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"
42 #include "mac-learning.h"
45 #include "ofp-print.h"
47 #include "ofproto/netflow.h"
48 #include "ofproto/ofproto.h"
49 #include "ovsdb-data.h"
51 #include "poll-loop.h"
52 #include "port-array.h"
53 #include "proc-net-compat.h"
57 #include "socket-util.h"
58 #include "stream-ssl.h"
64 #include "vswitchd/vswitch-idl.h"
65 #include "xenserver.h"
68 #include "sflow_api.h"
70 VLOG_DEFINE_THIS_MODULE(bridge)
78 /* These members are always valid. */
79 struct port *port; /* Containing port. */
80 size_t port_ifidx; /* Index within containing port. */
81 char *name; /* Host network device name. */
82 tag_type tag; /* Tag associated with this interface. */
83 long long delay_expires; /* Time after which 'enabled' may change. */
85 /* These members are valid only after bridge_reconfigure() causes them to
87 int dp_ifidx; /* Index within kernel datapath. */
88 struct netdev *netdev; /* Network device. */
89 bool enabled; /* May be chosen for flows? */
90 const struct ovsrec_interface *cfg;
93 #define BOND_MASK 0xff
95 int iface_idx; /* Index of assigned iface, or -1 if none. */
96 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
97 tag_type iface_tag; /* Tag associated with iface_idx. */
100 #define MAX_MIRRORS 32
101 typedef uint32_t mirror_mask_t;
102 #define MIRROR_MASK_C(X) UINT32_C(X)
103 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
105 struct bridge *bridge;
108 struct uuid uuid; /* UUID of this "mirror" record in database. */
110 /* Selection criteria. */
111 struct shash src_ports; /* Name is port name; data is always NULL. */
112 struct shash dst_ports; /* Name is port name; data is always NULL. */
117 struct port *out_port;
121 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
123 struct bridge *bridge;
125 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
126 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
127 * NULL if all VLANs are trunked. */
128 const struct ovsrec_port *cfg;
131 /* An ordinary bridge port has 1 interface.
132 * A bridge port for bonding has at least 2 interfaces. */
133 struct iface **ifaces;
134 size_t n_ifaces, allocated_ifaces;
137 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
138 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
139 tag_type active_iface_tag; /* Tag for bcast flows. */
140 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
141 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
142 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
143 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
144 long long int bond_next_fake_iface_update; /* Time of next update. */
145 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
146 long long int bond_next_rebalance; /* Next rebalancing time. */
148 /* Port mirroring info. */
149 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
150 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
151 bool is_mirror_output_port; /* Does port mirroring send frames here? */
154 #define DP_MAX_PORTS 255
156 struct list node; /* Node in global list of bridges. */
157 char *name; /* User-specified arbitrary name. */
158 struct mac_learning *ml; /* MAC learning table. */
159 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
160 const struct ovsrec_bridge *cfg;
162 /* OpenFlow switch processing. */
163 struct ofproto *ofproto; /* OpenFlow switch. */
165 /* Kernel datapath information. */
166 struct dpif *dpif; /* Datapath. */
167 struct port_array ifaces; /* Indexed by kernel datapath port number. */
171 size_t n_ports, allocated_ports;
172 struct shash iface_by_name; /* "struct iface"s indexed by name. */
173 struct shash port_by_name; /* "struct port"s indexed by name. */
176 bool has_bonded_ports;
181 /* Port mirroring. */
182 struct mirror *mirrors[MAX_MIRRORS];
185 /* List of all bridges. */
186 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
188 /* OVSDB IDL used to obtain configuration. */
189 static struct ovsdb_idl *idl;
191 /* Each time this timer expires, the bridge fetches statistics for every
192 * interface and pushes them into the database. */
193 #define IFACE_STATS_INTERVAL (5 * 1000) /* In milliseconds. */
194 static long long int iface_stats_timer = LLONG_MIN;
196 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
197 static void bridge_destroy(struct bridge *);
198 static struct bridge *bridge_lookup(const char *name);
199 static unixctl_cb_func bridge_unixctl_dump_flows;
200 static unixctl_cb_func bridge_unixctl_reconnect;
201 static int bridge_run_one(struct bridge *);
202 static size_t bridge_get_controllers(const struct bridge *br,
203 struct ovsrec_controller ***controllersp);
204 static void bridge_reconfigure_one(struct bridge *);
205 static void bridge_reconfigure_remotes(struct bridge *,
206 const struct sockaddr_in *managers,
208 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
209 static void bridge_fetch_dp_ifaces(struct bridge *);
210 static void bridge_flush(struct bridge *);
211 static void bridge_pick_local_hw_addr(struct bridge *,
212 uint8_t ea[ETH_ADDR_LEN],
213 struct iface **hw_addr_iface);
214 static uint64_t bridge_pick_datapath_id(struct bridge *,
215 const uint8_t bridge_ea[ETH_ADDR_LEN],
216 struct iface *hw_addr_iface);
217 static struct iface *bridge_get_local_iface(struct bridge *);
218 static uint64_t dpid_from_hash(const void *, size_t nbytes);
220 static unixctl_cb_func bridge_unixctl_fdb_show;
222 static void bond_init(void);
223 static void bond_run(struct bridge *);
224 static void bond_wait(struct bridge *);
225 static void bond_rebalance_port(struct port *);
226 static void bond_send_learning_packets(struct port *);
227 static void bond_enable_slave(struct iface *iface, bool enable);
229 static struct port *port_create(struct bridge *, const char *name);
230 static void port_reconfigure(struct port *, const struct ovsrec_port *);
231 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
232 static void port_destroy(struct port *);
233 static struct port *port_lookup(const struct bridge *, const char *name);
234 static struct iface *port_lookup_iface(const struct port *, const char *name);
235 static struct port *port_from_dp_ifidx(const struct bridge *,
237 static void port_update_bond_compat(struct port *);
238 static void port_update_vlan_compat(struct port *);
239 static void port_update_bonding(struct port *);
241 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
242 static void mirror_destroy(struct mirror *);
243 static void mirror_reconfigure(struct bridge *);
244 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
245 static bool vlan_is_mirrored(const struct mirror *, int vlan);
247 static struct iface *iface_create(struct port *port,
248 const struct ovsrec_interface *if_cfg);
249 static void iface_destroy(struct iface *);
250 static struct iface *iface_lookup(const struct bridge *, const char *name);
251 static struct iface *iface_from_dp_ifidx(const struct bridge *,
253 static bool iface_is_internal(const struct bridge *, const char *name);
254 static void iface_set_mac(struct iface *);
255 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
257 /* Hooks into ofproto processing. */
258 static struct ofhooks bridge_ofhooks;
260 /* Public functions. */
262 /* Initializes the bridge module, configuring it to obtain its configuration
263 * from an OVSDB server accessed over 'remote', which should be a string in a
264 * form acceptable to ovsdb_idl_create(). */
266 bridge_init(const char *remote)
268 /* Create connection to database. */
269 idl = ovsdb_idl_create(remote, &ovsrec_idl_class);
271 ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_cur_cfg);
272 ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_statistics);
273 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
275 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
277 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
278 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
280 ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_ofport);
281 ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_statistics);
282 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
284 /* Register unixctl commands. */
285 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
286 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
288 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
293 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
294 * but for which the ovs-vswitchd configuration 'cfg' is required. */
296 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
298 static bool already_configured_once;
299 struct svec bridge_names;
300 struct svec dpif_names, dpif_types;
303 /* Only do this once per ovs-vswitchd run. */
304 if (already_configured_once) {
307 already_configured_once = true;
309 iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
311 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
312 svec_init(&bridge_names);
313 for (i = 0; i < cfg->n_bridges; i++) {
314 svec_add(&bridge_names, cfg->bridges[i]->name);
316 svec_sort(&bridge_names);
318 /* Iterate over all system dpifs and delete any of them that do not appear
320 svec_init(&dpif_names);
321 svec_init(&dpif_types);
322 dp_enumerate_types(&dpif_types);
323 for (i = 0; i < dpif_types.n; i++) {
328 dp_enumerate_names(dpif_types.names[i], &dpif_names);
330 /* For each dpif... */
331 for (j = 0; j < dpif_names.n; j++) {
332 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
334 struct svec all_names;
337 /* ...check whether any of its names is in 'bridge_names'. */
338 svec_init(&all_names);
339 dpif_get_all_names(dpif, &all_names);
340 for (k = 0; k < all_names.n; k++) {
341 if (svec_contains(&bridge_names, all_names.names[k])) {
346 /* No. Delete the dpif. */
350 svec_destroy(&all_names);
355 svec_destroy(&bridge_names);
356 svec_destroy(&dpif_names);
357 svec_destroy(&dpif_types);
360 /* Attempt to create the network device 'iface_name' through the netdev
363 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
366 struct shash options;
370 shash_init(&options);
371 for (i = 0; i < iface_cfg->n_options; i++) {
372 shash_add(&options, iface_cfg->key_options[i],
373 xstrdup(iface_cfg->value_options[i]));
377 struct netdev_options netdev_options;
379 memset(&netdev_options, 0, sizeof netdev_options);
380 netdev_options.name = iface_cfg->name;
381 if (!strcmp(iface_cfg->type, "internal")) {
382 /* An "internal" config type maps to a netdev "system" type. */
383 netdev_options.type = "system";
385 netdev_options.type = iface_cfg->type;
387 netdev_options.args = &options;
388 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
390 error = netdev_open(&netdev_options, &iface->netdev);
393 netdev_get_carrier(iface->netdev, &iface->enabled);
395 } else if (iface->netdev) {
396 const char *netdev_type = netdev_get_type(iface->netdev);
397 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
398 ? iface_cfg->type : NULL;
400 /* An "internal" config type maps to a netdev "system" type. */
401 if (iface_type && !strcmp(iface_type, "internal")) {
402 iface_type = "system";
405 if (!iface_type || !strcmp(netdev_type, iface_type)) {
406 error = netdev_reconfigure(iface->netdev, &options);
408 VLOG_WARN("%s: attempting change device type from %s to %s",
409 iface_cfg->name, netdev_type, iface_type);
413 shash_destroy_free_data(&options);
419 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
421 return set_up_iface(iface_cfg, iface, false);
425 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
426 void *aux OVS_UNUSED)
428 if (!iface->netdev) {
429 int error = set_up_iface(iface->cfg, iface, true);
431 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
441 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
442 void *aux OVS_UNUSED)
444 if (iface->dp_ifidx >= 0) {
445 VLOG_DBG("%s has interface %s on port %d",
447 iface->name, iface->dp_ifidx);
450 VLOG_ERR("%s interface not in %s, dropping",
451 iface->name, dpif_name(br->dpif));
457 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
458 void *aux OVS_UNUSED)
460 /* Set policing attributes. */
461 netdev_set_policing(iface->netdev,
462 iface->cfg->ingress_policing_rate,
463 iface->cfg->ingress_policing_burst);
465 /* Set MAC address of internal interfaces other than the local
467 if (iface->dp_ifidx != ODPP_LOCAL
468 && iface_is_internal(br, iface->name)) {
469 iface_set_mac(iface);
475 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
476 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
477 * deletes from 'br' any ports that no longer have any interfaces. */
479 iterate_and_prune_ifaces(struct bridge *br,
480 bool (*cb)(struct bridge *, struct iface *,
486 for (i = 0; i < br->n_ports; ) {
487 struct port *port = br->ports[i];
488 for (j = 0; j < port->n_ifaces; ) {
489 struct iface *iface = port->ifaces[j];
490 if (cb(br, iface, aux)) {
493 iface_destroy(iface);
497 if (port->n_ifaces) {
500 VLOG_ERR("%s port has no interfaces, dropping", port->name);
506 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
507 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
508 * responsible for freeing '*managersp' (with free()).
510 * You may be asking yourself "why does ovs-vswitchd care?", because
511 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
512 * should not be and in fact is not directly involved in that. But
513 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
514 * it has to tell in-band control where the managers are to enable that.
517 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
518 struct sockaddr_in **managersp, size_t *n_managersp)
520 struct sockaddr_in *managers = NULL;
521 size_t n_managers = 0;
523 if (ovs_cfg->n_managers > 0) {
526 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
527 for (i = 0; i < ovs_cfg->n_managers; i++) {
528 const char *name = ovs_cfg->managers[i];
529 struct sockaddr_in *sin = &managers[i];
531 if ((!strncmp(name, "tcp:", 4)
532 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
533 (!strncmp(name, "ssl:", 4)
534 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
540 *managersp = managers;
541 *n_managersp = n_managers;
545 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
547 struct shash old_br, new_br;
548 struct shash_node *node;
549 struct bridge *br, *next;
550 struct sockaddr_in *managers;
553 int sflow_bridge_number;
555 COVERAGE_INC(bridge_reconfigure);
557 collect_managers(ovs_cfg, &managers, &n_managers);
559 /* Collect old and new bridges. */
562 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
563 shash_add(&old_br, br->name, br);
565 for (i = 0; i < ovs_cfg->n_bridges; i++) {
566 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
567 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
568 VLOG_WARN("more than one bridge named %s", br_cfg->name);
572 /* Get rid of deleted bridges and add new bridges. */
573 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
574 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
581 SHASH_FOR_EACH (node, &new_br) {
582 const char *br_name = node->name;
583 const struct ovsrec_bridge *br_cfg = node->data;
584 br = shash_find_data(&old_br, br_name);
586 /* If the bridge datapath type has changed, we need to tear it
587 * down and recreate. */
588 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
590 bridge_create(br_cfg);
593 bridge_create(br_cfg);
596 shash_destroy(&old_br);
597 shash_destroy(&new_br);
599 /* Reconfigure all bridges. */
600 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
601 bridge_reconfigure_one(br);
604 /* Add and delete ports on all datapaths.
606 * The kernel will reject any attempt to add a given port to a datapath if
607 * that port already belongs to a different datapath, so we must do all
608 * port deletions before any port additions. */
609 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
610 struct odp_port *dpif_ports;
612 struct shash want_ifaces;
614 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
615 bridge_get_all_ifaces(br, &want_ifaces);
616 for (i = 0; i < n_dpif_ports; i++) {
617 const struct odp_port *p = &dpif_ports[i];
618 if (!shash_find(&want_ifaces, p->devname)
619 && strcmp(p->devname, br->name)) {
620 int retval = dpif_port_del(br->dpif, p->port);
622 VLOG_ERR("failed to remove %s interface from %s: %s",
623 p->devname, dpif_name(br->dpif),
628 shash_destroy(&want_ifaces);
631 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
632 struct odp_port *dpif_ports;
634 struct shash cur_ifaces, want_ifaces;
636 /* Get the set of interfaces currently in this datapath. */
637 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
638 shash_init(&cur_ifaces);
639 for (i = 0; i < n_dpif_ports; i++) {
640 const char *name = dpif_ports[i].devname;
641 shash_add_once(&cur_ifaces, name, NULL);
645 /* Get the set of interfaces we want on this datapath. */
646 bridge_get_all_ifaces(br, &want_ifaces);
648 SHASH_FOR_EACH (node, &want_ifaces) {
649 const char *if_name = node->name;
650 struct iface *iface = node->data;
652 if (shash_find(&cur_ifaces, if_name)) {
653 /* Already exists, just reconfigure it. */
655 reconfigure_iface(iface->cfg, iface);
658 /* Need to add to datapath. */
662 /* Add to datapath. */
663 internal = iface_is_internal(br, if_name);
664 error = dpif_port_add(br->dpif, if_name,
665 internal ? ODP_PORT_INTERNAL : 0, NULL);
666 if (error == EFBIG) {
667 VLOG_ERR("ran out of valid port numbers on %s",
668 dpif_name(br->dpif));
671 VLOG_ERR("failed to add %s interface to %s: %s",
672 if_name, dpif_name(br->dpif), strerror(error));
676 shash_destroy(&cur_ifaces);
677 shash_destroy(&want_ifaces);
679 sflow_bridge_number = 0;
680 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
683 struct iface *local_iface;
684 struct iface *hw_addr_iface;
687 bridge_fetch_dp_ifaces(br);
689 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
690 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
692 /* Pick local port hardware address, datapath ID. */
693 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
694 local_iface = bridge_get_local_iface(br);
696 int error = netdev_set_etheraddr(local_iface->netdev, ea);
698 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
699 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
700 "Ethernet address: %s",
701 br->name, strerror(error));
705 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
706 ofproto_set_datapath_id(br->ofproto, dpid);
708 dpid_string = xasprintf("%016"PRIx64, dpid);
709 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
712 /* Set NetFlow configuration on this bridge. */
713 if (br->cfg->netflow) {
714 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
715 struct netflow_options opts;
717 memset(&opts, 0, sizeof opts);
719 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
720 if (nf_cfg->engine_type) {
721 opts.engine_type = *nf_cfg->engine_type;
723 if (nf_cfg->engine_id) {
724 opts.engine_id = *nf_cfg->engine_id;
727 opts.active_timeout = nf_cfg->active_timeout;
728 if (!opts.active_timeout) {
729 opts.active_timeout = -1;
730 } else if (opts.active_timeout < 0) {
731 VLOG_WARN("bridge %s: active timeout interval set to negative "
732 "value, using default instead (%d seconds)", br->name,
733 NF_ACTIVE_TIMEOUT_DEFAULT);
734 opts.active_timeout = -1;
737 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
738 if (opts.add_id_to_iface) {
739 if (opts.engine_id > 0x7f) {
740 VLOG_WARN("bridge %s: netflow port mangling may conflict "
741 "with another vswitch, choose an engine id less "
742 "than 128", br->name);
744 if (br->n_ports > 508) {
745 VLOG_WARN("bridge %s: netflow port mangling will conflict "
746 "with another port when more than 508 ports are "
751 opts.collectors.n = nf_cfg->n_targets;
752 opts.collectors.names = nf_cfg->targets;
753 if (ofproto_set_netflow(br->ofproto, &opts)) {
754 VLOG_ERR("bridge %s: problem setting netflow collectors",
758 ofproto_set_netflow(br->ofproto, NULL);
761 /* Set sFlow configuration on this bridge. */
762 if (br->cfg->sflow) {
763 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
764 struct ovsrec_controller **controllers;
765 struct ofproto_sflow_options oso;
766 size_t n_controllers;
768 memset(&oso, 0, sizeof oso);
770 oso.targets.n = sflow_cfg->n_targets;
771 oso.targets.names = sflow_cfg->targets;
773 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
774 if (sflow_cfg->sampling) {
775 oso.sampling_rate = *sflow_cfg->sampling;
778 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
779 if (sflow_cfg->polling) {
780 oso.polling_interval = *sflow_cfg->polling;
783 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
784 if (sflow_cfg->header) {
785 oso.header_len = *sflow_cfg->header;
788 oso.sub_id = sflow_bridge_number++;
789 oso.agent_device = sflow_cfg->agent;
791 oso.control_ip = NULL;
792 n_controllers = bridge_get_controllers(br, &controllers);
793 for (i = 0; i < n_controllers; i++) {
794 if (controllers[i]->local_ip) {
795 oso.control_ip = controllers[i]->local_ip;
799 ofproto_set_sflow(br->ofproto, &oso);
801 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
803 ofproto_set_sflow(br->ofproto, NULL);
806 /* Update the controller and related settings. It would be more
807 * straightforward to call this from bridge_reconfigure_one(), but we
808 * can't do it there for two reasons. First, and most importantly, at
809 * that point we don't know the dp_ifidx of any interfaces that have
810 * been added to the bridge (because we haven't actually added them to
811 * the datapath). Second, at that point we haven't set the datapath ID
812 * yet; when a controller is configured, resetting the datapath ID will
813 * immediately disconnect from the controller, so it's better to set
814 * the datapath ID before the controller. */
815 bridge_reconfigure_remotes(br, managers, n_managers);
817 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
818 for (i = 0; i < br->n_ports; i++) {
819 struct port *port = br->ports[i];
822 port_update_vlan_compat(port);
823 port_update_bonding(port);
825 for (j = 0; j < port->n_ifaces; j++) {
826 iface_update_qos(port->ifaces[j], port->cfg->qos);
830 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
831 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
838 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
839 const struct ovsdb_idl_column *column,
842 const struct ovsdb_datum *datum;
843 union ovsdb_atom atom;
846 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
847 atom.string = (char *) key;
848 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
849 return idx == UINT_MAX ? NULL : datum->values[idx].string;
853 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
855 return get_ovsrec_key_value(&br_cfg->header_,
856 &ovsrec_bridge_col_other_config, key);
860 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
861 struct iface **hw_addr_iface)
867 *hw_addr_iface = NULL;
869 /* Did the user request a particular MAC? */
870 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
871 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
872 if (eth_addr_is_multicast(ea)) {
873 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
874 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
875 } else if (eth_addr_is_zero(ea)) {
876 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
882 /* Otherwise choose the minimum non-local MAC address among all of the
884 memset(ea, 0xff, sizeof ea);
885 for (i = 0; i < br->n_ports; i++) {
886 struct port *port = br->ports[i];
887 uint8_t iface_ea[ETH_ADDR_LEN];
890 /* Mirror output ports don't participate. */
891 if (port->is_mirror_output_port) {
895 /* Choose the MAC address to represent the port. */
896 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
897 /* Find the interface with this Ethernet address (if any) so that
898 * we can provide the correct devname to the caller. */
900 for (j = 0; j < port->n_ifaces; j++) {
901 struct iface *candidate = port->ifaces[j];
902 uint8_t candidate_ea[ETH_ADDR_LEN];
903 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
904 && eth_addr_equals(iface_ea, candidate_ea)) {
909 /* Choose the interface whose MAC address will represent the port.
910 * The Linux kernel bonding code always chooses the MAC address of
911 * the first slave added to a bond, and the Fedora networking
912 * scripts always add slaves to a bond in alphabetical order, so
913 * for compatibility we choose the interface with the name that is
914 * first in alphabetical order. */
915 iface = port->ifaces[0];
916 for (j = 1; j < port->n_ifaces; j++) {
917 struct iface *candidate = port->ifaces[j];
918 if (strcmp(candidate->name, iface->name) < 0) {
923 /* The local port doesn't count (since we're trying to choose its
924 * MAC address anyway). */
925 if (iface->dp_ifidx == ODPP_LOCAL) {
930 error = netdev_get_etheraddr(iface->netdev, iface_ea);
932 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
933 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
934 iface->name, strerror(error));
939 /* Compare against our current choice. */
940 if (!eth_addr_is_multicast(iface_ea) &&
941 !eth_addr_is_local(iface_ea) &&
942 !eth_addr_is_reserved(iface_ea) &&
943 !eth_addr_is_zero(iface_ea) &&
944 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
946 memcpy(ea, iface_ea, ETH_ADDR_LEN);
947 *hw_addr_iface = iface;
950 if (eth_addr_is_multicast(ea)) {
951 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
952 *hw_addr_iface = NULL;
953 VLOG_WARN("bridge %s: using default bridge Ethernet "
954 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
956 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
957 br->name, ETH_ADDR_ARGS(ea));
961 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
962 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
963 * an interface on 'br', then that interface must be passed in as
964 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
965 * 'hw_addr_iface' must be passed in as a null pointer. */
967 bridge_pick_datapath_id(struct bridge *br,
968 const uint8_t bridge_ea[ETH_ADDR_LEN],
969 struct iface *hw_addr_iface)
972 * The procedure for choosing a bridge MAC address will, in the most
973 * ordinary case, also choose a unique MAC that we can use as a datapath
974 * ID. In some special cases, though, multiple bridges will end up with
975 * the same MAC address. This is OK for the bridges, but it will confuse
976 * the OpenFlow controller, because each datapath needs a unique datapath
979 * Datapath IDs must be unique. It is also very desirable that they be
980 * stable from one run to the next, so that policy set on a datapath
983 const char *datapath_id;
986 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
987 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
993 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
995 * A bridge whose MAC address is taken from a VLAN network device
996 * (that is, a network device created with vconfig(8) or similar
997 * tool) will have the same MAC address as a bridge on the VLAN
998 * device's physical network device.
1000 * Handle this case by hashing the physical network device MAC
1001 * along with the VLAN identifier.
1003 uint8_t buf[ETH_ADDR_LEN + 2];
1004 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1005 buf[ETH_ADDR_LEN] = vlan >> 8;
1006 buf[ETH_ADDR_LEN + 1] = vlan;
1007 return dpid_from_hash(buf, sizeof buf);
1010 * Assume that this bridge's MAC address is unique, since it
1011 * doesn't fit any of the cases we handle specially.
1016 * A purely internal bridge, that is, one that has no non-virtual
1017 * network devices on it at all, is more difficult because it has no
1018 * natural unique identifier at all.
1020 * When the host is a XenServer, we handle this case by hashing the
1021 * host's UUID with the name of the bridge. Names of bridges are
1022 * persistent across XenServer reboots, although they can be reused if
1023 * an internal network is destroyed and then a new one is later
1024 * created, so this is fairly effective.
1026 * When the host is not a XenServer, we punt by using a random MAC
1027 * address on each run.
1029 const char *host_uuid = xenserver_get_host_uuid();
1031 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1032 dpid = dpid_from_hash(combined, strlen(combined));
1038 return eth_addr_to_uint64(bridge_ea);
1042 dpid_from_hash(const void *data, size_t n)
1044 uint8_t hash[SHA1_DIGEST_SIZE];
1046 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1047 sha1_bytes(data, n, hash);
1048 eth_addr_mark_random(hash);
1049 return eth_addr_to_uint64(hash);
1053 iface_refresh_stats(struct iface *iface)
1059 static const struct iface_stat iface_stats[] = {
1060 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1061 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1062 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1063 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1064 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1065 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1066 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1067 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1068 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1069 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1070 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1071 { "collisions", offsetof(struct netdev_stats, collisions) },
1073 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1074 const struct iface_stat *s;
1076 char *keys[N_STATS];
1077 int64_t values[N_STATS];
1080 struct netdev_stats stats;
1082 /* Intentionally ignore return value, since errors will set 'stats' to
1083 * all-1s, and we will deal with that correctly below. */
1084 netdev_get_stats(iface->netdev, &stats);
1087 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1088 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1089 if (value != UINT64_MAX) {
1096 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1102 const struct ovsrec_open_vswitch *cfg;
1104 bool datapath_destroyed;
1105 bool database_changed;
1108 /* Let each bridge do the work that it needs to do. */
1109 datapath_destroyed = false;
1110 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1111 int error = bridge_run_one(br);
1113 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1114 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1115 "forcing reconfiguration", br->name);
1116 datapath_destroyed = true;
1120 /* (Re)configure if necessary. */
1121 database_changed = ovsdb_idl_run(idl);
1122 cfg = ovsrec_open_vswitch_first(idl);
1123 if (database_changed || datapath_destroyed) {
1125 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1127 bridge_configure_once(cfg);
1128 bridge_reconfigure(cfg);
1130 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1131 ovsdb_idl_txn_commit(txn);
1132 ovsdb_idl_txn_destroy(txn); /* XXX */
1134 /* We still need to reconfigure to avoid dangling pointers to
1135 * now-destroyed ovsrec structures inside bridge data. */
1136 static const struct ovsrec_open_vswitch null_cfg;
1138 bridge_reconfigure(&null_cfg);
1143 /* Re-configure SSL. We do this on every trip through the main loop,
1144 * instead of just when the database changes, because the contents of the
1145 * key and certificate files can change without the database changing. */
1146 if (cfg && cfg->ssl) {
1147 const struct ovsrec_ssl *ssl = cfg->ssl;
1149 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1150 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1154 /* Refresh interface stats if necessary. */
1155 if (time_msec() >= iface_stats_timer) {
1156 struct ovsdb_idl_txn *txn;
1158 txn = ovsdb_idl_txn_create(idl);
1159 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1162 for (i = 0; i < br->n_ports; i++) {
1163 struct port *port = br->ports[i];
1166 for (j = 0; j < port->n_ifaces; j++) {
1167 struct iface *iface = port->ifaces[j];
1168 iface_refresh_stats(iface);
1172 ovsdb_idl_txn_commit(txn);
1173 ovsdb_idl_txn_destroy(txn); /* XXX */
1175 iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
1184 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1185 ofproto_wait(br->ofproto);
1186 if (ofproto_has_primary_controller(br->ofproto)) {
1190 mac_learning_wait(br->ml);
1193 ovsdb_idl_wait(idl);
1194 poll_timer_wait_until(iface_stats_timer);
1197 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1198 * configuration changes. */
1200 bridge_flush(struct bridge *br)
1202 COVERAGE_INC(bridge_flush);
1204 mac_learning_flush(br->ml);
1207 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1208 * such interface. */
1209 static struct iface *
1210 bridge_get_local_iface(struct bridge *br)
1214 for (i = 0; i < br->n_ports; i++) {
1215 struct port *port = br->ports[i];
1216 for (j = 0; j < port->n_ifaces; j++) {
1217 struct iface *iface = port->ifaces[j];
1218 if (iface->dp_ifidx == ODPP_LOCAL) {
1227 /* Bridge unixctl user interface functions. */
1229 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1230 const char *args, void *aux OVS_UNUSED)
1232 struct ds ds = DS_EMPTY_INITIALIZER;
1233 const struct bridge *br;
1234 const struct mac_entry *e;
1236 br = bridge_lookup(args);
1238 unixctl_command_reply(conn, 501, "no such bridge");
1242 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1243 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1244 if (e->port < 0 || e->port >= br->n_ports) {
1247 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1248 br->ports[e->port]->ifaces[0]->dp_ifidx,
1249 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1251 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1255 /* Bridge reconfiguration functions. */
1256 static struct bridge *
1257 bridge_create(const struct ovsrec_bridge *br_cfg)
1262 assert(!bridge_lookup(br_cfg->name));
1263 br = xzalloc(sizeof *br);
1265 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1271 dpif_flow_flush(br->dpif);
1273 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1276 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1278 dpif_delete(br->dpif);
1279 dpif_close(br->dpif);
1284 br->name = xstrdup(br_cfg->name);
1286 br->ml = mac_learning_create();
1287 eth_addr_nicira_random(br->default_ea);
1289 port_array_init(&br->ifaces);
1291 shash_init(&br->port_by_name);
1292 shash_init(&br->iface_by_name);
1296 list_push_back(&all_bridges, &br->node);
1298 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1304 bridge_destroy(struct bridge *br)
1309 while (br->n_ports > 0) {
1310 port_destroy(br->ports[br->n_ports - 1]);
1312 list_remove(&br->node);
1313 error = dpif_delete(br->dpif);
1314 if (error && error != ENOENT) {
1315 VLOG_ERR("failed to delete %s: %s",
1316 dpif_name(br->dpif), strerror(error));
1318 dpif_close(br->dpif);
1319 ofproto_destroy(br->ofproto);
1320 mac_learning_destroy(br->ml);
1321 port_array_destroy(&br->ifaces);
1322 shash_destroy(&br->port_by_name);
1323 shash_destroy(&br->iface_by_name);
1330 static struct bridge *
1331 bridge_lookup(const char *name)
1335 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1336 if (!strcmp(br->name, name)) {
1343 /* Handle requests for a listing of all flows known by the OpenFlow
1344 * stack, including those normally hidden. */
1346 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1347 const char *args, void *aux OVS_UNUSED)
1352 br = bridge_lookup(args);
1354 unixctl_command_reply(conn, 501, "Unknown bridge");
1359 ofproto_get_all_flows(br->ofproto, &results);
1361 unixctl_command_reply(conn, 200, ds_cstr(&results));
1362 ds_destroy(&results);
1365 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1366 * connections and reconnect. If BRIDGE is not specified, then all bridges
1367 * drop their controller connections and reconnect. */
1369 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1370 const char *args, void *aux OVS_UNUSED)
1373 if (args[0] != '\0') {
1374 br = bridge_lookup(args);
1376 unixctl_command_reply(conn, 501, "Unknown bridge");
1379 ofproto_reconnect_controllers(br->ofproto);
1381 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1382 ofproto_reconnect_controllers(br->ofproto);
1385 unixctl_command_reply(conn, 200, NULL);
1389 bridge_run_one(struct bridge *br)
1393 error = ofproto_run1(br->ofproto);
1398 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1401 error = ofproto_run2(br->ofproto, br->flush);
1408 bridge_get_controllers(const struct bridge *br,
1409 struct ovsrec_controller ***controllersp)
1411 struct ovsrec_controller **controllers;
1412 size_t n_controllers;
1414 controllers = br->cfg->controller;
1415 n_controllers = br->cfg->n_controller;
1417 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1423 *controllersp = controllers;
1425 return n_controllers;
1429 bridge_reconfigure_one(struct bridge *br)
1431 struct shash old_ports, new_ports;
1432 struct svec snoops, old_snoops;
1433 struct shash_node *node;
1434 enum ofproto_fail_mode fail_mode;
1437 /* Collect old ports. */
1438 shash_init(&old_ports);
1439 for (i = 0; i < br->n_ports; i++) {
1440 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1443 /* Collect new ports. */
1444 shash_init(&new_ports);
1445 for (i = 0; i < br->cfg->n_ports; i++) {
1446 const char *name = br->cfg->ports[i]->name;
1447 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1448 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1453 /* If we have a controller, then we need a local port. Complain if the
1454 * user didn't specify one.
1456 * XXX perhaps we should synthesize a port ourselves in this case. */
1457 if (bridge_get_controllers(br, NULL)) {
1458 char local_name[IF_NAMESIZE];
1461 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1462 local_name, sizeof local_name);
1463 if (!error && !shash_find(&new_ports, local_name)) {
1464 VLOG_WARN("bridge %s: controller specified but no local port "
1465 "(port named %s) defined",
1466 br->name, local_name);
1470 /* Get rid of deleted ports.
1471 * Get rid of deleted interfaces on ports that still exist. */
1472 SHASH_FOR_EACH (node, &old_ports) {
1473 struct port *port = node->data;
1474 const struct ovsrec_port *port_cfg;
1476 port_cfg = shash_find_data(&new_ports, node->name);
1480 port_del_ifaces(port, port_cfg);
1484 /* Create new ports.
1485 * Add new interfaces to existing ports.
1486 * Reconfigure existing ports. */
1487 SHASH_FOR_EACH (node, &new_ports) {
1488 struct port *port = shash_find_data(&old_ports, node->name);
1490 port = port_create(br, node->name);
1493 port_reconfigure(port, node->data);
1494 if (!port->n_ifaces) {
1495 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1496 br->name, port->name);
1500 shash_destroy(&old_ports);
1501 shash_destroy(&new_ports);
1503 /* Set the fail-mode */
1504 fail_mode = !br->cfg->fail_mode
1505 || !strcmp(br->cfg->fail_mode, "standalone")
1506 ? OFPROTO_FAIL_STANDALONE
1507 : OFPROTO_FAIL_SECURE;
1508 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1509 && !ofproto_has_primary_controller(br->ofproto)) {
1510 ofproto_flush_flows(br->ofproto);
1512 ofproto_set_fail_mode(br->ofproto, fail_mode);
1514 /* Delete all flows if we're switching from connected to standalone or vice
1515 * versa. (XXX Should we delete all flows if we are switching from one
1516 * controller to another?) */
1518 /* Configure OpenFlow controller connection snooping. */
1520 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1521 ovs_rundir, br->name));
1522 svec_init(&old_snoops);
1523 ofproto_get_snoops(br->ofproto, &old_snoops);
1524 if (!svec_equal(&snoops, &old_snoops)) {
1525 ofproto_set_snoops(br->ofproto, &snoops);
1527 svec_destroy(&snoops);
1528 svec_destroy(&old_snoops);
1530 mirror_reconfigure(br);
1533 /* Initializes 'oc' appropriately as a management service controller for
1536 * The caller must free oc->target when it is no longer needed. */
1538 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1539 struct ofproto_controller *oc)
1541 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir, br->name);
1542 oc->max_backoff = 0;
1543 oc->probe_interval = 60;
1544 oc->band = OFPROTO_OUT_OF_BAND;
1545 oc->accept_re = NULL;
1546 oc->update_resolv_conf = false;
1548 oc->burst_limit = 0;
1551 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1553 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1554 struct ofproto_controller *oc)
1556 oc->target = c->target;
1557 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1558 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1559 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1560 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1561 oc->accept_re = c->discover_accept_regex;
1562 oc->update_resolv_conf = c->discover_update_resolv_conf;
1563 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1564 oc->burst_limit = (c->controller_burst_limit
1565 ? *c->controller_burst_limit : 0);
1568 /* Configures the IP stack for 'br''s local interface properly according to the
1569 * configuration in 'c'. */
1571 bridge_configure_local_iface_netdev(struct bridge *br,
1572 struct ovsrec_controller *c)
1574 struct netdev *netdev;
1575 struct in_addr mask, gateway;
1577 struct iface *local_iface;
1580 /* Controller discovery does its own TCP/IP configuration later. */
1581 if (strcmp(c->target, "discover")) {
1585 /* If there's no local interface or no IP address, give up. */
1586 local_iface = bridge_get_local_iface(br);
1587 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1591 /* Bring up the local interface. */
1592 netdev = local_iface->netdev;
1593 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1595 /* Configure the IP address and netmask. */
1596 if (!c->local_netmask
1597 || !inet_aton(c->local_netmask, &mask)
1599 mask.s_addr = guess_netmask(ip.s_addr);
1601 if (!netdev_set_in4(netdev, ip, mask)) {
1602 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1603 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1606 /* Configure the default gateway. */
1607 if (c->local_gateway
1608 && inet_aton(c->local_gateway, &gateway)
1609 && gateway.s_addr) {
1610 if (!netdev_add_router(netdev, gateway)) {
1611 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1612 br->name, IP_ARGS(&gateway.s_addr));
1618 bridge_reconfigure_remotes(struct bridge *br,
1619 const struct sockaddr_in *managers,
1622 struct ovsrec_controller **controllers;
1623 size_t n_controllers;
1626 struct ofproto_controller *ocs;
1630 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1631 had_primary = ofproto_has_primary_controller(br->ofproto);
1633 n_controllers = bridge_get_controllers(br, &controllers);
1635 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1638 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1639 for (i = 0; i < n_controllers; i++) {
1640 struct ovsrec_controller *c = controllers[i];
1642 if (!strncmp(c->target, "punix:", 6)
1643 || !strncmp(c->target, "unix:", 5)) {
1644 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1646 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1647 * domain sockets and overwriting arbitrary local files. */
1648 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1649 "\"%s\" due to possibility for remote exploit",
1650 dpif_name(br->dpif), c->target);
1654 bridge_configure_local_iface_netdev(br, c);
1655 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs++]);
1658 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1659 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1662 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1663 ofproto_flush_flows(br->ofproto);
1666 /* If there are no controllers and the bridge is in standalone
1667 * mode, set up a flow that matches every packet and directs
1668 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1669 * switch is in secure mode and we won't pass any traffic until
1670 * a controller has been defined and it tells us to do so. */
1672 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1673 union ofp_action action;
1676 memset(&action, 0, sizeof action);
1677 action.type = htons(OFPAT_OUTPUT);
1678 action.output.len = htons(sizeof action);
1679 action.output.port = htons(OFPP_NORMAL);
1680 memset(&flow, 0, sizeof flow);
1681 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1686 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1691 for (i = 0; i < br->n_ports; i++) {
1692 struct port *port = br->ports[i];
1693 for (j = 0; j < port->n_ifaces; j++) {
1694 struct iface *iface = port->ifaces[j];
1695 shash_add_once(ifaces, iface->name, iface);
1697 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1698 shash_add_once(ifaces, port->name, NULL);
1703 /* For robustness, in case the administrator moves around datapath ports behind
1704 * our back, we re-check all the datapath port numbers here.
1706 * This function will set the 'dp_ifidx' members of interfaces that have
1707 * disappeared to -1, so only call this function from a context where those
1708 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1709 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1710 * datapath, which doesn't support UINT16_MAX+1 ports. */
1712 bridge_fetch_dp_ifaces(struct bridge *br)
1714 struct odp_port *dpif_ports;
1715 size_t n_dpif_ports;
1718 /* Reset all interface numbers. */
1719 for (i = 0; i < br->n_ports; i++) {
1720 struct port *port = br->ports[i];
1721 for (j = 0; j < port->n_ifaces; j++) {
1722 struct iface *iface = port->ifaces[j];
1723 iface->dp_ifidx = -1;
1726 port_array_clear(&br->ifaces);
1728 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1729 for (i = 0; i < n_dpif_ports; i++) {
1730 struct odp_port *p = &dpif_ports[i];
1731 struct iface *iface = iface_lookup(br, p->devname);
1733 if (iface->dp_ifidx >= 0) {
1734 VLOG_WARN("%s reported interface %s twice",
1735 dpif_name(br->dpif), p->devname);
1736 } else if (iface_from_dp_ifidx(br, p->port)) {
1737 VLOG_WARN("%s reported interface %"PRIu16" twice",
1738 dpif_name(br->dpif), p->port);
1740 port_array_set(&br->ifaces, p->port, iface);
1741 iface->dp_ifidx = p->port;
1745 int64_t ofport = (iface->dp_ifidx >= 0
1746 ? odp_port_to_ofp_port(iface->dp_ifidx)
1748 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1755 /* Bridge packet processing functions. */
1758 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1760 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1763 static struct bond_entry *
1764 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1766 return &port->bond_hash[bond_hash(mac)];
1770 bond_choose_iface(const struct port *port)
1772 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1773 size_t i, best_down_slave = -1;
1774 long long next_delay_expiration = LLONG_MAX;
1776 for (i = 0; i < port->n_ifaces; i++) {
1777 struct iface *iface = port->ifaces[i];
1779 if (iface->enabled) {
1781 } else if (iface->delay_expires < next_delay_expiration) {
1782 best_down_slave = i;
1783 next_delay_expiration = iface->delay_expires;
1787 if (best_down_slave != -1) {
1788 struct iface *iface = port->ifaces[best_down_slave];
1790 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1791 "since no other interface is up", iface->name,
1792 iface->delay_expires - time_msec());
1793 bond_enable_slave(iface, true);
1796 return best_down_slave;
1800 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1801 uint16_t *dp_ifidx, tag_type *tags)
1803 struct iface *iface;
1805 assert(port->n_ifaces);
1806 if (port->n_ifaces == 1) {
1807 iface = port->ifaces[0];
1809 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1810 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1811 || !port->ifaces[e->iface_idx]->enabled) {
1812 /* XXX select interface properly. The current interface selection
1813 * is only good for testing the rebalancing code. */
1814 e->iface_idx = bond_choose_iface(port);
1815 if (e->iface_idx < 0) {
1816 *tags |= port->no_ifaces_tag;
1819 e->iface_tag = tag_create_random();
1820 ((struct port *) port)->bond_compat_is_stale = true;
1822 *tags |= e->iface_tag;
1823 iface = port->ifaces[e->iface_idx];
1825 *dp_ifidx = iface->dp_ifidx;
1826 *tags |= iface->tag; /* Currently only used for bonding. */
1831 bond_link_status_update(struct iface *iface, bool carrier)
1833 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1834 struct port *port = iface->port;
1836 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1837 /* Nothing to do. */
1840 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1841 iface->name, carrier ? "detected" : "dropped");
1842 if (carrier == iface->enabled) {
1843 iface->delay_expires = LLONG_MAX;
1844 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1845 iface->name, carrier ? "disabled" : "enabled");
1846 } else if (carrier && port->active_iface < 0) {
1847 bond_enable_slave(iface, true);
1848 if (port->updelay) {
1849 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1850 "other interface is up", iface->name, port->updelay);
1853 int delay = carrier ? port->updelay : port->downdelay;
1854 iface->delay_expires = time_msec() + delay;
1857 "interface %s: will be %s if it stays %s for %d ms",
1859 carrier ? "enabled" : "disabled",
1860 carrier ? "up" : "down",
1867 bond_choose_active_iface(struct port *port)
1869 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1871 port->active_iface = bond_choose_iface(port);
1872 port->active_iface_tag = tag_create_random();
1873 if (port->active_iface >= 0) {
1874 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1875 port->name, port->ifaces[port->active_iface]->name);
1877 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1883 bond_enable_slave(struct iface *iface, bool enable)
1885 struct port *port = iface->port;
1886 struct bridge *br = port->bridge;
1888 /* This acts as a recursion check. If the act of disabling a slave
1889 * causes a different slave to be enabled, the flag will allow us to
1890 * skip redundant work when we reenter this function. It must be
1891 * cleared on exit to keep things safe with multiple bonds. */
1892 static bool moving_active_iface = false;
1894 iface->delay_expires = LLONG_MAX;
1895 if (enable == iface->enabled) {
1899 iface->enabled = enable;
1900 if (!iface->enabled) {
1901 VLOG_WARN("interface %s: disabled", iface->name);
1902 ofproto_revalidate(br->ofproto, iface->tag);
1903 if (iface->port_ifidx == port->active_iface) {
1904 ofproto_revalidate(br->ofproto,
1905 port->active_iface_tag);
1907 /* Disabling a slave can lead to another slave being immediately
1908 * enabled if there will be no active slaves but one is waiting
1909 * on an updelay. In this case we do not need to run most of the
1910 * code for the newly enabled slave since there was no period
1911 * without an active slave and it is redundant with the disabling
1913 moving_active_iface = true;
1914 bond_choose_active_iface(port);
1916 bond_send_learning_packets(port);
1918 VLOG_WARN("interface %s: enabled", iface->name);
1919 if (port->active_iface < 0 && !moving_active_iface) {
1920 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1921 bond_choose_active_iface(port);
1922 bond_send_learning_packets(port);
1924 iface->tag = tag_create_random();
1927 moving_active_iface = false;
1928 port->bond_compat_is_stale = true;
1931 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1932 * bond interface. */
1934 bond_update_fake_iface_stats(struct port *port)
1936 struct netdev_stats bond_stats;
1937 struct netdev *bond_dev;
1940 memset(&bond_stats, 0, sizeof bond_stats);
1942 for (i = 0; i < port->n_ifaces; i++) {
1943 struct netdev_stats slave_stats;
1945 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1946 /* XXX: We swap the stats here because they are swapped back when
1947 * reported by the internal device. The reason for this is
1948 * internal devices normally represent packets going into the system
1949 * but when used as fake bond device they represent packets leaving
1950 * the system. We really should do this in the internal device
1951 * itself because changing it here reverses the counts from the
1952 * perspective of the switch. However, the internal device doesn't
1953 * know what type of device it represents so we have to do it here
1955 bond_stats.tx_packets += slave_stats.rx_packets;
1956 bond_stats.tx_bytes += slave_stats.rx_bytes;
1957 bond_stats.rx_packets += slave_stats.tx_packets;
1958 bond_stats.rx_bytes += slave_stats.tx_bytes;
1962 if (!netdev_open_default(port->name, &bond_dev)) {
1963 netdev_set_stats(bond_dev, &bond_stats);
1964 netdev_close(bond_dev);
1969 bond_run(struct bridge *br)
1973 for (i = 0; i < br->n_ports; i++) {
1974 struct port *port = br->ports[i];
1976 if (port->n_ifaces >= 2) {
1977 for (j = 0; j < port->n_ifaces; j++) {
1978 struct iface *iface = port->ifaces[j];
1979 if (time_msec() >= iface->delay_expires) {
1980 bond_enable_slave(iface, !iface->enabled);
1984 if (port->bond_fake_iface
1985 && time_msec() >= port->bond_next_fake_iface_update) {
1986 bond_update_fake_iface_stats(port);
1987 port->bond_next_fake_iface_update = time_msec() + 1000;
1991 if (port->bond_compat_is_stale) {
1992 port->bond_compat_is_stale = false;
1993 port_update_bond_compat(port);
1999 bond_wait(struct bridge *br)
2003 for (i = 0; i < br->n_ports; i++) {
2004 struct port *port = br->ports[i];
2005 if (port->n_ifaces < 2) {
2008 for (j = 0; j < port->n_ifaces; j++) {
2009 struct iface *iface = port->ifaces[j];
2010 if (iface->delay_expires != LLONG_MAX) {
2011 poll_timer_wait_until(iface->delay_expires);
2014 if (port->bond_fake_iface) {
2015 poll_timer_wait_until(port->bond_next_fake_iface_update);
2021 set_dst(struct dst *p, const flow_t *flow,
2022 const struct port *in_port, const struct port *out_port,
2025 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2026 : in_port->vlan >= 0 ? in_port->vlan
2027 : ntohs(flow->dl_vlan));
2028 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
2032 swap_dst(struct dst *p, struct dst *q)
2034 struct dst tmp = *p;
2039 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2040 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2041 * that we push to the datapath. We could in fact fully sort the array by
2042 * vlan, but in most cases there are at most two different vlan tags so that's
2043 * possibly overkill.) */
2045 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2047 struct dst *first = dsts;
2048 struct dst *last = dsts + n_dsts;
2050 while (first != last) {
2052 * - All dsts < first have vlan == 'vlan'.
2053 * - All dsts >= last have vlan != 'vlan'.
2054 * - first < last. */
2055 while (first->vlan == vlan) {
2056 if (++first == last) {
2061 /* Same invariants, plus one additional:
2062 * - first->vlan != vlan.
2064 while (last[-1].vlan != vlan) {
2065 if (--last == first) {
2070 /* Same invariants, plus one additional:
2071 * - last[-1].vlan == vlan.*/
2072 swap_dst(first++, --last);
2077 mirror_mask_ffs(mirror_mask_t mask)
2079 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2084 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2085 const struct dst *test)
2088 for (i = 0; i < n_dsts; i++) {
2089 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2097 port_trunks_vlan(const struct port *port, uint16_t vlan)
2099 return (port->vlan < 0
2100 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2104 port_includes_vlan(const struct port *port, uint16_t vlan)
2106 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2110 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2111 const struct port *in_port, const struct port *out_port,
2112 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2114 mirror_mask_t mirrors = in_port->src_mirrors;
2115 struct dst *dst = dsts;
2118 if (out_port == FLOOD_PORT) {
2119 /* XXX use ODP_FLOOD if no vlans or bonding. */
2120 /* XXX even better, define each VLAN as a datapath port group */
2121 for (i = 0; i < br->n_ports; i++) {
2122 struct port *port = br->ports[i];
2123 if (port != in_port && port_includes_vlan(port, vlan)
2124 && !port->is_mirror_output_port
2125 && set_dst(dst, flow, in_port, port, tags)) {
2126 mirrors |= port->dst_mirrors;
2130 *nf_output_iface = NF_OUT_FLOOD;
2131 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2132 *nf_output_iface = dst->dp_ifidx;
2133 mirrors |= out_port->dst_mirrors;
2138 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2139 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2141 if (set_dst(dst, flow, in_port, m->out_port, tags)
2142 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2146 for (i = 0; i < br->n_ports; i++) {
2147 struct port *port = br->ports[i];
2148 if (port_includes_vlan(port, m->out_vlan)
2149 && set_dst(dst, flow, in_port, port, tags))
2153 if (port->vlan < 0) {
2154 dst->vlan = m->out_vlan;
2156 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2160 /* Use the vlan tag on the original flow instead of
2161 * the one passed in the vlan parameter. This ensures
2162 * that we compare the vlan from before any implicit
2163 * tagging tags place. This is necessary because
2164 * dst->vlan is the final vlan, after removing implicit
2166 flow_vlan = ntohs(flow->dl_vlan);
2167 if (flow_vlan == 0) {
2168 flow_vlan = OFP_VLAN_NONE;
2170 if (port == in_port && dst->vlan == flow_vlan) {
2171 /* Don't send out input port on same VLAN. */
2179 mirrors &= mirrors - 1;
2182 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2186 static void OVS_UNUSED
2187 print_dsts(const struct dst *dsts, size_t n)
2189 for (; n--; dsts++) {
2190 printf(">p%"PRIu16, dsts->dp_ifidx);
2191 if (dsts->vlan != OFP_VLAN_NONE) {
2192 printf("v%"PRIu16, dsts->vlan);
2198 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2199 const struct port *in_port, const struct port *out_port,
2200 tag_type *tags, struct odp_actions *actions,
2201 uint16_t *nf_output_iface)
2203 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2205 const struct dst *p;
2208 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2211 cur_vlan = ntohs(flow->dl_vlan);
2212 for (p = dsts; p < &dsts[n_dsts]; p++) {
2213 union odp_action *a;
2214 if (p->vlan != cur_vlan) {
2215 if (p->vlan == OFP_VLAN_NONE) {
2216 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2218 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2219 a->vlan_vid.vlan_vid = htons(p->vlan);
2223 a = odp_actions_add(actions, ODPAT_OUTPUT);
2224 a->output.port = p->dp_ifidx;
2228 /* Returns the effective vlan of a packet, taking into account both the
2229 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2230 * the packet is untagged and -1 indicates it has an invalid header and
2231 * should be dropped. */
2232 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2233 struct port *in_port, bool have_packet)
2235 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2236 * belongs to VLAN 0, so we should treat both cases identically. (In the
2237 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2238 * presumably to allow a priority to be specified. In the latter case, the
2239 * packet does not have any 802.1Q header.) */
2240 int vlan = ntohs(flow->dl_vlan);
2241 if (vlan == OFP_VLAN_NONE) {
2244 if (in_port->vlan >= 0) {
2246 /* XXX support double tagging? */
2248 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2249 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2250 "packet received on port %s configured with "
2251 "implicit VLAN %"PRIu16,
2252 br->name, ntohs(flow->dl_vlan),
2253 in_port->name, in_port->vlan);
2257 vlan = in_port->vlan;
2259 if (!port_includes_vlan(in_port, vlan)) {
2261 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2262 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2263 "packet received on port %s not configured for "
2265 br->name, vlan, in_port->name, vlan);
2274 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2275 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2276 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2278 is_gratuitous_arp(const flow_t *flow)
2280 return (flow->dl_type == htons(ETH_TYPE_ARP)
2281 && eth_addr_is_broadcast(flow->dl_dst)
2282 && (flow->nw_proto == ARP_OP_REPLY
2283 || (flow->nw_proto == ARP_OP_REQUEST
2284 && flow->nw_src == flow->nw_dst)));
2288 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2289 struct port *in_port)
2291 enum grat_arp_lock_type lock_type;
2294 /* We don't want to learn from gratuitous ARP packets that are reflected
2295 * back over bond slaves so we lock the learning table. */
2296 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2297 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2298 GRAT_ARP_LOCK_CHECK;
2300 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2303 /* The log messages here could actually be useful in debugging,
2304 * so keep the rate limit relatively high. */
2305 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2307 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2308 "on port %s in VLAN %d",
2309 br->name, ETH_ADDR_ARGS(flow->dl_src),
2310 in_port->name, vlan);
2311 ofproto_revalidate(br->ofproto, rev_tag);
2315 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2316 * dropped. Returns true if they may be forwarded, false if they should be
2319 * If 'have_packet' is true, it indicates that the caller is processing a
2320 * received packet. If 'have_packet' is false, then the caller is just
2321 * revalidating an existing flow because configuration has changed. Either
2322 * way, 'have_packet' only affects logging (there is no point in logging errors
2323 * during revalidation).
2325 * Sets '*in_portp' to the input port. This will be a null pointer if
2326 * flow->in_port does not designate a known input port (in which case
2327 * is_admissible() returns false).
2329 * When returning true, sets '*vlanp' to the effective VLAN of the input
2330 * packet, as returned by flow_get_vlan().
2332 * May also add tags to '*tags', although the current implementation only does
2333 * so in one special case.
2336 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2337 tag_type *tags, int *vlanp, struct port **in_portp)
2339 struct iface *in_iface;
2340 struct port *in_port;
2343 /* Find the interface and port structure for the received packet. */
2344 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2346 /* No interface? Something fishy... */
2348 /* Odd. A few possible reasons here:
2350 * - We deleted an interface but there are still a few packets
2351 * queued up from it.
2353 * - Someone externally added an interface (e.g. with "ovs-dpctl
2354 * add-if") that we don't know about.
2356 * - Packet arrived on the local port but the local port is not
2357 * one of our bridge ports.
2359 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2361 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2362 "interface %"PRIu16, br->name, flow->in_port);
2368 *in_portp = in_port = in_iface->port;
2369 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2374 /* Drop frames for reserved multicast addresses. */
2375 if (eth_addr_is_reserved(flow->dl_dst)) {
2379 /* Drop frames on ports reserved for mirroring. */
2380 if (in_port->is_mirror_output_port) {
2382 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2383 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2384 "%s, which is reserved exclusively for mirroring",
2385 br->name, in_port->name);
2390 /* Packets received on bonds need special attention to avoid duplicates. */
2391 if (in_port->n_ifaces > 1) {
2393 bool is_grat_arp_locked;
2395 if (eth_addr_is_multicast(flow->dl_dst)) {
2396 *tags |= in_port->active_iface_tag;
2397 if (in_port->active_iface != in_iface->port_ifidx) {
2398 /* Drop all multicast packets on inactive slaves. */
2403 /* Drop all packets for which we have learned a different input
2404 * port, because we probably sent the packet on one slave and got
2405 * it back on the other. Gratuitous ARP packets are an exception
2406 * to this rule: the host has moved to another switch. The exception
2407 * to the exception is if we locked the learning table to avoid
2408 * reflections on bond slaves. If this is the case, just drop the
2410 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2411 &is_grat_arp_locked);
2412 if (src_idx != -1 && src_idx != in_port->port_idx &&
2413 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2421 /* If the composed actions may be applied to any packet in the given 'flow',
2422 * returns true. Otherwise, the actions should only be applied to 'packet', or
2423 * not at all, if 'packet' was NULL. */
2425 process_flow(struct bridge *br, const flow_t *flow,
2426 const struct ofpbuf *packet, struct odp_actions *actions,
2427 tag_type *tags, uint16_t *nf_output_iface)
2429 struct port *in_port;
2430 struct port *out_port;
2434 /* Check whether we should drop packets in this flow. */
2435 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2440 /* Learn source MAC (but don't try to learn from revalidation). */
2442 update_learning_table(br, flow, vlan, in_port);
2445 /* Determine output port. */
2446 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2448 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2449 out_port = br->ports[out_port_idx];
2450 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2451 /* If we are revalidating but don't have a learning entry then
2452 * eject the flow. Installing a flow that floods packets opens
2453 * up a window of time where we could learn from a packet reflected
2454 * on a bond and blackhole packets before the learning table is
2455 * updated to reflect the correct port. */
2458 out_port = FLOOD_PORT;
2461 /* Don't send packets out their input ports. */
2462 if (in_port == out_port) {
2468 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2475 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2478 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2479 const struct ofp_phy_port *opp,
2482 struct bridge *br = br_;
2483 struct iface *iface;
2486 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2492 if (reason == OFPPR_DELETE) {
2493 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2494 br->name, iface->name);
2495 iface_destroy(iface);
2496 if (!port->n_ifaces) {
2497 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2498 br->name, port->name);
2504 if (port->n_ifaces > 1) {
2505 bool up = !(opp->state & OFPPS_LINK_DOWN);
2506 bond_link_status_update(iface, up);
2507 port_update_bond_compat(port);
2513 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2514 struct odp_actions *actions, tag_type *tags,
2515 uint16_t *nf_output_iface, void *br_)
2517 struct bridge *br = br_;
2519 COVERAGE_INC(bridge_process_flow);
2521 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2525 bridge_account_flow_ofhook_cb(const flow_t *flow, tag_type tags,
2526 const union odp_action *actions,
2527 size_t n_actions, unsigned long long int n_bytes,
2530 struct bridge *br = br_;
2531 const union odp_action *a;
2532 struct port *in_port;
2536 /* Feed information from the active flows back into the learning table to
2537 * ensure that table is always in sync with what is actually flowing
2538 * through the datapath.
2540 * We test that 'tags' is nonzero to ensure that only flows that include an
2541 * OFPP_NORMAL action are used for learning. This works because
2542 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2543 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2544 update_learning_table(br, flow, vlan, in_port);
2547 /* Account for bond slave utilization. */
2548 if (!br->has_bonded_ports) {
2551 for (a = actions; a < &actions[n_actions]; a++) {
2552 if (a->type == ODPAT_OUTPUT) {
2553 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2554 if (out_port && out_port->n_ifaces >= 2) {
2555 struct bond_entry *e = lookup_bond_entry(out_port,
2557 e->tx_bytes += n_bytes;
2564 bridge_account_checkpoint_ofhook_cb(void *br_)
2566 struct bridge *br = br_;
2570 if (!br->has_bonded_ports) {
2575 for (i = 0; i < br->n_ports; i++) {
2576 struct port *port = br->ports[i];
2577 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2578 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2579 bond_rebalance_port(port);
2584 static struct ofhooks bridge_ofhooks = {
2585 bridge_port_changed_ofhook_cb,
2586 bridge_normal_ofhook_cb,
2587 bridge_account_flow_ofhook_cb,
2588 bridge_account_checkpoint_ofhook_cb,
2591 /* Bonding functions. */
2593 /* Statistics for a single interface on a bonded port, used for load-based
2594 * bond rebalancing. */
2595 struct slave_balance {
2596 struct iface *iface; /* The interface. */
2597 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2599 /* All the "bond_entry"s that are assigned to this interface, in order of
2600 * increasing tx_bytes. */
2601 struct bond_entry **hashes;
2605 /* Sorts pointers to pointers to bond_entries in ascending order by the
2606 * interface to which they are assigned, and within a single interface in
2607 * ascending order of bytes transmitted. */
2609 compare_bond_entries(const void *a_, const void *b_)
2611 const struct bond_entry *const *ap = a_;
2612 const struct bond_entry *const *bp = b_;
2613 const struct bond_entry *a = *ap;
2614 const struct bond_entry *b = *bp;
2615 if (a->iface_idx != b->iface_idx) {
2616 return a->iface_idx > b->iface_idx ? 1 : -1;
2617 } else if (a->tx_bytes != b->tx_bytes) {
2618 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2624 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2625 * *descending* order by number of bytes transmitted. */
2627 compare_slave_balance(const void *a_, const void *b_)
2629 const struct slave_balance *a = a_;
2630 const struct slave_balance *b = b_;
2631 if (a->iface->enabled != b->iface->enabled) {
2632 return a->iface->enabled ? -1 : 1;
2633 } else if (a->tx_bytes != b->tx_bytes) {
2634 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2641 swap_bals(struct slave_balance *a, struct slave_balance *b)
2643 struct slave_balance tmp = *a;
2648 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2649 * given that 'p' (and only 'p') might be in the wrong location.
2651 * This function invalidates 'p', since it might now be in a different memory
2654 resort_bals(struct slave_balance *p,
2655 struct slave_balance bals[], size_t n_bals)
2658 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2659 swap_bals(p, p - 1);
2661 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2662 swap_bals(p, p + 1);
2668 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2670 if (VLOG_IS_DBG_ENABLED()) {
2671 struct ds ds = DS_EMPTY_INITIALIZER;
2672 const struct slave_balance *b;
2674 for (b = bals; b < bals + n_bals; b++) {
2678 ds_put_char(&ds, ',');
2680 ds_put_format(&ds, " %s %"PRIu64"kB",
2681 b->iface->name, b->tx_bytes / 1024);
2683 if (!b->iface->enabled) {
2684 ds_put_cstr(&ds, " (disabled)");
2686 if (b->n_hashes > 0) {
2687 ds_put_cstr(&ds, " (");
2688 for (i = 0; i < b->n_hashes; i++) {
2689 const struct bond_entry *e = b->hashes[i];
2691 ds_put_cstr(&ds, " + ");
2693 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2694 e - port->bond_hash, e->tx_bytes / 1024);
2696 ds_put_cstr(&ds, ")");
2699 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2704 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2706 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2709 struct bond_entry *hash = from->hashes[hash_idx];
2710 struct port *port = from->iface->port;
2711 uint64_t delta = hash->tx_bytes;
2713 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2714 "from %s to %s (now carrying %"PRIu64"kB and "
2715 "%"PRIu64"kB load, respectively)",
2716 port->name, delta / 1024, hash - port->bond_hash,
2717 from->iface->name, to->iface->name,
2718 (from->tx_bytes - delta) / 1024,
2719 (to->tx_bytes + delta) / 1024);
2721 /* Delete element from from->hashes.
2723 * We don't bother to add the element to to->hashes because not only would
2724 * it require more work, the only purpose it would be to allow that hash to
2725 * be migrated to another slave in this rebalancing run, and there is no
2726 * point in doing that. */
2727 if (hash_idx == 0) {
2730 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2731 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2735 /* Shift load away from 'from' to 'to'. */
2736 from->tx_bytes -= delta;
2737 to->tx_bytes += delta;
2739 /* Arrange for flows to be revalidated. */
2740 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2741 hash->iface_idx = to->iface->port_ifidx;
2742 hash->iface_tag = tag_create_random();
2746 bond_rebalance_port(struct port *port)
2748 struct slave_balance bals[DP_MAX_PORTS];
2750 struct bond_entry *hashes[BOND_MASK + 1];
2751 struct slave_balance *b, *from, *to;
2752 struct bond_entry *e;
2755 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2756 * descending order of tx_bytes, so that bals[0] represents the most
2757 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2760 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2761 * array for each slave_balance structure, we sort our local array of
2762 * hashes in order by slave, so that all of the hashes for a given slave
2763 * become contiguous in memory, and then we point each 'hashes' members of
2764 * a slave_balance structure to the start of a contiguous group. */
2765 n_bals = port->n_ifaces;
2766 for (b = bals; b < &bals[n_bals]; b++) {
2767 b->iface = port->ifaces[b - bals];
2772 for (i = 0; i <= BOND_MASK; i++) {
2773 hashes[i] = &port->bond_hash[i];
2775 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2776 for (i = 0; i <= BOND_MASK; i++) {
2778 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2779 b = &bals[e->iface_idx];
2780 b->tx_bytes += e->tx_bytes;
2782 b->hashes = &hashes[i];
2787 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2788 log_bals(bals, n_bals, port);
2790 /* Discard slaves that aren't enabled (which were sorted to the back of the
2791 * array earlier). */
2792 while (!bals[n_bals - 1].iface->enabled) {
2799 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2800 to = &bals[n_bals - 1];
2801 for (from = bals; from < to; ) {
2802 uint64_t overload = from->tx_bytes - to->tx_bytes;
2803 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2804 /* The extra load on 'from' (and all less-loaded slaves), compared
2805 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2806 * it is less than ~1Mbps. No point in rebalancing. */
2808 } else if (from->n_hashes == 1) {
2809 /* 'from' only carries a single MAC hash, so we can't shift any
2810 * load away from it, even though we want to. */
2813 /* 'from' is carrying significantly more load than 'to', and that
2814 * load is split across at least two different hashes. Pick a hash
2815 * to migrate to 'to' (the least-loaded slave), given that doing so
2816 * must decrease the ratio of the load on the two slaves by at
2819 * The sort order we use means that we prefer to shift away the
2820 * smallest hashes instead of the biggest ones. There is little
2821 * reason behind this decision; we could use the opposite sort
2822 * order to shift away big hashes ahead of small ones. */
2825 for (i = 0; i < from->n_hashes; i++) {
2826 double old_ratio, new_ratio;
2827 uint64_t delta = from->hashes[i]->tx_bytes;
2829 if (delta == 0 || from->tx_bytes - delta == 0) {
2830 /* Pointless move. */
2834 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2836 if (to->tx_bytes == 0) {
2837 /* Nothing on the new slave, move it. */
2841 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2842 new_ratio = (double)(from->tx_bytes - delta) /
2843 (to->tx_bytes + delta);
2845 if (new_ratio == 0) {
2846 /* Should already be covered but check to prevent division
2851 if (new_ratio < 1) {
2852 new_ratio = 1 / new_ratio;
2855 if (old_ratio - new_ratio > 0.1) {
2856 /* Would decrease the ratio, move it. */
2860 if (i < from->n_hashes) {
2861 bond_shift_load(from, to, i);
2862 port->bond_compat_is_stale = true;
2864 /* If the result of the migration changed the relative order of
2865 * 'from' and 'to' swap them back to maintain invariants. */
2866 if (order_swapped) {
2867 swap_bals(from, to);
2870 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2871 * point to different slave_balance structures. It is only
2872 * valid to do these two operations in a row at all because we
2873 * know that 'from' will not move past 'to' and vice versa. */
2874 resort_bals(from, bals, n_bals);
2875 resort_bals(to, bals, n_bals);
2882 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2883 * historical data to decay to <1% in 7 rebalancing runs. */
2884 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2890 bond_send_learning_packets(struct port *port)
2892 struct bridge *br = port->bridge;
2893 struct mac_entry *e;
2894 struct ofpbuf packet;
2895 int error, n_packets, n_errors;
2897 if (!port->n_ifaces || port->active_iface < 0) {
2901 ofpbuf_init(&packet, 128);
2902 error = n_packets = n_errors = 0;
2903 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2904 union ofp_action actions[2], *a;
2910 if (e->port == port->port_idx
2911 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2915 /* Compose actions. */
2916 memset(actions, 0, sizeof actions);
2919 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2920 a->vlan_vid.len = htons(sizeof *a);
2921 a->vlan_vid.vlan_vid = htons(e->vlan);
2924 a->output.type = htons(OFPAT_OUTPUT);
2925 a->output.len = htons(sizeof *a);
2926 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2931 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2933 flow_extract(&packet, 0, ODPP_NONE, &flow);
2934 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2941 ofpbuf_uninit(&packet);
2944 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2945 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2946 "packets, last error was: %s",
2947 port->name, n_errors, n_packets, strerror(error));
2949 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2950 port->name, n_packets);
2954 /* Bonding unixctl user interface functions. */
2957 bond_unixctl_list(struct unixctl_conn *conn,
2958 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2960 struct ds ds = DS_EMPTY_INITIALIZER;
2961 const struct bridge *br;
2963 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2965 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2968 for (i = 0; i < br->n_ports; i++) {
2969 const struct port *port = br->ports[i];
2970 if (port->n_ifaces > 1) {
2973 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2974 for (j = 0; j < port->n_ifaces; j++) {
2975 const struct iface *iface = port->ifaces[j];
2977 ds_put_cstr(&ds, ", ");
2979 ds_put_cstr(&ds, iface->name);
2981 ds_put_char(&ds, '\n');
2985 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2989 static struct port *
2990 bond_find(const char *name)
2992 const struct bridge *br;
2994 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2997 for (i = 0; i < br->n_ports; i++) {
2998 struct port *port = br->ports[i];
2999 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3008 bond_unixctl_show(struct unixctl_conn *conn,
3009 const char *args, void *aux OVS_UNUSED)
3011 struct ds ds = DS_EMPTY_INITIALIZER;
3012 const struct port *port;
3015 port = bond_find(args);
3017 unixctl_command_reply(conn, 501, "no such bond");
3021 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3022 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3023 ds_put_format(&ds, "next rebalance: %lld ms\n",
3024 port->bond_next_rebalance - time_msec());
3025 for (j = 0; j < port->n_ifaces; j++) {
3026 const struct iface *iface = port->ifaces[j];
3027 struct bond_entry *be;
3030 ds_put_format(&ds, "slave %s: %s\n",
3031 iface->name, iface->enabled ? "enabled" : "disabled");
3032 if (j == port->active_iface) {
3033 ds_put_cstr(&ds, "\tactive slave\n");
3035 if (iface->delay_expires != LLONG_MAX) {
3036 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3037 iface->enabled ? "downdelay" : "updelay",
3038 iface->delay_expires - time_msec());
3042 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3043 int hash = be - port->bond_hash;
3044 struct mac_entry *me;
3046 if (be->iface_idx != j) {
3050 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3051 hash, be->tx_bytes / 1024);
3054 LIST_FOR_EACH (me, struct mac_entry, lru_node,
3055 &port->bridge->ml->lrus) {
3058 if (bond_hash(me->mac) == hash
3059 && me->port != port->port_idx
3060 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3061 && dp_ifidx == iface->dp_ifidx)
3063 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3064 ETH_ADDR_ARGS(me->mac));
3069 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3074 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3075 void *aux OVS_UNUSED)
3077 char *args = (char *) args_;
3078 char *save_ptr = NULL;
3079 char *bond_s, *hash_s, *slave_s;
3080 uint8_t mac[ETH_ADDR_LEN];
3082 struct iface *iface;
3083 struct bond_entry *entry;
3086 bond_s = strtok_r(args, " ", &save_ptr);
3087 hash_s = strtok_r(NULL, " ", &save_ptr);
3088 slave_s = strtok_r(NULL, " ", &save_ptr);
3090 unixctl_command_reply(conn, 501,
3091 "usage: bond/migrate BOND HASH SLAVE");
3095 port = bond_find(bond_s);
3097 unixctl_command_reply(conn, 501, "no such bond");
3101 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3102 == ETH_ADDR_SCAN_COUNT) {
3103 hash = bond_hash(mac);
3104 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3105 hash = atoi(hash_s) & BOND_MASK;
3107 unixctl_command_reply(conn, 501, "bad hash");
3111 iface = port_lookup_iface(port, slave_s);
3113 unixctl_command_reply(conn, 501, "no such slave");
3117 if (!iface->enabled) {
3118 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3122 entry = &port->bond_hash[hash];
3123 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3124 entry->iface_idx = iface->port_ifidx;
3125 entry->iface_tag = tag_create_random();
3126 port->bond_compat_is_stale = true;
3127 unixctl_command_reply(conn, 200, "migrated");
3131 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3132 void *aux OVS_UNUSED)
3134 char *args = (char *) args_;
3135 char *save_ptr = NULL;
3136 char *bond_s, *slave_s;
3138 struct iface *iface;
3140 bond_s = strtok_r(args, " ", &save_ptr);
3141 slave_s = strtok_r(NULL, " ", &save_ptr);
3143 unixctl_command_reply(conn, 501,
3144 "usage: bond/set-active-slave BOND SLAVE");
3148 port = bond_find(bond_s);
3150 unixctl_command_reply(conn, 501, "no such bond");
3154 iface = port_lookup_iface(port, slave_s);
3156 unixctl_command_reply(conn, 501, "no such slave");
3160 if (!iface->enabled) {
3161 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3165 if (port->active_iface != iface->port_ifidx) {
3166 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3167 port->active_iface = iface->port_ifidx;
3168 port->active_iface_tag = tag_create_random();
3169 VLOG_INFO("port %s: active interface is now %s",
3170 port->name, iface->name);
3171 bond_send_learning_packets(port);
3172 unixctl_command_reply(conn, 200, "done");
3174 unixctl_command_reply(conn, 200, "no change");
3179 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3181 char *args = (char *) args_;
3182 char *save_ptr = NULL;
3183 char *bond_s, *slave_s;
3185 struct iface *iface;
3187 bond_s = strtok_r(args, " ", &save_ptr);
3188 slave_s = strtok_r(NULL, " ", &save_ptr);
3190 unixctl_command_reply(conn, 501,
3191 "usage: bond/enable/disable-slave BOND SLAVE");
3195 port = bond_find(bond_s);
3197 unixctl_command_reply(conn, 501, "no such bond");
3201 iface = port_lookup_iface(port, slave_s);
3203 unixctl_command_reply(conn, 501, "no such slave");
3207 bond_enable_slave(iface, enable);
3208 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3212 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3213 void *aux OVS_UNUSED)
3215 enable_slave(conn, args, true);
3219 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3220 void *aux OVS_UNUSED)
3222 enable_slave(conn, args, false);
3226 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3227 void *aux OVS_UNUSED)
3229 uint8_t mac[ETH_ADDR_LEN];
3233 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3234 == ETH_ADDR_SCAN_COUNT) {
3235 hash = bond_hash(mac);
3237 hash_cstr = xasprintf("%u", hash);
3238 unixctl_command_reply(conn, 200, hash_cstr);
3241 unixctl_command_reply(conn, 501, "invalid mac");
3248 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3249 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3250 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3251 unixctl_command_register("bond/set-active-slave",
3252 bond_unixctl_set_active_slave, NULL);
3253 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3255 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3257 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3260 /* Port functions. */
3262 static struct port *
3263 port_create(struct bridge *br, const char *name)
3267 port = xzalloc(sizeof *port);
3269 port->port_idx = br->n_ports;
3271 port->trunks = NULL;
3272 port->name = xstrdup(name);
3273 port->active_iface = -1;
3275 if (br->n_ports >= br->allocated_ports) {
3276 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3279 br->ports[br->n_ports++] = port;
3280 shash_add_assert(&br->port_by_name, port->name, port);
3282 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3289 get_port_other_config(const struct ovsrec_port *port, const char *key,
3290 const char *default_value)
3294 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3296 return value ? value : default_value;
3300 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3302 struct shash new_ifaces;
3305 /* Collect list of new interfaces. */
3306 shash_init(&new_ifaces);
3307 for (i = 0; i < cfg->n_interfaces; i++) {
3308 const char *name = cfg->interfaces[i]->name;
3309 shash_add_once(&new_ifaces, name, NULL);
3312 /* Get rid of deleted interfaces. */
3313 for (i = 0; i < port->n_ifaces; ) {
3314 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3315 iface_destroy(port->ifaces[i]);
3321 shash_destroy(&new_ifaces);
3325 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3327 struct shash new_ifaces;
3328 long long int next_rebalance;
3329 unsigned long *trunks;
3335 /* Update settings. */
3336 port->updelay = cfg->bond_updelay;
3337 if (port->updelay < 0) {
3340 port->downdelay = cfg->bond_downdelay;
3341 if (port->downdelay < 0) {
3342 port->downdelay = 0;
3344 port->bond_rebalance_interval = atoi(
3345 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3346 if (port->bond_rebalance_interval < 1000) {
3347 port->bond_rebalance_interval = 1000;
3349 next_rebalance = time_msec() + port->bond_rebalance_interval;
3350 if (port->bond_next_rebalance > next_rebalance) {
3351 port->bond_next_rebalance = next_rebalance;
3354 /* Add new interfaces and update 'cfg' member of existing ones. */
3355 shash_init(&new_ifaces);
3356 for (i = 0; i < cfg->n_interfaces; i++) {
3357 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3358 struct iface *iface;
3360 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3361 VLOG_WARN("port %s: %s specified twice as port interface",
3362 port->name, if_cfg->name);
3366 iface = iface_lookup(port->bridge, if_cfg->name);
3368 if (iface->port != port) {
3369 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3371 port->bridge->name, if_cfg->name, iface->port->name);
3374 iface->cfg = if_cfg;
3376 iface_create(port, if_cfg);
3379 shash_destroy(&new_ifaces);
3384 if (port->n_ifaces < 2) {
3386 if (vlan >= 0 && vlan <= 4095) {
3387 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3392 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3393 * they even work as-is. But they have not been tested. */
3394 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3398 if (port->vlan != vlan) {
3400 bridge_flush(port->bridge);
3403 /* Get trunked VLANs. */
3405 if (vlan < 0 && cfg->n_trunks) {
3408 trunks = bitmap_allocate(4096);
3410 for (i = 0; i < cfg->n_trunks; i++) {
3411 int trunk = cfg->trunks[i];
3413 bitmap_set1(trunks, trunk);
3419 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3420 port->name, cfg->n_trunks);
3422 if (n_errors == cfg->n_trunks) {
3423 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3425 bitmap_free(trunks);
3428 } else if (vlan >= 0 && cfg->n_trunks) {
3429 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3433 ? port->trunks != NULL
3434 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3435 bridge_flush(port->bridge);
3437 bitmap_free(port->trunks);
3438 port->trunks = trunks;
3442 port_destroy(struct port *port)
3445 struct bridge *br = port->bridge;
3449 proc_net_compat_update_vlan(port->name, NULL, 0);
3450 proc_net_compat_update_bond(port->name, NULL);
3452 for (i = 0; i < MAX_MIRRORS; i++) {
3453 struct mirror *m = br->mirrors[i];
3454 if (m && m->out_port == port) {
3459 while (port->n_ifaces > 0) {
3460 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3463 shash_find_and_delete_assert(&br->port_by_name, port->name);
3465 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3466 del->port_idx = port->port_idx;
3469 bitmap_free(port->trunks);
3476 static struct port *
3477 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3479 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3480 return iface ? iface->port : NULL;
3483 static struct port *
3484 port_lookup(const struct bridge *br, const char *name)
3486 return shash_find_data(&br->port_by_name, name);
3489 static struct iface *
3490 port_lookup_iface(const struct port *port, const char *name)
3492 struct iface *iface = iface_lookup(port->bridge, name);
3493 return iface && iface->port == port ? iface : NULL;
3497 port_update_bonding(struct port *port)
3499 if (port->n_ifaces < 2) {
3500 /* Not a bonded port. */
3501 if (port->bond_hash) {
3502 free(port->bond_hash);
3503 port->bond_hash = NULL;
3504 port->bond_compat_is_stale = true;
3505 port->bond_fake_iface = false;
3508 if (!port->bond_hash) {
3511 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3512 for (i = 0; i <= BOND_MASK; i++) {
3513 struct bond_entry *e = &port->bond_hash[i];
3517 port->no_ifaces_tag = tag_create_random();
3518 bond_choose_active_iface(port);
3519 port->bond_next_rebalance
3520 = time_msec() + port->bond_rebalance_interval;
3522 if (port->cfg->bond_fake_iface) {
3523 port->bond_next_fake_iface_update = time_msec();
3526 port->bond_compat_is_stale = true;
3527 port->bond_fake_iface = port->cfg->bond_fake_iface;
3532 port_update_bond_compat(struct port *port)
3534 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3535 struct compat_bond bond;
3538 if (port->n_ifaces < 2) {
3539 proc_net_compat_update_bond(port->name, NULL);
3544 bond.updelay = port->updelay;
3545 bond.downdelay = port->downdelay;
3548 bond.hashes = compat_hashes;
3549 if (port->bond_hash) {
3550 const struct bond_entry *e;
3551 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3552 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3553 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3554 cbh->hash = e - port->bond_hash;
3555 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3560 bond.n_slaves = port->n_ifaces;
3561 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3562 for (i = 0; i < port->n_ifaces; i++) {
3563 struct iface *iface = port->ifaces[i];
3564 struct compat_bond_slave *slave = &bond.slaves[i];
3565 slave->name = iface->name;
3567 /* We need to make the same determination as the Linux bonding
3568 * code to determine whether a slave should be consider "up".
3569 * The Linux function bond_miimon_inspect() supports four
3570 * BOND_LINK_* states:
3572 * - BOND_LINK_UP: carrier detected, updelay has passed.
3573 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3574 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3575 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3577 * The function bond_info_show_slave() only considers BOND_LINK_UP
3578 * to be "up" and anything else to be "down".
3580 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3584 netdev_get_etheraddr(iface->netdev, slave->mac);
3587 if (port->bond_fake_iface) {
3588 struct netdev *bond_netdev;
3590 if (!netdev_open_default(port->name, &bond_netdev)) {
3592 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3594 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3596 netdev_close(bond_netdev);
3600 proc_net_compat_update_bond(port->name, &bond);
3605 port_update_vlan_compat(struct port *port)
3607 struct bridge *br = port->bridge;
3608 char *vlandev_name = NULL;
3610 if (port->vlan > 0) {
3611 /* Figure out the name that the VLAN device should actually have, if it
3612 * existed. This takes some work because the VLAN device would not
3613 * have port->name in its name; rather, it would have the trunk port's
3614 * name, and 'port' would be attached to a bridge that also had the
3615 * VLAN device one of its ports. So we need to find a trunk port that
3616 * includes port->vlan.
3618 * There might be more than one candidate. This doesn't happen on
3619 * XenServer, so if it happens we just pick the first choice in
3620 * alphabetical order instead of creating multiple VLAN devices. */
3622 for (i = 0; i < br->n_ports; i++) {
3623 struct port *p = br->ports[i];
3624 if (port_trunks_vlan(p, port->vlan)
3626 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3628 uint8_t ea[ETH_ADDR_LEN];
3629 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3630 if (!eth_addr_is_multicast(ea) &&
3631 !eth_addr_is_reserved(ea) &&
3632 !eth_addr_is_zero(ea)) {
3633 vlandev_name = p->name;
3638 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3641 /* Interface functions. */
3643 static struct iface *
3644 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3646 struct bridge *br = port->bridge;
3647 struct iface *iface;
3648 char *name = if_cfg->name;
3651 iface = xzalloc(sizeof *iface);
3653 iface->port_ifidx = port->n_ifaces;
3654 iface->name = xstrdup(name);
3655 iface->dp_ifidx = -1;
3656 iface->tag = tag_create_random();
3657 iface->delay_expires = LLONG_MAX;
3658 iface->netdev = NULL;
3659 iface->cfg = if_cfg;
3661 shash_add_assert(&br->iface_by_name, iface->name, iface);
3663 /* Attempt to create the network interface in case it doesn't exist yet. */
3664 if (!iface_is_internal(br, iface->name)) {
3665 error = set_up_iface(if_cfg, iface, true);
3667 VLOG_WARN("could not create iface %s: %s", iface->name,
3670 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3677 if (port->n_ifaces >= port->allocated_ifaces) {
3678 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3679 sizeof *port->ifaces);
3681 port->ifaces[port->n_ifaces++] = iface;
3682 if (port->n_ifaces > 1) {
3683 br->has_bonded_ports = true;
3686 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3694 iface_destroy(struct iface *iface)
3697 struct port *port = iface->port;
3698 struct bridge *br = port->bridge;
3699 bool del_active = port->active_iface == iface->port_ifidx;
3702 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3704 if (iface->dp_ifidx >= 0) {
3705 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3708 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3709 del->port_ifidx = iface->port_ifidx;
3711 netdev_close(iface->netdev);
3714 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3715 bond_choose_active_iface(port);
3716 bond_send_learning_packets(port);
3722 bridge_flush(port->bridge);
3726 static struct iface *
3727 iface_lookup(const struct bridge *br, const char *name)
3729 return shash_find_data(&br->iface_by_name, name);
3732 static struct iface *
3733 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3735 return port_array_get(&br->ifaces, dp_ifidx);
3738 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3739 * 'br', that is, an interface that is entirely simulated within the datapath.
3740 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3741 * interfaces are created by setting "iface.<iface>.internal = true".
3743 * In addition, we have a kluge-y feature that creates an internal port with
3744 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3745 * This feature needs to go away in the long term. Until then, this is one
3746 * reason why this function takes a name instead of a struct iface: the fake
3747 * interfaces created this way do not have a struct iface. */
3749 iface_is_internal(const struct bridge *br, const char *if_name)
3751 struct iface *iface;
3754 if (!strcmp(if_name, br->name)) {
3758 iface = iface_lookup(br, if_name);
3759 if (iface && !strcmp(iface->cfg->type, "internal")) {
3763 port = port_lookup(br, if_name);
3764 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3770 /* Set Ethernet address of 'iface', if one is specified in the configuration
3773 iface_set_mac(struct iface *iface)
3775 uint8_t ea[ETH_ADDR_LEN];
3777 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3778 if (eth_addr_is_multicast(ea)) {
3779 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3781 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3782 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3783 iface->name, iface->name);
3785 int error = netdev_set_etheraddr(iface->netdev, ea);
3787 VLOG_ERR("interface %s: setting MAC failed (%s)",
3788 iface->name, strerror(error));
3795 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3796 struct shash *shash)
3801 for (i = 0; i < n; i++) {
3802 shash_add(shash, keys[i], values[i]);
3806 struct iface_delete_queues_cbdata {
3807 struct netdev *netdev;
3808 const struct ovsdb_datum *queues;
3812 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
3814 union ovsdb_atom atom;
3816 atom.integer = target;
3817 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
3821 iface_delete_queues(unsigned int queue_id,
3822 const struct shash *details OVS_UNUSED, void *cbdata_)
3824 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3826 if (!queue_ids_include(cbdata->queues, queue_id)) {
3827 netdev_delete_queue(cbdata->netdev, queue_id);
3832 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3834 if (!qos || qos->type[0] == '\0') {
3835 netdev_set_qos(iface->netdev, NULL, NULL);
3837 struct iface_delete_queues_cbdata cbdata;
3838 struct shash details;
3841 /* Configure top-level Qos for 'iface'. */
3842 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3843 qos->n_other_config, &details);
3844 netdev_set_qos(iface->netdev, qos->type, &details);
3845 shash_destroy(&details);
3847 /* Deconfigure queues that were deleted. */
3848 cbdata.netdev = iface->netdev;
3849 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
3851 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3853 /* Configure queues for 'iface'. */
3854 for (i = 0; i < qos->n_queues; i++) {
3855 const struct ovsrec_queue *queue = qos->value_queues[i];
3856 unsigned int queue_id = qos->key_queues[i];
3858 shash_from_ovs_idl_map(queue->key_other_config,
3859 queue->value_other_config,
3860 queue->n_other_config, &details);
3861 netdev_set_queue(iface->netdev, queue_id, &details);
3862 shash_destroy(&details);
3867 /* Port mirroring. */
3869 static struct mirror *
3870 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
3874 for (i = 0; i < MAX_MIRRORS; i++) {
3875 struct mirror *m = br->mirrors[i];
3876 if (m && uuid_equals(uuid, &m->uuid)) {
3884 mirror_reconfigure(struct bridge *br)
3886 unsigned long *rspan_vlans;
3889 /* Get rid of deleted mirrors. */
3890 for (i = 0; i < MAX_MIRRORS; i++) {
3891 struct mirror *m = br->mirrors[i];
3893 const struct ovsdb_datum *mc;
3894 union ovsdb_atom atom;
3896 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
3897 atom.uuid = br->mirrors[i]->uuid;
3898 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
3904 /* Add new mirrors and reconfigure existing ones. */
3905 for (i = 0; i < br->cfg->n_mirrors; i++) {
3906 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3907 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
3909 mirror_reconfigure_one(m, cfg);
3911 mirror_create(br, cfg);
3915 /* Update port reserved status. */
3916 for (i = 0; i < br->n_ports; i++) {
3917 br->ports[i]->is_mirror_output_port = false;
3919 for (i = 0; i < MAX_MIRRORS; i++) {
3920 struct mirror *m = br->mirrors[i];
3921 if (m && m->out_port) {
3922 m->out_port->is_mirror_output_port = true;
3926 /* Update flooded vlans (for RSPAN). */
3928 if (br->cfg->n_flood_vlans) {
3929 rspan_vlans = bitmap_allocate(4096);
3931 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3932 int64_t vlan = br->cfg->flood_vlans[i];
3933 if (vlan >= 0 && vlan < 4096) {
3934 bitmap_set1(rspan_vlans, vlan);
3935 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3938 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3943 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3949 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
3954 for (i = 0; ; i++) {
3955 if (i >= MAX_MIRRORS) {
3956 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3957 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
3960 if (!br->mirrors[i]) {
3965 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
3968 br->mirrors[i] = m = xzalloc(sizeof *m);
3971 m->name = xstrdup(cfg->name);
3972 shash_init(&m->src_ports);
3973 shash_init(&m->dst_ports);
3979 mirror_reconfigure_one(m, cfg);
3983 mirror_destroy(struct mirror *m)
3986 struct bridge *br = m->bridge;
3989 for (i = 0; i < br->n_ports; i++) {
3990 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3991 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3994 shash_destroy(&m->src_ports);
3995 shash_destroy(&m->dst_ports);
3998 m->bridge->mirrors[m->idx] = NULL;
4007 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4008 struct shash *names)
4012 for (i = 0; i < n_ports; i++) {
4013 const char *name = ports[i]->name;
4014 if (port_lookup(m->bridge, name)) {
4015 shash_add_once(names, name, NULL);
4017 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4018 "port %s", m->bridge->name, m->name, name);
4024 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4030 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4032 for (i = 0; i < cfg->n_select_vlan; i++) {
4033 int64_t vlan = cfg->select_vlan[i];
4034 if (vlan < 0 || vlan > 4095) {
4035 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4036 m->bridge->name, m->name, vlan);
4038 (*vlans)[n_vlans++] = vlan;
4045 vlan_is_mirrored(const struct mirror *m, int vlan)
4049 for (i = 0; i < m->n_vlans; i++) {
4050 if (m->vlans[i] == vlan) {
4058 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4062 for (i = 0; i < m->n_vlans; i++) {
4063 if (port_trunks_vlan(p, m->vlans[i])) {
4071 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4073 struct shash src_ports, dst_ports;
4074 mirror_mask_t mirror_bit;
4075 struct port *out_port;
4082 if (strcmp(cfg->name, m->name)) {
4084 m->name = xstrdup(cfg->name);
4087 /* Get output port. */
4088 if (cfg->output_port) {
4089 out_port = port_lookup(m->bridge, cfg->output_port->name);
4091 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4092 m->bridge->name, m->name);
4098 if (cfg->output_vlan) {
4099 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4100 "output vlan; ignoring output vlan",
4101 m->bridge->name, m->name);
4103 } else if (cfg->output_vlan) {
4105 out_vlan = *cfg->output_vlan;
4107 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4108 m->bridge->name, m->name);
4113 shash_init(&src_ports);
4114 shash_init(&dst_ports);
4115 if (cfg->select_all) {
4116 for (i = 0; i < m->bridge->n_ports; i++) {
4117 const char *name = m->bridge->ports[i]->name;
4118 shash_add_once(&src_ports, name, NULL);
4119 shash_add_once(&dst_ports, name, NULL);
4124 /* Get ports, and drop duplicates and ports that don't exist. */
4125 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4127 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4130 /* Get all the vlans, and drop duplicate and invalid vlans. */
4131 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4134 /* Update mirror data. */
4135 if (!shash_equal_keys(&m->src_ports, &src_ports)
4136 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4137 || m->n_vlans != n_vlans
4138 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4139 || m->out_port != out_port
4140 || m->out_vlan != out_vlan) {
4141 bridge_flush(m->bridge);
4143 shash_swap(&m->src_ports, &src_ports);
4144 shash_swap(&m->dst_ports, &dst_ports);
4147 m->n_vlans = n_vlans;
4148 m->out_port = out_port;
4149 m->out_vlan = out_vlan;
4152 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4153 for (i = 0; i < m->bridge->n_ports; i++) {
4154 struct port *port = m->bridge->ports[i];
4156 if (shash_find(&m->src_ports, port->name)
4159 ? port_trunks_any_mirrored_vlan(m, port)
4160 : vlan_is_mirrored(m, port->vlan)))) {
4161 port->src_mirrors |= mirror_bit;
4163 port->src_mirrors &= ~mirror_bit;
4166 if (shash_find(&m->dst_ports, port->name)) {
4167 port->dst_mirrors |= mirror_bit;
4169 port->dst_mirrors &= ~mirror_bit;
4174 shash_destroy(&src_ports);
4175 shash_destroy(&dst_ports);