *oc)
1996 oc->target = c->target;
1997 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1998 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1999 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
2000 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
2001 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
2002 oc->burst_limit = (c->controller_burst_limit
2003 ? *c->controller_burst_limit : 0);
2006 /* Configures the IP stack for 'br''s local interface properly according to the
2007 * configuration in 'c'. */
2009 bridge_configure_local_iface_netdev(struct bridge *br,
2010 struct ovsrec_controller *c)
2012 struct netdev *netdev;
2013 struct in_addr mask, gateway;
2015 struct iface *local_iface;
2018 /* If there's no local interface or no IP address, give up. */
2019 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
2020 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
2024 /* Bring up the local interface. */
2025 netdev = local_iface->netdev;
2026 netdev_turn_flags_on(netdev, NETDEV_UP, true);
2028 /* Configure the IP address and netmask. */
2029 if (!c->local_netmask
2030 || !inet_aton(c->local_netmask, &mask)
2032 mask.s_addr = guess_netmask(ip.s_addr);
2034 if (!netdev_set_in4(netdev, ip, mask)) {
2035 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
2036 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
2039 /* Configure the default gateway. */
2040 if (c->local_gateway
2041 && inet_aton(c->local_gateway, &gateway)
2042 && gateway.s_addr) {
2043 if (!netdev_add_router(netdev, gateway)) {
2044 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
2045 br->name, IP_ARGS(&gateway.s_addr));
2051 bridge_reconfigure_remotes(struct bridge *br,
2052 const struct sockaddr_in *managers,
2055 const char *disable_ib_str, *queue_id_str;
2056 bool disable_in_band = false;
2059 struct ovsrec_controller **controllers;
2060 size_t n_controllers;
2063 struct ofproto_controller *ocs;
2067 /* Check if we should disable in-band control on this bridge. */
2068 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
2069 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
2070 disable_in_band = true;
2073 /* Set OpenFlow queue ID for in-band control. */
2074 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
2075 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
2076 ofproto_set_in_band_queue(br->ofproto, queue_id);
2078 if (disable_in_band) {
2079 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
2081 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
2083 had_primary = ofproto_has_primary_controller(br->ofproto);
2085 n_controllers = bridge_get_controllers(br, &controllers);
2087 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
2090 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
2091 for (i = 0; i < n_controllers; i++) {
2092 struct ovsrec_controller *c = controllers[i];
2094 if (!strncmp(c->target, "punix:", 6)
2095 || !strncmp(c->target, "unix:", 5)) {
2096 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2098 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
2099 * domain sockets and overwriting arbitrary local files. */
2100 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2101 "\"%s\" due to possibility for remote exploit",
2102 dpif_name(br->dpif), c->target);
2106 bridge_configure_local_iface_netdev(br, c);
2107 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2108 if (disable_in_band) {
2109 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2114 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2115 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2118 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2119 ofproto_flush_flows(br->ofproto);
2122 /* If there are no controllers and the bridge is in standalone
2123 * mode, set up a flow that matches every packet and directs
2124 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2125 * switch is in secure mode and we won't pass any traffic until
2126 * a controller has been defined and it tells us to do so. */
2128 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2129 union ofp_action action;
2130 struct cls_rule rule;
2132 memset(&action, 0, sizeof action);
2133 action.type = htons(OFPAT_OUTPUT);
2134 action.output.len = htons(sizeof action);
2135 action.output.port = htons(OFPP_NORMAL);
2136 cls_rule_init_catchall(&rule, 0);
2137 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2142 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2147 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2148 struct iface *iface;
2150 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2151 shash_add_once(ifaces, iface->name, iface);
2153 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2154 shash_add_once(ifaces, port->name, NULL);
2159 /* For robustness, in case the administrator moves around datapath ports behind
2160 * our back, we re-check all the datapath port numbers here.
2162 * This function will set the 'dp_ifidx' members of interfaces that have
2163 * disappeared to -1, so only call this function from a context where those
2164 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2165 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2166 * datapath, which doesn't support UINT16_MAX+1 ports. */
2168 bridge_fetch_dp_ifaces(struct bridge *br)
2170 struct dpif_port_dump dump;
2171 struct dpif_port dpif_port;
2174 /* Reset all interface numbers. */
2175 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2176 struct iface *iface;
2178 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2179 iface->dp_ifidx = -1;
2182 hmap_clear(&br->ifaces);
2184 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2185 struct iface *iface = iface_lookup(br, dpif_port.name);
2187 if (iface->dp_ifidx >= 0) {
2188 VLOG_WARN("%s reported interface %s twice",
2189 dpif_name(br->dpif), dpif_port.name);
2190 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2191 VLOG_WARN("%s reported interface %"PRIu16" twice",
2192 dpif_name(br->dpif), dpif_port.port_no);
2194 iface->dp_ifidx = dpif_port.port_no;
2195 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2196 hash_int(iface->dp_ifidx, 0));
2199 iface_set_ofport(iface->cfg,
2200 (iface->dp_ifidx >= 0
2201 ? odp_port_to_ofp_port(iface->dp_ifidx)
2207 /* Bridge packet processing functions. */
2210 bond_is_tcp_hash(const struct port *port)
2212 return port->bond_mode == BM_TCP && lacp_negotiated(port->lacp);
2216 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2218 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2221 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2223 struct flow hash_flow;
2225 memcpy(&hash_flow, flow, sizeof hash_flow);
2226 hash_flow.vlan_tci = 0;
2228 /* The symmetric quality of this hash function is not required, but
2229 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2230 * purposes, so we use it out of convenience. */
2231 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2234 static struct bond_entry *
2235 lookup_bond_entry(const struct port *port, const struct flow *flow,
2238 assert(port->bond_mode != BM_AB);
2240 if (bond_is_tcp_hash(port)) {
2241 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2243 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2247 static struct iface *
2248 bond_choose_iface(const struct port *port)
2250 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2251 struct iface *best_down_slave;
2252 struct iface *iface;
2254 best_down_slave = NULL;
2255 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2256 if (iface->enabled) {
2258 } else if ((!best_down_slave
2259 || iface->delay_expires < best_down_slave->delay_expires)
2260 && lacp_slave_may_enable(port->lacp, iface)) {
2261 best_down_slave = iface;
2265 if (best_down_slave) {
2266 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2267 "since no other interface is up",
2268 best_down_slave->name,
2269 best_down_slave->delay_expires - time_msec());
2270 bond_enable_slave(best_down_slave, true);
2273 return best_down_slave;
2277 choose_output_iface(const struct port *port, const struct flow *flow,
2278 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2280 struct iface *iface;
2282 assert(port->n_ifaces);
2283 if (port->n_ifaces == 1) {
2284 iface = port_get_an_iface(port);
2285 } else if (port->bond_mode == BM_AB) {
2286 iface = port->active_iface;
2288 *tags |= port->no_ifaces_tag;
2292 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2293 if (!e->iface || !e->iface->enabled) {
2294 /* XXX select interface properly. The current interface selection
2295 * is only good for testing the rebalancing code. */
2296 e->iface = bond_choose_iface(port);
2298 *tags |= port->no_ifaces_tag;
2301 e->tag = tag_create_random();
2306 *dp_ifidx = iface->dp_ifidx;
2307 *tags |= iface->tag; /* Currently only used for bonding. */
2312 bond_link_status_update(struct iface *iface)
2314 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2315 struct port *port = iface->port;
2316 bool up = iface->up && lacp_slave_may_enable(port->lacp, iface);
2317 int updelay, downdelay;
2319 updelay = port->updelay;
2320 downdelay = port->downdelay;
2322 if (lacp_negotiated(port->lacp)) {
2327 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2328 /* Nothing to do. */
2331 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2332 iface->name, up ? "up" : "down");
2333 if (up == iface->enabled) {
2334 iface->delay_expires = LLONG_MAX;
2335 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2336 iface->name, up ? "disabled" : "enabled");
2337 } else if (up && !port->active_iface) {
2338 bond_enable_slave(iface, true);
2340 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2341 "other interface is up", iface->name, updelay);
2344 int delay = up ? updelay : downdelay;
2345 iface->delay_expires = time_msec() + delay;
2348 "interface %s: will be %s if it stays %s for %d ms",
2350 up ? "enabled" : "disabled",
2358 bond_choose_active_iface(struct port *port)
2360 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2362 port->active_iface = bond_choose_iface(port);
2363 if (port->active_iface) {
2364 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2365 port->name, port->active_iface->name);
2367 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2373 bond_enable_slave(struct iface *iface, bool enable)
2375 struct port *port = iface->port;
2376 struct bridge *br = port->bridge;
2378 /* This acts as a recursion check. If the act of disabling a slave
2379 * causes a different slave to be enabled, the flag will allow us to
2380 * skip redundant work when we reenter this function. It must be
2381 * cleared on exit to keep things safe with multiple bonds. */
2382 static bool moving_active_iface = false;
2384 iface->delay_expires = LLONG_MAX;
2385 if (enable == iface->enabled) {
2389 iface->enabled = enable;
2390 if (!iface->enabled) {
2391 VLOG_WARN("interface %s: disabled", iface->name);
2392 ofproto_revalidate(br->ofproto, iface->tag);
2393 if (iface == port->active_iface) {
2394 /* Disabling a slave can lead to another slave being immediately
2395 * enabled if there will be no active slaves but one is waiting
2396 * on an updelay. In this case we do not need to run most of the
2397 * code for the newly enabled slave since there was no period
2398 * without an active slave and it is redundant with the disabling
2400 moving_active_iface = true;
2401 bond_choose_active_iface(port);
2403 bond_send_learning_packets(port);
2405 VLOG_WARN("interface %s: enabled", iface->name);
2406 if (!port->active_iface && !moving_active_iface) {
2407 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2408 bond_choose_active_iface(port);
2409 bond_send_learning_packets(port);
2411 iface->tag = tag_create_random();
2414 moving_active_iface = false;
2417 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2418 * bond interface. */
2420 bond_update_fake_iface_stats(struct port *port)
2422 struct netdev_stats bond_stats;
2423 struct netdev *bond_dev;
2424 struct iface *iface;
2426 memset(&bond_stats, 0, sizeof bond_stats);
2428 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2429 struct netdev_stats slave_stats;
2431 if (!netdev_get_stats(iface->netdev, &slave_stats)) {
2432 /* XXX: We swap the stats here because they are swapped back when
2433 * reported by the internal device. The reason for this is
2434 * internal devices normally represent packets going into the system
2435 * but when used as fake bond device they represent packets leaving
2436 * the system. We really should do this in the internal device
2437 * itself because changing it here reverses the counts from the
2438 * perspective of the switch. However, the internal device doesn't
2439 * know what type of device it represents so we have to do it here
2441 bond_stats.tx_packets += slave_stats.rx_packets;
2442 bond_stats.tx_bytes += slave_stats.rx_bytes;
2443 bond_stats.rx_packets += slave_stats.tx_packets;
2444 bond_stats.rx_bytes += slave_stats.tx_bytes;
2448 if (!netdev_open_default(port->name, &bond_dev)) {
2449 netdev_set_stats(bond_dev, &bond_stats);
2450 netdev_close(bond_dev);
2455 bond_run(struct port *port)
2457 struct iface *iface;
2459 if (port->n_ifaces < 2) {
2463 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2464 bond_link_status_update(iface);
2467 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2468 if (time_msec() >= iface->delay_expires) {
2469 bond_enable_slave(iface, !iface->enabled);
2473 if (port->bond_fake_iface
2474 && time_msec() >= port->bond_next_fake_iface_update) {
2475 bond_update_fake_iface_stats(port);
2476 port->bond_next_fake_iface_update = time_msec() + 1000;
2481 bond_wait(struct port *port)
2483 struct iface *iface;
2485 if (port->n_ifaces < 2) {
2489 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2490 if (iface->delay_expires != LLONG_MAX) {
2491 poll_timer_wait_until(iface->delay_expires);
2495 if (port->bond_fake_iface) {
2496 poll_timer_wait_until(port->bond_next_fake_iface_update);
2501 set_dst(struct dst *dst, const struct flow *flow,
2502 const struct port *in_port, const struct port *out_port,
2505 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2506 : in_port->vlan >= 0 ? in_port->vlan
2507 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2508 : vlan_tci_to_vid(flow->vlan_tci));
2509 return choose_output_iface(out_port, flow, dst->vlan,
2510 &dst->dp_ifidx, tags);
2514 swap_dst(struct dst *p, struct dst *q)
2516 struct dst tmp = *p;
2521 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2522 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2523 * that we push to the datapath. We could in fact fully sort the array by
2524 * vlan, but in most cases there are at most two different vlan tags so that's
2525 * possibly overkill.) */
2527 partition_dsts(struct dst_set *set, int vlan)
2529 struct dst *first = set->dsts;
2530 struct dst *last = set->dsts + set->n;
2532 while (first != last) {
2534 * - All dsts < first have vlan == 'vlan'.
2535 * - All dsts >= last have vlan != 'vlan'.
2536 * - first < last. */
2537 while (first->vlan == vlan) {
2538 if (++first == last) {
2543 /* Same invariants, plus one additional:
2544 * - first->vlan != vlan.
2546 while (last[-1].vlan != vlan) {
2547 if (--last == first) {
2552 /* Same invariants, plus one additional:
2553 * - last[-1].vlan == vlan.*/
2554 swap_dst(first++, --last);
2559 mirror_mask_ffs(mirror_mask_t mask)
2561 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2566 dst_set_init(struct dst_set *set)
2568 set->dsts = set->builtin;
2570 set->allocated = ARRAY_SIZE(set->builtin);
2574 dst_set_add(struct dst_set *set, const struct dst *dst)
2576 if (set->n >= set->allocated) {
2577 size_t new_allocated;
2578 struct dst *new_dsts;
2580 new_allocated = set->allocated * 2;
2581 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2582 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2586 set->dsts = new_dsts;
2587 set->allocated = new_allocated;
2589 set->dsts[set->n++] = *dst;
2593 dst_set_free(struct dst_set *set)
2595 if (set->dsts != set->builtin) {
2601 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2604 for (i = 0; i < set->n; i++) {
2605 if (set->dsts[i].vlan == test->vlan
2606 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2614 port_trunks_vlan(const struct port *port, uint16_t vlan)
2616 return (port->vlan < 0
2617 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2621 port_includes_vlan(const struct port *port, uint16_t vlan)
2623 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2627 port_is_floodable(const struct port *port)
2629 struct iface *iface;
2631 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2632 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2640 /* Returns the tag for 'port''s active iface, or 'port''s no_ifaces_tag if
2641 * there is no active iface. */
2643 port_get_active_iface_tag(const struct port *port)
2645 return (port->active_iface
2646 ? port->active_iface->tag
2647 : port->no_ifaces_tag);
2650 /* Returns an arbitrary interface within 'port'.
2652 * 'port' must have at least one interface. */
2653 static struct iface *
2654 port_get_an_iface(const struct port *port)
2656 return CONTAINER_OF(list_front(&port->ifaces), struct iface, port_elem);
2660 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2661 const struct port *in_port, const struct port *out_port,
2662 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2664 mirror_mask_t mirrors = in_port->src_mirrors;
2668 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2669 if (flow_vlan == 0) {
2670 flow_vlan = OFP_VLAN_NONE;
2673 if (out_port == FLOOD_PORT) {
2676 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2678 && port_is_floodable(port)
2679 && port_includes_vlan(port, vlan)
2680 && !port->is_mirror_output_port
2681 && set_dst(&dst, flow, in_port, port, tags)) {
2682 mirrors |= port->dst_mirrors;
2683 dst_set_add(set, &dst);
2686 *nf_output_iface = NF_OUT_FLOOD;
2687 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2688 dst_set_add(set, &dst);
2689 *nf_output_iface = dst.dp_ifidx;
2690 mirrors |= out_port->dst_mirrors;
2694 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2695 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2697 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2698 && !dst_is_duplicate(set, &dst)) {
2699 dst_set_add(set, &dst);
2704 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2705 if (port_includes_vlan(port, m->out_vlan)
2706 && set_dst(&dst, flow, in_port, port, tags))
2708 if (port->vlan < 0) {
2709 dst.vlan = m->out_vlan;
2711 if (dst_is_duplicate(set, &dst)) {
2715 /* Use the vlan tag on the original flow instead of
2716 * the one passed in the vlan parameter. This ensures
2717 * that we compare the vlan from before any implicit
2718 * tagging tags place. This is necessary because
2719 * dst->vlan is the final vlan, after removing implicit
2721 if (port == in_port && dst.vlan == flow_vlan) {
2722 /* Don't send out input port on same VLAN. */
2725 dst_set_add(set, &dst);
2730 mirrors &= mirrors - 1;
2733 partition_dsts(set, flow_vlan);
2736 static void OVS_UNUSED
2737 print_dsts(const struct dst_set *set)
2741 for (i = 0; i < set->n; i++) {
2742 const struct dst *dst = &set->dsts[i];
2744 printf(">p%"PRIu16, dst->dp_ifidx);
2745 if (dst->vlan != OFP_VLAN_NONE) {
2746 printf("v%"PRIu16, dst->vlan);
2752 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2753 const struct port *in_port, const struct port *out_port,
2754 tag_type *tags, struct ofpbuf *actions,
2755 uint16_t *nf_output_iface)
2762 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2765 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2766 if (cur_vlan == 0) {
2767 cur_vlan = OFP_VLAN_NONE;
2769 for (i = 0; i < set.n; i++) {
2770 const struct dst *dst = &set.dsts[i];
2771 if (dst->vlan != cur_vlan) {
2772 if (dst->vlan == OFP_VLAN_NONE) {
2773 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2776 tci = htons(dst->vlan & VLAN_VID_MASK);
2777 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2778 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2780 cur_vlan = dst->vlan;
2782 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2787 /* Returns the effective vlan of a packet, taking into account both the
2788 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2789 * the packet is untagged and -1 indicates it has an invalid header and
2790 * should be dropped. */
2791 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2792 struct port *in_port, bool have_packet)
2794 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2795 if (in_port->vlan >= 0) {
2797 /* XXX support double tagging? */
2799 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2800 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2801 "packet received on port %s configured with "
2802 "implicit VLAN %"PRIu16,
2803 br->name, vlan, in_port->name, in_port->vlan);
2807 vlan = in_port->vlan;
2809 if (!port_includes_vlan(in_port, vlan)) {
2811 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2812 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2813 "packet received on port %s not configured for "
2815 br->name, vlan, in_port->name, vlan);
2824 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2825 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2826 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2828 is_gratuitous_arp(const struct flow *flow)
2830 return (flow->dl_type == htons(ETH_TYPE_ARP)
2831 && eth_addr_is_broadcast(flow->dl_dst)
2832 && (flow->nw_proto == ARP_OP_REPLY
2833 || (flow->nw_proto == ARP_OP_REQUEST
2834 && flow->nw_src == flow->nw_dst)));
2838 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2839 struct port *in_port)
2841 struct mac_entry *mac;
2843 if (!mac_learning_may_learn(br->ml, flow->dl_src, vlan)) {
2847 mac = mac_learning_insert(br->ml, flow->dl_src, vlan);
2848 if (is_gratuitous_arp(flow)) {
2849 /* We don't want to learn from gratuitous ARP packets that are
2850 * reflected back over bond slaves so we lock the learning table. */
2851 if (in_port->n_ifaces == 1) {
2852 mac_entry_set_grat_arp_lock(mac);
2853 } else if (mac_entry_is_grat_arp_locked(mac)) {
2858 if (mac_entry_is_new(mac) || mac->port.p != in_port) {
2859 /* The log messages here could actually be useful in debugging,
2860 * so keep the rate limit relatively high. */
2861 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
2862 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2863 "on port %s in VLAN %d",
2864 br->name, ETH_ADDR_ARGS(flow->dl_src),
2865 in_port->name, vlan);
2867 mac->port.p = in_port;
2868 ofproto_revalidate(br->ofproto, mac_learning_changed(br->ml, mac));
2872 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2873 * dropped. Returns true if they may be forwarded, false if they should be
2876 * If 'have_packet' is true, it indicates that the caller is processing a
2877 * received packet. If 'have_packet' is false, then the caller is just
2878 * revalidating an existing flow because configuration has changed. Either
2879 * way, 'have_packet' only affects logging (there is no point in logging errors
2880 * during revalidation).
2882 * Sets '*in_portp' to the input port. This will be a null pointer if
2883 * flow->in_port does not designate a known input port (in which case
2884 * is_admissible() returns false).
2886 * When returning true, sets '*vlanp' to the effective VLAN of the input
2887 * packet, as returned by flow_get_vlan().
2889 * May also add tags to '*tags', although the current implementation only does
2890 * so in one special case.
2893 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2894 tag_type *tags, int *vlanp, struct port **in_portp)
2896 struct iface *in_iface;
2897 struct port *in_port;
2900 /* Find the interface and port structure for the received packet. */
2901 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2903 /* No interface? Something fishy... */
2905 /* Odd. A few possible reasons here:
2907 * - We deleted an interface but there are still a few packets
2908 * queued up from it.
2910 * - Someone externally added an interface (e.g. with "ovs-dpctl
2911 * add-if") that we don't know about.
2913 * - Packet arrived on the local port but the local port is not
2914 * one of our bridge ports.
2916 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2918 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2919 "interface %"PRIu16, br->name, flow->in_port);
2925 *in_portp = in_port = in_iface->port;
2926 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2931 /* Drop frames for reserved multicast addresses. */
2932 if (eth_addr_is_reserved(flow->dl_dst)) {
2936 /* Drop frames on ports reserved for mirroring. */
2937 if (in_port->is_mirror_output_port) {
2939 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2940 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2941 "%s, which is reserved exclusively for mirroring",
2942 br->name, in_port->name);
2947 /* When using LACP, do not accept packets from disabled interfaces. */
2948 if (lacp_negotiated(in_port->lacp) && !in_iface->enabled) {
2952 /* Packets received on non-LACP bonds need special attention to avoid
2954 if (in_port->n_ifaces > 1 && !lacp_negotiated(in_port->lacp)) {
2955 struct mac_entry *mac;
2957 if (eth_addr_is_multicast(flow->dl_dst)) {
2958 *tags |= port_get_active_iface_tag(in_port);
2959 if (in_port->active_iface != in_iface) {
2960 /* Drop all multicast packets on inactive slaves. */
2965 /* Drop all packets for which we have learned a different input
2966 * port, because we probably sent the packet on one slave and got
2967 * it back on the other. Gratuitous ARP packets are an exception
2968 * to this rule: the host has moved to another switch. The exception
2969 * to the exception is if we locked the learning table to avoid
2970 * reflections on bond slaves. If this is the case, just drop the
2972 mac = mac_learning_lookup(br->ml, flow->dl_src, vlan, NULL);
2973 if (mac && mac->port.p != in_port &&
2974 (!is_gratuitous_arp(flow) || mac_entry_is_grat_arp_locked(mac))) {
2982 /* If the composed actions may be applied to any packet in the given 'flow',
2983 * returns true. Otherwise, the actions should only be applied to 'packet', or
2984 * not at all, if 'packet' was NULL. */
2986 process_flow(struct bridge *br, const struct flow *flow,
2987 const struct ofpbuf *packet, struct ofpbuf *actions,
2988 tag_type *tags, uint16_t *nf_output_iface)
2990 struct port *in_port;
2991 struct port *out_port;
2992 struct mac_entry *mac;
2995 /* Check whether we should drop packets in this flow. */
2996 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
3001 /* Learn source MAC (but don't try to learn from revalidation). */
3003 update_learning_table(br, flow, vlan, in_port);
3006 /* Determine output port. */
3007 mac = mac_learning_lookup(br->ml, flow->dl_dst, vlan, tags);
3009 out_port = mac->port.p;
3010 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
3011 /* If we are revalidating but don't have a learning entry then
3012 * eject the flow. Installing a flow that floods packets opens
3013 * up a window of time where we could learn from a packet reflected
3014 * on a bond and blackhole packets before the learning table is
3015 * updated to reflect the correct port. */
3018 out_port = FLOOD_PORT;
3021 /* Don't send packets out their input ports. */
3022 if (in_port == out_port) {
3028 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
3036 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
3037 struct ofpbuf *actions, tag_type *tags,
3038 uint16_t *nf_output_iface, void *br_)
3040 struct bridge *br = br_;
3042 COVERAGE_INC(bridge_process_flow);
3043 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
3047 bridge_special_ofhook_cb(const struct flow *flow,
3048 const struct ofpbuf *packet, void *br_)
3050 struct iface *iface;
3051 struct bridge *br = br_;
3053 iface = iface_from_dp_ifidx(br, flow->in_port);
3055 if (flow->dl_type == htons(ETH_TYPE_LACP)) {
3057 if (iface && iface->port->lacp && packet) {
3058 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
3061 COVERAGE_INC(bridge_process_lacp);
3062 lacp_process_pdu(iface->port->lacp, iface, pdu);
3072 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
3073 const struct nlattr *actions,
3075 uint64_t n_bytes, void *br_)
3077 struct bridge *br = br_;
3078 const struct nlattr *a;
3079 struct port *in_port;
3084 /* Feed information from the active flows back into the learning table to
3085 * ensure that table is always in sync with what is actually flowing
3086 * through the datapath.
3088 * We test that 'tags' is nonzero to ensure that only flows that include an
3089 * OFPP_NORMAL action are used for learning. This works because
3090 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3091 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3092 update_learning_table(br, flow, vlan, in_port);
3095 /* Account for bond slave utilization. */
3096 if (!br->has_bonded_ports) {
3099 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3100 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3101 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3102 if (out_port && out_port->n_ifaces >= 2 &&
3103 out_port->bond_mode != BM_AB) {
3104 uint16_t vlan = (flow->vlan_tci
3105 ? vlan_tci_to_vid(flow->vlan_tci)
3107 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
3108 e->tx_bytes += n_bytes;
3115 bridge_account_checkpoint_ofhook_cb(void *br_)
3117 struct bridge *br = br_;
3121 if (!br->has_bonded_ports) {
3126 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3127 if (port->n_ifaces > 1 && port->bond_mode != BM_AB
3128 && now >= port->bond_next_rebalance) {
3129 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3130 bond_rebalance_port(port);
3135 static struct ofhooks bridge_ofhooks = {
3136 bridge_normal_ofhook_cb,
3137 bridge_special_ofhook_cb,
3138 bridge_account_flow_ofhook_cb,
3139 bridge_account_checkpoint_ofhook_cb,
3142 /* Bonding functions. */
3144 /* Statistics for a single interface on a bonded port, used for load-based
3145 * bond rebalancing. */
3146 struct slave_balance {
3147 struct iface *iface; /* The interface. */
3148 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3150 /* All the "bond_entry"s that are assigned to this interface, in order of
3151 * increasing tx_bytes. */
3152 struct bond_entry **hashes;
3157 bond_mode_to_string(enum bond_mode bm) {
3158 static char *bm_slb = "balance-slb";
3159 static char *bm_ab = "active-backup";
3160 static char *bm_tcp = "balance-tcp";
3163 case BM_SLB: return bm_slb;
3164 case BM_AB: return bm_ab;
3165 case BM_TCP: return bm_tcp;
3172 /* Sorts pointers to pointers to bond_entries in ascending order by the
3173 * interface to which they are assigned, and within a single interface in
3174 * ascending order of bytes transmitted. */
3176 compare_bond_entries(const void *a_, const void *b_)
3178 const struct bond_entry *const *ap = a_;
3179 const struct bond_entry *const *bp = b_;
3180 const struct bond_entry *a = *ap;
3181 const struct bond_entry *b = *bp;
3182 if (a->iface != b->iface) {
3183 return a->iface > b->iface ? 1 : -1;
3184 } else if (a->tx_bytes != b->tx_bytes) {
3185 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3191 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3192 * *descending* order by number of bytes transmitted. */
3194 compare_slave_balance(const void *a_, const void *b_)
3196 const struct slave_balance *a = a_;
3197 const struct slave_balance *b = b_;
3198 if (a->iface->enabled != b->iface->enabled) {
3199 return a->iface->enabled ? -1 : 1;
3200 } else if (a->tx_bytes != b->tx_bytes) {
3201 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3208 swap_bals(struct slave_balance *a, struct slave_balance *b)
3210 struct slave_balance tmp = *a;
3215 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3216 * given that 'p' (and only 'p') might be in the wrong location.
3218 * This function invalidates 'p', since it might now be in a different memory
3221 resort_bals(struct slave_balance *p,
3222 struct slave_balance bals[], size_t n_bals)
3225 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3226 swap_bals(p, p - 1);
3228 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3229 swap_bals(p, p + 1);
3235 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3237 if (VLOG_IS_DBG_ENABLED()) {
3238 struct ds ds = DS_EMPTY_INITIALIZER;
3239 const struct slave_balance *b;
3241 for (b = bals; b < bals + n_bals; b++) {
3245 ds_put_char(&ds, ',');
3247 ds_put_format(&ds, " %s %"PRIu64"kB",
3248 b->iface->name, b->tx_bytes / 1024);
3250 if (!b->iface->enabled) {
3251 ds_put_cstr(&ds, " (disabled)");
3253 if (b->n_hashes > 0) {
3254 ds_put_cstr(&ds, " (");
3255 for (i = 0; i < b->n_hashes; i++) {
3256 const struct bond_entry *e = b->hashes[i];
3258 ds_put_cstr(&ds, " + ");
3260 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3261 e - port->bond_hash, e->tx_bytes / 1024);
3263 ds_put_cstr(&ds, ")");
3266 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3271 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3273 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3276 struct bond_entry *hash = from->hashes[hash_idx];
3277 struct port *port = from->iface->port;
3278 uint64_t delta = hash->tx_bytes;
3280 assert(port->bond_mode != BM_AB);
3282 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3283 "from %s to %s (now carrying %"PRIu64"kB and "
3284 "%"PRIu64"kB load, respectively)",
3285 port->name, delta / 1024, hash - port->bond_hash,
3286 from->iface->name, to->iface->name,
3287 (from->tx_bytes - delta) / 1024,
3288 (to->tx_bytes + delta) / 1024);
3290 /* Delete element from from->hashes.
3292 * We don't bother to add the element to to->hashes because not only would
3293 * it require more work, the only purpose it would be to allow that hash to
3294 * be migrated to another slave in this rebalancing run, and there is no
3295 * point in doing that. */
3296 if (hash_idx == 0) {
3299 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3300 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3304 /* Shift load away from 'from' to 'to'. */
3305 from->tx_bytes -= delta;
3306 to->tx_bytes += delta;
3308 /* Arrange for flows to be revalidated. */
3309 ofproto_revalidate(port->bridge->ofproto, hash->tag);
3310 hash->iface = to->iface;
3311 hash->tag = tag_create_random();
3315 bond_rebalance_port(struct port *port)
3317 struct slave_balance *bals;
3319 struct bond_entry *hashes[BOND_MASK + 1];
3320 struct slave_balance *b, *from, *to;
3321 struct bond_entry *e;
3322 struct iface *iface;
3325 assert(port->bond_mode != BM_AB);
3327 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3328 * descending order of tx_bytes, so that bals[0] represents the most
3329 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3332 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3333 * array for each slave_balance structure, we sort our local array of
3334 * hashes in order by slave, so that all of the hashes for a given slave
3335 * become contiguous in memory, and then we point each 'hashes' members of
3336 * a slave_balance structure to the start of a contiguous group. */
3337 n_bals = port->n_ifaces;
3338 b = bals = xmalloc(n_bals * sizeof *bals);
3339 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3346 assert(b == &bals[n_bals]);
3347 for (i = 0; i <= BOND_MASK; i++) {
3348 hashes[i] = &port->bond_hash[i];
3350 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3351 for (i = 0; i <= BOND_MASK; i++) {
3357 for (b = bals; b < &bals[n_bals]; b++) {
3358 if (b->iface == e->iface) {
3359 b->tx_bytes += e->tx_bytes;
3361 b->hashes = &hashes[i];
3368 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3369 log_bals(bals, n_bals, port);
3371 /* Discard slaves that aren't enabled (which were sorted to the back of the
3372 * array earlier). */
3373 while (!bals[n_bals - 1].iface->enabled) {
3380 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3381 to = &bals[n_bals - 1];
3382 for (from = bals; from < to; ) {
3383 uint64_t overload = from->tx_bytes - to->tx_bytes;
3384 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3385 /* The extra load on 'from' (and all less-loaded slaves), compared
3386 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3387 * it is less than ~1Mbps. No point in rebalancing. */
3389 } else if (from->n_hashes == 1) {
3390 /* 'from' only carries a single MAC hash, so we can't shift any
3391 * load away from it, even though we want to. */
3394 /* 'from' is carrying significantly more load than 'to', and that
3395 * load is split across at least two different hashes. Pick a hash
3396 * to migrate to 'to' (the least-loaded slave), given that doing so
3397 * must decrease the ratio of the load on the two slaves by at
3400 * The sort order we use means that we prefer to shift away the
3401 * smallest hashes instead of the biggest ones. There is little
3402 * reason behind this decision; we could use the opposite sort
3403 * order to shift away big hashes ahead of small ones. */
3406 for (i = 0; i < from->n_hashes; i++) {
3407 double old_ratio, new_ratio;
3408 uint64_t delta = from->hashes[i]->tx_bytes;
3410 if (delta == 0 || from->tx_bytes - delta == 0) {
3411 /* Pointless move. */
3415 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3417 if (to->tx_bytes == 0) {
3418 /* Nothing on the new slave, move it. */
3422 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3423 new_ratio = (double)(from->tx_bytes - delta) /
3424 (to->tx_bytes + delta);
3426 if (new_ratio == 0) {
3427 /* Should already be covered but check to prevent division
3432 if (new_ratio < 1) {
3433 new_ratio = 1 / new_ratio;
3436 if (old_ratio - new_ratio > 0.1) {
3437 /* Would decrease the ratio, move it. */
3441 if (i < from->n_hashes) {
3442 bond_shift_load(from, to, i);
3444 /* If the result of the migration changed the relative order of
3445 * 'from' and 'to' swap them back to maintain invariants. */
3446 if (order_swapped) {
3447 swap_bals(from, to);
3450 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3451 * point to different slave_balance structures. It is only
3452 * valid to do these two operations in a row at all because we
3453 * know that 'from' will not move past 'to' and vice versa. */
3454 resort_bals(from, bals, n_bals);
3455 resort_bals(to, bals, n_bals);
3462 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3463 * historical data to decay to <1% in 7 rebalancing runs. */
3464 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3476 bond_send_learning_packets(struct port *port)
3478 struct bridge *br = port->bridge;
3479 struct mac_entry *e;
3480 struct ofpbuf packet;
3481 int error, n_packets, n_errors;
3483 if (!port->n_ifaces || !port->active_iface || bond_is_tcp_hash(port)) {
3487 ofpbuf_init(&packet, 128);
3488 error = n_packets = n_errors = 0;
3489 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3495 if (e->port.p == port) {
3499 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3501 flow_extract(&packet, 0, ODPP_NONE, &flow);
3503 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3509 retval = ofproto_send_packet(br->ofproto, dp_ifidx, e->vlan, &packet);
3515 ofpbuf_uninit(&packet);
3518 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3519 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3520 "packets, last error was: %s",
3521 port->name, n_errors, n_packets, strerror(error));
3523 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3524 port->name, n_packets);
3528 /* Bonding unixctl user interface functions. */
3531 bond_unixctl_list(struct unixctl_conn *conn,
3532 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3534 struct ds ds = DS_EMPTY_INITIALIZER;
3535 const struct bridge *br;
3537 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3539 LIST_FOR_EACH (br, node, &all_bridges) {
3542 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3543 if (port->n_ifaces > 1) {
3544 struct iface *iface;
3546 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3547 bond_mode_to_string(port->bond_mode));
3548 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3549 if (&iface->port_elem != list_front(&port->ifaces)) {
3550 ds_put_cstr(&ds, ", ");
3552 ds_put_cstr(&ds, iface->name);
3554 ds_put_char(&ds, '\n');
3558 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3562 static struct port *
3563 bond_find(const char *name)
3565 const struct bridge *br;
3567 LIST_FOR_EACH (br, node, &all_bridges) {
3570 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3571 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3580 bond_unixctl_show(struct unixctl_conn *conn,
3581 const char *args, void *aux OVS_UNUSED)
3583 struct ds ds = DS_EMPTY_INITIALIZER;
3584 const struct port *port;
3585 struct iface *iface;
3587 port = bond_find(args);
3589 unixctl_command_reply(conn, 501, "no such bond");
3593 ds_put_format(&ds, "bond_mode: %s\n",
3594 bond_mode_to_string(port->bond_mode));
3597 ds_put_format(&ds, "lacp: %s\n",
3598 port->lacp_active ? "active" : "passive");
3600 ds_put_cstr(&ds, "lacp: off\n");
3603 if (port->bond_mode != BM_AB) {
3604 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3605 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3609 ds_put_format(&ds, "bond-detect-mode: %s\n",
3610 port->monitor ? "carrier" : "miimon");
3612 if (!port->monitor) {
3613 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3614 port->miimon_interval);
3617 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3618 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3620 if (port->bond_mode != BM_AB) {
3621 ds_put_format(&ds, "next rebalance: %lld ms\n",
3622 port->bond_next_rebalance - time_msec());
3625 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3626 struct bond_entry *be;
3630 ds_put_format(&ds, "\nslave %s: %s\n",
3631 iface->name, iface->enabled ? "enabled" : "disabled");
3632 if (iface == port->active_iface) {
3633 ds_put_cstr(&ds, "\tactive slave\n");
3635 if (iface->delay_expires != LLONG_MAX) {
3636 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3637 iface->enabled ? "downdelay" : "updelay",
3638 iface->delay_expires - time_msec());
3641 if (port->bond_mode == BM_AB) {
3646 memset(&flow, 0, sizeof flow);
3647 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3648 int hash = be - port->bond_hash;
3649 struct mac_entry *me;
3651 if (be->iface != iface) {
3655 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3656 hash, be->tx_bytes / 1024);
3658 if (port->bond_mode != BM_SLB) {
3663 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3667 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
3668 if (bond_hash_src(me->mac, me->vlan) == hash
3669 && me->port.p != port
3670 && choose_output_iface(port, &flow, me->vlan,
3672 && dp_ifidx == iface->dp_ifidx)
3674 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3675 ETH_ADDR_ARGS(me->mac));
3680 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3685 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3686 void *aux OVS_UNUSED)
3688 char *args = (char *) args_;
3689 char *save_ptr = NULL;
3690 char *bond_s, *hash_s, *slave_s;
3692 struct iface *iface;
3693 struct bond_entry *entry;
3696 bond_s = strtok_r(args, " ", &save_ptr);
3697 hash_s = strtok_r(NULL, " ", &save_ptr);
3698 slave_s = strtok_r(NULL, " ", &save_ptr);
3700 unixctl_command_reply(conn, 501,
3701 "usage: bond/migrate BOND HASH SLAVE");
3705 port = bond_find(bond_s);
3707 unixctl_command_reply(conn, 501, "no such bond");
3711 if (port->bond_mode != BM_SLB) {
3712 unixctl_command_reply(conn, 501, "not an SLB bond");
3716 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3717 hash = atoi(hash_s) & BOND_MASK;
3719 unixctl_command_reply(conn, 501, "bad hash");
3723 iface = port_lookup_iface(port, slave_s);
3725 unixctl_command_reply(conn, 501, "no such slave");
3729 if (!iface->enabled) {
3730 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3734 entry = &port->bond_hash[hash];
3735 ofproto_revalidate(port->bridge->ofproto, entry->tag);
3736 entry->iface = iface;
3737 entry->tag = tag_create_random();
3738 unixctl_command_reply(conn, 200, "migrated");
3742 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3743 void *aux OVS_UNUSED)
3745 char *args = (char *) args_;
3746 char *save_ptr = NULL;
3747 char *bond_s, *slave_s;
3749 struct iface *iface;
3751 bond_s = strtok_r(args, " ", &save_ptr);
3752 slave_s = strtok_r(NULL, " ", &save_ptr);
3754 unixctl_command_reply(conn, 501,
3755 "usage: bond/set-active-slave BOND SLAVE");
3759 port = bond_find(bond_s);
3761 unixctl_command_reply(conn, 501, "no such bond");
3765 iface = port_lookup_iface(port, slave_s);
3767 unixctl_command_reply(conn, 501, "no such slave");
3771 if (!iface->enabled) {
3772 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3776 if (port->active_iface != iface) {
3777 ofproto_revalidate(port->bridge->ofproto,
3778 port_get_active_iface_tag(port));
3779 port->active_iface = iface;
3780 VLOG_INFO("port %s: active interface is now %s",
3781 port->name, iface->name);
3782 bond_send_learning_packets(port);
3783 unixctl_command_reply(conn, 200, "done");
3785 unixctl_command_reply(conn, 200, "no change");
3790 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3792 char *args = (char *) args_;
3793 char *save_ptr = NULL;
3794 char *bond_s, *slave_s;
3796 struct iface *iface;
3798 bond_s = strtok_r(args, " ", &save_ptr);
3799 slave_s = strtok_r(NULL, " ", &save_ptr);
3801 unixctl_command_reply(conn, 501,
3802 "usage: bond/enable/disable-slave BOND SLAVE");
3806 port = bond_find(bond_s);
3808 unixctl_command_reply(conn, 501, "no such bond");
3812 iface = port_lookup_iface(port, slave_s);
3814 unixctl_command_reply(conn, 501, "no such slave");
3818 bond_enable_slave(iface, enable);
3819 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3823 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3824 void *aux OVS_UNUSED)
3826 enable_slave(conn, args, true);
3830 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3831 void *aux OVS_UNUSED)
3833 enable_slave(conn, args, false);
3837 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3838 void *aux OVS_UNUSED)
3840 char *args = (char *) args_;
3841 uint8_t mac[ETH_ADDR_LEN];
3845 char *mac_s, *vlan_s;
3846 char *save_ptr = NULL;
3848 mac_s = strtok_r(args, " ", &save_ptr);
3849 vlan_s = strtok_r(NULL, " ", &save_ptr);
3852 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3853 unixctl_command_reply(conn, 501, "invalid vlan");
3857 vlan = OFP_VLAN_NONE;
3860 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3861 == ETH_ADDR_SCAN_COUNT) {
3862 hash = bond_hash_src(mac, vlan);
3864 hash_cstr = xasprintf("%u", hash);
3865 unixctl_command_reply(conn, 200, hash_cstr);
3868 unixctl_command_reply(conn, 501, "invalid mac");
3875 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3876 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3877 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3878 unixctl_command_register("bond/set-active-slave",
3879 bond_unixctl_set_active_slave, NULL);
3880 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3882 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3884 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3887 /* Port functions. */
3890 lacp_send_pdu_cb(void *aux, const struct lacp_pdu *pdu)
3892 struct iface *iface = aux;
3893 uint8_t ea[ETH_ADDR_LEN];
3896 error = netdev_get_etheraddr(iface->netdev, ea);
3898 struct ofpbuf packet;
3899 struct lacp_pdu *packet_pdu;
3901 ofpbuf_init(&packet, 0);
3902 packet_pdu = compose_packet(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
3903 sizeof *packet_pdu);
3904 memcpy(packet_pdu, pdu, sizeof *packet_pdu);
3905 ofproto_send_packet(iface->port->bridge->ofproto,
3906 iface->dp_ifidx, 0, &packet);
3907 ofpbuf_uninit(&packet);
3909 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3910 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3911 "(%s)", iface->name, strerror(error));
3916 port_run(struct port *port)
3918 if (port->monitor) {
3921 /* Track carrier going up and down on interfaces. */
3922 while (!netdev_monitor_poll(port->monitor, &devname)) {
3923 struct iface *iface;
3925 iface = port_lookup_iface(port, devname);
3927 iface_update_carrier(iface);
3931 } else if (time_msec() >= port->miimon_next_update) {
3932 struct iface *iface;
3934 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3935 iface_update_carrier(iface);
3937 port->miimon_next_update = time_msec() + port->miimon_interval;
3941 struct iface *iface;
3943 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3944 lacp_slave_enable(port->lacp, iface, iface->enabled);
3947 lacp_run(port->lacp, lacp_send_pdu_cb);
3954 port_wait(struct port *port)
3956 if (port->monitor) {
3957 netdev_monitor_poll_wait(port->monitor);
3959 poll_timer_wait_until(port->miimon_next_update);
3963 lacp_wait(port->lacp);
3969 static struct port *
3970 port_create(struct bridge *br, const char *name)
3974 port = xzalloc(sizeof *port);
3977 port->trunks = NULL;
3978 port->name = xstrdup(name);
3979 port->active_iface = NULL;
3980 list_init(&port->ifaces);
3982 hmap_insert(&br->ports, &port->hmap_node, hash_string(port->name, 0));
3984 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3991 get_port_other_config(const struct ovsrec_port *port, const char *key,
3992 const char *default_value)
3996 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3998 return value ? value : default_value;
4002 get_interface_other_config(const struct ovsrec_interface *iface,
4003 const char *key, const char *default_value)
4007 value = get_ovsrec_key_value(&iface->header_,
4008 &ovsrec_interface_col_other_config, key);
4009 return value ? value : default_value;
4013 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
4015 struct iface *iface, *next;
4016 struct sset new_ifaces;
4019 /* Collect list of new interfaces. */
4020 sset_init(&new_ifaces);
4021 for (i = 0; i < cfg->n_interfaces; i++) {
4022 const char *name = cfg->interfaces[i]->name;
4023 sset_add(&new_ifaces, name);
4026 /* Get rid of deleted interfaces. */
4027 LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) {
4028 if (!sset_contains(&new_ifaces, iface->name)) {
4029 iface_destroy(iface);
4033 sset_destroy(&new_ifaces);
4036 /* Expires all MAC learning entries associated with 'port' and forces ofproto
4037 * to revalidate every flow. */
4039 port_flush_macs(struct port *port)
4041 struct bridge *br = port->bridge;
4042 struct mac_learning *ml = br->ml;
4043 struct mac_entry *mac, *next_mac;
4046 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
4047 if (mac->port.p == port) {
4048 mac_learning_expire(ml, mac);
4054 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
4056 const char *detect_mode;
4057 struct sset new_ifaces;
4058 long long int next_rebalance, miimon_next_update, lacp_priority;
4059 bool need_flush = false;
4060 unsigned long *trunks;
4066 /* Update settings. */
4067 port->updelay = cfg->bond_updelay;
4068 if (port->updelay < 0) {
4071 port->downdelay = cfg->bond_downdelay;
4072 if (port->downdelay < 0) {
4073 port->downdelay = 0;
4075 port->bond_rebalance_interval = atoi(
4076 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
4077 if (port->bond_rebalance_interval < 1000) {
4078 port->bond_rebalance_interval = 1000;
4080 next_rebalance = time_msec() + port->bond_rebalance_interval;
4081 if (port->bond_next_rebalance > next_rebalance) {
4082 port->bond_next_rebalance = next_rebalance;
4085 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
4088 netdev_monitor_destroy(port->monitor);
4089 port->monitor = NULL;
4091 if (strcmp(detect_mode, "miimon")) {
4092 port->monitor = netdev_monitor_create();
4094 if (strcmp(detect_mode, "carrier")) {
4095 VLOG_WARN("port %s: unsupported bond-detect-mode %s, "
4096 "defaulting to carrier", port->name, detect_mode);
4100 port->miimon_interval = atoi(
4101 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4102 if (port->miimon_interval < 100) {
4103 port->miimon_interval = 100;
4105 miimon_next_update = time_msec() + port->miimon_interval;
4106 if (port->miimon_next_update > miimon_next_update) {
4107 port->miimon_next_update = miimon_next_update;
4110 if (!port->cfg->bond_mode ||
4111 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4112 port->bond_mode = BM_SLB;
4113 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4114 port->bond_mode = BM_AB;
4115 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
4116 port->bond_mode = BM_TCP;
4118 port->bond_mode = BM_SLB;
4119 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4120 port->name, port->cfg->bond_mode,
4121 bond_mode_to_string(port->bond_mode));
4124 /* Add new interfaces and update 'cfg' member of existing ones. */
4125 sset_init(&new_ifaces);
4126 for (i = 0; i < cfg->n_interfaces; i++) {
4127 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4128 struct iface *iface;
4130 if (!sset_add(&new_ifaces, if_cfg->name)) {
4131 VLOG_WARN("port %s: %s specified twice as port interface",
4132 port->name, if_cfg->name);
4133 iface_set_ofport(if_cfg, -1);
4137 iface = iface_lookup(port->bridge, if_cfg->name);
4139 if (iface->port != port) {
4140 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4142 port->bridge->name, if_cfg->name, iface->port->name);
4145 iface->cfg = if_cfg;
4147 iface = iface_create(port, if_cfg);
4150 /* Determine interface type. The local port always has type
4151 * "internal". Other ports take their type from the database and
4152 * default to "system" if none is specified. */
4153 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4154 : if_cfg->type[0] ? if_cfg->type
4158 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4161 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4162 iface->lacp_priority = UINT16_MAX;
4164 iface->lacp_priority = lacp_priority;
4167 sset_destroy(&new_ifaces);
4169 port->lacp_fast = !strcmp(get_port_other_config(cfg, "lacp-time", "slow"),
4173 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4175 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4176 /* Prefer bondable links if unspecified. */
4177 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4179 port->lacp_priority = lacp_priority;
4182 if (!port->cfg->lacp) {
4183 /* XXX when LACP implementation has been sufficiently tested, enable by
4184 * default and make active on bonded ports. */
4185 lacp_destroy(port->lacp);
4187 } else if (!strcmp(port->cfg->lacp, "off")) {
4188 lacp_destroy(port->lacp);
4190 } else if (!strcmp(port->cfg->lacp, "active")) {
4192 port->lacp = lacp_create();
4194 port->lacp_active = true;
4195 } else if (!strcmp(port->cfg->lacp, "passive")) {
4197 port->lacp = lacp_create();
4199 port->lacp_active = false;
4201 VLOG_WARN("port %s: unknown LACP mode %s",
4202 port->name, port->cfg->lacp);
4203 lacp_destroy(port->lacp);
4210 if (port->n_ifaces < 2) {
4212 if (vlan >= 0 && vlan <= 4095) {
4213 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4218 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4219 * they even work as-is. But they have not been tested. */
4220 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4224 if (port->vlan != vlan) {
4229 /* Get trunked VLANs. */
4231 if (vlan < 0 && cfg->n_trunks) {
4234 trunks = bitmap_allocate(4096);
4236 for (i = 0; i < cfg->n_trunks; i++) {
4237 int trunk = cfg->trunks[i];
4239 bitmap_set1(trunks, trunk);
4245 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4246 port->name, cfg->n_trunks);
4248 if (n_errors == cfg->n_trunks) {
4249 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4251 bitmap_free(trunks);
4254 } else if (vlan >= 0 && cfg->n_trunks) {
4255 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4259 ? port->trunks != NULL
4260 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4263 bitmap_free(port->trunks);
4264 port->trunks = trunks;
4267 port_flush_macs(port);
4272 port_destroy(struct port *port)
4275 struct bridge *br = port->bridge;
4276 struct iface *iface, *next;
4279 for (i = 0; i < MAX_MIRRORS; i++) {
4280 struct mirror *m = br->mirrors[i];
4281 if (m && m->out_port == port) {
4286 LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) {
4287 iface_destroy(iface);
4290 hmap_remove(&br->ports, &port->hmap_node);
4292 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4294 port_flush_macs(port);
4296 lacp_destroy(port->lacp);
4297 netdev_monitor_destroy(port->monitor);
4298 bitmap_free(port->trunks);
4299 free(port->bond_hash);
4305 static struct port *
4306 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4308 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4309 return iface ? iface->port : NULL;
4312 static struct port *
4313 port_lookup(const struct bridge *br, const char *name)
4317 HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_string(name, 0),
4319 if (!strcmp(port->name, name)) {
4326 static struct iface *
4327 port_lookup_iface(const struct port *port, const char *name)
4329 struct iface *iface = iface_lookup(port->bridge, name);
4330 return iface && iface->port == port ? iface : NULL;
4334 port_update_lacp(struct port *port)
4337 struct iface *iface;
4339 lacp_configure(port->lacp, port->name,
4340 port->bridge->ea, port->lacp_priority,
4341 port->lacp_active, port->lacp_fast);
4343 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
4344 lacp_slave_register(port->lacp, iface, iface->name,
4345 iface->dp_ifidx, iface->lacp_priority);
4351 port_update_bonding(struct port *port)
4353 if (port->n_ifaces < 2) {
4354 /* Not a bonded port. */
4355 free(port->bond_hash);
4356 port->bond_hash = NULL;
4357 port->bond_fake_iface = false;
4358 port->active_iface = NULL;
4359 port->no_ifaces_tag = 0;
4363 if (port->bond_mode != BM_AB && !port->bond_hash) {
4364 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4365 for (i = 0; i <= BOND_MASK; i++) {
4366 struct bond_entry *e = &port->bond_hash[i];
4370 port->bond_next_rebalance
4371 = time_msec() + port->bond_rebalance_interval;
4372 } else if (port->bond_mode == BM_AB) {
4373 free(port->bond_hash);
4374 port->bond_hash = NULL;
4377 if (!port->no_ifaces_tag) {
4378 port->no_ifaces_tag = tag_create_random();
4381 if (!port->active_iface) {
4382 bond_choose_active_iface(port);
4385 port->bond_fake_iface = port->cfg->bond_fake_iface;
4386 if (port->bond_fake_iface) {
4387 port->bond_next_fake_iface_update = time_msec();
4393 /* Interface functions. */
4395 static struct iface *
4396 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4398 struct bridge *br = port->bridge;
4399 struct iface *iface;
4400 char *name = if_cfg->name;
4402 iface = xzalloc(sizeof *iface);
4404 iface->name = xstrdup(name);
4405 iface->dp_ifidx = -1;
4406 iface->tag = tag_create_random();
4407 iface->delay_expires = LLONG_MAX;
4408 iface->netdev = NULL;
4409 iface->cfg = if_cfg;
4411 shash_add_assert(&br->iface_by_name, iface->name, iface);
4413 list_push_back(&port->ifaces, &iface->port_elem);
4416 if (port->n_ifaces > 1) {
4417 br->has_bonded_ports = true;
4420 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4428 iface_destroy(struct iface *iface)
4431 struct port *port = iface->port;
4432 struct bridge *br = port->bridge;
4433 bool del_active = port->active_iface == iface;
4435 if (port->bond_hash) {
4436 struct bond_entry *e;
4437 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
4438 if (e->iface == iface) {
4444 if (iface->port->lacp) {
4445 lacp_slave_unregister(iface->port->lacp, iface);
4448 if (port->monitor && iface->netdev) {
4449 netdev_monitor_remove(port->monitor, iface->netdev);
4452 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4454 if (iface->dp_ifidx >= 0) {
4455 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4458 list_remove(&iface->port_elem);
4461 netdev_close(iface->netdev);
4464 bond_choose_active_iface(port);
4465 bond_send_learning_packets(port);
4471 bridge_flush(port->bridge);
4475 static struct iface *
4476 iface_lookup(const struct bridge *br, const char *name)
4478 return shash_find_data(&br->iface_by_name, name);
4481 static struct iface *
4482 iface_find(const char *name)
4484 const struct bridge *br;
4486 LIST_FOR_EACH (br, node, &all_bridges) {
4487 struct iface *iface = iface_lookup(br, name);
4496 static struct iface *
4497 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4499 struct iface *iface;
4501 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4502 hash_int(dp_ifidx, 0), &br->ifaces) {
4503 if (iface->dp_ifidx == dp_ifidx) {
4510 /* Set Ethernet address of 'iface', if one is specified in the configuration
4513 iface_set_mac(struct iface *iface)
4515 uint8_t ea[ETH_ADDR_LEN];
4517 if (!strcmp(iface->type, "internal")
4518 && iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4519 if (iface->dp_ifidx == ODPP_LOCAL) {
4520 VLOG_ERR("interface %s: ignoring mac in Interface record "
4521 "(use Bridge record to set local port's mac)",
4523 } else if (eth_addr_is_multicast(ea)) {
4524 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4527 int error = netdev_set_etheraddr(iface->netdev, ea);
4529 VLOG_ERR("interface %s: setting MAC failed (%s)",
4530 iface->name, strerror(error));
4536 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4538 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4540 if (if_cfg && !ovsdb_idl_row_is_synthetic(&if_cfg->header_)) {
4541 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4545 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4547 * The value strings in '*shash' are taken directly from values[], not copied,
4548 * so the caller should not modify or free them. */
4550 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4551 struct shash *shash)
4556 for (i = 0; i < n; i++) {
4557 shash_add(shash, keys[i], values[i]);
4561 /* Creates 'keys' and 'values' arrays from 'shash'.
4563 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4564 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4565 * are populated with with strings taken directly from 'shash' and thus have
4566 * the same ownership of the key-value pairs in shash.
4569 shash_to_ovs_idl_map(struct shash *shash,
4570 char ***keys, char ***values, size_t *n)
4574 struct shash_node *sn;
4576 count = shash_count(shash);
4578 k = xmalloc(count * sizeof *k);
4579 v = xmalloc(count * sizeof *v);
4582 SHASH_FOR_EACH(sn, shash) {
4593 struct iface_delete_queues_cbdata {
4594 struct netdev *netdev;
4595 const struct ovsdb_datum *queues;
4599 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4601 union ovsdb_atom atom;
4603 atom.integer = target;
4604 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4608 iface_delete_queues(unsigned int queue_id,
4609 const struct shash *details OVS_UNUSED, void *cbdata_)
4611 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4613 if (!queue_ids_include(cbdata->queues, queue_id)) {
4614 netdev_delete_queue(cbdata->netdev, queue_id);
4619 iface_update_carrier(struct iface *iface)
4621 bool carrier = iface_get_carrier(iface);
4622 if (carrier == iface->up) {
4626 iface->up = carrier;
4627 if (iface->port->lacp) {
4628 lacp_slave_carrier_changed(iface->port->lacp, iface);
4633 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4635 if (!qos || qos->type[0] == '\0') {
4636 netdev_set_qos(iface->netdev, NULL, NULL);
4638 struct iface_delete_queues_cbdata cbdata;
4639 struct shash details;
4642 /* Configure top-level Qos for 'iface'. */
4643 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4644 qos->n_other_config, &details);
4645 netdev_set_qos(iface->netdev, qos->type, &details);
4646 shash_destroy(&details);
4648 /* Deconfigure queues that were deleted. */
4649 cbdata.netdev = iface->netdev;
4650 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4652 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4654 /* Configure queues for 'iface'. */
4655 for (i = 0; i < qos->n_queues; i++) {
4656 const struct ovsrec_queue *queue = qos->value_queues[i];
4657 unsigned int queue_id = qos->key_queues[i];
4659 shash_from_ovs_idl_map(queue->key_other_config,
4660 queue->value_other_config,
4661 queue->n_other_config, &details);
4662 netdev_set_queue(iface->netdev, queue_id, &details);
4663 shash_destroy(&details);
4669 iface_update_cfm(struct iface *iface)
4673 uint16_t *remote_mps;
4674 struct ovsrec_monitor *mon;
4675 uint8_t maid[CCM_MAID_LEN];
4677 mon = iface->cfg->monitor;
4680 ofproto_iface_clear_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
4684 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4685 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4689 cfm.mpid = mon->mpid;
4690 cfm.interval = mon->interval ? *mon->interval : 1000;
4692 memcpy(cfm.maid, maid, sizeof cfm.maid);
4694 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4695 for(i = 0; i < mon->n_remote_mps; i++) {
4696 remote_mps[i] = mon->remote_mps[i]->mpid;
4699 ofproto_iface_set_cfm(iface->port->bridge->ofproto, iface->dp_ifidx,
4700 &cfm, remote_mps, mon->n_remote_mps);
4704 /* Read carrier or miimon status directly from 'iface''s netdev, according to
4705 * how 'iface''s port is configured.
4707 * Returns true if 'iface' is up, false otherwise. */
4709 iface_get_carrier(const struct iface *iface)
4711 return (iface->port->monitor
4712 ? netdev_get_carrier(iface->netdev)
4713 : netdev_get_miimon(iface->netdev));
4716 /* Returns true if 'iface' is synthetic, that is, if we constructed it locally
4717 * instead of obtaining it from the database. */
4719 iface_is_synthetic(const struct iface *iface)
4721 return ovsdb_idl_row_is_synthetic(&iface->cfg->header_);
4724 /* Port mirroring. */
4726 static struct mirror *
4727 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4731 for (i = 0; i < MAX_MIRRORS; i++) {
4732 struct mirror *m = br->mirrors[i];
4733 if (m && uuid_equals(uuid, &m->uuid)) {
4741 mirror_reconfigure(struct bridge *br)
4743 unsigned long *rspan_vlans;
4747 /* Get rid of deleted mirrors. */
4748 for (i = 0; i < MAX_MIRRORS; i++) {
4749 struct mirror *m = br->mirrors[i];
4751 const struct ovsdb_datum *mc;
4752 union ovsdb_atom atom;
4754 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4755 atom.uuid = br->mirrors[i]->uuid;
4756 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4762 /* Add new mirrors and reconfigure existing ones. */
4763 for (i = 0; i < br->cfg->n_mirrors; i++) {
4764 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4765 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4767 mirror_reconfigure_one(m, cfg);
4769 mirror_create(br, cfg);
4773 /* Update port reserved status. */
4774 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
4775 port->is_mirror_output_port = false;
4777 for (i = 0; i < MAX_MIRRORS; i++) {
4778 struct mirror *m = br->mirrors[i];
4779 if (m && m->out_port) {
4780 m->out_port->is_mirror_output_port = true;
4784 /* Update flooded vlans (for RSPAN). */
4786 if (br->cfg->n_flood_vlans) {
4787 rspan_vlans = bitmap_allocate(4096);
4789 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4790 int64_t vlan = br->cfg->flood_vlans[i];
4791 if (vlan >= 0 && vlan < 4096) {
4792 bitmap_set1(rspan_vlans, vlan);
4793 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4796 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4801 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4803 mac_learning_flush(br->ml);
4808 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4813 for (i = 0; ; i++) {
4814 if (i >= MAX_MIRRORS) {
4815 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4816 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4819 if (!br->mirrors[i]) {
4824 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4826 mac_learning_flush(br->ml);
4828 br->mirrors[i] = m = xzalloc(sizeof *m);
4829 m->uuid = cfg->header_.uuid;
4832 m->name = xstrdup(cfg->name);
4833 sset_init(&m->src_ports);
4834 sset_init(&m->dst_ports);
4840 mirror_reconfigure_one(m, cfg);
4844 mirror_destroy(struct mirror *m)
4847 struct bridge *br = m->bridge;
4850 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
4851 port->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4852 port->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4855 sset_destroy(&m->src_ports);
4856 sset_destroy(&m->dst_ports);
4859 m->bridge->mirrors[m->idx] = NULL;
4864 mac_learning_flush(br->ml);
4869 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4874 for (i = 0; i < n_ports; i++) {
4875 const char *name = ports[i]->name;
4876 if (port_lookup(m->bridge, name)) {
4877 sset_add(names, name);
4879 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4880 "port %s", m->bridge->name, m->name, name);
4886 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4892 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4894 for (i = 0; i < cfg->n_select_vlan; i++) {
4895 int64_t vlan = cfg->select_vlan[i];
4896 if (vlan < 0 || vlan > 4095) {
4897 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4898 m->bridge->name, m->name, vlan);
4900 (*vlans)[n_vlans++] = vlan;
4907 vlan_is_mirrored(const struct mirror *m, int vlan)
4911 for (i = 0; i < m->n_vlans; i++) {
4912 if (m->vlans[i] == vlan) {
4920 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4922 struct sset src_ports, dst_ports;
4923 mirror_mask_t mirror_bit;
4924 struct port *out_port;
4931 if (strcmp(cfg->name, m->name)) {
4933 m->name = xstrdup(cfg->name);
4936 /* Get output port. */
4937 if (cfg->output_port) {
4938 out_port = port_lookup(m->bridge, cfg->output_port->name);
4940 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4941 m->bridge->name, m->name);
4947 if (cfg->output_vlan) {
4948 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4949 "output vlan; ignoring output vlan",
4950 m->bridge->name, m->name);
4952 } else if (cfg->output_vlan) {
4954 out_vlan = *cfg->output_vlan;
4956 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4957 m->bridge->name, m->name);
4962 sset_init(&src_ports);
4963 sset_init(&dst_ports);
4964 if (cfg->select_all) {
4965 HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) {
4966 sset_add(&src_ports, port->name);
4967 sset_add(&dst_ports, port->name);
4972 /* Get ports, and drop duplicates and ports that don't exist. */
4973 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4975 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4978 /* Get all the vlans, and drop duplicate and invalid vlans. */
4979 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4982 /* Update mirror data. */
4983 if (!sset_equals(&m->src_ports, &src_ports)
4984 || !sset_equals(&m->dst_ports, &dst_ports)
4985 || m->n_vlans != n_vlans
4986 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4987 || m->out_port != out_port
4988 || m->out_vlan != out_vlan) {
4989 bridge_flush(m->bridge);
4990 mac_learning_flush(m->bridge->ml);
4992 sset_swap(&m->src_ports, &src_ports);
4993 sset_swap(&m->dst_ports, &dst_ports);
4996 m->n_vlans = n_vlans;
4997 m->out_port = out_port;
4998 m->out_vlan = out_vlan;
5001 mirror_bit = MIRROR_MASK_C(1) << m->idx;
5002 HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) {
5003 if (sset_contains(&m->src_ports, port->name)) {
5004 port->src_mirrors |= mirror_bit;
5006 port->src_mirrors &= ~mirror_bit;
5009 if (sset_contains(&m->dst_ports, port->name)) {
5010 port->dst_mirrors |= mirror_bit;
5012 port->dst_mirrors &= ~mirror_bit;
5017 sset_destroy(&src_ports);
5018 sset_destroy(&dst_ports);