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"
65 #include "system-stats.h"
70 #include "vswitchd/vswitch-idl.h"
71 #include "xenserver.h"
73 #include "sflow_api.h"
75 VLOG_DEFINE_THIS_MODULE(bridge);
77 COVERAGE_DEFINE(bridge_flush);
78 COVERAGE_DEFINE(bridge_process_flow);
79 COVERAGE_DEFINE(bridge_process_cfm);
80 COVERAGE_DEFINE(bridge_process_lacp);
81 COVERAGE_DEFINE(bridge_reconfigure);
82 COVERAGE_DEFINE(bridge_lacp_update);
90 struct dst builtin[32];
95 static void dst_set_init(struct dst_set *);
96 static void dst_set_add(struct dst_set *, const struct dst *);
97 static void dst_set_free(struct dst_set *);
100 /* These members are always valid. */
101 struct list port_elem; /* Element in struct port's "ifaces" list. */
102 struct port *port; /* Containing port. */
103 char *name; /* Host network device name. */
104 tag_type tag; /* Tag associated with this interface. */
105 long long delay_expires; /* Time after which 'enabled' may change. */
107 /* These members are valid only after bridge_reconfigure() causes them to
109 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
110 int dp_ifidx; /* Index within kernel datapath. */
111 struct netdev *netdev; /* Network device. */
112 bool enabled; /* May be chosen for flows? */
113 bool up; /* Is the interface up? */
114 const char *type; /* Usually same as cfg->type. */
115 const struct ovsrec_interface *cfg;
117 /* LACP information. */
118 uint16_t lacp_priority; /* LACP port priority. */
121 #define BOND_MASK 0xff
123 struct iface *iface; /* Assigned iface, or NULL if none. */
124 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
125 tag_type tag; /* Tag for bond_entry<->iface association. */
129 BM_TCP, /* Transport Layer Load Balance. */
130 BM_SLB, /* Source Load Balance. */
131 BM_AB /* Active Backup. */
134 #define MAX_MIRRORS 32
135 typedef uint32_t mirror_mask_t;
136 #define MIRROR_MASK_C(X) UINT32_C(X)
137 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
139 struct bridge *bridge;
142 struct uuid uuid; /* UUID of this "mirror" record in database. */
144 /* Selection criteria. */
145 struct shash src_ports; /* Name is port name; data is always NULL. */
146 struct shash dst_ports; /* Name is port name; data is always NULL. */
151 struct port *out_port;
155 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
157 struct bridge *bridge;
158 struct hmap_node hmap_node; /* Element in struct bridge's "ports" hmap. */
161 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
162 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
163 * NULL if all VLANs are trunked. */
164 const struct ovsrec_port *cfg;
167 struct netdev_monitor *monitor; /* Tracks carrier. NULL if miimon. */
168 long long int miimon_interval; /* Miimon status refresh interval. */
169 long long int miimon_next_update; /* Time of next miimon update. */
171 /* An ordinary bridge port has 1 interface.
172 * A bridge port for bonding has at least 2 interfaces. */
173 struct list ifaces; /* List of "struct iface"s. */
174 size_t n_ifaces; /* list_size(ifaces). */
177 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
178 struct iface *active_iface; /* iface on which bcasts accepted, or NULL. */
179 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
180 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
181 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
182 long long int bond_next_fake_iface_update; /* Time of next update. */
184 /* LACP information. */
185 struct lacp *lacp; /* LACP object. NULL if LACP is disabled. */
186 bool lacp_active; /* True if LACP is active */
187 bool lacp_fast; /* True if LACP is in fast mode. */
188 uint16_t lacp_priority; /* LACP system priority. */
190 /* SLB specific bonding info. */
191 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
192 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
193 long long int bond_next_rebalance; /* Next rebalancing time. */
195 /* Port mirroring info. */
196 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
197 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
198 bool is_mirror_output_port; /* Does port mirroring send frames here? */
202 struct list node; /* Node in global list of bridges. */
203 char *name; /* User-specified arbitrary name. */
204 struct mac_learning *ml; /* MAC learning table. */
205 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
206 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
207 const struct ovsrec_bridge *cfg;
209 /* OpenFlow switch processing. */
210 struct ofproto *ofproto; /* OpenFlow switch. */
212 /* Kernel datapath information. */
213 struct dpif *dpif; /* Datapath. */
214 struct hmap ifaces; /* Contains "struct iface"s. */
217 struct hmap ports; /* "struct port"s indexed by name. */
218 struct shash iface_by_name; /* "struct iface"s indexed by name. */
221 bool has_bonded_ports;
226 /* Port mirroring. */
227 struct mirror *mirrors[MAX_MIRRORS];
230 /* List of all bridges. */
231 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
233 /* OVSDB IDL used to obtain configuration. */
234 static struct ovsdb_idl *idl;
236 /* Each time this timer expires, the bridge fetches systems and interface
237 * statistics and pushes them into the database. */
238 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
239 static long long int stats_timer = LLONG_MIN;
241 /* Stores the time after which CFM statistics may be written to the database.
242 * Only updated when changes to the database require rate limiting. */
243 #define CFM_LIMIT_INTERVAL (1 * 1000) /* In milliseconds. */
244 static long long int cfm_limiter = LLONG_MIN;
246 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
247 static void bridge_destroy(struct bridge *);
248 static struct bridge *bridge_lookup(const char *name);
249 static unixctl_cb_func bridge_unixctl_dump_flows;
250 static unixctl_cb_func bridge_unixctl_reconnect;
251 static int bridge_run_one(struct bridge *);
252 static size_t bridge_get_controllers(const struct bridge *br,
253 struct ovsrec_controller ***controllersp);
254 static void bridge_reconfigure_one(struct bridge *);
255 static void bridge_reconfigure_remotes(struct bridge *,
256 const struct sockaddr_in *managers,
258 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
259 static void bridge_fetch_dp_ifaces(struct bridge *);
260 static void bridge_flush(struct bridge *);
261 static void bridge_pick_local_hw_addr(struct bridge *,
262 uint8_t ea[ETH_ADDR_LEN],
263 struct iface **hw_addr_iface);
264 static uint64_t bridge_pick_datapath_id(struct bridge *,
265 const uint8_t bridge_ea[ETH_ADDR_LEN],
266 struct iface *hw_addr_iface);
267 static uint64_t dpid_from_hash(const void *, size_t nbytes);
269 static unixctl_cb_func bridge_unixctl_fdb_show;
270 static unixctl_cb_func qos_unixctl_show;
272 static void bond_init(void);
273 static void bond_run(struct port *);
274 static void bond_wait(struct port *);
275 static void bond_rebalance_port(struct port *);
276 static void bond_send_learning_packets(struct port *);
277 static void bond_enable_slave(struct iface *iface, bool enable);
279 static void port_run(struct port *);
280 static void port_wait(struct port *);
281 static struct port *port_create(struct bridge *, const char *name);
282 static void port_reconfigure(struct port *, const struct ovsrec_port *);
283 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
284 static void port_destroy(struct port *);
285 static struct port *port_lookup(const struct bridge *, const char *name);
286 static struct iface *port_lookup_iface(const struct port *, const char *name);
287 static struct iface *port_get_an_iface(const struct port *);
288 static struct port *port_from_dp_ifidx(const struct bridge *,
290 static void port_update_bonding(struct port *);
291 static void port_update_lacp(struct port *);
293 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
294 static void mirror_destroy(struct mirror *);
295 static void mirror_reconfigure(struct bridge *);
296 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
297 static bool vlan_is_mirrored(const struct mirror *, int vlan);
299 static struct iface *iface_create(struct port *port,
300 const struct ovsrec_interface *if_cfg);
301 static void iface_destroy(struct iface *);
302 static struct iface *iface_lookup(const struct bridge *, const char *name);
303 static struct iface *iface_find(const char *name);
304 static struct iface *iface_from_dp_ifidx(const struct bridge *,
306 static void iface_set_mac(struct iface *);
307 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
308 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
309 static void iface_update_cfm(struct iface *);
310 static bool iface_refresh_cfm_stats(struct iface *iface);
311 static void iface_update_carrier(struct iface *);
312 static bool iface_get_carrier(const struct iface *);
314 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
316 static void shash_to_ovs_idl_map(struct shash *,
317 char ***keys, char ***values, size_t *n);
319 /* Hooks into ofproto processing. */
320 static struct ofhooks bridge_ofhooks;
322 /* Public functions. */
324 /* Initializes the bridge module, configuring it to obtain its configuration
325 * from an OVSDB server accessed over 'remote', which should be a string in a
326 * form acceptable to ovsdb_idl_create(). */
328 bridge_init(const char *remote)
330 /* Create connection to database. */
331 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
333 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
334 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
335 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
337 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
339 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
340 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
342 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
343 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
344 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
346 /* Register unixctl commands. */
347 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
348 unixctl_command_register("qos/show", qos_unixctl_show, NULL);
349 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
351 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
360 struct bridge *br, *next_br;
362 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
365 ovsdb_idl_destroy(idl);
368 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
369 * but for which the ovs-vswitchd configuration 'cfg' is required. */
371 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
373 static bool already_configured_once;
374 struct svec bridge_names;
375 struct svec dpif_names, dpif_types;
378 /* Only do this once per ovs-vswitchd run. */
379 if (already_configured_once) {
382 already_configured_once = true;
384 stats_timer = time_msec() + STATS_INTERVAL;
386 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
387 svec_init(&bridge_names);
388 for (i = 0; i < cfg->n_bridges; i++) {
389 svec_add(&bridge_names, cfg->bridges[i]->name);
391 svec_sort(&bridge_names);
393 /* Iterate over all system dpifs and delete any of them that do not appear
395 svec_init(&dpif_names);
396 svec_init(&dpif_types);
397 dp_enumerate_types(&dpif_types);
398 for (i = 0; i < dpif_types.n; i++) {
401 dp_enumerate_names(dpif_types.names[i], &dpif_names);
403 /* Delete each dpif whose name is not in 'bridge_names'. */
404 for (j = 0; j < dpif_names.n; j++) {
405 if (!svec_contains(&bridge_names, dpif_names.names[j])) {
409 retval = dpif_open(dpif_names.names[j], dpif_types.names[i],
418 svec_destroy(&bridge_names);
419 svec_destroy(&dpif_names);
420 svec_destroy(&dpif_types);
423 /* Callback for iterate_and_prune_ifaces(). */
425 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
427 if (!iface->netdev) {
428 /* We already reported a related error, don't bother duplicating it. */
432 if (iface->dp_ifidx < 0) {
433 VLOG_ERR("%s interface not in %s, dropping",
434 iface->name, dpif_name(br->dpif));
438 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
439 iface->name, iface->dp_ifidx);
443 /* Callback for iterate_and_prune_ifaces(). */
445 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
446 void *aux OVS_UNUSED)
448 /* Set policing attributes. */
449 netdev_set_policing(iface->netdev,
450 iface->cfg->ingress_policing_rate,
451 iface->cfg->ingress_policing_burst);
453 /* Set MAC address of internal interfaces other than the local
455 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
456 iface_set_mac(iface);
462 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
463 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
464 * deletes from 'br' any ports that no longer have any interfaces. */
466 iterate_and_prune_ifaces(struct bridge *br,
467 bool (*cb)(struct bridge *, struct iface *,
471 struct port *port, *next_port;
473 HMAP_FOR_EACH_SAFE (port, next_port, hmap_node, &br->ports) {
474 struct iface *iface, *next_iface;
476 LIST_FOR_EACH_SAFE (iface, next_iface, port_elem, &port->ifaces) {
477 if (!cb(br, iface, aux)) {
478 iface_set_ofport(iface->cfg, -1);
479 iface_destroy(iface);
483 if (!port->n_ifaces) {
484 VLOG_WARN("%s port has no interfaces, dropping", port->name);
490 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
491 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
492 * responsible for freeing '*managersp' (with free()).
494 * You may be asking yourself "why does ovs-vswitchd care?", because
495 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
496 * should not be and in fact is not directly involved in that. But
497 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
498 * it has to tell in-band control where the managers are to enable that.
499 * (Thus, only managers connected in-band are collected.)
502 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
503 struct sockaddr_in **managersp, size_t *n_managersp)
505 struct sockaddr_in *managers = NULL;
506 size_t n_managers = 0;
507 struct shash targets;
510 /* Collect all of the potential targets from the "targets" columns of the
511 * rows pointed to by "manager_options", excluding any that are
513 shash_init(&targets);
514 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
515 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
517 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
518 shash_find_and_delete(&targets, m->target);
520 shash_add_once(&targets, m->target, NULL);
524 /* Now extract the targets' IP addresses. */
525 if (!shash_is_empty(&targets)) {
526 struct shash_node *node;
528 managers = xmalloc(shash_count(&targets) * sizeof *managers);
529 SHASH_FOR_EACH (node, &targets) {
530 const char *target = node->name;
531 struct sockaddr_in *sin = &managers[n_managers];
533 if ((!strncmp(target, "tcp:", 4)
534 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
535 (!strncmp(target, "ssl:", 4)
536 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
541 shash_destroy(&targets);
543 *managersp = managers;
544 *n_managersp = n_managers;
548 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
550 struct shash old_br, new_br;
551 struct shash_node *node;
552 struct bridge *br, *next;
553 struct sockaddr_in *managers;
556 int sflow_bridge_number;
558 COVERAGE_INC(bridge_reconfigure);
560 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
562 /* Collect old and new bridges. */
565 LIST_FOR_EACH (br, node, &all_bridges) {
566 shash_add(&old_br, br->name, br);
568 for (i = 0; i < ovs_cfg->n_bridges; i++) {
569 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
570 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
571 VLOG_WARN("more than one bridge named %s", br_cfg->name);
575 /* Get rid of deleted bridges and add new bridges. */
576 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
577 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
584 SHASH_FOR_EACH (node, &new_br) {
585 const char *br_name = node->name;
586 const struct ovsrec_bridge *br_cfg = node->data;
587 br = shash_find_data(&old_br, br_name);
589 /* If the bridge datapath type has changed, we need to tear it
590 * down and recreate. */
591 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
593 bridge_create(br_cfg);
596 bridge_create(br_cfg);
599 shash_destroy(&old_br);
600 shash_destroy(&new_br);
602 /* Reconfigure all bridges. */
603 LIST_FOR_EACH (br, node, &all_bridges) {
604 bridge_reconfigure_one(br);
607 /* Add and delete ports on all datapaths.
609 * The kernel will reject any attempt to add a given port to a datapath if
610 * that port already belongs to a different datapath, so we must do all
611 * port deletions before any port additions. */
612 LIST_FOR_EACH (br, node, &all_bridges) {
613 struct dpif_port_dump dump;
614 struct shash want_ifaces;
615 struct dpif_port dpif_port;
617 bridge_get_all_ifaces(br, &want_ifaces);
618 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
619 if (!shash_find(&want_ifaces, dpif_port.name)
620 && strcmp(dpif_port.name, br->name)) {
621 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
623 VLOG_WARN("failed to remove %s interface from %s: %s",
624 dpif_port.name, dpif_name(br->dpif),
629 shash_destroy(&want_ifaces);
631 LIST_FOR_EACH (br, node, &all_bridges) {
632 struct shash cur_ifaces, want_ifaces;
633 struct dpif_port_dump dump;
634 struct dpif_port dpif_port;
636 /* Get the set of interfaces currently in this datapath. */
637 shash_init(&cur_ifaces);
638 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
639 struct dpif_port *port_info = xmalloc(sizeof *port_info);
640 dpif_port_clone(port_info, &dpif_port);
641 shash_add(&cur_ifaces, dpif_port.name, port_info);
644 /* Get the set of interfaces we want on this datapath. */
645 bridge_get_all_ifaces(br, &want_ifaces);
647 hmap_clear(&br->ifaces);
648 SHASH_FOR_EACH (node, &want_ifaces) {
649 const char *if_name = node->name;
650 struct iface *iface = node->data;
651 struct dpif_port *dpif_port;
655 type = iface ? iface->type : "internal";
656 dpif_port = shash_find_data(&cur_ifaces, if_name);
658 /* If we have a port or a netdev already, and it's not the type we
659 * want, then delete the port (if any) and close the netdev (if
661 if ((dpif_port && strcmp(dpif_port->type, type))
662 || (iface && iface->netdev
663 && strcmp(type, netdev_get_type(iface->netdev)))) {
665 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
672 netdev_close(iface->netdev);
673 iface->netdev = NULL;
677 /* If the port doesn't exist or we don't have the netdev open,
678 * we need to do more work. */
679 if (!dpif_port || (iface && !iface->netdev)) {
680 struct netdev_options options;
681 struct netdev *netdev;
684 /* First open the network device. */
685 options.name = if_name;
687 options.args = &args;
688 options.ethertype = NETDEV_ETH_TYPE_NONE;
692 shash_from_ovs_idl_map(iface->cfg->key_options,
693 iface->cfg->value_options,
694 iface->cfg->n_options, &args);
696 error = netdev_open(&options, &netdev);
697 shash_destroy(&args);
700 VLOG_WARN("could not open network device %s (%s)",
701 if_name, strerror(error));
705 /* Then add the port if we haven't already. */
707 error = dpif_port_add(br->dpif, netdev, NULL);
709 netdev_close(netdev);
710 if (error == EFBIG) {
711 VLOG_ERR("ran out of valid port numbers on %s",
712 dpif_name(br->dpif));
715 VLOG_WARN("failed to add %s interface to %s: %s",
716 if_name, dpif_name(br->dpif),
723 /* Update 'iface'. */
725 iface->netdev = netdev;
726 iface->enabled = iface_get_carrier(iface);
727 iface->up = iface->enabled;
729 } else if (iface && iface->netdev) {
733 shash_from_ovs_idl_map(iface->cfg->key_options,
734 iface->cfg->value_options,
735 iface->cfg->n_options, &args);
736 netdev_set_config(iface->netdev, &args);
737 shash_destroy(&args);
740 shash_destroy(&want_ifaces);
742 SHASH_FOR_EACH (node, &cur_ifaces) {
743 struct dpif_port *port_info = node->data;
744 dpif_port_destroy(port_info);
747 shash_destroy(&cur_ifaces);
749 sflow_bridge_number = 0;
750 LIST_FOR_EACH (br, node, &all_bridges) {
753 struct iface *local_iface;
754 struct iface *hw_addr_iface;
757 bridge_fetch_dp_ifaces(br);
759 iterate_and_prune_ifaces(br, check_iface, NULL);
761 /* Pick local port hardware address, datapath ID. */
762 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
763 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
765 int error = netdev_set_etheraddr(local_iface->netdev, ea);
767 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
768 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
769 "Ethernet address: %s",
770 br->name, strerror(error));
773 memcpy(br->ea, ea, ETH_ADDR_LEN);
775 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
776 ofproto_set_datapath_id(br->ofproto, dpid);
778 dpid_string = xasprintf("%016"PRIx64, dpid);
779 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
782 /* Set NetFlow configuration on this bridge. */
783 if (br->cfg->netflow) {
784 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
785 struct netflow_options opts;
787 memset(&opts, 0, sizeof opts);
789 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
790 if (nf_cfg->engine_type) {
791 opts.engine_type = *nf_cfg->engine_type;
793 if (nf_cfg->engine_id) {
794 opts.engine_id = *nf_cfg->engine_id;
797 opts.active_timeout = nf_cfg->active_timeout;
798 if (!opts.active_timeout) {
799 opts.active_timeout = -1;
800 } else if (opts.active_timeout < 0) {
801 VLOG_WARN("bridge %s: active timeout interval set to negative "
802 "value, using default instead (%d seconds)", br->name,
803 NF_ACTIVE_TIMEOUT_DEFAULT);
804 opts.active_timeout = -1;
807 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
808 if (opts.add_id_to_iface) {
809 if (opts.engine_id > 0x7f) {
810 VLOG_WARN("bridge %s: netflow port mangling may conflict "
811 "with another vswitch, choose an engine id less "
812 "than 128", br->name);
814 if (hmap_count(&br->ports) > 508) {
815 VLOG_WARN("bridge %s: netflow port mangling will conflict "
816 "with another port when more than 508 ports are "
821 opts.collectors.n = nf_cfg->n_targets;
822 opts.collectors.names = nf_cfg->targets;
823 if (ofproto_set_netflow(br->ofproto, &opts)) {
824 VLOG_ERR("bridge %s: problem setting netflow collectors",
828 ofproto_set_netflow(br->ofproto, NULL);
831 /* Set sFlow configuration on this bridge. */
832 if (br->cfg->sflow) {
833 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
834 struct ovsrec_controller **controllers;
835 struct ofproto_sflow_options oso;
836 size_t n_controllers;
838 memset(&oso, 0, sizeof oso);
840 oso.targets.n = sflow_cfg->n_targets;
841 oso.targets.names = sflow_cfg->targets;
843 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
844 if (sflow_cfg->sampling) {
845 oso.sampling_rate = *sflow_cfg->sampling;
848 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
849 if (sflow_cfg->polling) {
850 oso.polling_interval = *sflow_cfg->polling;
853 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
854 if (sflow_cfg->header) {
855 oso.header_len = *sflow_cfg->header;
858 oso.sub_id = sflow_bridge_number++;
859 oso.agent_device = sflow_cfg->agent;
861 oso.control_ip = NULL;
862 n_controllers = bridge_get_controllers(br, &controllers);
863 for (i = 0; i < n_controllers; i++) {
864 if (controllers[i]->local_ip) {
865 oso.control_ip = controllers[i]->local_ip;
869 ofproto_set_sflow(br->ofproto, &oso);
871 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
873 ofproto_set_sflow(br->ofproto, NULL);
876 /* Update the controller and related settings. It would be more
877 * straightforward to call this from bridge_reconfigure_one(), but we
878 * can't do it there for two reasons. First, and most importantly, at
879 * that point we don't know the dp_ifidx of any interfaces that have
880 * been added to the bridge (because we haven't actually added them to
881 * the datapath). Second, at that point we haven't set the datapath ID
882 * yet; when a controller is configured, resetting the datapath ID will
883 * immediately disconnect from the controller, so it's better to set
884 * the datapath ID before the controller. */
885 bridge_reconfigure_remotes(br, managers, n_managers);
887 LIST_FOR_EACH (br, node, &all_bridges) {
890 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
894 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
895 netdev_monitor_add(port->monitor, iface->netdev);
898 port->miimon_next_update = 0;
901 port_update_lacp(port);
902 port_update_bonding(port);
904 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
905 iface_update_qos(iface, port->cfg->qos);
909 LIST_FOR_EACH (br, node, &all_bridges) {
910 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
913 LIST_FOR_EACH (br, node, &all_bridges) {
915 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
916 iface_update_cfm(iface);
922 /* ovs-vswitchd has completed initialization, so allow the process that
923 * forked us to exit successfully. */
924 daemonize_complete();
928 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
929 const struct ovsdb_idl_column *column,
932 const struct ovsdb_datum *datum;
933 union ovsdb_atom atom;
936 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
937 atom.string = (char *) key;
938 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
939 return idx == UINT_MAX ? NULL : datum->values[idx].string;
943 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
945 return get_ovsrec_key_value(&br_cfg->header_,
946 &ovsrec_bridge_col_other_config, key);
950 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
951 struct iface **hw_addr_iface)
957 *hw_addr_iface = NULL;
959 /* Did the user request a particular MAC? */
960 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
961 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
962 if (eth_addr_is_multicast(ea)) {
963 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
964 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
965 } else if (eth_addr_is_zero(ea)) {
966 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
972 /* Otherwise choose the minimum non-local MAC address among all of the
974 memset(ea, 0xff, ETH_ADDR_LEN);
975 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
976 uint8_t iface_ea[ETH_ADDR_LEN];
977 struct iface *candidate;
980 /* Mirror output ports don't participate. */
981 if (port->is_mirror_output_port) {
985 /* Choose the MAC address to represent the port. */
987 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
988 /* Find the interface with this Ethernet address (if any) so that
989 * we can provide the correct devname to the caller. */
990 LIST_FOR_EACH (candidate, port_elem, &port->ifaces) {
991 uint8_t candidate_ea[ETH_ADDR_LEN];
992 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
993 && eth_addr_equals(iface_ea, candidate_ea)) {
998 /* Choose the interface whose MAC address will represent the port.
999 * The Linux kernel bonding code always chooses the MAC address of
1000 * the first slave added to a bond, and the Fedora networking
1001 * scripts always add slaves to a bond in alphabetical order, so
1002 * for compatibility we choose the interface with the name that is
1003 * first in alphabetical order. */
1004 LIST_FOR_EACH (candidate, port_elem, &port->ifaces) {
1005 if (!iface || strcmp(candidate->name, iface->name) < 0) {
1010 /* The local port doesn't count (since we're trying to choose its
1011 * MAC address anyway). */
1012 if (iface->dp_ifidx == ODPP_LOCAL) {
1017 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1019 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1020 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1021 iface->name, strerror(error));
1026 /* Compare against our current choice. */
1027 if (!eth_addr_is_multicast(iface_ea) &&
1028 !eth_addr_is_local(iface_ea) &&
1029 !eth_addr_is_reserved(iface_ea) &&
1030 !eth_addr_is_zero(iface_ea) &&
1031 eth_addr_compare_3way(iface_ea, ea) < 0)
1033 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1034 *hw_addr_iface = iface;
1037 if (eth_addr_is_multicast(ea)) {
1038 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1039 *hw_addr_iface = NULL;
1040 VLOG_WARN("bridge %s: using default bridge Ethernet "
1041 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1043 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1044 br->name, ETH_ADDR_ARGS(ea));
1048 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1049 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1050 * an interface on 'br', then that interface must be passed in as
1051 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1052 * 'hw_addr_iface' must be passed in as a null pointer. */
1054 bridge_pick_datapath_id(struct bridge *br,
1055 const uint8_t bridge_ea[ETH_ADDR_LEN],
1056 struct iface *hw_addr_iface)
1059 * The procedure for choosing a bridge MAC address will, in the most
1060 * ordinary case, also choose a unique MAC that we can use as a datapath
1061 * ID. In some special cases, though, multiple bridges will end up with
1062 * the same MAC address. This is OK for the bridges, but it will confuse
1063 * the OpenFlow controller, because each datapath needs a unique datapath
1066 * Datapath IDs must be unique. It is also very desirable that they be
1067 * stable from one run to the next, so that policy set on a datapath
1070 const char *datapath_id;
1073 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1074 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1078 if (hw_addr_iface) {
1080 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1082 * A bridge whose MAC address is taken from a VLAN network device
1083 * (that is, a network device created with vconfig(8) or similar
1084 * tool) will have the same MAC address as a bridge on the VLAN
1085 * device's physical network device.
1087 * Handle this case by hashing the physical network device MAC
1088 * along with the VLAN identifier.
1090 uint8_t buf[ETH_ADDR_LEN + 2];
1091 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1092 buf[ETH_ADDR_LEN] = vlan >> 8;
1093 buf[ETH_ADDR_LEN + 1] = vlan;
1094 return dpid_from_hash(buf, sizeof buf);
1097 * Assume that this bridge's MAC address is unique, since it
1098 * doesn't fit any of the cases we handle specially.
1103 * A purely internal bridge, that is, one that has no non-virtual
1104 * network devices on it at all, is more difficult because it has no
1105 * natural unique identifier at all.
1107 * When the host is a XenServer, we handle this case by hashing the
1108 * host's UUID with the name of the bridge. Names of bridges are
1109 * persistent across XenServer reboots, although they can be reused if
1110 * an internal network is destroyed and then a new one is later
1111 * created, so this is fairly effective.
1113 * When the host is not a XenServer, we punt by using a random MAC
1114 * address on each run.
1116 const char *host_uuid = xenserver_get_host_uuid();
1118 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1119 dpid = dpid_from_hash(combined, strlen(combined));
1125 return eth_addr_to_uint64(bridge_ea);
1129 dpid_from_hash(const void *data, size_t n)
1131 uint8_t hash[SHA1_DIGEST_SIZE];
1133 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1134 sha1_bytes(data, n, hash);
1135 eth_addr_mark_random(hash);
1136 return eth_addr_to_uint64(hash);
1140 iface_refresh_status(struct iface *iface)
1144 enum netdev_flags flags;
1153 if (!netdev_get_status(iface->netdev, &sh)) {
1155 char **keys, **values;
1157 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1158 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1163 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1166 shash_destroy_free_data(&sh);
1168 error = netdev_get_flags(iface->netdev, &flags);
1170 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1173 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1176 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1178 ovsrec_interface_set_duplex(iface->cfg,
1179 netdev_features_is_full_duplex(current)
1181 /* warning: uint64_t -> int64_t conversion */
1182 bps = netdev_features_to_bps(current);
1183 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1186 ovsrec_interface_set_duplex(iface->cfg, NULL);
1187 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1191 ovsrec_interface_set_link_state(iface->cfg,
1192 iface_get_carrier(iface) ? "up" : "down");
1194 error = netdev_get_mtu(iface->netdev, &mtu);
1195 if (!error && mtu != INT_MAX) {
1197 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1200 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1204 /* Writes 'iface''s CFM statistics to the database. Returns true if anything
1205 * changed, false otherwise. */
1207 iface_refresh_cfm_stats(struct iface *iface)
1209 const struct ovsrec_monitor *mon;
1210 const struct cfm *cfm;
1211 bool changed = false;
1214 mon = iface->cfg->monitor;
1215 cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
1221 for (i = 0; i < mon->n_remote_mps; i++) {
1222 const struct ovsrec_maintenance_point *mp;
1223 const struct remote_mp *rmp;
1225 mp = mon->remote_mps[i];
1226 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1228 if (mp->n_fault != 1 || mp->fault[0] != rmp->fault) {
1229 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1234 if (mon->n_fault != 1 || mon->fault[0] != cfm->fault) {
1235 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1243 iface_refresh_stats(struct iface *iface)
1249 static const struct iface_stat iface_stats[] = {
1250 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1251 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1252 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1253 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1254 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1255 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1256 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1257 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1258 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1259 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1260 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1261 { "collisions", offsetof(struct netdev_stats, collisions) },
1263 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1264 const struct iface_stat *s;
1266 char *keys[N_STATS];
1267 int64_t values[N_STATS];
1270 struct netdev_stats stats;
1272 /* Intentionally ignore return value, since errors will set 'stats' to
1273 * all-1s, and we will deal with that correctly below. */
1274 netdev_get_stats(iface->netdev, &stats);
1277 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1278 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1279 if (value != UINT64_MAX) {
1286 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1290 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1292 struct ovsdb_datum datum;
1296 get_system_stats(&stats);
1298 ovsdb_datum_from_shash(&datum, &stats);
1299 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1303 static inline const char *
1304 nx_role_to_str(enum nx_role role)
1309 case NX_ROLE_MASTER:
1314 return "*** INVALID ROLE ***";
1319 bridge_refresh_controller_status(const struct bridge *br)
1322 const struct ovsrec_controller *cfg;
1324 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1326 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1327 struct ofproto_controller_info *cinfo =
1328 shash_find_data(&info, cfg->target);
1331 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1332 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1333 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1334 (char **) cinfo->pairs.values,
1337 ovsrec_controller_set_is_connected(cfg, false);
1338 ovsrec_controller_set_role(cfg, NULL);
1339 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1343 ofproto_free_ofproto_controller_info(&info);
1349 const struct ovsrec_open_vswitch *cfg;
1351 bool datapath_destroyed;
1352 bool database_changed;
1355 /* Let each bridge do the work that it needs to do. */
1356 datapath_destroyed = false;
1357 LIST_FOR_EACH (br, node, &all_bridges) {
1358 int error = bridge_run_one(br);
1360 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1361 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1362 "forcing reconfiguration", br->name);
1363 datapath_destroyed = true;
1367 /* (Re)configure if necessary. */
1368 database_changed = ovsdb_idl_run(idl);
1369 cfg = ovsrec_open_vswitch_first(idl);
1371 /* Re-configure SSL. We do this on every trip through the main loop,
1372 * instead of just when the database changes, because the contents of the
1373 * key and certificate files can change without the database changing.
1375 * We do this before bridge_reconfigure() because that function might
1376 * initiate SSL connections and thus requires SSL to be configured. */
1377 if (cfg && cfg->ssl) {
1378 const struct ovsrec_ssl *ssl = cfg->ssl;
1380 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1381 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1384 if (database_changed || datapath_destroyed) {
1386 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1388 bridge_configure_once(cfg);
1389 bridge_reconfigure(cfg);
1391 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1392 ovsdb_idl_txn_commit(txn);
1393 ovsdb_idl_txn_destroy(txn); /* XXX */
1395 /* We still need to reconfigure to avoid dangling pointers to
1396 * now-destroyed ovsrec structures inside bridge data. */
1397 static const struct ovsrec_open_vswitch null_cfg;
1399 bridge_reconfigure(&null_cfg);
1403 /* Refresh system and interface stats if necessary. */
1404 if (time_msec() >= stats_timer) {
1406 struct ovsdb_idl_txn *txn;
1408 txn = ovsdb_idl_txn_create(idl);
1409 LIST_FOR_EACH (br, node, &all_bridges) {
1412 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1413 struct iface *iface;
1415 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
1416 iface_refresh_stats(iface);
1417 iface_refresh_status(iface);
1420 bridge_refresh_controller_status(br);
1422 refresh_system_stats(cfg);
1423 ovsdb_idl_txn_commit(txn);
1424 ovsdb_idl_txn_destroy(txn); /* XXX */
1427 stats_timer = time_msec() + STATS_INTERVAL;
1430 if (time_msec() >= cfm_limiter) {
1431 struct ovsdb_idl_txn *txn;
1432 bool changed = false;
1434 txn = ovsdb_idl_txn_create(idl);
1435 LIST_FOR_EACH (br, node, &all_bridges) {
1438 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1439 struct iface *iface;
1441 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
1442 changed = iface_refresh_cfm_stats(iface) || changed;
1448 cfm_limiter = time_msec() + CFM_LIMIT_INTERVAL;
1451 ovsdb_idl_txn_commit(txn);
1452 ovsdb_idl_txn_destroy(txn);
1461 LIST_FOR_EACH (br, node, &all_bridges) {
1464 ofproto_wait(br->ofproto);
1465 if (ofproto_has_primary_controller(br->ofproto)) {
1469 mac_learning_wait(br->ml);
1471 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1475 ovsdb_idl_wait(idl);
1476 poll_timer_wait_until(stats_timer);
1478 if (cfm_limiter > time_msec()) {
1479 poll_timer_wait_until(cfm_limiter);
1483 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1484 * configuration changes. */
1486 bridge_flush(struct bridge *br)
1488 COVERAGE_INC(bridge_flush);
1492 /* Bridge unixctl user interface functions. */
1494 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1495 const char *args, void *aux OVS_UNUSED)
1497 struct ds ds = DS_EMPTY_INITIALIZER;
1498 const struct bridge *br;
1499 const struct mac_entry *e;
1501 br = bridge_lookup(args);
1503 unixctl_command_reply(conn, 501, "no such bridge");
1507 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1508 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1509 struct port *port = e->port.p;
1510 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1511 port_get_an_iface(port)->dp_ifidx,
1512 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1514 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1518 /* QoS unixctl user interface functions. */
1520 struct qos_unixctl_show_cbdata {
1522 struct iface *iface;
1526 qos_unixctl_show_cb(unsigned int queue_id,
1527 const struct shash *details,
1530 struct qos_unixctl_show_cbdata *data = aux;
1531 struct ds *ds = data->ds;
1532 struct iface *iface = data->iface;
1533 struct netdev_queue_stats stats;
1534 struct shash_node *node;
1537 ds_put_cstr(ds, "\n");
1539 ds_put_format(ds, "Queue %u:\n", queue_id);
1541 ds_put_cstr(ds, "Default:\n");
1544 SHASH_FOR_EACH (node, details) {
1545 ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data);
1548 error = netdev_get_queue_stats(iface->netdev, queue_id, &stats);
1550 if (stats.tx_packets != UINT64_MAX) {
1551 ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets);
1554 if (stats.tx_bytes != UINT64_MAX) {
1555 ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes);
1558 if (stats.tx_errors != UINT64_MAX) {
1559 ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors);
1562 ds_put_format(ds, "\tFailed to get statistics for queue %u: %s",
1563 queue_id, strerror(error));
1568 qos_unixctl_show(struct unixctl_conn *conn,
1569 const char *args, void *aux OVS_UNUSED)
1571 struct ds ds = DS_EMPTY_INITIALIZER;
1572 struct shash sh = SHASH_INITIALIZER(&sh);
1573 struct iface *iface;
1575 struct shash_node *node;
1576 struct qos_unixctl_show_cbdata data;
1579 iface = iface_find(args);
1581 unixctl_command_reply(conn, 501, "no such interface");
1585 netdev_get_qos(iface->netdev, &type, &sh);
1587 if (*type != '\0') {
1588 ds_put_format(&ds, "QoS: %s %s\n", iface->name, type);
1590 SHASH_FOR_EACH (node, &sh) {
1591 ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data);
1596 error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data);
1599 ds_put_format(&ds, "failed to dump queues: %s", strerror(error));
1601 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1603 ds_put_format(&ds, "QoS not configured on %s\n", iface->name);
1604 unixctl_command_reply(conn, 501, ds_cstr(&ds));
1607 shash_destroy_free_data(&sh);
1611 /* Bridge reconfiguration functions. */
1612 static struct bridge *
1613 bridge_create(const struct ovsrec_bridge *br_cfg)
1618 assert(!bridge_lookup(br_cfg->name));
1619 br = xzalloc(sizeof *br);
1621 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1627 dpif_flow_flush(br->dpif);
1629 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1632 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1634 dpif_delete(br->dpif);
1635 dpif_close(br->dpif);
1640 br->name = xstrdup(br_cfg->name);
1642 br->ml = mac_learning_create();
1643 eth_addr_nicira_random(br->default_ea);
1645 hmap_init(&br->ports);
1646 hmap_init(&br->ifaces);
1647 shash_init(&br->iface_by_name);
1651 list_push_back(&all_bridges, &br->node);
1653 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1659 bridge_destroy(struct bridge *br)
1662 struct port *port, *next;
1665 HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) {
1668 list_remove(&br->node);
1669 error = dpif_delete(br->dpif);
1670 if (error && error != ENOENT) {
1671 VLOG_ERR("failed to delete %s: %s",
1672 dpif_name(br->dpif), strerror(error));
1674 dpif_close(br->dpif);
1675 ofproto_destroy(br->ofproto);
1676 mac_learning_destroy(br->ml);
1677 hmap_destroy(&br->ifaces);
1678 hmap_destroy(&br->ports);
1679 shash_destroy(&br->iface_by_name);
1685 static struct bridge *
1686 bridge_lookup(const char *name)
1690 LIST_FOR_EACH (br, node, &all_bridges) {
1691 if (!strcmp(br->name, name)) {
1698 /* Handle requests for a listing of all flows known by the OpenFlow
1699 * stack, including those normally hidden. */
1701 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1702 const char *args, void *aux OVS_UNUSED)
1707 br = bridge_lookup(args);
1709 unixctl_command_reply(conn, 501, "Unknown bridge");
1714 ofproto_get_all_flows(br->ofproto, &results);
1716 unixctl_command_reply(conn, 200, ds_cstr(&results));
1717 ds_destroy(&results);
1720 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1721 * connections and reconnect. If BRIDGE is not specified, then all bridges
1722 * drop their controller connections and reconnect. */
1724 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1725 const char *args, void *aux OVS_UNUSED)
1728 if (args[0] != '\0') {
1729 br = bridge_lookup(args);
1731 unixctl_command_reply(conn, 501, "Unknown bridge");
1734 ofproto_reconnect_controllers(br->ofproto);
1736 LIST_FOR_EACH (br, node, &all_bridges) {
1737 ofproto_reconnect_controllers(br->ofproto);
1740 unixctl_command_reply(conn, 200, NULL);
1744 bridge_run_one(struct bridge *br)
1749 error = ofproto_run1(br->ofproto);
1754 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1756 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1760 error = ofproto_run2(br->ofproto, br->flush);
1767 bridge_get_controllers(const struct bridge *br,
1768 struct ovsrec_controller ***controllersp)
1770 struct ovsrec_controller **controllers;
1771 size_t n_controllers;
1773 controllers = br->cfg->controller;
1774 n_controllers = br->cfg->n_controller;
1776 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1782 *controllersp = controllers;
1784 return n_controllers;
1788 bridge_reconfigure_one(struct bridge *br)
1790 enum ofproto_fail_mode fail_mode;
1791 struct svec snoops, old_snoops;
1792 struct port *port, *next;
1793 struct shash_node *node;
1794 struct shash new_ports;
1797 /* Collect new ports. */
1798 shash_init(&new_ports);
1799 for (i = 0; i < br->cfg->n_ports; i++) {
1800 const char *name = br->cfg->ports[i]->name;
1801 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1802 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1807 /* If we have a controller, then we need a local port. Complain if the
1808 * user didn't specify one.
1810 * XXX perhaps we should synthesize a port ourselves in this case. */
1811 if (bridge_get_controllers(br, NULL)) {
1812 char local_name[IF_NAMESIZE];
1815 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1816 local_name, sizeof local_name);
1817 if (!error && !shash_find(&new_ports, local_name)) {
1818 VLOG_WARN("bridge %s: controller specified but no local port "
1819 "(port named %s) defined",
1820 br->name, local_name);
1824 /* Get rid of deleted ports.
1825 * Get rid of deleted interfaces on ports that still exist. */
1826 HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) {
1827 const struct ovsrec_port *port_cfg;
1829 port_cfg = shash_find_data(&new_ports, port->name);
1833 port_del_ifaces(port, port_cfg);
1837 /* Create new ports.
1838 * Add new interfaces to existing ports.
1839 * Reconfigure existing ports. */
1840 SHASH_FOR_EACH (node, &new_ports) {
1841 struct port *port = port_lookup(br, node->name);
1843 port = port_create(br, node->name);
1846 port_reconfigure(port, node->data);
1847 if (!port->n_ifaces) {
1848 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1849 br->name, port->name);
1853 shash_destroy(&new_ports);
1855 /* Set the fail-mode */
1856 fail_mode = !br->cfg->fail_mode
1857 || !strcmp(br->cfg->fail_mode, "standalone")
1858 ? OFPROTO_FAIL_STANDALONE
1859 : OFPROTO_FAIL_SECURE;
1860 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1861 && !ofproto_has_primary_controller(br->ofproto)) {
1862 ofproto_flush_flows(br->ofproto);
1864 ofproto_set_fail_mode(br->ofproto, fail_mode);
1866 /* Delete all flows if we're switching from connected to standalone or vice
1867 * versa. (XXX Should we delete all flows if we are switching from one
1868 * controller to another?) */
1870 /* Configure OpenFlow controller connection snooping. */
1872 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1873 ovs_rundir(), br->name));
1874 svec_init(&old_snoops);
1875 ofproto_get_snoops(br->ofproto, &old_snoops);
1876 if (!svec_equal(&snoops, &old_snoops)) {
1877 ofproto_set_snoops(br->ofproto, &snoops);
1879 svec_destroy(&snoops);
1880 svec_destroy(&old_snoops);
1882 mirror_reconfigure(br);
1885 /* Initializes 'oc' appropriately as a management service controller for
1888 * The caller must free oc->target when it is no longer needed. */
1890 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1891 struct ofproto_controller *oc)
1893 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1894 oc->max_backoff = 0;
1895 oc->probe_interval = 60;
1896 oc->band = OFPROTO_OUT_OF_BAND;
1898 oc->burst_limit = 0;
1901 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1903 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1904 struct ofproto_controller *oc)
1906 oc->target = c->target;
1907 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1908 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1909 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1910 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1911 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1912 oc->burst_limit = (c->controller_burst_limit
1913 ? *c->controller_burst_limit : 0);
1916 /* Configures the IP stack for 'br''s local interface properly according to the
1917 * configuration in 'c'. */
1919 bridge_configure_local_iface_netdev(struct bridge *br,
1920 struct ovsrec_controller *c)
1922 struct netdev *netdev;
1923 struct in_addr mask, gateway;
1925 struct iface *local_iface;
1928 /* If there's no local interface or no IP address, give up. */
1929 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
1930 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1934 /* Bring up the local interface. */
1935 netdev = local_iface->netdev;
1936 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1938 /* Configure the IP address and netmask. */
1939 if (!c->local_netmask
1940 || !inet_aton(c->local_netmask, &mask)
1942 mask.s_addr = guess_netmask(ip.s_addr);
1944 if (!netdev_set_in4(netdev, ip, mask)) {
1945 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1946 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1949 /* Configure the default gateway. */
1950 if (c->local_gateway
1951 && inet_aton(c->local_gateway, &gateway)
1952 && gateway.s_addr) {
1953 if (!netdev_add_router(netdev, gateway)) {
1954 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1955 br->name, IP_ARGS(&gateway.s_addr));
1961 bridge_reconfigure_remotes(struct bridge *br,
1962 const struct sockaddr_in *managers,
1965 const char *disable_ib_str, *queue_id_str;
1966 bool disable_in_band = false;
1969 struct ovsrec_controller **controllers;
1970 size_t n_controllers;
1973 struct ofproto_controller *ocs;
1977 /* Check if we should disable in-band control on this bridge. */
1978 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1979 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1980 disable_in_band = true;
1983 /* Set OpenFlow queue ID for in-band control. */
1984 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
1985 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
1986 ofproto_set_in_band_queue(br->ofproto, queue_id);
1988 if (disable_in_band) {
1989 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1991 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1993 had_primary = ofproto_has_primary_controller(br->ofproto);
1995 n_controllers = bridge_get_controllers(br, &controllers);
1997 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
2000 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
2001 for (i = 0; i < n_controllers; i++) {
2002 struct ovsrec_controller *c = controllers[i];
2004 if (!strncmp(c->target, "punix:", 6)
2005 || !strncmp(c->target, "unix:", 5)) {
2006 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2008 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
2009 * domain sockets and overwriting arbitrary local files. */
2010 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2011 "\"%s\" due to possibility for remote exploit",
2012 dpif_name(br->dpif), c->target);
2016 bridge_configure_local_iface_netdev(br, c);
2017 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2018 if (disable_in_band) {
2019 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2024 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2025 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2028 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2029 ofproto_flush_flows(br->ofproto);
2032 /* If there are no controllers and the bridge is in standalone
2033 * mode, set up a flow that matches every packet and directs
2034 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2035 * switch is in secure mode and we won't pass any traffic until
2036 * a controller has been defined and it tells us to do so. */
2038 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2039 union ofp_action action;
2040 struct cls_rule rule;
2042 memset(&action, 0, sizeof action);
2043 action.type = htons(OFPAT_OUTPUT);
2044 action.output.len = htons(sizeof action);
2045 action.output.port = htons(OFPP_NORMAL);
2046 cls_rule_init_catchall(&rule, 0);
2047 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2052 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2057 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2058 struct iface *iface;
2060 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2061 shash_add_once(ifaces, iface->name, iface);
2063 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2064 shash_add_once(ifaces, port->name, NULL);
2069 /* For robustness, in case the administrator moves around datapath ports behind
2070 * our back, we re-check all the datapath port numbers here.
2072 * This function will set the 'dp_ifidx' members of interfaces that have
2073 * disappeared to -1, so only call this function from a context where those
2074 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2075 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2076 * datapath, which doesn't support UINT16_MAX+1 ports. */
2078 bridge_fetch_dp_ifaces(struct bridge *br)
2080 struct dpif_port_dump dump;
2081 struct dpif_port dpif_port;
2084 /* Reset all interface numbers. */
2085 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2086 struct iface *iface;
2088 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2089 iface->dp_ifidx = -1;
2092 hmap_clear(&br->ifaces);
2094 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2095 struct iface *iface = iface_lookup(br, dpif_port.name);
2097 if (iface->dp_ifidx >= 0) {
2098 VLOG_WARN("%s reported interface %s twice",
2099 dpif_name(br->dpif), dpif_port.name);
2100 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2101 VLOG_WARN("%s reported interface %"PRIu16" twice",
2102 dpif_name(br->dpif), dpif_port.port_no);
2104 iface->dp_ifidx = dpif_port.port_no;
2105 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2106 hash_int(iface->dp_ifidx, 0));
2109 iface_set_ofport(iface->cfg,
2110 (iface->dp_ifidx >= 0
2111 ? odp_port_to_ofp_port(iface->dp_ifidx)
2117 /* Bridge packet processing functions. */
2120 bond_is_tcp_hash(const struct port *port)
2122 return port->bond_mode == BM_TCP && lacp_negotiated(port->lacp);
2126 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2128 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2131 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2133 struct flow hash_flow;
2135 memcpy(&hash_flow, flow, sizeof hash_flow);
2136 hash_flow.vlan_tci = 0;
2138 /* The symmetric quality of this hash function is not required, but
2139 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2140 * purposes, so we use it out of convenience. */
2141 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2144 static struct bond_entry *
2145 lookup_bond_entry(const struct port *port, const struct flow *flow,
2148 assert(port->bond_mode != BM_AB);
2150 if (bond_is_tcp_hash(port)) {
2151 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2153 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2157 static struct iface *
2158 bond_choose_iface(const struct port *port)
2160 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2161 struct iface *best_down_slave;
2162 struct iface *iface;
2164 best_down_slave = NULL;
2165 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2166 if (iface->enabled) {
2168 } else if ((!best_down_slave
2169 || iface->delay_expires < best_down_slave->delay_expires)
2170 && lacp_slave_may_enable(port->lacp, iface)) {
2171 best_down_slave = iface;
2175 if (best_down_slave) {
2176 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2177 "since no other interface is up",
2178 best_down_slave->name,
2179 best_down_slave->delay_expires - time_msec());
2180 bond_enable_slave(best_down_slave, true);
2183 return best_down_slave;
2187 choose_output_iface(const struct port *port, const struct flow *flow,
2188 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2190 struct iface *iface;
2192 assert(port->n_ifaces);
2193 if (port->n_ifaces == 1) {
2194 iface = port_get_an_iface(port);
2195 } else if (port->bond_mode == BM_AB) {
2196 iface = port->active_iface;
2198 *tags |= port->no_ifaces_tag;
2202 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2203 if (!e->iface || !e->iface->enabled) {
2204 /* XXX select interface properly. The current interface selection
2205 * is only good for testing the rebalancing code. */
2206 e->iface = bond_choose_iface(port);
2208 *tags |= port->no_ifaces_tag;
2211 e->tag = tag_create_random();
2216 *dp_ifidx = iface->dp_ifidx;
2217 *tags |= iface->tag; /* Currently only used for bonding. */
2222 bond_link_status_update(struct iface *iface)
2224 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2225 struct port *port = iface->port;
2226 bool up = iface->up && lacp_slave_may_enable(port->lacp, iface);
2227 int updelay, downdelay;
2229 updelay = port->updelay;
2230 downdelay = port->downdelay;
2232 if (lacp_negotiated(port->lacp)) {
2237 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2238 /* Nothing to do. */
2241 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2242 iface->name, up ? "up" : "down");
2243 if (up == iface->enabled) {
2244 iface->delay_expires = LLONG_MAX;
2245 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2246 iface->name, up ? "disabled" : "enabled");
2247 } else if (up && !port->active_iface) {
2248 bond_enable_slave(iface, true);
2250 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2251 "other interface is up", iface->name, updelay);
2254 int delay = up ? updelay : downdelay;
2255 iface->delay_expires = time_msec() + delay;
2258 "interface %s: will be %s if it stays %s for %d ms",
2260 up ? "enabled" : "disabled",
2268 bond_choose_active_iface(struct port *port)
2270 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2272 port->active_iface = bond_choose_iface(port);
2273 if (port->active_iface) {
2274 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2275 port->name, port->active_iface->name);
2277 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2283 bond_enable_slave(struct iface *iface, bool enable)
2285 struct port *port = iface->port;
2286 struct bridge *br = port->bridge;
2288 /* This acts as a recursion check. If the act of disabling a slave
2289 * causes a different slave to be enabled, the flag will allow us to
2290 * skip redundant work when we reenter this function. It must be
2291 * cleared on exit to keep things safe with multiple bonds. */
2292 static bool moving_active_iface = false;
2294 iface->delay_expires = LLONG_MAX;
2295 if (enable == iface->enabled) {
2299 iface->enabled = enable;
2300 if (!iface->enabled) {
2301 VLOG_WARN("interface %s: disabled", iface->name);
2302 ofproto_revalidate(br->ofproto, iface->tag);
2303 if (iface == port->active_iface) {
2304 /* Disabling a slave can lead to another slave being immediately
2305 * enabled if there will be no active slaves but one is waiting
2306 * on an updelay. In this case we do not need to run most of the
2307 * code for the newly enabled slave since there was no period
2308 * without an active slave and it is redundant with the disabling
2310 moving_active_iface = true;
2311 bond_choose_active_iface(port);
2313 bond_send_learning_packets(port);
2315 VLOG_WARN("interface %s: enabled", iface->name);
2316 if (!port->active_iface && !moving_active_iface) {
2317 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2318 bond_choose_active_iface(port);
2319 bond_send_learning_packets(port);
2321 iface->tag = tag_create_random();
2324 moving_active_iface = false;
2327 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2328 * bond interface. */
2330 bond_update_fake_iface_stats(struct port *port)
2332 struct netdev_stats bond_stats;
2333 struct netdev *bond_dev;
2334 struct iface *iface;
2336 memset(&bond_stats, 0, sizeof bond_stats);
2338 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2339 struct netdev_stats slave_stats;
2341 if (!netdev_get_stats(iface->netdev, &slave_stats)) {
2342 /* XXX: We swap the stats here because they are swapped back when
2343 * reported by the internal device. The reason for this is
2344 * internal devices normally represent packets going into the system
2345 * but when used as fake bond device they represent packets leaving
2346 * the system. We really should do this in the internal device
2347 * itself because changing it here reverses the counts from the
2348 * perspective of the switch. However, the internal device doesn't
2349 * know what type of device it represents so we have to do it here
2351 bond_stats.tx_packets += slave_stats.rx_packets;
2352 bond_stats.tx_bytes += slave_stats.rx_bytes;
2353 bond_stats.rx_packets += slave_stats.tx_packets;
2354 bond_stats.rx_bytes += slave_stats.tx_bytes;
2358 if (!netdev_open_default(port->name, &bond_dev)) {
2359 netdev_set_stats(bond_dev, &bond_stats);
2360 netdev_close(bond_dev);
2365 bond_run(struct port *port)
2367 struct iface *iface;
2369 if (port->n_ifaces < 2) {
2373 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2374 bond_link_status_update(iface);
2377 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2378 if (time_msec() >= iface->delay_expires) {
2379 bond_enable_slave(iface, !iface->enabled);
2383 if (port->bond_fake_iface
2384 && time_msec() >= port->bond_next_fake_iface_update) {
2385 bond_update_fake_iface_stats(port);
2386 port->bond_next_fake_iface_update = time_msec() + 1000;
2391 bond_wait(struct port *port)
2393 struct iface *iface;
2395 if (port->n_ifaces < 2) {
2399 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2400 if (iface->delay_expires != LLONG_MAX) {
2401 poll_timer_wait_until(iface->delay_expires);
2405 if (port->bond_fake_iface) {
2406 poll_timer_wait_until(port->bond_next_fake_iface_update);
2411 set_dst(struct dst *dst, const struct flow *flow,
2412 const struct port *in_port, const struct port *out_port,
2415 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2416 : in_port->vlan >= 0 ? in_port->vlan
2417 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2418 : vlan_tci_to_vid(flow->vlan_tci));
2419 return choose_output_iface(out_port, flow, dst->vlan,
2420 &dst->dp_ifidx, tags);
2424 swap_dst(struct dst *p, struct dst *q)
2426 struct dst tmp = *p;
2431 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2432 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2433 * that we push to the datapath. We could in fact fully sort the array by
2434 * vlan, but in most cases there are at most two different vlan tags so that's
2435 * possibly overkill.) */
2437 partition_dsts(struct dst_set *set, int vlan)
2439 struct dst *first = set->dsts;
2440 struct dst *last = set->dsts + set->n;
2442 while (first != last) {
2444 * - All dsts < first have vlan == 'vlan'.
2445 * - All dsts >= last have vlan != 'vlan'.
2446 * - first < last. */
2447 while (first->vlan == vlan) {
2448 if (++first == last) {
2453 /* Same invariants, plus one additional:
2454 * - first->vlan != vlan.
2456 while (last[-1].vlan != vlan) {
2457 if (--last == first) {
2462 /* Same invariants, plus one additional:
2463 * - last[-1].vlan == vlan.*/
2464 swap_dst(first++, --last);
2469 mirror_mask_ffs(mirror_mask_t mask)
2471 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2476 dst_set_init(struct dst_set *set)
2478 set->dsts = set->builtin;
2480 set->allocated = ARRAY_SIZE(set->builtin);
2484 dst_set_add(struct dst_set *set, const struct dst *dst)
2486 if (set->n >= set->allocated) {
2487 size_t new_allocated;
2488 struct dst *new_dsts;
2490 new_allocated = set->allocated * 2;
2491 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2492 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2496 set->dsts = new_dsts;
2497 set->allocated = new_allocated;
2499 set->dsts[set->n++] = *dst;
2503 dst_set_free(struct dst_set *set)
2505 if (set->dsts != set->builtin) {
2511 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2514 for (i = 0; i < set->n; i++) {
2515 if (set->dsts[i].vlan == test->vlan
2516 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2524 port_trunks_vlan(const struct port *port, uint16_t vlan)
2526 return (port->vlan < 0
2527 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2531 port_includes_vlan(const struct port *port, uint16_t vlan)
2533 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2537 port_is_floodable(const struct port *port)
2539 struct iface *iface;
2541 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2542 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2550 /* Returns the tag for 'port''s active iface, or 'port''s no_ifaces_tag if
2551 * there is no active iface. */
2553 port_get_active_iface_tag(const struct port *port)
2555 return (port->active_iface
2556 ? port->active_iface->tag
2557 : port->no_ifaces_tag);
2560 /* Returns an arbitrary interface within 'port'.
2562 * 'port' must have at least one interface. */
2563 static struct iface *
2564 port_get_an_iface(const struct port *port)
2566 return CONTAINER_OF(list_front(&port->ifaces), struct iface, port_elem);
2570 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2571 const struct port *in_port, const struct port *out_port,
2572 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2574 mirror_mask_t mirrors = in_port->src_mirrors;
2578 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2579 if (flow_vlan == 0) {
2580 flow_vlan = OFP_VLAN_NONE;
2583 if (out_port == FLOOD_PORT) {
2586 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2588 && port_is_floodable(port)
2589 && port_includes_vlan(port, vlan)
2590 && !port->is_mirror_output_port
2591 && set_dst(&dst, flow, in_port, port, tags)) {
2592 mirrors |= port->dst_mirrors;
2593 dst_set_add(set, &dst);
2596 *nf_output_iface = NF_OUT_FLOOD;
2597 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2598 dst_set_add(set, &dst);
2599 *nf_output_iface = dst.dp_ifidx;
2600 mirrors |= out_port->dst_mirrors;
2604 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2605 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2607 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2608 && !dst_is_duplicate(set, &dst)) {
2609 dst_set_add(set, &dst);
2614 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2615 if (port_includes_vlan(port, m->out_vlan)
2616 && set_dst(&dst, flow, in_port, port, tags))
2618 if (port->vlan < 0) {
2619 dst.vlan = m->out_vlan;
2621 if (dst_is_duplicate(set, &dst)) {
2625 /* Use the vlan tag on the original flow instead of
2626 * the one passed in the vlan parameter. This ensures
2627 * that we compare the vlan from before any implicit
2628 * tagging tags place. This is necessary because
2629 * dst->vlan is the final vlan, after removing implicit
2631 if (port == in_port && dst.vlan == flow_vlan) {
2632 /* Don't send out input port on same VLAN. */
2635 dst_set_add(set, &dst);
2640 mirrors &= mirrors - 1;
2643 partition_dsts(set, flow_vlan);
2646 static void OVS_UNUSED
2647 print_dsts(const struct dst_set *set)
2651 for (i = 0; i < set->n; i++) {
2652 const struct dst *dst = &set->dsts[i];
2654 printf(">p%"PRIu16, dst->dp_ifidx);
2655 if (dst->vlan != OFP_VLAN_NONE) {
2656 printf("v%"PRIu16, dst->vlan);
2662 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2663 const struct port *in_port, const struct port *out_port,
2664 tag_type *tags, struct ofpbuf *actions,
2665 uint16_t *nf_output_iface)
2672 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2675 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2676 if (cur_vlan == 0) {
2677 cur_vlan = OFP_VLAN_NONE;
2679 for (i = 0; i < set.n; i++) {
2680 const struct dst *dst = &set.dsts[i];
2681 if (dst->vlan != cur_vlan) {
2682 if (dst->vlan == OFP_VLAN_NONE) {
2683 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2686 tci = htons(dst->vlan & VLAN_VID_MASK);
2687 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2688 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2690 cur_vlan = dst->vlan;
2692 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2697 /* Returns the effective vlan of a packet, taking into account both the
2698 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2699 * the packet is untagged and -1 indicates it has an invalid header and
2700 * should be dropped. */
2701 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2702 struct port *in_port, bool have_packet)
2704 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2705 if (in_port->vlan >= 0) {
2707 /* XXX support double tagging? */
2709 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2710 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2711 "packet received on port %s configured with "
2712 "implicit VLAN %"PRIu16,
2713 br->name, vlan, in_port->name, in_port->vlan);
2717 vlan = in_port->vlan;
2719 if (!port_includes_vlan(in_port, vlan)) {
2721 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2722 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2723 "packet received on port %s not configured for "
2725 br->name, vlan, in_port->name, vlan);
2734 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2735 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2736 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2738 is_gratuitous_arp(const struct flow *flow)
2740 return (flow->dl_type == htons(ETH_TYPE_ARP)
2741 && eth_addr_is_broadcast(flow->dl_dst)
2742 && (flow->nw_proto == ARP_OP_REPLY
2743 || (flow->nw_proto == ARP_OP_REQUEST
2744 && flow->nw_src == flow->nw_dst)));
2748 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2749 struct port *in_port)
2751 struct mac_entry *mac;
2753 if (!mac_learning_may_learn(br->ml, flow->dl_src, vlan)) {
2757 mac = mac_learning_insert(br->ml, flow->dl_src, vlan);
2758 if (is_gratuitous_arp(flow)) {
2759 /* We don't want to learn from gratuitous ARP packets that are
2760 * reflected back over bond slaves so we lock the learning table. */
2761 if (in_port->n_ifaces == 1) {
2762 mac_entry_set_grat_arp_lock(mac);
2763 } else if (mac_entry_is_grat_arp_locked(mac)) {
2768 if (mac_entry_is_new(mac) || mac->port.p != in_port) {
2769 /* The log messages here could actually be useful in debugging,
2770 * so keep the rate limit relatively high. */
2771 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
2772 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2773 "on port %s in VLAN %d",
2774 br->name, ETH_ADDR_ARGS(flow->dl_src),
2775 in_port->name, vlan);
2777 mac->port.p = in_port;
2778 ofproto_revalidate(br->ofproto, mac_learning_changed(br->ml, mac));
2782 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2783 * dropped. Returns true if they may be forwarded, false if they should be
2786 * If 'have_packet' is true, it indicates that the caller is processing a
2787 * received packet. If 'have_packet' is false, then the caller is just
2788 * revalidating an existing flow because configuration has changed. Either
2789 * way, 'have_packet' only affects logging (there is no point in logging errors
2790 * during revalidation).
2792 * Sets '*in_portp' to the input port. This will be a null pointer if
2793 * flow->in_port does not designate a known input port (in which case
2794 * is_admissible() returns false).
2796 * When returning true, sets '*vlanp' to the effective VLAN of the input
2797 * packet, as returned by flow_get_vlan().
2799 * May also add tags to '*tags', although the current implementation only does
2800 * so in one special case.
2803 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2804 tag_type *tags, int *vlanp, struct port **in_portp)
2806 struct iface *in_iface;
2807 struct port *in_port;
2810 /* Find the interface and port structure for the received packet. */
2811 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2813 /* No interface? Something fishy... */
2815 /* Odd. A few possible reasons here:
2817 * - We deleted an interface but there are still a few packets
2818 * queued up from it.
2820 * - Someone externally added an interface (e.g. with "ovs-dpctl
2821 * add-if") that we don't know about.
2823 * - Packet arrived on the local port but the local port is not
2824 * one of our bridge ports.
2826 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2828 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2829 "interface %"PRIu16, br->name, flow->in_port);
2835 *in_portp = in_port = in_iface->port;
2836 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2841 /* Drop frames for reserved multicast addresses. */
2842 if (eth_addr_is_reserved(flow->dl_dst)) {
2846 /* Drop frames on ports reserved for mirroring. */
2847 if (in_port->is_mirror_output_port) {
2849 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2850 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2851 "%s, which is reserved exclusively for mirroring",
2852 br->name, in_port->name);
2857 /* When using LACP, do not accept packets from disabled interfaces. */
2858 if (lacp_negotiated(in_port->lacp) && !in_iface->enabled) {
2862 /* Packets received on non-LACP bonds need special attention to avoid
2864 if (in_port->n_ifaces > 1 && !lacp_negotiated(in_port->lacp)) {
2865 struct mac_entry *mac;
2867 if (eth_addr_is_multicast(flow->dl_dst)) {
2868 *tags |= port_get_active_iface_tag(in_port);
2869 if (in_port->active_iface != in_iface) {
2870 /* Drop all multicast packets on inactive slaves. */
2875 /* Drop all packets for which we have learned a different input
2876 * port, because we probably sent the packet on one slave and got
2877 * it back on the other. Gratuitous ARP packets are an exception
2878 * to this rule: the host has moved to another switch. The exception
2879 * to the exception is if we locked the learning table to avoid
2880 * reflections on bond slaves. If this is the case, just drop the
2882 mac = mac_learning_lookup(br->ml, flow->dl_src, vlan, NULL);
2883 if (mac && mac->port.p != in_port &&
2884 (!is_gratuitous_arp(flow) || mac_entry_is_grat_arp_locked(mac))) {
2892 /* If the composed actions may be applied to any packet in the given 'flow',
2893 * returns true. Otherwise, the actions should only be applied to 'packet', or
2894 * not at all, if 'packet' was NULL. */
2896 process_flow(struct bridge *br, const struct flow *flow,
2897 const struct ofpbuf *packet, struct ofpbuf *actions,
2898 tag_type *tags, uint16_t *nf_output_iface)
2900 struct port *in_port;
2901 struct port *out_port;
2902 struct mac_entry *mac;
2905 /* Check whether we should drop packets in this flow. */
2906 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2911 /* Learn source MAC (but don't try to learn from revalidation). */
2913 update_learning_table(br, flow, vlan, in_port);
2916 /* Determine output port. */
2917 mac = mac_learning_lookup(br->ml, flow->dl_dst, vlan, tags);
2919 out_port = mac->port.p;
2920 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2921 /* If we are revalidating but don't have a learning entry then
2922 * eject the flow. Installing a flow that floods packets opens
2923 * up a window of time where we could learn from a packet reflected
2924 * on a bond and blackhole packets before the learning table is
2925 * updated to reflect the correct port. */
2928 out_port = FLOOD_PORT;
2931 /* Don't send packets out their input ports. */
2932 if (in_port == out_port) {
2938 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2946 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2947 struct ofpbuf *actions, tag_type *tags,
2948 uint16_t *nf_output_iface, void *br_)
2950 struct bridge *br = br_;
2952 COVERAGE_INC(bridge_process_flow);
2953 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2957 bridge_special_ofhook_cb(const struct flow *flow,
2958 const struct ofpbuf *packet, void *br_)
2960 struct iface *iface;
2961 struct bridge *br = br_;
2963 iface = iface_from_dp_ifidx(br, flow->in_port);
2965 if (flow->dl_type == htons(ETH_TYPE_LACP)) {
2967 if (iface && iface->port->lacp && packet) {
2968 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
2971 COVERAGE_INC(bridge_process_lacp);
2972 lacp_process_pdu(iface->port->lacp, iface, pdu);
2982 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
2983 const struct nlattr *actions,
2985 uint64_t n_bytes, void *br_)
2987 struct bridge *br = br_;
2988 const struct nlattr *a;
2989 struct port *in_port;
2994 /* Feed information from the active flows back into the learning table to
2995 * ensure that table is always in sync with what is actually flowing
2996 * through the datapath.
2998 * We test that 'tags' is nonzero to ensure that only flows that include an
2999 * OFPP_NORMAL action are used for learning. This works because
3000 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3001 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3002 update_learning_table(br, flow, vlan, in_port);
3005 /* Account for bond slave utilization. */
3006 if (!br->has_bonded_ports) {
3009 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3010 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3011 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3012 if (out_port && out_port->n_ifaces >= 2 &&
3013 out_port->bond_mode != BM_AB) {
3014 uint16_t vlan = (flow->vlan_tci
3015 ? vlan_tci_to_vid(flow->vlan_tci)
3017 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
3018 e->tx_bytes += n_bytes;
3025 bridge_account_checkpoint_ofhook_cb(void *br_)
3027 struct bridge *br = br_;
3031 if (!br->has_bonded_ports) {
3036 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3037 if (port->n_ifaces > 1 && port->bond_mode != BM_AB
3038 && now >= port->bond_next_rebalance) {
3039 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3040 bond_rebalance_port(port);
3045 static struct ofhooks bridge_ofhooks = {
3046 bridge_normal_ofhook_cb,
3047 bridge_special_ofhook_cb,
3048 bridge_account_flow_ofhook_cb,
3049 bridge_account_checkpoint_ofhook_cb,
3052 /* Bonding functions. */
3054 /* Statistics for a single interface on a bonded port, used for load-based
3055 * bond rebalancing. */
3056 struct slave_balance {
3057 struct iface *iface; /* The interface. */
3058 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3060 /* All the "bond_entry"s that are assigned to this interface, in order of
3061 * increasing tx_bytes. */
3062 struct bond_entry **hashes;
3067 bond_mode_to_string(enum bond_mode bm) {
3068 static char *bm_slb = "balance-slb";
3069 static char *bm_ab = "active-backup";
3070 static char *bm_tcp = "balance-tcp";
3073 case BM_SLB: return bm_slb;
3074 case BM_AB: return bm_ab;
3075 case BM_TCP: return bm_tcp;
3082 /* Sorts pointers to pointers to bond_entries in ascending order by the
3083 * interface to which they are assigned, and within a single interface in
3084 * ascending order of bytes transmitted. */
3086 compare_bond_entries(const void *a_, const void *b_)
3088 const struct bond_entry *const *ap = a_;
3089 const struct bond_entry *const *bp = b_;
3090 const struct bond_entry *a = *ap;
3091 const struct bond_entry *b = *bp;
3092 if (a->iface != b->iface) {
3093 return a->iface > b->iface ? 1 : -1;
3094 } else if (a->tx_bytes != b->tx_bytes) {
3095 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3101 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3102 * *descending* order by number of bytes transmitted. */
3104 compare_slave_balance(const void *a_, const void *b_)
3106 const struct slave_balance *a = a_;
3107 const struct slave_balance *b = b_;
3108 if (a->iface->enabled != b->iface->enabled) {
3109 return a->iface->enabled ? -1 : 1;
3110 } else if (a->tx_bytes != b->tx_bytes) {
3111 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3118 swap_bals(struct slave_balance *a, struct slave_balance *b)
3120 struct slave_balance tmp = *a;
3125 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3126 * given that 'p' (and only 'p') might be in the wrong location.
3128 * This function invalidates 'p', since it might now be in a different memory
3131 resort_bals(struct slave_balance *p,
3132 struct slave_balance bals[], size_t n_bals)
3135 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3136 swap_bals(p, p - 1);
3138 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3139 swap_bals(p, p + 1);
3145 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3147 if (VLOG_IS_DBG_ENABLED()) {
3148 struct ds ds = DS_EMPTY_INITIALIZER;
3149 const struct slave_balance *b;
3151 for (b = bals; b < bals + n_bals; b++) {
3155 ds_put_char(&ds, ',');
3157 ds_put_format(&ds, " %s %"PRIu64"kB",
3158 b->iface->name, b->tx_bytes / 1024);
3160 if (!b->iface->enabled) {
3161 ds_put_cstr(&ds, " (disabled)");
3163 if (b->n_hashes > 0) {
3164 ds_put_cstr(&ds, " (");
3165 for (i = 0; i < b->n_hashes; i++) {
3166 const struct bond_entry *e = b->hashes[i];
3168 ds_put_cstr(&ds, " + ");
3170 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3171 e - port->bond_hash, e->tx_bytes / 1024);
3173 ds_put_cstr(&ds, ")");
3176 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3181 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3183 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3186 struct bond_entry *hash = from->hashes[hash_idx];
3187 struct port *port = from->iface->port;
3188 uint64_t delta = hash->tx_bytes;
3190 assert(port->bond_mode != BM_AB);
3192 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3193 "from %s to %s (now carrying %"PRIu64"kB and "
3194 "%"PRIu64"kB load, respectively)",
3195 port->name, delta / 1024, hash - port->bond_hash,
3196 from->iface->name, to->iface->name,
3197 (from->tx_bytes - delta) / 1024,
3198 (to->tx_bytes + delta) / 1024);
3200 /* Delete element from from->hashes.
3202 * We don't bother to add the element to to->hashes because not only would
3203 * it require more work, the only purpose it would be to allow that hash to
3204 * be migrated to another slave in this rebalancing run, and there is no
3205 * point in doing that. */
3206 if (hash_idx == 0) {
3209 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3210 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3214 /* Shift load away from 'from' to 'to'. */
3215 from->tx_bytes -= delta;
3216 to->tx_bytes += delta;
3218 /* Arrange for flows to be revalidated. */
3219 ofproto_revalidate(port->bridge->ofproto, hash->tag);
3220 hash->iface = to->iface;
3221 hash->tag = tag_create_random();
3225 bond_rebalance_port(struct port *port)
3227 struct slave_balance *bals;
3229 struct bond_entry *hashes[BOND_MASK + 1];
3230 struct slave_balance *b, *from, *to;
3231 struct bond_entry *e;
3232 struct iface *iface;
3235 assert(port->bond_mode != BM_AB);
3237 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3238 * descending order of tx_bytes, so that bals[0] represents the most
3239 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3242 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3243 * array for each slave_balance structure, we sort our local array of
3244 * hashes in order by slave, so that all of the hashes for a given slave
3245 * become contiguous in memory, and then we point each 'hashes' members of
3246 * a slave_balance structure to the start of a contiguous group. */
3247 n_bals = port->n_ifaces;
3248 b = bals = xmalloc(n_bals * sizeof *bals);
3249 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3256 assert(b == &bals[n_bals]);
3257 for (i = 0; i <= BOND_MASK; i++) {
3258 hashes[i] = &port->bond_hash[i];
3260 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3261 for (i = 0; i <= BOND_MASK; i++) {
3267 for (b = bals; b < &bals[n_bals]; b++) {
3268 if (b->iface == e->iface) {
3269 b->tx_bytes += e->tx_bytes;
3271 b->hashes = &hashes[i];
3278 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3279 log_bals(bals, n_bals, port);
3281 /* Discard slaves that aren't enabled (which were sorted to the back of the
3282 * array earlier). */
3283 while (!bals[n_bals - 1].iface->enabled) {
3290 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3291 to = &bals[n_bals - 1];
3292 for (from = bals; from < to; ) {
3293 uint64_t overload = from->tx_bytes - to->tx_bytes;
3294 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3295 /* The extra load on 'from' (and all less-loaded slaves), compared
3296 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3297 * it is less than ~1Mbps. No point in rebalancing. */
3299 } else if (from->n_hashes == 1) {
3300 /* 'from' only carries a single MAC hash, so we can't shift any
3301 * load away from it, even though we want to. */
3304 /* 'from' is carrying significantly more load than 'to', and that
3305 * load is split across at least two different hashes. Pick a hash
3306 * to migrate to 'to' (the least-loaded slave), given that doing so
3307 * must decrease the ratio of the load on the two slaves by at
3310 * The sort order we use means that we prefer to shift away the
3311 * smallest hashes instead of the biggest ones. There is little
3312 * reason behind this decision; we could use the opposite sort
3313 * order to shift away big hashes ahead of small ones. */
3316 for (i = 0; i < from->n_hashes; i++) {
3317 double old_ratio, new_ratio;
3318 uint64_t delta = from->hashes[i]->tx_bytes;
3320 if (delta == 0 || from->tx_bytes - delta == 0) {
3321 /* Pointless move. */
3325 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3327 if (to->tx_bytes == 0) {
3328 /* Nothing on the new slave, move it. */
3332 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3333 new_ratio = (double)(from->tx_bytes - delta) /
3334 (to->tx_bytes + delta);
3336 if (new_ratio == 0) {
3337 /* Should already be covered but check to prevent division
3342 if (new_ratio < 1) {
3343 new_ratio = 1 / new_ratio;
3346 if (old_ratio - new_ratio > 0.1) {
3347 /* Would decrease the ratio, move it. */
3351 if (i < from->n_hashes) {
3352 bond_shift_load(from, to, i);
3354 /* If the result of the migration changed the relative order of
3355 * 'from' and 'to' swap them back to maintain invariants. */
3356 if (order_swapped) {
3357 swap_bals(from, to);
3360 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3361 * point to different slave_balance structures. It is only
3362 * valid to do these two operations in a row at all because we
3363 * know that 'from' will not move past 'to' and vice versa. */
3364 resort_bals(from, bals, n_bals);
3365 resort_bals(to, bals, n_bals);
3372 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3373 * historical data to decay to <1% in 7 rebalancing runs. */
3374 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3386 bond_send_learning_packets(struct port *port)
3388 struct bridge *br = port->bridge;
3389 struct mac_entry *e;
3390 struct ofpbuf packet;
3391 int error, n_packets, n_errors;
3393 if (!port->n_ifaces || !port->active_iface || bond_is_tcp_hash(port)) {
3397 ofpbuf_init(&packet, 128);
3398 error = n_packets = n_errors = 0;
3399 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3405 if (e->port.p == port) {
3409 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3411 flow_extract(&packet, 0, ODPP_NONE, &flow);
3413 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3419 retval = ofproto_send_packet(br->ofproto, dp_ifidx, e->vlan, &packet);
3425 ofpbuf_uninit(&packet);
3428 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3429 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3430 "packets, last error was: %s",
3431 port->name, n_errors, n_packets, strerror(error));
3433 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3434 port->name, n_packets);
3438 /* Bonding unixctl user interface functions. */
3441 bond_unixctl_list(struct unixctl_conn *conn,
3442 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3444 struct ds ds = DS_EMPTY_INITIALIZER;
3445 const struct bridge *br;
3447 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3449 LIST_FOR_EACH (br, node, &all_bridges) {
3452 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3453 if (port->n_ifaces > 1) {
3454 struct iface *iface;
3456 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3457 bond_mode_to_string(port->bond_mode));
3458 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3459 if (&iface->port_elem != list_front(&port->ifaces)) {
3460 ds_put_cstr(&ds, ", ");
3462 ds_put_cstr(&ds, iface->name);
3464 ds_put_char(&ds, '\n');
3468 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3472 static struct port *
3473 bond_find(const char *name)
3475 const struct bridge *br;
3477 LIST_FOR_EACH (br, node, &all_bridges) {
3480 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3481 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3490 bond_unixctl_show(struct unixctl_conn *conn,
3491 const char *args, void *aux OVS_UNUSED)
3493 struct ds ds = DS_EMPTY_INITIALIZER;
3494 const struct port *port;
3495 struct iface *iface;
3497 port = bond_find(args);
3499 unixctl_command_reply(conn, 501, "no such bond");
3503 ds_put_format(&ds, "bond_mode: %s\n",
3504 bond_mode_to_string(port->bond_mode));
3507 ds_put_format(&ds, "lacp: %s\n",
3508 port->lacp_active ? "active" : "passive");
3510 ds_put_cstr(&ds, "lacp: off\n");
3513 if (port->bond_mode != BM_AB) {
3514 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3515 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3519 ds_put_format(&ds, "bond-detect-mode: %s\n",
3520 port->monitor ? "carrier" : "miimon");
3522 if (!port->monitor) {
3523 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3524 port->miimon_interval);
3527 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3528 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3530 if (port->bond_mode != BM_AB) {
3531 ds_put_format(&ds, "next rebalance: %lld ms\n",
3532 port->bond_next_rebalance - time_msec());
3535 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3536 struct bond_entry *be;
3540 ds_put_format(&ds, "\nslave %s: %s\n",
3541 iface->name, iface->enabled ? "enabled" : "disabled");
3542 if (iface == port->active_iface) {
3543 ds_put_cstr(&ds, "\tactive slave\n");
3545 if (iface->delay_expires != LLONG_MAX) {
3546 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3547 iface->enabled ? "downdelay" : "updelay",
3548 iface->delay_expires - time_msec());
3551 if (port->bond_mode == BM_AB) {
3556 memset(&flow, 0, sizeof flow);
3557 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3558 int hash = be - port->bond_hash;
3559 struct mac_entry *me;
3561 if (be->iface != iface) {
3565 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3566 hash, be->tx_bytes / 1024);
3568 if (port->bond_mode != BM_SLB) {
3573 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3577 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
3578 if (bond_hash_src(me->mac, me->vlan) == hash
3579 && me->port.p != port
3580 && choose_output_iface(port, &flow, me->vlan,
3582 && dp_ifidx == iface->dp_ifidx)
3584 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3585 ETH_ADDR_ARGS(me->mac));
3590 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3595 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3596 void *aux OVS_UNUSED)
3598 char *args = (char *) args_;
3599 char *save_ptr = NULL;
3600 char *bond_s, *hash_s, *slave_s;
3602 struct iface *iface;
3603 struct bond_entry *entry;
3606 bond_s = strtok_r(args, " ", &save_ptr);
3607 hash_s = strtok_r(NULL, " ", &save_ptr);
3608 slave_s = strtok_r(NULL, " ", &save_ptr);
3610 unixctl_command_reply(conn, 501,
3611 "usage: bond/migrate BOND HASH SLAVE");
3615 port = bond_find(bond_s);
3617 unixctl_command_reply(conn, 501, "no such bond");
3621 if (port->bond_mode != BM_SLB) {
3622 unixctl_command_reply(conn, 501, "not an SLB bond");
3626 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3627 hash = atoi(hash_s) & BOND_MASK;
3629 unixctl_command_reply(conn, 501, "bad hash");
3633 iface = port_lookup_iface(port, slave_s);
3635 unixctl_command_reply(conn, 501, "no such slave");
3639 if (!iface->enabled) {
3640 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3644 entry = &port->bond_hash[hash];
3645 ofproto_revalidate(port->bridge->ofproto, entry->tag);
3646 entry->iface = iface;
3647 entry->tag = tag_create_random();
3648 unixctl_command_reply(conn, 200, "migrated");
3652 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3653 void *aux OVS_UNUSED)
3655 char *args = (char *) args_;
3656 char *save_ptr = NULL;
3657 char *bond_s, *slave_s;
3659 struct iface *iface;
3661 bond_s = strtok_r(args, " ", &save_ptr);
3662 slave_s = strtok_r(NULL, " ", &save_ptr);
3664 unixctl_command_reply(conn, 501,
3665 "usage: bond/set-active-slave BOND SLAVE");
3669 port = bond_find(bond_s);
3671 unixctl_command_reply(conn, 501, "no such bond");
3675 iface = port_lookup_iface(port, slave_s);
3677 unixctl_command_reply(conn, 501, "no such slave");
3681 if (!iface->enabled) {
3682 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3686 if (port->active_iface != iface) {
3687 ofproto_revalidate(port->bridge->ofproto,
3688 port_get_active_iface_tag(port));
3689 port->active_iface = iface;
3690 VLOG_INFO("port %s: active interface is now %s",
3691 port->name, iface->name);
3692 bond_send_learning_packets(port);
3693 unixctl_command_reply(conn, 200, "done");
3695 unixctl_command_reply(conn, 200, "no change");
3700 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3702 char *args = (char *) args_;
3703 char *save_ptr = NULL;
3704 char *bond_s, *slave_s;
3706 struct iface *iface;
3708 bond_s = strtok_r(args, " ", &save_ptr);
3709 slave_s = strtok_r(NULL, " ", &save_ptr);
3711 unixctl_command_reply(conn, 501,
3712 "usage: bond/enable/disable-slave BOND SLAVE");
3716 port = bond_find(bond_s);
3718 unixctl_command_reply(conn, 501, "no such bond");
3722 iface = port_lookup_iface(port, slave_s);
3724 unixctl_command_reply(conn, 501, "no such slave");
3728 bond_enable_slave(iface, enable);
3729 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3733 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3734 void *aux OVS_UNUSED)
3736 enable_slave(conn, args, true);
3740 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3741 void *aux OVS_UNUSED)
3743 enable_slave(conn, args, false);
3747 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3748 void *aux OVS_UNUSED)
3750 char *args = (char *) args_;
3751 uint8_t mac[ETH_ADDR_LEN];
3755 char *mac_s, *vlan_s;
3756 char *save_ptr = NULL;
3758 mac_s = strtok_r(args, " ", &save_ptr);
3759 vlan_s = strtok_r(NULL, " ", &save_ptr);
3762 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3763 unixctl_command_reply(conn, 501, "invalid vlan");
3767 vlan = OFP_VLAN_NONE;
3770 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3771 == ETH_ADDR_SCAN_COUNT) {
3772 hash = bond_hash_src(mac, vlan);
3774 hash_cstr = xasprintf("%u", hash);
3775 unixctl_command_reply(conn, 200, hash_cstr);
3778 unixctl_command_reply(conn, 501, "invalid mac");
3785 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3786 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3787 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3788 unixctl_command_register("bond/set-active-slave",
3789 bond_unixctl_set_active_slave, NULL);
3790 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3792 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3794 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3797 /* Port functions. */
3800 lacp_send_pdu_cb(void *aux, const struct lacp_pdu *pdu)
3802 struct iface *iface = aux;
3803 uint8_t ea[ETH_ADDR_LEN];
3806 error = netdev_get_etheraddr(iface->netdev, ea);
3808 struct ofpbuf packet;
3809 struct lacp_pdu *packet_pdu;
3811 ofpbuf_init(&packet, 0);
3812 packet_pdu = compose_packet(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
3813 sizeof *packet_pdu);
3814 memcpy(packet_pdu, pdu, sizeof *packet_pdu);
3815 ofproto_send_packet(iface->port->bridge->ofproto,
3816 iface->dp_ifidx, 0, &packet);
3817 ofpbuf_uninit(&packet);
3819 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3820 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3821 "(%s)", iface->name, strerror(error));
3826 port_run(struct port *port)
3828 if (port->monitor) {
3831 /* Track carrier going up and down on interfaces. */
3832 while (!netdev_monitor_poll(port->monitor, &devname)) {
3833 struct iface *iface;
3835 iface = port_lookup_iface(port, devname);
3837 iface_update_carrier(iface);
3841 } else if (time_msec() >= port->miimon_next_update) {
3842 struct iface *iface;
3844 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3845 iface_update_carrier(iface);
3847 port->miimon_next_update = time_msec() + port->miimon_interval;
3851 struct iface *iface;
3853 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3854 lacp_slave_enable(port->lacp, iface, iface->enabled);
3857 lacp_run(port->lacp, lacp_send_pdu_cb);
3864 port_wait(struct port *port)
3866 if (port->monitor) {
3867 netdev_monitor_poll_wait(port->monitor);
3869 poll_timer_wait_until(port->miimon_next_update);
3873 lacp_wait(port->lacp);
3879 static struct port *
3880 port_create(struct bridge *br, const char *name)
3884 port = xzalloc(sizeof *port);
3887 port->trunks = NULL;
3888 port->name = xstrdup(name);
3889 port->active_iface = NULL;
3890 list_init(&port->ifaces);
3892 hmap_insert(&br->ports, &port->hmap_node, hash_string(port->name, 0));
3894 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3901 get_port_other_config(const struct ovsrec_port *port, const char *key,
3902 const char *default_value)
3906 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3908 return value ? value : default_value;
3912 get_interface_other_config(const struct ovsrec_interface *iface,
3913 const char *key, const char *default_value)
3917 value = get_ovsrec_key_value(&iface->header_,
3918 &ovsrec_interface_col_other_config, key);
3919 return value ? value : default_value;
3923 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3925 struct iface *iface, *next;
3926 struct shash new_ifaces;
3929 /* Collect list of new interfaces. */
3930 shash_init(&new_ifaces);
3931 for (i = 0; i < cfg->n_interfaces; i++) {
3932 const char *name = cfg->interfaces[i]->name;
3933 shash_add_once(&new_ifaces, name, NULL);
3936 /* Get rid of deleted interfaces. */
3937 LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) {
3938 if (!shash_find(&new_ifaces, iface->name)) {
3939 iface_destroy(iface);
3943 shash_destroy(&new_ifaces);
3946 /* Expires all MAC learning entries associated with 'port' and forces ofproto
3947 * to revalidate every flow. */
3949 port_flush_macs(struct port *port)
3951 struct bridge *br = port->bridge;
3952 struct mac_learning *ml = br->ml;
3953 struct mac_entry *mac, *next_mac;
3956 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
3957 if (mac->port.p == port) {
3958 mac_learning_expire(ml, mac);
3964 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3966 const char *detect_mode;
3967 struct shash new_ifaces;
3968 long long int next_rebalance, miimon_next_update, lacp_priority;
3969 bool need_flush = false;
3970 unsigned long *trunks;
3976 /* Update settings. */
3977 port->updelay = cfg->bond_updelay;
3978 if (port->updelay < 0) {
3981 port->downdelay = cfg->bond_downdelay;
3982 if (port->downdelay < 0) {
3983 port->downdelay = 0;
3985 port->bond_rebalance_interval = atoi(
3986 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3987 if (port->bond_rebalance_interval < 1000) {
3988 port->bond_rebalance_interval = 1000;
3990 next_rebalance = time_msec() + port->bond_rebalance_interval;
3991 if (port->bond_next_rebalance > next_rebalance) {
3992 port->bond_next_rebalance = next_rebalance;
3995 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
3998 netdev_monitor_destroy(port->monitor);
3999 port->monitor = NULL;
4001 if (strcmp(detect_mode, "miimon")) {
4002 port->monitor = netdev_monitor_create();
4004 if (strcmp(detect_mode, "carrier")) {
4005 VLOG_WARN("port %s: unsupported bond-detect-mode %s, "
4006 "defaulting to carrier", port->name, detect_mode);
4010 port->miimon_interval = atoi(
4011 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4012 if (port->miimon_interval < 100) {
4013 port->miimon_interval = 100;
4015 miimon_next_update = time_msec() + port->miimon_interval;
4016 if (port->miimon_next_update > miimon_next_update) {
4017 port->miimon_next_update = miimon_next_update;
4020 if (!port->cfg->bond_mode ||
4021 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4022 port->bond_mode = BM_SLB;
4023 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4024 port->bond_mode = BM_AB;
4025 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
4026 port->bond_mode = BM_TCP;
4028 port->bond_mode = BM_SLB;
4029 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4030 port->name, port->cfg->bond_mode,
4031 bond_mode_to_string(port->bond_mode));
4034 /* Add new interfaces and update 'cfg' member of existing ones. */
4035 shash_init(&new_ifaces);
4036 for (i = 0; i < cfg->n_interfaces; i++) {
4037 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4038 struct iface *iface;
4040 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
4041 VLOG_WARN("port %s: %s specified twice as port interface",
4042 port->name, if_cfg->name);
4043 iface_set_ofport(if_cfg, -1);
4047 iface = iface_lookup(port->bridge, if_cfg->name);
4049 if (iface->port != port) {
4050 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4052 port->bridge->name, if_cfg->name, iface->port->name);
4055 iface->cfg = if_cfg;
4057 iface = iface_create(port, if_cfg);
4060 /* Determine interface type. The local port always has type
4061 * "internal". Other ports take their type from the database and
4062 * default to "system" if none is specified. */
4063 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4064 : if_cfg->type[0] ? if_cfg->type
4068 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4071 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4072 iface->lacp_priority = UINT16_MAX;
4074 iface->lacp_priority = lacp_priority;
4077 shash_destroy(&new_ifaces);
4079 port->lacp_fast = !strcmp(get_port_other_config(cfg, "lacp-time", "slow"),
4083 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4085 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4086 /* Prefer bondable links if unspecified. */
4087 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4089 port->lacp_priority = lacp_priority;
4092 if (!port->cfg->lacp) {
4093 /* XXX when LACP implementation has been sufficiently tested, enable by
4094 * default and make active on bonded ports. */
4095 lacp_destroy(port->lacp);
4097 } else if (!strcmp(port->cfg->lacp, "off")) {
4098 lacp_destroy(port->lacp);
4100 } else if (!strcmp(port->cfg->lacp, "active")) {
4102 port->lacp = lacp_create();
4104 port->lacp_active = true;
4105 } else if (!strcmp(port->cfg->lacp, "passive")) {
4107 port->lacp = lacp_create();
4109 port->lacp_active = false;
4111 VLOG_WARN("port %s: unknown LACP mode %s",
4112 port->name, port->cfg->lacp);
4113 lacp_destroy(port->lacp);
4120 if (port->n_ifaces < 2) {
4122 if (vlan >= 0 && vlan <= 4095) {
4123 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4128 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4129 * they even work as-is. But they have not been tested. */
4130 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4134 if (port->vlan != vlan) {
4139 /* Get trunked VLANs. */
4141 if (vlan < 0 && cfg->n_trunks) {
4144 trunks = bitmap_allocate(4096);
4146 for (i = 0; i < cfg->n_trunks; i++) {
4147 int trunk = cfg->trunks[i];
4149 bitmap_set1(trunks, trunk);
4155 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4156 port->name, cfg->n_trunks);
4158 if (n_errors == cfg->n_trunks) {
4159 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4161 bitmap_free(trunks);
4164 } else if (vlan >= 0 && cfg->n_trunks) {
4165 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4169 ? port->trunks != NULL
4170 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4173 bitmap_free(port->trunks);
4174 port->trunks = trunks;
4177 port_flush_macs(port);
4182 port_destroy(struct port *port)
4185 struct bridge *br = port->bridge;
4186 struct iface *iface, *next;
4189 for (i = 0; i < MAX_MIRRORS; i++) {
4190 struct mirror *m = br->mirrors[i];
4191 if (m && m->out_port == port) {
4196 LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) {
4197 iface_destroy(iface);
4200 hmap_remove(&br->ports, &port->hmap_node);
4202 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4204 port_flush_macs(port);
4206 lacp_destroy(port->lacp);
4207 netdev_monitor_destroy(port->monitor);
4208 bitmap_free(port->trunks);
4209 free(port->bond_hash);
4215 static struct port *
4216 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4218 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4219 return iface ? iface->port : NULL;
4222 static struct port *
4223 port_lookup(const struct bridge *br, const char *name)
4227 HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_string(name, 0),
4229 if (!strcmp(port->name, name)) {
4236 static struct iface *
4237 port_lookup_iface(const struct port *port, const char *name)
4239 struct iface *iface = iface_lookup(port->bridge, name);
4240 return iface && iface->port == port ? iface : NULL;
4244 port_update_lacp(struct port *port)
4247 struct iface *iface;
4249 lacp_configure(port->lacp, port->name,
4250 port->bridge->ea, port->lacp_priority,
4251 port->lacp_active, port->lacp_fast);
4253 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
4254 lacp_slave_register(port->lacp, iface, iface->name,
4255 iface->dp_ifidx, iface->lacp_priority);
4261 port_update_bonding(struct port *port)
4263 if (port->n_ifaces < 2) {
4264 /* Not a bonded port. */
4265 free(port->bond_hash);
4266 port->bond_hash = NULL;
4267 port->bond_fake_iface = false;
4268 port->active_iface = NULL;
4269 port->no_ifaces_tag = 0;
4273 if (port->bond_mode != BM_AB && !port->bond_hash) {
4274 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4275 for (i = 0; i <= BOND_MASK; i++) {
4276 struct bond_entry *e = &port->bond_hash[i];
4280 port->bond_next_rebalance
4281 = time_msec() + port->bond_rebalance_interval;
4282 } else if (port->bond_mode == BM_AB) {
4283 free(port->bond_hash);
4284 port->bond_hash = NULL;
4287 if (!port->no_ifaces_tag) {
4288 port->no_ifaces_tag = tag_create_random();
4291 if (!port->active_iface) {
4292 bond_choose_active_iface(port);
4295 port->bond_fake_iface = port->cfg->bond_fake_iface;
4296 if (port->bond_fake_iface) {
4297 port->bond_next_fake_iface_update = time_msec();
4303 /* Interface functions. */
4305 static struct iface *
4306 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4308 struct bridge *br = port->bridge;
4309 struct iface *iface;
4310 char *name = if_cfg->name;
4312 iface = xzalloc(sizeof *iface);
4314 iface->name = xstrdup(name);
4315 iface->dp_ifidx = -1;
4316 iface->tag = tag_create_random();
4317 iface->delay_expires = LLONG_MAX;
4318 iface->netdev = NULL;
4319 iface->cfg = if_cfg;
4321 shash_add_assert(&br->iface_by_name, iface->name, iface);
4323 list_push_back(&port->ifaces, &iface->port_elem);
4326 if (port->n_ifaces > 1) {
4327 br->has_bonded_ports = true;
4330 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4338 iface_destroy(struct iface *iface)
4341 struct port *port = iface->port;
4342 struct bridge *br = port->bridge;
4343 bool del_active = port->active_iface == iface;
4345 if (port->bond_hash) {
4346 struct bond_entry *e;
4347 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
4348 if (e->iface == iface) {
4354 if (iface->port->lacp) {
4355 lacp_slave_unregister(iface->port->lacp, iface);
4358 if (port->monitor && iface->netdev) {
4359 netdev_monitor_remove(port->monitor, iface->netdev);
4362 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4364 if (iface->dp_ifidx >= 0) {
4365 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4368 list_remove(&iface->port_elem);
4371 netdev_close(iface->netdev);
4374 bond_choose_active_iface(port);
4375 bond_send_learning_packets(port);
4381 bridge_flush(port->bridge);
4385 static struct iface *
4386 iface_lookup(const struct bridge *br, const char *name)
4388 return shash_find_data(&br->iface_by_name, name);
4391 static struct iface *
4392 iface_find(const char *name)
4394 const struct bridge *br;
4396 LIST_FOR_EACH (br, node, &all_bridges) {
4397 struct iface *iface = iface_lookup(br, name);
4406 static struct iface *
4407 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4409 struct iface *iface;
4411 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4412 hash_int(dp_ifidx, 0), &br->ifaces) {
4413 if (iface->dp_ifidx == dp_ifidx) {
4420 /* Set Ethernet address of 'iface', if one is specified in the configuration
4423 iface_set_mac(struct iface *iface)
4425 uint8_t ea[ETH_ADDR_LEN];
4427 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4428 if (eth_addr_is_multicast(ea)) {
4429 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4431 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4432 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4433 iface->name, iface->name);
4435 int error = netdev_set_etheraddr(iface->netdev, ea);
4437 VLOG_ERR("interface %s: setting MAC failed (%s)",
4438 iface->name, strerror(error));
4444 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4446 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4449 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4453 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4455 * The value strings in '*shash' are taken directly from values[], not copied,
4456 * so the caller should not modify or free them. */
4458 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4459 struct shash *shash)
4464 for (i = 0; i < n; i++) {
4465 shash_add(shash, keys[i], values[i]);
4469 /* Creates 'keys' and 'values' arrays from 'shash'.
4471 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4472 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4473 * are populated with with strings taken directly from 'shash' and thus have
4474 * the same ownership of the key-value pairs in shash.
4477 shash_to_ovs_idl_map(struct shash *shash,
4478 char ***keys, char ***values, size_t *n)
4482 struct shash_node *sn;
4484 count = shash_count(shash);
4486 k = xmalloc(count * sizeof *k);
4487 v = xmalloc(count * sizeof *v);
4490 SHASH_FOR_EACH(sn, shash) {
4501 struct iface_delete_queues_cbdata {
4502 struct netdev *netdev;
4503 const struct ovsdb_datum *queues;
4507 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4509 union ovsdb_atom atom;
4511 atom.integer = target;
4512 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4516 iface_delete_queues(unsigned int queue_id,
4517 const struct shash *details OVS_UNUSED, void *cbdata_)
4519 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4521 if (!queue_ids_include(cbdata->queues, queue_id)) {
4522 netdev_delete_queue(cbdata->netdev, queue_id);
4527 iface_update_carrier(struct iface *iface)
4529 bool carrier = iface_get_carrier(iface);
4530 if (carrier == iface->up) {
4534 iface->up = carrier;
4535 if (iface->port->lacp) {
4536 lacp_slave_carrier_changed(iface->port->lacp, iface);
4541 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4543 if (!qos || qos->type[0] == '\0') {
4544 netdev_set_qos(iface->netdev, NULL, NULL);
4546 struct iface_delete_queues_cbdata cbdata;
4547 struct shash details;
4550 /* Configure top-level Qos for 'iface'. */
4551 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4552 qos->n_other_config, &details);
4553 netdev_set_qos(iface->netdev, qos->type, &details);
4554 shash_destroy(&details);
4556 /* Deconfigure queues that were deleted. */
4557 cbdata.netdev = iface->netdev;
4558 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4560 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4562 /* Configure queues for 'iface'. */
4563 for (i = 0; i < qos->n_queues; i++) {
4564 const struct ovsrec_queue *queue = qos->value_queues[i];
4565 unsigned int queue_id = qos->key_queues[i];
4567 shash_from_ovs_idl_map(queue->key_other_config,
4568 queue->value_other_config,
4569 queue->n_other_config, &details);
4570 netdev_set_queue(iface->netdev, queue_id, &details);
4571 shash_destroy(&details);
4577 iface_update_cfm(struct iface *iface)
4581 uint16_t *remote_mps;
4582 struct ovsrec_monitor *mon;
4583 uint8_t maid[CCM_MAID_LEN];
4585 mon = iface->cfg->monitor;
4588 ofproto_iface_clear_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
4592 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4593 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4597 cfm.mpid = mon->mpid;
4598 cfm.interval = mon->interval ? *mon->interval : 1000;
4600 memcpy(cfm.maid, maid, sizeof cfm.maid);
4602 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4603 for(i = 0; i < mon->n_remote_mps; i++) {
4604 remote_mps[i] = mon->remote_mps[i]->mpid;
4607 ofproto_iface_set_cfm(iface->port->bridge->ofproto, iface->dp_ifidx,
4608 &cfm, remote_mps, mon->n_remote_mps);
4612 /* Read carrier or miimon status directly from 'iface''s netdev, according to
4613 * how 'iface''s port is configured.
4615 * Returns true if 'iface' is up, false otherwise. */
4617 iface_get_carrier(const struct iface *iface)
4619 return (iface->port->monitor
4620 ? netdev_get_carrier(iface->netdev)
4621 : netdev_get_miimon(iface->netdev));
4624 /* Port mirroring. */
4626 static struct mirror *
4627 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4631 for (i = 0; i < MAX_MIRRORS; i++) {
4632 struct mirror *m = br->mirrors[i];
4633 if (m && uuid_equals(uuid, &m->uuid)) {
4641 mirror_reconfigure(struct bridge *br)
4643 unsigned long *rspan_vlans;
4647 /* Get rid of deleted mirrors. */
4648 for (i = 0; i < MAX_MIRRORS; i++) {
4649 struct mirror *m = br->mirrors[i];
4651 const struct ovsdb_datum *mc;
4652 union ovsdb_atom atom;
4654 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4655 atom.uuid = br->mirrors[i]->uuid;
4656 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4662 /* Add new mirrors and reconfigure existing ones. */
4663 for (i = 0; i < br->cfg->n_mirrors; i++) {
4664 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4665 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4667 mirror_reconfigure_one(m, cfg);
4669 mirror_create(br, cfg);
4673 /* Update port reserved status. */
4674 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
4675 port->is_mirror_output_port = false;
4677 for (i = 0; i < MAX_MIRRORS; i++) {
4678 struct mirror *m = br->mirrors[i];
4679 if (m && m->out_port) {
4680 m->out_port->is_mirror_output_port = true;
4684 /* Update flooded vlans (for RSPAN). */
4686 if (br->cfg->n_flood_vlans) {
4687 rspan_vlans = bitmap_allocate(4096);
4689 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4690 int64_t vlan = br->cfg->flood_vlans[i];
4691 if (vlan >= 0 && vlan < 4096) {
4692 bitmap_set1(rspan_vlans, vlan);
4693 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4696 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4701 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4703 mac_learning_flush(br->ml);
4708 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4713 for (i = 0; ; i++) {
4714 if (i >= MAX_MIRRORS) {
4715 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4716 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4719 if (!br->mirrors[i]) {
4724 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4726 mac_learning_flush(br->ml);
4728 br->mirrors[i] = m = xzalloc(sizeof *m);
4731 m->name = xstrdup(cfg->name);
4732 shash_init(&m->src_ports);
4733 shash_init(&m->dst_ports);
4739 mirror_reconfigure_one(m, cfg);
4743 mirror_destroy(struct mirror *m)
4746 struct bridge *br = m->bridge;
4749 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
4750 port->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4751 port->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4754 shash_destroy(&m->src_ports);
4755 shash_destroy(&m->dst_ports);
4758 m->bridge->mirrors[m->idx] = NULL;
4763 mac_learning_flush(br->ml);
4768 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4769 struct shash *names)
4773 for (i = 0; i < n_ports; i++) {
4774 const char *name = ports[i]->name;
4775 if (port_lookup(m->bridge, name)) {
4776 shash_add_once(names, name, NULL);
4778 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4779 "port %s", m->bridge->name, m->name, name);
4785 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4791 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4793 for (i = 0; i < cfg->n_select_vlan; i++) {
4794 int64_t vlan = cfg->select_vlan[i];
4795 if (vlan < 0 || vlan > 4095) {
4796 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4797 m->bridge->name, m->name, vlan);
4799 (*vlans)[n_vlans++] = vlan;
4806 vlan_is_mirrored(const struct mirror *m, int vlan)
4810 for (i = 0; i < m->n_vlans; i++) {
4811 if (m->vlans[i] == vlan) {
4819 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4823 for (i = 0; i < m->n_vlans; i++) {
4824 if (port_trunks_vlan(p, m->vlans[i])) {
4832 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4834 struct shash src_ports, dst_ports;
4835 mirror_mask_t mirror_bit;
4836 struct port *out_port;
4843 if (strcmp(cfg->name, m->name)) {
4845 m->name = xstrdup(cfg->name);
4848 /* Get output port. */
4849 if (cfg->output_port) {
4850 out_port = port_lookup(m->bridge, cfg->output_port->name);
4852 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4853 m->bridge->name, m->name);
4859 if (cfg->output_vlan) {
4860 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4861 "output vlan; ignoring output vlan",
4862 m->bridge->name, m->name);
4864 } else if (cfg->output_vlan) {
4866 out_vlan = *cfg->output_vlan;
4868 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4869 m->bridge->name, m->name);
4874 shash_init(&src_ports);
4875 shash_init(&dst_ports);
4876 if (cfg->select_all) {
4877 HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) {
4878 shash_add_once(&src_ports, port->name, NULL);
4879 shash_add_once(&dst_ports, port->name, NULL);
4884 /* Get ports, and drop duplicates and ports that don't exist. */
4885 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4887 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4890 /* Get all the vlans, and drop duplicate and invalid vlans. */
4891 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4894 /* Update mirror data. */
4895 if (!shash_equal_keys(&m->src_ports, &src_ports)
4896 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4897 || m->n_vlans != n_vlans
4898 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4899 || m->out_port != out_port
4900 || m->out_vlan != out_vlan) {
4901 bridge_flush(m->bridge);
4902 mac_learning_flush(m->bridge->ml);
4904 shash_swap(&m->src_ports, &src_ports);
4905 shash_swap(&m->dst_ports, &dst_ports);
4908 m->n_vlans = n_vlans;
4909 m->out_port = out_port;
4910 m->out_vlan = out_vlan;
4913 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4914 HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) {
4915 if (shash_find(&m->src_ports, port->name)
4918 ? port_trunks_any_mirrored_vlan(m, port)
4919 : vlan_is_mirrored(m, port->vlan)))) {
4920 port->src_mirrors |= mirror_bit;
4922 port->src_mirrors &= ~mirror_bit;
4925 if (shash_find(&m->dst_ports, port->name)) {
4926 port->dst_mirrors |= mirror_bit;
4928 port->dst_mirrors &= ~mirror_bit;
4933 shash_destroy(&src_ports);
4934 shash_destroy(&dst_ports);