1 /* Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
18 #include "byte-order.h"
21 #include <arpa/inet.h>
24 #include <sys/socket.h>
26 #include <openflow/openflow.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
36 #include "classifier.h"
41 #include "dynamic-string.h"
48 #include "mac-learning.h"
52 #include "ofp-print.h"
54 #include "ofproto/netflow.h"
55 #include "ofproto/ofproto.h"
56 #include "ovsdb-data.h"
58 #include "poll-loop.h"
62 #include "socket-util.h"
63 #include "stream-ssl.h"
66 #include "system-stats.h"
71 #include "vswitchd/vswitch-idl.h"
72 #include "xenserver.h"
74 #include "sflow_api.h"
76 VLOG_DEFINE_THIS_MODULE(bridge);
78 COVERAGE_DEFINE(bridge_flush);
79 COVERAGE_DEFINE(bridge_process_flow);
80 COVERAGE_DEFINE(bridge_process_cfm);
81 COVERAGE_DEFINE(bridge_process_lacp);
82 COVERAGE_DEFINE(bridge_reconfigure);
83 COVERAGE_DEFINE(bridge_lacp_update);
91 struct dst builtin[32];
96 static void dst_set_init(struct dst_set *);
97 static void dst_set_add(struct dst_set *, const struct dst *);
98 static void dst_set_free(struct dst_set *);
101 /* These members are always valid. */
102 struct list port_elem; /* Element in struct port's "ifaces" list. */
103 struct port *port; /* Containing port. */
104 char *name; /* Host network device name. */
105 tag_type tag; /* Tag associated with this interface. */
106 long long delay_expires; /* Time after which 'enabled' may change. */
108 /* These members are valid only after bridge_reconfigure() causes them to
110 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
111 int dp_ifidx; /* Index within kernel datapath. */
112 struct netdev *netdev; /* Network device. */
113 bool enabled; /* May be chosen for flows? */
114 bool up; /* Is the interface up? */
115 const char *type; /* Usually same as cfg->type. */
116 const struct ovsrec_interface *cfg;
119 #define BOND_MASK 0xff
121 struct iface *iface; /* Assigned iface, or NULL if none. */
122 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
123 tag_type tag; /* Tag for bond_entry<->iface association. */
127 BM_TCP, /* Transport Layer Load Balance. */
128 BM_SLB, /* Source Load Balance. */
129 BM_AB /* Active Backup. */
132 #define MAX_MIRRORS 32
133 typedef uint32_t mirror_mask_t;
134 #define MIRROR_MASK_C(X) UINT32_C(X)
135 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
137 struct bridge *bridge;
140 struct uuid uuid; /* UUID of this "mirror" record in database. */
142 /* Selection criteria. */
143 struct sset src_ports; /* Source port names. */
144 struct sset dst_ports; /* Destination port names. */
149 struct port *out_port;
153 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
155 struct bridge *bridge;
156 struct hmap_node hmap_node; /* Element in struct bridge's "ports" hmap. */
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;
165 struct netdev_monitor *monitor; /* Tracks carrier. NULL if miimon. */
166 long long int miimon_interval; /* Miimon status refresh interval. */
167 long long int miimon_next_update; /* Time of next miimon update. */
169 /* An ordinary bridge port has 1 interface.
170 * A bridge port for bonding has at least 2 interfaces. */
171 struct list ifaces; /* List of "struct iface"s. */
172 size_t n_ifaces; /* list_size(ifaces). */
175 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
176 struct iface *active_iface; /* iface on which bcasts accepted, or NULL. */
177 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
178 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
179 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
180 long long int bond_next_fake_iface_update; /* Time of next update. */
182 /* LACP information. */
183 struct lacp *lacp; /* LACP object. NULL if LACP is disabled. */
185 /* SLB specific bonding info. */
186 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
187 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
188 long long int bond_next_rebalance; /* Next rebalancing time. */
190 /* Port mirroring info. */
191 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
192 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
193 bool is_mirror_output_port; /* Does port mirroring send frames here? */
197 struct list node; /* Node in global list of bridges. */
198 char *name; /* User-specified arbitrary name. */
199 struct mac_learning *ml; /* MAC learning table. */
200 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
201 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
202 const struct ovsrec_bridge *cfg;
204 /* OpenFlow switch processing. */
205 struct ofproto *ofproto; /* OpenFlow switch. */
207 /* Kernel datapath information. */
208 struct dpif *dpif; /* Datapath. */
209 struct hmap ifaces; /* "struct iface"s indexed by dp_ifidx. */
212 struct hmap ports; /* "struct port"s indexed by name. */
213 struct shash iface_by_name; /* "struct iface"s indexed by name. */
216 bool has_bonded_ports;
221 /* Port mirroring. */
222 struct mirror *mirrors[MAX_MIRRORS];
225 /* List of all bridges. */
226 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
228 /* OVSDB IDL used to obtain configuration. */
229 static struct ovsdb_idl *idl;
231 /* Each time this timer expires, the bridge fetches systems and interface
232 * statistics and pushes them into the database. */
233 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
234 static long long int stats_timer = LLONG_MIN;
236 /* Stores the time after which CFM statistics may be written to the database.
237 * Only updated when changes to the database require rate limiting. */
238 #define CFM_LIMIT_INTERVAL (1 * 1000) /* In milliseconds. */
239 static long long int cfm_limiter = LLONG_MIN;
241 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
242 static void bridge_destroy(struct bridge *);
243 static struct bridge *bridge_lookup(const char *name);
244 static unixctl_cb_func bridge_unixctl_dump_flows;
245 static unixctl_cb_func bridge_unixctl_reconnect;
246 static int bridge_run_one(struct bridge *);
247 static size_t bridge_get_controllers(const struct bridge *br,
248 struct ovsrec_controller ***controllersp);
249 static void bridge_reconfigure_one(struct bridge *);
250 static void bridge_reconfigure_remotes(struct bridge *,
251 const struct sockaddr_in *managers,
253 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
254 static void bridge_fetch_dp_ifaces(struct bridge *);
255 static void bridge_flush(struct bridge *);
256 static void bridge_pick_local_hw_addr(struct bridge *,
257 uint8_t ea[ETH_ADDR_LEN],
258 struct iface **hw_addr_iface);
259 static uint64_t bridge_pick_datapath_id(struct bridge *,
260 const uint8_t bridge_ea[ETH_ADDR_LEN],
261 struct iface *hw_addr_iface);
262 static uint64_t dpid_from_hash(const void *, size_t nbytes);
264 static unixctl_cb_func bridge_unixctl_fdb_show;
265 static unixctl_cb_func cfm_unixctl_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 iface *port_get_an_iface(const struct port *);
284 static struct port *port_from_dp_ifidx(const struct bridge *,
286 static void port_update_bonding(struct port *);
287 static void port_update_lacp(struct port *);
289 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
290 static void mirror_destroy(struct mirror *);
291 static void mirror_reconfigure(struct bridge *);
292 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
293 static bool vlan_is_mirrored(const struct mirror *, int vlan);
295 static struct iface *iface_create(struct port *port,
296 const struct ovsrec_interface *if_cfg);
297 static void iface_destroy(struct iface *);
298 static struct iface *iface_lookup(const struct bridge *, const char *name);
299 static struct iface *iface_find(const char *name);
300 static struct iface *iface_from_dp_ifidx(const struct bridge *,
302 static void iface_set_mac(struct iface *);
303 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
304 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
305 static void iface_update_cfm(struct iface *);
306 static bool iface_refresh_cfm_stats(struct iface *iface);
307 static void iface_update_carrier(struct iface *);
308 static bool iface_get_carrier(const struct iface *);
310 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
312 static void shash_to_ovs_idl_map(struct shash *,
313 char ***keys, char ***values, size_t *n);
315 /* Hooks into ofproto processing. */
316 static struct ofhooks bridge_ofhooks;
318 /* Public functions. */
320 /* Initializes the bridge module, configuring it to obtain its configuration
321 * from an OVSDB server accessed over 'remote', which should be a string in a
322 * form acceptable to ovsdb_idl_create(). */
324 bridge_init(const char *remote)
326 /* Create connection to database. */
327 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
329 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
330 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
331 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
333 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
335 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
336 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
338 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
339 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
340 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
342 /* Register unixctl commands. */
343 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
344 unixctl_command_register("cfm/show", cfm_unixctl_show, NULL);
345 unixctl_command_register("qos/show", qos_unixctl_show, NULL);
346 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
348 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
357 struct bridge *br, *next_br;
359 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
362 ovsdb_idl_destroy(idl);
365 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
366 * but for which the ovs-vswitchd configuration 'cfg' is required. */
368 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
370 static bool already_configured_once;
371 struct sset bridge_names;
372 struct sset dpif_names, dpif_types;
376 /* Only do this once per ovs-vswitchd run. */
377 if (already_configured_once) {
380 already_configured_once = true;
382 stats_timer = time_msec() + STATS_INTERVAL;
384 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
385 sset_init(&bridge_names);
386 for (i = 0; i < cfg->n_bridges; i++) {
387 sset_add(&bridge_names, cfg->bridges[i]->name);
390 /* Iterate over all system dpifs and delete any of them that do not appear
392 sset_init(&dpif_names);
393 sset_init(&dpif_types);
394 dp_enumerate_types(&dpif_types);
395 SSET_FOR_EACH (type, &dpif_types) {
398 dp_enumerate_names(type, &dpif_names);
400 /* Delete each dpif whose name is not in 'bridge_names'. */
401 SSET_FOR_EACH (name, &dpif_names) {
402 if (!sset_contains(&bridge_names, name)) {
406 retval = dpif_open(name, type, &dpif);
414 sset_destroy(&bridge_names);
415 sset_destroy(&dpif_names);
416 sset_destroy(&dpif_types);
419 /* Callback for iterate_and_prune_ifaces(). */
421 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
423 if (!iface->netdev) {
424 /* We already reported a related error, don't bother duplicating it. */
428 if (iface->dp_ifidx < 0) {
429 VLOG_ERR("%s interface not in %s, dropping",
430 iface->name, dpif_name(br->dpif));
434 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
435 iface->name, iface->dp_ifidx);
439 /* Callback for iterate_and_prune_ifaces(). */
441 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
442 void *aux OVS_UNUSED)
444 /* Set policing attributes. */
445 netdev_set_policing(iface->netdev,
446 iface->cfg->ingress_policing_rate,
447 iface->cfg->ingress_policing_burst);
449 /* Set MAC address of internal interfaces other than the local
451 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
452 iface_set_mac(iface);
458 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
459 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
460 * deletes from 'br' any ports that no longer have any interfaces. */
462 iterate_and_prune_ifaces(struct bridge *br,
463 bool (*cb)(struct bridge *, struct iface *,
467 struct port *port, *next_port;
469 HMAP_FOR_EACH_SAFE (port, next_port, hmap_node, &br->ports) {
470 struct iface *iface, *next_iface;
472 LIST_FOR_EACH_SAFE (iface, next_iface, port_elem, &port->ifaces) {
473 if (!cb(br, iface, aux)) {
474 iface_set_ofport(iface->cfg, -1);
475 iface_destroy(iface);
479 if (!port->n_ifaces) {
480 VLOG_WARN("%s port has no interfaces, dropping", port->name);
486 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
487 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
488 * responsible for freeing '*managersp' (with free()).
490 * You may be asking yourself "why does ovs-vswitchd care?", because
491 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
492 * should not be and in fact is not directly involved in that. But
493 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
494 * it has to tell in-band control where the managers are to enable that.
495 * (Thus, only managers connected in-band are collected.)
498 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
499 struct sockaddr_in **managersp, size_t *n_managersp)
501 struct sockaddr_in *managers = NULL;
502 size_t n_managers = 0;
506 /* Collect all of the potential targets from the "targets" columns of the
507 * rows pointed to by "manager_options", excluding any that are
510 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
511 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
513 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
514 sset_find_and_delete(&targets, m->target);
516 sset_add(&targets, m->target);
520 /* Now extract the targets' IP addresses. */
521 if (!sset_is_empty(&targets)) {
524 managers = xmalloc(sset_count(&targets) * sizeof *managers);
525 SSET_FOR_EACH (target, &targets) {
526 struct sockaddr_in *sin = &managers[n_managers];
528 if ((!strncmp(target, "tcp:", 4)
529 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
530 (!strncmp(target, "ssl:", 4)
531 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
536 sset_destroy(&targets);
538 *managersp = managers;
539 *n_managersp = n_managers;
543 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
545 struct shash old_br, new_br;
546 struct shash_node *node;
547 struct bridge *br, *next;
548 struct sockaddr_in *managers;
551 int sflow_bridge_number;
553 COVERAGE_INC(bridge_reconfigure);
555 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
557 /* Collect old and new bridges. */
560 LIST_FOR_EACH (br, node, &all_bridges) {
561 shash_add(&old_br, br->name, br);
563 for (i = 0; i < ovs_cfg->n_bridges; i++) {
564 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
565 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
566 VLOG_WARN("more than one bridge named %s", br_cfg->name);
570 /* Get rid of deleted bridges and add new bridges. */
571 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
572 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
579 SHASH_FOR_EACH (node, &new_br) {
580 const char *br_name = node->name;
581 const struct ovsrec_bridge *br_cfg = node->data;
582 br = shash_find_data(&old_br, br_name);
584 /* If the bridge datapath type has changed, we need to tear it
585 * down and recreate. */
586 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
588 bridge_create(br_cfg);
591 bridge_create(br_cfg);
594 shash_destroy(&old_br);
595 shash_destroy(&new_br);
597 /* Reconfigure all bridges. */
598 LIST_FOR_EACH (br, node, &all_bridges) {
599 bridge_reconfigure_one(br);
602 /* Add and delete ports on all datapaths.
604 * The kernel will reject any attempt to add a given port to a datapath if
605 * that port already belongs to a different datapath, so we must do all
606 * port deletions before any port additions. */
607 LIST_FOR_EACH (br, node, &all_bridges) {
608 struct dpif_port_dump dump;
609 struct shash want_ifaces;
610 struct dpif_port dpif_port;
612 bridge_get_all_ifaces(br, &want_ifaces);
613 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
614 if (!shash_find(&want_ifaces, dpif_port.name)
615 && strcmp(dpif_port.name, br->name)) {
616 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
618 VLOG_WARN("failed to remove %s interface from %s: %s",
619 dpif_port.name, dpif_name(br->dpif),
624 shash_destroy(&want_ifaces);
626 LIST_FOR_EACH (br, node, &all_bridges) {
627 struct shash cur_ifaces, want_ifaces;
628 struct dpif_port_dump dump;
629 struct dpif_port dpif_port;
631 /* Get the set of interfaces currently in this datapath. */
632 shash_init(&cur_ifaces);
633 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
634 struct dpif_port *port_info = xmalloc(sizeof *port_info);
635 dpif_port_clone(port_info, &dpif_port);
636 shash_add(&cur_ifaces, dpif_port.name, port_info);
639 /* Get the set of interfaces we want on this datapath. */
640 bridge_get_all_ifaces(br, &want_ifaces);
642 hmap_clear(&br->ifaces);
643 SHASH_FOR_EACH (node, &want_ifaces) {
644 const char *if_name = node->name;
645 struct iface *iface = node->data;
646 struct dpif_port *dpif_port;
650 type = iface ? iface->type : "internal";
651 dpif_port = shash_find_data(&cur_ifaces, if_name);
653 /* If we have a port or a netdev already, and it's not the type we
654 * want, then delete the port (if any) and close the netdev (if
656 if ((dpif_port && strcmp(dpif_port->type, type))
657 || (iface && iface->netdev
658 && strcmp(type, netdev_get_type(iface->netdev)))) {
660 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
667 netdev_close(iface->netdev);
668 iface->netdev = NULL;
672 /* If the port doesn't exist or we don't have the netdev open,
673 * we need to do more work. */
674 if (!dpif_port || (iface && !iface->netdev)) {
675 struct netdev_options options;
676 struct netdev *netdev;
679 /* First open the network device. */
680 options.name = if_name;
682 options.args = &args;
683 options.ethertype = NETDEV_ETH_TYPE_NONE;
687 shash_from_ovs_idl_map(iface->cfg->key_options,
688 iface->cfg->value_options,
689 iface->cfg->n_options, &args);
691 error = netdev_open(&options, &netdev);
692 shash_destroy(&args);
695 VLOG_WARN("could not open network device %s (%s)",
696 if_name, strerror(error));
700 /* Then add the port if we haven't already. */
702 error = dpif_port_add(br->dpif, netdev, NULL);
704 netdev_close(netdev);
705 if (error == EFBIG) {
706 VLOG_ERR("ran out of valid port numbers on %s",
707 dpif_name(br->dpif));
710 VLOG_WARN("failed to add %s interface to %s: %s",
711 if_name, dpif_name(br->dpif),
718 /* Update 'iface'. */
720 iface->netdev = netdev;
721 iface->enabled = iface_get_carrier(iface);
722 iface->up = iface->enabled;
724 } else if (iface && iface->netdev) {
728 shash_from_ovs_idl_map(iface->cfg->key_options,
729 iface->cfg->value_options,
730 iface->cfg->n_options, &args);
731 netdev_set_config(iface->netdev, &args);
732 shash_destroy(&args);
735 shash_destroy(&want_ifaces);
737 SHASH_FOR_EACH (node, &cur_ifaces) {
738 struct dpif_port *port_info = node->data;
739 dpif_port_destroy(port_info);
742 shash_destroy(&cur_ifaces);
744 sflow_bridge_number = 0;
745 LIST_FOR_EACH (br, node, &all_bridges) {
746 uint8_t ea[ETH_ADDR_LEN];
748 struct iface *local_iface;
749 struct iface *hw_addr_iface;
752 bridge_fetch_dp_ifaces(br);
754 iterate_and_prune_ifaces(br, check_iface, NULL);
756 /* Pick local port hardware address, datapath ID. */
757 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
758 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
760 int error = netdev_set_etheraddr(local_iface->netdev, ea);
762 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
763 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
764 "Ethernet address: %s",
765 br->name, strerror(error));
768 memcpy(br->ea, ea, ETH_ADDR_LEN);
770 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
771 ofproto_set_datapath_id(br->ofproto, dpid);
773 dpid_string = xasprintf("%016"PRIx64, dpid);
774 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
777 /* Set NetFlow configuration on this bridge. */
778 if (br->cfg->netflow) {
779 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
780 struct netflow_options opts;
782 memset(&opts, 0, sizeof opts);
784 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
785 if (nf_cfg->engine_type) {
786 opts.engine_type = *nf_cfg->engine_type;
788 if (nf_cfg->engine_id) {
789 opts.engine_id = *nf_cfg->engine_id;
792 opts.active_timeout = nf_cfg->active_timeout;
793 if (!opts.active_timeout) {
794 opts.active_timeout = -1;
795 } else if (opts.active_timeout < 0) {
796 VLOG_WARN("bridge %s: active timeout interval set to negative "
797 "value, using default instead (%d seconds)", br->name,
798 NF_ACTIVE_TIMEOUT_DEFAULT);
799 opts.active_timeout = -1;
802 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
803 if (opts.add_id_to_iface) {
804 if (opts.engine_id > 0x7f) {
805 VLOG_WARN("bridge %s: netflow port mangling may conflict "
806 "with another vswitch, choose an engine id less "
807 "than 128", br->name);
809 if (hmap_count(&br->ports) > 508) {
810 VLOG_WARN("bridge %s: netflow port mangling will conflict "
811 "with another port when more than 508 ports are "
816 sset_init(&opts.collectors);
817 sset_add_array(&opts.collectors,
818 nf_cfg->targets, nf_cfg->n_targets);
819 if (ofproto_set_netflow(br->ofproto, &opts)) {
820 VLOG_ERR("bridge %s: problem setting netflow collectors",
823 sset_destroy(&opts.collectors);
825 ofproto_set_netflow(br->ofproto, NULL);
828 /* Set sFlow configuration on this bridge. */
829 if (br->cfg->sflow) {
830 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
831 struct ovsrec_controller **controllers;
832 struct ofproto_sflow_options oso;
833 size_t n_controllers;
835 memset(&oso, 0, sizeof oso);
837 sset_init(&oso.targets);
838 sset_add_array(&oso.targets,
839 sflow_cfg->targets, sflow_cfg->n_targets);
841 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
842 if (sflow_cfg->sampling) {
843 oso.sampling_rate = *sflow_cfg->sampling;
846 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
847 if (sflow_cfg->polling) {
848 oso.polling_interval = *sflow_cfg->polling;
851 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
852 if (sflow_cfg->header) {
853 oso.header_len = *sflow_cfg->header;
856 oso.sub_id = sflow_bridge_number++;
857 oso.agent_device = sflow_cfg->agent;
859 oso.control_ip = NULL;
860 n_controllers = bridge_get_controllers(br, &controllers);
861 for (i = 0; i < n_controllers; i++) {
862 if (controllers[i]->local_ip) {
863 oso.control_ip = controllers[i]->local_ip;
867 ofproto_set_sflow(br->ofproto, &oso);
869 sset_destroy(&oso.targets);
871 ofproto_set_sflow(br->ofproto, NULL);
874 /* Update the controller and related settings. It would be more
875 * straightforward to call this from bridge_reconfigure_one(), but we
876 * can't do it there for two reasons. First, and most importantly, at
877 * that point we don't know the dp_ifidx of any interfaces that have
878 * been added to the bridge (because we haven't actually added them to
879 * the datapath). Second, at that point we haven't set the datapath ID
880 * yet; when a controller is configured, resetting the datapath ID will
881 * immediately disconnect from the controller, so it's better to set
882 * the datapath ID before the controller. */
883 bridge_reconfigure_remotes(br, managers, n_managers);
885 LIST_FOR_EACH (br, node, &all_bridges) {
888 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
892 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
893 netdev_monitor_add(port->monitor, iface->netdev);
896 port->miimon_next_update = 0;
899 port_update_lacp(port);
900 port_update_bonding(port);
902 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
903 iface_update_qos(iface, 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 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
974 uint8_t iface_ea[ETH_ADDR_LEN];
975 struct iface *candidate;
978 /* Mirror output ports don't participate. */
979 if (port->is_mirror_output_port) {
983 /* Choose the MAC address to represent the port. */
985 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
986 /* Find the interface with this Ethernet address (if any) so that
987 * we can provide the correct devname to the caller. */
988 LIST_FOR_EACH (candidate, port_elem, &port->ifaces) {
989 uint8_t candidate_ea[ETH_ADDR_LEN];
990 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
991 && eth_addr_equals(iface_ea, candidate_ea)) {
996 /* Choose the interface whose MAC address will represent the port.
997 * The Linux kernel bonding code always chooses the MAC address of
998 * the first slave added to a bond, and the Fedora networking
999 * scripts always add slaves to a bond in alphabetical order, so
1000 * for compatibility we choose the interface with the name that is
1001 * first in alphabetical order. */
1002 LIST_FOR_EACH (candidate, port_elem, &port->ifaces) {
1003 if (!iface || strcmp(candidate->name, iface->name) < 0) {
1008 /* The local port doesn't count (since we're trying to choose its
1009 * MAC address anyway). */
1010 if (iface->dp_ifidx == ODPP_LOCAL) {
1015 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1017 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1018 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1019 iface->name, strerror(error));
1024 /* Compare against our current choice. */
1025 if (!eth_addr_is_multicast(iface_ea) &&
1026 !eth_addr_is_local(iface_ea) &&
1027 !eth_addr_is_reserved(iface_ea) &&
1028 !eth_addr_is_zero(iface_ea) &&
1029 eth_addr_compare_3way(iface_ea, ea) < 0)
1031 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1032 *hw_addr_iface = iface;
1035 if (eth_addr_is_multicast(ea)) {
1036 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1037 *hw_addr_iface = NULL;
1038 VLOG_WARN("bridge %s: using default bridge Ethernet "
1039 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1041 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1042 br->name, ETH_ADDR_ARGS(ea));
1046 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1047 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1048 * an interface on 'br', then that interface must be passed in as
1049 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1050 * 'hw_addr_iface' must be passed in as a null pointer. */
1052 bridge_pick_datapath_id(struct bridge *br,
1053 const uint8_t bridge_ea[ETH_ADDR_LEN],
1054 struct iface *hw_addr_iface)
1057 * The procedure for choosing a bridge MAC address will, in the most
1058 * ordinary case, also choose a unique MAC that we can use as a datapath
1059 * ID. In some special cases, though, multiple bridges will end up with
1060 * the same MAC address. This is OK for the bridges, but it will confuse
1061 * the OpenFlow controller, because each datapath needs a unique datapath
1064 * Datapath IDs must be unique. It is also very desirable that they be
1065 * stable from one run to the next, so that policy set on a datapath
1068 const char *datapath_id;
1071 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1072 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1076 if (hw_addr_iface) {
1078 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1080 * A bridge whose MAC address is taken from a VLAN network device
1081 * (that is, a network device created with vconfig(8) or similar
1082 * tool) will have the same MAC address as a bridge on the VLAN
1083 * device's physical network device.
1085 * Handle this case by hashing the physical network device MAC
1086 * along with the VLAN identifier.
1088 uint8_t buf[ETH_ADDR_LEN + 2];
1089 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1090 buf[ETH_ADDR_LEN] = vlan >> 8;
1091 buf[ETH_ADDR_LEN + 1] = vlan;
1092 return dpid_from_hash(buf, sizeof buf);
1095 * Assume that this bridge's MAC address is unique, since it
1096 * doesn't fit any of the cases we handle specially.
1101 * A purely internal bridge, that is, one that has no non-virtual
1102 * network devices on it at all, is more difficult because it has no
1103 * natural unique identifier at all.
1105 * When the host is a XenServer, we handle this case by hashing the
1106 * host's UUID with the name of the bridge. Names of bridges are
1107 * persistent across XenServer reboots, although they can be reused if
1108 * an internal network is destroyed and then a new one is later
1109 * created, so this is fairly effective.
1111 * When the host is not a XenServer, we punt by using a random MAC
1112 * address on each run.
1114 const char *host_uuid = xenserver_get_host_uuid();
1116 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1117 dpid = dpid_from_hash(combined, strlen(combined));
1123 return eth_addr_to_uint64(bridge_ea);
1127 dpid_from_hash(const void *data, size_t n)
1129 uint8_t hash[SHA1_DIGEST_SIZE];
1131 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1132 sha1_bytes(data, n, hash);
1133 eth_addr_mark_random(hash);
1134 return eth_addr_to_uint64(hash);
1138 iface_refresh_status(struct iface *iface)
1142 enum netdev_flags flags;
1151 if (!netdev_get_status(iface->netdev, &sh)) {
1153 char **keys, **values;
1155 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1156 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1161 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1164 shash_destroy_free_data(&sh);
1166 error = netdev_get_flags(iface->netdev, &flags);
1168 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1171 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1174 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1176 ovsrec_interface_set_duplex(iface->cfg,
1177 netdev_features_is_full_duplex(current)
1179 /* warning: uint64_t -> int64_t conversion */
1180 bps = netdev_features_to_bps(current);
1181 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1184 ovsrec_interface_set_duplex(iface->cfg, NULL);
1185 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1189 ovsrec_interface_set_link_state(iface->cfg,
1190 iface_get_carrier(iface) ? "up" : "down");
1192 error = netdev_get_mtu(iface->netdev, &mtu);
1193 if (!error && mtu != INT_MAX) {
1195 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1198 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1202 /* Writes 'iface''s CFM statistics to the database. Returns true if anything
1203 * changed, false otherwise. */
1205 iface_refresh_cfm_stats(struct iface *iface)
1207 const struct ovsrec_monitor *mon;
1208 const struct cfm *cfm;
1209 bool changed = false;
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 if (mp->n_fault != 1 || mp->fault[0] != rmp->fault) {
1227 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1232 if (mon->n_fault != 1 || mon->fault[0] != cfm->fault) {
1233 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1241 iface_refresh_stats(struct iface *iface)
1247 static const struct iface_stat iface_stats[] = {
1248 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1249 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1250 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1251 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1252 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1253 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1254 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1255 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1256 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1257 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1258 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1259 { "collisions", offsetof(struct netdev_stats, collisions) },
1261 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1262 const struct iface_stat *s;
1264 char *keys[N_STATS];
1265 int64_t values[N_STATS];
1268 struct netdev_stats stats;
1270 /* Intentionally ignore return value, since errors will set 'stats' to
1271 * all-1s, and we will deal with that correctly below. */
1272 netdev_get_stats(iface->netdev, &stats);
1275 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1276 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1277 if (value != UINT64_MAX) {
1284 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1288 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1290 struct ovsdb_datum datum;
1294 get_system_stats(&stats);
1296 ovsdb_datum_from_shash(&datum, &stats);
1297 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1301 static inline const char *
1302 nx_role_to_str(enum nx_role role)
1307 case NX_ROLE_MASTER:
1312 return "*** INVALID ROLE ***";
1317 bridge_refresh_controller_status(const struct bridge *br)
1320 const struct ovsrec_controller *cfg;
1322 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1324 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1325 struct ofproto_controller_info *cinfo =
1326 shash_find_data(&info, cfg->target);
1329 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1330 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1331 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1332 (char **) cinfo->pairs.values,
1335 ovsrec_controller_set_is_connected(cfg, false);
1336 ovsrec_controller_set_role(cfg, NULL);
1337 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1341 ofproto_free_ofproto_controller_info(&info);
1347 const struct ovsrec_open_vswitch *cfg;
1349 bool datapath_destroyed;
1350 bool database_changed;
1353 /* Let each bridge do the work that it needs to do. */
1354 datapath_destroyed = false;
1355 LIST_FOR_EACH (br, node, &all_bridges) {
1356 int error = bridge_run_one(br);
1358 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1359 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1360 "forcing reconfiguration", br->name);
1361 datapath_destroyed = true;
1365 /* (Re)configure if necessary. */
1366 database_changed = ovsdb_idl_run(idl);
1367 cfg = ovsrec_open_vswitch_first(idl);
1369 /* Re-configure SSL. We do this on every trip through the main loop,
1370 * instead of just when the database changes, because the contents of the
1371 * key and certificate files can change without the database changing.
1373 * We do this before bridge_reconfigure() because that function might
1374 * initiate SSL connections and thus requires SSL to be configured. */
1375 if (cfg && cfg->ssl) {
1376 const struct ovsrec_ssl *ssl = cfg->ssl;
1378 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1379 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1382 if (database_changed || datapath_destroyed) {
1384 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1386 bridge_configure_once(cfg);
1387 bridge_reconfigure(cfg);
1389 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1390 ovsdb_idl_txn_commit(txn);
1391 ovsdb_idl_txn_destroy(txn); /* XXX */
1393 /* We still need to reconfigure to avoid dangling pointers to
1394 * now-destroyed ovsrec structures inside bridge data. */
1395 static const struct ovsrec_open_vswitch null_cfg;
1397 bridge_reconfigure(&null_cfg);
1401 /* Refresh system and interface stats if necessary. */
1402 if (time_msec() >= stats_timer) {
1404 struct ovsdb_idl_txn *txn;
1406 txn = ovsdb_idl_txn_create(idl);
1407 LIST_FOR_EACH (br, node, &all_bridges) {
1410 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1411 struct iface *iface;
1413 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
1414 iface_refresh_stats(iface);
1415 iface_refresh_status(iface);
1418 bridge_refresh_controller_status(br);
1420 refresh_system_stats(cfg);
1421 ovsdb_idl_txn_commit(txn);
1422 ovsdb_idl_txn_destroy(txn); /* XXX */
1425 stats_timer = time_msec() + STATS_INTERVAL;
1428 if (time_msec() >= cfm_limiter) {
1429 struct ovsdb_idl_txn *txn;
1430 bool changed = false;
1432 txn = ovsdb_idl_txn_create(idl);
1433 LIST_FOR_EACH (br, node, &all_bridges) {
1436 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1437 struct iface *iface;
1439 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
1440 changed = iface_refresh_cfm_stats(iface) || changed;
1446 cfm_limiter = time_msec() + CFM_LIMIT_INTERVAL;
1449 ovsdb_idl_txn_commit(txn);
1450 ovsdb_idl_txn_destroy(txn);
1459 LIST_FOR_EACH (br, node, &all_bridges) {
1462 ofproto_wait(br->ofproto);
1463 mac_learning_wait(br->ml);
1464 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1468 ovsdb_idl_wait(idl);
1469 poll_timer_wait_until(stats_timer);
1471 if (cfm_limiter > time_msec()) {
1472 poll_timer_wait_until(cfm_limiter);
1476 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1477 * configuration changes. */
1479 bridge_flush(struct bridge *br)
1481 COVERAGE_INC(bridge_flush);
1485 /* Bridge unixctl user interface functions. */
1487 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1488 const char *args, void *aux OVS_UNUSED)
1490 struct ds ds = DS_EMPTY_INITIALIZER;
1491 const struct bridge *br;
1492 const struct mac_entry *e;
1494 br = bridge_lookup(args);
1496 unixctl_command_reply(conn, 501, "no such bridge");
1500 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1501 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1502 struct port *port = e->port.p;
1503 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1504 port_get_an_iface(port)->dp_ifidx,
1505 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1507 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1511 /* CFM unixctl user interface functions. */
1513 cfm_unixctl_show(struct unixctl_conn *conn,
1514 const char *args, void *aux OVS_UNUSED)
1516 struct ds ds = DS_EMPTY_INITIALIZER;
1517 struct iface *iface;
1518 const struct cfm *cfm;
1520 iface = iface_find(args);
1522 unixctl_command_reply(conn, 501, "no such interface");
1526 cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
1529 unixctl_command_reply(conn, 501, "CFM not enabled");
1533 cfm_dump_ds(cfm, &ds);
1534 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1538 /* QoS unixctl user interface functions. */
1540 struct qos_unixctl_show_cbdata {
1542 struct iface *iface;
1546 qos_unixctl_show_cb(unsigned int queue_id,
1547 const struct shash *details,
1550 struct qos_unixctl_show_cbdata *data = aux;
1551 struct ds *ds = data->ds;
1552 struct iface *iface = data->iface;
1553 struct netdev_queue_stats stats;
1554 struct shash_node *node;
1557 ds_put_cstr(ds, "\n");
1559 ds_put_format(ds, "Queue %u:\n", queue_id);
1561 ds_put_cstr(ds, "Default:\n");
1564 SHASH_FOR_EACH (node, details) {
1565 ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data);
1568 error = netdev_get_queue_stats(iface->netdev, queue_id, &stats);
1570 if (stats.tx_packets != UINT64_MAX) {
1571 ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets);
1574 if (stats.tx_bytes != UINT64_MAX) {
1575 ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes);
1578 if (stats.tx_errors != UINT64_MAX) {
1579 ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors);
1582 ds_put_format(ds, "\tFailed to get statistics for queue %u: %s",
1583 queue_id, strerror(error));
1588 qos_unixctl_show(struct unixctl_conn *conn,
1589 const char *args, void *aux OVS_UNUSED)
1591 struct ds ds = DS_EMPTY_INITIALIZER;
1592 struct shash sh = SHASH_INITIALIZER(&sh);
1593 struct iface *iface;
1595 struct shash_node *node;
1596 struct qos_unixctl_show_cbdata data;
1599 iface = iface_find(args);
1601 unixctl_command_reply(conn, 501, "no such interface");
1605 netdev_get_qos(iface->netdev, &type, &sh);
1607 if (*type != '\0') {
1608 ds_put_format(&ds, "QoS: %s %s\n", iface->name, type);
1610 SHASH_FOR_EACH (node, &sh) {
1611 ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data);
1616 error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data);
1619 ds_put_format(&ds, "failed to dump queues: %s", strerror(error));
1621 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1623 ds_put_format(&ds, "QoS not configured on %s\n", iface->name);
1624 unixctl_command_reply(conn, 501, ds_cstr(&ds));
1627 shash_destroy_free_data(&sh);
1631 /* Bridge reconfiguration functions. */
1632 static struct bridge *
1633 bridge_create(const struct ovsrec_bridge *br_cfg)
1638 assert(!bridge_lookup(br_cfg->name));
1639 br = xzalloc(sizeof *br);
1641 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1648 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1651 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1653 dpif_delete(br->dpif);
1654 dpif_close(br->dpif);
1659 br->name = xstrdup(br_cfg->name);
1661 br->ml = mac_learning_create();
1662 eth_addr_nicira_random(br->default_ea);
1664 hmap_init(&br->ports);
1665 hmap_init(&br->ifaces);
1666 shash_init(&br->iface_by_name);
1670 list_push_back(&all_bridges, &br->node);
1672 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1678 bridge_destroy(struct bridge *br)
1681 struct port *port, *next;
1684 HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) {
1687 list_remove(&br->node);
1688 ofproto_destroy(br->ofproto);
1689 error = dpif_delete(br->dpif);
1690 if (error && error != ENOENT) {
1691 VLOG_ERR("failed to delete %s: %s",
1692 dpif_name(br->dpif), strerror(error));
1694 dpif_close(br->dpif);
1695 mac_learning_destroy(br->ml);
1696 hmap_destroy(&br->ifaces);
1697 hmap_destroy(&br->ports);
1698 shash_destroy(&br->iface_by_name);
1704 static struct bridge *
1705 bridge_lookup(const char *name)
1709 LIST_FOR_EACH (br, node, &all_bridges) {
1710 if (!strcmp(br->name, name)) {
1717 /* Handle requests for a listing of all flows known by the OpenFlow
1718 * stack, including those normally hidden. */
1720 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1721 const char *args, void *aux OVS_UNUSED)
1726 br = bridge_lookup(args);
1728 unixctl_command_reply(conn, 501, "Unknown bridge");
1733 ofproto_get_all_flows(br->ofproto, &results);
1735 unixctl_command_reply(conn, 200, ds_cstr(&results));
1736 ds_destroy(&results);
1739 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1740 * connections and reconnect. If BRIDGE is not specified, then all bridges
1741 * drop their controller connections and reconnect. */
1743 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1744 const char *args, void *aux OVS_UNUSED)
1747 if (args[0] != '\0') {
1748 br = bridge_lookup(args);
1750 unixctl_command_reply(conn, 501, "Unknown bridge");
1753 ofproto_reconnect_controllers(br->ofproto);
1755 LIST_FOR_EACH (br, node, &all_bridges) {
1756 ofproto_reconnect_controllers(br->ofproto);
1759 unixctl_command_reply(conn, 200, NULL);
1763 bridge_run_one(struct bridge *br)
1768 error = ofproto_run1(br->ofproto);
1773 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1775 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1779 error = ofproto_run2(br->ofproto, br->flush);
1786 bridge_get_controllers(const struct bridge *br,
1787 struct ovsrec_controller ***controllersp)
1789 struct ovsrec_controller **controllers;
1790 size_t n_controllers;
1792 controllers = br->cfg->controller;
1793 n_controllers = br->cfg->n_controller;
1795 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1801 *controllersp = controllers;
1803 return n_controllers;
1807 bridge_reconfigure_one(struct bridge *br)
1809 enum ofproto_fail_mode fail_mode;
1810 struct port *port, *next;
1811 struct shash_node *node;
1812 struct shash new_ports;
1815 /* Collect new ports. */
1816 shash_init(&new_ports);
1817 for (i = 0; i < br->cfg->n_ports; i++) {
1818 const char *name = br->cfg->ports[i]->name;
1819 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1820 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1825 /* If we have a controller, then we need a local port. Complain if the
1826 * user didn't specify one.
1828 * XXX perhaps we should synthesize a port ourselves in this case. */
1829 if (bridge_get_controllers(br, NULL)) {
1830 char local_name[IF_NAMESIZE];
1833 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1834 local_name, sizeof local_name);
1835 if (!error && !shash_find(&new_ports, local_name)) {
1836 VLOG_WARN("bridge %s: controller specified but no local port "
1837 "(port named %s) defined",
1838 br->name, local_name);
1842 /* Get rid of deleted ports.
1843 * Get rid of deleted interfaces on ports that still exist. */
1844 HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) {
1845 const struct ovsrec_port *port_cfg;
1847 port_cfg = shash_find_data(&new_ports, port->name);
1851 port_del_ifaces(port, port_cfg);
1855 /* Create new ports.
1856 * Add new interfaces to existing ports.
1857 * Reconfigure existing ports. */
1858 SHASH_FOR_EACH (node, &new_ports) {
1859 struct port *port = port_lookup(br, node->name);
1861 port = port_create(br, node->name);
1864 port_reconfigure(port, node->data);
1865 if (!port->n_ifaces) {
1866 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1867 br->name, port->name);
1871 shash_destroy(&new_ports);
1873 /* Set the fail-mode */
1874 fail_mode = !br->cfg->fail_mode
1875 || !strcmp(br->cfg->fail_mode, "standalone")
1876 ? OFPROTO_FAIL_STANDALONE
1877 : OFPROTO_FAIL_SECURE;
1878 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1879 && !ofproto_has_primary_controller(br->ofproto)) {
1880 ofproto_flush_flows(br->ofproto);
1882 ofproto_set_fail_mode(br->ofproto, fail_mode);
1884 /* Delete all flows if we're switching from connected to standalone or vice
1885 * versa. (XXX Should we delete all flows if we are switching from one
1886 * controller to another?) */
1888 /* Configure OpenFlow controller connection snooping. */
1889 if (!ofproto_has_snoops(br->ofproto)) {
1893 sset_add_and_free(&snoops, xasprintf("punix:%s/%s.snoop",
1894 ovs_rundir(), br->name));
1895 ofproto_set_snoops(br->ofproto, &snoops);
1896 sset_destroy(&snoops);
1899 mirror_reconfigure(br);
1902 /* Initializes 'oc' appropriately as a management service controller for
1905 * The caller must free oc->target when it is no longer needed. */
1907 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1908 struct ofproto_controller *oc)
1910 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1911 oc->max_backoff = 0;
1912 oc->probe_interval = 60;
1913 oc->band = OFPROTO_OUT_OF_BAND;
1915 oc->burst_limit = 0;
1918 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1920 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1921 struct ofproto_controller *oc)
1923 oc->target = c->target;
1924 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1925 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1926 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1927 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1928 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1929 oc->burst_limit = (c->controller_burst_limit
1930 ? *c->controller_burst_limit : 0);
1933 /* Configures the IP stack for 'br''s local interface properly according to the
1934 * configuration in 'c'. */
1936 bridge_configure_local_iface_netdev(struct bridge *br,
1937 struct ovsrec_controller *c)
1939 struct netdev *netdev;
1940 struct in_addr mask, gateway;
1942 struct iface *local_iface;
1945 /* If there's no local interface or no IP address, give up. */
1946 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
1947 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1951 /* Bring up the local interface. */
1952 netdev = local_iface->netdev;
1953 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1955 /* Configure the IP address and netmask. */
1956 if (!c->local_netmask
1957 || !inet_aton(c->local_netmask, &mask)
1959 mask.s_addr = guess_netmask(ip.s_addr);
1961 if (!netdev_set_in4(netdev, ip, mask)) {
1962 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1963 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1966 /* Configure the default gateway. */
1967 if (c->local_gateway
1968 && inet_aton(c->local_gateway, &gateway)
1969 && gateway.s_addr) {
1970 if (!netdev_add_router(netdev, gateway)) {
1971 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1972 br->name, IP_ARGS(&gateway.s_addr));
1978 bridge_reconfigure_remotes(struct bridge *br,
1979 const struct sockaddr_in *managers,
1982 const char *disable_ib_str, *queue_id_str;
1983 bool disable_in_band = false;
1986 struct ovsrec_controller **controllers;
1987 size_t n_controllers;
1990 struct ofproto_controller *ocs;
1994 /* Check if we should disable in-band control on this bridge. */
1995 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1996 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1997 disable_in_band = true;
2000 /* Set OpenFlow queue ID for in-band control. */
2001 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
2002 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
2003 ofproto_set_in_band_queue(br->ofproto, queue_id);
2005 if (disable_in_band) {
2006 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
2008 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
2010 had_primary = ofproto_has_primary_controller(br->ofproto);
2012 n_controllers = bridge_get_controllers(br, &controllers);
2014 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
2017 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
2018 for (i = 0; i < n_controllers; i++) {
2019 struct ovsrec_controller *c = controllers[i];
2021 if (!strncmp(c->target, "punix:", 6)
2022 || !strncmp(c->target, "unix:", 5)) {
2023 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2025 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
2026 * domain sockets and overwriting arbitrary local files. */
2027 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2028 "\"%s\" due to possibility for remote exploit",
2029 dpif_name(br->dpif), c->target);
2033 bridge_configure_local_iface_netdev(br, c);
2034 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2035 if (disable_in_band) {
2036 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2041 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2042 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2045 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2046 ofproto_flush_flows(br->ofproto);
2049 /* If there are no controllers and the bridge is in standalone
2050 * mode, set up a flow that matches every packet and directs
2051 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2052 * switch is in secure mode and we won't pass any traffic until
2053 * a controller has been defined and it tells us to do so. */
2055 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2056 union ofp_action action;
2057 struct cls_rule rule;
2059 memset(&action, 0, sizeof action);
2060 action.type = htons(OFPAT_OUTPUT);
2061 action.output.len = htons(sizeof action);
2062 action.output.port = htons(OFPP_NORMAL);
2063 cls_rule_init_catchall(&rule, 0);
2064 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2069 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2074 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2075 struct iface *iface;
2077 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2078 shash_add_once(ifaces, iface->name, iface);
2080 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2081 shash_add_once(ifaces, port->name, NULL);
2086 /* For robustness, in case the administrator moves around datapath ports behind
2087 * our back, we re-check all the datapath port numbers here.
2089 * This function will set the 'dp_ifidx' members of interfaces that have
2090 * disappeared to -1, so only call this function from a context where those
2091 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2092 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2093 * datapath, which doesn't support UINT16_MAX+1 ports. */
2095 bridge_fetch_dp_ifaces(struct bridge *br)
2097 struct dpif_port_dump dump;
2098 struct dpif_port dpif_port;
2101 /* Reset all interface numbers. */
2102 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2103 struct iface *iface;
2105 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2106 iface->dp_ifidx = -1;
2109 hmap_clear(&br->ifaces);
2111 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2112 struct iface *iface = iface_lookup(br, dpif_port.name);
2114 if (iface->dp_ifidx >= 0) {
2115 VLOG_WARN("%s reported interface %s twice",
2116 dpif_name(br->dpif), dpif_port.name);
2117 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2118 VLOG_WARN("%s reported interface %"PRIu16" twice",
2119 dpif_name(br->dpif), dpif_port.port_no);
2121 iface->dp_ifidx = dpif_port.port_no;
2122 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2123 hash_int(iface->dp_ifidx, 0));
2126 iface_set_ofport(iface->cfg,
2127 (iface->dp_ifidx >= 0
2128 ? odp_port_to_ofp_port(iface->dp_ifidx)
2134 /* Bridge packet processing functions. */
2137 bond_is_tcp_hash(const struct port *port)
2139 return port->bond_mode == BM_TCP && lacp_negotiated(port->lacp);
2143 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2145 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2148 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2150 struct flow hash_flow;
2152 memcpy(&hash_flow, flow, sizeof hash_flow);
2153 hash_flow.vlan_tci = 0;
2155 /* The symmetric quality of this hash function is not required, but
2156 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2157 * purposes, so we use it out of convenience. */
2158 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2161 static struct bond_entry *
2162 lookup_bond_entry(const struct port *port, const struct flow *flow,
2165 assert(port->bond_mode != BM_AB);
2167 if (bond_is_tcp_hash(port)) {
2168 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2170 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2174 static struct iface *
2175 bond_choose_iface(const struct port *port)
2177 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2178 struct iface *best_down_slave;
2179 struct iface *iface;
2181 best_down_slave = NULL;
2182 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2183 if (iface->enabled) {
2185 } else if ((!best_down_slave
2186 || iface->delay_expires < best_down_slave->delay_expires)
2187 && lacp_slave_may_enable(port->lacp, iface)) {
2188 best_down_slave = iface;
2192 if (best_down_slave) {
2193 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2194 "since no other interface is up",
2195 best_down_slave->name,
2196 best_down_slave->delay_expires - time_msec());
2197 bond_enable_slave(best_down_slave, true);
2200 return best_down_slave;
2204 choose_output_iface(const struct port *port, const struct flow *flow,
2205 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2207 struct iface *iface;
2209 assert(port->n_ifaces);
2210 if (port->n_ifaces == 1) {
2211 iface = port_get_an_iface(port);
2212 } else if (port->bond_mode == BM_AB) {
2213 iface = port->active_iface;
2215 *tags |= port->no_ifaces_tag;
2219 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2220 if (!e->iface || !e->iface->enabled) {
2221 /* XXX select interface properly. The current interface selection
2222 * is only good for testing the rebalancing code. */
2223 e->iface = bond_choose_iface(port);
2225 *tags |= port->no_ifaces_tag;
2228 e->tag = tag_create_random();
2233 *dp_ifidx = iface->dp_ifidx;
2234 *tags |= iface->tag; /* Currently only used for bonding. */
2239 bond_link_status_update(struct iface *iface)
2241 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2242 struct port *port = iface->port;
2243 bool up = iface->up && lacp_slave_may_enable(port->lacp, iface);
2244 int updelay, downdelay;
2246 updelay = port->updelay;
2247 downdelay = port->downdelay;
2249 if (lacp_negotiated(port->lacp)) {
2254 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2255 /* Nothing to do. */
2258 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2259 iface->name, up ? "up" : "down");
2260 if (up == iface->enabled) {
2261 iface->delay_expires = LLONG_MAX;
2262 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2263 iface->name, up ? "disabled" : "enabled");
2264 } else if (up && !port->active_iface) {
2265 bond_enable_slave(iface, true);
2267 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2268 "other interface is up", iface->name, updelay);
2271 int delay = up ? updelay : downdelay;
2272 iface->delay_expires = time_msec() + delay;
2275 "interface %s: will be %s if it stays %s for %d ms",
2277 up ? "enabled" : "disabled",
2285 bond_choose_active_iface(struct port *port)
2287 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2289 port->active_iface = bond_choose_iface(port);
2290 if (port->active_iface) {
2291 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2292 port->name, port->active_iface->name);
2294 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2300 bond_enable_slave(struct iface *iface, bool enable)
2302 struct port *port = iface->port;
2303 struct bridge *br = port->bridge;
2305 /* This acts as a recursion check. If the act of disabling a slave
2306 * causes a different slave to be enabled, the flag will allow us to
2307 * skip redundant work when we reenter this function. It must be
2308 * cleared on exit to keep things safe with multiple bonds. */
2309 static bool moving_active_iface = false;
2311 iface->delay_expires = LLONG_MAX;
2312 if (enable == iface->enabled) {
2316 iface->enabled = enable;
2317 if (!iface->enabled) {
2318 VLOG_WARN("interface %s: disabled", iface->name);
2319 ofproto_revalidate(br->ofproto, iface->tag);
2320 if (iface == port->active_iface) {
2321 /* Disabling a slave can lead to another slave being immediately
2322 * enabled if there will be no active slaves but one is waiting
2323 * on an updelay. In this case we do not need to run most of the
2324 * code for the newly enabled slave since there was no period
2325 * without an active slave and it is redundant with the disabling
2327 moving_active_iface = true;
2328 bond_choose_active_iface(port);
2330 bond_send_learning_packets(port);
2332 VLOG_WARN("interface %s: enabled", iface->name);
2333 if (!port->active_iface && !moving_active_iface) {
2334 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2335 bond_choose_active_iface(port);
2336 bond_send_learning_packets(port);
2338 iface->tag = tag_create_random();
2341 moving_active_iface = false;
2344 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2345 * bond interface. */
2347 bond_update_fake_iface_stats(struct port *port)
2349 struct netdev_stats bond_stats;
2350 struct netdev *bond_dev;
2351 struct iface *iface;
2353 memset(&bond_stats, 0, sizeof bond_stats);
2355 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2356 struct netdev_stats slave_stats;
2358 if (!netdev_get_stats(iface->netdev, &slave_stats)) {
2359 /* XXX: We swap the stats here because they are swapped back when
2360 * reported by the internal device. The reason for this is
2361 * internal devices normally represent packets going into the system
2362 * but when used as fake bond device they represent packets leaving
2363 * the system. We really should do this in the internal device
2364 * itself because changing it here reverses the counts from the
2365 * perspective of the switch. However, the internal device doesn't
2366 * know what type of device it represents so we have to do it here
2368 bond_stats.tx_packets += slave_stats.rx_packets;
2369 bond_stats.tx_bytes += slave_stats.rx_bytes;
2370 bond_stats.rx_packets += slave_stats.tx_packets;
2371 bond_stats.rx_bytes += slave_stats.tx_bytes;
2375 if (!netdev_open_default(port->name, &bond_dev)) {
2376 netdev_set_stats(bond_dev, &bond_stats);
2377 netdev_close(bond_dev);
2382 bond_run(struct port *port)
2384 struct iface *iface;
2386 if (port->n_ifaces < 2) {
2390 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2391 bond_link_status_update(iface);
2394 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2395 if (time_msec() >= iface->delay_expires) {
2396 bond_enable_slave(iface, !iface->enabled);
2400 if (port->bond_fake_iface
2401 && time_msec() >= port->bond_next_fake_iface_update) {
2402 bond_update_fake_iface_stats(port);
2403 port->bond_next_fake_iface_update = time_msec() + 1000;
2408 bond_wait(struct port *port)
2410 struct iface *iface;
2412 if (port->n_ifaces < 2) {
2416 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2417 if (iface->delay_expires != LLONG_MAX) {
2418 poll_timer_wait_until(iface->delay_expires);
2422 if (port->bond_fake_iface) {
2423 poll_timer_wait_until(port->bond_next_fake_iface_update);
2428 set_dst(struct dst *dst, const struct flow *flow,
2429 const struct port *in_port, const struct port *out_port,
2432 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2433 : in_port->vlan >= 0 ? in_port->vlan
2434 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2435 : vlan_tci_to_vid(flow->vlan_tci));
2436 return choose_output_iface(out_port, flow, dst->vlan,
2437 &dst->dp_ifidx, tags);
2441 swap_dst(struct dst *p, struct dst *q)
2443 struct dst tmp = *p;
2448 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2449 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2450 * that we push to the datapath. We could in fact fully sort the array by
2451 * vlan, but in most cases there are at most two different vlan tags so that's
2452 * possibly overkill.) */
2454 partition_dsts(struct dst_set *set, int vlan)
2456 struct dst *first = set->dsts;
2457 struct dst *last = set->dsts + set->n;
2459 while (first != last) {
2461 * - All dsts < first have vlan == 'vlan'.
2462 * - All dsts >= last have vlan != 'vlan'.
2463 * - first < last. */
2464 while (first->vlan == vlan) {
2465 if (++first == last) {
2470 /* Same invariants, plus one additional:
2471 * - first->vlan != vlan.
2473 while (last[-1].vlan != vlan) {
2474 if (--last == first) {
2479 /* Same invariants, plus one additional:
2480 * - last[-1].vlan == vlan.*/
2481 swap_dst(first++, --last);
2486 mirror_mask_ffs(mirror_mask_t mask)
2488 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2493 dst_set_init(struct dst_set *set)
2495 set->dsts = set->builtin;
2497 set->allocated = ARRAY_SIZE(set->builtin);
2501 dst_set_add(struct dst_set *set, const struct dst *dst)
2503 if (set->n >= set->allocated) {
2504 size_t new_allocated;
2505 struct dst *new_dsts;
2507 new_allocated = set->allocated * 2;
2508 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2509 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2513 set->dsts = new_dsts;
2514 set->allocated = new_allocated;
2516 set->dsts[set->n++] = *dst;
2520 dst_set_free(struct dst_set *set)
2522 if (set->dsts != set->builtin) {
2528 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2531 for (i = 0; i < set->n; i++) {
2532 if (set->dsts[i].vlan == test->vlan
2533 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2541 port_trunks_vlan(const struct port *port, uint16_t vlan)
2543 return (port->vlan < 0
2544 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2548 port_includes_vlan(const struct port *port, uint16_t vlan)
2550 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2554 port_is_floodable(const struct port *port)
2556 struct iface *iface;
2558 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2559 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2567 /* Returns the tag for 'port''s active iface, or 'port''s no_ifaces_tag if
2568 * there is no active iface. */
2570 port_get_active_iface_tag(const struct port *port)
2572 return (port->active_iface
2573 ? port->active_iface->tag
2574 : port->no_ifaces_tag);
2577 /* Returns an arbitrary interface within 'port'.
2579 * 'port' must have at least one interface. */
2580 static struct iface *
2581 port_get_an_iface(const struct port *port)
2583 return CONTAINER_OF(list_front(&port->ifaces), struct iface, port_elem);
2587 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2588 const struct port *in_port, const struct port *out_port,
2589 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2591 mirror_mask_t mirrors = in_port->src_mirrors;
2595 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2596 if (flow_vlan == 0) {
2597 flow_vlan = OFP_VLAN_NONE;
2600 if (out_port == FLOOD_PORT) {
2603 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2605 && port_is_floodable(port)
2606 && port_includes_vlan(port, vlan)
2607 && !port->is_mirror_output_port
2608 && set_dst(&dst, flow, in_port, port, tags)) {
2609 mirrors |= port->dst_mirrors;
2610 dst_set_add(set, &dst);
2613 *nf_output_iface = NF_OUT_FLOOD;
2614 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2615 dst_set_add(set, &dst);
2616 *nf_output_iface = dst.dp_ifidx;
2617 mirrors |= out_port->dst_mirrors;
2621 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2622 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2624 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2625 && !dst_is_duplicate(set, &dst)) {
2626 dst_set_add(set, &dst);
2631 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2632 if (port_includes_vlan(port, m->out_vlan)
2633 && set_dst(&dst, flow, in_port, port, tags))
2635 if (port->vlan < 0) {
2636 dst.vlan = m->out_vlan;
2638 if (dst_is_duplicate(set, &dst)) {
2642 /* Use the vlan tag on the original flow instead of
2643 * the one passed in the vlan parameter. This ensures
2644 * that we compare the vlan from before any implicit
2645 * tagging tags place. This is necessary because
2646 * dst->vlan is the final vlan, after removing implicit
2648 if (port == in_port && dst.vlan == flow_vlan) {
2649 /* Don't send out input port on same VLAN. */
2652 dst_set_add(set, &dst);
2657 mirrors &= mirrors - 1;
2660 partition_dsts(set, flow_vlan);
2663 static void OVS_UNUSED
2664 print_dsts(const struct dst_set *set)
2668 for (i = 0; i < set->n; i++) {
2669 const struct dst *dst = &set->dsts[i];
2671 printf(">p%"PRIu16, dst->dp_ifidx);
2672 if (dst->vlan != OFP_VLAN_NONE) {
2673 printf("v%"PRIu16, dst->vlan);
2679 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2680 const struct port *in_port, const struct port *out_port,
2681 tag_type *tags, struct ofpbuf *actions,
2682 uint16_t *nf_output_iface)
2689 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2692 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2693 if (cur_vlan == 0) {
2694 cur_vlan = OFP_VLAN_NONE;
2696 for (i = 0; i < set.n; i++) {
2697 const struct dst *dst = &set.dsts[i];
2698 if (dst->vlan != cur_vlan) {
2699 if (dst->vlan == OFP_VLAN_NONE) {
2700 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2703 tci = htons(dst->vlan & VLAN_VID_MASK);
2704 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2705 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2707 cur_vlan = dst->vlan;
2709 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2714 /* Returns the effective vlan of a packet, taking into account both the
2715 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2716 * the packet is untagged and -1 indicates it has an invalid header and
2717 * should be dropped. */
2718 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2719 struct port *in_port, bool have_packet)
2721 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2722 if (in_port->vlan >= 0) {
2725 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2726 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2727 "packet received on port %s configured with "
2728 "implicit VLAN %"PRIu16,
2729 br->name, vlan, in_port->name, in_port->vlan);
2733 vlan = in_port->vlan;
2735 if (!port_includes_vlan(in_port, vlan)) {
2737 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2738 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2739 "packet received on port %s not configured for "
2741 br->name, vlan, in_port->name, vlan);
2750 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2751 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2752 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2754 is_gratuitous_arp(const struct flow *flow)
2756 return (flow->dl_type == htons(ETH_TYPE_ARP)
2757 && eth_addr_is_broadcast(flow->dl_dst)
2758 && (flow->nw_proto == ARP_OP_REPLY
2759 || (flow->nw_proto == ARP_OP_REQUEST
2760 && flow->nw_src == flow->nw_dst)));
2764 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2765 struct port *in_port)
2767 struct mac_entry *mac;
2769 if (!mac_learning_may_learn(br->ml, flow->dl_src, vlan)) {
2773 mac = mac_learning_insert(br->ml, flow->dl_src, vlan);
2774 if (is_gratuitous_arp(flow)) {
2775 /* We don't want to learn from gratuitous ARP packets that are
2776 * reflected back over bond slaves so we lock the learning table. */
2777 if (in_port->n_ifaces == 1) {
2778 mac_entry_set_grat_arp_lock(mac);
2779 } else if (mac_entry_is_grat_arp_locked(mac)) {
2784 if (mac_entry_is_new(mac) || mac->port.p != in_port) {
2785 /* The log messages here could actually be useful in debugging,
2786 * so keep the rate limit relatively high. */
2787 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
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);
2793 mac->port.p = in_port;
2794 ofproto_revalidate(br->ofproto, mac_learning_changed(br->ml, mac));
2798 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2799 * dropped. Returns true if they may be forwarded, false if they should be
2802 * If 'have_packet' is true, it indicates that the caller is processing a
2803 * received packet. If 'have_packet' is false, then the caller is just
2804 * revalidating an existing flow because configuration has changed. Either
2805 * way, 'have_packet' only affects logging (there is no point in logging errors
2806 * during revalidation).
2808 * Sets '*in_portp' to the input port. This will be a null pointer if
2809 * flow->in_port does not designate a known input port (in which case
2810 * is_admissible() returns false).
2812 * When returning true, sets '*vlanp' to the effective VLAN of the input
2813 * packet, as returned by flow_get_vlan().
2815 * May also add tags to '*tags', although the current implementation only does
2816 * so in one special case.
2819 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2820 tag_type *tags, int *vlanp, struct port **in_portp)
2822 struct iface *in_iface;
2823 struct port *in_port;
2826 /* Find the interface and port structure for the received packet. */
2827 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2829 /* No interface? Something fishy... */
2831 /* Odd. A few possible reasons here:
2833 * - We deleted an interface but there are still a few packets
2834 * queued up from it.
2836 * - Someone externally added an interface (e.g. with "ovs-dpctl
2837 * add-if") that we don't know about.
2839 * - Packet arrived on the local port but the local port is not
2840 * one of our bridge ports.
2842 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2844 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2845 "interface %"PRIu16, br->name, flow->in_port);
2851 *in_portp = in_port = in_iface->port;
2852 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2857 /* Drop frames for reserved multicast addresses. */
2858 if (eth_addr_is_reserved(flow->dl_dst)) {
2862 /* Drop frames on ports reserved for mirroring. */
2863 if (in_port->is_mirror_output_port) {
2865 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2866 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2867 "%s, which is reserved exclusively for mirroring",
2868 br->name, in_port->name);
2873 /* When using LACP, do not accept packets from disabled interfaces. */
2874 if (lacp_negotiated(in_port->lacp) && !in_iface->enabled) {
2878 /* Packets received on non-LACP bonds need special attention to avoid
2880 if (in_port->n_ifaces > 1 && !lacp_negotiated(in_port->lacp)) {
2881 struct mac_entry *mac;
2883 if (eth_addr_is_multicast(flow->dl_dst)) {
2884 *tags |= port_get_active_iface_tag(in_port);
2885 if (in_port->active_iface != in_iface) {
2886 /* Drop all multicast packets on inactive slaves. */
2891 /* Drop all packets for which we have learned a different input
2892 * port, because we probably sent the packet on one slave and got
2893 * it back on the other. Gratuitous ARP packets are an exception
2894 * to this rule: the host has moved to another switch. The exception
2895 * to the exception is if we locked the learning table to avoid
2896 * reflections on bond slaves. If this is the case, just drop the
2898 mac = mac_learning_lookup(br->ml, flow->dl_src, vlan, NULL);
2899 if (mac && mac->port.p != in_port &&
2900 (!is_gratuitous_arp(flow) || mac_entry_is_grat_arp_locked(mac))) {
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;
2918 struct mac_entry *mac;
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 mac = mac_learning_lookup(br->ml, flow->dl_dst, vlan, tags);
2935 out_port = mac->port.p;
2936 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2937 /* If we are revalidating but don't have a learning entry then
2938 * eject the flow. Installing a flow that floods packets opens
2939 * up a window of time where we could learn from a packet reflected
2940 * on a bond and blackhole packets before the learning table is
2941 * updated to reflect the correct port. */
2944 out_port = FLOOD_PORT;
2947 /* Don't send packets out their input ports. */
2948 if (in_port == out_port) {
2954 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2962 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2963 struct ofpbuf *actions, tag_type *tags,
2964 uint16_t *nf_output_iface, void *br_)
2966 struct bridge *br = br_;
2968 COVERAGE_INC(bridge_process_flow);
2969 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2973 bridge_special_ofhook_cb(const struct flow *flow,
2974 const struct ofpbuf *packet, void *br_)
2976 struct iface *iface;
2977 struct bridge *br = br_;
2979 iface = iface_from_dp_ifidx(br, flow->in_port);
2981 if (flow->dl_type == htons(ETH_TYPE_LACP)) {
2983 if (iface && iface->port->lacp && packet) {
2984 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
2987 COVERAGE_INC(bridge_process_lacp);
2988 lacp_process_pdu(iface->port->lacp, iface, pdu);
2998 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
2999 const struct nlattr *actions,
3001 uint64_t n_bytes, void *br_)
3003 struct bridge *br = br_;
3004 const struct nlattr *a;
3005 struct port *in_port;
3010 /* Feed information from the active flows back into the learning table to
3011 * ensure that table is always in sync with what is actually flowing
3012 * through the datapath.
3014 * We test that 'tags' is nonzero to ensure that only flows that include an
3015 * OFPP_NORMAL action are used for learning. This works because
3016 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3017 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3018 update_learning_table(br, flow, vlan, in_port);
3021 /* Account for bond slave utilization. */
3022 if (!br->has_bonded_ports) {
3025 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3026 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3027 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3028 if (out_port && out_port->n_ifaces >= 2 &&
3029 out_port->bond_mode != BM_AB) {
3030 uint16_t vlan = (flow->vlan_tci
3031 ? vlan_tci_to_vid(flow->vlan_tci)
3033 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
3034 e->tx_bytes += n_bytes;
3041 bridge_account_checkpoint_ofhook_cb(void *br_)
3043 struct bridge *br = br_;
3047 if (!br->has_bonded_ports) {
3052 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3053 if (port->n_ifaces > 1 && port->bond_mode != BM_AB
3054 && now >= port->bond_next_rebalance) {
3055 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3056 bond_rebalance_port(port);
3061 static struct ofhooks bridge_ofhooks = {
3062 bridge_normal_ofhook_cb,
3063 bridge_special_ofhook_cb,
3064 bridge_account_flow_ofhook_cb,
3065 bridge_account_checkpoint_ofhook_cb,
3068 /* Bonding functions. */
3070 /* Statistics for a single interface on a bonded port, used for load-based
3071 * bond rebalancing. */
3072 struct slave_balance {
3073 struct iface *iface; /* The interface. */
3074 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3076 /* All the "bond_entry"s that are assigned to this interface, in order of
3077 * increasing tx_bytes. */
3078 struct bond_entry **hashes;
3083 bond_mode_to_string(enum bond_mode bm) {
3084 static char *bm_slb = "balance-slb";
3085 static char *bm_ab = "active-backup";
3086 static char *bm_tcp = "balance-tcp";
3089 case BM_SLB: return bm_slb;
3090 case BM_AB: return bm_ab;
3091 case BM_TCP: return bm_tcp;
3098 /* Sorts pointers to pointers to bond_entries in ascending order by the
3099 * interface to which they are assigned, and within a single interface in
3100 * ascending order of bytes transmitted. */
3102 compare_bond_entries(const void *a_, const void *b_)
3104 const struct bond_entry *const *ap = a_;
3105 const struct bond_entry *const *bp = b_;
3106 const struct bond_entry *a = *ap;
3107 const struct bond_entry *b = *bp;
3108 if (a->iface != b->iface) {
3109 return a->iface > b->iface ? 1 : -1;
3110 } else if (a->tx_bytes != b->tx_bytes) {
3111 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3117 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3118 * *descending* order by number of bytes transmitted. */
3120 compare_slave_balance(const void *a_, const void *b_)
3122 const struct slave_balance *a = a_;
3123 const struct slave_balance *b = b_;
3124 if (a->iface->enabled != b->iface->enabled) {
3125 return a->iface->enabled ? -1 : 1;
3126 } else if (a->tx_bytes != b->tx_bytes) {
3127 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3134 swap_bals(struct slave_balance *a, struct slave_balance *b)
3136 struct slave_balance tmp = *a;
3141 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3142 * given that 'p' (and only 'p') might be in the wrong location.
3144 * This function invalidates 'p', since it might now be in a different memory
3147 resort_bals(struct slave_balance *p,
3148 struct slave_balance bals[], size_t n_bals)
3151 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3152 swap_bals(p, p - 1);
3154 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3155 swap_bals(p, p + 1);
3161 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3163 if (VLOG_IS_DBG_ENABLED()) {
3164 struct ds ds = DS_EMPTY_INITIALIZER;
3165 const struct slave_balance *b;
3167 for (b = bals; b < bals + n_bals; b++) {
3171 ds_put_char(&ds, ',');
3173 ds_put_format(&ds, " %s %"PRIu64"kB",
3174 b->iface->name, b->tx_bytes / 1024);
3176 if (!b->iface->enabled) {
3177 ds_put_cstr(&ds, " (disabled)");
3179 if (b->n_hashes > 0) {
3180 ds_put_cstr(&ds, " (");
3181 for (i = 0; i < b->n_hashes; i++) {
3182 const struct bond_entry *e = b->hashes[i];
3184 ds_put_cstr(&ds, " + ");
3186 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3187 e - port->bond_hash, e->tx_bytes / 1024);
3189 ds_put_cstr(&ds, ")");
3192 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3197 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3199 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3202 struct bond_entry *hash = from->hashes[hash_idx];
3203 struct port *port = from->iface->port;
3204 uint64_t delta = hash->tx_bytes;
3206 assert(port->bond_mode != BM_AB);
3208 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3209 "from %s to %s (now carrying %"PRIu64"kB and "
3210 "%"PRIu64"kB load, respectively)",
3211 port->name, delta / 1024, hash - port->bond_hash,
3212 from->iface->name, to->iface->name,
3213 (from->tx_bytes - delta) / 1024,
3214 (to->tx_bytes + delta) / 1024);
3216 /* Delete element from from->hashes.
3218 * We don't bother to add the element to to->hashes because not only would
3219 * it require more work, the only purpose it would be to allow that hash to
3220 * be migrated to another slave in this rebalancing run, and there is no
3221 * point in doing that. */
3222 if (hash_idx == 0) {
3225 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3226 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3230 /* Shift load away from 'from' to 'to'. */
3231 from->tx_bytes -= delta;
3232 to->tx_bytes += delta;
3234 /* Arrange for flows to be revalidated. */
3235 ofproto_revalidate(port->bridge->ofproto, hash->tag);
3236 hash->iface = to->iface;
3237 hash->tag = tag_create_random();
3241 bond_rebalance_port(struct port *port)
3243 struct slave_balance *bals;
3245 struct bond_entry *hashes[BOND_MASK + 1];
3246 struct slave_balance *b, *from, *to;
3247 struct bond_entry *e;
3248 struct iface *iface;
3251 assert(port->bond_mode != BM_AB);
3253 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3254 * descending order of tx_bytes, so that bals[0] represents the most
3255 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3258 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3259 * array for each slave_balance structure, we sort our local array of
3260 * hashes in order by slave, so that all of the hashes for a given slave
3261 * become contiguous in memory, and then we point each 'hashes' members of
3262 * a slave_balance structure to the start of a contiguous group. */
3263 n_bals = port->n_ifaces;
3264 b = bals = xmalloc(n_bals * sizeof *bals);
3265 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3272 assert(b == &bals[n_bals]);
3273 for (i = 0; i <= BOND_MASK; i++) {
3274 hashes[i] = &port->bond_hash[i];
3276 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3277 for (i = 0; i <= BOND_MASK; i++) {
3283 for (b = bals; b < &bals[n_bals]; b++) {
3284 if (b->iface == e->iface) {
3285 b->tx_bytes += e->tx_bytes;
3287 b->hashes = &hashes[i];
3294 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3295 log_bals(bals, n_bals, port);
3297 /* Discard slaves that aren't enabled (which were sorted to the back of the
3298 * array earlier). */
3299 while (!bals[n_bals - 1].iface->enabled) {
3306 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3307 to = &bals[n_bals - 1];
3308 for (from = bals; from < to; ) {
3309 uint64_t overload = from->tx_bytes - to->tx_bytes;
3310 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3311 /* The extra load on 'from' (and all less-loaded slaves), compared
3312 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3313 * it is less than ~1Mbps. No point in rebalancing. */
3315 } else if (from->n_hashes == 1) {
3316 /* 'from' only carries a single MAC hash, so we can't shift any
3317 * load away from it, even though we want to. */
3320 /* 'from' is carrying significantly more load than 'to', and that
3321 * load is split across at least two different hashes. Pick a hash
3322 * to migrate to 'to' (the least-loaded slave), given that doing so
3323 * must decrease the ratio of the load on the two slaves by at
3326 * The sort order we use means that we prefer to shift away the
3327 * smallest hashes instead of the biggest ones. There is little
3328 * reason behind this decision; we could use the opposite sort
3329 * order to shift away big hashes ahead of small ones. */
3332 for (i = 0; i < from->n_hashes; i++) {
3333 double old_ratio, new_ratio;
3334 uint64_t delta = from->hashes[i]->tx_bytes;
3336 if (delta == 0 || from->tx_bytes - delta == 0) {
3337 /* Pointless move. */
3341 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3343 if (to->tx_bytes == 0) {
3344 /* Nothing on the new slave, move it. */
3348 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3349 new_ratio = (double)(from->tx_bytes - delta) /
3350 (to->tx_bytes + delta);
3352 if (new_ratio == 0) {
3353 /* Should already be covered but check to prevent division
3358 if (new_ratio < 1) {
3359 new_ratio = 1 / new_ratio;
3362 if (old_ratio - new_ratio > 0.1) {
3363 /* Would decrease the ratio, move it. */
3367 if (i < from->n_hashes) {
3368 bond_shift_load(from, to, i);
3370 /* If the result of the migration changed the relative order of
3371 * 'from' and 'to' swap them back to maintain invariants. */
3372 if (order_swapped) {
3373 swap_bals(from, to);
3376 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3377 * point to different slave_balance structures. It is only
3378 * valid to do these two operations in a row at all because we
3379 * know that 'from' will not move past 'to' and vice versa. */
3380 resort_bals(from, bals, n_bals);
3381 resort_bals(to, bals, n_bals);
3388 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3389 * historical data to decay to <1% in 7 rebalancing runs. */
3390 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3402 bond_send_learning_packets(struct port *port)
3404 struct bridge *br = port->bridge;
3405 struct mac_entry *e;
3406 struct ofpbuf packet;
3407 int error, n_packets, n_errors;
3409 if (!port->n_ifaces || !port->active_iface || bond_is_tcp_hash(port)) {
3413 ofpbuf_init(&packet, 128);
3414 error = n_packets = n_errors = 0;
3415 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3421 if (e->port.p == port) {
3425 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3427 flow_extract(&packet, 0, ODPP_NONE, &flow);
3429 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3435 retval = ofproto_send_packet(br->ofproto, dp_ifidx, e->vlan, &packet);
3441 ofpbuf_uninit(&packet);
3444 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3445 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3446 "packets, last error was: %s",
3447 port->name, n_errors, n_packets, strerror(error));
3449 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3450 port->name, n_packets);
3454 /* Bonding unixctl user interface functions. */
3457 bond_unixctl_list(struct unixctl_conn *conn,
3458 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3460 struct ds ds = DS_EMPTY_INITIALIZER;
3461 const struct bridge *br;
3463 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3465 LIST_FOR_EACH (br, node, &all_bridges) {
3468 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3469 if (port->n_ifaces > 1) {
3470 struct iface *iface;
3472 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3473 bond_mode_to_string(port->bond_mode));
3474 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3475 if (&iface->port_elem != list_front(&port->ifaces)) {
3476 ds_put_cstr(&ds, ", ");
3478 ds_put_cstr(&ds, iface->name);
3480 ds_put_char(&ds, '\n');
3484 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3488 static struct port *
3489 bond_find(const char *name)
3491 const struct bridge *br;
3493 LIST_FOR_EACH (br, node, &all_bridges) {
3496 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3497 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3506 bond_unixctl_show(struct unixctl_conn *conn,
3507 const char *args, void *aux OVS_UNUSED)
3509 struct ds ds = DS_EMPTY_INITIALIZER;
3510 const struct port *port;
3511 struct iface *iface;
3513 port = bond_find(args);
3515 unixctl_command_reply(conn, 501, "no such bond");
3519 ds_put_format(&ds, "bond_mode: %s\n",
3520 bond_mode_to_string(port->bond_mode));
3523 ds_put_format(&ds, "lacp: %s\n",
3524 lacp_is_active(port->lacp) ? "active" : "passive");
3526 ds_put_cstr(&ds, "lacp: off\n");
3529 if (port->bond_mode != BM_AB) {
3530 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3531 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3535 ds_put_format(&ds, "bond-detect-mode: %s\n",
3536 port->monitor ? "carrier" : "miimon");
3538 if (!port->monitor) {
3539 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3540 port->miimon_interval);
3543 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3544 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3546 if (port->bond_mode != BM_AB) {
3547 ds_put_format(&ds, "next rebalance: %lld ms\n",
3548 port->bond_next_rebalance - time_msec());
3551 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3552 struct bond_entry *be;
3556 ds_put_format(&ds, "\nslave %s: %s\n",
3557 iface->name, iface->enabled ? "enabled" : "disabled");
3558 if (iface == port->active_iface) {
3559 ds_put_cstr(&ds, "\tactive slave\n");
3561 if (iface->delay_expires != LLONG_MAX) {
3562 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3563 iface->enabled ? "downdelay" : "updelay",
3564 iface->delay_expires - time_msec());
3567 if (port->bond_mode == BM_AB) {
3572 memset(&flow, 0, sizeof flow);
3573 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3574 int hash = be - port->bond_hash;
3575 struct mac_entry *me;
3577 if (be->iface != iface) {
3581 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3582 hash, be->tx_bytes / 1024);
3584 if (port->bond_mode != BM_SLB) {
3589 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3593 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
3594 if (bond_hash_src(me->mac, me->vlan) == hash
3595 && me->port.p != port
3596 && choose_output_iface(port, &flow, me->vlan,
3598 && dp_ifidx == iface->dp_ifidx)
3600 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3601 ETH_ADDR_ARGS(me->mac));
3606 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3611 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3612 void *aux OVS_UNUSED)
3614 char *args = (char *) args_;
3615 char *save_ptr = NULL;
3616 char *bond_s, *hash_s, *slave_s;
3618 struct iface *iface;
3619 struct bond_entry *entry;
3622 bond_s = strtok_r(args, " ", &save_ptr);
3623 hash_s = strtok_r(NULL, " ", &save_ptr);
3624 slave_s = strtok_r(NULL, " ", &save_ptr);
3626 unixctl_command_reply(conn, 501,
3627 "usage: bond/migrate BOND HASH SLAVE");
3631 port = bond_find(bond_s);
3633 unixctl_command_reply(conn, 501, "no such bond");
3637 if (port->bond_mode != BM_SLB) {
3638 unixctl_command_reply(conn, 501, "not an SLB bond");
3642 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3643 hash = atoi(hash_s) & BOND_MASK;
3645 unixctl_command_reply(conn, 501, "bad hash");
3649 iface = port_lookup_iface(port, slave_s);
3651 unixctl_command_reply(conn, 501, "no such slave");
3655 if (!iface->enabled) {
3656 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3660 entry = &port->bond_hash[hash];
3661 ofproto_revalidate(port->bridge->ofproto, entry->tag);
3662 entry->iface = iface;
3663 entry->tag = tag_create_random();
3664 unixctl_command_reply(conn, 200, "migrated");
3668 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3669 void *aux OVS_UNUSED)
3671 char *args = (char *) args_;
3672 char *save_ptr = NULL;
3673 char *bond_s, *slave_s;
3675 struct iface *iface;
3677 bond_s = strtok_r(args, " ", &save_ptr);
3678 slave_s = strtok_r(NULL, " ", &save_ptr);
3680 unixctl_command_reply(conn, 501,
3681 "usage: bond/set-active-slave BOND SLAVE");
3685 port = bond_find(bond_s);
3687 unixctl_command_reply(conn, 501, "no such bond");
3691 iface = port_lookup_iface(port, slave_s);
3693 unixctl_command_reply(conn, 501, "no such slave");
3697 if (!iface->enabled) {
3698 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3702 if (port->active_iface != iface) {
3703 ofproto_revalidate(port->bridge->ofproto,
3704 port_get_active_iface_tag(port));
3705 port->active_iface = iface;
3706 VLOG_INFO("port %s: active interface is now %s",
3707 port->name, iface->name);
3708 bond_send_learning_packets(port);
3709 unixctl_command_reply(conn, 200, "done");
3711 unixctl_command_reply(conn, 200, "no change");
3716 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3718 char *args = (char *) args_;
3719 char *save_ptr = NULL;
3720 char *bond_s, *slave_s;
3722 struct iface *iface;
3724 bond_s = strtok_r(args, " ", &save_ptr);
3725 slave_s = strtok_r(NULL, " ", &save_ptr);
3727 unixctl_command_reply(conn, 501,
3728 "usage: bond/enable/disable-slave BOND SLAVE");
3732 port = bond_find(bond_s);
3734 unixctl_command_reply(conn, 501, "no such bond");
3738 iface = port_lookup_iface(port, slave_s);
3740 unixctl_command_reply(conn, 501, "no such slave");
3744 bond_enable_slave(iface, enable);
3745 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3749 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3750 void *aux OVS_UNUSED)
3752 enable_slave(conn, args, true);
3756 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3757 void *aux OVS_UNUSED)
3759 enable_slave(conn, args, false);
3763 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3764 void *aux OVS_UNUSED)
3766 char *args = (char *) args_;
3767 uint8_t mac[ETH_ADDR_LEN];
3771 char *mac_s, *vlan_s;
3772 char *save_ptr = NULL;
3774 mac_s = strtok_r(args, " ", &save_ptr);
3775 vlan_s = strtok_r(NULL, " ", &save_ptr);
3778 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3779 unixctl_command_reply(conn, 501, "invalid vlan");
3783 vlan = OFP_VLAN_NONE;
3786 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3787 == ETH_ADDR_SCAN_COUNT) {
3788 hash = bond_hash_src(mac, vlan);
3790 hash_cstr = xasprintf("%u", hash);
3791 unixctl_command_reply(conn, 200, hash_cstr);
3794 unixctl_command_reply(conn, 501, "invalid mac");
3801 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3802 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3803 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3804 unixctl_command_register("bond/set-active-slave",
3805 bond_unixctl_set_active_slave, NULL);
3806 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3808 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3810 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3813 /* Port functions. */
3816 lacp_send_pdu_cb(void *aux, const struct lacp_pdu *pdu)
3818 struct iface *iface = aux;
3819 uint8_t ea[ETH_ADDR_LEN];
3822 error = netdev_get_etheraddr(iface->netdev, ea);
3824 struct ofpbuf packet;
3825 struct lacp_pdu *packet_pdu;
3827 ofpbuf_init(&packet, 0);
3828 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
3829 sizeof *packet_pdu);
3830 memcpy(packet_pdu, pdu, sizeof *packet_pdu);
3831 ofproto_send_packet(iface->port->bridge->ofproto,
3832 iface->dp_ifidx, 0, &packet);
3833 ofpbuf_uninit(&packet);
3835 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3836 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3837 "(%s)", iface->name, strerror(error));
3842 port_run(struct port *port)
3844 if (port->monitor) {
3847 /* Track carrier going up and down on interfaces. */
3848 while (!netdev_monitor_poll(port->monitor, &devname)) {
3849 struct iface *iface;
3851 iface = port_lookup_iface(port, devname);
3853 iface_update_carrier(iface);
3857 } else if (time_msec() >= port->miimon_next_update) {
3858 struct iface *iface;
3860 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3861 iface_update_carrier(iface);
3863 port->miimon_next_update = time_msec() + port->miimon_interval;
3867 struct iface *iface;
3869 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3870 lacp_slave_enable(port->lacp, iface, iface->enabled);
3873 lacp_run(port->lacp, lacp_send_pdu_cb);
3880 port_wait(struct port *port)
3882 if (port->monitor) {
3883 netdev_monitor_poll_wait(port->monitor);
3885 poll_timer_wait_until(port->miimon_next_update);
3889 lacp_wait(port->lacp);
3895 static struct port *
3896 port_create(struct bridge *br, const char *name)
3900 port = xzalloc(sizeof *port);
3903 port->trunks = NULL;
3904 port->name = xstrdup(name);
3905 port->active_iface = NULL;
3906 list_init(&port->ifaces);
3908 hmap_insert(&br->ports, &port->hmap_node, hash_string(port->name, 0));
3910 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3917 get_port_other_config(const struct ovsrec_port *port, const char *key,
3918 const char *default_value)
3922 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3924 return value ? value : default_value;
3928 get_interface_other_config(const struct ovsrec_interface *iface,
3929 const char *key, const char *default_value)
3933 value = get_ovsrec_key_value(&iface->header_,
3934 &ovsrec_interface_col_other_config, key);
3935 return value ? value : default_value;
3939 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3941 struct iface *iface, *next;
3942 struct sset new_ifaces;
3945 /* Collect list of new interfaces. */
3946 sset_init(&new_ifaces);
3947 for (i = 0; i < cfg->n_interfaces; i++) {
3948 const char *name = cfg->interfaces[i]->name;
3949 sset_add(&new_ifaces, name);
3952 /* Get rid of deleted interfaces. */
3953 LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) {
3954 if (!sset_contains(&new_ifaces, iface->name)) {
3955 iface_destroy(iface);
3959 sset_destroy(&new_ifaces);
3962 /* Expires all MAC learning entries associated with 'port' and forces ofproto
3963 * to revalidate every flow. */
3965 port_flush_macs(struct port *port)
3967 struct bridge *br = port->bridge;
3968 struct mac_learning *ml = br->ml;
3969 struct mac_entry *mac, *next_mac;
3972 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
3973 if (mac->port.p == port) {
3974 mac_learning_expire(ml, mac);
3980 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3982 const char *detect_mode;
3983 struct sset new_ifaces;
3984 long long int next_rebalance, miimon_next_update;
3985 bool need_flush = false;
3986 unsigned long *trunks;
3992 /* Update settings. */
3993 port->updelay = cfg->bond_updelay;
3994 if (port->updelay < 0) {
3997 port->downdelay = cfg->bond_downdelay;
3998 if (port->downdelay < 0) {
3999 port->downdelay = 0;
4001 port->bond_rebalance_interval = atoi(
4002 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
4003 if (port->bond_rebalance_interval < 1000) {
4004 port->bond_rebalance_interval = 1000;
4006 next_rebalance = time_msec() + port->bond_rebalance_interval;
4007 if (port->bond_next_rebalance > next_rebalance) {
4008 port->bond_next_rebalance = next_rebalance;
4011 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
4014 netdev_monitor_destroy(port->monitor);
4015 port->monitor = NULL;
4017 if (strcmp(detect_mode, "miimon")) {
4018 port->monitor = netdev_monitor_create();
4020 if (strcmp(detect_mode, "carrier")) {
4021 VLOG_WARN("port %s: unsupported bond-detect-mode %s, "
4022 "defaulting to carrier", port->name, detect_mode);
4026 port->miimon_interval = atoi(
4027 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4028 if (port->miimon_interval < 100) {
4029 port->miimon_interval = 100;
4031 miimon_next_update = time_msec() + port->miimon_interval;
4032 if (port->miimon_next_update > miimon_next_update) {
4033 port->miimon_next_update = miimon_next_update;
4036 if (!port->cfg->bond_mode ||
4037 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4038 port->bond_mode = BM_SLB;
4039 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4040 port->bond_mode = BM_AB;
4041 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
4042 port->bond_mode = BM_TCP;
4044 port->bond_mode = BM_SLB;
4045 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4046 port->name, port->cfg->bond_mode,
4047 bond_mode_to_string(port->bond_mode));
4050 /* Add new interfaces and update 'cfg' member of existing ones. */
4051 sset_init(&new_ifaces);
4052 for (i = 0; i < cfg->n_interfaces; i++) {
4053 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4054 struct iface *iface;
4056 if (!sset_add(&new_ifaces, if_cfg->name)) {
4057 VLOG_WARN("port %s: %s specified twice as port interface",
4058 port->name, if_cfg->name);
4059 iface_set_ofport(if_cfg, -1);
4063 iface = iface_lookup(port->bridge, if_cfg->name);
4065 if (iface->port != port) {
4066 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4068 port->bridge->name, if_cfg->name, iface->port->name);
4071 iface->cfg = if_cfg;
4073 iface = iface_create(port, if_cfg);
4076 /* Determine interface type. The local port always has type
4077 * "internal". Other ports take their type from the database and
4078 * default to "system" if none is specified. */
4079 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4080 : if_cfg->type[0] ? if_cfg->type
4083 sset_destroy(&new_ifaces);
4088 if (port->n_ifaces < 2) {
4090 if (vlan >= 0 && vlan <= 4095) {
4091 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4096 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4097 * they even work as-is. But they have not been tested. */
4098 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4102 if (port->vlan != vlan) {
4107 /* Get trunked VLANs. */
4109 if (vlan < 0 && cfg->n_trunks) {
4112 trunks = bitmap_allocate(4096);
4114 for (i = 0; i < cfg->n_trunks; i++) {
4115 int trunk = cfg->trunks[i];
4117 bitmap_set1(trunks, trunk);
4123 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4124 port->name, cfg->n_trunks);
4126 if (n_errors == cfg->n_trunks) {
4127 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4129 bitmap_free(trunks);
4132 } else if (vlan >= 0 && cfg->n_trunks) {
4133 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4137 ? port->trunks != NULL
4138 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4141 bitmap_free(port->trunks);
4142 port->trunks = trunks;
4145 port_flush_macs(port);
4150 port_destroy(struct port *port)
4153 struct bridge *br = port->bridge;
4154 struct iface *iface, *next;
4157 for (i = 0; i < MAX_MIRRORS; i++) {
4158 struct mirror *m = br->mirrors[i];
4159 if (m && m->out_port == port) {
4164 LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) {
4165 iface_destroy(iface);
4168 hmap_remove(&br->ports, &port->hmap_node);
4170 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4172 port_flush_macs(port);
4174 lacp_destroy(port->lacp);
4175 netdev_monitor_destroy(port->monitor);
4176 bitmap_free(port->trunks);
4177 free(port->bond_hash);
4183 static struct port *
4184 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4186 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4187 return iface ? iface->port : NULL;
4190 static struct port *
4191 port_lookup(const struct bridge *br, const char *name)
4195 HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_string(name, 0),
4197 if (!strcmp(port->name, name)) {
4204 static struct iface *
4205 port_lookup_iface(const struct port *port, const char *name)
4207 struct iface *iface = iface_lookup(port->bridge, name);
4208 return iface && iface->port == port ? iface : NULL;
4212 enable_lacp(struct port *port, bool *activep)
4214 if (!port->cfg->lacp) {
4215 /* XXX when LACP implementation has been sufficiently tested, enable by
4216 * default and make active on bonded ports. */
4218 } else if (!strcmp(port->cfg->lacp, "off")) {
4220 } else if (!strcmp(port->cfg->lacp, "active")) {
4223 } else if (!strcmp(port->cfg->lacp, "passive")) {
4227 VLOG_WARN("port %s: unknown LACP mode %s",
4228 port->name, port->cfg->lacp);
4234 iface_update_lacp(struct iface *iface)
4236 struct lacp_slave_settings s;
4239 s.name = iface->name;
4240 s.id = iface->dp_ifidx;
4241 priority = atoi(get_interface_other_config(
4242 iface->cfg, "lacp-port-priority", "0"));
4243 s.priority = (priority >= 0 && priority <= UINT16_MAX ? priority
4246 lacp_slave_register(iface->port->lacp, iface, &s);
4250 port_update_lacp(struct port *port)
4252 struct lacp_settings s;
4253 struct iface *iface;
4255 if (!enable_lacp(port, &s.active)) {
4256 lacp_destroy(port->lacp);
4262 port->lacp = lacp_create();
4265 s.name = port->name;
4266 memcpy(s.id, port->bridge->ea, ETH_ADDR_LEN);
4267 s.priority = atoi(get_port_other_config(port->cfg, "lacp-system-priority",
4269 s.fast = !strcmp(get_port_other_config(port->cfg, "lacp-time", "slow"),
4272 if (s.priority <= 0 || s.priority > UINT16_MAX) {
4273 /* Prefer bondable links if unspecified. */
4274 s.priority = UINT16_MAX - (port->n_ifaces > 1);
4277 lacp_configure(port->lacp, &s);
4278 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
4279 iface_update_lacp(iface);
4284 port_update_bonding(struct port *port)
4286 if (port->n_ifaces < 2) {
4287 /* Not a bonded port. */
4288 free(port->bond_hash);
4289 port->bond_hash = NULL;
4290 port->bond_fake_iface = false;
4291 port->active_iface = NULL;
4292 port->no_ifaces_tag = 0;
4296 if (port->bond_mode != BM_AB && !port->bond_hash) {
4297 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4298 for (i = 0; i <= BOND_MASK; i++) {
4299 struct bond_entry *e = &port->bond_hash[i];
4303 port->bond_next_rebalance
4304 = time_msec() + port->bond_rebalance_interval;
4305 } else if (port->bond_mode == BM_AB) {
4306 free(port->bond_hash);
4307 port->bond_hash = NULL;
4310 if (!port->no_ifaces_tag) {
4311 port->no_ifaces_tag = tag_create_random();
4314 if (!port->active_iface) {
4315 bond_choose_active_iface(port);
4318 port->bond_fake_iface = port->cfg->bond_fake_iface;
4319 if (port->bond_fake_iface) {
4320 port->bond_next_fake_iface_update = time_msec();
4326 /* Interface functions. */
4328 static struct iface *
4329 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4331 struct bridge *br = port->bridge;
4332 struct iface *iface;
4333 char *name = if_cfg->name;
4335 iface = xzalloc(sizeof *iface);
4337 iface->name = xstrdup(name);
4338 iface->dp_ifidx = -1;
4339 iface->tag = tag_create_random();
4340 iface->delay_expires = LLONG_MAX;
4341 iface->netdev = NULL;
4342 iface->cfg = if_cfg;
4344 shash_add_assert(&br->iface_by_name, iface->name, iface);
4346 list_push_back(&port->ifaces, &iface->port_elem);
4349 if (port->n_ifaces > 1) {
4350 br->has_bonded_ports = true;
4353 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4361 iface_destroy(struct iface *iface)
4364 struct port *port = iface->port;
4365 struct bridge *br = port->bridge;
4366 bool del_active = port->active_iface == iface;
4368 if (port->bond_hash) {
4369 struct bond_entry *e;
4370 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
4371 if (e->iface == iface) {
4377 if (iface->port->lacp) {
4378 lacp_slave_unregister(iface->port->lacp, iface);
4381 if (port->monitor && iface->netdev) {
4382 netdev_monitor_remove(port->monitor, iface->netdev);
4385 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4387 if (iface->dp_ifidx >= 0) {
4388 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4391 list_remove(&iface->port_elem);
4394 netdev_close(iface->netdev);
4397 bond_choose_active_iface(port);
4398 bond_send_learning_packets(port);
4404 bridge_flush(port->bridge);
4408 static struct iface *
4409 iface_lookup(const struct bridge *br, const char *name)
4411 return shash_find_data(&br->iface_by_name, name);
4414 static struct iface *
4415 iface_find(const char *name)
4417 const struct bridge *br;
4419 LIST_FOR_EACH (br, node, &all_bridges) {
4420 struct iface *iface = iface_lookup(br, name);
4429 static struct iface *
4430 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4432 struct iface *iface;
4434 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4435 hash_int(dp_ifidx, 0), &br->ifaces) {
4436 if (iface->dp_ifidx == dp_ifidx) {
4443 /* Set Ethernet address of 'iface', if one is specified in the configuration
4446 iface_set_mac(struct iface *iface)
4448 uint8_t ea[ETH_ADDR_LEN];
4450 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4451 if (eth_addr_is_multicast(ea)) {
4452 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4454 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4455 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4456 iface->name, iface->name);
4458 int error = netdev_set_etheraddr(iface->netdev, ea);
4460 VLOG_ERR("interface %s: setting MAC failed (%s)",
4461 iface->name, strerror(error));
4467 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4469 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4472 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4476 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4478 * The value strings in '*shash' are taken directly from values[], not copied,
4479 * so the caller should not modify or free them. */
4481 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4482 struct shash *shash)
4487 for (i = 0; i < n; i++) {
4488 shash_add(shash, keys[i], values[i]);
4492 /* Creates 'keys' and 'values' arrays from 'shash'.
4494 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4495 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4496 * are populated with with strings taken directly from 'shash' and thus have
4497 * the same ownership of the key-value pairs in shash.
4500 shash_to_ovs_idl_map(struct shash *shash,
4501 char ***keys, char ***values, size_t *n)
4505 struct shash_node *sn;
4507 count = shash_count(shash);
4509 k = xmalloc(count * sizeof *k);
4510 v = xmalloc(count * sizeof *v);
4513 SHASH_FOR_EACH(sn, shash) {
4524 struct iface_delete_queues_cbdata {
4525 struct netdev *netdev;
4526 const struct ovsdb_datum *queues;
4530 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4532 union ovsdb_atom atom;
4534 atom.integer = target;
4535 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4539 iface_delete_queues(unsigned int queue_id,
4540 const struct shash *details OVS_UNUSED, void *cbdata_)
4542 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4544 if (!queue_ids_include(cbdata->queues, queue_id)) {
4545 netdev_delete_queue(cbdata->netdev, queue_id);
4550 iface_update_carrier(struct iface *iface)
4552 bool carrier = iface_get_carrier(iface);
4553 if (carrier == iface->up) {
4557 iface->up = carrier;
4558 if (iface->port->lacp) {
4559 lacp_slave_carrier_changed(iface->port->lacp, iface);
4564 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4566 if (!qos || qos->type[0] == '\0') {
4567 netdev_set_qos(iface->netdev, NULL, NULL);
4569 struct iface_delete_queues_cbdata cbdata;
4570 struct shash details;
4573 /* Configure top-level Qos for 'iface'. */
4574 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4575 qos->n_other_config, &details);
4576 netdev_set_qos(iface->netdev, qos->type, &details);
4577 shash_destroy(&details);
4579 /* Deconfigure queues that were deleted. */
4580 cbdata.netdev = iface->netdev;
4581 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4583 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4585 /* Configure queues for 'iface'. */
4586 for (i = 0; i < qos->n_queues; i++) {
4587 const struct ovsrec_queue *queue = qos->value_queues[i];
4588 unsigned int queue_id = qos->key_queues[i];
4590 shash_from_ovs_idl_map(queue->key_other_config,
4591 queue->value_other_config,
4592 queue->n_other_config, &details);
4593 netdev_set_queue(iface->netdev, queue_id, &details);
4594 shash_destroy(&details);
4600 iface_update_cfm(struct iface *iface)
4604 uint16_t *remote_mps;
4605 struct ovsrec_monitor *mon;
4606 uint8_t maid[CCM_MAID_LEN];
4608 mon = iface->cfg->monitor;
4611 ofproto_iface_clear_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
4615 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4616 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4620 cfm.mpid = mon->mpid;
4621 cfm.interval = mon->interval ? *mon->interval : 1000;
4623 memcpy(cfm.maid, maid, sizeof cfm.maid);
4625 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4626 for(i = 0; i < mon->n_remote_mps; i++) {
4627 remote_mps[i] = mon->remote_mps[i]->mpid;
4630 ofproto_iface_set_cfm(iface->port->bridge->ofproto, iface->dp_ifidx,
4631 &cfm, remote_mps, mon->n_remote_mps);
4635 /* Read carrier or miimon status directly from 'iface''s netdev, according to
4636 * how 'iface''s port is configured.
4638 * Returns true if 'iface' is up, false otherwise. */
4640 iface_get_carrier(const struct iface *iface)
4642 return (iface->port->monitor
4643 ? netdev_get_carrier(iface->netdev)
4644 : netdev_get_miimon(iface->netdev));
4647 /* Port mirroring. */
4649 static struct mirror *
4650 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4654 for (i = 0; i < MAX_MIRRORS; i++) {
4655 struct mirror *m = br->mirrors[i];
4656 if (m && uuid_equals(uuid, &m->uuid)) {
4664 mirror_reconfigure(struct bridge *br)
4666 unsigned long *rspan_vlans;
4670 /* Get rid of deleted mirrors. */
4671 for (i = 0; i < MAX_MIRRORS; i++) {
4672 struct mirror *m = br->mirrors[i];
4674 const struct ovsdb_datum *mc;
4675 union ovsdb_atom atom;
4677 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4678 atom.uuid = br->mirrors[i]->uuid;
4679 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4685 /* Add new mirrors and reconfigure existing ones. */
4686 for (i = 0; i < br->cfg->n_mirrors; i++) {
4687 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4688 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4690 mirror_reconfigure_one(m, cfg);
4692 mirror_create(br, cfg);
4696 /* Update port reserved status. */
4697 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
4698 port->is_mirror_output_port = false;
4700 for (i = 0; i < MAX_MIRRORS; i++) {
4701 struct mirror *m = br->mirrors[i];
4702 if (m && m->out_port) {
4703 m->out_port->is_mirror_output_port = true;
4707 /* Update flooded vlans (for RSPAN). */
4709 if (br->cfg->n_flood_vlans) {
4710 rspan_vlans = bitmap_allocate(4096);
4712 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4713 int64_t vlan = br->cfg->flood_vlans[i];
4714 if (vlan >= 0 && vlan < 4096) {
4715 bitmap_set1(rspan_vlans, vlan);
4716 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4719 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4724 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4726 mac_learning_flush(br->ml);
4731 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4736 for (i = 0; ; i++) {
4737 if (i >= MAX_MIRRORS) {
4738 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4739 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4742 if (!br->mirrors[i]) {
4747 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4749 mac_learning_flush(br->ml);
4751 br->mirrors[i] = m = xzalloc(sizeof *m);
4754 m->name = xstrdup(cfg->name);
4755 sset_init(&m->src_ports);
4756 sset_init(&m->dst_ports);
4762 mirror_reconfigure_one(m, cfg);
4766 mirror_destroy(struct mirror *m)
4769 struct bridge *br = m->bridge;
4772 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
4773 port->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4774 port->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4777 sset_destroy(&m->src_ports);
4778 sset_destroy(&m->dst_ports);
4781 m->bridge->mirrors[m->idx] = NULL;
4786 mac_learning_flush(br->ml);
4791 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4796 for (i = 0; i < n_ports; i++) {
4797 const char *name = ports[i]->name;
4798 if (port_lookup(m->bridge, name)) {
4799 sset_add(names, name);
4801 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4802 "port %s", m->bridge->name, m->name, name);
4808 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4814 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4816 for (i = 0; i < cfg->n_select_vlan; i++) {
4817 int64_t vlan = cfg->select_vlan[i];
4818 if (vlan < 0 || vlan > 4095) {
4819 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4820 m->bridge->name, m->name, vlan);
4822 (*vlans)[n_vlans++] = vlan;
4829 vlan_is_mirrored(const struct mirror *m, int vlan)
4833 for (i = 0; i < m->n_vlans; i++) {
4834 if (m->vlans[i] == vlan) {
4842 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4846 for (i = 0; i < m->n_vlans; i++) {
4847 if (port_trunks_vlan(p, m->vlans[i])) {
4855 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4857 struct sset src_ports, dst_ports;
4858 mirror_mask_t mirror_bit;
4859 struct port *out_port;
4866 if (strcmp(cfg->name, m->name)) {
4868 m->name = xstrdup(cfg->name);
4871 /* Get output port. */
4872 if (cfg->output_port) {
4873 out_port = port_lookup(m->bridge, cfg->output_port->name);
4875 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4876 m->bridge->name, m->name);
4882 if (cfg->output_vlan) {
4883 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4884 "output vlan; ignoring output vlan",
4885 m->bridge->name, m->name);
4887 } else if (cfg->output_vlan) {
4889 out_vlan = *cfg->output_vlan;
4891 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4892 m->bridge->name, m->name);
4897 sset_init(&src_ports);
4898 sset_init(&dst_ports);
4899 if (cfg->select_all) {
4900 HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) {
4901 sset_add(&src_ports, port->name);
4902 sset_add(&dst_ports, port->name);
4907 /* Get ports, and drop duplicates and ports that don't exist. */
4908 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4910 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4913 /* Get all the vlans, and drop duplicate and invalid vlans. */
4914 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4917 /* Update mirror data. */
4918 if (!sset_equals(&m->src_ports, &src_ports)
4919 || !sset_equals(&m->dst_ports, &dst_ports)
4920 || m->n_vlans != n_vlans
4921 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4922 || m->out_port != out_port
4923 || m->out_vlan != out_vlan) {
4924 bridge_flush(m->bridge);
4925 mac_learning_flush(m->bridge->ml);
4927 sset_swap(&m->src_ports, &src_ports);
4928 sset_swap(&m->dst_ports, &dst_ports);
4931 m->n_vlans = n_vlans;
4932 m->out_port = out_port;
4933 m->out_vlan = out_vlan;
4936 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4937 HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) {
4938 if (sset_contains(&m->src_ports, port->name)
4941 ? port_trunks_any_mirrored_vlan(m, port)
4942 : vlan_is_mirrored(m, port->vlan)))) {
4943 port->src_mirrors |= mirror_bit;
4945 port->src_mirrors &= ~mirror_bit;
4948 if (sset_contains(&m->dst_ports, port->name)) {
4949 port->dst_mirrors |= mirror_bit;
4951 port->dst_mirrors &= ~mirror_bit;
4956 sset_destroy(&src_ports);
4957 sset_destroy(&dst_ports);