1 /* Copyright (c) 2008, 2009, 2010 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.
20 #include <arpa/inet.h>
23 #include <sys/socket.h>
25 #include <openflow/openflow.h>
30 #include <sys/socket.h>
31 #include <sys/types.h>
37 #include "dynamic-string.h"
43 #include "mac-learning.h"
46 #include "ofp-print.h"
48 #include "ofproto/netflow.h"
49 #include "ofproto/ofproto.h"
50 #include "ovsdb-data.h"
52 #include "poll-loop.h"
53 #include "proc-net-compat.h"
57 #include "socket-util.h"
58 #include "stream-ssl.h"
60 #include "system-stats.h"
65 #include "vswitchd/vswitch-idl.h"
66 #include "xenserver.h"
69 #include "sflow_api.h"
71 VLOG_DEFINE_THIS_MODULE(bridge)
79 /* These members are always valid. */
80 struct port *port; /* Containing port. */
81 size_t port_ifidx; /* Index within containing port. */
82 char *name; /* Host network device name. */
83 tag_type tag; /* Tag associated with this interface. */
84 long long delay_expires; /* Time after which 'enabled' may change. */
86 /* These members are valid only after bridge_reconfigure() causes them to
88 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
89 int dp_ifidx; /* Index within kernel datapath. */
90 struct netdev *netdev; /* Network device. */
91 bool enabled; /* May be chosen for flows? */
92 const char *type; /* Usually same as cfg->type. */
93 const struct ovsrec_interface *cfg;
96 #define BOND_MASK 0xff
98 int iface_idx; /* Index of assigned iface, or -1 if none. */
99 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
100 tag_type iface_tag; /* Tag associated with iface_idx. */
103 #define MAX_MIRRORS 32
104 typedef uint32_t mirror_mask_t;
105 #define MIRROR_MASK_C(X) UINT32_C(X)
106 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
108 struct bridge *bridge;
111 struct uuid uuid; /* UUID of this "mirror" record in database. */
113 /* Selection criteria. */
114 struct shash src_ports; /* Name is port name; data is always NULL. */
115 struct shash dst_ports; /* Name is port name; data is always NULL. */
120 struct port *out_port;
124 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
126 struct bridge *bridge;
128 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
129 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
130 * NULL if all VLANs are trunked. */
131 const struct ovsrec_port *cfg;
134 /* An ordinary bridge port has 1 interface.
135 * A bridge port for bonding has at least 2 interfaces. */
136 struct iface **ifaces;
137 size_t n_ifaces, allocated_ifaces;
140 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
141 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
142 tag_type active_iface_tag; /* Tag for bcast flows. */
143 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
144 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
145 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
146 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
147 long long int bond_next_fake_iface_update; /* Time of next update. */
148 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
149 long long int bond_next_rebalance; /* Next rebalancing time. */
151 /* Port mirroring info. */
152 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
153 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
154 bool is_mirror_output_port; /* Does port mirroring send frames here? */
157 #define DP_MAX_PORTS 255
159 struct list node; /* Node in global list of bridges. */
160 char *name; /* User-specified arbitrary name. */
161 struct mac_learning *ml; /* MAC learning table. */
162 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
163 const struct ovsrec_bridge *cfg;
165 /* OpenFlow switch processing. */
166 struct ofproto *ofproto; /* OpenFlow switch. */
168 /* Kernel datapath information. */
169 struct dpif *dpif; /* Datapath. */
170 struct hmap ifaces; /* Contains "struct iface"s. */
174 size_t n_ports, allocated_ports;
175 struct shash iface_by_name; /* "struct iface"s indexed by name. */
176 struct shash port_by_name; /* "struct port"s indexed by name. */
179 bool has_bonded_ports;
184 /* Port mirroring. */
185 struct mirror *mirrors[MAX_MIRRORS];
188 /* List of all bridges. */
189 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
191 /* OVSDB IDL used to obtain configuration. */
192 static struct ovsdb_idl *idl;
194 /* Each time this timer expires, the bridge fetches systems and interface
195 * statistics and pushes them into the database. */
196 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
197 static long long int stats_timer = LLONG_MIN;
199 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
200 static void bridge_destroy(struct bridge *);
201 static struct bridge *bridge_lookup(const char *name);
202 static unixctl_cb_func bridge_unixctl_dump_flows;
203 static unixctl_cb_func bridge_unixctl_reconnect;
204 static int bridge_run_one(struct bridge *);
205 static size_t bridge_get_controllers(const struct bridge *br,
206 struct ovsrec_controller ***controllersp);
207 static void bridge_reconfigure_one(struct bridge *);
208 static void bridge_reconfigure_remotes(struct bridge *,
209 const struct sockaddr_in *managers,
211 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
212 static void bridge_fetch_dp_ifaces(struct bridge *);
213 static void bridge_flush(struct bridge *);
214 static void bridge_pick_local_hw_addr(struct bridge *,
215 uint8_t ea[ETH_ADDR_LEN],
216 struct iface **hw_addr_iface);
217 static uint64_t bridge_pick_datapath_id(struct bridge *,
218 const uint8_t bridge_ea[ETH_ADDR_LEN],
219 struct iface *hw_addr_iface);
220 static struct iface *bridge_get_local_iface(struct bridge *);
221 static uint64_t dpid_from_hash(const void *, size_t nbytes);
223 static unixctl_cb_func bridge_unixctl_fdb_show;
225 static void bond_init(void);
226 static void bond_run(struct bridge *);
227 static void bond_wait(struct bridge *);
228 static void bond_rebalance_port(struct port *);
229 static void bond_send_learning_packets(struct port *);
230 static void bond_enable_slave(struct iface *iface, bool enable);
232 static struct port *port_create(struct bridge *, const char *name);
233 static void port_reconfigure(struct port *, const struct ovsrec_port *);
234 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
235 static void port_destroy(struct port *);
236 static struct port *port_lookup(const struct bridge *, const char *name);
237 static struct iface *port_lookup_iface(const struct port *, const char *name);
238 static struct port *port_from_dp_ifidx(const struct bridge *,
240 static void port_update_bond_compat(struct port *);
241 static void port_update_vlan_compat(struct port *);
242 static void port_update_bonding(struct port *);
244 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
245 static void mirror_destroy(struct mirror *);
246 static void mirror_reconfigure(struct bridge *);
247 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
248 static bool vlan_is_mirrored(const struct mirror *, int vlan);
250 static struct iface *iface_create(struct port *port,
251 const struct ovsrec_interface *if_cfg);
252 static void iface_destroy(struct iface *);
253 static struct iface *iface_lookup(const struct bridge *, const char *name);
254 static struct iface *iface_from_dp_ifidx(const struct bridge *,
256 static void iface_set_mac(struct iface *);
257 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
259 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
262 /* Hooks into ofproto processing. */
263 static struct ofhooks bridge_ofhooks;
265 /* Public functions. */
267 /* Initializes the bridge module, configuring it to obtain its configuration
268 * from an OVSDB server accessed over 'remote', which should be a string in a
269 * form acceptable to ovsdb_idl_create(). */
271 bridge_init(const char *remote)
273 /* Create connection to database. */
274 idl = ovsdb_idl_create(remote, &ovsrec_idl_class);
276 ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_cur_cfg);
277 ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_statistics);
278 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
280 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
282 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
283 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
285 ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_ofport);
286 ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_statistics);
287 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
289 /* Register unixctl commands. */
290 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
291 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
293 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
298 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
299 * but for which the ovs-vswitchd configuration 'cfg' is required. */
301 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
303 static bool already_configured_once;
304 struct svec bridge_names;
305 struct svec dpif_names, dpif_types;
308 /* Only do this once per ovs-vswitchd run. */
309 if (already_configured_once) {
312 already_configured_once = true;
314 stats_timer = time_msec() + STATS_INTERVAL;
316 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
317 svec_init(&bridge_names);
318 for (i = 0; i < cfg->n_bridges; i++) {
319 svec_add(&bridge_names, cfg->bridges[i]->name);
321 svec_sort(&bridge_names);
323 /* Iterate over all system dpifs and delete any of them that do not appear
325 svec_init(&dpif_names);
326 svec_init(&dpif_types);
327 dp_enumerate_types(&dpif_types);
328 for (i = 0; i < dpif_types.n; i++) {
333 dp_enumerate_names(dpif_types.names[i], &dpif_names);
335 /* For each dpif... */
336 for (j = 0; j < dpif_names.n; j++) {
337 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
339 struct svec all_names;
342 /* ...check whether any of its names is in 'bridge_names'. */
343 svec_init(&all_names);
344 dpif_get_all_names(dpif, &all_names);
345 for (k = 0; k < all_names.n; k++) {
346 if (svec_contains(&bridge_names, all_names.names[k])) {
351 /* No. Delete the dpif. */
355 svec_destroy(&all_names);
360 svec_destroy(&bridge_names);
361 svec_destroy(&dpif_names);
362 svec_destroy(&dpif_types);
365 /* Initializes 'options' and fills it with the options for 'if_cfg'. Merges
366 * keys from "options" and "other_config", preferring "options" keys over
367 * "other_config" keys.
369 * The value strings in '*options' are taken directly from if_cfg, not copied,
370 * so the caller should not modify or free them. */
372 iface_get_options(const struct ovsrec_interface *if_cfg, struct shash *options)
376 shash_from_ovs_idl_map(if_cfg->key_options, if_cfg->value_options,
377 if_cfg->n_options, options);
379 for (i = 0; i < if_cfg->n_other_config; i++) {
380 char *key = if_cfg->key_other_config[i];
381 char *value = if_cfg->value_other_config[i];
383 if (!shash_find_data(options, key)) {
384 shash_add(options, key, value);
386 VLOG_WARN("%s: ignoring \"other_config\" key %s that conflicts "
387 "with \"options\" key %s", if_cfg->name, key, key);
392 /* Returns the type of network device that 'iface' should have. (This is
393 * ordinarily the same type as the interface, but the network devices for
394 * "internal" ports have type "system".) */
396 iface_get_netdev_type(const struct iface *iface)
398 return !strcmp(iface->type, "internal") ? "system" : iface->type;
401 /* Attempt to create the network device for 'iface' through the netdev
404 create_iface_netdev(struct iface *iface)
406 struct netdev_options netdev_options;
407 struct shash options;
410 memset(&netdev_options, 0, sizeof netdev_options);
411 netdev_options.name = iface->cfg->name;
412 netdev_options.type = iface_get_netdev_type(iface);
413 netdev_options.args = &options;
414 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
416 iface_get_options(iface->cfg, &options);
418 error = netdev_open(&netdev_options, &iface->netdev);
421 netdev_get_carrier(iface->netdev, &iface->enabled);
424 shash_destroy(&options);
430 reconfigure_iface_netdev(struct iface *iface)
432 const char *netdev_type, *iface_type;
433 struct shash options;
436 /* Skip reconfiguration if the device has the wrong type. This shouldn't
438 iface_type = iface_get_netdev_type(iface);
439 netdev_type = netdev_get_type(iface->netdev);
440 if (iface_type && strcmp(netdev_type, iface_type)) {
441 VLOG_WARN("%s: attempting change device type from %s to %s",
442 iface->cfg->name, netdev_type, iface_type);
446 /* Reconfigure device. */
447 iface_get_options(iface->cfg, &options);
448 error = netdev_reconfigure(iface->netdev, &options);
449 shash_destroy(&options);
454 /* Callback for iterate_and_prune_ifaces(). */
456 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
458 if (!iface->netdev) {
459 /* We already reported a related error, don't bother duplicating it. */
463 if (iface->dp_ifidx < 0) {
464 VLOG_ERR("%s interface not in %s, dropping",
465 iface->name, dpif_name(br->dpif));
469 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
470 iface->name, iface->dp_ifidx);
474 /* Callback for iterate_and_prune_ifaces(). */
476 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
477 void *aux OVS_UNUSED)
479 /* Set policing attributes. */
480 netdev_set_policing(iface->netdev,
481 iface->cfg->ingress_policing_rate,
482 iface->cfg->ingress_policing_burst);
484 /* Set MAC address of internal interfaces other than the local
486 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
487 iface_set_mac(iface);
493 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
494 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
495 * deletes from 'br' any ports that no longer have any interfaces. */
497 iterate_and_prune_ifaces(struct bridge *br,
498 bool (*cb)(struct bridge *, struct iface *,
504 for (i = 0; i < br->n_ports; ) {
505 struct port *port = br->ports[i];
506 for (j = 0; j < port->n_ifaces; ) {
507 struct iface *iface = port->ifaces[j];
508 if (cb(br, iface, aux)) {
511 iface_destroy(iface);
515 if (port->n_ifaces) {
518 VLOG_ERR("%s port has no interfaces, dropping", port->name);
524 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
525 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
526 * responsible for freeing '*managersp' (with free()).
528 * You may be asking yourself "why does ovs-vswitchd care?", because
529 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
530 * should not be and in fact is not directly involved in that. But
531 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
532 * it has to tell in-band control where the managers are to enable that.
535 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
536 struct sockaddr_in **managersp, size_t *n_managersp)
538 struct sockaddr_in *managers = NULL;
539 size_t n_managers = 0;
541 if (ovs_cfg->n_managers > 0) {
544 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
545 for (i = 0; i < ovs_cfg->n_managers; i++) {
546 const char *name = ovs_cfg->managers[i];
547 struct sockaddr_in *sin = &managers[i];
549 if ((!strncmp(name, "tcp:", 4)
550 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
551 (!strncmp(name, "ssl:", 4)
552 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
558 *managersp = managers;
559 *n_managersp = n_managers;
563 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
565 struct shash old_br, new_br;
566 struct shash_node *node;
567 struct bridge *br, *next;
568 struct sockaddr_in *managers;
571 int sflow_bridge_number;
573 COVERAGE_INC(bridge_reconfigure);
575 collect_managers(ovs_cfg, &managers, &n_managers);
577 /* Collect old and new bridges. */
580 LIST_FOR_EACH (br, node, &all_bridges) {
581 shash_add(&old_br, br->name, br);
583 for (i = 0; i < ovs_cfg->n_bridges; i++) {
584 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
585 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
586 VLOG_WARN("more than one bridge named %s", br_cfg->name);
590 /* Get rid of deleted bridges and add new bridges. */
591 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
592 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
599 SHASH_FOR_EACH (node, &new_br) {
600 const char *br_name = node->name;
601 const struct ovsrec_bridge *br_cfg = node->data;
602 br = shash_find_data(&old_br, br_name);
604 /* If the bridge datapath type has changed, we need to tear it
605 * down and recreate. */
606 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
608 bridge_create(br_cfg);
611 bridge_create(br_cfg);
614 shash_destroy(&old_br);
615 shash_destroy(&new_br);
617 /* Reconfigure all bridges. */
618 LIST_FOR_EACH (br, node, &all_bridges) {
619 bridge_reconfigure_one(br);
622 /* Add and delete ports on all datapaths.
624 * The kernel will reject any attempt to add a given port to a datapath if
625 * that port already belongs to a different datapath, so we must do all
626 * port deletions before any port additions. */
627 LIST_FOR_EACH (br, node, &all_bridges) {
628 struct odp_port *dpif_ports;
630 struct shash want_ifaces;
632 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
633 bridge_get_all_ifaces(br, &want_ifaces);
634 for (i = 0; i < n_dpif_ports; i++) {
635 const struct odp_port *p = &dpif_ports[i];
636 if (!shash_find(&want_ifaces, p->devname)
637 && strcmp(p->devname, br->name)) {
638 int retval = dpif_port_del(br->dpif, p->port);
640 VLOG_ERR("failed to remove %s interface from %s: %s",
641 p->devname, dpif_name(br->dpif),
646 shash_destroy(&want_ifaces);
649 LIST_FOR_EACH (br, node, &all_bridges) {
650 struct odp_port *dpif_ports;
652 struct shash cur_ifaces, want_ifaces;
654 /* Get the set of interfaces currently in this datapath. */
655 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
656 shash_init(&cur_ifaces);
657 for (i = 0; i < n_dpif_ports; i++) {
658 const char *name = dpif_ports[i].devname;
659 shash_add_once(&cur_ifaces, name, &dpif_ports[i]);
662 /* Get the set of interfaces we want on this datapath. */
663 bridge_get_all_ifaces(br, &want_ifaces);
665 hmap_clear(&br->ifaces);
666 SHASH_FOR_EACH (node, &want_ifaces) {
667 const char *if_name = node->name;
668 struct iface *iface = node->data;
669 bool internal = !iface || !strcmp(iface->type, "internal");
670 struct odp_port *dpif_port = shash_find_data(&cur_ifaces, if_name);
673 /* If we have a port or a netdev already, and it's not the type we
674 * want, then delete the port (if any) and close the netdev (if
677 ? dpif_port && !(dpif_port->flags & ODP_PORT_INTERNAL)
679 && strcmp(iface->type, netdev_get_type(iface->netdev))))
682 error = ofproto_port_del(br->ofproto, dpif_port->port);
689 netdev_close(iface->netdev);
690 iface->netdev = NULL;
694 /* If it's not an internal port, open (possibly create) the
697 if (!iface->netdev) {
698 error = create_iface_netdev(iface);
700 VLOG_WARN("could not create iface %s: %s", iface->name,
705 reconfigure_iface_netdev(iface);
709 /* If it's not part of the datapath, add it. */
711 error = dpif_port_add(br->dpif, if_name,
712 internal ? ODP_PORT_INTERNAL : 0, NULL);
713 if (error == EFBIG) {
714 VLOG_ERR("ran out of valid port numbers on %s",
715 dpif_name(br->dpif));
718 VLOG_ERR("failed to add %s interface to %s: %s",
719 if_name, dpif_name(br->dpif), strerror(error));
724 /* If it's an internal port, open the netdev. */
726 if (iface && !iface->netdev) {
727 error = create_iface_netdev(iface);
729 VLOG_WARN("could not create iface %s: %s", iface->name,
735 assert(iface->netdev != NULL);
739 shash_destroy(&cur_ifaces);
740 shash_destroy(&want_ifaces);
742 sflow_bridge_number = 0;
743 LIST_FOR_EACH (br, node, &all_bridges) {
746 struct iface *local_iface;
747 struct iface *hw_addr_iface;
750 bridge_fetch_dp_ifaces(br);
752 iterate_and_prune_ifaces(br, check_iface, NULL);
754 /* Pick local port hardware address, datapath ID. */
755 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
756 local_iface = bridge_get_local_iface(br);
758 int error = netdev_set_etheraddr(local_iface->netdev, ea);
760 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
761 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
762 "Ethernet address: %s",
763 br->name, strerror(error));
767 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
768 ofproto_set_datapath_id(br->ofproto, dpid);
770 dpid_string = xasprintf("%016"PRIx64, dpid);
771 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
774 /* Set NetFlow configuration on this bridge. */
775 if (br->cfg->netflow) {
776 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
777 struct netflow_options opts;
779 memset(&opts, 0, sizeof opts);
781 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
782 if (nf_cfg->engine_type) {
783 opts.engine_type = *nf_cfg->engine_type;
785 if (nf_cfg->engine_id) {
786 opts.engine_id = *nf_cfg->engine_id;
789 opts.active_timeout = nf_cfg->active_timeout;
790 if (!opts.active_timeout) {
791 opts.active_timeout = -1;
792 } else if (opts.active_timeout < 0) {
793 VLOG_WARN("bridge %s: active timeout interval set to negative "
794 "value, using default instead (%d seconds)", br->name,
795 NF_ACTIVE_TIMEOUT_DEFAULT);
796 opts.active_timeout = -1;
799 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
800 if (opts.add_id_to_iface) {
801 if (opts.engine_id > 0x7f) {
802 VLOG_WARN("bridge %s: netflow port mangling may conflict "
803 "with another vswitch, choose an engine id less "
804 "than 128", br->name);
806 if (br->n_ports > 508) {
807 VLOG_WARN("bridge %s: netflow port mangling will conflict "
808 "with another port when more than 508 ports are "
813 opts.collectors.n = nf_cfg->n_targets;
814 opts.collectors.names = nf_cfg->targets;
815 if (ofproto_set_netflow(br->ofproto, &opts)) {
816 VLOG_ERR("bridge %s: problem setting netflow collectors",
820 ofproto_set_netflow(br->ofproto, NULL);
823 /* Set sFlow configuration on this bridge. */
824 if (br->cfg->sflow) {
825 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
826 struct ovsrec_controller **controllers;
827 struct ofproto_sflow_options oso;
828 size_t n_controllers;
830 memset(&oso, 0, sizeof oso);
832 oso.targets.n = sflow_cfg->n_targets;
833 oso.targets.names = sflow_cfg->targets;
835 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
836 if (sflow_cfg->sampling) {
837 oso.sampling_rate = *sflow_cfg->sampling;
840 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
841 if (sflow_cfg->polling) {
842 oso.polling_interval = *sflow_cfg->polling;
845 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
846 if (sflow_cfg->header) {
847 oso.header_len = *sflow_cfg->header;
850 oso.sub_id = sflow_bridge_number++;
851 oso.agent_device = sflow_cfg->agent;
853 oso.control_ip = NULL;
854 n_controllers = bridge_get_controllers(br, &controllers);
855 for (i = 0; i < n_controllers; i++) {
856 if (controllers[i]->local_ip) {
857 oso.control_ip = controllers[i]->local_ip;
861 ofproto_set_sflow(br->ofproto, &oso);
863 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
865 ofproto_set_sflow(br->ofproto, NULL);
868 /* Update the controller and related settings. It would be more
869 * straightforward to call this from bridge_reconfigure_one(), but we
870 * can't do it there for two reasons. First, and most importantly, at
871 * that point we don't know the dp_ifidx of any interfaces that have
872 * been added to the bridge (because we haven't actually added them to
873 * the datapath). Second, at that point we haven't set the datapath ID
874 * yet; when a controller is configured, resetting the datapath ID will
875 * immediately disconnect from the controller, so it's better to set
876 * the datapath ID before the controller. */
877 bridge_reconfigure_remotes(br, managers, n_managers);
879 LIST_FOR_EACH (br, node, &all_bridges) {
880 for (i = 0; i < br->n_ports; i++) {
881 struct port *port = br->ports[i];
884 port_update_vlan_compat(port);
885 port_update_bonding(port);
887 for (j = 0; j < port->n_ifaces; j++) {
888 iface_update_qos(port->ifaces[j], port->cfg->qos);
892 LIST_FOR_EACH (br, node, &all_bridges) {
893 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
900 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
901 const struct ovsdb_idl_column *column,
904 const struct ovsdb_datum *datum;
905 union ovsdb_atom atom;
908 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
909 atom.string = (char *) key;
910 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
911 return idx == UINT_MAX ? NULL : datum->values[idx].string;
915 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
917 return get_ovsrec_key_value(&br_cfg->header_,
918 &ovsrec_bridge_col_other_config, key);
922 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
923 struct iface **hw_addr_iface)
929 *hw_addr_iface = NULL;
931 /* Did the user request a particular MAC? */
932 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
933 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
934 if (eth_addr_is_multicast(ea)) {
935 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
936 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
937 } else if (eth_addr_is_zero(ea)) {
938 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
944 /* Otherwise choose the minimum non-local MAC address among all of the
946 memset(ea, 0xff, sizeof ea);
947 for (i = 0; i < br->n_ports; i++) {
948 struct port *port = br->ports[i];
949 uint8_t iface_ea[ETH_ADDR_LEN];
952 /* Mirror output ports don't participate. */
953 if (port->is_mirror_output_port) {
957 /* Choose the MAC address to represent the port. */
958 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
959 /* Find the interface with this Ethernet address (if any) so that
960 * we can provide the correct devname to the caller. */
962 for (j = 0; j < port->n_ifaces; j++) {
963 struct iface *candidate = port->ifaces[j];
964 uint8_t candidate_ea[ETH_ADDR_LEN];
965 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
966 && eth_addr_equals(iface_ea, candidate_ea)) {
971 /* Choose the interface whose MAC address will represent the port.
972 * The Linux kernel bonding code always chooses the MAC address of
973 * the first slave added to a bond, and the Fedora networking
974 * scripts always add slaves to a bond in alphabetical order, so
975 * for compatibility we choose the interface with the name that is
976 * first in alphabetical order. */
977 iface = port->ifaces[0];
978 for (j = 1; j < port->n_ifaces; j++) {
979 struct iface *candidate = port->ifaces[j];
980 if (strcmp(candidate->name, iface->name) < 0) {
985 /* The local port doesn't count (since we're trying to choose its
986 * MAC address anyway). */
987 if (iface->dp_ifidx == ODPP_LOCAL) {
992 error = netdev_get_etheraddr(iface->netdev, iface_ea);
994 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
995 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
996 iface->name, strerror(error));
1001 /* Compare against our current choice. */
1002 if (!eth_addr_is_multicast(iface_ea) &&
1003 !eth_addr_is_local(iface_ea) &&
1004 !eth_addr_is_reserved(iface_ea) &&
1005 !eth_addr_is_zero(iface_ea) &&
1006 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
1008 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1009 *hw_addr_iface = iface;
1012 if (eth_addr_is_multicast(ea)) {
1013 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1014 *hw_addr_iface = NULL;
1015 VLOG_WARN("bridge %s: using default bridge Ethernet "
1016 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1018 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1019 br->name, ETH_ADDR_ARGS(ea));
1023 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1024 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1025 * an interface on 'br', then that interface must be passed in as
1026 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1027 * 'hw_addr_iface' must be passed in as a null pointer. */
1029 bridge_pick_datapath_id(struct bridge *br,
1030 const uint8_t bridge_ea[ETH_ADDR_LEN],
1031 struct iface *hw_addr_iface)
1034 * The procedure for choosing a bridge MAC address will, in the most
1035 * ordinary case, also choose a unique MAC that we can use as a datapath
1036 * ID. In some special cases, though, multiple bridges will end up with
1037 * the same MAC address. This is OK for the bridges, but it will confuse
1038 * the OpenFlow controller, because each datapath needs a unique datapath
1041 * Datapath IDs must be unique. It is also very desirable that they be
1042 * stable from one run to the next, so that policy set on a datapath
1045 const char *datapath_id;
1048 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1049 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1053 if (hw_addr_iface) {
1055 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1057 * A bridge whose MAC address is taken from a VLAN network device
1058 * (that is, a network device created with vconfig(8) or similar
1059 * tool) will have the same MAC address as a bridge on the VLAN
1060 * device's physical network device.
1062 * Handle this case by hashing the physical network device MAC
1063 * along with the VLAN identifier.
1065 uint8_t buf[ETH_ADDR_LEN + 2];
1066 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1067 buf[ETH_ADDR_LEN] = vlan >> 8;
1068 buf[ETH_ADDR_LEN + 1] = vlan;
1069 return dpid_from_hash(buf, sizeof buf);
1072 * Assume that this bridge's MAC address is unique, since it
1073 * doesn't fit any of the cases we handle specially.
1078 * A purely internal bridge, that is, one that has no non-virtual
1079 * network devices on it at all, is more difficult because it has no
1080 * natural unique identifier at all.
1082 * When the host is a XenServer, we handle this case by hashing the
1083 * host's UUID with the name of the bridge. Names of bridges are
1084 * persistent across XenServer reboots, although they can be reused if
1085 * an internal network is destroyed and then a new one is later
1086 * created, so this is fairly effective.
1088 * When the host is not a XenServer, we punt by using a random MAC
1089 * address on each run.
1091 const char *host_uuid = xenserver_get_host_uuid();
1093 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1094 dpid = dpid_from_hash(combined, strlen(combined));
1100 return eth_addr_to_uint64(bridge_ea);
1104 dpid_from_hash(const void *data, size_t n)
1106 uint8_t hash[SHA1_DIGEST_SIZE];
1108 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1109 sha1_bytes(data, n, hash);
1110 eth_addr_mark_random(hash);
1111 return eth_addr_to_uint64(hash);
1115 iface_refresh_stats(struct iface *iface)
1121 static const struct iface_stat iface_stats[] = {
1122 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1123 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1124 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1125 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1126 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1127 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1128 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1129 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1130 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1131 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1132 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1133 { "collisions", offsetof(struct netdev_stats, collisions) },
1135 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1136 const struct iface_stat *s;
1138 char *keys[N_STATS];
1139 int64_t values[N_STATS];
1142 struct netdev_stats stats;
1144 /* Intentionally ignore return value, since errors will set 'stats' to
1145 * all-1s, and we will deal with that correctly below. */
1146 netdev_get_stats(iface->netdev, &stats);
1149 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1150 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1151 if (value != UINT64_MAX) {
1158 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1162 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1164 struct ovsdb_datum datum;
1168 get_system_stats(&stats);
1170 ovsdb_datum_from_shash(&datum, &stats);
1171 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1178 const struct ovsrec_open_vswitch *cfg;
1180 bool datapath_destroyed;
1181 bool database_changed;
1184 /* Let each bridge do the work that it needs to do. */
1185 datapath_destroyed = false;
1186 LIST_FOR_EACH (br, node, &all_bridges) {
1187 int error = bridge_run_one(br);
1189 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1190 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1191 "forcing reconfiguration", br->name);
1192 datapath_destroyed = true;
1196 /* (Re)configure if necessary. */
1197 database_changed = ovsdb_idl_run(idl);
1198 cfg = ovsrec_open_vswitch_first(idl);
1199 if (database_changed || datapath_destroyed) {
1201 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1203 bridge_configure_once(cfg);
1204 bridge_reconfigure(cfg);
1206 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1207 ovsdb_idl_txn_commit(txn);
1208 ovsdb_idl_txn_destroy(txn); /* XXX */
1210 /* We still need to reconfigure to avoid dangling pointers to
1211 * now-destroyed ovsrec structures inside bridge data. */
1212 static const struct ovsrec_open_vswitch null_cfg;
1214 bridge_reconfigure(&null_cfg);
1219 /* Re-configure SSL. We do this on every trip through the main loop,
1220 * instead of just when the database changes, because the contents of the
1221 * key and certificate files can change without the database changing. */
1222 if (cfg && cfg->ssl) {
1223 const struct ovsrec_ssl *ssl = cfg->ssl;
1225 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1226 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1230 /* Refresh system and interface stats if necessary. */
1231 if (time_msec() >= stats_timer) {
1233 struct ovsdb_idl_txn *txn;
1235 txn = ovsdb_idl_txn_create(idl);
1236 LIST_FOR_EACH (br, node, &all_bridges) {
1239 for (i = 0; i < br->n_ports; i++) {
1240 struct port *port = br->ports[i];
1243 for (j = 0; j < port->n_ifaces; j++) {
1244 struct iface *iface = port->ifaces[j];
1245 iface_refresh_stats(iface);
1249 refresh_system_stats(cfg);
1250 ovsdb_idl_txn_commit(txn);
1251 ovsdb_idl_txn_destroy(txn); /* XXX */
1254 stats_timer = time_msec() + STATS_INTERVAL;
1263 LIST_FOR_EACH (br, node, &all_bridges) {
1264 ofproto_wait(br->ofproto);
1265 if (ofproto_has_primary_controller(br->ofproto)) {
1269 mac_learning_wait(br->ml);
1272 ovsdb_idl_wait(idl);
1273 poll_timer_wait_until(stats_timer);
1276 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1277 * configuration changes. */
1279 bridge_flush(struct bridge *br)
1281 COVERAGE_INC(bridge_flush);
1283 mac_learning_flush(br->ml);
1286 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1287 * such interface. */
1288 static struct iface *
1289 bridge_get_local_iface(struct bridge *br)
1293 for (i = 0; i < br->n_ports; i++) {
1294 struct port *port = br->ports[i];
1295 for (j = 0; j < port->n_ifaces; j++) {
1296 struct iface *iface = port->ifaces[j];
1297 if (iface->dp_ifidx == ODPP_LOCAL) {
1306 /* Bridge unixctl user interface functions. */
1308 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1309 const char *args, void *aux OVS_UNUSED)
1311 struct ds ds = DS_EMPTY_INITIALIZER;
1312 const struct bridge *br;
1313 const struct mac_entry *e;
1315 br = bridge_lookup(args);
1317 unixctl_command_reply(conn, 501, "no such bridge");
1321 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1322 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1323 if (e->port < 0 || e->port >= br->n_ports) {
1326 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1327 br->ports[e->port]->ifaces[0]->dp_ifidx,
1328 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1330 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1334 /* Bridge reconfiguration functions. */
1335 static struct bridge *
1336 bridge_create(const struct ovsrec_bridge *br_cfg)
1341 assert(!bridge_lookup(br_cfg->name));
1342 br = xzalloc(sizeof *br);
1344 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1350 dpif_flow_flush(br->dpif);
1352 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1355 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1357 dpif_delete(br->dpif);
1358 dpif_close(br->dpif);
1363 br->name = xstrdup(br_cfg->name);
1365 br->ml = mac_learning_create();
1366 eth_addr_nicira_random(br->default_ea);
1368 hmap_init(&br->ifaces);
1370 shash_init(&br->port_by_name);
1371 shash_init(&br->iface_by_name);
1375 list_push_back(&all_bridges, &br->node);
1377 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1383 bridge_destroy(struct bridge *br)
1388 while (br->n_ports > 0) {
1389 port_destroy(br->ports[br->n_ports - 1]);
1391 list_remove(&br->node);
1392 error = dpif_delete(br->dpif);
1393 if (error && error != ENOENT) {
1394 VLOG_ERR("failed to delete %s: %s",
1395 dpif_name(br->dpif), strerror(error));
1397 dpif_close(br->dpif);
1398 ofproto_destroy(br->ofproto);
1399 mac_learning_destroy(br->ml);
1400 hmap_destroy(&br->ifaces);
1401 shash_destroy(&br->port_by_name);
1402 shash_destroy(&br->iface_by_name);
1409 static struct bridge *
1410 bridge_lookup(const char *name)
1414 LIST_FOR_EACH (br, node, &all_bridges) {
1415 if (!strcmp(br->name, name)) {
1422 /* Handle requests for a listing of all flows known by the OpenFlow
1423 * stack, including those normally hidden. */
1425 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1426 const char *args, void *aux OVS_UNUSED)
1431 br = bridge_lookup(args);
1433 unixctl_command_reply(conn, 501, "Unknown bridge");
1438 ofproto_get_all_flows(br->ofproto, &results);
1440 unixctl_command_reply(conn, 200, ds_cstr(&results));
1441 ds_destroy(&results);
1444 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1445 * connections and reconnect. If BRIDGE is not specified, then all bridges
1446 * drop their controller connections and reconnect. */
1448 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1449 const char *args, void *aux OVS_UNUSED)
1452 if (args[0] != '\0') {
1453 br = bridge_lookup(args);
1455 unixctl_command_reply(conn, 501, "Unknown bridge");
1458 ofproto_reconnect_controllers(br->ofproto);
1460 LIST_FOR_EACH (br, node, &all_bridges) {
1461 ofproto_reconnect_controllers(br->ofproto);
1464 unixctl_command_reply(conn, 200, NULL);
1468 bridge_run_one(struct bridge *br)
1472 error = ofproto_run1(br->ofproto);
1477 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1480 error = ofproto_run2(br->ofproto, br->flush);
1487 bridge_get_controllers(const struct bridge *br,
1488 struct ovsrec_controller ***controllersp)
1490 struct ovsrec_controller **controllers;
1491 size_t n_controllers;
1493 controllers = br->cfg->controller;
1494 n_controllers = br->cfg->n_controller;
1496 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1502 *controllersp = controllers;
1504 return n_controllers;
1508 bridge_reconfigure_one(struct bridge *br)
1510 struct shash old_ports, new_ports;
1511 struct svec snoops, old_snoops;
1512 struct shash_node *node;
1513 enum ofproto_fail_mode fail_mode;
1516 /* Collect old ports. */
1517 shash_init(&old_ports);
1518 for (i = 0; i < br->n_ports; i++) {
1519 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1522 /* Collect new ports. */
1523 shash_init(&new_ports);
1524 for (i = 0; i < br->cfg->n_ports; i++) {
1525 const char *name = br->cfg->ports[i]->name;
1526 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1527 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1532 /* If we have a controller, then we need a local port. Complain if the
1533 * user didn't specify one.
1535 * XXX perhaps we should synthesize a port ourselves in this case. */
1536 if (bridge_get_controllers(br, NULL)) {
1537 char local_name[IF_NAMESIZE];
1540 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1541 local_name, sizeof local_name);
1542 if (!error && !shash_find(&new_ports, local_name)) {
1543 VLOG_WARN("bridge %s: controller specified but no local port "
1544 "(port named %s) defined",
1545 br->name, local_name);
1549 /* Get rid of deleted ports.
1550 * Get rid of deleted interfaces on ports that still exist. */
1551 SHASH_FOR_EACH (node, &old_ports) {
1552 struct port *port = node->data;
1553 const struct ovsrec_port *port_cfg;
1555 port_cfg = shash_find_data(&new_ports, node->name);
1559 port_del_ifaces(port, port_cfg);
1563 /* Create new ports.
1564 * Add new interfaces to existing ports.
1565 * Reconfigure existing ports. */
1566 SHASH_FOR_EACH (node, &new_ports) {
1567 struct port *port = shash_find_data(&old_ports, node->name);
1569 port = port_create(br, node->name);
1572 port_reconfigure(port, node->data);
1573 if (!port->n_ifaces) {
1574 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1575 br->name, port->name);
1579 shash_destroy(&old_ports);
1580 shash_destroy(&new_ports);
1582 /* Set the fail-mode */
1583 fail_mode = !br->cfg->fail_mode
1584 || !strcmp(br->cfg->fail_mode, "standalone")
1585 ? OFPROTO_FAIL_STANDALONE
1586 : OFPROTO_FAIL_SECURE;
1587 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1588 && !ofproto_has_primary_controller(br->ofproto)) {
1589 ofproto_flush_flows(br->ofproto);
1591 ofproto_set_fail_mode(br->ofproto, fail_mode);
1593 /* Delete all flows if we're switching from connected to standalone or vice
1594 * versa. (XXX Should we delete all flows if we are switching from one
1595 * controller to another?) */
1597 /* Configure OpenFlow controller connection snooping. */
1599 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1600 ovs_rundir, br->name));
1601 svec_init(&old_snoops);
1602 ofproto_get_snoops(br->ofproto, &old_snoops);
1603 if (!svec_equal(&snoops, &old_snoops)) {
1604 ofproto_set_snoops(br->ofproto, &snoops);
1606 svec_destroy(&snoops);
1607 svec_destroy(&old_snoops);
1609 mirror_reconfigure(br);
1612 /* Initializes 'oc' appropriately as a management service controller for
1615 * The caller must free oc->target when it is no longer needed. */
1617 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1618 struct ofproto_controller *oc)
1620 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir, br->name);
1621 oc->max_backoff = 0;
1622 oc->probe_interval = 60;
1623 oc->band = OFPROTO_OUT_OF_BAND;
1624 oc->accept_re = NULL;
1625 oc->update_resolv_conf = false;
1627 oc->burst_limit = 0;
1630 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1632 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1633 struct ofproto_controller *oc)
1635 oc->target = c->target;
1636 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1637 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1638 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1639 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1640 oc->accept_re = c->discover_accept_regex;
1641 oc->update_resolv_conf = c->discover_update_resolv_conf;
1642 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1643 oc->burst_limit = (c->controller_burst_limit
1644 ? *c->controller_burst_limit : 0);
1647 /* Configures the IP stack for 'br''s local interface properly according to the
1648 * configuration in 'c'. */
1650 bridge_configure_local_iface_netdev(struct bridge *br,
1651 struct ovsrec_controller *c)
1653 struct netdev *netdev;
1654 struct in_addr mask, gateway;
1656 struct iface *local_iface;
1659 /* Controller discovery does its own TCP/IP configuration later. */
1660 if (strcmp(c->target, "discover")) {
1664 /* If there's no local interface or no IP address, give up. */
1665 local_iface = bridge_get_local_iface(br);
1666 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1670 /* Bring up the local interface. */
1671 netdev = local_iface->netdev;
1672 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1674 /* Configure the IP address and netmask. */
1675 if (!c->local_netmask
1676 || !inet_aton(c->local_netmask, &mask)
1678 mask.s_addr = guess_netmask(ip.s_addr);
1680 if (!netdev_set_in4(netdev, ip, mask)) {
1681 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1682 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1685 /* Configure the default gateway. */
1686 if (c->local_gateway
1687 && inet_aton(c->local_gateway, &gateway)
1688 && gateway.s_addr) {
1689 if (!netdev_add_router(netdev, gateway)) {
1690 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1691 br->name, IP_ARGS(&gateway.s_addr));
1697 bridge_reconfigure_remotes(struct bridge *br,
1698 const struct sockaddr_in *managers,
1701 struct ovsrec_controller **controllers;
1702 size_t n_controllers;
1705 struct ofproto_controller *ocs;
1709 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1710 had_primary = ofproto_has_primary_controller(br->ofproto);
1712 n_controllers = bridge_get_controllers(br, &controllers);
1714 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1717 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1718 for (i = 0; i < n_controllers; i++) {
1719 struct ovsrec_controller *c = controllers[i];
1721 if (!strncmp(c->target, "punix:", 6)
1722 || !strncmp(c->target, "unix:", 5)) {
1723 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1725 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1726 * domain sockets and overwriting arbitrary local files. */
1727 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1728 "\"%s\" due to possibility for remote exploit",
1729 dpif_name(br->dpif), c->target);
1733 bridge_configure_local_iface_netdev(br, c);
1734 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs++]);
1737 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1738 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1741 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1742 ofproto_flush_flows(br->ofproto);
1745 /* If there are no controllers and the bridge is in standalone
1746 * mode, set up a flow that matches every packet and directs
1747 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1748 * switch is in secure mode and we won't pass any traffic until
1749 * a controller has been defined and it tells us to do so. */
1751 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1752 union ofp_action action;
1755 memset(&action, 0, sizeof action);
1756 action.type = htons(OFPAT_OUTPUT);
1757 action.output.len = htons(sizeof action);
1758 action.output.port = htons(OFPP_NORMAL);
1759 memset(&flow, 0, sizeof flow);
1760 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1765 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1770 for (i = 0; i < br->n_ports; i++) {
1771 struct port *port = br->ports[i];
1772 for (j = 0; j < port->n_ifaces; j++) {
1773 struct iface *iface = port->ifaces[j];
1774 shash_add_once(ifaces, iface->name, iface);
1776 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1777 shash_add_once(ifaces, port->name, NULL);
1782 /* For robustness, in case the administrator moves around datapath ports behind
1783 * our back, we re-check all the datapath port numbers here.
1785 * This function will set the 'dp_ifidx' members of interfaces that have
1786 * disappeared to -1, so only call this function from a context where those
1787 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1788 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1789 * datapath, which doesn't support UINT16_MAX+1 ports. */
1791 bridge_fetch_dp_ifaces(struct bridge *br)
1793 struct odp_port *dpif_ports;
1794 size_t n_dpif_ports;
1797 /* Reset all interface numbers. */
1798 for (i = 0; i < br->n_ports; i++) {
1799 struct port *port = br->ports[i];
1800 for (j = 0; j < port->n_ifaces; j++) {
1801 struct iface *iface = port->ifaces[j];
1802 iface->dp_ifidx = -1;
1805 hmap_clear(&br->ifaces);
1807 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1808 for (i = 0; i < n_dpif_ports; i++) {
1809 struct odp_port *p = &dpif_ports[i];
1810 struct iface *iface = iface_lookup(br, p->devname);
1812 if (iface->dp_ifidx >= 0) {
1813 VLOG_WARN("%s reported interface %s twice",
1814 dpif_name(br->dpif), p->devname);
1815 } else if (iface_from_dp_ifidx(br, p->port)) {
1816 VLOG_WARN("%s reported interface %"PRIu16" twice",
1817 dpif_name(br->dpif), p->port);
1819 iface->dp_ifidx = p->port;
1820 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
1821 hash_int(iface->dp_ifidx, 0));
1825 int64_t ofport = (iface->dp_ifidx >= 0
1826 ? odp_port_to_ofp_port(iface->dp_ifidx)
1828 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1835 /* Bridge packet processing functions. */
1838 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1840 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1843 static struct bond_entry *
1844 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1846 return &port->bond_hash[bond_hash(mac)];
1850 bond_choose_iface(const struct port *port)
1852 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1853 size_t i, best_down_slave = -1;
1854 long long next_delay_expiration = LLONG_MAX;
1856 for (i = 0; i < port->n_ifaces; i++) {
1857 struct iface *iface = port->ifaces[i];
1859 if (iface->enabled) {
1861 } else if (iface->delay_expires < next_delay_expiration) {
1862 best_down_slave = i;
1863 next_delay_expiration = iface->delay_expires;
1867 if (best_down_slave != -1) {
1868 struct iface *iface = port->ifaces[best_down_slave];
1870 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1871 "since no other interface is up", iface->name,
1872 iface->delay_expires - time_msec());
1873 bond_enable_slave(iface, true);
1876 return best_down_slave;
1880 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1881 uint16_t *dp_ifidx, tag_type *tags)
1883 struct iface *iface;
1885 assert(port->n_ifaces);
1886 if (port->n_ifaces == 1) {
1887 iface = port->ifaces[0];
1889 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1890 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1891 || !port->ifaces[e->iface_idx]->enabled) {
1892 /* XXX select interface properly. The current interface selection
1893 * is only good for testing the rebalancing code. */
1894 e->iface_idx = bond_choose_iface(port);
1895 if (e->iface_idx < 0) {
1896 *tags |= port->no_ifaces_tag;
1899 e->iface_tag = tag_create_random();
1900 ((struct port *) port)->bond_compat_is_stale = true;
1902 *tags |= e->iface_tag;
1903 iface = port->ifaces[e->iface_idx];
1905 *dp_ifidx = iface->dp_ifidx;
1906 *tags |= iface->tag; /* Currently only used for bonding. */
1911 bond_link_status_update(struct iface *iface, bool carrier)
1913 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1914 struct port *port = iface->port;
1916 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1917 /* Nothing to do. */
1920 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1921 iface->name, carrier ? "detected" : "dropped");
1922 if (carrier == iface->enabled) {
1923 iface->delay_expires = LLONG_MAX;
1924 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1925 iface->name, carrier ? "disabled" : "enabled");
1926 } else if (carrier && port->active_iface < 0) {
1927 bond_enable_slave(iface, true);
1928 if (port->updelay) {
1929 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1930 "other interface is up", iface->name, port->updelay);
1933 int delay = carrier ? port->updelay : port->downdelay;
1934 iface->delay_expires = time_msec() + delay;
1937 "interface %s: will be %s if it stays %s for %d ms",
1939 carrier ? "enabled" : "disabled",
1940 carrier ? "up" : "down",
1947 bond_choose_active_iface(struct port *port)
1949 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1951 port->active_iface = bond_choose_iface(port);
1952 port->active_iface_tag = tag_create_random();
1953 if (port->active_iface >= 0) {
1954 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1955 port->name, port->ifaces[port->active_iface]->name);
1957 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1963 bond_enable_slave(struct iface *iface, bool enable)
1965 struct port *port = iface->port;
1966 struct bridge *br = port->bridge;
1968 /* This acts as a recursion check. If the act of disabling a slave
1969 * causes a different slave to be enabled, the flag will allow us to
1970 * skip redundant work when we reenter this function. It must be
1971 * cleared on exit to keep things safe with multiple bonds. */
1972 static bool moving_active_iface = false;
1974 iface->delay_expires = LLONG_MAX;
1975 if (enable == iface->enabled) {
1979 iface->enabled = enable;
1980 if (!iface->enabled) {
1981 VLOG_WARN("interface %s: disabled", iface->name);
1982 ofproto_revalidate(br->ofproto, iface->tag);
1983 if (iface->port_ifidx == port->active_iface) {
1984 ofproto_revalidate(br->ofproto,
1985 port->active_iface_tag);
1987 /* Disabling a slave can lead to another slave being immediately
1988 * enabled if there will be no active slaves but one is waiting
1989 * on an updelay. In this case we do not need to run most of the
1990 * code for the newly enabled slave since there was no period
1991 * without an active slave and it is redundant with the disabling
1993 moving_active_iface = true;
1994 bond_choose_active_iface(port);
1996 bond_send_learning_packets(port);
1998 VLOG_WARN("interface %s: enabled", iface->name);
1999 if (port->active_iface < 0 && !moving_active_iface) {
2000 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2001 bond_choose_active_iface(port);
2002 bond_send_learning_packets(port);
2004 iface->tag = tag_create_random();
2007 moving_active_iface = false;
2008 port->bond_compat_is_stale = true;
2011 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2012 * bond interface. */
2014 bond_update_fake_iface_stats(struct port *port)
2016 struct netdev_stats bond_stats;
2017 struct netdev *bond_dev;
2020 memset(&bond_stats, 0, sizeof bond_stats);
2022 for (i = 0; i < port->n_ifaces; i++) {
2023 struct netdev_stats slave_stats;
2025 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2026 /* XXX: We swap the stats here because they are swapped back when
2027 * reported by the internal device. The reason for this is
2028 * internal devices normally represent packets going into the system
2029 * but when used as fake bond device they represent packets leaving
2030 * the system. We really should do this in the internal device
2031 * itself because changing it here reverses the counts from the
2032 * perspective of the switch. However, the internal device doesn't
2033 * know what type of device it represents so we have to do it here
2035 bond_stats.tx_packets += slave_stats.rx_packets;
2036 bond_stats.tx_bytes += slave_stats.rx_bytes;
2037 bond_stats.rx_packets += slave_stats.tx_packets;
2038 bond_stats.rx_bytes += slave_stats.tx_bytes;
2042 if (!netdev_open_default(port->name, &bond_dev)) {
2043 netdev_set_stats(bond_dev, &bond_stats);
2044 netdev_close(bond_dev);
2049 bond_run(struct bridge *br)
2053 for (i = 0; i < br->n_ports; i++) {
2054 struct port *port = br->ports[i];
2056 if (port->n_ifaces >= 2) {
2057 for (j = 0; j < port->n_ifaces; j++) {
2058 struct iface *iface = port->ifaces[j];
2059 if (time_msec() >= iface->delay_expires) {
2060 bond_enable_slave(iface, !iface->enabled);
2064 if (port->bond_fake_iface
2065 && time_msec() >= port->bond_next_fake_iface_update) {
2066 bond_update_fake_iface_stats(port);
2067 port->bond_next_fake_iface_update = time_msec() + 1000;
2071 if (port->bond_compat_is_stale) {
2072 port->bond_compat_is_stale = false;
2073 port_update_bond_compat(port);
2079 bond_wait(struct bridge *br)
2083 for (i = 0; i < br->n_ports; i++) {
2084 struct port *port = br->ports[i];
2085 if (port->n_ifaces < 2) {
2088 for (j = 0; j < port->n_ifaces; j++) {
2089 struct iface *iface = port->ifaces[j];
2090 if (iface->delay_expires != LLONG_MAX) {
2091 poll_timer_wait_until(iface->delay_expires);
2094 if (port->bond_fake_iface) {
2095 poll_timer_wait_until(port->bond_next_fake_iface_update);
2101 set_dst(struct dst *p, const flow_t *flow,
2102 const struct port *in_port, const struct port *out_port,
2105 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2106 : in_port->vlan >= 0 ? in_port->vlan
2107 : ntohs(flow->dl_vlan));
2108 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
2112 swap_dst(struct dst *p, struct dst *q)
2114 struct dst tmp = *p;
2119 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2120 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2121 * that we push to the datapath. We could in fact fully sort the array by
2122 * vlan, but in most cases there are at most two different vlan tags so that's
2123 * possibly overkill.) */
2125 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2127 struct dst *first = dsts;
2128 struct dst *last = dsts + n_dsts;
2130 while (first != last) {
2132 * - All dsts < first have vlan == 'vlan'.
2133 * - All dsts >= last have vlan != 'vlan'.
2134 * - first < last. */
2135 while (first->vlan == vlan) {
2136 if (++first == last) {
2141 /* Same invariants, plus one additional:
2142 * - first->vlan != vlan.
2144 while (last[-1].vlan != vlan) {
2145 if (--last == first) {
2150 /* Same invariants, plus one additional:
2151 * - last[-1].vlan == vlan.*/
2152 swap_dst(first++, --last);
2157 mirror_mask_ffs(mirror_mask_t mask)
2159 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2164 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2165 const struct dst *test)
2168 for (i = 0; i < n_dsts; i++) {
2169 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2177 port_trunks_vlan(const struct port *port, uint16_t vlan)
2179 return (port->vlan < 0
2180 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2184 port_includes_vlan(const struct port *port, uint16_t vlan)
2186 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2190 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2191 const struct port *in_port, const struct port *out_port,
2192 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2194 mirror_mask_t mirrors = in_port->src_mirrors;
2195 struct dst *dst = dsts;
2198 if (out_port == FLOOD_PORT) {
2199 /* XXX use ODP_FLOOD if no vlans or bonding. */
2200 /* XXX even better, define each VLAN as a datapath port group */
2201 for (i = 0; i < br->n_ports; i++) {
2202 struct port *port = br->ports[i];
2203 if (port != in_port && port_includes_vlan(port, vlan)
2204 && !port->is_mirror_output_port
2205 && set_dst(dst, flow, in_port, port, tags)) {
2206 mirrors |= port->dst_mirrors;
2210 *nf_output_iface = NF_OUT_FLOOD;
2211 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2212 *nf_output_iface = dst->dp_ifidx;
2213 mirrors |= out_port->dst_mirrors;
2218 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2219 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2221 if (set_dst(dst, flow, in_port, m->out_port, tags)
2222 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2226 for (i = 0; i < br->n_ports; i++) {
2227 struct port *port = br->ports[i];
2228 if (port_includes_vlan(port, m->out_vlan)
2229 && set_dst(dst, flow, in_port, port, tags))
2233 if (port->vlan < 0) {
2234 dst->vlan = m->out_vlan;
2236 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2240 /* Use the vlan tag on the original flow instead of
2241 * the one passed in the vlan parameter. This ensures
2242 * that we compare the vlan from before any implicit
2243 * tagging tags place. This is necessary because
2244 * dst->vlan is the final vlan, after removing implicit
2246 flow_vlan = ntohs(flow->dl_vlan);
2247 if (flow_vlan == 0) {
2248 flow_vlan = OFP_VLAN_NONE;
2250 if (port == in_port && dst->vlan == flow_vlan) {
2251 /* Don't send out input port on same VLAN. */
2259 mirrors &= mirrors - 1;
2262 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2266 static void OVS_UNUSED
2267 print_dsts(const struct dst *dsts, size_t n)
2269 for (; n--; dsts++) {
2270 printf(">p%"PRIu16, dsts->dp_ifidx);
2271 if (dsts->vlan != OFP_VLAN_NONE) {
2272 printf("v%"PRIu16, dsts->vlan);
2278 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2279 const struct port *in_port, const struct port *out_port,
2280 tag_type *tags, struct odp_actions *actions,
2281 uint16_t *nf_output_iface)
2283 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2285 const struct dst *p;
2288 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2291 cur_vlan = ntohs(flow->dl_vlan);
2292 for (p = dsts; p < &dsts[n_dsts]; p++) {
2293 union odp_action *a;
2294 if (p->vlan != cur_vlan) {
2295 if (p->vlan == OFP_VLAN_NONE) {
2296 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2298 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2299 a->vlan_vid.vlan_vid = htons(p->vlan);
2303 a = odp_actions_add(actions, ODPAT_OUTPUT);
2304 a->output.port = p->dp_ifidx;
2308 /* Returns the effective vlan of a packet, taking into account both the
2309 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2310 * the packet is untagged and -1 indicates it has an invalid header and
2311 * should be dropped. */
2312 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2313 struct port *in_port, bool have_packet)
2315 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2316 * belongs to VLAN 0, so we should treat both cases identically. (In the
2317 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2318 * presumably to allow a priority to be specified. In the latter case, the
2319 * packet does not have any 802.1Q header.) */
2320 int vlan = ntohs(flow->dl_vlan);
2321 if (vlan == OFP_VLAN_NONE) {
2324 if (in_port->vlan >= 0) {
2326 /* XXX support double tagging? */
2328 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2329 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2330 "packet received on port %s configured with "
2331 "implicit VLAN %"PRIu16,
2332 br->name, ntohs(flow->dl_vlan),
2333 in_port->name, in_port->vlan);
2337 vlan = in_port->vlan;
2339 if (!port_includes_vlan(in_port, vlan)) {
2341 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2342 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2343 "packet received on port %s not configured for "
2345 br->name, vlan, in_port->name, vlan);
2354 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2355 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2356 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2358 is_gratuitous_arp(const flow_t *flow)
2360 return (flow->dl_type == htons(ETH_TYPE_ARP)
2361 && eth_addr_is_broadcast(flow->dl_dst)
2362 && (flow->nw_proto == ARP_OP_REPLY
2363 || (flow->nw_proto == ARP_OP_REQUEST
2364 && flow->nw_src == flow->nw_dst)));
2368 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2369 struct port *in_port)
2371 enum grat_arp_lock_type lock_type;
2374 /* We don't want to learn from gratuitous ARP packets that are reflected
2375 * back over bond slaves so we lock the learning table. */
2376 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2377 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2378 GRAT_ARP_LOCK_CHECK;
2380 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2383 /* The log messages here could actually be useful in debugging,
2384 * so keep the rate limit relatively high. */
2385 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2387 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2388 "on port %s in VLAN %d",
2389 br->name, ETH_ADDR_ARGS(flow->dl_src),
2390 in_port->name, vlan);
2391 ofproto_revalidate(br->ofproto, rev_tag);
2395 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2396 * dropped. Returns true if they may be forwarded, false if they should be
2399 * If 'have_packet' is true, it indicates that the caller is processing a
2400 * received packet. If 'have_packet' is false, then the caller is just
2401 * revalidating an existing flow because configuration has changed. Either
2402 * way, 'have_packet' only affects logging (there is no point in logging errors
2403 * during revalidation).
2405 * Sets '*in_portp' to the input port. This will be a null pointer if
2406 * flow->in_port does not designate a known input port (in which case
2407 * is_admissible() returns false).
2409 * When returning true, sets '*vlanp' to the effective VLAN of the input
2410 * packet, as returned by flow_get_vlan().
2412 * May also add tags to '*tags', although the current implementation only does
2413 * so in one special case.
2416 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2417 tag_type *tags, int *vlanp, struct port **in_portp)
2419 struct iface *in_iface;
2420 struct port *in_port;
2423 /* Find the interface and port structure for the received packet. */
2424 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2426 /* No interface? Something fishy... */
2428 /* Odd. A few possible reasons here:
2430 * - We deleted an interface but there are still a few packets
2431 * queued up from it.
2433 * - Someone externally added an interface (e.g. with "ovs-dpctl
2434 * add-if") that we don't know about.
2436 * - Packet arrived on the local port but the local port is not
2437 * one of our bridge ports.
2439 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2441 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2442 "interface %"PRIu16, br->name, flow->in_port);
2448 *in_portp = in_port = in_iface->port;
2449 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2454 /* Drop frames for reserved multicast addresses. */
2455 if (eth_addr_is_reserved(flow->dl_dst)) {
2459 /* Drop frames on ports reserved for mirroring. */
2460 if (in_port->is_mirror_output_port) {
2462 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2463 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2464 "%s, which is reserved exclusively for mirroring",
2465 br->name, in_port->name);
2470 /* Packets received on bonds need special attention to avoid duplicates. */
2471 if (in_port->n_ifaces > 1) {
2473 bool is_grat_arp_locked;
2475 if (eth_addr_is_multicast(flow->dl_dst)) {
2476 *tags |= in_port->active_iface_tag;
2477 if (in_port->active_iface != in_iface->port_ifidx) {
2478 /* Drop all multicast packets on inactive slaves. */
2483 /* Drop all packets for which we have learned a different input
2484 * port, because we probably sent the packet on one slave and got
2485 * it back on the other. Gratuitous ARP packets are an exception
2486 * to this rule: the host has moved to another switch. The exception
2487 * to the exception is if we locked the learning table to avoid
2488 * reflections on bond slaves. If this is the case, just drop the
2490 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2491 &is_grat_arp_locked);
2492 if (src_idx != -1 && src_idx != in_port->port_idx &&
2493 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2501 /* If the composed actions may be applied to any packet in the given 'flow',
2502 * returns true. Otherwise, the actions should only be applied to 'packet', or
2503 * not at all, if 'packet' was NULL. */
2505 process_flow(struct bridge *br, const flow_t *flow,
2506 const struct ofpbuf *packet, struct odp_actions *actions,
2507 tag_type *tags, uint16_t *nf_output_iface)
2509 struct port *in_port;
2510 struct port *out_port;
2514 /* Check whether we should drop packets in this flow. */
2515 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2520 /* Learn source MAC (but don't try to learn from revalidation). */
2522 update_learning_table(br, flow, vlan, in_port);
2525 /* Determine output port. */
2526 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2528 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2529 out_port = br->ports[out_port_idx];
2530 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2531 /* If we are revalidating but don't have a learning entry then
2532 * eject the flow. Installing a flow that floods packets opens
2533 * up a window of time where we could learn from a packet reflected
2534 * on a bond and blackhole packets before the learning table is
2535 * updated to reflect the correct port. */
2538 out_port = FLOOD_PORT;
2541 /* Don't send packets out their input ports. */
2542 if (in_port == out_port) {
2548 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2555 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2558 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2559 const struct ofp_phy_port *opp,
2562 struct bridge *br = br_;
2563 struct iface *iface;
2566 if (reason == OFPPR_DELETE || !br->has_bonded_ports) {
2570 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2576 if (port->n_ifaces > 1) {
2577 bool up = !(opp->state & OFPPS_LINK_DOWN);
2578 bond_link_status_update(iface, up);
2579 port_update_bond_compat(port);
2584 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2585 struct odp_actions *actions, tag_type *tags,
2586 uint16_t *nf_output_iface, void *br_)
2588 struct bridge *br = br_;
2590 COVERAGE_INC(bridge_process_flow);
2592 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2596 bridge_account_flow_ofhook_cb(const flow_t *flow, tag_type tags,
2597 const union odp_action *actions,
2598 size_t n_actions, unsigned long long int n_bytes,
2601 struct bridge *br = br_;
2602 const union odp_action *a;
2603 struct port *in_port;
2607 /* Feed information from the active flows back into the learning table to
2608 * ensure that table is always in sync with what is actually flowing
2609 * through the datapath.
2611 * We test that 'tags' is nonzero to ensure that only flows that include an
2612 * OFPP_NORMAL action are used for learning. This works because
2613 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2614 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2615 update_learning_table(br, flow, vlan, in_port);
2618 /* Account for bond slave utilization. */
2619 if (!br->has_bonded_ports) {
2622 for (a = actions; a < &actions[n_actions]; a++) {
2623 if (a->type == ODPAT_OUTPUT) {
2624 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2625 if (out_port && out_port->n_ifaces >= 2) {
2626 struct bond_entry *e = lookup_bond_entry(out_port,
2628 e->tx_bytes += n_bytes;
2635 bridge_account_checkpoint_ofhook_cb(void *br_)
2637 struct bridge *br = br_;
2641 if (!br->has_bonded_ports) {
2646 for (i = 0; i < br->n_ports; i++) {
2647 struct port *port = br->ports[i];
2648 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2649 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2650 bond_rebalance_port(port);
2655 static struct ofhooks bridge_ofhooks = {
2656 bridge_port_changed_ofhook_cb,
2657 bridge_normal_ofhook_cb,
2658 bridge_account_flow_ofhook_cb,
2659 bridge_account_checkpoint_ofhook_cb,
2662 /* Bonding functions. */
2664 /* Statistics for a single interface on a bonded port, used for load-based
2665 * bond rebalancing. */
2666 struct slave_balance {
2667 struct iface *iface; /* The interface. */
2668 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2670 /* All the "bond_entry"s that are assigned to this interface, in order of
2671 * increasing tx_bytes. */
2672 struct bond_entry **hashes;
2676 /* Sorts pointers to pointers to bond_entries in ascending order by the
2677 * interface to which they are assigned, and within a single interface in
2678 * ascending order of bytes transmitted. */
2680 compare_bond_entries(const void *a_, const void *b_)
2682 const struct bond_entry *const *ap = a_;
2683 const struct bond_entry *const *bp = b_;
2684 const struct bond_entry *a = *ap;
2685 const struct bond_entry *b = *bp;
2686 if (a->iface_idx != b->iface_idx) {
2687 return a->iface_idx > b->iface_idx ? 1 : -1;
2688 } else if (a->tx_bytes != b->tx_bytes) {
2689 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2695 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2696 * *descending* order by number of bytes transmitted. */
2698 compare_slave_balance(const void *a_, const void *b_)
2700 const struct slave_balance *a = a_;
2701 const struct slave_balance *b = b_;
2702 if (a->iface->enabled != b->iface->enabled) {
2703 return a->iface->enabled ? -1 : 1;
2704 } else if (a->tx_bytes != b->tx_bytes) {
2705 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2712 swap_bals(struct slave_balance *a, struct slave_balance *b)
2714 struct slave_balance tmp = *a;
2719 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2720 * given that 'p' (and only 'p') might be in the wrong location.
2722 * This function invalidates 'p', since it might now be in a different memory
2725 resort_bals(struct slave_balance *p,
2726 struct slave_balance bals[], size_t n_bals)
2729 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2730 swap_bals(p, p - 1);
2732 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2733 swap_bals(p, p + 1);
2739 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2741 if (VLOG_IS_DBG_ENABLED()) {
2742 struct ds ds = DS_EMPTY_INITIALIZER;
2743 const struct slave_balance *b;
2745 for (b = bals; b < bals + n_bals; b++) {
2749 ds_put_char(&ds, ',');
2751 ds_put_format(&ds, " %s %"PRIu64"kB",
2752 b->iface->name, b->tx_bytes / 1024);
2754 if (!b->iface->enabled) {
2755 ds_put_cstr(&ds, " (disabled)");
2757 if (b->n_hashes > 0) {
2758 ds_put_cstr(&ds, " (");
2759 for (i = 0; i < b->n_hashes; i++) {
2760 const struct bond_entry *e = b->hashes[i];
2762 ds_put_cstr(&ds, " + ");
2764 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2765 e - port->bond_hash, e->tx_bytes / 1024);
2767 ds_put_cstr(&ds, ")");
2770 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2775 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2777 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2780 struct bond_entry *hash = from->hashes[hash_idx];
2781 struct port *port = from->iface->port;
2782 uint64_t delta = hash->tx_bytes;
2784 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2785 "from %s to %s (now carrying %"PRIu64"kB and "
2786 "%"PRIu64"kB load, respectively)",
2787 port->name, delta / 1024, hash - port->bond_hash,
2788 from->iface->name, to->iface->name,
2789 (from->tx_bytes - delta) / 1024,
2790 (to->tx_bytes + delta) / 1024);
2792 /* Delete element from from->hashes.
2794 * We don't bother to add the element to to->hashes because not only would
2795 * it require more work, the only purpose it would be to allow that hash to
2796 * be migrated to another slave in this rebalancing run, and there is no
2797 * point in doing that. */
2798 if (hash_idx == 0) {
2801 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2802 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2806 /* Shift load away from 'from' to 'to'. */
2807 from->tx_bytes -= delta;
2808 to->tx_bytes += delta;
2810 /* Arrange for flows to be revalidated. */
2811 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2812 hash->iface_idx = to->iface->port_ifidx;
2813 hash->iface_tag = tag_create_random();
2817 bond_rebalance_port(struct port *port)
2819 struct slave_balance bals[DP_MAX_PORTS];
2821 struct bond_entry *hashes[BOND_MASK + 1];
2822 struct slave_balance *b, *from, *to;
2823 struct bond_entry *e;
2826 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2827 * descending order of tx_bytes, so that bals[0] represents the most
2828 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2831 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2832 * array for each slave_balance structure, we sort our local array of
2833 * hashes in order by slave, so that all of the hashes for a given slave
2834 * become contiguous in memory, and then we point each 'hashes' members of
2835 * a slave_balance structure to the start of a contiguous group. */
2836 n_bals = port->n_ifaces;
2837 for (b = bals; b < &bals[n_bals]; b++) {
2838 b->iface = port->ifaces[b - bals];
2843 for (i = 0; i <= BOND_MASK; i++) {
2844 hashes[i] = &port->bond_hash[i];
2846 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2847 for (i = 0; i <= BOND_MASK; i++) {
2849 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2850 b = &bals[e->iface_idx];
2851 b->tx_bytes += e->tx_bytes;
2853 b->hashes = &hashes[i];
2858 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2859 log_bals(bals, n_bals, port);
2861 /* Discard slaves that aren't enabled (which were sorted to the back of the
2862 * array earlier). */
2863 while (!bals[n_bals - 1].iface->enabled) {
2870 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2871 to = &bals[n_bals - 1];
2872 for (from = bals; from < to; ) {
2873 uint64_t overload = from->tx_bytes - to->tx_bytes;
2874 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2875 /* The extra load on 'from' (and all less-loaded slaves), compared
2876 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2877 * it is less than ~1Mbps. No point in rebalancing. */
2879 } else if (from->n_hashes == 1) {
2880 /* 'from' only carries a single MAC hash, so we can't shift any
2881 * load away from it, even though we want to. */
2884 /* 'from' is carrying significantly more load than 'to', and that
2885 * load is split across at least two different hashes. Pick a hash
2886 * to migrate to 'to' (the least-loaded slave), given that doing so
2887 * must decrease the ratio of the load on the two slaves by at
2890 * The sort order we use means that we prefer to shift away the
2891 * smallest hashes instead of the biggest ones. There is little
2892 * reason behind this decision; we could use the opposite sort
2893 * order to shift away big hashes ahead of small ones. */
2896 for (i = 0; i < from->n_hashes; i++) {
2897 double old_ratio, new_ratio;
2898 uint64_t delta = from->hashes[i]->tx_bytes;
2900 if (delta == 0 || from->tx_bytes - delta == 0) {
2901 /* Pointless move. */
2905 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2907 if (to->tx_bytes == 0) {
2908 /* Nothing on the new slave, move it. */
2912 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2913 new_ratio = (double)(from->tx_bytes - delta) /
2914 (to->tx_bytes + delta);
2916 if (new_ratio == 0) {
2917 /* Should already be covered but check to prevent division
2922 if (new_ratio < 1) {
2923 new_ratio = 1 / new_ratio;
2926 if (old_ratio - new_ratio > 0.1) {
2927 /* Would decrease the ratio, move it. */
2931 if (i < from->n_hashes) {
2932 bond_shift_load(from, to, i);
2933 port->bond_compat_is_stale = true;
2935 /* If the result of the migration changed the relative order of
2936 * 'from' and 'to' swap them back to maintain invariants. */
2937 if (order_swapped) {
2938 swap_bals(from, to);
2941 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2942 * point to different slave_balance structures. It is only
2943 * valid to do these two operations in a row at all because we
2944 * know that 'from' will not move past 'to' and vice versa. */
2945 resort_bals(from, bals, n_bals);
2946 resort_bals(to, bals, n_bals);
2953 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2954 * historical data to decay to <1% in 7 rebalancing runs. */
2955 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2961 bond_send_learning_packets(struct port *port)
2963 struct bridge *br = port->bridge;
2964 struct mac_entry *e;
2965 struct ofpbuf packet;
2966 int error, n_packets, n_errors;
2968 if (!port->n_ifaces || port->active_iface < 0) {
2972 ofpbuf_init(&packet, 128);
2973 error = n_packets = n_errors = 0;
2974 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
2975 union ofp_action actions[2], *a;
2981 if (e->port == port->port_idx
2982 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2986 /* Compose actions. */
2987 memset(actions, 0, sizeof actions);
2990 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2991 a->vlan_vid.len = htons(sizeof *a);
2992 a->vlan_vid.vlan_vid = htons(e->vlan);
2995 a->output.type = htons(OFPAT_OUTPUT);
2996 a->output.len = htons(sizeof *a);
2997 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3002 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3004 flow_extract(&packet, 0, ODPP_NONE, &flow);
3005 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3012 ofpbuf_uninit(&packet);
3015 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3016 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3017 "packets, last error was: %s",
3018 port->name, n_errors, n_packets, strerror(error));
3020 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3021 port->name, n_packets);
3025 /* Bonding unixctl user interface functions. */
3028 bond_unixctl_list(struct unixctl_conn *conn,
3029 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3031 struct ds ds = DS_EMPTY_INITIALIZER;
3032 const struct bridge *br;
3034 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
3036 LIST_FOR_EACH (br, node, &all_bridges) {
3039 for (i = 0; i < br->n_ports; i++) {
3040 const struct port *port = br->ports[i];
3041 if (port->n_ifaces > 1) {
3044 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
3045 for (j = 0; j < port->n_ifaces; j++) {
3046 const struct iface *iface = port->ifaces[j];
3048 ds_put_cstr(&ds, ", ");
3050 ds_put_cstr(&ds, iface->name);
3052 ds_put_char(&ds, '\n');
3056 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3060 static struct port *
3061 bond_find(const char *name)
3063 const struct bridge *br;
3065 LIST_FOR_EACH (br, node, &all_bridges) {
3068 for (i = 0; i < br->n_ports; i++) {
3069 struct port *port = br->ports[i];
3070 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3079 bond_unixctl_show(struct unixctl_conn *conn,
3080 const char *args, void *aux OVS_UNUSED)
3082 struct ds ds = DS_EMPTY_INITIALIZER;
3083 const struct port *port;
3086 port = bond_find(args);
3088 unixctl_command_reply(conn, 501, "no such bond");
3092 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3093 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3094 ds_put_format(&ds, "next rebalance: %lld ms\n",
3095 port->bond_next_rebalance - time_msec());
3096 for (j = 0; j < port->n_ifaces; j++) {
3097 const struct iface *iface = port->ifaces[j];
3098 struct bond_entry *be;
3101 ds_put_format(&ds, "slave %s: %s\n",
3102 iface->name, iface->enabled ? "enabled" : "disabled");
3103 if (j == port->active_iface) {
3104 ds_put_cstr(&ds, "\tactive slave\n");
3106 if (iface->delay_expires != LLONG_MAX) {
3107 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3108 iface->enabled ? "downdelay" : "updelay",
3109 iface->delay_expires - time_msec());
3113 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3114 int hash = be - port->bond_hash;
3115 struct mac_entry *me;
3117 if (be->iface_idx != j) {
3121 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3122 hash, be->tx_bytes / 1024);
3125 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3128 if (bond_hash(me->mac) == hash
3129 && me->port != port->port_idx
3130 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3131 && dp_ifidx == iface->dp_ifidx)
3133 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3134 ETH_ADDR_ARGS(me->mac));
3139 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3144 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3145 void *aux OVS_UNUSED)
3147 char *args = (char *) args_;
3148 char *save_ptr = NULL;
3149 char *bond_s, *hash_s, *slave_s;
3150 uint8_t mac[ETH_ADDR_LEN];
3152 struct iface *iface;
3153 struct bond_entry *entry;
3156 bond_s = strtok_r(args, " ", &save_ptr);
3157 hash_s = strtok_r(NULL, " ", &save_ptr);
3158 slave_s = strtok_r(NULL, " ", &save_ptr);
3160 unixctl_command_reply(conn, 501,
3161 "usage: bond/migrate BOND HASH SLAVE");
3165 port = bond_find(bond_s);
3167 unixctl_command_reply(conn, 501, "no such bond");
3171 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3172 == ETH_ADDR_SCAN_COUNT) {
3173 hash = bond_hash(mac);
3174 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3175 hash = atoi(hash_s) & BOND_MASK;
3177 unixctl_command_reply(conn, 501, "bad hash");
3181 iface = port_lookup_iface(port, slave_s);
3183 unixctl_command_reply(conn, 501, "no such slave");
3187 if (!iface->enabled) {
3188 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3192 entry = &port->bond_hash[hash];
3193 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3194 entry->iface_idx = iface->port_ifidx;
3195 entry->iface_tag = tag_create_random();
3196 port->bond_compat_is_stale = true;
3197 unixctl_command_reply(conn, 200, "migrated");
3201 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3202 void *aux OVS_UNUSED)
3204 char *args = (char *) args_;
3205 char *save_ptr = NULL;
3206 char *bond_s, *slave_s;
3208 struct iface *iface;
3210 bond_s = strtok_r(args, " ", &save_ptr);
3211 slave_s = strtok_r(NULL, " ", &save_ptr);
3213 unixctl_command_reply(conn, 501,
3214 "usage: bond/set-active-slave BOND SLAVE");
3218 port = bond_find(bond_s);
3220 unixctl_command_reply(conn, 501, "no such bond");
3224 iface = port_lookup_iface(port, slave_s);
3226 unixctl_command_reply(conn, 501, "no such slave");
3230 if (!iface->enabled) {
3231 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3235 if (port->active_iface != iface->port_ifidx) {
3236 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3237 port->active_iface = iface->port_ifidx;
3238 port->active_iface_tag = tag_create_random();
3239 VLOG_INFO("port %s: active interface is now %s",
3240 port->name, iface->name);
3241 bond_send_learning_packets(port);
3242 unixctl_command_reply(conn, 200, "done");
3244 unixctl_command_reply(conn, 200, "no change");
3249 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3251 char *args = (char *) args_;
3252 char *save_ptr = NULL;
3253 char *bond_s, *slave_s;
3255 struct iface *iface;
3257 bond_s = strtok_r(args, " ", &save_ptr);
3258 slave_s = strtok_r(NULL, " ", &save_ptr);
3260 unixctl_command_reply(conn, 501,
3261 "usage: bond/enable/disable-slave BOND SLAVE");
3265 port = bond_find(bond_s);
3267 unixctl_command_reply(conn, 501, "no such bond");
3271 iface = port_lookup_iface(port, slave_s);
3273 unixctl_command_reply(conn, 501, "no such slave");
3277 bond_enable_slave(iface, enable);
3278 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3282 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3283 void *aux OVS_UNUSED)
3285 enable_slave(conn, args, true);
3289 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3290 void *aux OVS_UNUSED)
3292 enable_slave(conn, args, false);
3296 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3297 void *aux OVS_UNUSED)
3299 uint8_t mac[ETH_ADDR_LEN];
3303 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3304 == ETH_ADDR_SCAN_COUNT) {
3305 hash = bond_hash(mac);
3307 hash_cstr = xasprintf("%u", hash);
3308 unixctl_command_reply(conn, 200, hash_cstr);
3311 unixctl_command_reply(conn, 501, "invalid mac");
3318 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3319 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3320 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3321 unixctl_command_register("bond/set-active-slave",
3322 bond_unixctl_set_active_slave, NULL);
3323 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3325 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3327 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3330 /* Port functions. */
3332 static struct port *
3333 port_create(struct bridge *br, const char *name)
3337 port = xzalloc(sizeof *port);
3339 port->port_idx = br->n_ports;
3341 port->trunks = NULL;
3342 port->name = xstrdup(name);
3343 port->active_iface = -1;
3345 if (br->n_ports >= br->allocated_ports) {
3346 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3349 br->ports[br->n_ports++] = port;
3350 shash_add_assert(&br->port_by_name, port->name, port);
3352 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3359 get_port_other_config(const struct ovsrec_port *port, const char *key,
3360 const char *default_value)
3364 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3366 return value ? value : default_value;
3370 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3372 struct shash new_ifaces;
3375 /* Collect list of new interfaces. */
3376 shash_init(&new_ifaces);
3377 for (i = 0; i < cfg->n_interfaces; i++) {
3378 const char *name = cfg->interfaces[i]->name;
3379 shash_add_once(&new_ifaces, name, NULL);
3382 /* Get rid of deleted interfaces. */
3383 for (i = 0; i < port->n_ifaces; ) {
3384 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3385 iface_destroy(port->ifaces[i]);
3391 shash_destroy(&new_ifaces);
3395 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3397 struct shash new_ifaces;
3398 long long int next_rebalance;
3399 unsigned long *trunks;
3405 /* Update settings. */
3406 port->updelay = cfg->bond_updelay;
3407 if (port->updelay < 0) {
3410 port->downdelay = cfg->bond_downdelay;
3411 if (port->downdelay < 0) {
3412 port->downdelay = 0;
3414 port->bond_rebalance_interval = atoi(
3415 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3416 if (port->bond_rebalance_interval < 1000) {
3417 port->bond_rebalance_interval = 1000;
3419 next_rebalance = time_msec() + port->bond_rebalance_interval;
3420 if (port->bond_next_rebalance > next_rebalance) {
3421 port->bond_next_rebalance = next_rebalance;
3424 /* Add new interfaces and update 'cfg' member of existing ones. */
3425 shash_init(&new_ifaces);
3426 for (i = 0; i < cfg->n_interfaces; i++) {
3427 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3428 struct iface *iface;
3430 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3431 VLOG_WARN("port %s: %s specified twice as port interface",
3432 port->name, if_cfg->name);
3436 iface = iface_lookup(port->bridge, if_cfg->name);
3438 if (iface->port != port) {
3439 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3441 port->bridge->name, if_cfg->name, iface->port->name);
3444 iface->cfg = if_cfg;
3446 iface = iface_create(port, if_cfg);
3449 /* Determine interface type. The local port always has type
3450 * "internal". Other ports take their type from the database and
3451 * default to "system" if none is specified. */
3452 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
3453 : if_cfg->type[0] ? if_cfg->type
3456 shash_destroy(&new_ifaces);
3461 if (port->n_ifaces < 2) {
3463 if (vlan >= 0 && vlan <= 4095) {
3464 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3469 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3470 * they even work as-is. But they have not been tested. */
3471 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3475 if (port->vlan != vlan) {
3477 bridge_flush(port->bridge);
3480 /* Get trunked VLANs. */
3482 if (vlan < 0 && cfg->n_trunks) {
3485 trunks = bitmap_allocate(4096);
3487 for (i = 0; i < cfg->n_trunks; i++) {
3488 int trunk = cfg->trunks[i];
3490 bitmap_set1(trunks, trunk);
3496 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3497 port->name, cfg->n_trunks);
3499 if (n_errors == cfg->n_trunks) {
3500 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3502 bitmap_free(trunks);
3505 } else if (vlan >= 0 && cfg->n_trunks) {
3506 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3510 ? port->trunks != NULL
3511 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3512 bridge_flush(port->bridge);
3514 bitmap_free(port->trunks);
3515 port->trunks = trunks;
3519 port_destroy(struct port *port)
3522 struct bridge *br = port->bridge;
3526 proc_net_compat_update_vlan(port->name, NULL, 0);
3527 proc_net_compat_update_bond(port->name, NULL);
3529 for (i = 0; i < MAX_MIRRORS; i++) {
3530 struct mirror *m = br->mirrors[i];
3531 if (m && m->out_port == port) {
3536 while (port->n_ifaces > 0) {
3537 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3540 shash_find_and_delete_assert(&br->port_by_name, port->name);
3542 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3543 del->port_idx = port->port_idx;
3546 bitmap_free(port->trunks);
3553 static struct port *
3554 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3556 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3557 return iface ? iface->port : NULL;
3560 static struct port *
3561 port_lookup(const struct bridge *br, const char *name)
3563 return shash_find_data(&br->port_by_name, name);
3566 static struct iface *
3567 port_lookup_iface(const struct port *port, const char *name)
3569 struct iface *iface = iface_lookup(port->bridge, name);
3570 return iface && iface->port == port ? iface : NULL;
3574 port_update_bonding(struct port *port)
3576 if (port->n_ifaces < 2) {
3577 /* Not a bonded port. */
3578 if (port->bond_hash) {
3579 free(port->bond_hash);
3580 port->bond_hash = NULL;
3581 port->bond_compat_is_stale = true;
3582 port->bond_fake_iface = false;
3585 if (!port->bond_hash) {
3588 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3589 for (i = 0; i <= BOND_MASK; i++) {
3590 struct bond_entry *e = &port->bond_hash[i];
3594 port->no_ifaces_tag = tag_create_random();
3595 bond_choose_active_iface(port);
3596 port->bond_next_rebalance
3597 = time_msec() + port->bond_rebalance_interval;
3599 if (port->cfg->bond_fake_iface) {
3600 port->bond_next_fake_iface_update = time_msec();
3603 port->bond_compat_is_stale = true;
3604 port->bond_fake_iface = port->cfg->bond_fake_iface;
3609 port_update_bond_compat(struct port *port)
3611 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3612 struct compat_bond bond;
3615 if (port->n_ifaces < 2) {
3616 proc_net_compat_update_bond(port->name, NULL);
3621 bond.updelay = port->updelay;
3622 bond.downdelay = port->downdelay;
3625 bond.hashes = compat_hashes;
3626 if (port->bond_hash) {
3627 const struct bond_entry *e;
3628 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3629 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3630 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3631 cbh->hash = e - port->bond_hash;
3632 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3637 bond.n_slaves = port->n_ifaces;
3638 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3639 for (i = 0; i < port->n_ifaces; i++) {
3640 struct iface *iface = port->ifaces[i];
3641 struct compat_bond_slave *slave = &bond.slaves[i];
3642 slave->name = iface->name;
3644 /* We need to make the same determination as the Linux bonding
3645 * code to determine whether a slave should be consider "up".
3646 * The Linux function bond_miimon_inspect() supports four
3647 * BOND_LINK_* states:
3649 * - BOND_LINK_UP: carrier detected, updelay has passed.
3650 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3651 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3652 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3654 * The function bond_info_show_slave() only considers BOND_LINK_UP
3655 * to be "up" and anything else to be "down".
3657 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3661 netdev_get_etheraddr(iface->netdev, slave->mac);
3664 if (port->bond_fake_iface) {
3665 struct netdev *bond_netdev;
3667 if (!netdev_open_default(port->name, &bond_netdev)) {
3669 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3671 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3673 netdev_close(bond_netdev);
3677 proc_net_compat_update_bond(port->name, &bond);
3682 port_update_vlan_compat(struct port *port)
3684 struct bridge *br = port->bridge;
3685 char *vlandev_name = NULL;
3687 if (port->vlan > 0) {
3688 /* Figure out the name that the VLAN device should actually have, if it
3689 * existed. This takes some work because the VLAN device would not
3690 * have port->name in its name; rather, it would have the trunk port's
3691 * name, and 'port' would be attached to a bridge that also had the
3692 * VLAN device one of its ports. So we need to find a trunk port that
3693 * includes port->vlan.
3695 * There might be more than one candidate. This doesn't happen on
3696 * XenServer, so if it happens we just pick the first choice in
3697 * alphabetical order instead of creating multiple VLAN devices. */
3699 for (i = 0; i < br->n_ports; i++) {
3700 struct port *p = br->ports[i];
3701 if (port_trunks_vlan(p, port->vlan)
3703 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3705 uint8_t ea[ETH_ADDR_LEN];
3706 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3707 if (!eth_addr_is_multicast(ea) &&
3708 !eth_addr_is_reserved(ea) &&
3709 !eth_addr_is_zero(ea)) {
3710 vlandev_name = p->name;
3715 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3718 /* Interface functions. */
3720 static struct iface *
3721 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3723 struct bridge *br = port->bridge;
3724 struct iface *iface;
3725 char *name = if_cfg->name;
3727 iface = xzalloc(sizeof *iface);
3729 iface->port_ifidx = port->n_ifaces;
3730 iface->name = xstrdup(name);
3731 iface->dp_ifidx = -1;
3732 iface->tag = tag_create_random();
3733 iface->delay_expires = LLONG_MAX;
3734 iface->netdev = NULL;
3735 iface->cfg = if_cfg;
3737 shash_add_assert(&br->iface_by_name, iface->name, iface);
3739 if (port->n_ifaces >= port->allocated_ifaces) {
3740 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3741 sizeof *port->ifaces);
3743 port->ifaces[port->n_ifaces++] = iface;
3744 if (port->n_ifaces > 1) {
3745 br->has_bonded_ports = true;
3748 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3756 iface_destroy(struct iface *iface)
3759 struct port *port = iface->port;
3760 struct bridge *br = port->bridge;
3761 bool del_active = port->active_iface == iface->port_ifidx;
3764 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3766 if (iface->dp_ifidx >= 0) {
3767 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
3770 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3771 del->port_ifidx = iface->port_ifidx;
3773 netdev_close(iface->netdev);
3776 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3777 bond_choose_active_iface(port);
3778 bond_send_learning_packets(port);
3784 bridge_flush(port->bridge);
3788 static struct iface *
3789 iface_lookup(const struct bridge *br, const char *name)
3791 return shash_find_data(&br->iface_by_name, name);
3794 static struct iface *
3795 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3797 struct iface *iface;
3799 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
3800 hash_int(dp_ifidx, 0), &br->ifaces) {
3801 if (iface->dp_ifidx == dp_ifidx) {
3808 /* Set Ethernet address of 'iface', if one is specified in the configuration
3811 iface_set_mac(struct iface *iface)
3813 uint8_t ea[ETH_ADDR_LEN];
3815 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3816 if (eth_addr_is_multicast(ea)) {
3817 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3819 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3820 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3821 iface->name, iface->name);
3823 int error = netdev_set_etheraddr(iface->netdev, ea);
3825 VLOG_ERR("interface %s: setting MAC failed (%s)",
3826 iface->name, strerror(error));
3832 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
3834 * The value strings in '*shash' are taken directly from values[], not copied,
3835 * so the caller should not modify or free them. */
3837 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3838 struct shash *shash)
3843 for (i = 0; i < n; i++) {
3844 shash_add(shash, keys[i], values[i]);
3848 struct iface_delete_queues_cbdata {
3849 struct netdev *netdev;
3850 const struct ovsdb_datum *queues;
3854 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
3856 union ovsdb_atom atom;
3858 atom.integer = target;
3859 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
3863 iface_delete_queues(unsigned int queue_id,
3864 const struct shash *details OVS_UNUSED, void *cbdata_)
3866 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3868 if (!queue_ids_include(cbdata->queues, queue_id)) {
3869 netdev_delete_queue(cbdata->netdev, queue_id);
3874 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3876 if (!qos || qos->type[0] == '\0') {
3877 netdev_set_qos(iface->netdev, NULL, NULL);
3879 struct iface_delete_queues_cbdata cbdata;
3880 struct shash details;
3883 /* Configure top-level Qos for 'iface'. */
3884 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3885 qos->n_other_config, &details);
3886 netdev_set_qos(iface->netdev, qos->type, &details);
3887 shash_destroy(&details);
3889 /* Deconfigure queues that were deleted. */
3890 cbdata.netdev = iface->netdev;
3891 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
3893 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3895 /* Configure queues for 'iface'. */
3896 for (i = 0; i < qos->n_queues; i++) {
3897 const struct ovsrec_queue *queue = qos->value_queues[i];
3898 unsigned int queue_id = qos->key_queues[i];
3900 shash_from_ovs_idl_map(queue->key_other_config,
3901 queue->value_other_config,
3902 queue->n_other_config, &details);
3903 netdev_set_queue(iface->netdev, queue_id, &details);
3904 shash_destroy(&details);
3909 /* Port mirroring. */
3911 static struct mirror *
3912 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
3916 for (i = 0; i < MAX_MIRRORS; i++) {
3917 struct mirror *m = br->mirrors[i];
3918 if (m && uuid_equals(uuid, &m->uuid)) {
3926 mirror_reconfigure(struct bridge *br)
3928 unsigned long *rspan_vlans;
3931 /* Get rid of deleted mirrors. */
3932 for (i = 0; i < MAX_MIRRORS; i++) {
3933 struct mirror *m = br->mirrors[i];
3935 const struct ovsdb_datum *mc;
3936 union ovsdb_atom atom;
3938 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
3939 atom.uuid = br->mirrors[i]->uuid;
3940 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
3946 /* Add new mirrors and reconfigure existing ones. */
3947 for (i = 0; i < br->cfg->n_mirrors; i++) {
3948 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3949 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
3951 mirror_reconfigure_one(m, cfg);
3953 mirror_create(br, cfg);
3957 /* Update port reserved status. */
3958 for (i = 0; i < br->n_ports; i++) {
3959 br->ports[i]->is_mirror_output_port = false;
3961 for (i = 0; i < MAX_MIRRORS; i++) {
3962 struct mirror *m = br->mirrors[i];
3963 if (m && m->out_port) {
3964 m->out_port->is_mirror_output_port = true;
3968 /* Update flooded vlans (for RSPAN). */
3970 if (br->cfg->n_flood_vlans) {
3971 rspan_vlans = bitmap_allocate(4096);
3973 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3974 int64_t vlan = br->cfg->flood_vlans[i];
3975 if (vlan >= 0 && vlan < 4096) {
3976 bitmap_set1(rspan_vlans, vlan);
3977 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3980 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3985 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3991 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
3996 for (i = 0; ; i++) {
3997 if (i >= MAX_MIRRORS) {
3998 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3999 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4002 if (!br->mirrors[i]) {
4007 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4010 br->mirrors[i] = m = xzalloc(sizeof *m);
4013 m->name = xstrdup(cfg->name);
4014 shash_init(&m->src_ports);
4015 shash_init(&m->dst_ports);
4021 mirror_reconfigure_one(m, cfg);
4025 mirror_destroy(struct mirror *m)
4028 struct bridge *br = m->bridge;
4031 for (i = 0; i < br->n_ports; i++) {
4032 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4033 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4036 shash_destroy(&m->src_ports);
4037 shash_destroy(&m->dst_ports);
4040 m->bridge->mirrors[m->idx] = NULL;
4049 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4050 struct shash *names)
4054 for (i = 0; i < n_ports; i++) {
4055 const char *name = ports[i]->name;
4056 if (port_lookup(m->bridge, name)) {
4057 shash_add_once(names, name, NULL);
4059 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4060 "port %s", m->bridge->name, m->name, name);
4066 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4072 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4074 for (i = 0; i < cfg->n_select_vlan; i++) {
4075 int64_t vlan = cfg->select_vlan[i];
4076 if (vlan < 0 || vlan > 4095) {
4077 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4078 m->bridge->name, m->name, vlan);
4080 (*vlans)[n_vlans++] = vlan;
4087 vlan_is_mirrored(const struct mirror *m, int vlan)
4091 for (i = 0; i < m->n_vlans; i++) {
4092 if (m->vlans[i] == vlan) {
4100 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4104 for (i = 0; i < m->n_vlans; i++) {
4105 if (port_trunks_vlan(p, m->vlans[i])) {
4113 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4115 struct shash src_ports, dst_ports;
4116 mirror_mask_t mirror_bit;
4117 struct port *out_port;
4124 if (strcmp(cfg->name, m->name)) {
4126 m->name = xstrdup(cfg->name);
4129 /* Get output port. */
4130 if (cfg->output_port) {
4131 out_port = port_lookup(m->bridge, cfg->output_port->name);
4133 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4134 m->bridge->name, m->name);
4140 if (cfg->output_vlan) {
4141 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4142 "output vlan; ignoring output vlan",
4143 m->bridge->name, m->name);
4145 } else if (cfg->output_vlan) {
4147 out_vlan = *cfg->output_vlan;
4149 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4150 m->bridge->name, m->name);
4155 shash_init(&src_ports);
4156 shash_init(&dst_ports);
4157 if (cfg->select_all) {
4158 for (i = 0; i < m->bridge->n_ports; i++) {
4159 const char *name = m->bridge->ports[i]->name;
4160 shash_add_once(&src_ports, name, NULL);
4161 shash_add_once(&dst_ports, name, NULL);
4166 /* Get ports, and drop duplicates and ports that don't exist. */
4167 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4169 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4172 /* Get all the vlans, and drop duplicate and invalid vlans. */
4173 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4176 /* Update mirror data. */
4177 if (!shash_equal_keys(&m->src_ports, &src_ports)
4178 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4179 || m->n_vlans != n_vlans
4180 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4181 || m->out_port != out_port
4182 || m->out_vlan != out_vlan) {
4183 bridge_flush(m->bridge);
4185 shash_swap(&m->src_ports, &src_ports);
4186 shash_swap(&m->dst_ports, &dst_ports);
4189 m->n_vlans = n_vlans;
4190 m->out_port = out_port;
4191 m->out_vlan = out_vlan;
4194 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4195 for (i = 0; i < m->bridge->n_ports; i++) {
4196 struct port *port = m->bridge->ports[i];
4198 if (shash_find(&m->src_ports, port->name)
4201 ? port_trunks_any_mirrored_vlan(m, port)
4202 : vlan_is_mirrored(m, port->vlan)))) {
4203 port->src_mirrors |= mirror_bit;
4205 port->src_mirrors &= ~mirror_bit;
4208 if (shash_find(&m->dst_ports, port->name)) {
4209 port->dst_mirrors |= mirror_bit;
4211 port->dst_mirrors &= ~mirror_bit;
4216 shash_destroy(&src_ports);
4217 shash_destroy(&dst_ports);