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 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
242 static void bridge_destroy(struct bridge *);
243 static struct bridge *bridge_lookup(const char *name);
244 static unixctl_cb_func bridge_unixctl_dump_flows;
245 static unixctl_cb_func bridge_unixctl_reconnect;
246 static int bridge_run_one(struct bridge *);
247 static size_t bridge_get_controllers(const struct bridge *br,
248 struct ovsrec_controller ***controllersp);
249 static void bridge_reconfigure_one(struct bridge *);
250 static void bridge_reconfigure_remotes(struct bridge *,
251 const struct sockaddr_in *managers,
253 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
254 static void bridge_fetch_dp_ifaces(struct bridge *);
255 static void bridge_flush(struct bridge *);
256 static void bridge_pick_local_hw_addr(struct bridge *,
257 uint8_t ea[ETH_ADDR_LEN],
258 struct iface **hw_addr_iface);
259 static uint64_t bridge_pick_datapath_id(struct bridge *,
260 const uint8_t bridge_ea[ETH_ADDR_LEN],
261 struct iface *hw_addr_iface);
262 static uint64_t dpid_from_hash(const void *, size_t nbytes);
264 static unixctl_cb_func bridge_unixctl_fdb_show;
265 static unixctl_cb_func qos_unixctl_show;
267 static void bond_init(void);
268 static void bond_run(struct port *);
269 static void bond_wait(struct port *);
270 static void bond_rebalance_port(struct port *);
271 static void bond_send_learning_packets(struct port *);
272 static void bond_enable_slave(struct iface *iface, bool enable);
274 static void port_run(struct port *);
275 static void port_wait(struct port *);
276 static struct port *port_create(struct bridge *, const char *name);
277 static void port_reconfigure(struct port *, const struct ovsrec_port *);
278 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
279 static void port_destroy(struct port *);
280 static struct port *port_lookup(const struct bridge *, const char *name);
281 static struct iface *port_lookup_iface(const struct port *, const char *name);
282 static struct iface *port_get_an_iface(const struct port *);
283 static struct port *port_from_dp_ifidx(const struct bridge *,
285 static void port_update_bonding(struct port *);
286 static void port_update_lacp(struct port *);
288 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
289 static void mirror_destroy(struct mirror *);
290 static void mirror_reconfigure(struct bridge *);
291 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
292 static bool vlan_is_mirrored(const struct mirror *, int vlan);
294 static struct iface *iface_create(struct port *port,
295 const struct ovsrec_interface *if_cfg);
296 static void iface_destroy(struct iface *);
297 static struct iface *iface_lookup(const struct bridge *, const char *name);
298 static struct iface *iface_find(const char *name);
299 static struct iface *iface_from_dp_ifidx(const struct bridge *,
301 static void iface_set_mac(struct iface *);
302 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
303 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
304 static void iface_update_cfm(struct iface *);
305 static void iface_refresh_cfm_stats(struct iface *iface);
306 static void iface_update_carrier(struct iface *);
307 static bool iface_get_carrier(const struct iface *);
309 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
311 static void shash_to_ovs_idl_map(struct shash *,
312 char ***keys, char ***values, size_t *n);
314 /* Hooks into ofproto processing. */
315 static struct ofhooks bridge_ofhooks;
317 /* Public functions. */
319 /* Initializes the bridge module, configuring it to obtain its configuration
320 * from an OVSDB server accessed over 'remote', which should be a string in a
321 * form acceptable to ovsdb_idl_create(). */
323 bridge_init(const char *remote)
325 /* Create connection to database. */
326 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
328 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
329 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
330 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
332 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
334 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
335 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
337 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
338 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
339 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
341 /* Register unixctl commands. */
342 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
343 unixctl_command_register("qos/show", qos_unixctl_show, NULL);
344 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
346 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
355 struct bridge *br, *next_br;
357 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
360 ovsdb_idl_destroy(idl);
363 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
364 * but for which the ovs-vswitchd configuration 'cfg' is required. */
366 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
368 static bool already_configured_once;
369 struct svec bridge_names;
370 struct svec dpif_names, dpif_types;
373 /* Only do this once per ovs-vswitchd run. */
374 if (already_configured_once) {
377 already_configured_once = true;
379 stats_timer = time_msec() + STATS_INTERVAL;
381 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
382 svec_init(&bridge_names);
383 for (i = 0; i < cfg->n_bridges; i++) {
384 svec_add(&bridge_names, cfg->bridges[i]->name);
386 svec_sort(&bridge_names);
388 /* Iterate over all system dpifs and delete any of them that do not appear
390 svec_init(&dpif_names);
391 svec_init(&dpif_types);
392 dp_enumerate_types(&dpif_types);
393 for (i = 0; i < dpif_types.n; i++) {
396 dp_enumerate_names(dpif_types.names[i], &dpif_names);
398 /* Delete each dpif whose name is not in 'bridge_names'. */
399 for (j = 0; j < dpif_names.n; j++) {
400 if (!svec_contains(&bridge_names, dpif_names.names[j])) {
404 retval = dpif_open(dpif_names.names[j], dpif_types.names[i],
413 svec_destroy(&bridge_names);
414 svec_destroy(&dpif_names);
415 svec_destroy(&dpif_types);
418 /* Callback for iterate_and_prune_ifaces(). */
420 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
422 if (!iface->netdev) {
423 /* We already reported a related error, don't bother duplicating it. */
427 if (iface->dp_ifidx < 0) {
428 VLOG_ERR("%s interface not in %s, dropping",
429 iface->name, dpif_name(br->dpif));
433 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
434 iface->name, iface->dp_ifidx);
438 /* Callback for iterate_and_prune_ifaces(). */
440 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
441 void *aux OVS_UNUSED)
443 /* Set policing attributes. */
444 netdev_set_policing(iface->netdev,
445 iface->cfg->ingress_policing_rate,
446 iface->cfg->ingress_policing_burst);
448 /* Set MAC address of internal interfaces other than the local
450 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
451 iface_set_mac(iface);
457 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
458 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
459 * deletes from 'br' any ports that no longer have any interfaces. */
461 iterate_and_prune_ifaces(struct bridge *br,
462 bool (*cb)(struct bridge *, struct iface *,
466 struct port *port, *next_port;
468 HMAP_FOR_EACH_SAFE (port, next_port, hmap_node, &br->ports) {
469 struct iface *iface, *next_iface;
471 LIST_FOR_EACH_SAFE (iface, next_iface, port_elem, &port->ifaces) {
472 if (!cb(br, iface, aux)) {
473 iface_set_ofport(iface->cfg, -1);
474 iface_destroy(iface);
478 if (!port->n_ifaces) {
479 VLOG_WARN("%s port has no interfaces, dropping", port->name);
485 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
486 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
487 * responsible for freeing '*managersp' (with free()).
489 * You may be asking yourself "why does ovs-vswitchd care?", because
490 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
491 * should not be and in fact is not directly involved in that. But
492 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
493 * it has to tell in-band control where the managers are to enable that.
494 * (Thus, only managers connected in-band are collected.)
497 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
498 struct sockaddr_in **managersp, size_t *n_managersp)
500 struct sockaddr_in *managers = NULL;
501 size_t n_managers = 0;
502 struct shash targets;
505 /* Collect all of the potential targets from the "targets" columns of the
506 * rows pointed to by "manager_options", excluding any that are
508 shash_init(&targets);
509 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
510 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
512 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
513 shash_find_and_delete(&targets, m->target);
515 shash_add_once(&targets, m->target, NULL);
519 /* Now extract the targets' IP addresses. */
520 if (!shash_is_empty(&targets)) {
521 struct shash_node *node;
523 managers = xmalloc(shash_count(&targets) * sizeof *managers);
524 SHASH_FOR_EACH (node, &targets) {
525 const char *target = node->name;
526 struct sockaddr_in *sin = &managers[n_managers];
528 if ((!strncmp(target, "tcp:", 4)
529 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
530 (!strncmp(target, "ssl:", 4)
531 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
536 shash_destroy(&targets);
538 *managersp = managers;
539 *n_managersp = n_managers;
543 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
545 struct shash old_br, new_br;
546 struct shash_node *node;
547 struct bridge *br, *next;
548 struct sockaddr_in *managers;
551 int sflow_bridge_number;
553 COVERAGE_INC(bridge_reconfigure);
555 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
557 /* Collect old and new bridges. */
560 LIST_FOR_EACH (br, node, &all_bridges) {
561 shash_add(&old_br, br->name, br);
563 for (i = 0; i < ovs_cfg->n_bridges; i++) {
564 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
565 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
566 VLOG_WARN("more than one bridge named %s", br_cfg->name);
570 /* Get rid of deleted bridges and add new bridges. */
571 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
572 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
579 SHASH_FOR_EACH (node, &new_br) {
580 const char *br_name = node->name;
581 const struct ovsrec_bridge *br_cfg = node->data;
582 br = shash_find_data(&old_br, br_name);
584 /* If the bridge datapath type has changed, we need to tear it
585 * down and recreate. */
586 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
588 bridge_create(br_cfg);
591 bridge_create(br_cfg);
594 shash_destroy(&old_br);
595 shash_destroy(&new_br);
597 /* Reconfigure all bridges. */
598 LIST_FOR_EACH (br, node, &all_bridges) {
599 bridge_reconfigure_one(br);
602 /* Add and delete ports on all datapaths.
604 * The kernel will reject any attempt to add a given port to a datapath if
605 * that port already belongs to a different datapath, so we must do all
606 * port deletions before any port additions. */
607 LIST_FOR_EACH (br, node, &all_bridges) {
608 struct dpif_port_dump dump;
609 struct shash want_ifaces;
610 struct dpif_port dpif_port;
612 bridge_get_all_ifaces(br, &want_ifaces);
613 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
614 if (!shash_find(&want_ifaces, dpif_port.name)
615 && strcmp(dpif_port.name, br->name)) {
616 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
618 VLOG_WARN("failed to remove %s interface from %s: %s",
619 dpif_port.name, dpif_name(br->dpif),
624 shash_destroy(&want_ifaces);
626 LIST_FOR_EACH (br, node, &all_bridges) {
627 struct shash cur_ifaces, want_ifaces;
628 struct dpif_port_dump dump;
629 struct dpif_port dpif_port;
631 /* Get the set of interfaces currently in this datapath. */
632 shash_init(&cur_ifaces);
633 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
634 struct dpif_port *port_info = xmalloc(sizeof *port_info);
635 dpif_port_clone(port_info, &dpif_port);
636 shash_add(&cur_ifaces, dpif_port.name, port_info);
639 /* Get the set of interfaces we want on this datapath. */
640 bridge_get_all_ifaces(br, &want_ifaces);
642 hmap_clear(&br->ifaces);
643 SHASH_FOR_EACH (node, &want_ifaces) {
644 const char *if_name = node->name;
645 struct iface *iface = node->data;
646 struct dpif_port *dpif_port;
650 type = iface ? iface->type : "internal";
651 dpif_port = shash_find_data(&cur_ifaces, if_name);
653 /* If we have a port or a netdev already, and it's not the type we
654 * want, then delete the port (if any) and close the netdev (if
656 if ((dpif_port && strcmp(dpif_port->type, type))
657 || (iface && iface->netdev
658 && strcmp(type, netdev_get_type(iface->netdev)))) {
660 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
667 netdev_close(iface->netdev);
668 iface->netdev = NULL;
672 /* If the port doesn't exist or we don't have the netdev open,
673 * we need to do more work. */
674 if (!dpif_port || (iface && !iface->netdev)) {
675 struct netdev_options options;
676 struct netdev *netdev;
679 /* First open the network device. */
680 options.name = if_name;
682 options.args = &args;
683 options.ethertype = NETDEV_ETH_TYPE_NONE;
687 shash_from_ovs_idl_map(iface->cfg->key_options,
688 iface->cfg->value_options,
689 iface->cfg->n_options, &args);
691 error = netdev_open(&options, &netdev);
692 shash_destroy(&args);
695 VLOG_WARN("could not open network device %s (%s)",
696 if_name, strerror(error));
700 /* Then add the port if we haven't already. */
702 error = dpif_port_add(br->dpif, netdev, NULL);
704 netdev_close(netdev);
705 if (error == EFBIG) {
706 VLOG_ERR("ran out of valid port numbers on %s",
707 dpif_name(br->dpif));
710 VLOG_WARN("failed to add %s interface to %s: %s",
711 if_name, dpif_name(br->dpif),
718 /* Update 'iface'. */
720 iface->netdev = netdev;
721 iface->enabled = iface_get_carrier(iface);
722 iface->up = iface->enabled;
724 } else if (iface && iface->netdev) {
728 shash_from_ovs_idl_map(iface->cfg->key_options,
729 iface->cfg->value_options,
730 iface->cfg->n_options, &args);
731 netdev_set_config(iface->netdev, &args);
732 shash_destroy(&args);
735 shash_destroy(&want_ifaces);
737 SHASH_FOR_EACH (node, &cur_ifaces) {
738 struct dpif_port *port_info = node->data;
739 dpif_port_destroy(port_info);
742 shash_destroy(&cur_ifaces);
744 sflow_bridge_number = 0;
745 LIST_FOR_EACH (br, node, &all_bridges) {
748 struct iface *local_iface;
749 struct iface *hw_addr_iface;
752 bridge_fetch_dp_ifaces(br);
754 iterate_and_prune_ifaces(br, check_iface, NULL);
756 /* Pick local port hardware address, datapath ID. */
757 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
758 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
760 int error = netdev_set_etheraddr(local_iface->netdev, ea);
762 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
763 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
764 "Ethernet address: %s",
765 br->name, strerror(error));
768 memcpy(br->ea, ea, ETH_ADDR_LEN);
770 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
771 ofproto_set_datapath_id(br->ofproto, dpid);
773 dpid_string = xasprintf("%016"PRIx64, dpid);
774 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
777 /* Set NetFlow configuration on this bridge. */
778 if (br->cfg->netflow) {
779 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
780 struct netflow_options opts;
782 memset(&opts, 0, sizeof opts);
784 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
785 if (nf_cfg->engine_type) {
786 opts.engine_type = *nf_cfg->engine_type;
788 if (nf_cfg->engine_id) {
789 opts.engine_id = *nf_cfg->engine_id;
792 opts.active_timeout = nf_cfg->active_timeout;
793 if (!opts.active_timeout) {
794 opts.active_timeout = -1;
795 } else if (opts.active_timeout < 0) {
796 VLOG_WARN("bridge %s: active timeout interval set to negative "
797 "value, using default instead (%d seconds)", br->name,
798 NF_ACTIVE_TIMEOUT_DEFAULT);
799 opts.active_timeout = -1;
802 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
803 if (opts.add_id_to_iface) {
804 if (opts.engine_id > 0x7f) {
805 VLOG_WARN("bridge %s: netflow port mangling may conflict "
806 "with another vswitch, choose an engine id less "
807 "than 128", br->name);
809 if (hmap_count(&br->ports) > 508) {
810 VLOG_WARN("bridge %s: netflow port mangling will conflict "
811 "with another port when more than 508 ports are "
816 opts.collectors.n = nf_cfg->n_targets;
817 opts.collectors.names = nf_cfg->targets;
818 if (ofproto_set_netflow(br->ofproto, &opts)) {
819 VLOG_ERR("bridge %s: problem setting netflow collectors",
823 ofproto_set_netflow(br->ofproto, NULL);
826 /* Set sFlow configuration on this bridge. */
827 if (br->cfg->sflow) {
828 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
829 struct ovsrec_controller **controllers;
830 struct ofproto_sflow_options oso;
831 size_t n_controllers;
833 memset(&oso, 0, sizeof oso);
835 oso.targets.n = sflow_cfg->n_targets;
836 oso.targets.names = sflow_cfg->targets;
838 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
839 if (sflow_cfg->sampling) {
840 oso.sampling_rate = *sflow_cfg->sampling;
843 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
844 if (sflow_cfg->polling) {
845 oso.polling_interval = *sflow_cfg->polling;
848 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
849 if (sflow_cfg->header) {
850 oso.header_len = *sflow_cfg->header;
853 oso.sub_id = sflow_bridge_number++;
854 oso.agent_device = sflow_cfg->agent;
856 oso.control_ip = NULL;
857 n_controllers = bridge_get_controllers(br, &controllers);
858 for (i = 0; i < n_controllers; i++) {
859 if (controllers[i]->local_ip) {
860 oso.control_ip = controllers[i]->local_ip;
864 ofproto_set_sflow(br->ofproto, &oso);
866 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
868 ofproto_set_sflow(br->ofproto, NULL);
871 /* Update the controller and related settings. It would be more
872 * straightforward to call this from bridge_reconfigure_one(), but we
873 * can't do it there for two reasons. First, and most importantly, at
874 * that point we don't know the dp_ifidx of any interfaces that have
875 * been added to the bridge (because we haven't actually added them to
876 * the datapath). Second, at that point we haven't set the datapath ID
877 * yet; when a controller is configured, resetting the datapath ID will
878 * immediately disconnect from the controller, so it's better to set
879 * the datapath ID before the controller. */
880 bridge_reconfigure_remotes(br, managers, n_managers);
882 LIST_FOR_EACH (br, node, &all_bridges) {
885 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
889 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
890 netdev_monitor_add(port->monitor, iface->netdev);
893 port->miimon_next_update = 0;
896 port_update_lacp(port);
897 port_update_bonding(port);
899 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
900 iface_update_qos(iface, port->cfg->qos);
904 LIST_FOR_EACH (br, node, &all_bridges) {
905 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
908 LIST_FOR_EACH (br, node, &all_bridges) {
910 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
911 iface_update_cfm(iface);
917 /* ovs-vswitchd has completed initialization, so allow the process that
918 * forked us to exit successfully. */
919 daemonize_complete();
923 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
924 const struct ovsdb_idl_column *column,
927 const struct ovsdb_datum *datum;
928 union ovsdb_atom atom;
931 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
932 atom.string = (char *) key;
933 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
934 return idx == UINT_MAX ? NULL : datum->values[idx].string;
938 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
940 return get_ovsrec_key_value(&br_cfg->header_,
941 &ovsrec_bridge_col_other_config, key);
945 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
946 struct iface **hw_addr_iface)
952 *hw_addr_iface = NULL;
954 /* Did the user request a particular MAC? */
955 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
956 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
957 if (eth_addr_is_multicast(ea)) {
958 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
959 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
960 } else if (eth_addr_is_zero(ea)) {
961 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
967 /* Otherwise choose the minimum non-local MAC address among all of the
969 memset(ea, 0xff, ETH_ADDR_LEN);
970 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
971 uint8_t iface_ea[ETH_ADDR_LEN];
972 struct iface *candidate;
975 /* Mirror output ports don't participate. */
976 if (port->is_mirror_output_port) {
980 /* Choose the MAC address to represent the port. */
982 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
983 /* Find the interface with this Ethernet address (if any) so that
984 * we can provide the correct devname to the caller. */
985 LIST_FOR_EACH (candidate, port_elem, &port->ifaces) {
986 uint8_t candidate_ea[ETH_ADDR_LEN];
987 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
988 && eth_addr_equals(iface_ea, candidate_ea)) {
993 /* Choose the interface whose MAC address will represent the port.
994 * The Linux kernel bonding code always chooses the MAC address of
995 * the first slave added to a bond, and the Fedora networking
996 * scripts always add slaves to a bond in alphabetical order, so
997 * for compatibility we choose the interface with the name that is
998 * first in alphabetical order. */
999 LIST_FOR_EACH (candidate, port_elem, &port->ifaces) {
1000 if (!iface || strcmp(candidate->name, iface->name) < 0) {
1005 /* The local port doesn't count (since we're trying to choose its
1006 * MAC address anyway). */
1007 if (iface->dp_ifidx == ODPP_LOCAL) {
1012 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1014 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1015 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1016 iface->name, strerror(error));
1021 /* Compare against our current choice. */
1022 if (!eth_addr_is_multicast(iface_ea) &&
1023 !eth_addr_is_local(iface_ea) &&
1024 !eth_addr_is_reserved(iface_ea) &&
1025 !eth_addr_is_zero(iface_ea) &&
1026 eth_addr_compare_3way(iface_ea, ea) < 0)
1028 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1029 *hw_addr_iface = iface;
1032 if (eth_addr_is_multicast(ea)) {
1033 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1034 *hw_addr_iface = NULL;
1035 VLOG_WARN("bridge %s: using default bridge Ethernet "
1036 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1038 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1039 br->name, ETH_ADDR_ARGS(ea));
1043 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1044 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1045 * an interface on 'br', then that interface must be passed in as
1046 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1047 * 'hw_addr_iface' must be passed in as a null pointer. */
1049 bridge_pick_datapath_id(struct bridge *br,
1050 const uint8_t bridge_ea[ETH_ADDR_LEN],
1051 struct iface *hw_addr_iface)
1054 * The procedure for choosing a bridge MAC address will, in the most
1055 * ordinary case, also choose a unique MAC that we can use as a datapath
1056 * ID. In some special cases, though, multiple bridges will end up with
1057 * the same MAC address. This is OK for the bridges, but it will confuse
1058 * the OpenFlow controller, because each datapath needs a unique datapath
1061 * Datapath IDs must be unique. It is also very desirable that they be
1062 * stable from one run to the next, so that policy set on a datapath
1065 const char *datapath_id;
1068 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1069 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1073 if (hw_addr_iface) {
1075 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1077 * A bridge whose MAC address is taken from a VLAN network device
1078 * (that is, a network device created with vconfig(8) or similar
1079 * tool) will have the same MAC address as a bridge on the VLAN
1080 * device's physical network device.
1082 * Handle this case by hashing the physical network device MAC
1083 * along with the VLAN identifier.
1085 uint8_t buf[ETH_ADDR_LEN + 2];
1086 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1087 buf[ETH_ADDR_LEN] = vlan >> 8;
1088 buf[ETH_ADDR_LEN + 1] = vlan;
1089 return dpid_from_hash(buf, sizeof buf);
1092 * Assume that this bridge's MAC address is unique, since it
1093 * doesn't fit any of the cases we handle specially.
1098 * A purely internal bridge, that is, one that has no non-virtual
1099 * network devices on it at all, is more difficult because it has no
1100 * natural unique identifier at all.
1102 * When the host is a XenServer, we handle this case by hashing the
1103 * host's UUID with the name of the bridge. Names of bridges are
1104 * persistent across XenServer reboots, although they can be reused if
1105 * an internal network is destroyed and then a new one is later
1106 * created, so this is fairly effective.
1108 * When the host is not a XenServer, we punt by using a random MAC
1109 * address on each run.
1111 const char *host_uuid = xenserver_get_host_uuid();
1113 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1114 dpid = dpid_from_hash(combined, strlen(combined));
1120 return eth_addr_to_uint64(bridge_ea);
1124 dpid_from_hash(const void *data, size_t n)
1126 uint8_t hash[SHA1_DIGEST_SIZE];
1128 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1129 sha1_bytes(data, n, hash);
1130 eth_addr_mark_random(hash);
1131 return eth_addr_to_uint64(hash);
1135 iface_refresh_status(struct iface *iface)
1139 enum netdev_flags flags;
1148 if (!netdev_get_status(iface->netdev, &sh)) {
1150 char **keys, **values;
1152 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1153 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1158 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1161 shash_destroy_free_data(&sh);
1163 error = netdev_get_flags(iface->netdev, &flags);
1165 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1168 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1171 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1173 ovsrec_interface_set_duplex(iface->cfg,
1174 netdev_features_is_full_duplex(current)
1176 /* warning: uint64_t -> int64_t conversion */
1177 bps = netdev_features_to_bps(current);
1178 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1181 ovsrec_interface_set_duplex(iface->cfg, NULL);
1182 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1186 ovsrec_interface_set_link_state(iface->cfg,
1187 iface_get_carrier(iface) ? "up" : "down");
1189 error = netdev_get_mtu(iface->netdev, &mtu);
1190 if (!error && mtu != INT_MAX) {
1192 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1195 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1200 iface_refresh_cfm_stats(struct iface *iface)
1202 const struct ovsrec_monitor *mon;
1203 const struct cfm *cfm;
1206 mon = iface->cfg->monitor;
1207 cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
1213 for (i = 0; i < mon->n_remote_mps; i++) {
1214 const struct ovsrec_maintenance_point *mp;
1215 const struct remote_mp *rmp;
1217 mp = mon->remote_mps[i];
1218 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1220 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1223 if (hmap_is_empty(&cfm->x_remote_mps)) {
1224 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1227 struct remote_mp *rmp;
1228 int64_t *x_remote_mps;
1230 length = hmap_count(&cfm->x_remote_mps);
1231 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1234 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1235 x_remote_mps[i++] = rmp->mpid;
1238 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1242 if (hmap_is_empty(&cfm->x_remote_maids)) {
1243 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1246 char **x_remote_maids;
1247 struct remote_maid *rmaid;
1249 length = hmap_count(&cfm->x_remote_maids);
1250 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1253 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1256 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1258 for (j = 0; j < CCM_MAID_LEN; j++) {
1259 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1264 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1266 for (i = 0; i < length; i++) {
1267 free(x_remote_maids[i]);
1269 free(x_remote_maids);
1272 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1276 iface_refresh_stats(struct iface *iface)
1282 static const struct iface_stat iface_stats[] = {
1283 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1284 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1285 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1286 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1287 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1288 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1289 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1290 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1291 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1292 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1293 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1294 { "collisions", offsetof(struct netdev_stats, collisions) },
1296 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1297 const struct iface_stat *s;
1299 char *keys[N_STATS];
1300 int64_t values[N_STATS];
1303 struct netdev_stats stats;
1305 /* Intentionally ignore return value, since errors will set 'stats' to
1306 * all-1s, and we will deal with that correctly below. */
1307 netdev_get_stats(iface->netdev, &stats);
1310 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1311 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1312 if (value != UINT64_MAX) {
1319 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1323 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1325 struct ovsdb_datum datum;
1329 get_system_stats(&stats);
1331 ovsdb_datum_from_shash(&datum, &stats);
1332 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1336 static inline const char *
1337 nx_role_to_str(enum nx_role role)
1342 case NX_ROLE_MASTER:
1347 return "*** INVALID ROLE ***";
1352 bridge_refresh_controller_status(const struct bridge *br)
1355 const struct ovsrec_controller *cfg;
1357 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1359 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1360 struct ofproto_controller_info *cinfo =
1361 shash_find_data(&info, cfg->target);
1364 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1365 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1366 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1367 (char **) cinfo->pairs.values,
1370 ovsrec_controller_set_is_connected(cfg, false);
1371 ovsrec_controller_set_role(cfg, NULL);
1372 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1376 ofproto_free_ofproto_controller_info(&info);
1382 const struct ovsrec_open_vswitch *cfg;
1384 bool datapath_destroyed;
1385 bool database_changed;
1388 /* Let each bridge do the work that it needs to do. */
1389 datapath_destroyed = false;
1390 LIST_FOR_EACH (br, node, &all_bridges) {
1391 int error = bridge_run_one(br);
1393 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1394 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1395 "forcing reconfiguration", br->name);
1396 datapath_destroyed = true;
1400 /* (Re)configure if necessary. */
1401 database_changed = ovsdb_idl_run(idl);
1402 cfg = ovsrec_open_vswitch_first(idl);
1404 /* Re-configure SSL. We do this on every trip through the main loop,
1405 * instead of just when the database changes, because the contents of the
1406 * key and certificate files can change without the database changing.
1408 * We do this before bridge_reconfigure() because that function might
1409 * initiate SSL connections and thus requires SSL to be configured. */
1410 if (cfg && cfg->ssl) {
1411 const struct ovsrec_ssl *ssl = cfg->ssl;
1413 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1414 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1417 if (database_changed || datapath_destroyed) {
1419 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1421 bridge_configure_once(cfg);
1422 bridge_reconfigure(cfg);
1424 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1425 ovsdb_idl_txn_commit(txn);
1426 ovsdb_idl_txn_destroy(txn); /* XXX */
1428 /* We still need to reconfigure to avoid dangling pointers to
1429 * now-destroyed ovsrec structures inside bridge data. */
1430 static const struct ovsrec_open_vswitch null_cfg;
1432 bridge_reconfigure(&null_cfg);
1436 /* Refresh system and interface stats if necessary. */
1437 if (time_msec() >= stats_timer) {
1439 struct ovsdb_idl_txn *txn;
1441 txn = ovsdb_idl_txn_create(idl);
1442 LIST_FOR_EACH (br, node, &all_bridges) {
1445 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1446 struct iface *iface;
1448 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
1449 iface_refresh_stats(iface);
1450 iface_refresh_cfm_stats(iface);
1451 iface_refresh_status(iface);
1454 bridge_refresh_controller_status(br);
1456 refresh_system_stats(cfg);
1457 ovsdb_idl_txn_commit(txn);
1458 ovsdb_idl_txn_destroy(txn); /* XXX */
1461 stats_timer = time_msec() + STATS_INTERVAL;
1470 LIST_FOR_EACH (br, node, &all_bridges) {
1473 ofproto_wait(br->ofproto);
1474 if (ofproto_has_primary_controller(br->ofproto)) {
1478 mac_learning_wait(br->ml);
1480 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1484 ovsdb_idl_wait(idl);
1485 poll_timer_wait_until(stats_timer);
1488 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1489 * configuration changes. */
1491 bridge_flush(struct bridge *br)
1493 COVERAGE_INC(bridge_flush);
1497 /* Bridge unixctl user interface functions. */
1499 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1500 const char *args, void *aux OVS_UNUSED)
1502 struct ds ds = DS_EMPTY_INITIALIZER;
1503 const struct bridge *br;
1504 const struct mac_entry *e;
1506 br = bridge_lookup(args);
1508 unixctl_command_reply(conn, 501, "no such bridge");
1512 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1513 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1514 struct port *port = e->port.p;
1515 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1516 port_get_an_iface(port)->dp_ifidx,
1517 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1519 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1523 /* QoS unixctl user interface functions. */
1525 struct qos_unixctl_show_cbdata {
1527 struct iface *iface;
1531 qos_unixctl_show_cb(unsigned int queue_id,
1532 const struct shash *details,
1535 struct qos_unixctl_show_cbdata *data = aux;
1536 struct ds *ds = data->ds;
1537 struct iface *iface = data->iface;
1538 struct netdev_queue_stats stats;
1539 struct shash_node *node;
1542 ds_put_cstr(ds, "\n");
1544 ds_put_format(ds, "Queue %u:\n", queue_id);
1546 ds_put_cstr(ds, "Default:\n");
1549 SHASH_FOR_EACH (node, details) {
1550 ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data);
1553 error = netdev_get_queue_stats(iface->netdev, queue_id, &stats);
1555 if (stats.tx_packets != UINT64_MAX) {
1556 ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets);
1559 if (stats.tx_bytes != UINT64_MAX) {
1560 ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes);
1563 if (stats.tx_errors != UINT64_MAX) {
1564 ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors);
1567 ds_put_format(ds, "\tFailed to get statistics for queue %u: %s",
1568 queue_id, strerror(error));
1573 qos_unixctl_show(struct unixctl_conn *conn,
1574 const char *args, void *aux OVS_UNUSED)
1576 struct ds ds = DS_EMPTY_INITIALIZER;
1577 struct shash sh = SHASH_INITIALIZER(&sh);
1578 struct iface *iface;
1580 struct shash_node *node;
1581 struct qos_unixctl_show_cbdata data;
1584 iface = iface_find(args);
1586 unixctl_command_reply(conn, 501, "no such interface");
1590 netdev_get_qos(iface->netdev, &type, &sh);
1592 if (*type != '\0') {
1593 ds_put_format(&ds, "QoS: %s %s\n", iface->name, type);
1595 SHASH_FOR_EACH (node, &sh) {
1596 ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data);
1601 error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data);
1604 ds_put_format(&ds, "failed to dump queues: %s", strerror(error));
1606 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1608 ds_put_format(&ds, "QoS not configured on %s\n", iface->name);
1609 unixctl_command_reply(conn, 501, ds_cstr(&ds));
1612 shash_destroy_free_data(&sh);
1616 /* Bridge reconfiguration functions. */
1617 static struct bridge *
1618 bridge_create(const struct ovsrec_bridge *br_cfg)
1623 assert(!bridge_lookup(br_cfg->name));
1624 br = xzalloc(sizeof *br);
1626 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1632 dpif_flow_flush(br->dpif);
1634 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1637 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1639 dpif_delete(br->dpif);
1640 dpif_close(br->dpif);
1645 br->name = xstrdup(br_cfg->name);
1647 br->ml = mac_learning_create();
1648 eth_addr_nicira_random(br->default_ea);
1650 hmap_init(&br->ports);
1651 hmap_init(&br->ifaces);
1652 shash_init(&br->iface_by_name);
1656 list_push_back(&all_bridges, &br->node);
1658 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1664 bridge_destroy(struct bridge *br)
1667 struct port *port, *next;
1670 HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) {
1673 list_remove(&br->node);
1674 error = dpif_delete(br->dpif);
1675 if (error && error != ENOENT) {
1676 VLOG_ERR("failed to delete %s: %s",
1677 dpif_name(br->dpif), strerror(error));
1679 dpif_close(br->dpif);
1680 ofproto_destroy(br->ofproto);
1681 mac_learning_destroy(br->ml);
1682 hmap_destroy(&br->ifaces);
1683 hmap_destroy(&br->ports);
1684 shash_destroy(&br->iface_by_name);
1690 static struct bridge *
1691 bridge_lookup(const char *name)
1695 LIST_FOR_EACH (br, node, &all_bridges) {
1696 if (!strcmp(br->name, name)) {
1703 /* Handle requests for a listing of all flows known by the OpenFlow
1704 * stack, including those normally hidden. */
1706 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1707 const char *args, void *aux OVS_UNUSED)
1712 br = bridge_lookup(args);
1714 unixctl_command_reply(conn, 501, "Unknown bridge");
1719 ofproto_get_all_flows(br->ofproto, &results);
1721 unixctl_command_reply(conn, 200, ds_cstr(&results));
1722 ds_destroy(&results);
1725 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1726 * connections and reconnect. If BRIDGE is not specified, then all bridges
1727 * drop their controller connections and reconnect. */
1729 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1730 const char *args, void *aux OVS_UNUSED)
1733 if (args[0] != '\0') {
1734 br = bridge_lookup(args);
1736 unixctl_command_reply(conn, 501, "Unknown bridge");
1739 ofproto_reconnect_controllers(br->ofproto);
1741 LIST_FOR_EACH (br, node, &all_bridges) {
1742 ofproto_reconnect_controllers(br->ofproto);
1745 unixctl_command_reply(conn, 200, NULL);
1749 bridge_run_one(struct bridge *br)
1754 error = ofproto_run1(br->ofproto);
1759 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1761 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1765 error = ofproto_run2(br->ofproto, br->flush);
1772 bridge_get_controllers(const struct bridge *br,
1773 struct ovsrec_controller ***controllersp)
1775 struct ovsrec_controller **controllers;
1776 size_t n_controllers;
1778 controllers = br->cfg->controller;
1779 n_controllers = br->cfg->n_controller;
1781 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1787 *controllersp = controllers;
1789 return n_controllers;
1793 bridge_reconfigure_one(struct bridge *br)
1795 enum ofproto_fail_mode fail_mode;
1796 struct svec snoops, old_snoops;
1797 struct port *port, *next;
1798 struct shash_node *node;
1799 struct shash new_ports;
1802 /* Collect new ports. */
1803 shash_init(&new_ports);
1804 for (i = 0; i < br->cfg->n_ports; i++) {
1805 const char *name = br->cfg->ports[i]->name;
1806 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1807 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1812 /* If we have a controller, then we need a local port. Complain if the
1813 * user didn't specify one.
1815 * XXX perhaps we should synthesize a port ourselves in this case. */
1816 if (bridge_get_controllers(br, NULL)) {
1817 char local_name[IF_NAMESIZE];
1820 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1821 local_name, sizeof local_name);
1822 if (!error && !shash_find(&new_ports, local_name)) {
1823 VLOG_WARN("bridge %s: controller specified but no local port "
1824 "(port named %s) defined",
1825 br->name, local_name);
1829 /* Get rid of deleted ports.
1830 * Get rid of deleted interfaces on ports that still exist. */
1831 HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) {
1832 const struct ovsrec_port *port_cfg;
1834 port_cfg = shash_find_data(&new_ports, port->name);
1838 port_del_ifaces(port, port_cfg);
1842 /* Create new ports.
1843 * Add new interfaces to existing ports.
1844 * Reconfigure existing ports. */
1845 SHASH_FOR_EACH (node, &new_ports) {
1846 struct port *port = port_lookup(br, node->name);
1848 port = port_create(br, node->name);
1851 port_reconfigure(port, node->data);
1852 if (!port->n_ifaces) {
1853 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1854 br->name, port->name);
1858 shash_destroy(&new_ports);
1860 /* Set the fail-mode */
1861 fail_mode = !br->cfg->fail_mode
1862 || !strcmp(br->cfg->fail_mode, "standalone")
1863 ? OFPROTO_FAIL_STANDALONE
1864 : OFPROTO_FAIL_SECURE;
1865 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1866 && !ofproto_has_primary_controller(br->ofproto)) {
1867 ofproto_flush_flows(br->ofproto);
1869 ofproto_set_fail_mode(br->ofproto, fail_mode);
1871 /* Delete all flows if we're switching from connected to standalone or vice
1872 * versa. (XXX Should we delete all flows if we are switching from one
1873 * controller to another?) */
1875 /* Configure OpenFlow controller connection snooping. */
1877 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1878 ovs_rundir(), br->name));
1879 svec_init(&old_snoops);
1880 ofproto_get_snoops(br->ofproto, &old_snoops);
1881 if (!svec_equal(&snoops, &old_snoops)) {
1882 ofproto_set_snoops(br->ofproto, &snoops);
1884 svec_destroy(&snoops);
1885 svec_destroy(&old_snoops);
1887 mirror_reconfigure(br);
1890 /* Initializes 'oc' appropriately as a management service controller for
1893 * The caller must free oc->target when it is no longer needed. */
1895 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1896 struct ofproto_controller *oc)
1898 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1899 oc->max_backoff = 0;
1900 oc->probe_interval = 60;
1901 oc->band = OFPROTO_OUT_OF_BAND;
1903 oc->burst_limit = 0;
1906 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1908 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1909 struct ofproto_controller *oc)
1911 oc->target = c->target;
1912 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1913 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1914 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1915 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1916 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1917 oc->burst_limit = (c->controller_burst_limit
1918 ? *c->controller_burst_limit : 0);
1921 /* Configures the IP stack for 'br''s local interface properly according to the
1922 * configuration in 'c'. */
1924 bridge_configure_local_iface_netdev(struct bridge *br,
1925 struct ovsrec_controller *c)
1927 struct netdev *netdev;
1928 struct in_addr mask, gateway;
1930 struct iface *local_iface;
1933 /* If there's no local interface or no IP address, give up. */
1934 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
1935 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1939 /* Bring up the local interface. */
1940 netdev = local_iface->netdev;
1941 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1943 /* Configure the IP address and netmask. */
1944 if (!c->local_netmask
1945 || !inet_aton(c->local_netmask, &mask)
1947 mask.s_addr = guess_netmask(ip.s_addr);
1949 if (!netdev_set_in4(netdev, ip, mask)) {
1950 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1951 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1954 /* Configure the default gateway. */
1955 if (c->local_gateway
1956 && inet_aton(c->local_gateway, &gateway)
1957 && gateway.s_addr) {
1958 if (!netdev_add_router(netdev, gateway)) {
1959 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1960 br->name, IP_ARGS(&gateway.s_addr));
1966 bridge_reconfigure_remotes(struct bridge *br,
1967 const struct sockaddr_in *managers,
1970 const char *disable_ib_str, *queue_id_str;
1971 bool disable_in_band = false;
1974 struct ovsrec_controller **controllers;
1975 size_t n_controllers;
1978 struct ofproto_controller *ocs;
1982 /* Check if we should disable in-band control on this bridge. */
1983 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1984 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1985 disable_in_band = true;
1988 /* Set OpenFlow queue ID for in-band control. */
1989 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
1990 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
1991 ofproto_set_in_band_queue(br->ofproto, queue_id);
1993 if (disable_in_band) {
1994 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1996 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1998 had_primary = ofproto_has_primary_controller(br->ofproto);
2000 n_controllers = bridge_get_controllers(br, &controllers);
2002 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
2005 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
2006 for (i = 0; i < n_controllers; i++) {
2007 struct ovsrec_controller *c = controllers[i];
2009 if (!strncmp(c->target, "punix:", 6)
2010 || !strncmp(c->target, "unix:", 5)) {
2011 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2013 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
2014 * domain sockets and overwriting arbitrary local files. */
2015 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2016 "\"%s\" due to possibility for remote exploit",
2017 dpif_name(br->dpif), c->target);
2021 bridge_configure_local_iface_netdev(br, c);
2022 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2023 if (disable_in_band) {
2024 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2029 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2030 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2033 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2034 ofproto_flush_flows(br->ofproto);
2037 /* If there are no controllers and the bridge is in standalone
2038 * mode, set up a flow that matches every packet and directs
2039 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2040 * switch is in secure mode and we won't pass any traffic until
2041 * a controller has been defined and it tells us to do so. */
2043 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2044 union ofp_action action;
2045 struct cls_rule rule;
2047 memset(&action, 0, sizeof action);
2048 action.type = htons(OFPAT_OUTPUT);
2049 action.output.len = htons(sizeof action);
2050 action.output.port = htons(OFPP_NORMAL);
2051 cls_rule_init_catchall(&rule, 0);
2052 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2057 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2062 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2063 struct iface *iface;
2065 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2066 shash_add_once(ifaces, iface->name, iface);
2068 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2069 shash_add_once(ifaces, port->name, NULL);
2074 /* For robustness, in case the administrator moves around datapath ports behind
2075 * our back, we re-check all the datapath port numbers here.
2077 * This function will set the 'dp_ifidx' members of interfaces that have
2078 * disappeared to -1, so only call this function from a context where those
2079 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2080 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2081 * datapath, which doesn't support UINT16_MAX+1 ports. */
2083 bridge_fetch_dp_ifaces(struct bridge *br)
2085 struct dpif_port_dump dump;
2086 struct dpif_port dpif_port;
2089 /* Reset all interface numbers. */
2090 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2091 struct iface *iface;
2093 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2094 iface->dp_ifidx = -1;
2097 hmap_clear(&br->ifaces);
2099 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2100 struct iface *iface = iface_lookup(br, dpif_port.name);
2102 if (iface->dp_ifidx >= 0) {
2103 VLOG_WARN("%s reported interface %s twice",
2104 dpif_name(br->dpif), dpif_port.name);
2105 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2106 VLOG_WARN("%s reported interface %"PRIu16" twice",
2107 dpif_name(br->dpif), dpif_port.port_no);
2109 iface->dp_ifidx = dpif_port.port_no;
2110 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2111 hash_int(iface->dp_ifidx, 0));
2114 iface_set_ofport(iface->cfg,
2115 (iface->dp_ifidx >= 0
2116 ? odp_port_to_ofp_port(iface->dp_ifidx)
2122 /* Bridge packet processing functions. */
2125 bond_is_tcp_hash(const struct port *port)
2127 return port->bond_mode == BM_TCP && lacp_negotiated(port->lacp);
2131 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2133 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2136 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2138 struct flow hash_flow;
2140 memcpy(&hash_flow, flow, sizeof hash_flow);
2141 hash_flow.vlan_tci = 0;
2143 /* The symmetric quality of this hash function is not required, but
2144 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2145 * purposes, so we use it out of convenience. */
2146 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2149 static struct bond_entry *
2150 lookup_bond_entry(const struct port *port, const struct flow *flow,
2153 assert(port->bond_mode != BM_AB);
2155 if (bond_is_tcp_hash(port)) {
2156 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2158 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2162 static struct iface *
2163 bond_choose_iface(const struct port *port)
2165 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2166 struct iface *best_down_slave;
2167 struct iface *iface;
2169 best_down_slave = NULL;
2170 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2171 if (iface->enabled) {
2173 } else if ((!best_down_slave
2174 || iface->delay_expires < best_down_slave->delay_expires)
2175 && lacp_slave_may_enable(port->lacp, iface)) {
2176 best_down_slave = iface;
2180 if (best_down_slave) {
2181 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2182 "since no other interface is up",
2183 best_down_slave->name,
2184 best_down_slave->delay_expires - time_msec());
2185 bond_enable_slave(best_down_slave, true);
2188 return best_down_slave;
2192 choose_output_iface(const struct port *port, const struct flow *flow,
2193 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2195 struct iface *iface;
2197 assert(port->n_ifaces);
2198 if (port->n_ifaces == 1) {
2199 iface = port_get_an_iface(port);
2200 } else if (port->bond_mode == BM_AB) {
2201 iface = port->active_iface;
2203 *tags |= port->no_ifaces_tag;
2207 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2208 if (!e->iface || !e->iface->enabled) {
2209 /* XXX select interface properly. The current interface selection
2210 * is only good for testing the rebalancing code. */
2211 e->iface = bond_choose_iface(port);
2213 *tags |= port->no_ifaces_tag;
2216 e->tag = tag_create_random();
2221 *dp_ifidx = iface->dp_ifidx;
2222 *tags |= iface->tag; /* Currently only used for bonding. */
2227 bond_link_status_update(struct iface *iface)
2229 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2230 struct port *port = iface->port;
2231 bool up = iface->up && lacp_slave_may_enable(port->lacp, iface);
2232 int updelay, downdelay;
2234 updelay = port->updelay;
2235 downdelay = port->downdelay;
2237 if (lacp_negotiated(port->lacp)) {
2242 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2243 /* Nothing to do. */
2246 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2247 iface->name, up ? "up" : "down");
2248 if (up == iface->enabled) {
2249 iface->delay_expires = LLONG_MAX;
2250 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2251 iface->name, up ? "disabled" : "enabled");
2252 } else if (up && !port->active_iface) {
2253 bond_enable_slave(iface, true);
2255 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2256 "other interface is up", iface->name, updelay);
2259 int delay = up ? updelay : downdelay;
2260 iface->delay_expires = time_msec() + delay;
2263 "interface %s: will be %s if it stays %s for %d ms",
2265 up ? "enabled" : "disabled",
2273 bond_choose_active_iface(struct port *port)
2275 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2277 port->active_iface = bond_choose_iface(port);
2278 if (port->active_iface) {
2279 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2280 port->name, port->active_iface->name);
2282 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2288 bond_enable_slave(struct iface *iface, bool enable)
2290 struct port *port = iface->port;
2291 struct bridge *br = port->bridge;
2293 /* This acts as a recursion check. If the act of disabling a slave
2294 * causes a different slave to be enabled, the flag will allow us to
2295 * skip redundant work when we reenter this function. It must be
2296 * cleared on exit to keep things safe with multiple bonds. */
2297 static bool moving_active_iface = false;
2299 iface->delay_expires = LLONG_MAX;
2300 if (enable == iface->enabled) {
2304 iface->enabled = enable;
2305 if (!iface->enabled) {
2306 VLOG_WARN("interface %s: disabled", iface->name);
2307 ofproto_revalidate(br->ofproto, iface->tag);
2308 if (iface == port->active_iface) {
2309 /* Disabling a slave can lead to another slave being immediately
2310 * enabled if there will be no active slaves but one is waiting
2311 * on an updelay. In this case we do not need to run most of the
2312 * code for the newly enabled slave since there was no period
2313 * without an active slave and it is redundant with the disabling
2315 moving_active_iface = true;
2316 bond_choose_active_iface(port);
2318 bond_send_learning_packets(port);
2320 VLOG_WARN("interface %s: enabled", iface->name);
2321 if (!port->active_iface && !moving_active_iface) {
2322 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2323 bond_choose_active_iface(port);
2324 bond_send_learning_packets(port);
2326 iface->tag = tag_create_random();
2329 moving_active_iface = false;
2332 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2333 * bond interface. */
2335 bond_update_fake_iface_stats(struct port *port)
2337 struct netdev_stats bond_stats;
2338 struct netdev *bond_dev;
2339 struct iface *iface;
2341 memset(&bond_stats, 0, sizeof bond_stats);
2343 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2344 struct netdev_stats slave_stats;
2346 if (!netdev_get_stats(iface->netdev, &slave_stats)) {
2347 /* XXX: We swap the stats here because they are swapped back when
2348 * reported by the internal device. The reason for this is
2349 * internal devices normally represent packets going into the system
2350 * but when used as fake bond device they represent packets leaving
2351 * the system. We really should do this in the internal device
2352 * itself because changing it here reverses the counts from the
2353 * perspective of the switch. However, the internal device doesn't
2354 * know what type of device it represents so we have to do it here
2356 bond_stats.tx_packets += slave_stats.rx_packets;
2357 bond_stats.tx_bytes += slave_stats.rx_bytes;
2358 bond_stats.rx_packets += slave_stats.tx_packets;
2359 bond_stats.rx_bytes += slave_stats.tx_bytes;
2363 if (!netdev_open_default(port->name, &bond_dev)) {
2364 netdev_set_stats(bond_dev, &bond_stats);
2365 netdev_close(bond_dev);
2370 bond_run(struct port *port)
2372 struct iface *iface;
2374 if (port->n_ifaces < 2) {
2378 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2379 bond_link_status_update(iface);
2382 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2383 if (time_msec() >= iface->delay_expires) {
2384 bond_enable_slave(iface, !iface->enabled);
2388 if (port->bond_fake_iface
2389 && time_msec() >= port->bond_next_fake_iface_update) {
2390 bond_update_fake_iface_stats(port);
2391 port->bond_next_fake_iface_update = time_msec() + 1000;
2396 bond_wait(struct port *port)
2398 struct iface *iface;
2400 if (port->n_ifaces < 2) {
2404 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2405 if (iface->delay_expires != LLONG_MAX) {
2406 poll_timer_wait_until(iface->delay_expires);
2410 if (port->bond_fake_iface) {
2411 poll_timer_wait_until(port->bond_next_fake_iface_update);
2416 set_dst(struct dst *dst, const struct flow *flow,
2417 const struct port *in_port, const struct port *out_port,
2420 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2421 : in_port->vlan >= 0 ? in_port->vlan
2422 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2423 : vlan_tci_to_vid(flow->vlan_tci));
2424 return choose_output_iface(out_port, flow, dst->vlan,
2425 &dst->dp_ifidx, tags);
2429 swap_dst(struct dst *p, struct dst *q)
2431 struct dst tmp = *p;
2436 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2437 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2438 * that we push to the datapath. We could in fact fully sort the array by
2439 * vlan, but in most cases there are at most two different vlan tags so that's
2440 * possibly overkill.) */
2442 partition_dsts(struct dst_set *set, int vlan)
2444 struct dst *first = set->dsts;
2445 struct dst *last = set->dsts + set->n;
2447 while (first != last) {
2449 * - All dsts < first have vlan == 'vlan'.
2450 * - All dsts >= last have vlan != 'vlan'.
2451 * - first < last. */
2452 while (first->vlan == vlan) {
2453 if (++first == last) {
2458 /* Same invariants, plus one additional:
2459 * - first->vlan != vlan.
2461 while (last[-1].vlan != vlan) {
2462 if (--last == first) {
2467 /* Same invariants, plus one additional:
2468 * - last[-1].vlan == vlan.*/
2469 swap_dst(first++, --last);
2474 mirror_mask_ffs(mirror_mask_t mask)
2476 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2481 dst_set_init(struct dst_set *set)
2483 set->dsts = set->builtin;
2485 set->allocated = ARRAY_SIZE(set->builtin);
2489 dst_set_add(struct dst_set *set, const struct dst *dst)
2491 if (set->n >= set->allocated) {
2492 size_t new_allocated;
2493 struct dst *new_dsts;
2495 new_allocated = set->allocated * 2;
2496 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2497 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2501 set->dsts = new_dsts;
2502 set->allocated = new_allocated;
2504 set->dsts[set->n++] = *dst;
2508 dst_set_free(struct dst_set *set)
2510 if (set->dsts != set->builtin) {
2516 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2519 for (i = 0; i < set->n; i++) {
2520 if (set->dsts[i].vlan == test->vlan
2521 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2529 port_trunks_vlan(const struct port *port, uint16_t vlan)
2531 return (port->vlan < 0
2532 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2536 port_includes_vlan(const struct port *port, uint16_t vlan)
2538 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2542 port_is_floodable(const struct port *port)
2544 struct iface *iface;
2546 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2547 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2555 /* Returns the tag for 'port''s active iface, or 'port''s no_ifaces_tag if
2556 * there is no active iface. */
2558 port_get_active_iface_tag(const struct port *port)
2560 return (port->active_iface
2561 ? port->active_iface->tag
2562 : port->no_ifaces_tag);
2565 /* Returns an arbitrary interface within 'port'.
2567 * 'port' must have at least one interface. */
2568 static struct iface *
2569 port_get_an_iface(const struct port *port)
2571 return CONTAINER_OF(list_front(&port->ifaces), struct iface, port_elem);
2575 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2576 const struct port *in_port, const struct port *out_port,
2577 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2579 mirror_mask_t mirrors = in_port->src_mirrors;
2583 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2584 if (flow_vlan == 0) {
2585 flow_vlan = OFP_VLAN_NONE;
2588 if (out_port == FLOOD_PORT) {
2591 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2593 && port_is_floodable(port)
2594 && port_includes_vlan(port, vlan)
2595 && !port->is_mirror_output_port
2596 && set_dst(&dst, flow, in_port, port, tags)) {
2597 mirrors |= port->dst_mirrors;
2598 dst_set_add(set, &dst);
2601 *nf_output_iface = NF_OUT_FLOOD;
2602 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2603 dst_set_add(set, &dst);
2604 *nf_output_iface = dst.dp_ifidx;
2605 mirrors |= out_port->dst_mirrors;
2609 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2610 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2612 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2613 && !dst_is_duplicate(set, &dst)) {
2614 dst_set_add(set, &dst);
2619 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2620 if (port_includes_vlan(port, m->out_vlan)
2621 && set_dst(&dst, flow, in_port, port, tags))
2623 if (port->vlan < 0) {
2624 dst.vlan = m->out_vlan;
2626 if (dst_is_duplicate(set, &dst)) {
2630 /* Use the vlan tag on the original flow instead of
2631 * the one passed in the vlan parameter. This ensures
2632 * that we compare the vlan from before any implicit
2633 * tagging tags place. This is necessary because
2634 * dst->vlan is the final vlan, after removing implicit
2636 if (port == in_port && dst.vlan == flow_vlan) {
2637 /* Don't send out input port on same VLAN. */
2640 dst_set_add(set, &dst);
2645 mirrors &= mirrors - 1;
2648 partition_dsts(set, flow_vlan);
2651 static void OVS_UNUSED
2652 print_dsts(const struct dst_set *set)
2656 for (i = 0; i < set->n; i++) {
2657 const struct dst *dst = &set->dsts[i];
2659 printf(">p%"PRIu16, dst->dp_ifidx);
2660 if (dst->vlan != OFP_VLAN_NONE) {
2661 printf("v%"PRIu16, dst->vlan);
2667 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2668 const struct port *in_port, const struct port *out_port,
2669 tag_type *tags, struct ofpbuf *actions,
2670 uint16_t *nf_output_iface)
2677 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2680 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2681 if (cur_vlan == 0) {
2682 cur_vlan = OFP_VLAN_NONE;
2684 for (i = 0; i < set.n; i++) {
2685 const struct dst *dst = &set.dsts[i];
2686 if (dst->vlan != cur_vlan) {
2687 if (dst->vlan == OFP_VLAN_NONE) {
2688 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2691 tci = htons(dst->vlan & VLAN_VID_MASK);
2692 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2693 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2695 cur_vlan = dst->vlan;
2697 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2702 /* Returns the effective vlan of a packet, taking into account both the
2703 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2704 * the packet is untagged and -1 indicates it has an invalid header and
2705 * should be dropped. */
2706 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2707 struct port *in_port, bool have_packet)
2709 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2710 if (in_port->vlan >= 0) {
2712 /* XXX support double tagging? */
2714 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2715 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2716 "packet received on port %s configured with "
2717 "implicit VLAN %"PRIu16,
2718 br->name, vlan, in_port->name, in_port->vlan);
2722 vlan = in_port->vlan;
2724 if (!port_includes_vlan(in_port, vlan)) {
2726 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2727 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2728 "packet received on port %s not configured for "
2730 br->name, vlan, in_port->name, vlan);
2739 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2740 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2741 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2743 is_gratuitous_arp(const struct flow *flow)
2745 return (flow->dl_type == htons(ETH_TYPE_ARP)
2746 && eth_addr_is_broadcast(flow->dl_dst)
2747 && (flow->nw_proto == ARP_OP_REPLY
2748 || (flow->nw_proto == ARP_OP_REQUEST
2749 && flow->nw_src == flow->nw_dst)));
2753 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2754 struct port *in_port)
2756 struct mac_entry *mac;
2758 if (!mac_learning_may_learn(br->ml, flow->dl_src, vlan)) {
2762 mac = mac_learning_insert(br->ml, flow->dl_src, vlan);
2763 if (is_gratuitous_arp(flow)) {
2764 /* We don't want to learn from gratuitous ARP packets that are
2765 * reflected back over bond slaves so we lock the learning table. */
2766 if (in_port->n_ifaces == 1) {
2767 mac_entry_set_grat_arp_lock(mac);
2768 } else if (mac_entry_is_grat_arp_locked(mac)) {
2773 if (mac_entry_is_new(mac) || mac->port.p != in_port) {
2774 /* The log messages here could actually be useful in debugging,
2775 * so keep the rate limit relatively high. */
2776 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
2777 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2778 "on port %s in VLAN %d",
2779 br->name, ETH_ADDR_ARGS(flow->dl_src),
2780 in_port->name, vlan);
2782 mac->port.p = in_port;
2783 ofproto_revalidate(br->ofproto, mac_learning_changed(br->ml, mac));
2787 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2788 * dropped. Returns true if they may be forwarded, false if they should be
2791 * If 'have_packet' is true, it indicates that the caller is processing a
2792 * received packet. If 'have_packet' is false, then the caller is just
2793 * revalidating an existing flow because configuration has changed. Either
2794 * way, 'have_packet' only affects logging (there is no point in logging errors
2795 * during revalidation).
2797 * Sets '*in_portp' to the input port. This will be a null pointer if
2798 * flow->in_port does not designate a known input port (in which case
2799 * is_admissible() returns false).
2801 * When returning true, sets '*vlanp' to the effective VLAN of the input
2802 * packet, as returned by flow_get_vlan().
2804 * May also add tags to '*tags', although the current implementation only does
2805 * so in one special case.
2808 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2809 tag_type *tags, int *vlanp, struct port **in_portp)
2811 struct iface *in_iface;
2812 struct port *in_port;
2815 /* Find the interface and port structure for the received packet. */
2816 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2818 /* No interface? Something fishy... */
2820 /* Odd. A few possible reasons here:
2822 * - We deleted an interface but there are still a few packets
2823 * queued up from it.
2825 * - Someone externally added an interface (e.g. with "ovs-dpctl
2826 * add-if") that we don't know about.
2828 * - Packet arrived on the local port but the local port is not
2829 * one of our bridge ports.
2831 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2833 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2834 "interface %"PRIu16, br->name, flow->in_port);
2840 *in_portp = in_port = in_iface->port;
2841 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2846 /* Drop frames for reserved multicast addresses. */
2847 if (eth_addr_is_reserved(flow->dl_dst)) {
2851 /* Drop frames on ports reserved for mirroring. */
2852 if (in_port->is_mirror_output_port) {
2854 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2855 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2856 "%s, which is reserved exclusively for mirroring",
2857 br->name, in_port->name);
2862 /* When using LACP, do not accept packets from disabled interfaces. */
2863 if (lacp_negotiated(in_port->lacp) && !in_iface->enabled) {
2867 /* Packets received on non-LACP bonds need special attention to avoid
2869 if (in_port->n_ifaces > 1 && !lacp_negotiated(in_port->lacp)) {
2870 struct mac_entry *mac;
2872 if (eth_addr_is_multicast(flow->dl_dst)) {
2873 *tags |= port_get_active_iface_tag(in_port);
2874 if (in_port->active_iface != in_iface) {
2875 /* Drop all multicast packets on inactive slaves. */
2880 /* Drop all packets for which we have learned a different input
2881 * port, because we probably sent the packet on one slave and got
2882 * it back on the other. Gratuitous ARP packets are an exception
2883 * to this rule: the host has moved to another switch. The exception
2884 * to the exception is if we locked the learning table to avoid
2885 * reflections on bond slaves. If this is the case, just drop the
2887 mac = mac_learning_lookup(br->ml, flow->dl_src, vlan, NULL);
2888 if (mac && mac->port.p != in_port &&
2889 (!is_gratuitous_arp(flow) || mac_entry_is_grat_arp_locked(mac))) {
2897 /* If the composed actions may be applied to any packet in the given 'flow',
2898 * returns true. Otherwise, the actions should only be applied to 'packet', or
2899 * not at all, if 'packet' was NULL. */
2901 process_flow(struct bridge *br, const struct flow *flow,
2902 const struct ofpbuf *packet, struct ofpbuf *actions,
2903 tag_type *tags, uint16_t *nf_output_iface)
2905 struct port *in_port;
2906 struct port *out_port;
2907 struct mac_entry *mac;
2910 /* Check whether we should drop packets in this flow. */
2911 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2916 /* Learn source MAC (but don't try to learn from revalidation). */
2918 update_learning_table(br, flow, vlan, in_port);
2921 /* Determine output port. */
2922 mac = mac_learning_lookup(br->ml, flow->dl_dst, vlan, tags);
2925 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2926 /* If we are revalidating but don't have a learning entry then
2927 * eject the flow. Installing a flow that floods packets opens
2928 * up a window of time where we could learn from a packet reflected
2929 * on a bond and blackhole packets before the learning table is
2930 * updated to reflect the correct port. */
2933 out_port = FLOOD_PORT;
2936 /* Don't send packets out their input ports. */
2937 if (in_port == out_port) {
2943 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2951 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2952 struct ofpbuf *actions, tag_type *tags,
2953 uint16_t *nf_output_iface, void *br_)
2955 struct bridge *br = br_;
2957 COVERAGE_INC(bridge_process_flow);
2958 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2962 bridge_special_ofhook_cb(const struct flow *flow,
2963 const struct ofpbuf *packet, void *br_)
2965 struct iface *iface;
2966 struct bridge *br = br_;
2968 iface = iface_from_dp_ifidx(br, flow->in_port);
2970 if (flow->dl_type == htons(ETH_TYPE_LACP)) {
2972 if (iface && iface->port->lacp && packet) {
2973 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
2976 COVERAGE_INC(bridge_process_lacp);
2977 lacp_process_pdu(iface->port->lacp, iface, pdu);
2987 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
2988 const struct nlattr *actions,
2990 uint64_t n_bytes, void *br_)
2992 struct bridge *br = br_;
2993 const struct nlattr *a;
2994 struct port *in_port;
2999 /* Feed information from the active flows back into the learning table to
3000 * ensure that table is always in sync with what is actually flowing
3001 * through the datapath.
3003 * We test that 'tags' is nonzero to ensure that only flows that include an
3004 * OFPP_NORMAL action are used for learning. This works because
3005 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3006 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3007 update_learning_table(br, flow, vlan, in_port);
3010 /* Account for bond slave utilization. */
3011 if (!br->has_bonded_ports) {
3014 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3015 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3016 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3017 if (out_port && out_port->n_ifaces >= 2 &&
3018 out_port->bond_mode != BM_AB) {
3019 uint16_t vlan = (flow->vlan_tci
3020 ? vlan_tci_to_vid(flow->vlan_tci)
3022 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
3023 e->tx_bytes += n_bytes;
3030 bridge_account_checkpoint_ofhook_cb(void *br_)
3032 struct bridge *br = br_;
3036 if (!br->has_bonded_ports) {
3041 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3042 if (port->n_ifaces > 1 && port->bond_mode != BM_AB
3043 && now >= port->bond_next_rebalance) {
3044 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3045 bond_rebalance_port(port);
3050 static struct ofhooks bridge_ofhooks = {
3051 bridge_normal_ofhook_cb,
3052 bridge_special_ofhook_cb,
3053 bridge_account_flow_ofhook_cb,
3054 bridge_account_checkpoint_ofhook_cb,
3057 /* Bonding functions. */
3059 /* Statistics for a single interface on a bonded port, used for load-based
3060 * bond rebalancing. */
3061 struct slave_balance {
3062 struct iface *iface; /* The interface. */
3063 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3065 /* All the "bond_entry"s that are assigned to this interface, in order of
3066 * increasing tx_bytes. */
3067 struct bond_entry **hashes;
3072 bond_mode_to_string(enum bond_mode bm) {
3073 static char *bm_slb = "balance-slb";
3074 static char *bm_ab = "active-backup";
3075 static char *bm_tcp = "balance-tcp";
3078 case BM_SLB: return bm_slb;
3079 case BM_AB: return bm_ab;
3080 case BM_TCP: return bm_tcp;
3087 /* Sorts pointers to pointers to bond_entries in ascending order by the
3088 * interface to which they are assigned, and within a single interface in
3089 * ascending order of bytes transmitted. */
3091 compare_bond_entries(const void *a_, const void *b_)
3093 const struct bond_entry *const *ap = a_;
3094 const struct bond_entry *const *bp = b_;
3095 const struct bond_entry *a = *ap;
3096 const struct bond_entry *b = *bp;
3097 if (a->iface != b->iface) {
3098 return a->iface > b->iface ? 1 : -1;
3099 } else if (a->tx_bytes != b->tx_bytes) {
3100 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3106 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3107 * *descending* order by number of bytes transmitted. */
3109 compare_slave_balance(const void *a_, const void *b_)
3111 const struct slave_balance *a = a_;
3112 const struct slave_balance *b = b_;
3113 if (a->iface->enabled != b->iface->enabled) {
3114 return a->iface->enabled ? -1 : 1;
3115 } else if (a->tx_bytes != b->tx_bytes) {
3116 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3123 swap_bals(struct slave_balance *a, struct slave_balance *b)
3125 struct slave_balance tmp = *a;
3130 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3131 * given that 'p' (and only 'p') might be in the wrong location.
3133 * This function invalidates 'p', since it might now be in a different memory
3136 resort_bals(struct slave_balance *p,
3137 struct slave_balance bals[], size_t n_bals)
3140 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3141 swap_bals(p, p - 1);
3143 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3144 swap_bals(p, p + 1);
3150 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3152 if (VLOG_IS_DBG_ENABLED()) {
3153 struct ds ds = DS_EMPTY_INITIALIZER;
3154 const struct slave_balance *b;
3156 for (b = bals; b < bals + n_bals; b++) {
3160 ds_put_char(&ds, ',');
3162 ds_put_format(&ds, " %s %"PRIu64"kB",
3163 b->iface->name, b->tx_bytes / 1024);
3165 if (!b->iface->enabled) {
3166 ds_put_cstr(&ds, " (disabled)");
3168 if (b->n_hashes > 0) {
3169 ds_put_cstr(&ds, " (");
3170 for (i = 0; i < b->n_hashes; i++) {
3171 const struct bond_entry *e = b->hashes[i];
3173 ds_put_cstr(&ds, " + ");
3175 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3176 e - port->bond_hash, e->tx_bytes / 1024);
3178 ds_put_cstr(&ds, ")");
3181 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3186 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3188 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3191 struct bond_entry *hash = from->hashes[hash_idx];
3192 struct port *port = from->iface->port;
3193 uint64_t delta = hash->tx_bytes;
3195 assert(port->bond_mode != BM_AB);
3197 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3198 "from %s to %s (now carrying %"PRIu64"kB and "
3199 "%"PRIu64"kB load, respectively)",
3200 port->name, delta / 1024, hash - port->bond_hash,
3201 from->iface->name, to->iface->name,
3202 (from->tx_bytes - delta) / 1024,
3203 (to->tx_bytes + delta) / 1024);
3205 /* Delete element from from->hashes.
3207 * We don't bother to add the element to to->hashes because not only would
3208 * it require more work, the only purpose it would be to allow that hash to
3209 * be migrated to another slave in this rebalancing run, and there is no
3210 * point in doing that. */
3211 if (hash_idx == 0) {
3214 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3215 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3219 /* Shift load away from 'from' to 'to'. */
3220 from->tx_bytes -= delta;
3221 to->tx_bytes += delta;
3223 /* Arrange for flows to be revalidated. */
3224 ofproto_revalidate(port->bridge->ofproto, hash->tag);
3225 hash->iface = to->iface;
3226 hash->tag = tag_create_random();
3230 bond_rebalance_port(struct port *port)
3232 struct slave_balance *bals;
3234 struct bond_entry *hashes[BOND_MASK + 1];
3235 struct slave_balance *b, *from, *to;
3236 struct bond_entry *e;
3237 struct iface *iface;
3240 assert(port->bond_mode != BM_AB);
3242 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3243 * descending order of tx_bytes, so that bals[0] represents the most
3244 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3247 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3248 * array for each slave_balance structure, we sort our local array of
3249 * hashes in order by slave, so that all of the hashes for a given slave
3250 * become contiguous in memory, and then we point each 'hashes' members of
3251 * a slave_balance structure to the start of a contiguous group. */
3252 n_bals = port->n_ifaces;
3253 b = bals = xmalloc(n_bals * sizeof *bals);
3254 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3261 assert(b == &bals[n_bals]);
3262 for (i = 0; i <= BOND_MASK; i++) {
3263 hashes[i] = &port->bond_hash[i];
3265 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3266 for (i = 0; i <= BOND_MASK; i++) {
3272 for (b = bals; b < &bals[n_bals]; b++) {
3273 if (b->iface == e->iface) {
3274 b->tx_bytes += e->tx_bytes;
3276 b->hashes = &hashes[i];
3283 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3284 log_bals(bals, n_bals, port);
3286 /* Discard slaves that aren't enabled (which were sorted to the back of the
3287 * array earlier). */
3288 while (!bals[n_bals - 1].iface->enabled) {
3295 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3296 to = &bals[n_bals - 1];
3297 for (from = bals; from < to; ) {
3298 uint64_t overload = from->tx_bytes - to->tx_bytes;
3299 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3300 /* The extra load on 'from' (and all less-loaded slaves), compared
3301 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3302 * it is less than ~1Mbps. No point in rebalancing. */
3304 } else if (from->n_hashes == 1) {
3305 /* 'from' only carries a single MAC hash, so we can't shift any
3306 * load away from it, even though we want to. */
3309 /* 'from' is carrying significantly more load than 'to', and that
3310 * load is split across at least two different hashes. Pick a hash
3311 * to migrate to 'to' (the least-loaded slave), given that doing so
3312 * must decrease the ratio of the load on the two slaves by at
3315 * The sort order we use means that we prefer to shift away the
3316 * smallest hashes instead of the biggest ones. There is little
3317 * reason behind this decision; we could use the opposite sort
3318 * order to shift away big hashes ahead of small ones. */
3321 for (i = 0; i < from->n_hashes; i++) {
3322 double old_ratio, new_ratio;
3323 uint64_t delta = from->hashes[i]->tx_bytes;
3325 if (delta == 0 || from->tx_bytes - delta == 0) {
3326 /* Pointless move. */
3330 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3332 if (to->tx_bytes == 0) {
3333 /* Nothing on the new slave, move it. */
3337 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3338 new_ratio = (double)(from->tx_bytes - delta) /
3339 (to->tx_bytes + delta);
3341 if (new_ratio == 0) {
3342 /* Should already be covered but check to prevent division
3347 if (new_ratio < 1) {
3348 new_ratio = 1 / new_ratio;
3351 if (old_ratio - new_ratio > 0.1) {
3352 /* Would decrease the ratio, move it. */
3356 if (i < from->n_hashes) {
3357 bond_shift_load(from, to, i);
3359 /* If the result of the migration changed the relative order of
3360 * 'from' and 'to' swap them back to maintain invariants. */
3361 if (order_swapped) {
3362 swap_bals(from, to);
3365 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3366 * point to different slave_balance structures. It is only
3367 * valid to do these two operations in a row at all because we
3368 * know that 'from' will not move past 'to' and vice versa. */
3369 resort_bals(from, bals, n_bals);
3370 resort_bals(to, bals, n_bals);
3377 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3378 * historical data to decay to <1% in 7 rebalancing runs. */
3379 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3391 bond_send_learning_packets(struct port *port)
3393 struct bridge *br = port->bridge;
3394 struct mac_entry *e;
3395 struct ofpbuf packet;
3396 int error, n_packets, n_errors;
3398 if (!port->n_ifaces || !port->active_iface || bond_is_tcp_hash(port)) {
3402 ofpbuf_init(&packet, 128);
3403 error = n_packets = n_errors = 0;
3404 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3410 if (e->port.p == port) {
3414 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3416 flow_extract(&packet, 0, ODPP_NONE, &flow);
3418 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3424 retval = ofproto_send_packet(br->ofproto, dp_ifidx, e->vlan, &packet);
3430 ofpbuf_uninit(&packet);
3433 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3434 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3435 "packets, last error was: %s",
3436 port->name, n_errors, n_packets, strerror(error));
3438 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3439 port->name, n_packets);
3443 /* Bonding unixctl user interface functions. */
3446 bond_unixctl_list(struct unixctl_conn *conn,
3447 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3449 struct ds ds = DS_EMPTY_INITIALIZER;
3450 const struct bridge *br;
3452 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3454 LIST_FOR_EACH (br, node, &all_bridges) {
3457 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3458 if (port->n_ifaces > 1) {
3459 struct iface *iface;
3461 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3462 bond_mode_to_string(port->bond_mode));
3463 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3464 if (&iface->port_elem != list_front(&port->ifaces)) {
3465 ds_put_cstr(&ds, ", ");
3467 ds_put_cstr(&ds, iface->name);
3469 ds_put_char(&ds, '\n');
3473 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3477 static struct port *
3478 bond_find(const char *name)
3480 const struct bridge *br;
3482 LIST_FOR_EACH (br, node, &all_bridges) {
3485 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3486 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3495 bond_unixctl_show(struct unixctl_conn *conn,
3496 const char *args, void *aux OVS_UNUSED)
3498 struct ds ds = DS_EMPTY_INITIALIZER;
3499 const struct port *port;
3500 struct iface *iface;
3502 port = bond_find(args);
3504 unixctl_command_reply(conn, 501, "no such bond");
3508 ds_put_format(&ds, "bond_mode: %s\n",
3509 bond_mode_to_string(port->bond_mode));
3512 ds_put_format(&ds, "lacp: %s\n",
3513 port->lacp_active ? "active" : "passive");
3515 ds_put_cstr(&ds, "lacp: off\n");
3518 if (port->bond_mode != BM_AB) {
3519 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3520 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3524 ds_put_format(&ds, "bond-detect-mode: %s\n",
3525 port->monitor ? "carrier" : "miimon");
3527 if (!port->monitor) {
3528 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3529 port->miimon_interval);
3532 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3533 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3535 if (port->bond_mode != BM_AB) {
3536 ds_put_format(&ds, "next rebalance: %lld ms\n",
3537 port->bond_next_rebalance - time_msec());
3540 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3541 struct bond_entry *be;
3545 ds_put_format(&ds, "\nslave %s: %s\n",
3546 iface->name, iface->enabled ? "enabled" : "disabled");
3547 if (iface == port->active_iface) {
3548 ds_put_cstr(&ds, "\tactive slave\n");
3550 if (iface->delay_expires != LLONG_MAX) {
3551 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3552 iface->enabled ? "downdelay" : "updelay",
3553 iface->delay_expires - time_msec());
3556 if (port->bond_mode == BM_AB) {
3561 memset(&flow, 0, sizeof flow);
3562 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3563 int hash = be - port->bond_hash;
3564 struct mac_entry *me;
3566 if (be->iface != iface) {
3570 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3571 hash, be->tx_bytes / 1024);
3573 if (port->bond_mode != BM_SLB) {
3578 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3582 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
3583 if (bond_hash_src(me->mac, me->vlan) == hash
3584 && me->port.p != port
3585 && choose_output_iface(port, &flow, me->vlan,
3587 && dp_ifidx == iface->dp_ifidx)
3589 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3590 ETH_ADDR_ARGS(me->mac));
3595 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3600 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3601 void *aux OVS_UNUSED)
3603 char *args = (char *) args_;
3604 char *save_ptr = NULL;
3605 char *bond_s, *hash_s, *slave_s;
3607 struct iface *iface;
3608 struct bond_entry *entry;
3611 bond_s = strtok_r(args, " ", &save_ptr);
3612 hash_s = strtok_r(NULL, " ", &save_ptr);
3613 slave_s = strtok_r(NULL, " ", &save_ptr);
3615 unixctl_command_reply(conn, 501,
3616 "usage: bond/migrate BOND HASH SLAVE");
3620 port = bond_find(bond_s);
3622 unixctl_command_reply(conn, 501, "no such bond");
3626 if (port->bond_mode != BM_SLB) {
3627 unixctl_command_reply(conn, 501, "not an SLB bond");
3631 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3632 hash = atoi(hash_s) & BOND_MASK;
3634 unixctl_command_reply(conn, 501, "bad hash");
3638 iface = port_lookup_iface(port, slave_s);
3640 unixctl_command_reply(conn, 501, "no such slave");
3644 if (!iface->enabled) {
3645 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3649 entry = &port->bond_hash[hash];
3650 ofproto_revalidate(port->bridge->ofproto, entry->tag);
3651 entry->iface = iface;
3652 entry->tag = tag_create_random();
3653 unixctl_command_reply(conn, 200, "migrated");
3657 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3658 void *aux OVS_UNUSED)
3660 char *args = (char *) args_;
3661 char *save_ptr = NULL;
3662 char *bond_s, *slave_s;
3664 struct iface *iface;
3666 bond_s = strtok_r(args, " ", &save_ptr);
3667 slave_s = strtok_r(NULL, " ", &save_ptr);
3669 unixctl_command_reply(conn, 501,
3670 "usage: bond/set-active-slave BOND SLAVE");
3674 port = bond_find(bond_s);
3676 unixctl_command_reply(conn, 501, "no such bond");
3680 iface = port_lookup_iface(port, slave_s);
3682 unixctl_command_reply(conn, 501, "no such slave");
3686 if (!iface->enabled) {
3687 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3691 if (port->active_iface != iface) {
3692 ofproto_revalidate(port->bridge->ofproto,
3693 port_get_active_iface_tag(port));
3694 port->active_iface = iface;
3695 VLOG_INFO("port %s: active interface is now %s",
3696 port->name, iface->name);
3697 bond_send_learning_packets(port);
3698 unixctl_command_reply(conn, 200, "done");
3700 unixctl_command_reply(conn, 200, "no change");
3705 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3707 char *args = (char *) args_;
3708 char *save_ptr = NULL;
3709 char *bond_s, *slave_s;
3711 struct iface *iface;
3713 bond_s = strtok_r(args, " ", &save_ptr);
3714 slave_s = strtok_r(NULL, " ", &save_ptr);
3716 unixctl_command_reply(conn, 501,
3717 "usage: bond/enable/disable-slave BOND SLAVE");
3721 port = bond_find(bond_s);
3723 unixctl_command_reply(conn, 501, "no such bond");
3727 iface = port_lookup_iface(port, slave_s);
3729 unixctl_command_reply(conn, 501, "no such slave");
3733 bond_enable_slave(iface, enable);
3734 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3738 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3739 void *aux OVS_UNUSED)
3741 enable_slave(conn, args, true);
3745 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3746 void *aux OVS_UNUSED)
3748 enable_slave(conn, args, false);
3752 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3753 void *aux OVS_UNUSED)
3755 char *args = (char *) args_;
3756 uint8_t mac[ETH_ADDR_LEN];
3760 char *mac_s, *vlan_s;
3761 char *save_ptr = NULL;
3763 mac_s = strtok_r(args, " ", &save_ptr);
3764 vlan_s = strtok_r(NULL, " ", &save_ptr);
3767 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3768 unixctl_command_reply(conn, 501, "invalid vlan");
3772 vlan = OFP_VLAN_NONE;
3775 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3776 == ETH_ADDR_SCAN_COUNT) {
3777 hash = bond_hash_src(mac, vlan);
3779 hash_cstr = xasprintf("%u", hash);
3780 unixctl_command_reply(conn, 200, hash_cstr);
3783 unixctl_command_reply(conn, 501, "invalid mac");
3790 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3791 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3792 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3793 unixctl_command_register("bond/set-active-slave",
3794 bond_unixctl_set_active_slave, NULL);
3795 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3797 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3799 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3802 /* Port functions. */
3805 lacp_send_pdu_cb(void *aux, const struct lacp_pdu *pdu)
3807 struct iface *iface = aux;
3808 uint8_t ea[ETH_ADDR_LEN];
3811 error = netdev_get_etheraddr(iface->netdev, ea);
3813 struct ofpbuf packet;
3815 ofpbuf_init(&packet, ETH_HEADER_LEN + LACP_PDU_LEN);
3816 compose_lacp_packet(&packet, ea, pdu);
3817 ofproto_send_packet(iface->port->bridge->ofproto,
3818 iface->dp_ifidx, 0, &packet);
3819 ofpbuf_uninit(&packet);
3821 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3822 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3823 "(%s)", iface->name, strerror(error));
3828 port_run(struct port *port)
3830 if (port->monitor) {
3833 /* Track carrier going up and down on interfaces. */
3834 while (!netdev_monitor_poll(port->monitor, &devname)) {
3835 struct iface *iface;
3837 iface = port_lookup_iface(port, devname);
3839 iface_update_carrier(iface);
3843 } else if (time_msec() >= port->miimon_next_update) {
3844 struct iface *iface;
3846 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3847 iface_update_carrier(iface);
3849 port->miimon_next_update = time_msec() + port->miimon_interval;
3853 struct iface *iface;
3855 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3856 lacp_slave_enable(port->lacp, iface, iface->enabled);
3859 lacp_run(port->lacp, lacp_send_pdu_cb);
3866 port_wait(struct port *port)
3868 if (port->monitor) {
3869 netdev_monitor_poll_wait(port->monitor);
3871 poll_timer_wait_until(port->miimon_next_update);
3875 lacp_wait(port->lacp);
3881 static struct port *
3882 port_create(struct bridge *br, const char *name)
3886 port = xzalloc(sizeof *port);
3889 port->trunks = NULL;
3890 port->name = xstrdup(name);
3891 port->active_iface = NULL;
3892 list_init(&port->ifaces);
3894 hmap_insert(&br->ports, &port->hmap_node, hash_string(port->name, 0));
3896 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3903 get_port_other_config(const struct ovsrec_port *port, const char *key,
3904 const char *default_value)
3908 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3910 return value ? value : default_value;
3914 get_interface_other_config(const struct ovsrec_interface *iface,
3915 const char *key, const char *default_value)
3919 value = get_ovsrec_key_value(&iface->header_,
3920 &ovsrec_interface_col_other_config, key);
3921 return value ? value : default_value;
3925 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3927 struct iface *iface, *next;
3928 struct shash new_ifaces;
3931 /* Collect list of new interfaces. */
3932 shash_init(&new_ifaces);
3933 for (i = 0; i < cfg->n_interfaces; i++) {
3934 const char *name = cfg->interfaces[i]->name;
3935 shash_add_once(&new_ifaces, name, NULL);
3938 /* Get rid of deleted interfaces. */
3939 LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) {
3940 if (!shash_find(&new_ifaces, iface->name)) {
3941 iface_destroy(iface);
3945 shash_destroy(&new_ifaces);
3948 /* Expires all MAC learning entries associated with 'port' and forces ofproto
3949 * to revalidate every flow. */
3951 port_flush_macs(struct port *port)
3953 struct bridge *br = port->bridge;
3954 struct mac_learning *ml = br->ml;
3955 struct mac_entry *mac, *next_mac;
3958 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
3959 if (mac->port.p == port) {
3960 mac_learning_expire(ml, mac);
3966 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3968 const char *detect_mode;
3969 struct shash new_ifaces;
3970 long long int next_rebalance, miimon_next_update, lacp_priority;
3971 bool need_flush = false;
3972 unsigned long *trunks;
3978 /* Update settings. */
3979 port->updelay = cfg->bond_updelay;
3980 if (port->updelay < 0) {
3983 port->downdelay = cfg->bond_downdelay;
3984 if (port->downdelay < 0) {
3985 port->downdelay = 0;
3987 port->bond_rebalance_interval = atoi(
3988 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3989 if (port->bond_rebalance_interval < 1000) {
3990 port->bond_rebalance_interval = 1000;
3992 next_rebalance = time_msec() + port->bond_rebalance_interval;
3993 if (port->bond_next_rebalance > next_rebalance) {
3994 port->bond_next_rebalance = next_rebalance;
3997 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
4000 netdev_monitor_destroy(port->monitor);
4001 port->monitor = NULL;
4003 if (strcmp(detect_mode, "miimon")) {
4004 port->monitor = netdev_monitor_create();
4006 if (strcmp(detect_mode, "carrier")) {
4007 VLOG_WARN("port %s: unsupported bond-detect-mode %s, "
4008 "defaulting to carrier", port->name, detect_mode);
4012 port->miimon_interval = atoi(
4013 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4014 if (port->miimon_interval < 100) {
4015 port->miimon_interval = 100;
4017 miimon_next_update = time_msec() + port->miimon_interval;
4018 if (port->miimon_next_update > miimon_next_update) {
4019 port->miimon_next_update = miimon_next_update;
4022 if (!port->cfg->bond_mode ||
4023 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4024 port->bond_mode = BM_SLB;
4025 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4026 port->bond_mode = BM_AB;
4027 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
4028 port->bond_mode = BM_TCP;
4030 port->bond_mode = BM_SLB;
4031 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4032 port->name, port->cfg->bond_mode,
4033 bond_mode_to_string(port->bond_mode));
4036 /* Add new interfaces and update 'cfg' member of existing ones. */
4037 shash_init(&new_ifaces);
4038 for (i = 0; i < cfg->n_interfaces; i++) {
4039 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4040 struct iface *iface;
4042 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
4043 VLOG_WARN("port %s: %s specified twice as port interface",
4044 port->name, if_cfg->name);
4045 iface_set_ofport(if_cfg, -1);
4049 iface = iface_lookup(port->bridge, if_cfg->name);
4051 if (iface->port != port) {
4052 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4054 port->bridge->name, if_cfg->name, iface->port->name);
4057 iface->cfg = if_cfg;
4059 iface = iface_create(port, if_cfg);
4062 /* Determine interface type. The local port always has type
4063 * "internal". Other ports take their type from the database and
4064 * default to "system" if none is specified. */
4065 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4066 : if_cfg->type[0] ? if_cfg->type
4070 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4073 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4074 iface->lacp_priority = UINT16_MAX;
4076 iface->lacp_priority = lacp_priority;
4079 shash_destroy(&new_ifaces);
4081 port->lacp_fast = !strcmp(get_port_other_config(cfg, "lacp-time", "slow"),
4085 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4087 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4088 /* Prefer bondable links if unspecified. */
4089 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4091 port->lacp_priority = lacp_priority;
4094 if (!port->cfg->lacp) {
4095 /* XXX when LACP implementation has been sufficiently tested, enable by
4096 * default and make active on bonded ports. */
4097 lacp_destroy(port->lacp);
4099 } else if (!strcmp(port->cfg->lacp, "off")) {
4100 lacp_destroy(port->lacp);
4102 } else if (!strcmp(port->cfg->lacp, "active")) {
4104 port->lacp = lacp_create();
4106 port->lacp_active = true;
4107 } else if (!strcmp(port->cfg->lacp, "passive")) {
4109 port->lacp = lacp_create();
4111 port->lacp_active = false;
4113 VLOG_WARN("port %s: unknown LACP mode %s",
4114 port->name, port->cfg->lacp);
4115 lacp_destroy(port->lacp);
4122 if (port->n_ifaces < 2) {
4124 if (vlan >= 0 && vlan <= 4095) {
4125 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4130 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4131 * they even work as-is. But they have not been tested. */
4132 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4136 if (port->vlan != vlan) {
4141 /* Get trunked VLANs. */
4143 if (vlan < 0 && cfg->n_trunks) {
4146 trunks = bitmap_allocate(4096);
4148 for (i = 0; i < cfg->n_trunks; i++) {
4149 int trunk = cfg->trunks[i];
4151 bitmap_set1(trunks, trunk);
4157 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4158 port->name, cfg->n_trunks);
4160 if (n_errors == cfg->n_trunks) {
4161 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4163 bitmap_free(trunks);
4166 } else if (vlan >= 0 && cfg->n_trunks) {
4167 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4171 ? port->trunks != NULL
4172 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4175 bitmap_free(port->trunks);
4176 port->trunks = trunks;
4179 port_flush_macs(port);
4184 port_destroy(struct port *port)
4187 struct bridge *br = port->bridge;
4188 struct iface *iface, *next;
4191 for (i = 0; i < MAX_MIRRORS; i++) {
4192 struct mirror *m = br->mirrors[i];
4193 if (m && m->out_port == port) {
4198 LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) {
4199 iface_destroy(iface);
4202 hmap_remove(&br->ports, &port->hmap_node);
4204 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4206 port_flush_macs(port);
4208 netdev_monitor_destroy(port->monitor);
4209 bitmap_free(port->trunks);
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 ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
4585 mon = iface->cfg->monitor;
4588 ofproto_iface_clear_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
4592 if (netdev_get_etheraddr(iface->netdev, ea)) {
4593 VLOG_WARN("interface %s: Failed to get ethernet address. "
4594 "Skipping Monitor.", iface->name);
4598 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4599 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4603 cfm.mpid = mon->mpid;
4604 cfm.interval = mon->interval ? *mon->interval : 1000;
4606 memcpy(cfm.eth_src, ea, sizeof cfm.eth_src);
4607 memcpy(cfm.maid, maid, sizeof cfm.maid);
4609 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4610 for(i = 0; i < mon->n_remote_mps; i++) {
4611 remote_mps[i] = mon->remote_mps[i]->mpid;
4614 ofproto_iface_set_cfm(iface->port->bridge->ofproto, iface->dp_ifidx,
4615 &cfm, remote_mps, mon->n_remote_mps);
4619 /* Read carrier or miimon status directly from 'iface''s netdev, according to
4620 * how 'iface''s port is configured.
4622 * Returns true if 'iface' is up, false otherwise. */
4624 iface_get_carrier(const struct iface *iface)
4626 return (iface->port->monitor
4627 ? netdev_get_carrier(iface->netdev)
4628 : netdev_get_miimon(iface->netdev));
4631 /* Port mirroring. */
4633 static struct mirror *
4634 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4638 for (i = 0; i < MAX_MIRRORS; i++) {
4639 struct mirror *m = br->mirrors[i];
4640 if (m && uuid_equals(uuid, &m->uuid)) {
4648 mirror_reconfigure(struct bridge *br)
4650 unsigned long *rspan_vlans;
4654 /* Get rid of deleted mirrors. */
4655 for (i = 0; i < MAX_MIRRORS; i++) {
4656 struct mirror *m = br->mirrors[i];
4658 const struct ovsdb_datum *mc;
4659 union ovsdb_atom atom;
4661 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4662 atom.uuid = br->mirrors[i]->uuid;
4663 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4669 /* Add new mirrors and reconfigure existing ones. */
4670 for (i = 0; i < br->cfg->n_mirrors; i++) {
4671 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4672 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4674 mirror_reconfigure_one(m, cfg);
4676 mirror_create(br, cfg);
4680 /* Update port reserved status. */
4681 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
4682 port->is_mirror_output_port = false;
4684 for (i = 0; i < MAX_MIRRORS; i++) {
4685 struct mirror *m = br->mirrors[i];
4686 if (m && m->out_port) {
4687 m->out_port->is_mirror_output_port = true;
4691 /* Update flooded vlans (for RSPAN). */
4693 if (br->cfg->n_flood_vlans) {
4694 rspan_vlans = bitmap_allocate(4096);
4696 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4697 int64_t vlan = br->cfg->flood_vlans[i];
4698 if (vlan >= 0 && vlan < 4096) {
4699 bitmap_set1(rspan_vlans, vlan);
4700 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4703 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4708 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4710 mac_learning_flush(br->ml);
4715 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4720 for (i = 0; ; i++) {
4721 if (i >= MAX_MIRRORS) {
4722 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4723 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4726 if (!br->mirrors[i]) {
4731 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4733 mac_learning_flush(br->ml);
4735 br->mirrors[i] = m = xzalloc(sizeof *m);
4738 m->name = xstrdup(cfg->name);
4739 shash_init(&m->src_ports);
4740 shash_init(&m->dst_ports);
4746 mirror_reconfigure_one(m, cfg);
4750 mirror_destroy(struct mirror *m)
4753 struct bridge *br = m->bridge;
4756 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
4757 port->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4758 port->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4761 shash_destroy(&m->src_ports);
4762 shash_destroy(&m->dst_ports);
4765 m->bridge->mirrors[m->idx] = NULL;
4770 mac_learning_flush(br->ml);
4775 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4776 struct shash *names)
4780 for (i = 0; i < n_ports; i++) {
4781 const char *name = ports[i]->name;
4782 if (port_lookup(m->bridge, name)) {
4783 shash_add_once(names, name, NULL);
4785 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4786 "port %s", m->bridge->name, m->name, name);
4792 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4798 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4800 for (i = 0; i < cfg->n_select_vlan; i++) {
4801 int64_t vlan = cfg->select_vlan[i];
4802 if (vlan < 0 || vlan > 4095) {
4803 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4804 m->bridge->name, m->name, vlan);
4806 (*vlans)[n_vlans++] = vlan;
4813 vlan_is_mirrored(const struct mirror *m, int vlan)
4817 for (i = 0; i < m->n_vlans; i++) {
4818 if (m->vlans[i] == vlan) {
4826 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4830 for (i = 0; i < m->n_vlans; i++) {
4831 if (port_trunks_vlan(p, m->vlans[i])) {
4839 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4841 struct shash src_ports, dst_ports;
4842 mirror_mask_t mirror_bit;
4843 struct port *out_port;
4850 if (strcmp(cfg->name, m->name)) {
4852 m->name = xstrdup(cfg->name);
4855 /* Get output port. */
4856 if (cfg->output_port) {
4857 out_port = port_lookup(m->bridge, cfg->output_port->name);
4859 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4860 m->bridge->name, m->name);
4866 if (cfg->output_vlan) {
4867 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4868 "output vlan; ignoring output vlan",
4869 m->bridge->name, m->name);
4871 } else if (cfg->output_vlan) {
4873 out_vlan = *cfg->output_vlan;
4875 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4876 m->bridge->name, m->name);
4881 shash_init(&src_ports);
4882 shash_init(&dst_ports);
4883 if (cfg->select_all) {
4884 HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) {
4885 shash_add_once(&src_ports, port->name, NULL);
4886 shash_add_once(&dst_ports, port->name, NULL);
4891 /* Get ports, and drop duplicates and ports that don't exist. */
4892 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4894 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4897 /* Get all the vlans, and drop duplicate and invalid vlans. */
4898 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4901 /* Update mirror data. */
4902 if (!shash_equal_keys(&m->src_ports, &src_ports)
4903 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4904 || m->n_vlans != n_vlans
4905 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4906 || m->out_port != out_port
4907 || m->out_vlan != out_vlan) {
4908 bridge_flush(m->bridge);
4909 mac_learning_flush(m->bridge->ml);
4911 shash_swap(&m->src_ports, &src_ports);
4912 shash_swap(&m->dst_ports, &dst_ports);
4915 m->n_vlans = n_vlans;
4916 m->out_port = out_port;
4917 m->out_vlan = out_vlan;
4920 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4921 HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) {
4922 if (shash_find(&m->src_ports, port->name)
4925 ? port_trunks_any_mirrored_vlan(m, port)
4926 : vlan_is_mirrored(m, port->vlan)))) {
4927 port->src_mirrors |= mirror_bit;
4929 port->src_mirrors &= ~mirror_bit;
4932 if (shash_find(&m->dst_ports, port->name)) {
4933 port->dst_mirrors |= mirror_bit;
4935 port->dst_mirrors &= ~mirror_bit;
4940 shash_destroy(&src_ports);
4941 shash_destroy(&dst_ports);