1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
23 #include <sys/socket.h>
25 #include <openflow/openflow.h>
30 #include <sys/socket.h>
31 #include <sys/types.h>
37 #include "dynamic-string.h"
43 #include "mac-learning.h"
46 #include "ofp-print.h"
48 #include "ofproto/netflow.h"
49 #include "ofproto/ofproto.h"
50 #include "ovsdb-data.h"
52 #include "poll-loop.h"
53 #include "proc-net-compat.h"
57 #include "socket-util.h"
58 #include "stream-ssl.h"
60 #include "system-stats.h"
65 #include "vswitchd/vswitch-idl.h"
66 #include "xenserver.h"
69 #include "sflow_api.h"
71 VLOG_DEFINE_THIS_MODULE(bridge)
79 /* These members are always valid. */
80 struct port *port; /* Containing port. */
81 size_t port_ifidx; /* Index within containing port. */
82 char *name; /* Host network device name. */
83 tag_type tag; /* Tag associated with this interface. */
84 long long delay_expires; /* Time after which 'enabled' may change. */
86 /* These members are valid only after bridge_reconfigure() causes them to
88 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
89 int dp_ifidx; /* Index within kernel datapath. */
90 struct netdev *netdev; /* Network device. */
91 bool enabled; /* May be chosen for flows? */
92 const struct ovsrec_interface *cfg;
95 #define BOND_MASK 0xff
97 int iface_idx; /* Index of assigned iface, or -1 if none. */
98 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
99 tag_type iface_tag; /* Tag associated with iface_idx. */
102 #define MAX_MIRRORS 32
103 typedef uint32_t mirror_mask_t;
104 #define MIRROR_MASK_C(X) UINT32_C(X)
105 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
107 struct bridge *bridge;
110 struct uuid uuid; /* UUID of this "mirror" record in database. */
112 /* Selection criteria. */
113 struct shash src_ports; /* Name is port name; data is always NULL. */
114 struct shash dst_ports; /* Name is port name; data is always NULL. */
119 struct port *out_port;
123 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
125 struct bridge *bridge;
127 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
128 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
129 * NULL if all VLANs are trunked. */
130 const struct ovsrec_port *cfg;
133 /* An ordinary bridge port has 1 interface.
134 * A bridge port for bonding has at least 2 interfaces. */
135 struct iface **ifaces;
136 size_t n_ifaces, allocated_ifaces;
139 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
140 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
141 tag_type active_iface_tag; /* Tag for bcast flows. */
142 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
143 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
144 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
145 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
146 long long int bond_next_fake_iface_update; /* Time of next update. */
147 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
148 long long int bond_next_rebalance; /* Next rebalancing time. */
150 /* Port mirroring info. */
151 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
152 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
153 bool is_mirror_output_port; /* Does port mirroring send frames here? */
156 #define DP_MAX_PORTS 255
158 struct list node; /* Node in global list of bridges. */
159 char *name; /* User-specified arbitrary name. */
160 struct mac_learning *ml; /* MAC learning table. */
161 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
162 const struct ovsrec_bridge *cfg;
164 /* OpenFlow switch processing. */
165 struct ofproto *ofproto; /* OpenFlow switch. */
167 /* Kernel datapath information. */
168 struct dpif *dpif; /* Datapath. */
169 struct hmap ifaces; /* Contains "struct iface"s. */
173 size_t n_ports, allocated_ports;
174 struct shash iface_by_name; /* "struct iface"s indexed by name. */
175 struct shash port_by_name; /* "struct port"s indexed by name. */
178 bool has_bonded_ports;
183 /* Port mirroring. */
184 struct mirror *mirrors[MAX_MIRRORS];
187 /* List of all bridges. */
188 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
190 /* OVSDB IDL used to obtain configuration. */
191 static struct ovsdb_idl *idl;
193 /* Each time this timer expires, the bridge fetches systems and interface
194 * statistics and pushes them into the database. */
195 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
196 static long long int stats_timer = LLONG_MIN;
198 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
199 static void bridge_destroy(struct bridge *);
200 static struct bridge *bridge_lookup(const char *name);
201 static unixctl_cb_func bridge_unixctl_dump_flows;
202 static unixctl_cb_func bridge_unixctl_reconnect;
203 static int bridge_run_one(struct bridge *);
204 static size_t bridge_get_controllers(const struct bridge *br,
205 struct ovsrec_controller ***controllersp);
206 static void bridge_reconfigure_one(struct bridge *);
207 static void bridge_reconfigure_remotes(struct bridge *,
208 const struct sockaddr_in *managers,
210 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
211 static void bridge_fetch_dp_ifaces(struct bridge *);
212 static void bridge_flush(struct bridge *);
213 static void bridge_pick_local_hw_addr(struct bridge *,
214 uint8_t ea[ETH_ADDR_LEN],
215 struct iface **hw_addr_iface);
216 static uint64_t bridge_pick_datapath_id(struct bridge *,
217 const uint8_t bridge_ea[ETH_ADDR_LEN],
218 struct iface *hw_addr_iface);
219 static struct iface *bridge_get_local_iface(struct bridge *);
220 static uint64_t dpid_from_hash(const void *, size_t nbytes);
222 static unixctl_cb_func bridge_unixctl_fdb_show;
224 static void bond_init(void);
225 static void bond_run(struct bridge *);
226 static void bond_wait(struct bridge *);
227 static void bond_rebalance_port(struct port *);
228 static void bond_send_learning_packets(struct port *);
229 static void bond_enable_slave(struct iface *iface, bool enable);
231 static struct port *port_create(struct bridge *, const char *name);
232 static void port_reconfigure(struct port *, const struct ovsrec_port *);
233 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
234 static void port_destroy(struct port *);
235 static struct port *port_lookup(const struct bridge *, const char *name);
236 static struct iface *port_lookup_iface(const struct port *, const char *name);
237 static struct port *port_from_dp_ifidx(const struct bridge *,
239 static void port_update_bond_compat(struct port *);
240 static void port_update_vlan_compat(struct port *);
241 static void port_update_bonding(struct port *);
243 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
244 static void mirror_destroy(struct mirror *);
245 static void mirror_reconfigure(struct bridge *);
246 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
247 static bool vlan_is_mirrored(const struct mirror *, int vlan);
249 static struct iface *iface_create(struct port *port,
250 const struct ovsrec_interface *if_cfg);
251 static void iface_destroy(struct iface *);
252 static struct iface *iface_lookup(const struct bridge *, const char *name);
253 static struct iface *iface_from_dp_ifidx(const struct bridge *,
255 static bool iface_is_internal(const struct bridge *, const char *name);
256 static void iface_set_mac(struct iface *);
257 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
259 /* Hooks into ofproto processing. */
260 static struct ofhooks bridge_ofhooks;
262 /* Public functions. */
264 /* Initializes the bridge module, configuring it to obtain its configuration
265 * from an OVSDB server accessed over 'remote', which should be a string in a
266 * form acceptable to ovsdb_idl_create(). */
268 bridge_init(const char *remote)
270 /* Create connection to database. */
271 idl = ovsdb_idl_create(remote, &ovsrec_idl_class);
273 ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_cur_cfg);
274 ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_statistics);
275 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
277 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
279 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
280 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
282 ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_ofport);
283 ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_statistics);
284 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
286 /* Register unixctl commands. */
287 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
288 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
290 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
295 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
296 * but for which the ovs-vswitchd configuration 'cfg' is required. */
298 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
300 static bool already_configured_once;
301 struct svec bridge_names;
302 struct svec dpif_names, dpif_types;
305 /* Only do this once per ovs-vswitchd run. */
306 if (already_configured_once) {
309 already_configured_once = true;
311 stats_timer = time_msec() + STATS_INTERVAL;
313 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
314 svec_init(&bridge_names);
315 for (i = 0; i < cfg->n_bridges; i++) {
316 svec_add(&bridge_names, cfg->bridges[i]->name);
318 svec_sort(&bridge_names);
320 /* Iterate over all system dpifs and delete any of them that do not appear
322 svec_init(&dpif_names);
323 svec_init(&dpif_types);
324 dp_enumerate_types(&dpif_types);
325 for (i = 0; i < dpif_types.n; i++) {
330 dp_enumerate_names(dpif_types.names[i], &dpif_names);
332 /* For each dpif... */
333 for (j = 0; j < dpif_names.n; j++) {
334 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
336 struct svec all_names;
339 /* ...check whether any of its names is in 'bridge_names'. */
340 svec_init(&all_names);
341 dpif_get_all_names(dpif, &all_names);
342 for (k = 0; k < all_names.n; k++) {
343 if (svec_contains(&bridge_names, all_names.names[k])) {
348 /* No. Delete the dpif. */
352 svec_destroy(&all_names);
357 svec_destroy(&bridge_names);
358 svec_destroy(&dpif_names);
359 svec_destroy(&dpif_types);
362 /* Attempt to create the network device 'iface_name' through the netdev
365 set_up_iface(struct iface *iface, bool create)
367 struct shash options;
371 shash_init(&options);
372 for (i = 0; i < iface->cfg->n_options; i++) {
373 shash_add(&options, iface->cfg->key_options[i],
374 xstrdup(iface->cfg->value_options[i]));
377 /* Include 'other_config' keys in hash of netdev options. The
378 * namespace of 'other_config' and 'options' must be disjoint.
379 * Prefer 'options' keys over 'other_config' keys. */
380 for (i = 0; i < iface->cfg->n_other_config; i++) {
381 char *value = xstrdup(iface->cfg->value_other_config[i]);
382 if (!shash_add_once(&options, iface->cfg->key_other_config[i],
384 VLOG_WARN("%s: \"other_config\" key %s conflicts with existing "
385 "\"other_config\" or \"options\" entry...ignoring",
386 iface->cfg->name, iface->cfg->key_other_config[i]);
392 struct netdev_options netdev_options;
394 memset(&netdev_options, 0, sizeof netdev_options);
395 netdev_options.name = iface->cfg->name;
396 if (!strcmp(iface->cfg->type, "internal")) {
397 /* An "internal" config type maps to a netdev "system" type. */
398 netdev_options.type = "system";
400 netdev_options.type = iface->cfg->type;
402 netdev_options.args = &options;
403 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
405 error = netdev_open(&netdev_options, &iface->netdev);
408 netdev_get_carrier(iface->netdev, &iface->enabled);
410 } else if (iface->netdev) {
411 const char *netdev_type = netdev_get_type(iface->netdev);
412 const char *iface_type = iface->cfg->type && strlen(iface->cfg->type)
413 ? iface->cfg->type : NULL;
415 /* An "internal" config type maps to a netdev "system" type. */
416 if (iface_type && !strcmp(iface_type, "internal")) {
417 iface_type = "system";
420 if (!iface_type || !strcmp(netdev_type, iface_type)) {
421 error = netdev_reconfigure(iface->netdev, &options);
423 VLOG_WARN("%s: attempting change device type from %s to %s",
424 iface->cfg->name, netdev_type, iface_type);
428 shash_destroy_free_data(&options);
434 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
435 void *aux OVS_UNUSED)
437 if (!iface->netdev) {
438 int error = set_up_iface(iface, true);
440 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
450 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
451 void *aux OVS_UNUSED)
453 if (iface->dp_ifidx >= 0) {
454 VLOG_DBG("%s has interface %s on port %d",
456 iface->name, iface->dp_ifidx);
459 VLOG_ERR("%s interface not in %s, dropping",
460 iface->name, dpif_name(br->dpif));
466 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
467 void *aux OVS_UNUSED)
469 /* Set policing attributes. */
470 netdev_set_policing(iface->netdev,
471 iface->cfg->ingress_policing_rate,
472 iface->cfg->ingress_policing_burst);
474 /* Set MAC address of internal interfaces other than the local
476 if (iface->dp_ifidx != ODPP_LOCAL
477 && iface_is_internal(br, iface->name)) {
478 iface_set_mac(iface);
484 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
485 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
486 * deletes from 'br' any ports that no longer have any interfaces. */
488 iterate_and_prune_ifaces(struct bridge *br,
489 bool (*cb)(struct bridge *, struct iface *,
495 for (i = 0; i < br->n_ports; ) {
496 struct port *port = br->ports[i];
497 for (j = 0; j < port->n_ifaces; ) {
498 struct iface *iface = port->ifaces[j];
499 if (cb(br, iface, aux)) {
502 iface_destroy(iface);
506 if (port->n_ifaces) {
509 VLOG_ERR("%s port has no interfaces, dropping", port->name);
515 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
516 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
517 * responsible for freeing '*managersp' (with free()).
519 * You may be asking yourself "why does ovs-vswitchd care?", because
520 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
521 * should not be and in fact is not directly involved in that. But
522 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
523 * it has to tell in-band control where the managers are to enable that.
526 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
527 struct sockaddr_in **managersp, size_t *n_managersp)
529 struct sockaddr_in *managers = NULL;
530 size_t n_managers = 0;
532 if (ovs_cfg->n_managers > 0) {
535 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
536 for (i = 0; i < ovs_cfg->n_managers; i++) {
537 const char *name = ovs_cfg->managers[i];
538 struct sockaddr_in *sin = &managers[i];
540 if ((!strncmp(name, "tcp:", 4)
541 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
542 (!strncmp(name, "ssl:", 4)
543 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
549 *managersp = managers;
550 *n_managersp = n_managers;
554 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
556 struct shash old_br, new_br;
557 struct shash_node *node;
558 struct bridge *br, *next;
559 struct sockaddr_in *managers;
562 int sflow_bridge_number;
564 COVERAGE_INC(bridge_reconfigure);
566 collect_managers(ovs_cfg, &managers, &n_managers);
568 /* Collect old and new bridges. */
571 LIST_FOR_EACH (br, node, &all_bridges) {
572 shash_add(&old_br, br->name, br);
574 for (i = 0; i < ovs_cfg->n_bridges; i++) {
575 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
576 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
577 VLOG_WARN("more than one bridge named %s", br_cfg->name);
581 /* Get rid of deleted bridges and add new bridges. */
582 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
583 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
590 SHASH_FOR_EACH (node, &new_br) {
591 const char *br_name = node->name;
592 const struct ovsrec_bridge *br_cfg = node->data;
593 br = shash_find_data(&old_br, br_name);
595 /* If the bridge datapath type has changed, we need to tear it
596 * down and recreate. */
597 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
599 bridge_create(br_cfg);
602 bridge_create(br_cfg);
605 shash_destroy(&old_br);
606 shash_destroy(&new_br);
608 /* Reconfigure all bridges. */
609 LIST_FOR_EACH (br, node, &all_bridges) {
610 bridge_reconfigure_one(br);
613 /* Add and delete ports on all datapaths.
615 * The kernel will reject any attempt to add a given port to a datapath if
616 * that port already belongs to a different datapath, so we must do all
617 * port deletions before any port additions. */
618 LIST_FOR_EACH (br, node, &all_bridges) {
619 struct odp_port *dpif_ports;
621 struct shash want_ifaces;
623 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
624 bridge_get_all_ifaces(br, &want_ifaces);
625 for (i = 0; i < n_dpif_ports; i++) {
626 const struct odp_port *p = &dpif_ports[i];
627 if (!shash_find(&want_ifaces, p->devname)
628 && strcmp(p->devname, br->name)) {
629 int retval = dpif_port_del(br->dpif, p->port);
631 VLOG_ERR("failed to remove %s interface from %s: %s",
632 p->devname, dpif_name(br->dpif),
637 shash_destroy(&want_ifaces);
640 LIST_FOR_EACH (br, node, &all_bridges) {
641 struct odp_port *dpif_ports;
643 struct shash cur_ifaces, want_ifaces;
645 /* Get the set of interfaces currently in this datapath. */
646 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
647 shash_init(&cur_ifaces);
648 for (i = 0; i < n_dpif_ports; i++) {
649 const char *name = dpif_ports[i].devname;
650 shash_add_once(&cur_ifaces, name, NULL);
654 /* Get the set of interfaces we want on this datapath. */
655 bridge_get_all_ifaces(br, &want_ifaces);
657 SHASH_FOR_EACH (node, &want_ifaces) {
658 const char *if_name = node->name;
659 struct iface *iface = node->data;
661 if (shash_find(&cur_ifaces, if_name)) {
662 /* Already exists, just reconfigure it. */
664 set_up_iface(iface, false);
667 /* Need to add to datapath. */
671 /* Add to datapath. */
672 internal = iface_is_internal(br, if_name);
673 error = dpif_port_add(br->dpif, if_name,
674 internal ? ODP_PORT_INTERNAL : 0, NULL);
675 if (error == EFBIG) {
676 VLOG_ERR("ran out of valid port numbers on %s",
677 dpif_name(br->dpif));
680 VLOG_ERR("failed to add %s interface to %s: %s",
681 if_name, dpif_name(br->dpif), strerror(error));
685 shash_destroy(&cur_ifaces);
686 shash_destroy(&want_ifaces);
688 sflow_bridge_number = 0;
689 LIST_FOR_EACH (br, node, &all_bridges) {
692 struct iface *local_iface;
693 struct iface *hw_addr_iface;
696 bridge_fetch_dp_ifaces(br);
698 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
699 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
701 /* Pick local port hardware address, datapath ID. */
702 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
703 local_iface = bridge_get_local_iface(br);
705 int error = netdev_set_etheraddr(local_iface->netdev, ea);
707 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
708 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
709 "Ethernet address: %s",
710 br->name, strerror(error));
714 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
715 ofproto_set_datapath_id(br->ofproto, dpid);
717 dpid_string = xasprintf("%016"PRIx64, dpid);
718 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
721 /* Set NetFlow configuration on this bridge. */
722 if (br->cfg->netflow) {
723 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
724 struct netflow_options opts;
726 memset(&opts, 0, sizeof opts);
728 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
729 if (nf_cfg->engine_type) {
730 opts.engine_type = *nf_cfg->engine_type;
732 if (nf_cfg->engine_id) {
733 opts.engine_id = *nf_cfg->engine_id;
736 opts.active_timeout = nf_cfg->active_timeout;
737 if (!opts.active_timeout) {
738 opts.active_timeout = -1;
739 } else if (opts.active_timeout < 0) {
740 VLOG_WARN("bridge %s: active timeout interval set to negative "
741 "value, using default instead (%d seconds)", br->name,
742 NF_ACTIVE_TIMEOUT_DEFAULT);
743 opts.active_timeout = -1;
746 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
747 if (opts.add_id_to_iface) {
748 if (opts.engine_id > 0x7f) {
749 VLOG_WARN("bridge %s: netflow port mangling may conflict "
750 "with another vswitch, choose an engine id less "
751 "than 128", br->name);
753 if (br->n_ports > 508) {
754 VLOG_WARN("bridge %s: netflow port mangling will conflict "
755 "with another port when more than 508 ports are "
760 opts.collectors.n = nf_cfg->n_targets;
761 opts.collectors.names = nf_cfg->targets;
762 if (ofproto_set_netflow(br->ofproto, &opts)) {
763 VLOG_ERR("bridge %s: problem setting netflow collectors",
767 ofproto_set_netflow(br->ofproto, NULL);
770 /* Set sFlow configuration on this bridge. */
771 if (br->cfg->sflow) {
772 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
773 struct ovsrec_controller **controllers;
774 struct ofproto_sflow_options oso;
775 size_t n_controllers;
777 memset(&oso, 0, sizeof oso);
779 oso.targets.n = sflow_cfg->n_targets;
780 oso.targets.names = sflow_cfg->targets;
782 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
783 if (sflow_cfg->sampling) {
784 oso.sampling_rate = *sflow_cfg->sampling;
787 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
788 if (sflow_cfg->polling) {
789 oso.polling_interval = *sflow_cfg->polling;
792 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
793 if (sflow_cfg->header) {
794 oso.header_len = *sflow_cfg->header;
797 oso.sub_id = sflow_bridge_number++;
798 oso.agent_device = sflow_cfg->agent;
800 oso.control_ip = NULL;
801 n_controllers = bridge_get_controllers(br, &controllers);
802 for (i = 0; i < n_controllers; i++) {
803 if (controllers[i]->local_ip) {
804 oso.control_ip = controllers[i]->local_ip;
808 ofproto_set_sflow(br->ofproto, &oso);
810 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
812 ofproto_set_sflow(br->ofproto, NULL);
815 /* Update the controller and related settings. It would be more
816 * straightforward to call this from bridge_reconfigure_one(), but we
817 * can't do it there for two reasons. First, and most importantly, at
818 * that point we don't know the dp_ifidx of any interfaces that have
819 * been added to the bridge (because we haven't actually added them to
820 * the datapath). Second, at that point we haven't set the datapath ID
821 * yet; when a controller is configured, resetting the datapath ID will
822 * immediately disconnect from the controller, so it's better to set
823 * the datapath ID before the controller. */
824 bridge_reconfigure_remotes(br, managers, n_managers);
826 LIST_FOR_EACH (br, node, &all_bridges) {
827 for (i = 0; i < br->n_ports; i++) {
828 struct port *port = br->ports[i];
831 port_update_vlan_compat(port);
832 port_update_bonding(port);
834 for (j = 0; j < port->n_ifaces; j++) {
835 iface_update_qos(port->ifaces[j], port->cfg->qos);
839 LIST_FOR_EACH (br, node, &all_bridges) {
840 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
847 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
848 const struct ovsdb_idl_column *column,
851 const struct ovsdb_datum *datum;
852 union ovsdb_atom atom;
855 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
856 atom.string = (char *) key;
857 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
858 return idx == UINT_MAX ? NULL : datum->values[idx].string;
862 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
864 return get_ovsrec_key_value(&br_cfg->header_,
865 &ovsrec_bridge_col_other_config, key);
869 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
870 struct iface **hw_addr_iface)
876 *hw_addr_iface = NULL;
878 /* Did the user request a particular MAC? */
879 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
880 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
881 if (eth_addr_is_multicast(ea)) {
882 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
883 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
884 } else if (eth_addr_is_zero(ea)) {
885 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
891 /* Otherwise choose the minimum non-local MAC address among all of the
893 memset(ea, 0xff, sizeof ea);
894 for (i = 0; i < br->n_ports; i++) {
895 struct port *port = br->ports[i];
896 uint8_t iface_ea[ETH_ADDR_LEN];
899 /* Mirror output ports don't participate. */
900 if (port->is_mirror_output_port) {
904 /* Choose the MAC address to represent the port. */
905 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
906 /* Find the interface with this Ethernet address (if any) so that
907 * we can provide the correct devname to the caller. */
909 for (j = 0; j < port->n_ifaces; j++) {
910 struct iface *candidate = port->ifaces[j];
911 uint8_t candidate_ea[ETH_ADDR_LEN];
912 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
913 && eth_addr_equals(iface_ea, candidate_ea)) {
918 /* Choose the interface whose MAC address will represent the port.
919 * The Linux kernel bonding code always chooses the MAC address of
920 * the first slave added to a bond, and the Fedora networking
921 * scripts always add slaves to a bond in alphabetical order, so
922 * for compatibility we choose the interface with the name that is
923 * first in alphabetical order. */
924 iface = port->ifaces[0];
925 for (j = 1; j < port->n_ifaces; j++) {
926 struct iface *candidate = port->ifaces[j];
927 if (strcmp(candidate->name, iface->name) < 0) {
932 /* The local port doesn't count (since we're trying to choose its
933 * MAC address anyway). */
934 if (iface->dp_ifidx == ODPP_LOCAL) {
939 error = netdev_get_etheraddr(iface->netdev, iface_ea);
941 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
942 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
943 iface->name, strerror(error));
948 /* Compare against our current choice. */
949 if (!eth_addr_is_multicast(iface_ea) &&
950 !eth_addr_is_local(iface_ea) &&
951 !eth_addr_is_reserved(iface_ea) &&
952 !eth_addr_is_zero(iface_ea) &&
953 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
955 memcpy(ea, iface_ea, ETH_ADDR_LEN);
956 *hw_addr_iface = iface;
959 if (eth_addr_is_multicast(ea)) {
960 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
961 *hw_addr_iface = NULL;
962 VLOG_WARN("bridge %s: using default bridge Ethernet "
963 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
965 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
966 br->name, ETH_ADDR_ARGS(ea));
970 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
971 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
972 * an interface on 'br', then that interface must be passed in as
973 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
974 * 'hw_addr_iface' must be passed in as a null pointer. */
976 bridge_pick_datapath_id(struct bridge *br,
977 const uint8_t bridge_ea[ETH_ADDR_LEN],
978 struct iface *hw_addr_iface)
981 * The procedure for choosing a bridge MAC address will, in the most
982 * ordinary case, also choose a unique MAC that we can use as a datapath
983 * ID. In some special cases, though, multiple bridges will end up with
984 * the same MAC address. This is OK for the bridges, but it will confuse
985 * the OpenFlow controller, because each datapath needs a unique datapath
988 * Datapath IDs must be unique. It is also very desirable that they be
989 * stable from one run to the next, so that policy set on a datapath
992 const char *datapath_id;
995 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
996 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1000 if (hw_addr_iface) {
1002 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1004 * A bridge whose MAC address is taken from a VLAN network device
1005 * (that is, a network device created with vconfig(8) or similar
1006 * tool) will have the same MAC address as a bridge on the VLAN
1007 * device's physical network device.
1009 * Handle this case by hashing the physical network device MAC
1010 * along with the VLAN identifier.
1012 uint8_t buf[ETH_ADDR_LEN + 2];
1013 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1014 buf[ETH_ADDR_LEN] = vlan >> 8;
1015 buf[ETH_ADDR_LEN + 1] = vlan;
1016 return dpid_from_hash(buf, sizeof buf);
1019 * Assume that this bridge's MAC address is unique, since it
1020 * doesn't fit any of the cases we handle specially.
1025 * A purely internal bridge, that is, one that has no non-virtual
1026 * network devices on it at all, is more difficult because it has no
1027 * natural unique identifier at all.
1029 * When the host is a XenServer, we handle this case by hashing the
1030 * host's UUID with the name of the bridge. Names of bridges are
1031 * persistent across XenServer reboots, although they can be reused if
1032 * an internal network is destroyed and then a new one is later
1033 * created, so this is fairly effective.
1035 * When the host is not a XenServer, we punt by using a random MAC
1036 * address on each run.
1038 const char *host_uuid = xenserver_get_host_uuid();
1040 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1041 dpid = dpid_from_hash(combined, strlen(combined));
1047 return eth_addr_to_uint64(bridge_ea);
1051 dpid_from_hash(const void *data, size_t n)
1053 uint8_t hash[SHA1_DIGEST_SIZE];
1055 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1056 sha1_bytes(data, n, hash);
1057 eth_addr_mark_random(hash);
1058 return eth_addr_to_uint64(hash);
1062 iface_refresh_stats(struct iface *iface)
1068 static const struct iface_stat iface_stats[] = {
1069 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1070 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1071 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1072 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1073 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1074 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1075 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1076 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1077 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1078 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1079 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1080 { "collisions", offsetof(struct netdev_stats, collisions) },
1082 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1083 const struct iface_stat *s;
1085 char *keys[N_STATS];
1086 int64_t values[N_STATS];
1089 struct netdev_stats stats;
1091 /* Intentionally ignore return value, since errors will set 'stats' to
1092 * all-1s, and we will deal with that correctly below. */
1093 netdev_get_stats(iface->netdev, &stats);
1096 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1097 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1098 if (value != UINT64_MAX) {
1105 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1109 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1111 struct ovsdb_datum datum;
1115 get_system_stats(&stats);
1117 ovsdb_datum_from_shash(&datum, &stats);
1118 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1125 const struct ovsrec_open_vswitch *cfg;
1127 bool datapath_destroyed;
1128 bool database_changed;
1131 /* Let each bridge do the work that it needs to do. */
1132 datapath_destroyed = false;
1133 LIST_FOR_EACH (br, node, &all_bridges) {
1134 int error = bridge_run_one(br);
1136 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1137 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1138 "forcing reconfiguration", br->name);
1139 datapath_destroyed = true;
1143 /* (Re)configure if necessary. */
1144 database_changed = ovsdb_idl_run(idl);
1145 cfg = ovsrec_open_vswitch_first(idl);
1146 if (database_changed || datapath_destroyed) {
1148 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1150 bridge_configure_once(cfg);
1151 bridge_reconfigure(cfg);
1153 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1154 ovsdb_idl_txn_commit(txn);
1155 ovsdb_idl_txn_destroy(txn); /* XXX */
1157 /* We still need to reconfigure to avoid dangling pointers to
1158 * now-destroyed ovsrec structures inside bridge data. */
1159 static const struct ovsrec_open_vswitch null_cfg;
1161 bridge_reconfigure(&null_cfg);
1166 /* Re-configure SSL. We do this on every trip through the main loop,
1167 * instead of just when the database changes, because the contents of the
1168 * key and certificate files can change without the database changing. */
1169 if (cfg && cfg->ssl) {
1170 const struct ovsrec_ssl *ssl = cfg->ssl;
1172 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1173 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1177 /* Refresh system and interface stats if necessary. */
1178 if (time_msec() >= stats_timer) {
1180 struct ovsdb_idl_txn *txn;
1182 txn = ovsdb_idl_txn_create(idl);
1183 LIST_FOR_EACH (br, node, &all_bridges) {
1186 for (i = 0; i < br->n_ports; i++) {
1187 struct port *port = br->ports[i];
1190 for (j = 0; j < port->n_ifaces; j++) {
1191 struct iface *iface = port->ifaces[j];
1192 iface_refresh_stats(iface);
1196 refresh_system_stats(cfg);
1197 ovsdb_idl_txn_commit(txn);
1198 ovsdb_idl_txn_destroy(txn); /* XXX */
1201 stats_timer = time_msec() + STATS_INTERVAL;
1210 LIST_FOR_EACH (br, node, &all_bridges) {
1211 ofproto_wait(br->ofproto);
1212 if (ofproto_has_primary_controller(br->ofproto)) {
1216 mac_learning_wait(br->ml);
1219 ovsdb_idl_wait(idl);
1220 poll_timer_wait_until(stats_timer);
1223 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1224 * configuration changes. */
1226 bridge_flush(struct bridge *br)
1228 COVERAGE_INC(bridge_flush);
1230 mac_learning_flush(br->ml);
1233 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1234 * such interface. */
1235 static struct iface *
1236 bridge_get_local_iface(struct bridge *br)
1240 for (i = 0; i < br->n_ports; i++) {
1241 struct port *port = br->ports[i];
1242 for (j = 0; j < port->n_ifaces; j++) {
1243 struct iface *iface = port->ifaces[j];
1244 if (iface->dp_ifidx == ODPP_LOCAL) {
1253 /* Bridge unixctl user interface functions. */
1255 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1256 const char *args, void *aux OVS_UNUSED)
1258 struct ds ds = DS_EMPTY_INITIALIZER;
1259 const struct bridge *br;
1260 const struct mac_entry *e;
1262 br = bridge_lookup(args);
1264 unixctl_command_reply(conn, 501, "no such bridge");
1268 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1269 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1270 if (e->port < 0 || e->port >= br->n_ports) {
1273 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1274 br->ports[e->port]->ifaces[0]->dp_ifidx,
1275 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1277 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1281 /* Bridge reconfiguration functions. */
1282 static struct bridge *
1283 bridge_create(const struct ovsrec_bridge *br_cfg)
1288 assert(!bridge_lookup(br_cfg->name));
1289 br = xzalloc(sizeof *br);
1291 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1297 dpif_flow_flush(br->dpif);
1299 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1302 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1304 dpif_delete(br->dpif);
1305 dpif_close(br->dpif);
1310 br->name = xstrdup(br_cfg->name);
1312 br->ml = mac_learning_create();
1313 eth_addr_nicira_random(br->default_ea);
1315 hmap_init(&br->ifaces);
1317 shash_init(&br->port_by_name);
1318 shash_init(&br->iface_by_name);
1322 list_push_back(&all_bridges, &br->node);
1324 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1330 bridge_destroy(struct bridge *br)
1335 while (br->n_ports > 0) {
1336 port_destroy(br->ports[br->n_ports - 1]);
1338 list_remove(&br->node);
1339 error = dpif_delete(br->dpif);
1340 if (error && error != ENOENT) {
1341 VLOG_ERR("failed to delete %s: %s",
1342 dpif_name(br->dpif), strerror(error));
1344 dpif_close(br->dpif);
1345 ofproto_destroy(br->ofproto);
1346 mac_learning_destroy(br->ml);
1347 hmap_destroy(&br->ifaces);
1348 shash_destroy(&br->port_by_name);
1349 shash_destroy(&br->iface_by_name);
1356 static struct bridge *
1357 bridge_lookup(const char *name)
1361 LIST_FOR_EACH (br, node, &all_bridges) {
1362 if (!strcmp(br->name, name)) {
1369 /* Handle requests for a listing of all flows known by the OpenFlow
1370 * stack, including those normally hidden. */
1372 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1373 const char *args, void *aux OVS_UNUSED)
1378 br = bridge_lookup(args);
1380 unixctl_command_reply(conn, 501, "Unknown bridge");
1385 ofproto_get_all_flows(br->ofproto, &results);
1387 unixctl_command_reply(conn, 200, ds_cstr(&results));
1388 ds_destroy(&results);
1391 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1392 * connections and reconnect. If BRIDGE is not specified, then all bridges
1393 * drop their controller connections and reconnect. */
1395 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1396 const char *args, void *aux OVS_UNUSED)
1399 if (args[0] != '\0') {
1400 br = bridge_lookup(args);
1402 unixctl_command_reply(conn, 501, "Unknown bridge");
1405 ofproto_reconnect_controllers(br->ofproto);
1407 LIST_FOR_EACH (br, node, &all_bridges) {
1408 ofproto_reconnect_controllers(br->ofproto);
1411 unixctl_command_reply(conn, 200, NULL);
1415 bridge_run_one(struct bridge *br)
1419 error = ofproto_run1(br->ofproto);
1424 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1427 error = ofproto_run2(br->ofproto, br->flush);
1434 bridge_get_controllers(const struct bridge *br,
1435 struct ovsrec_controller ***controllersp)
1437 struct ovsrec_controller **controllers;
1438 size_t n_controllers;
1440 controllers = br->cfg->controller;
1441 n_controllers = br->cfg->n_controller;
1443 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1449 *controllersp = controllers;
1451 return n_controllers;
1455 bridge_reconfigure_one(struct bridge *br)
1457 struct shash old_ports, new_ports;
1458 struct svec snoops, old_snoops;
1459 struct shash_node *node;
1460 enum ofproto_fail_mode fail_mode;
1463 /* Collect old ports. */
1464 shash_init(&old_ports);
1465 for (i = 0; i < br->n_ports; i++) {
1466 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1469 /* Collect new ports. */
1470 shash_init(&new_ports);
1471 for (i = 0; i < br->cfg->n_ports; i++) {
1472 const char *name = br->cfg->ports[i]->name;
1473 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1474 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1479 /* If we have a controller, then we need a local port. Complain if the
1480 * user didn't specify one.
1482 * XXX perhaps we should synthesize a port ourselves in this case. */
1483 if (bridge_get_controllers(br, NULL)) {
1484 char local_name[IF_NAMESIZE];
1487 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1488 local_name, sizeof local_name);
1489 if (!error && !shash_find(&new_ports, local_name)) {
1490 VLOG_WARN("bridge %s: controller specified but no local port "
1491 "(port named %s) defined",
1492 br->name, local_name);
1496 /* Get rid of deleted ports.
1497 * Get rid of deleted interfaces on ports that still exist. */
1498 SHASH_FOR_EACH (node, &old_ports) {
1499 struct port *port = node->data;
1500 const struct ovsrec_port *port_cfg;
1502 port_cfg = shash_find_data(&new_ports, node->name);
1506 port_del_ifaces(port, port_cfg);
1510 /* Create new ports.
1511 * Add new interfaces to existing ports.
1512 * Reconfigure existing ports. */
1513 SHASH_FOR_EACH (node, &new_ports) {
1514 struct port *port = shash_find_data(&old_ports, node->name);
1516 port = port_create(br, node->name);
1519 port_reconfigure(port, node->data);
1520 if (!port->n_ifaces) {
1521 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1522 br->name, port->name);
1526 shash_destroy(&old_ports);
1527 shash_destroy(&new_ports);
1529 /* Set the fail-mode */
1530 fail_mode = !br->cfg->fail_mode
1531 || !strcmp(br->cfg->fail_mode, "standalone")
1532 ? OFPROTO_FAIL_STANDALONE
1533 : OFPROTO_FAIL_SECURE;
1534 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1535 && !ofproto_has_primary_controller(br->ofproto)) {
1536 ofproto_flush_flows(br->ofproto);
1538 ofproto_set_fail_mode(br->ofproto, fail_mode);
1540 /* Delete all flows if we're switching from connected to standalone or vice
1541 * versa. (XXX Should we delete all flows if we are switching from one
1542 * controller to another?) */
1544 /* Configure OpenFlow controller connection snooping. */
1546 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1547 ovs_rundir, br->name));
1548 svec_init(&old_snoops);
1549 ofproto_get_snoops(br->ofproto, &old_snoops);
1550 if (!svec_equal(&snoops, &old_snoops)) {
1551 ofproto_set_snoops(br->ofproto, &snoops);
1553 svec_destroy(&snoops);
1554 svec_destroy(&old_snoops);
1556 mirror_reconfigure(br);
1559 /* Initializes 'oc' appropriately as a management service controller for
1562 * The caller must free oc->target when it is no longer needed. */
1564 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1565 struct ofproto_controller *oc)
1567 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir, br->name);
1568 oc->max_backoff = 0;
1569 oc->probe_interval = 60;
1570 oc->band = OFPROTO_OUT_OF_BAND;
1571 oc->accept_re = NULL;
1572 oc->update_resolv_conf = false;
1574 oc->burst_limit = 0;
1577 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1579 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1580 struct ofproto_controller *oc)
1582 oc->target = c->target;
1583 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1584 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1585 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1586 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1587 oc->accept_re = c->discover_accept_regex;
1588 oc->update_resolv_conf = c->discover_update_resolv_conf;
1589 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1590 oc->burst_limit = (c->controller_burst_limit
1591 ? *c->controller_burst_limit : 0);
1594 /* Configures the IP stack for 'br''s local interface properly according to the
1595 * configuration in 'c'. */
1597 bridge_configure_local_iface_netdev(struct bridge *br,
1598 struct ovsrec_controller *c)
1600 struct netdev *netdev;
1601 struct in_addr mask, gateway;
1603 struct iface *local_iface;
1606 /* Controller discovery does its own TCP/IP configuration later. */
1607 if (strcmp(c->target, "discover")) {
1611 /* If there's no local interface or no IP address, give up. */
1612 local_iface = bridge_get_local_iface(br);
1613 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1617 /* Bring up the local interface. */
1618 netdev = local_iface->netdev;
1619 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1621 /* Configure the IP address and netmask. */
1622 if (!c->local_netmask
1623 || !inet_aton(c->local_netmask, &mask)
1625 mask.s_addr = guess_netmask(ip.s_addr);
1627 if (!netdev_set_in4(netdev, ip, mask)) {
1628 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1629 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1632 /* Configure the default gateway. */
1633 if (c->local_gateway
1634 && inet_aton(c->local_gateway, &gateway)
1635 && gateway.s_addr) {
1636 if (!netdev_add_router(netdev, gateway)) {
1637 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1638 br->name, IP_ARGS(&gateway.s_addr));
1644 bridge_reconfigure_remotes(struct bridge *br,
1645 const struct sockaddr_in *managers,
1648 struct ovsrec_controller **controllers;
1649 size_t n_controllers;
1652 struct ofproto_controller *ocs;
1656 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1657 had_primary = ofproto_has_primary_controller(br->ofproto);
1659 n_controllers = bridge_get_controllers(br, &controllers);
1661 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1664 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1665 for (i = 0; i < n_controllers; i++) {
1666 struct ovsrec_controller *c = controllers[i];
1668 if (!strncmp(c->target, "punix:", 6)
1669 || !strncmp(c->target, "unix:", 5)) {
1670 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1672 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1673 * domain sockets and overwriting arbitrary local files. */
1674 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1675 "\"%s\" due to possibility for remote exploit",
1676 dpif_name(br->dpif), c->target);
1680 bridge_configure_local_iface_netdev(br, c);
1681 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs++]);
1684 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1685 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1688 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1689 ofproto_flush_flows(br->ofproto);
1692 /* If there are no controllers and the bridge is in standalone
1693 * mode, set up a flow that matches every packet and directs
1694 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1695 * switch is in secure mode and we won't pass any traffic until
1696 * a controller has been defined and it tells us to do so. */
1698 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1699 union ofp_action action;
1702 memset(&action, 0, sizeof action);
1703 action.type = htons(OFPAT_OUTPUT);
1704 action.output.len = htons(sizeof action);
1705 action.output.port = htons(OFPP_NORMAL);
1706 memset(&flow, 0, sizeof flow);
1707 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1712 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1717 for (i = 0; i < br->n_ports; i++) {
1718 struct port *port = br->ports[i];
1719 for (j = 0; j < port->n_ifaces; j++) {
1720 struct iface *iface = port->ifaces[j];
1721 shash_add_once(ifaces, iface->name, iface);
1723 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1724 shash_add_once(ifaces, port->name, NULL);
1729 /* For robustness, in case the administrator moves around datapath ports behind
1730 * our back, we re-check all the datapath port numbers here.
1732 * This function will set the 'dp_ifidx' members of interfaces that have
1733 * disappeared to -1, so only call this function from a context where those
1734 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1735 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1736 * datapath, which doesn't support UINT16_MAX+1 ports. */
1738 bridge_fetch_dp_ifaces(struct bridge *br)
1740 struct odp_port *dpif_ports;
1741 size_t n_dpif_ports;
1744 /* Reset all interface numbers. */
1745 for (i = 0; i < br->n_ports; i++) {
1746 struct port *port = br->ports[i];
1747 for (j = 0; j < port->n_ifaces; j++) {
1748 struct iface *iface = port->ifaces[j];
1749 iface->dp_ifidx = -1;
1752 hmap_clear(&br->ifaces);
1754 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1755 for (i = 0; i < n_dpif_ports; i++) {
1756 struct odp_port *p = &dpif_ports[i];
1757 struct iface *iface = iface_lookup(br, p->devname);
1759 if (iface->dp_ifidx >= 0) {
1760 VLOG_WARN("%s reported interface %s twice",
1761 dpif_name(br->dpif), p->devname);
1762 } else if (iface_from_dp_ifidx(br, p->port)) {
1763 VLOG_WARN("%s reported interface %"PRIu16" twice",
1764 dpif_name(br->dpif), p->port);
1766 iface->dp_ifidx = p->port;
1767 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
1768 hash_int(iface->dp_ifidx, 0));
1772 int64_t ofport = (iface->dp_ifidx >= 0
1773 ? odp_port_to_ofp_port(iface->dp_ifidx)
1775 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1782 /* Bridge packet processing functions. */
1785 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1787 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1790 static struct bond_entry *
1791 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1793 return &port->bond_hash[bond_hash(mac)];
1797 bond_choose_iface(const struct port *port)
1799 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1800 size_t i, best_down_slave = -1;
1801 long long next_delay_expiration = LLONG_MAX;
1803 for (i = 0; i < port->n_ifaces; i++) {
1804 struct iface *iface = port->ifaces[i];
1806 if (iface->enabled) {
1808 } else if (iface->delay_expires < next_delay_expiration) {
1809 best_down_slave = i;
1810 next_delay_expiration = iface->delay_expires;
1814 if (best_down_slave != -1) {
1815 struct iface *iface = port->ifaces[best_down_slave];
1817 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1818 "since no other interface is up", iface->name,
1819 iface->delay_expires - time_msec());
1820 bond_enable_slave(iface, true);
1823 return best_down_slave;
1827 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1828 uint16_t *dp_ifidx, tag_type *tags)
1830 struct iface *iface;
1832 assert(port->n_ifaces);
1833 if (port->n_ifaces == 1) {
1834 iface = port->ifaces[0];
1836 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1837 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1838 || !port->ifaces[e->iface_idx]->enabled) {
1839 /* XXX select interface properly. The current interface selection
1840 * is only good for testing the rebalancing code. */
1841 e->iface_idx = bond_choose_iface(port);
1842 if (e->iface_idx < 0) {
1843 *tags |= port->no_ifaces_tag;
1846 e->iface_tag = tag_create_random();
1847 ((struct port *) port)->bond_compat_is_stale = true;
1849 *tags |= e->iface_tag;
1850 iface = port->ifaces[e->iface_idx];
1852 *dp_ifidx = iface->dp_ifidx;
1853 *tags |= iface->tag; /* Currently only used for bonding. */
1858 bond_link_status_update(struct iface *iface, bool carrier)
1860 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1861 struct port *port = iface->port;
1863 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1864 /* Nothing to do. */
1867 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1868 iface->name, carrier ? "detected" : "dropped");
1869 if (carrier == iface->enabled) {
1870 iface->delay_expires = LLONG_MAX;
1871 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1872 iface->name, carrier ? "disabled" : "enabled");
1873 } else if (carrier && port->active_iface < 0) {
1874 bond_enable_slave(iface, true);
1875 if (port->updelay) {
1876 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1877 "other interface is up", iface->name, port->updelay);
1880 int delay = carrier ? port->updelay : port->downdelay;
1881 iface->delay_expires = time_msec() + delay;
1884 "interface %s: will be %s if it stays %s for %d ms",
1886 carrier ? "enabled" : "disabled",
1887 carrier ? "up" : "down",
1894 bond_choose_active_iface(struct port *port)
1896 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1898 port->active_iface = bond_choose_iface(port);
1899 port->active_iface_tag = tag_create_random();
1900 if (port->active_iface >= 0) {
1901 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1902 port->name, port->ifaces[port->active_iface]->name);
1904 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1910 bond_enable_slave(struct iface *iface, bool enable)
1912 struct port *port = iface->port;
1913 struct bridge *br = port->bridge;
1915 /* This acts as a recursion check. If the act of disabling a slave
1916 * causes a different slave to be enabled, the flag will allow us to
1917 * skip redundant work when we reenter this function. It must be
1918 * cleared on exit to keep things safe with multiple bonds. */
1919 static bool moving_active_iface = false;
1921 iface->delay_expires = LLONG_MAX;
1922 if (enable == iface->enabled) {
1926 iface->enabled = enable;
1927 if (!iface->enabled) {
1928 VLOG_WARN("interface %s: disabled", iface->name);
1929 ofproto_revalidate(br->ofproto, iface->tag);
1930 if (iface->port_ifidx == port->active_iface) {
1931 ofproto_revalidate(br->ofproto,
1932 port->active_iface_tag);
1934 /* Disabling a slave can lead to another slave being immediately
1935 * enabled if there will be no active slaves but one is waiting
1936 * on an updelay. In this case we do not need to run most of the
1937 * code for the newly enabled slave since there was no period
1938 * without an active slave and it is redundant with the disabling
1940 moving_active_iface = true;
1941 bond_choose_active_iface(port);
1943 bond_send_learning_packets(port);
1945 VLOG_WARN("interface %s: enabled", iface->name);
1946 if (port->active_iface < 0 && !moving_active_iface) {
1947 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1948 bond_choose_active_iface(port);
1949 bond_send_learning_packets(port);
1951 iface->tag = tag_create_random();
1954 moving_active_iface = false;
1955 port->bond_compat_is_stale = true;
1958 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1959 * bond interface. */
1961 bond_update_fake_iface_stats(struct port *port)
1963 struct netdev_stats bond_stats;
1964 struct netdev *bond_dev;
1967 memset(&bond_stats, 0, sizeof bond_stats);
1969 for (i = 0; i < port->n_ifaces; i++) {
1970 struct netdev_stats slave_stats;
1972 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1973 /* XXX: We swap the stats here because they are swapped back when
1974 * reported by the internal device. The reason for this is
1975 * internal devices normally represent packets going into the system
1976 * but when used as fake bond device they represent packets leaving
1977 * the system. We really should do this in the internal device
1978 * itself because changing it here reverses the counts from the
1979 * perspective of the switch. However, the internal device doesn't
1980 * know what type of device it represents so we have to do it here
1982 bond_stats.tx_packets += slave_stats.rx_packets;
1983 bond_stats.tx_bytes += slave_stats.rx_bytes;
1984 bond_stats.rx_packets += slave_stats.tx_packets;
1985 bond_stats.rx_bytes += slave_stats.tx_bytes;
1989 if (!netdev_open_default(port->name, &bond_dev)) {
1990 netdev_set_stats(bond_dev, &bond_stats);
1991 netdev_close(bond_dev);
1996 bond_run(struct bridge *br)
2000 for (i = 0; i < br->n_ports; i++) {
2001 struct port *port = br->ports[i];
2003 if (port->n_ifaces >= 2) {
2004 for (j = 0; j < port->n_ifaces; j++) {
2005 struct iface *iface = port->ifaces[j];
2006 if (time_msec() >= iface->delay_expires) {
2007 bond_enable_slave(iface, !iface->enabled);
2011 if (port->bond_fake_iface
2012 && time_msec() >= port->bond_next_fake_iface_update) {
2013 bond_update_fake_iface_stats(port);
2014 port->bond_next_fake_iface_update = time_msec() + 1000;
2018 if (port->bond_compat_is_stale) {
2019 port->bond_compat_is_stale = false;
2020 port_update_bond_compat(port);
2026 bond_wait(struct bridge *br)
2030 for (i = 0; i < br->n_ports; i++) {
2031 struct port *port = br->ports[i];
2032 if (port->n_ifaces < 2) {
2035 for (j = 0; j < port->n_ifaces; j++) {
2036 struct iface *iface = port->ifaces[j];
2037 if (iface->delay_expires != LLONG_MAX) {
2038 poll_timer_wait_until(iface->delay_expires);
2041 if (port->bond_fake_iface) {
2042 poll_timer_wait_until(port->bond_next_fake_iface_update);
2048 set_dst(struct dst *p, const flow_t *flow,
2049 const struct port *in_port, const struct port *out_port,
2052 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2053 : in_port->vlan >= 0 ? in_port->vlan
2054 : ntohs(flow->dl_vlan));
2055 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
2059 swap_dst(struct dst *p, struct dst *q)
2061 struct dst tmp = *p;
2066 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2067 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2068 * that we push to the datapath. We could in fact fully sort the array by
2069 * vlan, but in most cases there are at most two different vlan tags so that's
2070 * possibly overkill.) */
2072 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2074 struct dst *first = dsts;
2075 struct dst *last = dsts + n_dsts;
2077 while (first != last) {
2079 * - All dsts < first have vlan == 'vlan'.
2080 * - All dsts >= last have vlan != 'vlan'.
2081 * - first < last. */
2082 while (first->vlan == vlan) {
2083 if (++first == last) {
2088 /* Same invariants, plus one additional:
2089 * - first->vlan != vlan.
2091 while (last[-1].vlan != vlan) {
2092 if (--last == first) {
2097 /* Same invariants, plus one additional:
2098 * - last[-1].vlan == vlan.*/
2099 swap_dst(first++, --last);
2104 mirror_mask_ffs(mirror_mask_t mask)
2106 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2111 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2112 const struct dst *test)
2115 for (i = 0; i < n_dsts; i++) {
2116 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2124 port_trunks_vlan(const struct port *port, uint16_t vlan)
2126 return (port->vlan < 0
2127 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2131 port_includes_vlan(const struct port *port, uint16_t vlan)
2133 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2137 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2138 const struct port *in_port, const struct port *out_port,
2139 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2141 mirror_mask_t mirrors = in_port->src_mirrors;
2142 struct dst *dst = dsts;
2145 if (out_port == FLOOD_PORT) {
2146 /* XXX use ODP_FLOOD if no vlans or bonding. */
2147 /* XXX even better, define each VLAN as a datapath port group */
2148 for (i = 0; i < br->n_ports; i++) {
2149 struct port *port = br->ports[i];
2150 if (port != in_port && port_includes_vlan(port, vlan)
2151 && !port->is_mirror_output_port
2152 && set_dst(dst, flow, in_port, port, tags)) {
2153 mirrors |= port->dst_mirrors;
2157 *nf_output_iface = NF_OUT_FLOOD;
2158 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2159 *nf_output_iface = dst->dp_ifidx;
2160 mirrors |= out_port->dst_mirrors;
2165 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2166 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2168 if (set_dst(dst, flow, in_port, m->out_port, tags)
2169 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2173 for (i = 0; i < br->n_ports; i++) {
2174 struct port *port = br->ports[i];
2175 if (port_includes_vlan(port, m->out_vlan)
2176 && set_dst(dst, flow, in_port, port, tags))
2180 if (port->vlan < 0) {
2181 dst->vlan = m->out_vlan;
2183 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2187 /* Use the vlan tag on the original flow instead of
2188 * the one passed in the vlan parameter. This ensures
2189 * that we compare the vlan from before any implicit
2190 * tagging tags place. This is necessary because
2191 * dst->vlan is the final vlan, after removing implicit
2193 flow_vlan = ntohs(flow->dl_vlan);
2194 if (flow_vlan == 0) {
2195 flow_vlan = OFP_VLAN_NONE;
2197 if (port == in_port && dst->vlan == flow_vlan) {
2198 /* Don't send out input port on same VLAN. */
2206 mirrors &= mirrors - 1;
2209 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2213 static void OVS_UNUSED
2214 print_dsts(const struct dst *dsts, size_t n)
2216 for (; n--; dsts++) {
2217 printf(">p%"PRIu16, dsts->dp_ifidx);
2218 if (dsts->vlan != OFP_VLAN_NONE) {
2219 printf("v%"PRIu16, dsts->vlan);
2225 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2226 const struct port *in_port, const struct port *out_port,
2227 tag_type *tags, struct odp_actions *actions,
2228 uint16_t *nf_output_iface)
2230 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2232 const struct dst *p;
2235 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2238 cur_vlan = ntohs(flow->dl_vlan);
2239 for (p = dsts; p < &dsts[n_dsts]; p++) {
2240 union odp_action *a;
2241 if (p->vlan != cur_vlan) {
2242 if (p->vlan == OFP_VLAN_NONE) {
2243 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2245 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2246 a->vlan_vid.vlan_vid = htons(p->vlan);
2250 a = odp_actions_add(actions, ODPAT_OUTPUT);
2251 a->output.port = p->dp_ifidx;
2255 /* Returns the effective vlan of a packet, taking into account both the
2256 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2257 * the packet is untagged and -1 indicates it has an invalid header and
2258 * should be dropped. */
2259 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2260 struct port *in_port, bool have_packet)
2262 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2263 * belongs to VLAN 0, so we should treat both cases identically. (In the
2264 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2265 * presumably to allow a priority to be specified. In the latter case, the
2266 * packet does not have any 802.1Q header.) */
2267 int vlan = ntohs(flow->dl_vlan);
2268 if (vlan == OFP_VLAN_NONE) {
2271 if (in_port->vlan >= 0) {
2273 /* XXX support double tagging? */
2275 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2276 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2277 "packet received on port %s configured with "
2278 "implicit VLAN %"PRIu16,
2279 br->name, ntohs(flow->dl_vlan),
2280 in_port->name, in_port->vlan);
2284 vlan = in_port->vlan;
2286 if (!port_includes_vlan(in_port, vlan)) {
2288 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2289 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2290 "packet received on port %s not configured for "
2292 br->name, vlan, in_port->name, vlan);
2301 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2302 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2303 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2305 is_gratuitous_arp(const flow_t *flow)
2307 return (flow->dl_type == htons(ETH_TYPE_ARP)
2308 && eth_addr_is_broadcast(flow->dl_dst)
2309 && (flow->nw_proto == ARP_OP_REPLY
2310 || (flow->nw_proto == ARP_OP_REQUEST
2311 && flow->nw_src == flow->nw_dst)));
2315 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2316 struct port *in_port)
2318 enum grat_arp_lock_type lock_type;
2321 /* We don't want to learn from gratuitous ARP packets that are reflected
2322 * back over bond slaves so we lock the learning table. */
2323 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2324 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2325 GRAT_ARP_LOCK_CHECK;
2327 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2330 /* The log messages here could actually be useful in debugging,
2331 * so keep the rate limit relatively high. */
2332 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2334 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2335 "on port %s in VLAN %d",
2336 br->name, ETH_ADDR_ARGS(flow->dl_src),
2337 in_port->name, vlan);
2338 ofproto_revalidate(br->ofproto, rev_tag);
2342 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2343 * dropped. Returns true if they may be forwarded, false if they should be
2346 * If 'have_packet' is true, it indicates that the caller is processing a
2347 * received packet. If 'have_packet' is false, then the caller is just
2348 * revalidating an existing flow because configuration has changed. Either
2349 * way, 'have_packet' only affects logging (there is no point in logging errors
2350 * during revalidation).
2352 * Sets '*in_portp' to the input port. This will be a null pointer if
2353 * flow->in_port does not designate a known input port (in which case
2354 * is_admissible() returns false).
2356 * When returning true, sets '*vlanp' to the effective VLAN of the input
2357 * packet, as returned by flow_get_vlan().
2359 * May also add tags to '*tags', although the current implementation only does
2360 * so in one special case.
2363 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2364 tag_type *tags, int *vlanp, struct port **in_portp)
2366 struct iface *in_iface;
2367 struct port *in_port;
2370 /* Find the interface and port structure for the received packet. */
2371 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2373 /* No interface? Something fishy... */
2375 /* Odd. A few possible reasons here:
2377 * - We deleted an interface but there are still a few packets
2378 * queued up from it.
2380 * - Someone externally added an interface (e.g. with "ovs-dpctl
2381 * add-if") that we don't know about.
2383 * - Packet arrived on the local port but the local port is not
2384 * one of our bridge ports.
2386 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2388 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2389 "interface %"PRIu16, br->name, flow->in_port);
2395 *in_portp = in_port = in_iface->port;
2396 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2401 /* Drop frames for reserved multicast addresses. */
2402 if (eth_addr_is_reserved(flow->dl_dst)) {
2406 /* Drop frames on ports reserved for mirroring. */
2407 if (in_port->is_mirror_output_port) {
2409 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2410 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2411 "%s, which is reserved exclusively for mirroring",
2412 br->name, in_port->name);
2417 /* Packets received on bonds need special attention to avoid duplicates. */
2418 if (in_port->n_ifaces > 1) {
2420 bool is_grat_arp_locked;
2422 if (eth_addr_is_multicast(flow->dl_dst)) {
2423 *tags |= in_port->active_iface_tag;
2424 if (in_port->active_iface != in_iface->port_ifidx) {
2425 /* Drop all multicast packets on inactive slaves. */
2430 /* Drop all packets for which we have learned a different input
2431 * port, because we probably sent the packet on one slave and got
2432 * it back on the other. Gratuitous ARP packets are an exception
2433 * to this rule: the host has moved to another switch. The exception
2434 * to the exception is if we locked the learning table to avoid
2435 * reflections on bond slaves. If this is the case, just drop the
2437 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2438 &is_grat_arp_locked);
2439 if (src_idx != -1 && src_idx != in_port->port_idx &&
2440 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2448 /* If the composed actions may be applied to any packet in the given 'flow',
2449 * returns true. Otherwise, the actions should only be applied to 'packet', or
2450 * not at all, if 'packet' was NULL. */
2452 process_flow(struct bridge *br, const flow_t *flow,
2453 const struct ofpbuf *packet, struct odp_actions *actions,
2454 tag_type *tags, uint16_t *nf_output_iface)
2456 struct port *in_port;
2457 struct port *out_port;
2461 /* Check whether we should drop packets in this flow. */
2462 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2467 /* Learn source MAC (but don't try to learn from revalidation). */
2469 update_learning_table(br, flow, vlan, in_port);
2472 /* Determine output port. */
2473 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2475 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2476 out_port = br->ports[out_port_idx];
2477 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2478 /* If we are revalidating but don't have a learning entry then
2479 * eject the flow. Installing a flow that floods packets opens
2480 * up a window of time where we could learn from a packet reflected
2481 * on a bond and blackhole packets before the learning table is
2482 * updated to reflect the correct port. */
2485 out_port = FLOOD_PORT;
2488 /* Don't send packets out their input ports. */
2489 if (in_port == out_port) {
2495 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2502 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2505 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2506 const struct ofp_phy_port *opp,
2509 struct bridge *br = br_;
2510 struct iface *iface;
2513 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2519 if (reason == OFPPR_DELETE) {
2520 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2521 br->name, iface->name);
2522 iface_destroy(iface);
2523 if (!port->n_ifaces) {
2524 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2525 br->name, port->name);
2531 if (port->n_ifaces > 1) {
2532 bool up = !(opp->state & OFPPS_LINK_DOWN);
2533 bond_link_status_update(iface, up);
2534 port_update_bond_compat(port);
2540 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2541 struct odp_actions *actions, tag_type *tags,
2542 uint16_t *nf_output_iface, void *br_)
2544 struct bridge *br = br_;
2546 COVERAGE_INC(bridge_process_flow);
2548 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2552 bridge_account_flow_ofhook_cb(const flow_t *flow, tag_type tags,
2553 const union odp_action *actions,
2554 size_t n_actions, unsigned long long int n_bytes,
2557 struct bridge *br = br_;
2558 const union odp_action *a;
2559 struct port *in_port;
2563 /* Feed information from the active flows back into the learning table to
2564 * ensure that table is always in sync with what is actually flowing
2565 * through the datapath.
2567 * We test that 'tags' is nonzero to ensure that only flows that include an
2568 * OFPP_NORMAL action are used for learning. This works because
2569 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2570 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2571 update_learning_table(br, flow, vlan, in_port);
2574 /* Account for bond slave utilization. */
2575 if (!br->has_bonded_ports) {
2578 for (a = actions; a < &actions[n_actions]; a++) {
2579 if (a->type == ODPAT_OUTPUT) {
2580 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2581 if (out_port && out_port->n_ifaces >= 2) {
2582 struct bond_entry *e = lookup_bond_entry(out_port,
2584 e->tx_bytes += n_bytes;
2591 bridge_account_checkpoint_ofhook_cb(void *br_)
2593 struct bridge *br = br_;
2597 if (!br->has_bonded_ports) {
2602 for (i = 0; i < br->n_ports; i++) {
2603 struct port *port = br->ports[i];
2604 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2605 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2606 bond_rebalance_port(port);
2611 static struct ofhooks bridge_ofhooks = {
2612 bridge_port_changed_ofhook_cb,
2613 bridge_normal_ofhook_cb,
2614 bridge_account_flow_ofhook_cb,
2615 bridge_account_checkpoint_ofhook_cb,
2618 /* Bonding functions. */
2620 /* Statistics for a single interface on a bonded port, used for load-based
2621 * bond rebalancing. */
2622 struct slave_balance {
2623 struct iface *iface; /* The interface. */
2624 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2626 /* All the "bond_entry"s that are assigned to this interface, in order of
2627 * increasing tx_bytes. */
2628 struct bond_entry **hashes;
2632 /* Sorts pointers to pointers to bond_entries in ascending order by the
2633 * interface to which they are assigned, and within a single interface in
2634 * ascending order of bytes transmitted. */
2636 compare_bond_entries(const void *a_, const void *b_)
2638 const struct bond_entry *const *ap = a_;
2639 const struct bond_entry *const *bp = b_;
2640 const struct bond_entry *a = *ap;
2641 const struct bond_entry *b = *bp;
2642 if (a->iface_idx != b->iface_idx) {
2643 return a->iface_idx > b->iface_idx ? 1 : -1;
2644 } else if (a->tx_bytes != b->tx_bytes) {
2645 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2651 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2652 * *descending* order by number of bytes transmitted. */
2654 compare_slave_balance(const void *a_, const void *b_)
2656 const struct slave_balance *a = a_;
2657 const struct slave_balance *b = b_;
2658 if (a->iface->enabled != b->iface->enabled) {
2659 return a->iface->enabled ? -1 : 1;
2660 } else if (a->tx_bytes != b->tx_bytes) {
2661 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2668 swap_bals(struct slave_balance *a, struct slave_balance *b)
2670 struct slave_balance tmp = *a;
2675 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2676 * given that 'p' (and only 'p') might be in the wrong location.
2678 * This function invalidates 'p', since it might now be in a different memory
2681 resort_bals(struct slave_balance *p,
2682 struct slave_balance bals[], size_t n_bals)
2685 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2686 swap_bals(p, p - 1);
2688 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2689 swap_bals(p, p + 1);
2695 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2697 if (VLOG_IS_DBG_ENABLED()) {
2698 struct ds ds = DS_EMPTY_INITIALIZER;
2699 const struct slave_balance *b;
2701 for (b = bals; b < bals + n_bals; b++) {
2705 ds_put_char(&ds, ',');
2707 ds_put_format(&ds, " %s %"PRIu64"kB",
2708 b->iface->name, b->tx_bytes / 1024);
2710 if (!b->iface->enabled) {
2711 ds_put_cstr(&ds, " (disabled)");
2713 if (b->n_hashes > 0) {
2714 ds_put_cstr(&ds, " (");
2715 for (i = 0; i < b->n_hashes; i++) {
2716 const struct bond_entry *e = b->hashes[i];
2718 ds_put_cstr(&ds, " + ");
2720 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2721 e - port->bond_hash, e->tx_bytes / 1024);
2723 ds_put_cstr(&ds, ")");
2726 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2731 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2733 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2736 struct bond_entry *hash = from->hashes[hash_idx];
2737 struct port *port = from->iface->port;
2738 uint64_t delta = hash->tx_bytes;
2740 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2741 "from %s to %s (now carrying %"PRIu64"kB and "
2742 "%"PRIu64"kB load, respectively)",
2743 port->name, delta / 1024, hash - port->bond_hash,
2744 from->iface->name, to->iface->name,
2745 (from->tx_bytes - delta) / 1024,
2746 (to->tx_bytes + delta) / 1024);
2748 /* Delete element from from->hashes.
2750 * We don't bother to add the element to to->hashes because not only would
2751 * it require more work, the only purpose it would be to allow that hash to
2752 * be migrated to another slave in this rebalancing run, and there is no
2753 * point in doing that. */
2754 if (hash_idx == 0) {
2757 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2758 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2762 /* Shift load away from 'from' to 'to'. */
2763 from->tx_bytes -= delta;
2764 to->tx_bytes += delta;
2766 /* Arrange for flows to be revalidated. */
2767 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2768 hash->iface_idx = to->iface->port_ifidx;
2769 hash->iface_tag = tag_create_random();
2773 bond_rebalance_port(struct port *port)
2775 struct slave_balance bals[DP_MAX_PORTS];
2777 struct bond_entry *hashes[BOND_MASK + 1];
2778 struct slave_balance *b, *from, *to;
2779 struct bond_entry *e;
2782 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2783 * descending order of tx_bytes, so that bals[0] represents the most
2784 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2787 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2788 * array for each slave_balance structure, we sort our local array of
2789 * hashes in order by slave, so that all of the hashes for a given slave
2790 * become contiguous in memory, and then we point each 'hashes' members of
2791 * a slave_balance structure to the start of a contiguous group. */
2792 n_bals = port->n_ifaces;
2793 for (b = bals; b < &bals[n_bals]; b++) {
2794 b->iface = port->ifaces[b - bals];
2799 for (i = 0; i <= BOND_MASK; i++) {
2800 hashes[i] = &port->bond_hash[i];
2802 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2803 for (i = 0; i <= BOND_MASK; i++) {
2805 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2806 b = &bals[e->iface_idx];
2807 b->tx_bytes += e->tx_bytes;
2809 b->hashes = &hashes[i];
2814 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2815 log_bals(bals, n_bals, port);
2817 /* Discard slaves that aren't enabled (which were sorted to the back of the
2818 * array earlier). */
2819 while (!bals[n_bals - 1].iface->enabled) {
2826 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2827 to = &bals[n_bals - 1];
2828 for (from = bals; from < to; ) {
2829 uint64_t overload = from->tx_bytes - to->tx_bytes;
2830 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2831 /* The extra load on 'from' (and all less-loaded slaves), compared
2832 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2833 * it is less than ~1Mbps. No point in rebalancing. */
2835 } else if (from->n_hashes == 1) {
2836 /* 'from' only carries a single MAC hash, so we can't shift any
2837 * load away from it, even though we want to. */
2840 /* 'from' is carrying significantly more load than 'to', and that
2841 * load is split across at least two different hashes. Pick a hash
2842 * to migrate to 'to' (the least-loaded slave), given that doing so
2843 * must decrease the ratio of the load on the two slaves by at
2846 * The sort order we use means that we prefer to shift away the
2847 * smallest hashes instead of the biggest ones. There is little
2848 * reason behind this decision; we could use the opposite sort
2849 * order to shift away big hashes ahead of small ones. */
2852 for (i = 0; i < from->n_hashes; i++) {
2853 double old_ratio, new_ratio;
2854 uint64_t delta = from->hashes[i]->tx_bytes;
2856 if (delta == 0 || from->tx_bytes - delta == 0) {
2857 /* Pointless move. */
2861 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2863 if (to->tx_bytes == 0) {
2864 /* Nothing on the new slave, move it. */
2868 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2869 new_ratio = (double)(from->tx_bytes - delta) /
2870 (to->tx_bytes + delta);
2872 if (new_ratio == 0) {
2873 /* Should already be covered but check to prevent division
2878 if (new_ratio < 1) {
2879 new_ratio = 1 / new_ratio;
2882 if (old_ratio - new_ratio > 0.1) {
2883 /* Would decrease the ratio, move it. */
2887 if (i < from->n_hashes) {
2888 bond_shift_load(from, to, i);
2889 port->bond_compat_is_stale = true;
2891 /* If the result of the migration changed the relative order of
2892 * 'from' and 'to' swap them back to maintain invariants. */
2893 if (order_swapped) {
2894 swap_bals(from, to);
2897 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2898 * point to different slave_balance structures. It is only
2899 * valid to do these two operations in a row at all because we
2900 * know that 'from' will not move past 'to' and vice versa. */
2901 resort_bals(from, bals, n_bals);
2902 resort_bals(to, bals, n_bals);
2909 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2910 * historical data to decay to <1% in 7 rebalancing runs. */
2911 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2917 bond_send_learning_packets(struct port *port)
2919 struct bridge *br = port->bridge;
2920 struct mac_entry *e;
2921 struct ofpbuf packet;
2922 int error, n_packets, n_errors;
2924 if (!port->n_ifaces || port->active_iface < 0) {
2928 ofpbuf_init(&packet, 128);
2929 error = n_packets = n_errors = 0;
2930 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
2931 union ofp_action actions[2], *a;
2937 if (e->port == port->port_idx
2938 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2942 /* Compose actions. */
2943 memset(actions, 0, sizeof actions);
2946 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2947 a->vlan_vid.len = htons(sizeof *a);
2948 a->vlan_vid.vlan_vid = htons(e->vlan);
2951 a->output.type = htons(OFPAT_OUTPUT);
2952 a->output.len = htons(sizeof *a);
2953 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2958 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2960 flow_extract(&packet, 0, ODPP_NONE, &flow);
2961 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2968 ofpbuf_uninit(&packet);
2971 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2972 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2973 "packets, last error was: %s",
2974 port->name, n_errors, n_packets, strerror(error));
2976 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2977 port->name, n_packets);
2981 /* Bonding unixctl user interface functions. */
2984 bond_unixctl_list(struct unixctl_conn *conn,
2985 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2987 struct ds ds = DS_EMPTY_INITIALIZER;
2988 const struct bridge *br;
2990 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2992 LIST_FOR_EACH (br, node, &all_bridges) {
2995 for (i = 0; i < br->n_ports; i++) {
2996 const struct port *port = br->ports[i];
2997 if (port->n_ifaces > 1) {
3000 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
3001 for (j = 0; j < port->n_ifaces; j++) {
3002 const struct iface *iface = port->ifaces[j];
3004 ds_put_cstr(&ds, ", ");
3006 ds_put_cstr(&ds, iface->name);
3008 ds_put_char(&ds, '\n');
3012 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3016 static struct port *
3017 bond_find(const char *name)
3019 const struct bridge *br;
3021 LIST_FOR_EACH (br, node, &all_bridges) {
3024 for (i = 0; i < br->n_ports; i++) {
3025 struct port *port = br->ports[i];
3026 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3035 bond_unixctl_show(struct unixctl_conn *conn,
3036 const char *args, void *aux OVS_UNUSED)
3038 struct ds ds = DS_EMPTY_INITIALIZER;
3039 const struct port *port;
3042 port = bond_find(args);
3044 unixctl_command_reply(conn, 501, "no such bond");
3048 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3049 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3050 ds_put_format(&ds, "next rebalance: %lld ms\n",
3051 port->bond_next_rebalance - time_msec());
3052 for (j = 0; j < port->n_ifaces; j++) {
3053 const struct iface *iface = port->ifaces[j];
3054 struct bond_entry *be;
3057 ds_put_format(&ds, "slave %s: %s\n",
3058 iface->name, iface->enabled ? "enabled" : "disabled");
3059 if (j == port->active_iface) {
3060 ds_put_cstr(&ds, "\tactive slave\n");
3062 if (iface->delay_expires != LLONG_MAX) {
3063 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3064 iface->enabled ? "downdelay" : "updelay",
3065 iface->delay_expires - time_msec());
3069 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3070 int hash = be - port->bond_hash;
3071 struct mac_entry *me;
3073 if (be->iface_idx != j) {
3077 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3078 hash, be->tx_bytes / 1024);
3081 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3084 if (bond_hash(me->mac) == hash
3085 && me->port != port->port_idx
3086 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3087 && dp_ifidx == iface->dp_ifidx)
3089 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3090 ETH_ADDR_ARGS(me->mac));
3095 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3100 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3101 void *aux OVS_UNUSED)
3103 char *args = (char *) args_;
3104 char *save_ptr = NULL;
3105 char *bond_s, *hash_s, *slave_s;
3106 uint8_t mac[ETH_ADDR_LEN];
3108 struct iface *iface;
3109 struct bond_entry *entry;
3112 bond_s = strtok_r(args, " ", &save_ptr);
3113 hash_s = strtok_r(NULL, " ", &save_ptr);
3114 slave_s = strtok_r(NULL, " ", &save_ptr);
3116 unixctl_command_reply(conn, 501,
3117 "usage: bond/migrate BOND HASH SLAVE");
3121 port = bond_find(bond_s);
3123 unixctl_command_reply(conn, 501, "no such bond");
3127 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3128 == ETH_ADDR_SCAN_COUNT) {
3129 hash = bond_hash(mac);
3130 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3131 hash = atoi(hash_s) & BOND_MASK;
3133 unixctl_command_reply(conn, 501, "bad hash");
3137 iface = port_lookup_iface(port, slave_s);
3139 unixctl_command_reply(conn, 501, "no such slave");
3143 if (!iface->enabled) {
3144 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3148 entry = &port->bond_hash[hash];
3149 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3150 entry->iface_idx = iface->port_ifidx;
3151 entry->iface_tag = tag_create_random();
3152 port->bond_compat_is_stale = true;
3153 unixctl_command_reply(conn, 200, "migrated");
3157 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3158 void *aux OVS_UNUSED)
3160 char *args = (char *) args_;
3161 char *save_ptr = NULL;
3162 char *bond_s, *slave_s;
3164 struct iface *iface;
3166 bond_s = strtok_r(args, " ", &save_ptr);
3167 slave_s = strtok_r(NULL, " ", &save_ptr);
3169 unixctl_command_reply(conn, 501,
3170 "usage: bond/set-active-slave BOND SLAVE");
3174 port = bond_find(bond_s);
3176 unixctl_command_reply(conn, 501, "no such bond");
3180 iface = port_lookup_iface(port, slave_s);
3182 unixctl_command_reply(conn, 501, "no such slave");
3186 if (!iface->enabled) {
3187 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3191 if (port->active_iface != iface->port_ifidx) {
3192 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3193 port->active_iface = iface->port_ifidx;
3194 port->active_iface_tag = tag_create_random();
3195 VLOG_INFO("port %s: active interface is now %s",
3196 port->name, iface->name);
3197 bond_send_learning_packets(port);
3198 unixctl_command_reply(conn, 200, "done");
3200 unixctl_command_reply(conn, 200, "no change");
3205 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3207 char *args = (char *) args_;
3208 char *save_ptr = NULL;
3209 char *bond_s, *slave_s;
3211 struct iface *iface;
3213 bond_s = strtok_r(args, " ", &save_ptr);
3214 slave_s = strtok_r(NULL, " ", &save_ptr);
3216 unixctl_command_reply(conn, 501,
3217 "usage: bond/enable/disable-slave BOND SLAVE");
3221 port = bond_find(bond_s);
3223 unixctl_command_reply(conn, 501, "no such bond");
3227 iface = port_lookup_iface(port, slave_s);
3229 unixctl_command_reply(conn, 501, "no such slave");
3233 bond_enable_slave(iface, enable);
3234 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3238 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3239 void *aux OVS_UNUSED)
3241 enable_slave(conn, args, true);
3245 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3246 void *aux OVS_UNUSED)
3248 enable_slave(conn, args, false);
3252 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3253 void *aux OVS_UNUSED)
3255 uint8_t mac[ETH_ADDR_LEN];
3259 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3260 == ETH_ADDR_SCAN_COUNT) {
3261 hash = bond_hash(mac);
3263 hash_cstr = xasprintf("%u", hash);
3264 unixctl_command_reply(conn, 200, hash_cstr);
3267 unixctl_command_reply(conn, 501, "invalid mac");
3274 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3275 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3276 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3277 unixctl_command_register("bond/set-active-slave",
3278 bond_unixctl_set_active_slave, NULL);
3279 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3281 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3283 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3286 /* Port functions. */
3288 static struct port *
3289 port_create(struct bridge *br, const char *name)
3293 port = xzalloc(sizeof *port);
3295 port->port_idx = br->n_ports;
3297 port->trunks = NULL;
3298 port->name = xstrdup(name);
3299 port->active_iface = -1;
3301 if (br->n_ports >= br->allocated_ports) {
3302 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3305 br->ports[br->n_ports++] = port;
3306 shash_add_assert(&br->port_by_name, port->name, port);
3308 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3315 get_port_other_config(const struct ovsrec_port *port, const char *key,
3316 const char *default_value)
3320 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3322 return value ? value : default_value;
3326 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3328 struct shash new_ifaces;
3331 /* Collect list of new interfaces. */
3332 shash_init(&new_ifaces);
3333 for (i = 0; i < cfg->n_interfaces; i++) {
3334 const char *name = cfg->interfaces[i]->name;
3335 shash_add_once(&new_ifaces, name, NULL);
3338 /* Get rid of deleted interfaces. */
3339 for (i = 0; i < port->n_ifaces; ) {
3340 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3341 iface_destroy(port->ifaces[i]);
3347 shash_destroy(&new_ifaces);
3351 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3353 struct shash new_ifaces;
3354 long long int next_rebalance;
3355 unsigned long *trunks;
3361 /* Update settings. */
3362 port->updelay = cfg->bond_updelay;
3363 if (port->updelay < 0) {
3366 port->downdelay = cfg->bond_downdelay;
3367 if (port->downdelay < 0) {
3368 port->downdelay = 0;
3370 port->bond_rebalance_interval = atoi(
3371 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3372 if (port->bond_rebalance_interval < 1000) {
3373 port->bond_rebalance_interval = 1000;
3375 next_rebalance = time_msec() + port->bond_rebalance_interval;
3376 if (port->bond_next_rebalance > next_rebalance) {
3377 port->bond_next_rebalance = next_rebalance;
3380 /* Add new interfaces and update 'cfg' member of existing ones. */
3381 shash_init(&new_ifaces);
3382 for (i = 0; i < cfg->n_interfaces; i++) {
3383 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3384 struct iface *iface;
3386 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3387 VLOG_WARN("port %s: %s specified twice as port interface",
3388 port->name, if_cfg->name);
3392 iface = iface_lookup(port->bridge, if_cfg->name);
3394 if (iface->port != port) {
3395 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3397 port->bridge->name, if_cfg->name, iface->port->name);
3400 iface->cfg = if_cfg;
3402 iface_create(port, if_cfg);
3405 shash_destroy(&new_ifaces);
3410 if (port->n_ifaces < 2) {
3412 if (vlan >= 0 && vlan <= 4095) {
3413 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3418 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3419 * they even work as-is. But they have not been tested. */
3420 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3424 if (port->vlan != vlan) {
3426 bridge_flush(port->bridge);
3429 /* Get trunked VLANs. */
3431 if (vlan < 0 && cfg->n_trunks) {
3434 trunks = bitmap_allocate(4096);
3436 for (i = 0; i < cfg->n_trunks; i++) {
3437 int trunk = cfg->trunks[i];
3439 bitmap_set1(trunks, trunk);
3445 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3446 port->name, cfg->n_trunks);
3448 if (n_errors == cfg->n_trunks) {
3449 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3451 bitmap_free(trunks);
3454 } else if (vlan >= 0 && cfg->n_trunks) {
3455 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3459 ? port->trunks != NULL
3460 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3461 bridge_flush(port->bridge);
3463 bitmap_free(port->trunks);
3464 port->trunks = trunks;
3468 port_destroy(struct port *port)
3471 struct bridge *br = port->bridge;
3475 proc_net_compat_update_vlan(port->name, NULL, 0);
3476 proc_net_compat_update_bond(port->name, NULL);
3478 for (i = 0; i < MAX_MIRRORS; i++) {
3479 struct mirror *m = br->mirrors[i];
3480 if (m && m->out_port == port) {
3485 while (port->n_ifaces > 0) {
3486 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3489 shash_find_and_delete_assert(&br->port_by_name, port->name);
3491 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3492 del->port_idx = port->port_idx;
3495 bitmap_free(port->trunks);
3502 static struct port *
3503 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3505 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3506 return iface ? iface->port : NULL;
3509 static struct port *
3510 port_lookup(const struct bridge *br, const char *name)
3512 return shash_find_data(&br->port_by_name, name);
3515 static struct iface *
3516 port_lookup_iface(const struct port *port, const char *name)
3518 struct iface *iface = iface_lookup(port->bridge, name);
3519 return iface && iface->port == port ? iface : NULL;
3523 port_update_bonding(struct port *port)
3525 if (port->n_ifaces < 2) {
3526 /* Not a bonded port. */
3527 if (port->bond_hash) {
3528 free(port->bond_hash);
3529 port->bond_hash = NULL;
3530 port->bond_compat_is_stale = true;
3531 port->bond_fake_iface = false;
3534 if (!port->bond_hash) {
3537 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3538 for (i = 0; i <= BOND_MASK; i++) {
3539 struct bond_entry *e = &port->bond_hash[i];
3543 port->no_ifaces_tag = tag_create_random();
3544 bond_choose_active_iface(port);
3545 port->bond_next_rebalance
3546 = time_msec() + port->bond_rebalance_interval;
3548 if (port->cfg->bond_fake_iface) {
3549 port->bond_next_fake_iface_update = time_msec();
3552 port->bond_compat_is_stale = true;
3553 port->bond_fake_iface = port->cfg->bond_fake_iface;
3558 port_update_bond_compat(struct port *port)
3560 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3561 struct compat_bond bond;
3564 if (port->n_ifaces < 2) {
3565 proc_net_compat_update_bond(port->name, NULL);
3570 bond.updelay = port->updelay;
3571 bond.downdelay = port->downdelay;
3574 bond.hashes = compat_hashes;
3575 if (port->bond_hash) {
3576 const struct bond_entry *e;
3577 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3578 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3579 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3580 cbh->hash = e - port->bond_hash;
3581 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3586 bond.n_slaves = port->n_ifaces;
3587 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3588 for (i = 0; i < port->n_ifaces; i++) {
3589 struct iface *iface = port->ifaces[i];
3590 struct compat_bond_slave *slave = &bond.slaves[i];
3591 slave->name = iface->name;
3593 /* We need to make the same determination as the Linux bonding
3594 * code to determine whether a slave should be consider "up".
3595 * The Linux function bond_miimon_inspect() supports four
3596 * BOND_LINK_* states:
3598 * - BOND_LINK_UP: carrier detected, updelay has passed.
3599 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3600 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3601 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3603 * The function bond_info_show_slave() only considers BOND_LINK_UP
3604 * to be "up" and anything else to be "down".
3606 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3610 netdev_get_etheraddr(iface->netdev, slave->mac);
3613 if (port->bond_fake_iface) {
3614 struct netdev *bond_netdev;
3616 if (!netdev_open_default(port->name, &bond_netdev)) {
3618 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3620 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3622 netdev_close(bond_netdev);
3626 proc_net_compat_update_bond(port->name, &bond);
3631 port_update_vlan_compat(struct port *port)
3633 struct bridge *br = port->bridge;
3634 char *vlandev_name = NULL;
3636 if (port->vlan > 0) {
3637 /* Figure out the name that the VLAN device should actually have, if it
3638 * existed. This takes some work because the VLAN device would not
3639 * have port->name in its name; rather, it would have the trunk port's
3640 * name, and 'port' would be attached to a bridge that also had the
3641 * VLAN device one of its ports. So we need to find a trunk port that
3642 * includes port->vlan.
3644 * There might be more than one candidate. This doesn't happen on
3645 * XenServer, so if it happens we just pick the first choice in
3646 * alphabetical order instead of creating multiple VLAN devices. */
3648 for (i = 0; i < br->n_ports; i++) {
3649 struct port *p = br->ports[i];
3650 if (port_trunks_vlan(p, port->vlan)
3652 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3654 uint8_t ea[ETH_ADDR_LEN];
3655 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3656 if (!eth_addr_is_multicast(ea) &&
3657 !eth_addr_is_reserved(ea) &&
3658 !eth_addr_is_zero(ea)) {
3659 vlandev_name = p->name;
3664 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3667 /* Interface functions. */
3669 static struct iface *
3670 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3672 struct bridge *br = port->bridge;
3673 struct iface *iface;
3674 char *name = if_cfg->name;
3677 iface = xzalloc(sizeof *iface);
3679 iface->port_ifidx = port->n_ifaces;
3680 iface->name = xstrdup(name);
3681 iface->dp_ifidx = -1;
3682 iface->tag = tag_create_random();
3683 iface->delay_expires = LLONG_MAX;
3684 iface->netdev = NULL;
3685 iface->cfg = if_cfg;
3687 shash_add_assert(&br->iface_by_name, iface->name, iface);
3689 /* Attempt to create the network interface in case it doesn't exist yet. */
3690 if (!iface_is_internal(br, iface->name)) {
3691 error = set_up_iface(iface, true);
3693 VLOG_WARN("could not create iface %s: %s", iface->name,
3696 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3703 if (port->n_ifaces >= port->allocated_ifaces) {
3704 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3705 sizeof *port->ifaces);
3707 port->ifaces[port->n_ifaces++] = iface;
3708 if (port->n_ifaces > 1) {
3709 br->has_bonded_ports = true;
3712 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3720 iface_destroy(struct iface *iface)
3723 struct port *port = iface->port;
3724 struct bridge *br = port->bridge;
3725 bool del_active = port->active_iface == iface->port_ifidx;
3728 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3730 if (iface->dp_ifidx >= 0) {
3731 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
3734 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3735 del->port_ifidx = iface->port_ifidx;
3737 netdev_close(iface->netdev);
3740 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3741 bond_choose_active_iface(port);
3742 bond_send_learning_packets(port);
3748 bridge_flush(port->bridge);
3752 static struct iface *
3753 iface_lookup(const struct bridge *br, const char *name)
3755 return shash_find_data(&br->iface_by_name, name);
3758 static struct iface *
3759 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3761 struct iface *iface;
3763 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
3764 hash_int(dp_ifidx, 0), &br->ifaces) {
3765 if (iface->dp_ifidx == dp_ifidx) {
3772 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3773 * 'br', that is, an interface that is entirely simulated within the datapath.
3774 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3775 * interfaces are created by setting "iface.<iface>.internal = true".
3777 * In addition, we have a kluge-y feature that creates an internal port with
3778 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3779 * This feature needs to go away in the long term. Until then, this is one
3780 * reason why this function takes a name instead of a struct iface: the fake
3781 * interfaces created this way do not have a struct iface. */
3783 iface_is_internal(const struct bridge *br, const char *if_name)
3785 struct iface *iface;
3788 if (!strcmp(if_name, br->name)) {
3792 iface = iface_lookup(br, if_name);
3793 if (iface && !strcmp(iface->cfg->type, "internal")) {
3797 port = port_lookup(br, if_name);
3798 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3804 /* Set Ethernet address of 'iface', if one is specified in the configuration
3807 iface_set_mac(struct iface *iface)
3809 uint8_t ea[ETH_ADDR_LEN];
3811 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3812 if (eth_addr_is_multicast(ea)) {
3813 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3815 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3816 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3817 iface->name, iface->name);
3819 int error = netdev_set_etheraddr(iface->netdev, ea);
3821 VLOG_ERR("interface %s: setting MAC failed (%s)",
3822 iface->name, strerror(error));
3829 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3830 struct shash *shash)
3835 for (i = 0; i < n; i++) {
3836 shash_add(shash, keys[i], values[i]);
3840 struct iface_delete_queues_cbdata {
3841 struct netdev *netdev;
3842 const struct ovsdb_datum *queues;
3846 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
3848 union ovsdb_atom atom;
3850 atom.integer = target;
3851 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
3855 iface_delete_queues(unsigned int queue_id,
3856 const struct shash *details OVS_UNUSED, void *cbdata_)
3858 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3860 if (!queue_ids_include(cbdata->queues, queue_id)) {
3861 netdev_delete_queue(cbdata->netdev, queue_id);
3866 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3868 if (!qos || qos->type[0] == '\0') {
3869 netdev_set_qos(iface->netdev, NULL, NULL);
3871 struct iface_delete_queues_cbdata cbdata;
3872 struct shash details;
3875 /* Configure top-level Qos for 'iface'. */
3876 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3877 qos->n_other_config, &details);
3878 netdev_set_qos(iface->netdev, qos->type, &details);
3879 shash_destroy(&details);
3881 /* Deconfigure queues that were deleted. */
3882 cbdata.netdev = iface->netdev;
3883 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
3885 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3887 /* Configure queues for 'iface'. */
3888 for (i = 0; i < qos->n_queues; i++) {
3889 const struct ovsrec_queue *queue = qos->value_queues[i];
3890 unsigned int queue_id = qos->key_queues[i];
3892 shash_from_ovs_idl_map(queue->key_other_config,
3893 queue->value_other_config,
3894 queue->n_other_config, &details);
3895 netdev_set_queue(iface->netdev, queue_id, &details);
3896 shash_destroy(&details);
3901 /* Port mirroring. */
3903 static struct mirror *
3904 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
3908 for (i = 0; i < MAX_MIRRORS; i++) {
3909 struct mirror *m = br->mirrors[i];
3910 if (m && uuid_equals(uuid, &m->uuid)) {
3918 mirror_reconfigure(struct bridge *br)
3920 unsigned long *rspan_vlans;
3923 /* Get rid of deleted mirrors. */
3924 for (i = 0; i < MAX_MIRRORS; i++) {
3925 struct mirror *m = br->mirrors[i];
3927 const struct ovsdb_datum *mc;
3928 union ovsdb_atom atom;
3930 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
3931 atom.uuid = br->mirrors[i]->uuid;
3932 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
3938 /* Add new mirrors and reconfigure existing ones. */
3939 for (i = 0; i < br->cfg->n_mirrors; i++) {
3940 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3941 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
3943 mirror_reconfigure_one(m, cfg);
3945 mirror_create(br, cfg);
3949 /* Update port reserved status. */
3950 for (i = 0; i < br->n_ports; i++) {
3951 br->ports[i]->is_mirror_output_port = false;
3953 for (i = 0; i < MAX_MIRRORS; i++) {
3954 struct mirror *m = br->mirrors[i];
3955 if (m && m->out_port) {
3956 m->out_port->is_mirror_output_port = true;
3960 /* Update flooded vlans (for RSPAN). */
3962 if (br->cfg->n_flood_vlans) {
3963 rspan_vlans = bitmap_allocate(4096);
3965 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3966 int64_t vlan = br->cfg->flood_vlans[i];
3967 if (vlan >= 0 && vlan < 4096) {
3968 bitmap_set1(rspan_vlans, vlan);
3969 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3972 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3977 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3983 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
3988 for (i = 0; ; i++) {
3989 if (i >= MAX_MIRRORS) {
3990 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3991 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
3994 if (!br->mirrors[i]) {
3999 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4002 br->mirrors[i] = m = xzalloc(sizeof *m);
4005 m->name = xstrdup(cfg->name);
4006 shash_init(&m->src_ports);
4007 shash_init(&m->dst_ports);
4013 mirror_reconfigure_one(m, cfg);
4017 mirror_destroy(struct mirror *m)
4020 struct bridge *br = m->bridge;
4023 for (i = 0; i < br->n_ports; i++) {
4024 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4025 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4028 shash_destroy(&m->src_ports);
4029 shash_destroy(&m->dst_ports);
4032 m->bridge->mirrors[m->idx] = NULL;
4041 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4042 struct shash *names)
4046 for (i = 0; i < n_ports; i++) {
4047 const char *name = ports[i]->name;
4048 if (port_lookup(m->bridge, name)) {
4049 shash_add_once(names, name, NULL);
4051 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4052 "port %s", m->bridge->name, m->name, name);
4058 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4064 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4066 for (i = 0; i < cfg->n_select_vlan; i++) {
4067 int64_t vlan = cfg->select_vlan[i];
4068 if (vlan < 0 || vlan > 4095) {
4069 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4070 m->bridge->name, m->name, vlan);
4072 (*vlans)[n_vlans++] = vlan;
4079 vlan_is_mirrored(const struct mirror *m, int vlan)
4083 for (i = 0; i < m->n_vlans; i++) {
4084 if (m->vlans[i] == vlan) {
4092 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4096 for (i = 0; i < m->n_vlans; i++) {
4097 if (port_trunks_vlan(p, m->vlans[i])) {
4105 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4107 struct shash src_ports, dst_ports;
4108 mirror_mask_t mirror_bit;
4109 struct port *out_port;
4116 if (strcmp(cfg->name, m->name)) {
4118 m->name = xstrdup(cfg->name);
4121 /* Get output port. */
4122 if (cfg->output_port) {
4123 out_port = port_lookup(m->bridge, cfg->output_port->name);
4125 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4126 m->bridge->name, m->name);
4132 if (cfg->output_vlan) {
4133 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4134 "output vlan; ignoring output vlan",
4135 m->bridge->name, m->name);
4137 } else if (cfg->output_vlan) {
4139 out_vlan = *cfg->output_vlan;
4141 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4142 m->bridge->name, m->name);
4147 shash_init(&src_ports);
4148 shash_init(&dst_ports);
4149 if (cfg->select_all) {
4150 for (i = 0; i < m->bridge->n_ports; i++) {
4151 const char *name = m->bridge->ports[i]->name;
4152 shash_add_once(&src_ports, name, NULL);
4153 shash_add_once(&dst_ports, name, NULL);
4158 /* Get ports, and drop duplicates and ports that don't exist. */
4159 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4161 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4164 /* Get all the vlans, and drop duplicate and invalid vlans. */
4165 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4168 /* Update mirror data. */
4169 if (!shash_equal_keys(&m->src_ports, &src_ports)
4170 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4171 || m->n_vlans != n_vlans
4172 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4173 || m->out_port != out_port
4174 || m->out_vlan != out_vlan) {
4175 bridge_flush(m->bridge);
4177 shash_swap(&m->src_ports, &src_ports);
4178 shash_swap(&m->dst_ports, &dst_ports);
4181 m->n_vlans = n_vlans;
4182 m->out_port = out_port;
4183 m->out_vlan = out_vlan;
4186 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4187 for (i = 0; i < m->bridge->n_ports; i++) {
4188 struct port *port = m->bridge->ports[i];
4190 if (shash_find(&m->src_ports, port->name)
4193 ? port_trunks_any_mirrored_vlan(m, port)
4194 : vlan_is_mirrored(m, port->vlan)))) {
4195 port->src_mirrors |= mirror_bit;
4197 port->src_mirrors &= ~mirror_bit;
4200 if (shash_find(&m->dst_ports, port->name)) {
4201 port->dst_mirrors |= mirror_bit;
4203 port->dst_mirrors &= ~mirror_bit;
4208 shash_destroy(&src_ports);
4209 shash_destroy(&dst_ports);