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 port *port; /* Containing port. */
102 size_t port_ifidx; /* Index within 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 int iface_idx; /* Index of assigned iface, or -1 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;
159 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
160 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
161 * NULL if all VLANs are trunked. */
162 const struct ovsrec_port *cfg;
166 struct netdev_monitor *monitor; /* Tracks carrier. NULL if miimon. */
167 long long int miimon_interval; /* Miimon status refresh interval. */
168 long long int miimon_next_update; /* Time of next miimon update. */
170 /* An ordinary bridge port has 1 interface.
171 * A bridge port for bonding has at least 2 interfaces. */
172 struct iface **ifaces;
173 size_t n_ifaces, allocated_ifaces;
176 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
177 struct iface *active_iface; /* iface on which bcasts accepted, or NULL. */
178 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
179 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
180 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
181 long long int bond_next_fake_iface_update; /* Time of next update. */
183 /* LACP information. */
184 struct lacp *lacp; /* LACP object. NULL if LACP is disabled. */
185 bool lacp_active; /* True if LACP is active */
186 bool lacp_fast; /* True if LACP is in fast mode. */
187 uint16_t lacp_priority; /* LACP system priority. */
189 /* SLB specific bonding info. */
190 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
191 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
192 long long int bond_next_rebalance; /* Next rebalancing time. */
194 /* Port mirroring info. */
195 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
196 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
197 bool is_mirror_output_port; /* Does port mirroring send frames here? */
201 struct list node; /* Node in global list of bridges. */
202 char *name; /* User-specified arbitrary name. */
203 struct mac_learning *ml; /* MAC learning table. */
204 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
205 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
206 const struct ovsrec_bridge *cfg;
208 /* OpenFlow switch processing. */
209 struct ofproto *ofproto; /* OpenFlow switch. */
211 /* Kernel datapath information. */
212 struct dpif *dpif; /* Datapath. */
213 struct hmap ifaces; /* Contains "struct iface"s. */
217 size_t n_ports, allocated_ports;
218 struct shash iface_by_name; /* "struct iface"s indexed by name. */
219 struct shash port_by_name; /* "struct port"s indexed by name. */
222 bool has_bonded_ports;
227 /* Port mirroring. */
228 struct mirror *mirrors[MAX_MIRRORS];
231 /* List of all bridges. */
232 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
234 /* OVSDB IDL used to obtain configuration. */
235 static struct ovsdb_idl *idl;
237 /* Each time this timer expires, the bridge fetches systems and interface
238 * statistics and pushes them into the database. */
239 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
240 static long long int stats_timer = LLONG_MIN;
242 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
243 static void bridge_destroy(struct bridge *);
244 static struct bridge *bridge_lookup(const char *name);
245 static unixctl_cb_func bridge_unixctl_dump_flows;
246 static unixctl_cb_func bridge_unixctl_reconnect;
247 static int bridge_run_one(struct bridge *);
248 static size_t bridge_get_controllers(const struct bridge *br,
249 struct ovsrec_controller ***controllersp);
250 static void bridge_reconfigure_one(struct bridge *);
251 static void bridge_reconfigure_remotes(struct bridge *,
252 const struct sockaddr_in *managers,
254 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
255 static void bridge_fetch_dp_ifaces(struct bridge *);
256 static void bridge_flush(struct bridge *);
257 static void bridge_pick_local_hw_addr(struct bridge *,
258 uint8_t ea[ETH_ADDR_LEN],
259 struct iface **hw_addr_iface);
260 static uint64_t bridge_pick_datapath_id(struct bridge *,
261 const uint8_t bridge_ea[ETH_ADDR_LEN],
262 struct iface *hw_addr_iface);
263 static uint64_t dpid_from_hash(const void *, size_t nbytes);
265 static unixctl_cb_func bridge_unixctl_fdb_show;
266 static unixctl_cb_func qos_unixctl_show;
268 static void bond_init(void);
269 static void bond_run(struct port *);
270 static void bond_wait(struct port *);
271 static void bond_rebalance_port(struct port *);
272 static void bond_send_learning_packets(struct port *);
273 static void bond_enable_slave(struct iface *iface, bool enable);
275 static void port_run(struct port *);
276 static void port_wait(struct port *);
277 static struct port *port_create(struct bridge *, const char *name);
278 static void port_reconfigure(struct port *, const struct ovsrec_port *);
279 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
280 static void port_destroy(struct port *);
281 static struct port *port_lookup(const struct bridge *, const char *name);
282 static struct iface *port_lookup_iface(const struct port *, const char *name);
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 *,
468 for (i = 0; i < br->n_ports; ) {
469 struct port *port = br->ports[i];
470 for (j = 0; j < port->n_ifaces; ) {
471 struct iface *iface = port->ifaces[j];
472 if (cb(br, iface, aux)) {
475 iface_set_ofport(iface->cfg, -1);
476 iface_destroy(iface);
480 if (port->n_ifaces) {
483 VLOG_WARN("%s port has no interfaces, dropping", port->name);
489 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
490 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
491 * responsible for freeing '*managersp' (with free()).
493 * You may be asking yourself "why does ovs-vswitchd care?", because
494 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
495 * should not be and in fact is not directly involved in that. But
496 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
497 * it has to tell in-band control where the managers are to enable that.
498 * (Thus, only managers connected in-band are collected.)
501 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
502 struct sockaddr_in **managersp, size_t *n_managersp)
504 struct sockaddr_in *managers = NULL;
505 size_t n_managers = 0;
506 struct shash targets;
509 /* Collect all of the potential targets from the "targets" columns of the
510 * rows pointed to by "manager_options", excluding any that are
512 shash_init(&targets);
513 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
514 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
516 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
517 shash_find_and_delete(&targets, m->target);
519 shash_add_once(&targets, m->target, NULL);
523 /* Now extract the targets' IP addresses. */
524 if (!shash_is_empty(&targets)) {
525 struct shash_node *node;
527 managers = xmalloc(shash_count(&targets) * sizeof *managers);
528 SHASH_FOR_EACH (node, &targets) {
529 const char *target = node->name;
530 struct sockaddr_in *sin = &managers[n_managers];
532 if ((!strncmp(target, "tcp:", 4)
533 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
534 (!strncmp(target, "ssl:", 4)
535 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
540 shash_destroy(&targets);
542 *managersp = managers;
543 *n_managersp = n_managers;
547 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
549 struct shash old_br, new_br;
550 struct shash_node *node;
551 struct bridge *br, *next;
552 struct sockaddr_in *managers;
555 int sflow_bridge_number;
557 COVERAGE_INC(bridge_reconfigure);
559 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
561 /* Collect old and new bridges. */
564 LIST_FOR_EACH (br, node, &all_bridges) {
565 shash_add(&old_br, br->name, br);
567 for (i = 0; i < ovs_cfg->n_bridges; i++) {
568 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
569 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
570 VLOG_WARN("more than one bridge named %s", br_cfg->name);
574 /* Get rid of deleted bridges and add new bridges. */
575 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
576 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
583 SHASH_FOR_EACH (node, &new_br) {
584 const char *br_name = node->name;
585 const struct ovsrec_bridge *br_cfg = node->data;
586 br = shash_find_data(&old_br, br_name);
588 /* If the bridge datapath type has changed, we need to tear it
589 * down and recreate. */
590 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
592 bridge_create(br_cfg);
595 bridge_create(br_cfg);
598 shash_destroy(&old_br);
599 shash_destroy(&new_br);
601 /* Reconfigure all bridges. */
602 LIST_FOR_EACH (br, node, &all_bridges) {
603 bridge_reconfigure_one(br);
606 /* Add and delete ports on all datapaths.
608 * The kernel will reject any attempt to add a given port to a datapath if
609 * that port already belongs to a different datapath, so we must do all
610 * port deletions before any port additions. */
611 LIST_FOR_EACH (br, node, &all_bridges) {
612 struct dpif_port_dump dump;
613 struct shash want_ifaces;
614 struct dpif_port dpif_port;
616 bridge_get_all_ifaces(br, &want_ifaces);
617 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
618 if (!shash_find(&want_ifaces, dpif_port.name)
619 && strcmp(dpif_port.name, br->name)) {
620 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
622 VLOG_WARN("failed to remove %s interface from %s: %s",
623 dpif_port.name, dpif_name(br->dpif),
628 shash_destroy(&want_ifaces);
630 LIST_FOR_EACH (br, node, &all_bridges) {
631 struct shash cur_ifaces, want_ifaces;
632 struct dpif_port_dump dump;
633 struct dpif_port dpif_port;
635 /* Get the set of interfaces currently in this datapath. */
636 shash_init(&cur_ifaces);
637 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
638 struct dpif_port *port_info = xmalloc(sizeof *port_info);
639 dpif_port_clone(port_info, &dpif_port);
640 shash_add(&cur_ifaces, dpif_port.name, port_info);
643 /* Get the set of interfaces we want on this datapath. */
644 bridge_get_all_ifaces(br, &want_ifaces);
646 hmap_clear(&br->ifaces);
647 SHASH_FOR_EACH (node, &want_ifaces) {
648 const char *if_name = node->name;
649 struct iface *iface = node->data;
650 struct dpif_port *dpif_port;
654 type = iface ? iface->type : "internal";
655 dpif_port = shash_find_data(&cur_ifaces, if_name);
657 /* If we have a port or a netdev already, and it's not the type we
658 * want, then delete the port (if any) and close the netdev (if
660 if ((dpif_port && strcmp(dpif_port->type, type))
661 || (iface && iface->netdev
662 && strcmp(type, netdev_get_type(iface->netdev)))) {
664 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
671 netdev_close(iface->netdev);
672 iface->netdev = NULL;
676 /* If the port doesn't exist or we don't have the netdev open,
677 * we need to do more work. */
678 if (!dpif_port || (iface && !iface->netdev)) {
679 struct netdev_options options;
680 struct netdev *netdev;
683 /* First open the network device. */
684 options.name = if_name;
686 options.args = &args;
687 options.ethertype = NETDEV_ETH_TYPE_NONE;
691 shash_from_ovs_idl_map(iface->cfg->key_options,
692 iface->cfg->value_options,
693 iface->cfg->n_options, &args);
695 error = netdev_open(&options, &netdev);
696 shash_destroy(&args);
699 VLOG_WARN("could not open network device %s (%s)",
700 if_name, strerror(error));
704 /* Then add the port if we haven't already. */
706 error = dpif_port_add(br->dpif, netdev, NULL);
708 netdev_close(netdev);
709 if (error == EFBIG) {
710 VLOG_ERR("ran out of valid port numbers on %s",
711 dpif_name(br->dpif));
714 VLOG_WARN("failed to add %s interface to %s: %s",
715 if_name, dpif_name(br->dpif),
722 /* Update 'iface'. */
724 iface->netdev = netdev;
725 iface->enabled = iface_get_carrier(iface);
726 iface->up = iface->enabled;
728 } else if (iface && iface->netdev) {
732 shash_from_ovs_idl_map(iface->cfg->key_options,
733 iface->cfg->value_options,
734 iface->cfg->n_options, &args);
735 netdev_set_config(iface->netdev, &args);
736 shash_destroy(&args);
739 shash_destroy(&want_ifaces);
741 SHASH_FOR_EACH (node, &cur_ifaces) {
742 struct dpif_port *port_info = node->data;
743 dpif_port_destroy(port_info);
746 shash_destroy(&cur_ifaces);
748 sflow_bridge_number = 0;
749 LIST_FOR_EACH (br, node, &all_bridges) {
752 struct iface *local_iface;
753 struct iface *hw_addr_iface;
756 bridge_fetch_dp_ifaces(br);
758 iterate_and_prune_ifaces(br, check_iface, NULL);
760 /* Pick local port hardware address, datapath ID. */
761 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
762 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
764 int error = netdev_set_etheraddr(local_iface->netdev, ea);
766 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
767 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
768 "Ethernet address: %s",
769 br->name, strerror(error));
772 memcpy(br->ea, ea, ETH_ADDR_LEN);
774 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
775 ofproto_set_datapath_id(br->ofproto, dpid);
777 dpid_string = xasprintf("%016"PRIx64, dpid);
778 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
781 /* Set NetFlow configuration on this bridge. */
782 if (br->cfg->netflow) {
783 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
784 struct netflow_options opts;
786 memset(&opts, 0, sizeof opts);
788 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
789 if (nf_cfg->engine_type) {
790 opts.engine_type = *nf_cfg->engine_type;
792 if (nf_cfg->engine_id) {
793 opts.engine_id = *nf_cfg->engine_id;
796 opts.active_timeout = nf_cfg->active_timeout;
797 if (!opts.active_timeout) {
798 opts.active_timeout = -1;
799 } else if (opts.active_timeout < 0) {
800 VLOG_WARN("bridge %s: active timeout interval set to negative "
801 "value, using default instead (%d seconds)", br->name,
802 NF_ACTIVE_TIMEOUT_DEFAULT);
803 opts.active_timeout = -1;
806 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
807 if (opts.add_id_to_iface) {
808 if (opts.engine_id > 0x7f) {
809 VLOG_WARN("bridge %s: netflow port mangling may conflict "
810 "with another vswitch, choose an engine id less "
811 "than 128", br->name);
813 if (br->n_ports > 508) {
814 VLOG_WARN("bridge %s: netflow port mangling will conflict "
815 "with another port when more than 508 ports are "
820 opts.collectors.n = nf_cfg->n_targets;
821 opts.collectors.names = nf_cfg->targets;
822 if (ofproto_set_netflow(br->ofproto, &opts)) {
823 VLOG_ERR("bridge %s: problem setting netflow collectors",
827 ofproto_set_netflow(br->ofproto, NULL);
830 /* Set sFlow configuration on this bridge. */
831 if (br->cfg->sflow) {
832 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
833 struct ovsrec_controller **controllers;
834 struct ofproto_sflow_options oso;
835 size_t n_controllers;
837 memset(&oso, 0, sizeof oso);
839 oso.targets.n = sflow_cfg->n_targets;
840 oso.targets.names = sflow_cfg->targets;
842 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
843 if (sflow_cfg->sampling) {
844 oso.sampling_rate = *sflow_cfg->sampling;
847 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
848 if (sflow_cfg->polling) {
849 oso.polling_interval = *sflow_cfg->polling;
852 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
853 if (sflow_cfg->header) {
854 oso.header_len = *sflow_cfg->header;
857 oso.sub_id = sflow_bridge_number++;
858 oso.agent_device = sflow_cfg->agent;
860 oso.control_ip = NULL;
861 n_controllers = bridge_get_controllers(br, &controllers);
862 for (i = 0; i < n_controllers; i++) {
863 if (controllers[i]->local_ip) {
864 oso.control_ip = controllers[i]->local_ip;
868 ofproto_set_sflow(br->ofproto, &oso);
870 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
872 ofproto_set_sflow(br->ofproto, NULL);
875 /* Update the controller and related settings. It would be more
876 * straightforward to call this from bridge_reconfigure_one(), but we
877 * can't do it there for two reasons. First, and most importantly, at
878 * that point we don't know the dp_ifidx of any interfaces that have
879 * been added to the bridge (because we haven't actually added them to
880 * the datapath). Second, at that point we haven't set the datapath ID
881 * yet; when a controller is configured, resetting the datapath ID will
882 * immediately disconnect from the controller, so it's better to set
883 * the datapath ID before the controller. */
884 bridge_reconfigure_remotes(br, managers, n_managers);
886 LIST_FOR_EACH (br, node, &all_bridges) {
887 for (i = 0; i < br->n_ports; i++) {
888 struct port *port = br->ports[i];
892 for (j = 0; j < port->n_ifaces; j++) {
893 netdev_monitor_add(port->monitor, port->ifaces[j]->netdev);
896 port->miimon_next_update = 0;
899 port_update_lacp(port);
900 port_update_bonding(port);
902 for (j = 0; j < port->n_ifaces; j++) {
903 iface_update_qos(port->ifaces[j], port->cfg->qos);
907 LIST_FOR_EACH (br, node, &all_bridges) {
908 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
911 LIST_FOR_EACH (br, node, &all_bridges) {
913 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
914 iface_update_cfm(iface);
920 /* ovs-vswitchd has completed initialization, so allow the process that
921 * forked us to exit successfully. */
922 daemonize_complete();
926 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
927 const struct ovsdb_idl_column *column,
930 const struct ovsdb_datum *datum;
931 union ovsdb_atom atom;
934 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
935 atom.string = (char *) key;
936 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
937 return idx == UINT_MAX ? NULL : datum->values[idx].string;
941 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
943 return get_ovsrec_key_value(&br_cfg->header_,
944 &ovsrec_bridge_col_other_config, key);
948 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
949 struct iface **hw_addr_iface)
955 *hw_addr_iface = NULL;
957 /* Did the user request a particular MAC? */
958 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
959 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
960 if (eth_addr_is_multicast(ea)) {
961 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
962 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
963 } else if (eth_addr_is_zero(ea)) {
964 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
970 /* Otherwise choose the minimum non-local MAC address among all of the
972 memset(ea, 0xff, ETH_ADDR_LEN);
973 for (i = 0; i < br->n_ports; i++) {
974 struct port *port = br->ports[i];
975 uint8_t iface_ea[ETH_ADDR_LEN];
978 /* Mirror output ports don't participate. */
979 if (port->is_mirror_output_port) {
983 /* Choose the MAC address to represent the port. */
984 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
985 /* Find the interface with this Ethernet address (if any) so that
986 * we can provide the correct devname to the caller. */
988 for (j = 0; j < port->n_ifaces; j++) {
989 struct iface *candidate = port->ifaces[j];
990 uint8_t candidate_ea[ETH_ADDR_LEN];
991 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
992 && eth_addr_equals(iface_ea, candidate_ea)) {
997 /* Choose the interface whose MAC address will represent the port.
998 * The Linux kernel bonding code always chooses the MAC address of
999 * the first slave added to a bond, and the Fedora networking
1000 * scripts always add slaves to a bond in alphabetical order, so
1001 * for compatibility we choose the interface with the name that is
1002 * first in alphabetical order. */
1003 iface = port->ifaces[0];
1004 for (j = 1; j < port->n_ifaces; j++) {
1005 struct iface *candidate = port->ifaces[j];
1006 if (strcmp(candidate->name, iface->name) < 0) {
1011 /* The local port doesn't count (since we're trying to choose its
1012 * MAC address anyway). */
1013 if (iface->dp_ifidx == ODPP_LOCAL) {
1018 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1020 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1021 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1022 iface->name, strerror(error));
1027 /* Compare against our current choice. */
1028 if (!eth_addr_is_multicast(iface_ea) &&
1029 !eth_addr_is_local(iface_ea) &&
1030 !eth_addr_is_reserved(iface_ea) &&
1031 !eth_addr_is_zero(iface_ea) &&
1032 eth_addr_compare_3way(iface_ea, ea) < 0)
1034 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1035 *hw_addr_iface = iface;
1038 if (eth_addr_is_multicast(ea)) {
1039 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1040 *hw_addr_iface = NULL;
1041 VLOG_WARN("bridge %s: using default bridge Ethernet "
1042 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1044 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1045 br->name, ETH_ADDR_ARGS(ea));
1049 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1050 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1051 * an interface on 'br', then that interface must be passed in as
1052 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1053 * 'hw_addr_iface' must be passed in as a null pointer. */
1055 bridge_pick_datapath_id(struct bridge *br,
1056 const uint8_t bridge_ea[ETH_ADDR_LEN],
1057 struct iface *hw_addr_iface)
1060 * The procedure for choosing a bridge MAC address will, in the most
1061 * ordinary case, also choose a unique MAC that we can use as a datapath
1062 * ID. In some special cases, though, multiple bridges will end up with
1063 * the same MAC address. This is OK for the bridges, but it will confuse
1064 * the OpenFlow controller, because each datapath needs a unique datapath
1067 * Datapath IDs must be unique. It is also very desirable that they be
1068 * stable from one run to the next, so that policy set on a datapath
1071 const char *datapath_id;
1074 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1075 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1079 if (hw_addr_iface) {
1081 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1083 * A bridge whose MAC address is taken from a VLAN network device
1084 * (that is, a network device created with vconfig(8) or similar
1085 * tool) will have the same MAC address as a bridge on the VLAN
1086 * device's physical network device.
1088 * Handle this case by hashing the physical network device MAC
1089 * along with the VLAN identifier.
1091 uint8_t buf[ETH_ADDR_LEN + 2];
1092 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1093 buf[ETH_ADDR_LEN] = vlan >> 8;
1094 buf[ETH_ADDR_LEN + 1] = vlan;
1095 return dpid_from_hash(buf, sizeof buf);
1098 * Assume that this bridge's MAC address is unique, since it
1099 * doesn't fit any of the cases we handle specially.
1104 * A purely internal bridge, that is, one that has no non-virtual
1105 * network devices on it at all, is more difficult because it has no
1106 * natural unique identifier at all.
1108 * When the host is a XenServer, we handle this case by hashing the
1109 * host's UUID with the name of the bridge. Names of bridges are
1110 * persistent across XenServer reboots, although they can be reused if
1111 * an internal network is destroyed and then a new one is later
1112 * created, so this is fairly effective.
1114 * When the host is not a XenServer, we punt by using a random MAC
1115 * address on each run.
1117 const char *host_uuid = xenserver_get_host_uuid();
1119 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1120 dpid = dpid_from_hash(combined, strlen(combined));
1126 return eth_addr_to_uint64(bridge_ea);
1130 dpid_from_hash(const void *data, size_t n)
1132 uint8_t hash[SHA1_DIGEST_SIZE];
1134 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1135 sha1_bytes(data, n, hash);
1136 eth_addr_mark_random(hash);
1137 return eth_addr_to_uint64(hash);
1141 iface_refresh_status(struct iface *iface)
1145 enum netdev_flags flags;
1154 if (!netdev_get_status(iface->netdev, &sh)) {
1156 char **keys, **values;
1158 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1159 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1164 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1167 shash_destroy_free_data(&sh);
1169 error = netdev_get_flags(iface->netdev, &flags);
1171 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1174 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1177 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1179 ovsrec_interface_set_duplex(iface->cfg,
1180 netdev_features_is_full_duplex(current)
1182 /* warning: uint64_t -> int64_t conversion */
1183 bps = netdev_features_to_bps(current);
1184 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1187 ovsrec_interface_set_duplex(iface->cfg, NULL);
1188 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1192 ovsrec_interface_set_link_state(iface->cfg,
1193 iface_get_carrier(iface) ? "up" : "down");
1195 error = netdev_get_mtu(iface->netdev, &mtu);
1196 if (!error && mtu != INT_MAX) {
1198 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1201 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1206 iface_refresh_cfm_stats(struct iface *iface)
1208 const struct ovsrec_monitor *mon;
1209 const struct cfm *cfm;
1212 mon = iface->cfg->monitor;
1213 cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
1219 for (i = 0; i < mon->n_remote_mps; i++) {
1220 const struct ovsrec_maintenance_point *mp;
1221 const struct remote_mp *rmp;
1223 mp = mon->remote_mps[i];
1224 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1226 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1229 if (hmap_is_empty(&cfm->x_remote_mps)) {
1230 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1233 struct remote_mp *rmp;
1234 int64_t *x_remote_mps;
1236 length = hmap_count(&cfm->x_remote_mps);
1237 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1240 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1241 x_remote_mps[i++] = rmp->mpid;
1244 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1248 if (hmap_is_empty(&cfm->x_remote_maids)) {
1249 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1252 char **x_remote_maids;
1253 struct remote_maid *rmaid;
1255 length = hmap_count(&cfm->x_remote_maids);
1256 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1259 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1262 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1264 for (j = 0; j < CCM_MAID_LEN; j++) {
1265 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1270 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1272 for (i = 0; i < length; i++) {
1273 free(x_remote_maids[i]);
1275 free(x_remote_maids);
1278 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1282 iface_refresh_stats(struct iface *iface)
1288 static const struct iface_stat iface_stats[] = {
1289 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1290 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1291 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1292 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1293 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1294 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1295 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1296 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1297 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1298 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1299 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1300 { "collisions", offsetof(struct netdev_stats, collisions) },
1302 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1303 const struct iface_stat *s;
1305 char *keys[N_STATS];
1306 int64_t values[N_STATS];
1309 struct netdev_stats stats;
1311 /* Intentionally ignore return value, since errors will set 'stats' to
1312 * all-1s, and we will deal with that correctly below. */
1313 netdev_get_stats(iface->netdev, &stats);
1316 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1317 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1318 if (value != UINT64_MAX) {
1325 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1329 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1331 struct ovsdb_datum datum;
1335 get_system_stats(&stats);
1337 ovsdb_datum_from_shash(&datum, &stats);
1338 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1342 static inline const char *
1343 nx_role_to_str(enum nx_role role)
1348 case NX_ROLE_MASTER:
1353 return "*** INVALID ROLE ***";
1358 bridge_refresh_controller_status(const struct bridge *br)
1361 const struct ovsrec_controller *cfg;
1363 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1365 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1366 struct ofproto_controller_info *cinfo =
1367 shash_find_data(&info, cfg->target);
1370 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1371 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1372 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1373 (char **) cinfo->pairs.values,
1376 ovsrec_controller_set_is_connected(cfg, false);
1377 ovsrec_controller_set_role(cfg, NULL);
1378 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1382 ofproto_free_ofproto_controller_info(&info);
1388 const struct ovsrec_open_vswitch *cfg;
1390 bool datapath_destroyed;
1391 bool database_changed;
1394 /* Let each bridge do the work that it needs to do. */
1395 datapath_destroyed = false;
1396 LIST_FOR_EACH (br, node, &all_bridges) {
1397 int error = bridge_run_one(br);
1399 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1400 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1401 "forcing reconfiguration", br->name);
1402 datapath_destroyed = true;
1406 /* (Re)configure if necessary. */
1407 database_changed = ovsdb_idl_run(idl);
1408 cfg = ovsrec_open_vswitch_first(idl);
1410 /* Re-configure SSL. We do this on every trip through the main loop,
1411 * instead of just when the database changes, because the contents of the
1412 * key and certificate files can change without the database changing.
1414 * We do this before bridge_reconfigure() because that function might
1415 * initiate SSL connections and thus requires SSL to be configured. */
1416 if (cfg && cfg->ssl) {
1417 const struct ovsrec_ssl *ssl = cfg->ssl;
1419 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1420 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1423 if (database_changed || datapath_destroyed) {
1425 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1427 bridge_configure_once(cfg);
1428 bridge_reconfigure(cfg);
1430 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1431 ovsdb_idl_txn_commit(txn);
1432 ovsdb_idl_txn_destroy(txn); /* XXX */
1434 /* We still need to reconfigure to avoid dangling pointers to
1435 * now-destroyed ovsrec structures inside bridge data. */
1436 static const struct ovsrec_open_vswitch null_cfg;
1438 bridge_reconfigure(&null_cfg);
1442 /* Refresh system and interface stats if necessary. */
1443 if (time_msec() >= stats_timer) {
1445 struct ovsdb_idl_txn *txn;
1447 txn = ovsdb_idl_txn_create(idl);
1448 LIST_FOR_EACH (br, node, &all_bridges) {
1451 for (i = 0; i < br->n_ports; i++) {
1452 struct port *port = br->ports[i];
1455 for (j = 0; j < port->n_ifaces; j++) {
1456 struct iface *iface = port->ifaces[j];
1457 iface_refresh_stats(iface);
1458 iface_refresh_cfm_stats(iface);
1459 iface_refresh_status(iface);
1462 bridge_refresh_controller_status(br);
1464 refresh_system_stats(cfg);
1465 ovsdb_idl_txn_commit(txn);
1466 ovsdb_idl_txn_destroy(txn); /* XXX */
1469 stats_timer = time_msec() + STATS_INTERVAL;
1478 LIST_FOR_EACH (br, node, &all_bridges) {
1481 ofproto_wait(br->ofproto);
1482 if (ofproto_has_primary_controller(br->ofproto)) {
1486 mac_learning_wait(br->ml);
1488 for (i = 0; i < br->n_ports; i++) {
1489 port_wait(br->ports[i]);
1492 ovsdb_idl_wait(idl);
1493 poll_timer_wait_until(stats_timer);
1496 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1497 * configuration changes. */
1499 bridge_flush(struct bridge *br)
1501 COVERAGE_INC(bridge_flush);
1503 mac_learning_flush(br->ml);
1506 /* Bridge unixctl user interface functions. */
1508 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1509 const char *args, void *aux OVS_UNUSED)
1511 struct ds ds = DS_EMPTY_INITIALIZER;
1512 const struct bridge *br;
1513 const struct mac_entry *e;
1515 br = bridge_lookup(args);
1517 unixctl_command_reply(conn, 501, "no such bridge");
1521 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1522 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1523 if (e->port < 0 || e->port >= br->n_ports) {
1526 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1527 br->ports[e->port]->ifaces[0]->dp_ifidx,
1528 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1530 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1534 /* QoS unixctl user interface functions. */
1536 struct qos_unixctl_show_cbdata {
1538 struct iface *iface;
1542 qos_unixctl_show_cb(unsigned int queue_id,
1543 const struct shash *details,
1546 struct qos_unixctl_show_cbdata *data = aux;
1547 struct ds *ds = data->ds;
1548 struct iface *iface = data->iface;
1549 struct netdev_queue_stats stats;
1550 struct shash_node *node;
1553 ds_put_cstr(ds, "\n");
1555 ds_put_format(ds, "Queue %u:\n", queue_id);
1557 ds_put_cstr(ds, "Default:\n");
1560 SHASH_FOR_EACH (node, details) {
1561 ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data);
1564 error = netdev_get_queue_stats(iface->netdev, queue_id, &stats);
1566 if (stats.tx_packets != UINT64_MAX) {
1567 ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets);
1570 if (stats.tx_bytes != UINT64_MAX) {
1571 ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes);
1574 if (stats.tx_errors != UINT64_MAX) {
1575 ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors);
1578 ds_put_format(ds, "\tFailed to get statistics for queue %u: %s",
1579 queue_id, strerror(error));
1584 qos_unixctl_show(struct unixctl_conn *conn,
1585 const char *args, void *aux OVS_UNUSED)
1587 struct ds ds = DS_EMPTY_INITIALIZER;
1588 struct shash sh = SHASH_INITIALIZER(&sh);
1589 struct iface *iface;
1591 struct shash_node *node;
1592 struct qos_unixctl_show_cbdata data;
1595 iface = iface_find(args);
1597 unixctl_command_reply(conn, 501, "no such interface");
1601 netdev_get_qos(iface->netdev, &type, &sh);
1603 if (*type != '\0') {
1604 ds_put_format(&ds, "QoS: %s %s\n", iface->name, type);
1606 SHASH_FOR_EACH (node, &sh) {
1607 ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data);
1612 error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data);
1615 ds_put_format(&ds, "failed to dump queues: %s", strerror(error));
1617 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1619 ds_put_format(&ds, "QoS not configured on %s\n", iface->name);
1620 unixctl_command_reply(conn, 501, ds_cstr(&ds));
1623 shash_destroy_free_data(&sh);
1627 /* Bridge reconfiguration functions. */
1628 static struct bridge *
1629 bridge_create(const struct ovsrec_bridge *br_cfg)
1634 assert(!bridge_lookup(br_cfg->name));
1635 br = xzalloc(sizeof *br);
1637 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1643 dpif_flow_flush(br->dpif);
1645 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1648 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1650 dpif_delete(br->dpif);
1651 dpif_close(br->dpif);
1656 br->name = xstrdup(br_cfg->name);
1658 br->ml = mac_learning_create();
1659 eth_addr_nicira_random(br->default_ea);
1661 hmap_init(&br->ifaces);
1663 shash_init(&br->port_by_name);
1664 shash_init(&br->iface_by_name);
1668 list_push_back(&all_bridges, &br->node);
1670 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1676 bridge_destroy(struct bridge *br)
1681 while (br->n_ports > 0) {
1682 port_destroy(br->ports[br->n_ports - 1]);
1684 list_remove(&br->node);
1685 error = dpif_delete(br->dpif);
1686 if (error && error != ENOENT) {
1687 VLOG_ERR("failed to delete %s: %s",
1688 dpif_name(br->dpif), strerror(error));
1690 dpif_close(br->dpif);
1691 ofproto_destroy(br->ofproto);
1692 mac_learning_destroy(br->ml);
1693 hmap_destroy(&br->ifaces);
1694 shash_destroy(&br->port_by_name);
1695 shash_destroy(&br->iface_by_name);
1702 static struct bridge *
1703 bridge_lookup(const char *name)
1707 LIST_FOR_EACH (br, node, &all_bridges) {
1708 if (!strcmp(br->name, name)) {
1715 /* Handle requests for a listing of all flows known by the OpenFlow
1716 * stack, including those normally hidden. */
1718 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1719 const char *args, void *aux OVS_UNUSED)
1724 br = bridge_lookup(args);
1726 unixctl_command_reply(conn, 501, "Unknown bridge");
1731 ofproto_get_all_flows(br->ofproto, &results);
1733 unixctl_command_reply(conn, 200, ds_cstr(&results));
1734 ds_destroy(&results);
1737 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1738 * connections and reconnect. If BRIDGE is not specified, then all bridges
1739 * drop their controller connections and reconnect. */
1741 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1742 const char *args, void *aux OVS_UNUSED)
1745 if (args[0] != '\0') {
1746 br = bridge_lookup(args);
1748 unixctl_command_reply(conn, 501, "Unknown bridge");
1751 ofproto_reconnect_controllers(br->ofproto);
1753 LIST_FOR_EACH (br, node, &all_bridges) {
1754 ofproto_reconnect_controllers(br->ofproto);
1757 unixctl_command_reply(conn, 200, NULL);
1761 bridge_run_one(struct bridge *br)
1766 error = ofproto_run1(br->ofproto);
1771 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1773 for (i = 0; i < br->n_ports; i++) {
1774 port_run(br->ports[i]);
1777 error = ofproto_run2(br->ofproto, br->flush);
1784 bridge_get_controllers(const struct bridge *br,
1785 struct ovsrec_controller ***controllersp)
1787 struct ovsrec_controller **controllers;
1788 size_t n_controllers;
1790 controllers = br->cfg->controller;
1791 n_controllers = br->cfg->n_controller;
1793 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1799 *controllersp = controllers;
1801 return n_controllers;
1805 bridge_reconfigure_one(struct bridge *br)
1807 struct shash old_ports, new_ports;
1808 struct svec snoops, old_snoops;
1809 struct shash_node *node;
1810 enum ofproto_fail_mode fail_mode;
1813 /* Collect old ports. */
1814 shash_init(&old_ports);
1815 for (i = 0; i < br->n_ports; i++) {
1816 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1819 /* Collect new ports. */
1820 shash_init(&new_ports);
1821 for (i = 0; i < br->cfg->n_ports; i++) {
1822 const char *name = br->cfg->ports[i]->name;
1823 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1824 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1829 /* If we have a controller, then we need a local port. Complain if the
1830 * user didn't specify one.
1832 * XXX perhaps we should synthesize a port ourselves in this case. */
1833 if (bridge_get_controllers(br, NULL)) {
1834 char local_name[IF_NAMESIZE];
1837 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1838 local_name, sizeof local_name);
1839 if (!error && !shash_find(&new_ports, local_name)) {
1840 VLOG_WARN("bridge %s: controller specified but no local port "
1841 "(port named %s) defined",
1842 br->name, local_name);
1846 /* Get rid of deleted ports.
1847 * Get rid of deleted interfaces on ports that still exist. */
1848 SHASH_FOR_EACH (node, &old_ports) {
1849 struct port *port = node->data;
1850 const struct ovsrec_port *port_cfg;
1852 port_cfg = shash_find_data(&new_ports, node->name);
1856 port_del_ifaces(port, port_cfg);
1860 /* Create new ports.
1861 * Add new interfaces to existing ports.
1862 * Reconfigure existing ports. */
1863 SHASH_FOR_EACH (node, &new_ports) {
1864 struct port *port = shash_find_data(&old_ports, node->name);
1866 port = port_create(br, node->name);
1869 port_reconfigure(port, node->data);
1870 if (!port->n_ifaces) {
1871 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1872 br->name, port->name);
1876 shash_destroy(&old_ports);
1877 shash_destroy(&new_ports);
1879 /* Set the fail-mode */
1880 fail_mode = !br->cfg->fail_mode
1881 || !strcmp(br->cfg->fail_mode, "standalone")
1882 ? OFPROTO_FAIL_STANDALONE
1883 : OFPROTO_FAIL_SECURE;
1884 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1885 && !ofproto_has_primary_controller(br->ofproto)) {
1886 ofproto_flush_flows(br->ofproto);
1888 ofproto_set_fail_mode(br->ofproto, fail_mode);
1890 /* Delete all flows if we're switching from connected to standalone or vice
1891 * versa. (XXX Should we delete all flows if we are switching from one
1892 * controller to another?) */
1894 /* Configure OpenFlow controller connection snooping. */
1896 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1897 ovs_rundir(), br->name));
1898 svec_init(&old_snoops);
1899 ofproto_get_snoops(br->ofproto, &old_snoops);
1900 if (!svec_equal(&snoops, &old_snoops)) {
1901 ofproto_set_snoops(br->ofproto, &snoops);
1903 svec_destroy(&snoops);
1904 svec_destroy(&old_snoops);
1906 mirror_reconfigure(br);
1909 /* Initializes 'oc' appropriately as a management service controller for
1912 * The caller must free oc->target when it is no longer needed. */
1914 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1915 struct ofproto_controller *oc)
1917 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1918 oc->max_backoff = 0;
1919 oc->probe_interval = 60;
1920 oc->band = OFPROTO_OUT_OF_BAND;
1922 oc->burst_limit = 0;
1925 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1927 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1928 struct ofproto_controller *oc)
1930 oc->target = c->target;
1931 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1932 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1933 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1934 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1935 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1936 oc->burst_limit = (c->controller_burst_limit
1937 ? *c->controller_burst_limit : 0);
1940 /* Configures the IP stack for 'br''s local interface properly according to the
1941 * configuration in 'c'. */
1943 bridge_configure_local_iface_netdev(struct bridge *br,
1944 struct ovsrec_controller *c)
1946 struct netdev *netdev;
1947 struct in_addr mask, gateway;
1949 struct iface *local_iface;
1952 /* If there's no local interface or no IP address, give up. */
1953 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
1954 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1958 /* Bring up the local interface. */
1959 netdev = local_iface->netdev;
1960 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1962 /* Configure the IP address and netmask. */
1963 if (!c->local_netmask
1964 || !inet_aton(c->local_netmask, &mask)
1966 mask.s_addr = guess_netmask(ip.s_addr);
1968 if (!netdev_set_in4(netdev, ip, mask)) {
1969 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1970 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1973 /* Configure the default gateway. */
1974 if (c->local_gateway
1975 && inet_aton(c->local_gateway, &gateway)
1976 && gateway.s_addr) {
1977 if (!netdev_add_router(netdev, gateway)) {
1978 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1979 br->name, IP_ARGS(&gateway.s_addr));
1985 bridge_reconfigure_remotes(struct bridge *br,
1986 const struct sockaddr_in *managers,
1989 const char *disable_ib_str, *queue_id_str;
1990 bool disable_in_band = false;
1993 struct ovsrec_controller **controllers;
1994 size_t n_controllers;
1997 struct ofproto_controller *ocs;
2001 /* Check if we should disable in-band control on this bridge. */
2002 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
2003 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
2004 disable_in_band = true;
2007 /* Set OpenFlow queue ID for in-band control. */
2008 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
2009 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
2010 ofproto_set_in_band_queue(br->ofproto, queue_id);
2012 if (disable_in_band) {
2013 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
2015 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
2017 had_primary = ofproto_has_primary_controller(br->ofproto);
2019 n_controllers = bridge_get_controllers(br, &controllers);
2021 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
2024 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
2025 for (i = 0; i < n_controllers; i++) {
2026 struct ovsrec_controller *c = controllers[i];
2028 if (!strncmp(c->target, "punix:", 6)
2029 || !strncmp(c->target, "unix:", 5)) {
2030 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2032 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
2033 * domain sockets and overwriting arbitrary local files. */
2034 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2035 "\"%s\" due to possibility for remote exploit",
2036 dpif_name(br->dpif), c->target);
2040 bridge_configure_local_iface_netdev(br, c);
2041 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2042 if (disable_in_band) {
2043 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2048 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2049 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2052 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2053 ofproto_flush_flows(br->ofproto);
2056 /* If there are no controllers and the bridge is in standalone
2057 * mode, set up a flow that matches every packet and directs
2058 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2059 * switch is in secure mode and we won't pass any traffic until
2060 * a controller has been defined and it tells us to do so. */
2062 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2063 union ofp_action action;
2064 struct cls_rule rule;
2066 memset(&action, 0, sizeof action);
2067 action.type = htons(OFPAT_OUTPUT);
2068 action.output.len = htons(sizeof action);
2069 action.output.port = htons(OFPP_NORMAL);
2070 cls_rule_init_catchall(&rule, 0);
2071 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2076 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2081 for (i = 0; i < br->n_ports; i++) {
2082 struct port *port = br->ports[i];
2083 for (j = 0; j < port->n_ifaces; j++) {
2084 struct iface *iface = port->ifaces[j];
2085 shash_add_once(ifaces, iface->name, iface);
2087 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2088 shash_add_once(ifaces, port->name, NULL);
2093 /* For robustness, in case the administrator moves around datapath ports behind
2094 * our back, we re-check all the datapath port numbers here.
2096 * This function will set the 'dp_ifidx' members of interfaces that have
2097 * disappeared to -1, so only call this function from a context where those
2098 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2099 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2100 * datapath, which doesn't support UINT16_MAX+1 ports. */
2102 bridge_fetch_dp_ifaces(struct bridge *br)
2104 struct dpif_port_dump dump;
2105 struct dpif_port dpif_port;
2108 /* Reset all interface numbers. */
2109 for (i = 0; i < br->n_ports; i++) {
2110 struct port *port = br->ports[i];
2111 for (j = 0; j < port->n_ifaces; j++) {
2112 struct iface *iface = port->ifaces[j];
2113 iface->dp_ifidx = -1;
2116 hmap_clear(&br->ifaces);
2118 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2119 struct iface *iface = iface_lookup(br, dpif_port.name);
2121 if (iface->dp_ifidx >= 0) {
2122 VLOG_WARN("%s reported interface %s twice",
2123 dpif_name(br->dpif), dpif_port.name);
2124 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2125 VLOG_WARN("%s reported interface %"PRIu16" twice",
2126 dpif_name(br->dpif), dpif_port.port_no);
2128 iface->dp_ifidx = dpif_port.port_no;
2129 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2130 hash_int(iface->dp_ifidx, 0));
2133 iface_set_ofport(iface->cfg,
2134 (iface->dp_ifidx >= 0
2135 ? odp_port_to_ofp_port(iface->dp_ifidx)
2141 /* Bridge packet processing functions. */
2144 bond_is_tcp_hash(const struct port *port)
2146 return port->bond_mode == BM_TCP && lacp_negotiated(port->lacp);
2150 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2152 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2155 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2157 struct flow hash_flow;
2159 memcpy(&hash_flow, flow, sizeof hash_flow);
2160 hash_flow.vlan_tci = 0;
2162 /* The symmetric quality of this hash function is not required, but
2163 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2164 * purposes, so we use it out of convenience. */
2165 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2168 static struct bond_entry *
2169 lookup_bond_entry(const struct port *port, const struct flow *flow,
2172 assert(port->bond_mode != BM_AB);
2174 if (bond_is_tcp_hash(port)) {
2175 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2177 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2181 static struct iface *
2182 bond_choose_iface(const struct port *port)
2184 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2185 struct iface *best_down_slave;
2188 best_down_slave = NULL;
2189 for (i = 0; i < port->n_ifaces; i++) {
2190 struct iface *iface = port->ifaces[i];
2192 if (iface->enabled) {
2194 } else if ((!best_down_slave
2195 || iface->delay_expires < best_down_slave->delay_expires)
2196 && lacp_slave_may_enable(port->lacp, iface)) {
2197 best_down_slave = iface;
2201 if (best_down_slave) {
2202 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2203 "since no other interface is up",
2204 best_down_slave->name,
2205 best_down_slave->delay_expires - time_msec());
2206 bond_enable_slave(best_down_slave, true);
2209 return best_down_slave;
2213 choose_output_iface(const struct port *port, const struct flow *flow,
2214 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2216 struct iface *iface;
2218 assert(port->n_ifaces);
2219 if (port->n_ifaces == 1) {
2220 iface = port->ifaces[0];
2221 } else if (port->bond_mode == BM_AB) {
2222 iface = port->active_iface;
2224 *tags |= port->no_ifaces_tag;
2228 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2229 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2230 || !port->ifaces[e->iface_idx]->enabled) {
2231 /* XXX select interface properly. The current interface selection
2232 * is only good for testing the rebalancing code. */
2233 iface = bond_choose_iface(port);
2236 *tags |= port->no_ifaces_tag;
2239 e->iface_idx = iface->port_ifidx;
2240 e->tag = tag_create_random();
2243 iface = port->ifaces[e->iface_idx];
2245 *dp_ifidx = iface->dp_ifidx;
2246 *tags |= iface->tag; /* Currently only used for bonding. */
2251 bond_link_status_update(struct iface *iface)
2253 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2254 struct port *port = iface->port;
2255 bool up = iface->up && lacp_slave_may_enable(port->lacp, iface);
2256 int updelay, downdelay;
2258 updelay = port->updelay;
2259 downdelay = port->downdelay;
2261 if (lacp_negotiated(port->lacp)) {
2266 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2267 /* Nothing to do. */
2270 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2271 iface->name, up ? "up" : "down");
2272 if (up == iface->enabled) {
2273 iface->delay_expires = LLONG_MAX;
2274 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2275 iface->name, up ? "disabled" : "enabled");
2276 } else if (up && !port->active_iface) {
2277 bond_enable_slave(iface, true);
2279 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2280 "other interface is up", iface->name, updelay);
2283 int delay = up ? updelay : downdelay;
2284 iface->delay_expires = time_msec() + delay;
2287 "interface %s: will be %s if it stays %s for %d ms",
2289 up ? "enabled" : "disabled",
2297 bond_choose_active_iface(struct port *port)
2299 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2301 port->active_iface = bond_choose_iface(port);
2302 if (port->active_iface) {
2303 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2304 port->name, port->active_iface->name);
2306 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2312 bond_enable_slave(struct iface *iface, bool enable)
2314 struct port *port = iface->port;
2315 struct bridge *br = port->bridge;
2317 /* This acts as a recursion check. If the act of disabling a slave
2318 * causes a different slave to be enabled, the flag will allow us to
2319 * skip redundant work when we reenter this function. It must be
2320 * cleared on exit to keep things safe with multiple bonds. */
2321 static bool moving_active_iface = false;
2323 iface->delay_expires = LLONG_MAX;
2324 if (enable == iface->enabled) {
2328 iface->enabled = enable;
2329 if (!iface->enabled) {
2330 VLOG_WARN("interface %s: disabled", iface->name);
2331 ofproto_revalidate(br->ofproto, iface->tag);
2332 if (iface == port->active_iface) {
2333 /* Disabling a slave can lead to another slave being immediately
2334 * enabled if there will be no active slaves but one is waiting
2335 * on an updelay. In this case we do not need to run most of the
2336 * code for the newly enabled slave since there was no period
2337 * without an active slave and it is redundant with the disabling
2339 moving_active_iface = true;
2340 bond_choose_active_iface(port);
2342 bond_send_learning_packets(port);
2344 VLOG_WARN("interface %s: enabled", iface->name);
2345 if (!port->active_iface && !moving_active_iface) {
2346 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2347 bond_choose_active_iface(port);
2348 bond_send_learning_packets(port);
2350 iface->tag = tag_create_random();
2353 moving_active_iface = false;
2356 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2357 * bond interface. */
2359 bond_update_fake_iface_stats(struct port *port)
2361 struct netdev_stats bond_stats;
2362 struct netdev *bond_dev;
2365 memset(&bond_stats, 0, sizeof bond_stats);
2367 for (i = 0; i < port->n_ifaces; i++) {
2368 struct netdev_stats slave_stats;
2370 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2371 /* XXX: We swap the stats here because they are swapped back when
2372 * reported by the internal device. The reason for this is
2373 * internal devices normally represent packets going into the system
2374 * but when used as fake bond device they represent packets leaving
2375 * the system. We really should do this in the internal device
2376 * itself because changing it here reverses the counts from the
2377 * perspective of the switch. However, the internal device doesn't
2378 * know what type of device it represents so we have to do it here
2380 bond_stats.tx_packets += slave_stats.rx_packets;
2381 bond_stats.tx_bytes += slave_stats.rx_bytes;
2382 bond_stats.rx_packets += slave_stats.tx_packets;
2383 bond_stats.rx_bytes += slave_stats.tx_bytes;
2387 if (!netdev_open_default(port->name, &bond_dev)) {
2388 netdev_set_stats(bond_dev, &bond_stats);
2389 netdev_close(bond_dev);
2394 bond_run(struct port *port)
2398 if (port->n_ifaces < 2) {
2402 for (i = 0; i < port->n_ifaces; i++) {
2403 bond_link_status_update(port->ifaces[i]);
2406 for (i = 0; i < port->n_ifaces; i++) {
2407 struct iface *iface = port->ifaces[i];
2408 if (time_msec() >= iface->delay_expires) {
2409 bond_enable_slave(iface, !iface->enabled);
2413 if (port->bond_fake_iface
2414 && time_msec() >= port->bond_next_fake_iface_update) {
2415 bond_update_fake_iface_stats(port);
2416 port->bond_next_fake_iface_update = time_msec() + 1000;
2421 bond_wait(struct port *port)
2425 if (port->n_ifaces < 2) {
2429 for (i = 0; i < port->n_ifaces; i++) {
2430 struct iface *iface = port->ifaces[i];
2431 if (iface->delay_expires != LLONG_MAX) {
2432 poll_timer_wait_until(iface->delay_expires);
2436 if (port->bond_fake_iface) {
2437 poll_timer_wait_until(port->bond_next_fake_iface_update);
2442 set_dst(struct dst *dst, const struct flow *flow,
2443 const struct port *in_port, const struct port *out_port,
2446 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2447 : in_port->vlan >= 0 ? in_port->vlan
2448 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2449 : vlan_tci_to_vid(flow->vlan_tci));
2450 return choose_output_iface(out_port, flow, dst->vlan,
2451 &dst->dp_ifidx, tags);
2455 swap_dst(struct dst *p, struct dst *q)
2457 struct dst tmp = *p;
2462 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2463 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2464 * that we push to the datapath. We could in fact fully sort the array by
2465 * vlan, but in most cases there are at most two different vlan tags so that's
2466 * possibly overkill.) */
2468 partition_dsts(struct dst_set *set, int vlan)
2470 struct dst *first = set->dsts;
2471 struct dst *last = set->dsts + set->n;
2473 while (first != last) {
2475 * - All dsts < first have vlan == 'vlan'.
2476 * - All dsts >= last have vlan != 'vlan'.
2477 * - first < last. */
2478 while (first->vlan == vlan) {
2479 if (++first == last) {
2484 /* Same invariants, plus one additional:
2485 * - first->vlan != vlan.
2487 while (last[-1].vlan != vlan) {
2488 if (--last == first) {
2493 /* Same invariants, plus one additional:
2494 * - last[-1].vlan == vlan.*/
2495 swap_dst(first++, --last);
2500 mirror_mask_ffs(mirror_mask_t mask)
2502 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2507 dst_set_init(struct dst_set *set)
2509 set->dsts = set->builtin;
2511 set->allocated = ARRAY_SIZE(set->builtin);
2515 dst_set_add(struct dst_set *set, const struct dst *dst)
2517 if (set->n >= set->allocated) {
2518 size_t new_allocated;
2519 struct dst *new_dsts;
2521 new_allocated = set->allocated * 2;
2522 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2523 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2527 set->dsts = new_dsts;
2528 set->allocated = new_allocated;
2530 set->dsts[set->n++] = *dst;
2534 dst_set_free(struct dst_set *set)
2536 if (set->dsts != set->builtin) {
2542 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2545 for (i = 0; i < set->n; i++) {
2546 if (set->dsts[i].vlan == test->vlan
2547 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2555 port_trunks_vlan(const struct port *port, uint16_t vlan)
2557 return (port->vlan < 0
2558 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2562 port_includes_vlan(const struct port *port, uint16_t vlan)
2564 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2568 port_is_floodable(const struct port *port)
2572 for (i = 0; i < port->n_ifaces; i++) {
2573 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2574 port->ifaces[i]->dp_ifidx)) {
2581 /* Returns the tag for 'port''s active iface, or 'port''s no_ifaces_tag if
2582 * there is no active iface. */
2584 port_get_active_iface_tag(const struct port *port)
2586 return (port->active_iface
2587 ? port->active_iface->tag
2588 : port->no_ifaces_tag);
2592 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2593 const struct port *in_port, const struct port *out_port,
2594 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2596 mirror_mask_t mirrors = in_port->src_mirrors;
2601 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2602 if (flow_vlan == 0) {
2603 flow_vlan = OFP_VLAN_NONE;
2606 if (out_port == FLOOD_PORT) {
2607 for (i = 0; i < br->n_ports; i++) {
2608 struct port *port = br->ports[i];
2610 && port_is_floodable(port)
2611 && port_includes_vlan(port, vlan)
2612 && !port->is_mirror_output_port
2613 && set_dst(&dst, flow, in_port, port, tags)) {
2614 mirrors |= port->dst_mirrors;
2615 dst_set_add(set, &dst);
2618 *nf_output_iface = NF_OUT_FLOOD;
2619 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2620 dst_set_add(set, &dst);
2621 *nf_output_iface = dst.dp_ifidx;
2622 mirrors |= out_port->dst_mirrors;
2626 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2627 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2629 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2630 && !dst_is_duplicate(set, &dst)) {
2631 dst_set_add(set, &dst);
2634 for (i = 0; i < br->n_ports; i++) {
2635 struct port *port = br->ports[i];
2636 if (port_includes_vlan(port, m->out_vlan)
2637 && set_dst(&dst, flow, in_port, port, tags))
2639 if (port->vlan < 0) {
2640 dst.vlan = m->out_vlan;
2642 if (dst_is_duplicate(set, &dst)) {
2646 /* Use the vlan tag on the original flow instead of
2647 * the one passed in the vlan parameter. This ensures
2648 * that we compare the vlan from before any implicit
2649 * tagging tags place. This is necessary because
2650 * dst->vlan is the final vlan, after removing implicit
2652 if (port == in_port && dst.vlan == flow_vlan) {
2653 /* Don't send out input port on same VLAN. */
2656 dst_set_add(set, &dst);
2661 mirrors &= mirrors - 1;
2664 partition_dsts(set, flow_vlan);
2667 static void OVS_UNUSED
2668 print_dsts(const struct dst_set *set)
2672 for (i = 0; i < set->n; i++) {
2673 const struct dst *dst = &set->dsts[i];
2675 printf(">p%"PRIu16, dst->dp_ifidx);
2676 if (dst->vlan != OFP_VLAN_NONE) {
2677 printf("v%"PRIu16, dst->vlan);
2683 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2684 const struct port *in_port, const struct port *out_port,
2685 tag_type *tags, struct ofpbuf *actions,
2686 uint16_t *nf_output_iface)
2693 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2696 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2697 if (cur_vlan == 0) {
2698 cur_vlan = OFP_VLAN_NONE;
2700 for (i = 0; i < set.n; i++) {
2701 const struct dst *dst = &set.dsts[i];
2702 if (dst->vlan != cur_vlan) {
2703 if (dst->vlan == OFP_VLAN_NONE) {
2704 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2707 tci = htons(dst->vlan & VLAN_VID_MASK);
2708 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2709 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2711 cur_vlan = dst->vlan;
2713 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2718 /* Returns the effective vlan of a packet, taking into account both the
2719 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2720 * the packet is untagged and -1 indicates it has an invalid header and
2721 * should be dropped. */
2722 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2723 struct port *in_port, bool have_packet)
2725 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2726 if (in_port->vlan >= 0) {
2728 /* XXX support double tagging? */
2730 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2731 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2732 "packet received on port %s configured with "
2733 "implicit VLAN %"PRIu16,
2734 br->name, vlan, in_port->name, in_port->vlan);
2738 vlan = in_port->vlan;
2740 if (!port_includes_vlan(in_port, vlan)) {
2742 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2743 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2744 "packet received on port %s not configured for "
2746 br->name, vlan, in_port->name, vlan);
2755 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2756 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2757 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2759 is_gratuitous_arp(const struct flow *flow)
2761 return (flow->dl_type == htons(ETH_TYPE_ARP)
2762 && eth_addr_is_broadcast(flow->dl_dst)
2763 && (flow->nw_proto == ARP_OP_REPLY
2764 || (flow->nw_proto == ARP_OP_REQUEST
2765 && flow->nw_src == flow->nw_dst)));
2769 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2770 struct port *in_port)
2772 enum grat_arp_lock_type lock_type;
2775 /* We don't want to learn from gratuitous ARP packets that are reflected
2776 * back over bond slaves so we lock the learning table. */
2777 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2778 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2779 GRAT_ARP_LOCK_CHECK;
2781 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2784 /* The log messages here could actually be useful in debugging,
2785 * so keep the rate limit relatively high. */
2786 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2788 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2789 "on port %s in VLAN %d",
2790 br->name, ETH_ADDR_ARGS(flow->dl_src),
2791 in_port->name, vlan);
2792 ofproto_revalidate(br->ofproto, rev_tag);
2796 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2797 * dropped. Returns true if they may be forwarded, false if they should be
2800 * If 'have_packet' is true, it indicates that the caller is processing a
2801 * received packet. If 'have_packet' is false, then the caller is just
2802 * revalidating an existing flow because configuration has changed. Either
2803 * way, 'have_packet' only affects logging (there is no point in logging errors
2804 * during revalidation).
2806 * Sets '*in_portp' to the input port. This will be a null pointer if
2807 * flow->in_port does not designate a known input port (in which case
2808 * is_admissible() returns false).
2810 * When returning true, sets '*vlanp' to the effective VLAN of the input
2811 * packet, as returned by flow_get_vlan().
2813 * May also add tags to '*tags', although the current implementation only does
2814 * so in one special case.
2817 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2818 tag_type *tags, int *vlanp, struct port **in_portp)
2820 struct iface *in_iface;
2821 struct port *in_port;
2824 /* Find the interface and port structure for the received packet. */
2825 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2827 /* No interface? Something fishy... */
2829 /* Odd. A few possible reasons here:
2831 * - We deleted an interface but there are still a few packets
2832 * queued up from it.
2834 * - Someone externally added an interface (e.g. with "ovs-dpctl
2835 * add-if") that we don't know about.
2837 * - Packet arrived on the local port but the local port is not
2838 * one of our bridge ports.
2840 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2842 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2843 "interface %"PRIu16, br->name, flow->in_port);
2849 *in_portp = in_port = in_iface->port;
2850 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2855 /* Drop frames for reserved multicast addresses. */
2856 if (eth_addr_is_reserved(flow->dl_dst)) {
2860 /* Drop frames on ports reserved for mirroring. */
2861 if (in_port->is_mirror_output_port) {
2863 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2864 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2865 "%s, which is reserved exclusively for mirroring",
2866 br->name, in_port->name);
2871 /* When using LACP, do not accept packets from disabled interfaces. */
2872 if (lacp_negotiated(in_port->lacp) && !in_iface->enabled) {
2876 /* Packets received on non-LACP bonds need special attention to avoid
2878 if (in_port->n_ifaces > 1 && !lacp_negotiated(in_port->lacp)) {
2880 bool is_grat_arp_locked;
2882 if (eth_addr_is_multicast(flow->dl_dst)) {
2883 *tags |= port_get_active_iface_tag(in_port);
2884 if (in_port->active_iface != in_iface) {
2885 /* Drop all multicast packets on inactive slaves. */
2890 /* Drop all packets for which we have learned a different input
2891 * port, because we probably sent the packet on one slave and got
2892 * it back on the other. Gratuitous ARP packets are an exception
2893 * to this rule: the host has moved to another switch. The exception
2894 * to the exception is if we locked the learning table to avoid
2895 * reflections on bond slaves. If this is the case, just drop the
2897 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2898 &is_grat_arp_locked);
2899 if (src_idx != -1 && src_idx != in_port->port_idx &&
2900 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2908 /* If the composed actions may be applied to any packet in the given 'flow',
2909 * returns true. Otherwise, the actions should only be applied to 'packet', or
2910 * not at all, if 'packet' was NULL. */
2912 process_flow(struct bridge *br, const struct flow *flow,
2913 const struct ofpbuf *packet, struct ofpbuf *actions,
2914 tag_type *tags, uint16_t *nf_output_iface)
2916 struct port *in_port;
2917 struct port *out_port;
2921 /* Check whether we should drop packets in this flow. */
2922 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2927 /* Learn source MAC (but don't try to learn from revalidation). */
2929 update_learning_table(br, flow, vlan, in_port);
2932 /* Determine output port. */
2933 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2935 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2936 out_port = br->ports[out_port_idx];
2937 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2938 /* If we are revalidating but don't have a learning entry then
2939 * eject the flow. Installing a flow that floods packets opens
2940 * up a window of time where we could learn from a packet reflected
2941 * on a bond and blackhole packets before the learning table is
2942 * updated to reflect the correct port. */
2945 out_port = FLOOD_PORT;
2948 /* Don't send packets out their input ports. */
2949 if (in_port == out_port) {
2955 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2963 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2964 struct ofpbuf *actions, tag_type *tags,
2965 uint16_t *nf_output_iface, void *br_)
2967 struct bridge *br = br_;
2969 COVERAGE_INC(bridge_process_flow);
2970 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2974 bridge_special_ofhook_cb(const struct flow *flow,
2975 const struct ofpbuf *packet, void *br_)
2977 struct iface *iface;
2978 struct bridge *br = br_;
2980 iface = iface_from_dp_ifidx(br, flow->in_port);
2982 if (flow->dl_type == htons(ETH_TYPE_LACP)) {
2984 if (iface && iface->port->lacp && packet) {
2985 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
2988 COVERAGE_INC(bridge_process_lacp);
2989 lacp_process_pdu(iface->port->lacp, iface, pdu);
2999 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
3000 const struct nlattr *actions,
3002 uint64_t n_bytes, void *br_)
3004 struct bridge *br = br_;
3005 const struct nlattr *a;
3006 struct port *in_port;
3011 /* Feed information from the active flows back into the learning table to
3012 * ensure that table is always in sync with what is actually flowing
3013 * through the datapath.
3015 * We test that 'tags' is nonzero to ensure that only flows that include an
3016 * OFPP_NORMAL action are used for learning. This works because
3017 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3018 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3019 update_learning_table(br, flow, vlan, in_port);
3022 /* Account for bond slave utilization. */
3023 if (!br->has_bonded_ports) {
3026 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3027 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3028 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3029 if (out_port && out_port->n_ifaces >= 2 &&
3030 out_port->bond_mode != BM_AB) {
3031 uint16_t vlan = (flow->vlan_tci
3032 ? vlan_tci_to_vid(flow->vlan_tci)
3034 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
3035 e->tx_bytes += n_bytes;
3042 bridge_account_checkpoint_ofhook_cb(void *br_)
3044 struct bridge *br = br_;
3048 if (!br->has_bonded_ports) {
3053 for (i = 0; i < br->n_ports; i++) {
3054 struct port *port = br->ports[i];
3055 if (port->n_ifaces > 1 && port->bond_mode != BM_AB
3056 && now >= port->bond_next_rebalance) {
3057 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3058 bond_rebalance_port(port);
3063 static struct ofhooks bridge_ofhooks = {
3064 bridge_normal_ofhook_cb,
3065 bridge_special_ofhook_cb,
3066 bridge_account_flow_ofhook_cb,
3067 bridge_account_checkpoint_ofhook_cb,
3070 /* Bonding functions. */
3072 /* Statistics for a single interface on a bonded port, used for load-based
3073 * bond rebalancing. */
3074 struct slave_balance {
3075 struct iface *iface; /* The interface. */
3076 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3078 /* All the "bond_entry"s that are assigned to this interface, in order of
3079 * increasing tx_bytes. */
3080 struct bond_entry **hashes;
3085 bond_mode_to_string(enum bond_mode bm) {
3086 static char *bm_slb = "balance-slb";
3087 static char *bm_ab = "active-backup";
3088 static char *bm_tcp = "balance-tcp";
3091 case BM_SLB: return bm_slb;
3092 case BM_AB: return bm_ab;
3093 case BM_TCP: return bm_tcp;
3100 /* Sorts pointers to pointers to bond_entries in ascending order by the
3101 * interface to which they are assigned, and within a single interface in
3102 * ascending order of bytes transmitted. */
3104 compare_bond_entries(const void *a_, const void *b_)
3106 const struct bond_entry *const *ap = a_;
3107 const struct bond_entry *const *bp = b_;
3108 const struct bond_entry *a = *ap;
3109 const struct bond_entry *b = *bp;
3110 if (a->iface_idx != b->iface_idx) {
3111 return a->iface_idx > b->iface_idx ? 1 : -1;
3112 } else if (a->tx_bytes != b->tx_bytes) {
3113 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3119 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3120 * *descending* order by number of bytes transmitted. */
3122 compare_slave_balance(const void *a_, const void *b_)
3124 const struct slave_balance *a = a_;
3125 const struct slave_balance *b = b_;
3126 if (a->iface->enabled != b->iface->enabled) {
3127 return a->iface->enabled ? -1 : 1;
3128 } else if (a->tx_bytes != b->tx_bytes) {
3129 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3136 swap_bals(struct slave_balance *a, struct slave_balance *b)
3138 struct slave_balance tmp = *a;
3143 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3144 * given that 'p' (and only 'p') might be in the wrong location.
3146 * This function invalidates 'p', since it might now be in a different memory
3149 resort_bals(struct slave_balance *p,
3150 struct slave_balance bals[], size_t n_bals)
3153 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3154 swap_bals(p, p - 1);
3156 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3157 swap_bals(p, p + 1);
3163 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3165 if (VLOG_IS_DBG_ENABLED()) {
3166 struct ds ds = DS_EMPTY_INITIALIZER;
3167 const struct slave_balance *b;
3169 for (b = bals; b < bals + n_bals; b++) {
3173 ds_put_char(&ds, ',');
3175 ds_put_format(&ds, " %s %"PRIu64"kB",
3176 b->iface->name, b->tx_bytes / 1024);
3178 if (!b->iface->enabled) {
3179 ds_put_cstr(&ds, " (disabled)");
3181 if (b->n_hashes > 0) {
3182 ds_put_cstr(&ds, " (");
3183 for (i = 0; i < b->n_hashes; i++) {
3184 const struct bond_entry *e = b->hashes[i];
3186 ds_put_cstr(&ds, " + ");
3188 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3189 e - port->bond_hash, e->tx_bytes / 1024);
3191 ds_put_cstr(&ds, ")");
3194 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3199 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3201 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3204 struct bond_entry *hash = from->hashes[hash_idx];
3205 struct port *port = from->iface->port;
3206 uint64_t delta = hash->tx_bytes;
3208 assert(port->bond_mode != BM_AB);
3210 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3211 "from %s to %s (now carrying %"PRIu64"kB and "
3212 "%"PRIu64"kB load, respectively)",
3213 port->name, delta / 1024, hash - port->bond_hash,
3214 from->iface->name, to->iface->name,
3215 (from->tx_bytes - delta) / 1024,
3216 (to->tx_bytes + delta) / 1024);
3218 /* Delete element from from->hashes.
3220 * We don't bother to add the element to to->hashes because not only would
3221 * it require more work, the only purpose it would be to allow that hash to
3222 * be migrated to another slave in this rebalancing run, and there is no
3223 * point in doing that. */
3224 if (hash_idx == 0) {
3227 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3228 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3232 /* Shift load away from 'from' to 'to'. */
3233 from->tx_bytes -= delta;
3234 to->tx_bytes += delta;
3236 /* Arrange for flows to be revalidated. */
3237 ofproto_revalidate(port->bridge->ofproto, hash->tag);
3238 hash->iface_idx = to->iface->port_ifidx;
3239 hash->tag = tag_create_random();
3243 bond_rebalance_port(struct port *port)
3245 struct slave_balance *bals;
3247 struct bond_entry *hashes[BOND_MASK + 1];
3248 struct slave_balance *b, *from, *to;
3249 struct bond_entry *e;
3252 assert(port->bond_mode != BM_AB);
3254 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3255 * descending order of tx_bytes, so that bals[0] represents the most
3256 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3259 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3260 * array for each slave_balance structure, we sort our local array of
3261 * hashes in order by slave, so that all of the hashes for a given slave
3262 * become contiguous in memory, and then we point each 'hashes' members of
3263 * a slave_balance structure to the start of a contiguous group. */
3264 n_bals = port->n_ifaces;
3265 bals = xmalloc(n_bals * sizeof *bals);
3266 for (b = bals; b < &bals[n_bals]; b++) {
3267 b->iface = port->ifaces[b - bals];
3272 for (i = 0; i <= BOND_MASK; i++) {
3273 hashes[i] = &port->bond_hash[i];
3275 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3276 for (i = 0; i <= BOND_MASK; i++) {
3278 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3279 b = &bals[e->iface_idx];
3280 b->tx_bytes += e->tx_bytes;
3282 b->hashes = &hashes[i];
3287 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3288 log_bals(bals, n_bals, port);
3290 /* Discard slaves that aren't enabled (which were sorted to the back of the
3291 * array earlier). */
3292 while (!bals[n_bals - 1].iface->enabled) {
3299 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3300 to = &bals[n_bals - 1];
3301 for (from = bals; from < to; ) {
3302 uint64_t overload = from->tx_bytes - to->tx_bytes;
3303 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3304 /* The extra load on 'from' (and all less-loaded slaves), compared
3305 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3306 * it is less than ~1Mbps. No point in rebalancing. */
3308 } else if (from->n_hashes == 1) {
3309 /* 'from' only carries a single MAC hash, so we can't shift any
3310 * load away from it, even though we want to. */
3313 /* 'from' is carrying significantly more load than 'to', and that
3314 * load is split across at least two different hashes. Pick a hash
3315 * to migrate to 'to' (the least-loaded slave), given that doing so
3316 * must decrease the ratio of the load on the two slaves by at
3319 * The sort order we use means that we prefer to shift away the
3320 * smallest hashes instead of the biggest ones. There is little
3321 * reason behind this decision; we could use the opposite sort
3322 * order to shift away big hashes ahead of small ones. */
3325 for (i = 0; i < from->n_hashes; i++) {
3326 double old_ratio, new_ratio;
3327 uint64_t delta = from->hashes[i]->tx_bytes;
3329 if (delta == 0 || from->tx_bytes - delta == 0) {
3330 /* Pointless move. */
3334 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3336 if (to->tx_bytes == 0) {
3337 /* Nothing on the new slave, move it. */
3341 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3342 new_ratio = (double)(from->tx_bytes - delta) /
3343 (to->tx_bytes + delta);
3345 if (new_ratio == 0) {
3346 /* Should already be covered but check to prevent division
3351 if (new_ratio < 1) {
3352 new_ratio = 1 / new_ratio;
3355 if (old_ratio - new_ratio > 0.1) {
3356 /* Would decrease the ratio, move it. */
3360 if (i < from->n_hashes) {
3361 bond_shift_load(from, to, i);
3363 /* If the result of the migration changed the relative order of
3364 * 'from' and 'to' swap them back to maintain invariants. */
3365 if (order_swapped) {
3366 swap_bals(from, to);
3369 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3370 * point to different slave_balance structures. It is only
3371 * valid to do these two operations in a row at all because we
3372 * know that 'from' will not move past 'to' and vice versa. */
3373 resort_bals(from, bals, n_bals);
3374 resort_bals(to, bals, n_bals);
3381 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3382 * historical data to decay to <1% in 7 rebalancing runs. */
3383 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3395 bond_send_learning_packets(struct port *port)
3397 struct bridge *br = port->bridge;
3398 struct mac_entry *e;
3399 struct ofpbuf packet;
3400 int error, n_packets, n_errors;
3402 if (!port->n_ifaces || !port->active_iface || bond_is_tcp_hash(port)) {
3406 ofpbuf_init(&packet, 128);
3407 error = n_packets = n_errors = 0;
3408 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3414 if (e->port == port->port_idx) {
3418 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3420 flow_extract(&packet, 0, ODPP_NONE, &flow);
3422 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3428 retval = ofproto_send_packet(br->ofproto, dp_ifidx, e->vlan, &packet);
3434 ofpbuf_uninit(&packet);
3437 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3438 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3439 "packets, last error was: %s",
3440 port->name, n_errors, n_packets, strerror(error));
3442 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3443 port->name, n_packets);
3447 /* Bonding unixctl user interface functions. */
3450 bond_unixctl_list(struct unixctl_conn *conn,
3451 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3453 struct ds ds = DS_EMPTY_INITIALIZER;
3454 const struct bridge *br;
3456 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3458 LIST_FOR_EACH (br, node, &all_bridges) {
3461 for (i = 0; i < br->n_ports; i++) {
3462 const struct port *port = br->ports[i];
3463 if (port->n_ifaces > 1) {
3466 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3467 bond_mode_to_string(port->bond_mode));
3468 for (j = 0; j < port->n_ifaces; j++) {
3469 const struct iface *iface = port->ifaces[j];
3471 ds_put_cstr(&ds, ", ");
3473 ds_put_cstr(&ds, iface->name);
3475 ds_put_char(&ds, '\n');
3479 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3483 static struct port *
3484 bond_find(const char *name)
3486 const struct bridge *br;
3488 LIST_FOR_EACH (br, node, &all_bridges) {
3491 for (i = 0; i < br->n_ports; i++) {
3492 struct port *port = br->ports[i];
3493 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3502 bond_unixctl_show(struct unixctl_conn *conn,
3503 const char *args, void *aux OVS_UNUSED)
3505 struct ds ds = DS_EMPTY_INITIALIZER;
3506 const struct port *port;
3509 port = bond_find(args);
3511 unixctl_command_reply(conn, 501, "no such bond");
3515 ds_put_format(&ds, "bond_mode: %s\n",
3516 bond_mode_to_string(port->bond_mode));
3519 ds_put_format(&ds, "lacp: %s\n",
3520 port->lacp_active ? "active" : "passive");
3522 ds_put_cstr(&ds, "lacp: off\n");
3525 if (port->bond_mode != BM_AB) {
3526 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3527 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3531 ds_put_format(&ds, "bond-detect-mode: %s\n",
3532 port->monitor ? "carrier" : "miimon");
3534 if (!port->monitor) {
3535 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3536 port->miimon_interval);
3539 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3540 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3542 if (port->bond_mode != BM_AB) {
3543 ds_put_format(&ds, "next rebalance: %lld ms\n",
3544 port->bond_next_rebalance - time_msec());
3547 for (j = 0; j < port->n_ifaces; j++) {
3548 const struct iface *iface = port->ifaces[j];
3549 struct bond_entry *be;
3553 ds_put_format(&ds, "\nslave %s: %s\n",
3554 iface->name, iface->enabled ? "enabled" : "disabled");
3555 if (iface == port->active_iface) {
3556 ds_put_cstr(&ds, "\tactive slave\n");
3558 if (iface->delay_expires != LLONG_MAX) {
3559 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3560 iface->enabled ? "downdelay" : "updelay",
3561 iface->delay_expires - time_msec());
3564 if (port->bond_mode == BM_AB) {
3569 memset(&flow, 0, sizeof flow);
3570 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3571 int hash = be - port->bond_hash;
3572 struct mac_entry *me;
3574 if (be->iface_idx != j) {
3578 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3579 hash, be->tx_bytes / 1024);
3581 if (port->bond_mode != BM_SLB) {
3586 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3590 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
3591 if (bond_hash_src(me->mac, me->vlan) == hash
3592 && me->port != port->port_idx
3593 && choose_output_iface(port, &flow, me->vlan,
3595 && dp_ifidx == iface->dp_ifidx)
3597 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3598 ETH_ADDR_ARGS(me->mac));
3603 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3608 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3609 void *aux OVS_UNUSED)
3611 char *args = (char *) args_;
3612 char *save_ptr = NULL;
3613 char *bond_s, *hash_s, *slave_s;
3615 struct iface *iface;
3616 struct bond_entry *entry;
3619 bond_s = strtok_r(args, " ", &save_ptr);
3620 hash_s = strtok_r(NULL, " ", &save_ptr);
3621 slave_s = strtok_r(NULL, " ", &save_ptr);
3623 unixctl_command_reply(conn, 501,
3624 "usage: bond/migrate BOND HASH SLAVE");
3628 port = bond_find(bond_s);
3630 unixctl_command_reply(conn, 501, "no such bond");
3634 if (port->bond_mode != BM_SLB) {
3635 unixctl_command_reply(conn, 501, "not an SLB bond");
3639 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3640 hash = atoi(hash_s) & BOND_MASK;
3642 unixctl_command_reply(conn, 501, "bad hash");
3646 iface = port_lookup_iface(port, slave_s);
3648 unixctl_command_reply(conn, 501, "no such slave");
3652 if (!iface->enabled) {
3653 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3657 entry = &port->bond_hash[hash];
3658 ofproto_revalidate(port->bridge->ofproto, entry->tag);
3659 entry->iface_idx = iface->port_ifidx;
3660 entry->tag = tag_create_random();
3661 unixctl_command_reply(conn, 200, "migrated");
3665 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3666 void *aux OVS_UNUSED)
3668 char *args = (char *) args_;
3669 char *save_ptr = NULL;
3670 char *bond_s, *slave_s;
3672 struct iface *iface;
3674 bond_s = strtok_r(args, " ", &save_ptr);
3675 slave_s = strtok_r(NULL, " ", &save_ptr);
3677 unixctl_command_reply(conn, 501,
3678 "usage: bond/set-active-slave BOND SLAVE");
3682 port = bond_find(bond_s);
3684 unixctl_command_reply(conn, 501, "no such bond");
3688 iface = port_lookup_iface(port, slave_s);
3690 unixctl_command_reply(conn, 501, "no such slave");
3694 if (!iface->enabled) {
3695 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3699 if (port->active_iface != iface) {
3700 ofproto_revalidate(port->bridge->ofproto,
3701 port_get_active_iface_tag(port));
3702 port->active_iface = iface;
3703 VLOG_INFO("port %s: active interface is now %s",
3704 port->name, iface->name);
3705 bond_send_learning_packets(port);
3706 unixctl_command_reply(conn, 200, "done");
3708 unixctl_command_reply(conn, 200, "no change");
3713 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3715 char *args = (char *) args_;
3716 char *save_ptr = NULL;
3717 char *bond_s, *slave_s;
3719 struct iface *iface;
3721 bond_s = strtok_r(args, " ", &save_ptr);
3722 slave_s = strtok_r(NULL, " ", &save_ptr);
3724 unixctl_command_reply(conn, 501,
3725 "usage: bond/enable/disable-slave BOND SLAVE");
3729 port = bond_find(bond_s);
3731 unixctl_command_reply(conn, 501, "no such bond");
3735 iface = port_lookup_iface(port, slave_s);
3737 unixctl_command_reply(conn, 501, "no such slave");
3741 bond_enable_slave(iface, enable);
3742 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3746 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3747 void *aux OVS_UNUSED)
3749 enable_slave(conn, args, true);
3753 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3754 void *aux OVS_UNUSED)
3756 enable_slave(conn, args, false);
3760 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3761 void *aux OVS_UNUSED)
3763 char *args = (char *) args_;
3764 uint8_t mac[ETH_ADDR_LEN];
3768 char *mac_s, *vlan_s;
3769 char *save_ptr = NULL;
3771 mac_s = strtok_r(args, " ", &save_ptr);
3772 vlan_s = strtok_r(NULL, " ", &save_ptr);
3775 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3776 unixctl_command_reply(conn, 501, "invalid vlan");
3780 vlan = OFP_VLAN_NONE;
3783 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3784 == ETH_ADDR_SCAN_COUNT) {
3785 hash = bond_hash_src(mac, vlan);
3787 hash_cstr = xasprintf("%u", hash);
3788 unixctl_command_reply(conn, 200, hash_cstr);
3791 unixctl_command_reply(conn, 501, "invalid mac");
3798 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3799 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3800 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3801 unixctl_command_register("bond/set-active-slave",
3802 bond_unixctl_set_active_slave, NULL);
3803 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3805 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3807 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3810 /* Port functions. */
3813 lacp_send_pdu_cb(void *aux, const struct lacp_pdu *pdu)
3815 struct iface *iface = aux;
3816 uint8_t ea[ETH_ADDR_LEN];
3819 error = netdev_get_etheraddr(iface->netdev, ea);
3821 struct ofpbuf packet;
3823 ofpbuf_init(&packet, ETH_HEADER_LEN + LACP_PDU_LEN);
3824 compose_lacp_packet(&packet, ea, pdu);
3825 ofproto_send_packet(iface->port->bridge->ofproto,
3826 iface->dp_ifidx, 0, &packet);
3827 ofpbuf_uninit(&packet);
3829 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3830 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3831 "(%s)", iface->name, strerror(error));
3836 port_run(struct port *port)
3838 if (port->monitor) {
3841 /* Track carrier going up and down on interfaces. */
3842 while (!netdev_monitor_poll(port->monitor, &devname)) {
3843 struct iface *iface;
3845 iface = port_lookup_iface(port, devname);
3847 iface_update_carrier(iface);
3851 } else if (time_msec() >= port->miimon_next_update) {
3854 for (i = 0; i < port->n_ifaces; i++) {
3855 struct iface *iface = port->ifaces[i];
3856 iface_update_carrier(iface);
3858 port->miimon_next_update = time_msec() + port->miimon_interval;
3864 for (i = 0; i < port->n_ifaces; i++) {
3865 struct iface *iface = port->ifaces[i];
3866 lacp_slave_enable(port->lacp, iface, iface->enabled);
3869 lacp_run(port->lacp, lacp_send_pdu_cb);
3876 port_wait(struct port *port)
3878 if (port->monitor) {
3879 netdev_monitor_poll_wait(port->monitor);
3881 poll_timer_wait_until(port->miimon_next_update);
3885 lacp_wait(port->lacp);
3891 static struct port *
3892 port_create(struct bridge *br, const char *name)
3896 port = xzalloc(sizeof *port);
3898 port->port_idx = br->n_ports;
3900 port->trunks = NULL;
3901 port->name = xstrdup(name);
3902 port->active_iface = NULL;
3904 if (br->n_ports >= br->allocated_ports) {
3905 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3908 br->ports[br->n_ports++] = port;
3909 shash_add_assert(&br->port_by_name, port->name, port);
3911 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3918 get_port_other_config(const struct ovsrec_port *port, const char *key,
3919 const char *default_value)
3923 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3925 return value ? value : default_value;
3929 get_interface_other_config(const struct ovsrec_interface *iface,
3930 const char *key, const char *default_value)
3934 value = get_ovsrec_key_value(&iface->header_,
3935 &ovsrec_interface_col_other_config, key);
3936 return value ? value : default_value;
3940 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3942 struct shash new_ifaces;
3945 /* Collect list of new interfaces. */
3946 shash_init(&new_ifaces);
3947 for (i = 0; i < cfg->n_interfaces; i++) {
3948 const char *name = cfg->interfaces[i]->name;
3949 shash_add_once(&new_ifaces, name, NULL);
3952 /* Get rid of deleted interfaces. */
3953 for (i = 0; i < port->n_ifaces; ) {
3954 struct iface *iface = port->ifaces[i];
3955 if (!shash_find(&new_ifaces, iface->name)) {
3956 iface_destroy(iface);
3962 shash_destroy(&new_ifaces);
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 unsigned long *trunks;
3977 /* Update settings. */
3978 port->updelay = cfg->bond_updelay;
3979 if (port->updelay < 0) {
3982 port->downdelay = cfg->bond_downdelay;
3983 if (port->downdelay < 0) {
3984 port->downdelay = 0;
3986 port->bond_rebalance_interval = atoi(
3987 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3988 if (port->bond_rebalance_interval < 1000) {
3989 port->bond_rebalance_interval = 1000;
3991 next_rebalance = time_msec() + port->bond_rebalance_interval;
3992 if (port->bond_next_rebalance > next_rebalance) {
3993 port->bond_next_rebalance = next_rebalance;
3996 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
3999 netdev_monitor_destroy(port->monitor);
4000 port->monitor = NULL;
4002 if (strcmp(detect_mode, "miimon")) {
4003 port->monitor = netdev_monitor_create();
4005 if (strcmp(detect_mode, "carrier")) {
4006 VLOG_WARN("port %s: unsupported bond-detect-mode %s, "
4007 "defaulting to carrier", port->name, detect_mode);
4011 port->miimon_interval = atoi(
4012 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4013 if (port->miimon_interval < 100) {
4014 port->miimon_interval = 100;
4016 miimon_next_update = time_msec() + port->miimon_interval;
4017 if (port->miimon_next_update > miimon_next_update) {
4018 port->miimon_next_update = miimon_next_update;
4021 if (!port->cfg->bond_mode ||
4022 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4023 port->bond_mode = BM_SLB;
4024 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4025 port->bond_mode = BM_AB;
4026 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
4027 port->bond_mode = BM_TCP;
4029 port->bond_mode = BM_SLB;
4030 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4031 port->name, port->cfg->bond_mode,
4032 bond_mode_to_string(port->bond_mode));
4035 /* Add new interfaces and update 'cfg' member of existing ones. */
4036 shash_init(&new_ifaces);
4037 for (i = 0; i < cfg->n_interfaces; i++) {
4038 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4039 struct iface *iface;
4041 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
4042 VLOG_WARN("port %s: %s specified twice as port interface",
4043 port->name, if_cfg->name);
4044 iface_set_ofport(if_cfg, -1);
4048 iface = iface_lookup(port->bridge, if_cfg->name);
4050 if (iface->port != port) {
4051 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4053 port->bridge->name, if_cfg->name, iface->port->name);
4056 iface->cfg = if_cfg;
4058 iface = iface_create(port, if_cfg);
4061 /* Determine interface type. The local port always has type
4062 * "internal". Other ports take their type from the database and
4063 * default to "system" if none is specified. */
4064 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4065 : if_cfg->type[0] ? if_cfg->type
4069 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4072 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4073 iface->lacp_priority = UINT16_MAX;
4075 iface->lacp_priority = lacp_priority;
4078 shash_destroy(&new_ifaces);
4080 port->lacp_fast = !strcmp(get_port_other_config(cfg, "lacp-time", "slow"),
4084 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4086 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4087 /* Prefer bondable links if unspecified. */
4088 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4090 port->lacp_priority = lacp_priority;
4093 if (!port->cfg->lacp) {
4094 /* XXX when LACP implementation has been sufficiently tested, enable by
4095 * default and make active on bonded ports. */
4096 lacp_destroy(port->lacp);
4098 } else if (!strcmp(port->cfg->lacp, "off")) {
4099 lacp_destroy(port->lacp);
4101 } else if (!strcmp(port->cfg->lacp, "active")) {
4103 port->lacp = lacp_create();
4105 port->lacp_active = true;
4106 } else if (!strcmp(port->cfg->lacp, "passive")) {
4108 port->lacp = lacp_create();
4110 port->lacp_active = false;
4112 VLOG_WARN("port %s: unknown LACP mode %s",
4113 port->name, port->cfg->lacp);
4114 lacp_destroy(port->lacp);
4121 if (port->n_ifaces < 2) {
4123 if (vlan >= 0 && vlan <= 4095) {
4124 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4129 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4130 * they even work as-is. But they have not been tested. */
4131 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4135 if (port->vlan != vlan) {
4137 bridge_flush(port->bridge);
4140 /* Get trunked VLANs. */
4142 if (vlan < 0 && cfg->n_trunks) {
4145 trunks = bitmap_allocate(4096);
4147 for (i = 0; i < cfg->n_trunks; i++) {
4148 int trunk = cfg->trunks[i];
4150 bitmap_set1(trunks, trunk);
4156 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4157 port->name, cfg->n_trunks);
4159 if (n_errors == cfg->n_trunks) {
4160 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4162 bitmap_free(trunks);
4165 } else if (vlan >= 0 && cfg->n_trunks) {
4166 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4170 ? port->trunks != NULL
4171 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4172 bridge_flush(port->bridge);
4174 bitmap_free(port->trunks);
4175 port->trunks = trunks;
4179 port_destroy(struct port *port)
4182 struct bridge *br = port->bridge;
4186 for (i = 0; i < MAX_MIRRORS; i++) {
4187 struct mirror *m = br->mirrors[i];
4188 if (m && m->out_port == port) {
4193 while (port->n_ifaces > 0) {
4194 iface_destroy(port->ifaces[port->n_ifaces - 1]);
4197 shash_find_and_delete_assert(&br->port_by_name, port->name);
4199 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
4200 del->port_idx = port->port_idx;
4202 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4204 netdev_monitor_destroy(port->monitor);
4206 bitmap_free(port->trunks);
4213 static struct port *
4214 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4216 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4217 return iface ? iface->port : NULL;
4220 static struct port *
4221 port_lookup(const struct bridge *br, const char *name)
4223 return shash_find_data(&br->port_by_name, name);
4226 static struct iface *
4227 port_lookup_iface(const struct port *port, const char *name)
4229 struct iface *iface = iface_lookup(port->bridge, name);
4230 return iface && iface->port == port ? iface : NULL;
4234 port_update_lacp(struct port *port)
4239 lacp_configure(port->lacp, port->name,
4240 port->bridge->ea, port->lacp_priority,
4241 port->lacp_active, port->lacp_fast);
4243 for (i = 0; i < port->n_ifaces; i++) {
4244 struct iface *iface = port->ifaces[i];
4245 lacp_slave_register(port->lacp, iface, iface->name,
4246 iface->dp_ifidx, iface->lacp_priority);
4252 port_update_bonding(struct port *port)
4254 if (port->n_ifaces < 2) {
4255 /* Not a bonded port. */
4256 free(port->bond_hash);
4257 port->bond_hash = NULL;
4258 port->bond_fake_iface = false;
4259 port->active_iface = NULL;
4260 port->no_ifaces_tag = 0;
4264 if (port->bond_mode != BM_AB && !port->bond_hash) {
4265 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4266 for (i = 0; i <= BOND_MASK; i++) {
4267 struct bond_entry *e = &port->bond_hash[i];
4271 port->bond_next_rebalance
4272 = time_msec() + port->bond_rebalance_interval;
4273 } else if (port->bond_mode == BM_AB) {
4274 free(port->bond_hash);
4275 port->bond_hash = NULL;
4278 if (!port->no_ifaces_tag) {
4279 port->no_ifaces_tag = tag_create_random();
4282 if (!port->active_iface) {
4283 bond_choose_active_iface(port);
4286 port->bond_fake_iface = port->cfg->bond_fake_iface;
4287 if (port->bond_fake_iface) {
4288 port->bond_next_fake_iface_update = time_msec();
4294 /* Interface functions. */
4296 static struct iface *
4297 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4299 struct bridge *br = port->bridge;
4300 struct iface *iface;
4301 char *name = if_cfg->name;
4303 iface = xzalloc(sizeof *iface);
4305 iface->port_ifidx = port->n_ifaces;
4306 iface->name = xstrdup(name);
4307 iface->dp_ifidx = -1;
4308 iface->tag = tag_create_random();
4309 iface->delay_expires = LLONG_MAX;
4310 iface->netdev = NULL;
4311 iface->cfg = if_cfg;
4313 shash_add_assert(&br->iface_by_name, iface->name, iface);
4315 if (port->n_ifaces >= port->allocated_ifaces) {
4316 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
4317 sizeof *port->ifaces);
4319 port->ifaces[port->n_ifaces++] = iface;
4320 if (port->n_ifaces > 1) {
4321 br->has_bonded_ports = true;
4324 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4332 iface_destroy(struct iface *iface)
4335 struct port *port = iface->port;
4336 struct bridge *br = port->bridge;
4337 bool del_active = port->active_iface == iface;
4340 if (iface->port->lacp) {
4341 lacp_slave_unregister(iface->port->lacp, iface);
4344 if (port->monitor && iface->netdev) {
4345 netdev_monitor_remove(port->monitor, iface->netdev);
4348 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4350 if (iface->dp_ifidx >= 0) {
4351 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4354 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
4355 del->port_ifidx = iface->port_ifidx;
4357 netdev_close(iface->netdev);
4360 bond_choose_active_iface(port);
4361 bond_send_learning_packets(port);
4367 bridge_flush(port->bridge);
4371 static struct iface *
4372 iface_lookup(const struct bridge *br, const char *name)
4374 return shash_find_data(&br->iface_by_name, name);
4377 static struct iface *
4378 iface_find(const char *name)
4380 const struct bridge *br;
4382 LIST_FOR_EACH (br, node, &all_bridges) {
4383 struct iface *iface = iface_lookup(br, name);
4392 static struct iface *
4393 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4395 struct iface *iface;
4397 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4398 hash_int(dp_ifidx, 0), &br->ifaces) {
4399 if (iface->dp_ifidx == dp_ifidx) {
4406 /* Set Ethernet address of 'iface', if one is specified in the configuration
4409 iface_set_mac(struct iface *iface)
4411 uint8_t ea[ETH_ADDR_LEN];
4413 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4414 if (eth_addr_is_multicast(ea)) {
4415 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4417 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4418 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4419 iface->name, iface->name);
4421 int error = netdev_set_etheraddr(iface->netdev, ea);
4423 VLOG_ERR("interface %s: setting MAC failed (%s)",
4424 iface->name, strerror(error));
4430 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4432 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4435 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4439 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4441 * The value strings in '*shash' are taken directly from values[], not copied,
4442 * so the caller should not modify or free them. */
4444 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4445 struct shash *shash)
4450 for (i = 0; i < n; i++) {
4451 shash_add(shash, keys[i], values[i]);
4455 /* Creates 'keys' and 'values' arrays from 'shash'.
4457 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4458 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4459 * are populated with with strings taken directly from 'shash' and thus have
4460 * the same ownership of the key-value pairs in shash.
4463 shash_to_ovs_idl_map(struct shash *shash,
4464 char ***keys, char ***values, size_t *n)
4468 struct shash_node *sn;
4470 count = shash_count(shash);
4472 k = xmalloc(count * sizeof *k);
4473 v = xmalloc(count * sizeof *v);
4476 SHASH_FOR_EACH(sn, shash) {
4487 struct iface_delete_queues_cbdata {
4488 struct netdev *netdev;
4489 const struct ovsdb_datum *queues;
4493 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4495 union ovsdb_atom atom;
4497 atom.integer = target;
4498 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4502 iface_delete_queues(unsigned int queue_id,
4503 const struct shash *details OVS_UNUSED, void *cbdata_)
4505 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4507 if (!queue_ids_include(cbdata->queues, queue_id)) {
4508 netdev_delete_queue(cbdata->netdev, queue_id);
4513 iface_update_carrier(struct iface *iface)
4515 bool carrier = iface_get_carrier(iface);
4516 if (carrier == iface->up) {
4520 iface->up = carrier;
4521 if (iface->port->lacp) {
4522 lacp_slave_carrier_changed(iface->port->lacp, iface);
4527 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4529 if (!qos || qos->type[0] == '\0') {
4530 netdev_set_qos(iface->netdev, NULL, NULL);
4532 struct iface_delete_queues_cbdata cbdata;
4533 struct shash details;
4536 /* Configure top-level Qos for 'iface'. */
4537 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4538 qos->n_other_config, &details);
4539 netdev_set_qos(iface->netdev, qos->type, &details);
4540 shash_destroy(&details);
4542 /* Deconfigure queues that were deleted. */
4543 cbdata.netdev = iface->netdev;
4544 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4546 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4548 /* Configure queues for 'iface'. */
4549 for (i = 0; i < qos->n_queues; i++) {
4550 const struct ovsrec_queue *queue = qos->value_queues[i];
4551 unsigned int queue_id = qos->key_queues[i];
4553 shash_from_ovs_idl_map(queue->key_other_config,
4554 queue->value_other_config,
4555 queue->n_other_config, &details);
4556 netdev_set_queue(iface->netdev, queue_id, &details);
4557 shash_destroy(&details);
4563 iface_update_cfm(struct iface *iface)
4567 uint16_t *remote_mps;
4568 struct ovsrec_monitor *mon;
4569 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
4571 mon = iface->cfg->monitor;
4574 ofproto_iface_clear_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
4578 if (netdev_get_etheraddr(iface->netdev, ea)) {
4579 VLOG_WARN("interface %s: Failed to get ethernet address. "
4580 "Skipping Monitor.", iface->name);
4584 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4585 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4589 cfm.mpid = mon->mpid;
4590 cfm.interval = mon->interval ? *mon->interval : 1000;
4592 memcpy(cfm.eth_src, ea, sizeof cfm.eth_src);
4593 memcpy(cfm.maid, maid, sizeof cfm.maid);
4595 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4596 for(i = 0; i < mon->n_remote_mps; i++) {
4597 remote_mps[i] = mon->remote_mps[i]->mpid;
4600 ofproto_iface_set_cfm(iface->port->bridge->ofproto, iface->dp_ifidx,
4601 &cfm, remote_mps, mon->n_remote_mps);
4605 /* Read carrier or miimon status directly from 'iface''s netdev, according to
4606 * how 'iface''s port is configured.
4608 * Returns true if 'iface' is up, false otherwise. */
4610 iface_get_carrier(const struct iface *iface)
4612 return (iface->port->monitor
4613 ? netdev_get_carrier(iface->netdev)
4614 : netdev_get_miimon(iface->netdev));
4617 /* Port mirroring. */
4619 static struct mirror *
4620 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4624 for (i = 0; i < MAX_MIRRORS; i++) {
4625 struct mirror *m = br->mirrors[i];
4626 if (m && uuid_equals(uuid, &m->uuid)) {
4634 mirror_reconfigure(struct bridge *br)
4636 unsigned long *rspan_vlans;
4639 /* Get rid of deleted mirrors. */
4640 for (i = 0; i < MAX_MIRRORS; i++) {
4641 struct mirror *m = br->mirrors[i];
4643 const struct ovsdb_datum *mc;
4644 union ovsdb_atom atom;
4646 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4647 atom.uuid = br->mirrors[i]->uuid;
4648 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4654 /* Add new mirrors and reconfigure existing ones. */
4655 for (i = 0; i < br->cfg->n_mirrors; i++) {
4656 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4657 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4659 mirror_reconfigure_one(m, cfg);
4661 mirror_create(br, cfg);
4665 /* Update port reserved status. */
4666 for (i = 0; i < br->n_ports; i++) {
4667 br->ports[i]->is_mirror_output_port = false;
4669 for (i = 0; i < MAX_MIRRORS; i++) {
4670 struct mirror *m = br->mirrors[i];
4671 if (m && m->out_port) {
4672 m->out_port->is_mirror_output_port = true;
4676 /* Update flooded vlans (for RSPAN). */
4678 if (br->cfg->n_flood_vlans) {
4679 rspan_vlans = bitmap_allocate(4096);
4681 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4682 int64_t vlan = br->cfg->flood_vlans[i];
4683 if (vlan >= 0 && vlan < 4096) {
4684 bitmap_set1(rspan_vlans, vlan);
4685 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4688 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4693 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4699 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4704 for (i = 0; ; i++) {
4705 if (i >= MAX_MIRRORS) {
4706 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4707 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4710 if (!br->mirrors[i]) {
4715 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4718 br->mirrors[i] = m = xzalloc(sizeof *m);
4721 m->name = xstrdup(cfg->name);
4722 shash_init(&m->src_ports);
4723 shash_init(&m->dst_ports);
4729 mirror_reconfigure_one(m, cfg);
4733 mirror_destroy(struct mirror *m)
4736 struct bridge *br = m->bridge;
4739 for (i = 0; i < br->n_ports; i++) {
4740 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4741 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4744 shash_destroy(&m->src_ports);
4745 shash_destroy(&m->dst_ports);
4748 m->bridge->mirrors[m->idx] = NULL;
4757 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4758 struct shash *names)
4762 for (i = 0; i < n_ports; i++) {
4763 const char *name = ports[i]->name;
4764 if (port_lookup(m->bridge, name)) {
4765 shash_add_once(names, name, NULL);
4767 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4768 "port %s", m->bridge->name, m->name, name);
4774 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4780 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4782 for (i = 0; i < cfg->n_select_vlan; i++) {
4783 int64_t vlan = cfg->select_vlan[i];
4784 if (vlan < 0 || vlan > 4095) {
4785 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4786 m->bridge->name, m->name, vlan);
4788 (*vlans)[n_vlans++] = vlan;
4795 vlan_is_mirrored(const struct mirror *m, int vlan)
4799 for (i = 0; i < m->n_vlans; i++) {
4800 if (m->vlans[i] == vlan) {
4808 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4812 for (i = 0; i < m->n_vlans; i++) {
4813 if (port_trunks_vlan(p, m->vlans[i])) {
4821 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4823 struct shash src_ports, dst_ports;
4824 mirror_mask_t mirror_bit;
4825 struct port *out_port;
4832 if (strcmp(cfg->name, m->name)) {
4834 m->name = xstrdup(cfg->name);
4837 /* Get output port. */
4838 if (cfg->output_port) {
4839 out_port = port_lookup(m->bridge, cfg->output_port->name);
4841 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4842 m->bridge->name, m->name);
4848 if (cfg->output_vlan) {
4849 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4850 "output vlan; ignoring output vlan",
4851 m->bridge->name, m->name);
4853 } else if (cfg->output_vlan) {
4855 out_vlan = *cfg->output_vlan;
4857 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4858 m->bridge->name, m->name);
4863 shash_init(&src_ports);
4864 shash_init(&dst_ports);
4865 if (cfg->select_all) {
4866 for (i = 0; i < m->bridge->n_ports; i++) {
4867 const char *name = m->bridge->ports[i]->name;
4868 shash_add_once(&src_ports, name, NULL);
4869 shash_add_once(&dst_ports, name, NULL);
4874 /* Get ports, and drop duplicates and ports that don't exist. */
4875 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4877 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4880 /* Get all the vlans, and drop duplicate and invalid vlans. */
4881 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4884 /* Update mirror data. */
4885 if (!shash_equal_keys(&m->src_ports, &src_ports)
4886 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4887 || m->n_vlans != n_vlans
4888 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4889 || m->out_port != out_port
4890 || m->out_vlan != out_vlan) {
4891 bridge_flush(m->bridge);
4893 shash_swap(&m->src_ports, &src_ports);
4894 shash_swap(&m->dst_ports, &dst_ports);
4897 m->n_vlans = n_vlans;
4898 m->out_port = out_port;
4899 m->out_vlan = out_vlan;
4902 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4903 for (i = 0; i < m->bridge->n_ports; i++) {
4904 struct port *port = m->bridge->ports[i];
4906 if (shash_find(&m->src_ports, port->name)
4909 ? port_trunks_any_mirrored_vlan(m, port)
4910 : vlan_is_mirrored(m, port->vlan)))) {
4911 port->src_mirrors |= mirror_bit;
4913 port->src_mirrors &= ~mirror_bit;
4916 if (shash_find(&m->dst_ports, port->name)) {
4917 port->dst_mirrors |= mirror_bit;
4919 port->dst_mirrors &= ~mirror_bit;
4924 shash_destroy(&src_ports);
4925 shash_destroy(&dst_ports);