1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
23 #include <sys/socket.h>
25 #include <openflow/openflow.h>
30 #include <sys/socket.h>
31 #include <sys/types.h>
37 #include "dynamic-string.h"
42 #include "mac-learning.h"
45 #include "ofp-print.h"
47 #include "ofproto/netflow.h"
48 #include "ofproto/ofproto.h"
49 #include "ovsdb-data.h"
51 #include "poll-loop.h"
52 #include "port-array.h"
53 #include "proc-net-compat.h"
57 #include "socket-util.h"
58 #include "stream-ssl.h"
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 int dp_ifidx; /* Index within kernel datapath. */
89 struct netdev *netdev; /* Network device. */
90 bool enabled; /* May be chosen for flows? */
91 const struct ovsrec_interface *cfg;
94 #define BOND_MASK 0xff
96 int iface_idx; /* Index of assigned iface, or -1 if none. */
97 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
98 tag_type iface_tag; /* Tag associated with iface_idx. */
101 #define MAX_MIRRORS 32
102 typedef uint32_t mirror_mask_t;
103 #define MIRROR_MASK_C(X) UINT32_C(X)
104 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
106 struct bridge *bridge;
109 struct uuid uuid; /* UUID of this "mirror" record in database. */
111 /* Selection criteria. */
112 struct shash src_ports; /* Name is port name; data is always NULL. */
113 struct shash dst_ports; /* Name is port name; data is always NULL. */
118 struct port *out_port;
122 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
124 struct bridge *bridge;
126 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
127 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
128 * NULL if all VLANs are trunked. */
129 const struct ovsrec_port *cfg;
132 /* An ordinary bridge port has 1 interface.
133 * A bridge port for bonding has at least 2 interfaces. */
134 struct iface **ifaces;
135 size_t n_ifaces, allocated_ifaces;
138 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
139 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
140 tag_type active_iface_tag; /* Tag for bcast flows. */
141 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
142 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
143 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
144 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
145 long long int bond_next_fake_iface_update; /* Time of next update. */
146 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
147 long long int bond_next_rebalance; /* Next rebalancing time. */
149 /* Port mirroring info. */
150 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
151 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
152 bool is_mirror_output_port; /* Does port mirroring send frames here? */
155 #define DP_MAX_PORTS 255
157 struct list node; /* Node in global list of bridges. */
158 char *name; /* User-specified arbitrary name. */
159 struct mac_learning *ml; /* MAC learning table. */
160 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
161 const struct ovsrec_bridge *cfg;
163 /* OpenFlow switch processing. */
164 struct ofproto *ofproto; /* OpenFlow switch. */
166 /* Kernel datapath information. */
167 struct dpif *dpif; /* Datapath. */
168 struct port_array ifaces; /* Indexed by kernel datapath port number. */
172 size_t n_ports, allocated_ports;
173 struct shash iface_by_name; /* "struct iface"s indexed by name. */
174 struct shash port_by_name; /* "struct port"s indexed by name. */
177 bool has_bonded_ports;
182 /* Port mirroring. */
183 struct mirror *mirrors[MAX_MIRRORS];
186 /* List of all bridges. */
187 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
189 /* OVSDB IDL used to obtain configuration. */
190 static struct ovsdb_idl *idl;
192 /* Each time this timer expires, the bridge fetches statistics for every
193 * interface and pushes them into the database. */
194 #define IFACE_STATS_INTERVAL (5 * 1000) /* In milliseconds. */
195 static long long int iface_stats_timer = LLONG_MIN;
197 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
198 static void bridge_destroy(struct bridge *);
199 static struct bridge *bridge_lookup(const char *name);
200 static unixctl_cb_func bridge_unixctl_dump_flows;
201 static unixctl_cb_func bridge_unixctl_reconnect;
202 static int bridge_run_one(struct bridge *);
203 static size_t bridge_get_controllers(const struct bridge *br,
204 struct ovsrec_controller ***controllersp);
205 static void bridge_reconfigure_one(struct bridge *);
206 static void bridge_reconfigure_remotes(struct bridge *,
207 const struct sockaddr_in *managers,
209 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
210 static void bridge_fetch_dp_ifaces(struct bridge *);
211 static void bridge_flush(struct bridge *);
212 static void bridge_pick_local_hw_addr(struct bridge *,
213 uint8_t ea[ETH_ADDR_LEN],
214 struct iface **hw_addr_iface);
215 static uint64_t bridge_pick_datapath_id(struct bridge *,
216 const uint8_t bridge_ea[ETH_ADDR_LEN],
217 struct iface *hw_addr_iface);
218 static struct iface *bridge_get_local_iface(struct bridge *);
219 static uint64_t dpid_from_hash(const void *, size_t nbytes);
221 static unixctl_cb_func bridge_unixctl_fdb_show;
223 static void bond_init(void);
224 static void bond_run(struct bridge *);
225 static void bond_wait(struct bridge *);
226 static void bond_rebalance_port(struct port *);
227 static void bond_send_learning_packets(struct port *);
228 static void bond_enable_slave(struct iface *iface, bool enable);
230 static struct port *port_create(struct bridge *, const char *name);
231 static void port_reconfigure(struct port *, const struct ovsrec_port *);
232 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
233 static void port_destroy(struct port *);
234 static struct port *port_lookup(const struct bridge *, const char *name);
235 static struct iface *port_lookup_iface(const struct port *, const char *name);
236 static struct port *port_from_dp_ifidx(const struct bridge *,
238 static void port_update_bond_compat(struct port *);
239 static void port_update_vlan_compat(struct port *);
240 static void port_update_bonding(struct port *);
242 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
243 static void mirror_destroy(struct mirror *);
244 static void mirror_reconfigure(struct bridge *);
245 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
246 static bool vlan_is_mirrored(const struct mirror *, int vlan);
248 static struct iface *iface_create(struct port *port,
249 const struct ovsrec_interface *if_cfg);
250 static void iface_destroy(struct iface *);
251 static struct iface *iface_lookup(const struct bridge *, const char *name);
252 static struct iface *iface_from_dp_ifidx(const struct bridge *,
254 static bool iface_is_internal(const struct bridge *, const char *name);
255 static void iface_set_mac(struct iface *);
256 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
258 /* Hooks into ofproto processing. */
259 static struct ofhooks bridge_ofhooks;
261 /* Public functions. */
263 /* Initializes the bridge module, configuring it to obtain its configuration
264 * from an OVSDB server accessed over 'remote', which should be a string in a
265 * form acceptable to ovsdb_idl_create(). */
267 bridge_init(const char *remote)
269 /* Create connection to database. */
270 idl = ovsdb_idl_create(remote, &ovsrec_idl_class);
272 ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_cur_cfg);
273 ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_statistics);
274 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
276 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
278 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
279 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
281 ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_ofport);
282 ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_statistics);
283 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
285 /* Register unixctl commands. */
286 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
287 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
289 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
294 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
295 * but for which the ovs-vswitchd configuration 'cfg' is required. */
297 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
299 static bool already_configured_once;
300 struct svec bridge_names;
301 struct svec dpif_names, dpif_types;
304 /* Only do this once per ovs-vswitchd run. */
305 if (already_configured_once) {
308 already_configured_once = true;
310 iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
312 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
313 svec_init(&bridge_names);
314 for (i = 0; i < cfg->n_bridges; i++) {
315 svec_add(&bridge_names, cfg->bridges[i]->name);
317 svec_sort(&bridge_names);
319 /* Iterate over all system dpifs and delete any of them that do not appear
321 svec_init(&dpif_names);
322 svec_init(&dpif_types);
323 dp_enumerate_types(&dpif_types);
324 for (i = 0; i < dpif_types.n; i++) {
329 dp_enumerate_names(dpif_types.names[i], &dpif_names);
331 /* For each dpif... */
332 for (j = 0; j < dpif_names.n; j++) {
333 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
335 struct svec all_names;
338 /* ...check whether any of its names is in 'bridge_names'. */
339 svec_init(&all_names);
340 dpif_get_all_names(dpif, &all_names);
341 for (k = 0; k < all_names.n; k++) {
342 if (svec_contains(&bridge_names, all_names.names[k])) {
347 /* No. Delete the dpif. */
351 svec_destroy(&all_names);
356 svec_destroy(&bridge_names);
357 svec_destroy(&dpif_names);
358 svec_destroy(&dpif_types);
361 /* Attempt to create the network device 'iface_name' through the netdev
364 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
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]));
378 struct netdev_options netdev_options;
380 memset(&netdev_options, 0, sizeof netdev_options);
381 netdev_options.name = iface_cfg->name;
382 if (!strcmp(iface_cfg->type, "internal")) {
383 /* An "internal" config type maps to a netdev "system" type. */
384 netdev_options.type = "system";
386 netdev_options.type = iface_cfg->type;
388 netdev_options.args = &options;
389 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
391 error = netdev_open(&netdev_options, &iface->netdev);
394 netdev_get_carrier(iface->netdev, &iface->enabled);
396 } else if (iface->netdev) {
397 const char *netdev_type = netdev_get_type(iface->netdev);
398 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
399 ? iface_cfg->type : NULL;
401 /* An "internal" config type maps to a netdev "system" type. */
402 if (iface_type && !strcmp(iface_type, "internal")) {
403 iface_type = "system";
406 if (!iface_type || !strcmp(netdev_type, iface_type)) {
407 error = netdev_reconfigure(iface->netdev, &options);
409 VLOG_WARN("%s: attempting change device type from %s to %s",
410 iface_cfg->name, netdev_type, iface_type);
414 shash_destroy_free_data(&options);
420 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
422 return set_up_iface(iface_cfg, iface, false);
426 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
427 void *aux OVS_UNUSED)
429 if (!iface->netdev) {
430 int error = set_up_iface(iface->cfg, iface, true);
432 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
442 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
443 void *aux OVS_UNUSED)
445 if (iface->dp_ifidx >= 0) {
446 VLOG_DBG("%s has interface %s on port %d",
448 iface->name, iface->dp_ifidx);
451 VLOG_ERR("%s interface not in %s, dropping",
452 iface->name, dpif_name(br->dpif));
458 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
459 void *aux OVS_UNUSED)
461 /* Set policing attributes. */
462 netdev_set_policing(iface->netdev,
463 iface->cfg->ingress_policing_rate,
464 iface->cfg->ingress_policing_burst);
466 /* Set MAC address of internal interfaces other than the local
468 if (iface->dp_ifidx != ODPP_LOCAL
469 && iface_is_internal(br, iface->name)) {
470 iface_set_mac(iface);
476 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
477 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
478 * deletes from 'br' any ports that no longer have any interfaces. */
480 iterate_and_prune_ifaces(struct bridge *br,
481 bool (*cb)(struct bridge *, struct iface *,
487 for (i = 0; i < br->n_ports; ) {
488 struct port *port = br->ports[i];
489 for (j = 0; j < port->n_ifaces; ) {
490 struct iface *iface = port->ifaces[j];
491 if (cb(br, iface, aux)) {
494 iface_destroy(iface);
498 if (port->n_ifaces) {
501 VLOG_ERR("%s port has no interfaces, dropping", port->name);
507 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
508 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
509 * responsible for freeing '*managersp' (with free()).
511 * You may be asking yourself "why does ovs-vswitchd care?", because
512 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
513 * should not be and in fact is not directly involved in that. But
514 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
515 * it has to tell in-band control where the managers are to enable that.
518 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
519 struct sockaddr_in **managersp, size_t *n_managersp)
521 struct sockaddr_in *managers = NULL;
522 size_t n_managers = 0;
524 if (ovs_cfg->n_managers > 0) {
527 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
528 for (i = 0; i < ovs_cfg->n_managers; i++) {
529 const char *name = ovs_cfg->managers[i];
530 struct sockaddr_in *sin = &managers[i];
532 if ((!strncmp(name, "tcp:", 4)
533 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
534 (!strncmp(name, "ssl:", 4)
535 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
541 *managersp = managers;
542 *n_managersp = n_managers;
546 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
548 struct shash old_br, new_br;
549 struct shash_node *node;
550 struct bridge *br, *next;
551 struct sockaddr_in *managers;
554 int sflow_bridge_number;
556 COVERAGE_INC(bridge_reconfigure);
558 collect_managers(ovs_cfg, &managers, &n_managers);
560 /* Collect old and new bridges. */
563 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
564 shash_add(&old_br, br->name, br);
566 for (i = 0; i < ovs_cfg->n_bridges; i++) {
567 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
568 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
569 VLOG_WARN("more than one bridge named %s", br_cfg->name);
573 /* Get rid of deleted bridges and add new bridges. */
574 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
575 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
582 SHASH_FOR_EACH (node, &new_br) {
583 const char *br_name = node->name;
584 const struct ovsrec_bridge *br_cfg = node->data;
585 br = shash_find_data(&old_br, br_name);
587 /* If the bridge datapath type has changed, we need to tear it
588 * down and recreate. */
589 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
591 bridge_create(br_cfg);
594 bridge_create(br_cfg);
597 shash_destroy(&old_br);
598 shash_destroy(&new_br);
600 /* Reconfigure all bridges. */
601 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
602 bridge_reconfigure_one(br);
605 /* Add and delete ports on all datapaths.
607 * The kernel will reject any attempt to add a given port to a datapath if
608 * that port already belongs to a different datapath, so we must do all
609 * port deletions before any port additions. */
610 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
611 struct odp_port *dpif_ports;
613 struct shash want_ifaces;
615 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
616 bridge_get_all_ifaces(br, &want_ifaces);
617 for (i = 0; i < n_dpif_ports; i++) {
618 const struct odp_port *p = &dpif_ports[i];
619 if (!shash_find(&want_ifaces, p->devname)
620 && strcmp(p->devname, br->name)) {
621 int retval = dpif_port_del(br->dpif, p->port);
623 VLOG_ERR("failed to remove %s interface from %s: %s",
624 p->devname, dpif_name(br->dpif),
629 shash_destroy(&want_ifaces);
632 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
633 struct odp_port *dpif_ports;
635 struct shash cur_ifaces, want_ifaces;
637 /* Get the set of interfaces currently in this datapath. */
638 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
639 shash_init(&cur_ifaces);
640 for (i = 0; i < n_dpif_ports; i++) {
641 const char *name = dpif_ports[i].devname;
642 shash_add_once(&cur_ifaces, name, NULL);
646 /* Get the set of interfaces we want on this datapath. */
647 bridge_get_all_ifaces(br, &want_ifaces);
649 SHASH_FOR_EACH (node, &want_ifaces) {
650 const char *if_name = node->name;
651 struct iface *iface = node->data;
653 if (shash_find(&cur_ifaces, if_name)) {
654 /* Already exists, just reconfigure it. */
656 reconfigure_iface(iface->cfg, iface);
659 /* Need to add to datapath. */
663 /* Add to datapath. */
664 internal = iface_is_internal(br, if_name);
665 error = dpif_port_add(br->dpif, if_name,
666 internal ? ODP_PORT_INTERNAL : 0, NULL);
667 if (error == EFBIG) {
668 VLOG_ERR("ran out of valid port numbers on %s",
669 dpif_name(br->dpif));
672 VLOG_ERR("failed to add %s interface to %s: %s",
673 if_name, dpif_name(br->dpif), strerror(error));
677 shash_destroy(&cur_ifaces);
678 shash_destroy(&want_ifaces);
680 sflow_bridge_number = 0;
681 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
684 struct iface *local_iface;
685 struct iface *hw_addr_iface;
688 bridge_fetch_dp_ifaces(br);
690 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
691 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
693 /* Pick local port hardware address, datapath ID. */
694 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
695 local_iface = bridge_get_local_iface(br);
697 int error = netdev_set_etheraddr(local_iface->netdev, ea);
699 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
700 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
701 "Ethernet address: %s",
702 br->name, strerror(error));
706 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
707 ofproto_set_datapath_id(br->ofproto, dpid);
709 dpid_string = xasprintf("%016"PRIx64, dpid);
710 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
713 /* Set NetFlow configuration on this bridge. */
714 if (br->cfg->netflow) {
715 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
716 struct netflow_options opts;
718 memset(&opts, 0, sizeof opts);
720 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
721 if (nf_cfg->engine_type) {
722 opts.engine_type = *nf_cfg->engine_type;
724 if (nf_cfg->engine_id) {
725 opts.engine_id = *nf_cfg->engine_id;
728 opts.active_timeout = nf_cfg->active_timeout;
729 if (!opts.active_timeout) {
730 opts.active_timeout = -1;
731 } else if (opts.active_timeout < 0) {
732 VLOG_WARN("bridge %s: active timeout interval set to negative "
733 "value, using default instead (%d seconds)", br->name,
734 NF_ACTIVE_TIMEOUT_DEFAULT);
735 opts.active_timeout = -1;
738 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
739 if (opts.add_id_to_iface) {
740 if (opts.engine_id > 0x7f) {
741 VLOG_WARN("bridge %s: netflow port mangling may conflict "
742 "with another vswitch, choose an engine id less "
743 "than 128", br->name);
745 if (br->n_ports > 508) {
746 VLOG_WARN("bridge %s: netflow port mangling will conflict "
747 "with another port when more than 508 ports are "
752 opts.collectors.n = nf_cfg->n_targets;
753 opts.collectors.names = nf_cfg->targets;
754 if (ofproto_set_netflow(br->ofproto, &opts)) {
755 VLOG_ERR("bridge %s: problem setting netflow collectors",
759 ofproto_set_netflow(br->ofproto, NULL);
762 /* Set sFlow configuration on this bridge. */
763 if (br->cfg->sflow) {
764 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
765 struct ovsrec_controller **controllers;
766 struct ofproto_sflow_options oso;
767 size_t n_controllers;
769 memset(&oso, 0, sizeof oso);
771 oso.targets.n = sflow_cfg->n_targets;
772 oso.targets.names = sflow_cfg->targets;
774 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
775 if (sflow_cfg->sampling) {
776 oso.sampling_rate = *sflow_cfg->sampling;
779 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
780 if (sflow_cfg->polling) {
781 oso.polling_interval = *sflow_cfg->polling;
784 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
785 if (sflow_cfg->header) {
786 oso.header_len = *sflow_cfg->header;
789 oso.sub_id = sflow_bridge_number++;
790 oso.agent_device = sflow_cfg->agent;
792 oso.control_ip = NULL;
793 n_controllers = bridge_get_controllers(br, &controllers);
794 for (i = 0; i < n_controllers; i++) {
795 if (controllers[i]->local_ip) {
796 oso.control_ip = controllers[i]->local_ip;
800 ofproto_set_sflow(br->ofproto, &oso);
802 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
804 ofproto_set_sflow(br->ofproto, NULL);
807 /* Update the controller and related settings. It would be more
808 * straightforward to call this from bridge_reconfigure_one(), but we
809 * can't do it there for two reasons. First, and most importantly, at
810 * that point we don't know the dp_ifidx of any interfaces that have
811 * been added to the bridge (because we haven't actually added them to
812 * the datapath). Second, at that point we haven't set the datapath ID
813 * yet; when a controller is configured, resetting the datapath ID will
814 * immediately disconnect from the controller, so it's better to set
815 * the datapath ID before the controller. */
816 bridge_reconfigure_remotes(br, managers, n_managers);
818 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
819 for (i = 0; i < br->n_ports; i++) {
820 struct port *port = br->ports[i];
823 port_update_vlan_compat(port);
824 port_update_bonding(port);
826 for (j = 0; j < port->n_ifaces; j++) {
827 iface_update_qos(port->ifaces[j], port->cfg->qos);
831 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
832 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
839 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
840 const struct ovsdb_idl_column *column,
843 const struct ovsdb_datum *datum;
844 union ovsdb_atom atom;
847 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
848 atom.string = (char *) key;
849 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
850 return idx == UINT_MAX ? NULL : datum->values[idx].string;
854 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
856 return get_ovsrec_key_value(&br_cfg->header_,
857 &ovsrec_bridge_col_other_config, key);
861 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
862 struct iface **hw_addr_iface)
868 *hw_addr_iface = NULL;
870 /* Did the user request a particular MAC? */
871 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
872 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
873 if (eth_addr_is_multicast(ea)) {
874 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
875 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
876 } else if (eth_addr_is_zero(ea)) {
877 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
883 /* Otherwise choose the minimum non-local MAC address among all of the
885 memset(ea, 0xff, sizeof ea);
886 for (i = 0; i < br->n_ports; i++) {
887 struct port *port = br->ports[i];
888 uint8_t iface_ea[ETH_ADDR_LEN];
891 /* Mirror output ports don't participate. */
892 if (port->is_mirror_output_port) {
896 /* Choose the MAC address to represent the port. */
897 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
898 /* Find the interface with this Ethernet address (if any) so that
899 * we can provide the correct devname to the caller. */
901 for (j = 0; j < port->n_ifaces; j++) {
902 struct iface *candidate = port->ifaces[j];
903 uint8_t candidate_ea[ETH_ADDR_LEN];
904 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
905 && eth_addr_equals(iface_ea, candidate_ea)) {
910 /* Choose the interface whose MAC address will represent the port.
911 * The Linux kernel bonding code always chooses the MAC address of
912 * the first slave added to a bond, and the Fedora networking
913 * scripts always add slaves to a bond in alphabetical order, so
914 * for compatibility we choose the interface with the name that is
915 * first in alphabetical order. */
916 iface = port->ifaces[0];
917 for (j = 1; j < port->n_ifaces; j++) {
918 struct iface *candidate = port->ifaces[j];
919 if (strcmp(candidate->name, iface->name) < 0) {
924 /* The local port doesn't count (since we're trying to choose its
925 * MAC address anyway). */
926 if (iface->dp_ifidx == ODPP_LOCAL) {
931 error = netdev_get_etheraddr(iface->netdev, iface_ea);
933 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
934 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
935 iface->name, strerror(error));
940 /* Compare against our current choice. */
941 if (!eth_addr_is_multicast(iface_ea) &&
942 !eth_addr_is_local(iface_ea) &&
943 !eth_addr_is_reserved(iface_ea) &&
944 !eth_addr_is_zero(iface_ea) &&
945 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
947 memcpy(ea, iface_ea, ETH_ADDR_LEN);
948 *hw_addr_iface = iface;
951 if (eth_addr_is_multicast(ea)) {
952 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
953 *hw_addr_iface = NULL;
954 VLOG_WARN("bridge %s: using default bridge Ethernet "
955 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
957 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
958 br->name, ETH_ADDR_ARGS(ea));
962 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
963 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
964 * an interface on 'br', then that interface must be passed in as
965 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
966 * 'hw_addr_iface' must be passed in as a null pointer. */
968 bridge_pick_datapath_id(struct bridge *br,
969 const uint8_t bridge_ea[ETH_ADDR_LEN],
970 struct iface *hw_addr_iface)
973 * The procedure for choosing a bridge MAC address will, in the most
974 * ordinary case, also choose a unique MAC that we can use as a datapath
975 * ID. In some special cases, though, multiple bridges will end up with
976 * the same MAC address. This is OK for the bridges, but it will confuse
977 * the OpenFlow controller, because each datapath needs a unique datapath
980 * Datapath IDs must be unique. It is also very desirable that they be
981 * stable from one run to the next, so that policy set on a datapath
984 const char *datapath_id;
987 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
988 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
994 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
996 * A bridge whose MAC address is taken from a VLAN network device
997 * (that is, a network device created with vconfig(8) or similar
998 * tool) will have the same MAC address as a bridge on the VLAN
999 * device's physical network device.
1001 * Handle this case by hashing the physical network device MAC
1002 * along with the VLAN identifier.
1004 uint8_t buf[ETH_ADDR_LEN + 2];
1005 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1006 buf[ETH_ADDR_LEN] = vlan >> 8;
1007 buf[ETH_ADDR_LEN + 1] = vlan;
1008 return dpid_from_hash(buf, sizeof buf);
1011 * Assume that this bridge's MAC address is unique, since it
1012 * doesn't fit any of the cases we handle specially.
1017 * A purely internal bridge, that is, one that has no non-virtual
1018 * network devices on it at all, is more difficult because it has no
1019 * natural unique identifier at all.
1021 * When the host is a XenServer, we handle this case by hashing the
1022 * host's UUID with the name of the bridge. Names of bridges are
1023 * persistent across XenServer reboots, although they can be reused if
1024 * an internal network is destroyed and then a new one is later
1025 * created, so this is fairly effective.
1027 * When the host is not a XenServer, we punt by using a random MAC
1028 * address on each run.
1030 const char *host_uuid = xenserver_get_host_uuid();
1032 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1033 dpid = dpid_from_hash(combined, strlen(combined));
1039 return eth_addr_to_uint64(bridge_ea);
1043 dpid_from_hash(const void *data, size_t n)
1045 uint8_t hash[SHA1_DIGEST_SIZE];
1047 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1048 sha1_bytes(data, n, hash);
1049 eth_addr_mark_random(hash);
1050 return eth_addr_to_uint64(hash);
1054 iface_refresh_stats(struct iface *iface)
1060 static const struct iface_stat iface_stats[] = {
1061 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1062 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1063 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1064 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1065 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1066 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1067 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1068 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1069 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1070 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1071 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1072 { "collisions", offsetof(struct netdev_stats, collisions) },
1074 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1075 const struct iface_stat *s;
1077 char *keys[N_STATS];
1078 int64_t values[N_STATS];
1081 struct netdev_stats stats;
1083 /* Intentionally ignore return value, since errors will set 'stats' to
1084 * all-1s, and we will deal with that correctly below. */
1085 netdev_get_stats(iface->netdev, &stats);
1088 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1089 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1090 if (value != UINT64_MAX) {
1097 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1101 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1103 struct ovsdb_datum datum;
1107 get_system_stats(&stats);
1109 ovsdb_datum_from_shash(&datum, &stats);
1110 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1117 const struct ovsrec_open_vswitch *cfg;
1119 bool datapath_destroyed;
1120 bool database_changed;
1123 /* Let each bridge do the work that it needs to do. */
1124 datapath_destroyed = false;
1125 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1126 int error = bridge_run_one(br);
1128 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1129 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1130 "forcing reconfiguration", br->name);
1131 datapath_destroyed = true;
1135 /* (Re)configure if necessary. */
1136 database_changed = ovsdb_idl_run(idl);
1137 cfg = ovsrec_open_vswitch_first(idl);
1138 if (database_changed || datapath_destroyed) {
1140 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1142 bridge_configure_once(cfg);
1143 bridge_reconfigure(cfg);
1145 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1146 ovsdb_idl_txn_commit(txn);
1147 ovsdb_idl_txn_destroy(txn); /* XXX */
1149 /* We still need to reconfigure to avoid dangling pointers to
1150 * now-destroyed ovsrec structures inside bridge data. */
1151 static const struct ovsrec_open_vswitch null_cfg;
1153 bridge_reconfigure(&null_cfg);
1158 /* Re-configure SSL. We do this on every trip through the main loop,
1159 * instead of just when the database changes, because the contents of the
1160 * key and certificate files can change without the database changing. */
1161 if (cfg && cfg->ssl) {
1162 const struct ovsrec_ssl *ssl = cfg->ssl;
1164 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1165 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1169 /* Refresh interface stats if necessary. */
1170 if (time_msec() >= iface_stats_timer) {
1172 struct ovsdb_idl_txn *txn;
1174 txn = ovsdb_idl_txn_create(idl);
1175 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1178 for (i = 0; i < br->n_ports; i++) {
1179 struct port *port = br->ports[i];
1182 for (j = 0; j < port->n_ifaces; j++) {
1183 struct iface *iface = port->ifaces[j];
1184 iface_refresh_stats(iface);
1188 refresh_system_stats(cfg);
1189 ovsdb_idl_txn_commit(txn);
1190 ovsdb_idl_txn_destroy(txn); /* XXX */
1193 iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
1202 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1203 ofproto_wait(br->ofproto);
1204 if (ofproto_has_primary_controller(br->ofproto)) {
1208 mac_learning_wait(br->ml);
1211 ovsdb_idl_wait(idl);
1212 poll_timer_wait_until(iface_stats_timer);
1215 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1216 * configuration changes. */
1218 bridge_flush(struct bridge *br)
1220 COVERAGE_INC(bridge_flush);
1222 mac_learning_flush(br->ml);
1225 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1226 * such interface. */
1227 static struct iface *
1228 bridge_get_local_iface(struct bridge *br)
1232 for (i = 0; i < br->n_ports; i++) {
1233 struct port *port = br->ports[i];
1234 for (j = 0; j < port->n_ifaces; j++) {
1235 struct iface *iface = port->ifaces[j];
1236 if (iface->dp_ifidx == ODPP_LOCAL) {
1245 /* Bridge unixctl user interface functions. */
1247 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1248 const char *args, void *aux OVS_UNUSED)
1250 struct ds ds = DS_EMPTY_INITIALIZER;
1251 const struct bridge *br;
1252 const struct mac_entry *e;
1254 br = bridge_lookup(args);
1256 unixctl_command_reply(conn, 501, "no such bridge");
1260 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1261 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1262 if (e->port < 0 || e->port >= br->n_ports) {
1265 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1266 br->ports[e->port]->ifaces[0]->dp_ifidx,
1267 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1269 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1273 /* Bridge reconfiguration functions. */
1274 static struct bridge *
1275 bridge_create(const struct ovsrec_bridge *br_cfg)
1280 assert(!bridge_lookup(br_cfg->name));
1281 br = xzalloc(sizeof *br);
1283 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1289 dpif_flow_flush(br->dpif);
1291 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1294 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1296 dpif_delete(br->dpif);
1297 dpif_close(br->dpif);
1302 br->name = xstrdup(br_cfg->name);
1304 br->ml = mac_learning_create();
1305 eth_addr_nicira_random(br->default_ea);
1307 port_array_init(&br->ifaces);
1309 shash_init(&br->port_by_name);
1310 shash_init(&br->iface_by_name);
1314 list_push_back(&all_bridges, &br->node);
1316 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1322 bridge_destroy(struct bridge *br)
1327 while (br->n_ports > 0) {
1328 port_destroy(br->ports[br->n_ports - 1]);
1330 list_remove(&br->node);
1331 error = dpif_delete(br->dpif);
1332 if (error && error != ENOENT) {
1333 VLOG_ERR("failed to delete %s: %s",
1334 dpif_name(br->dpif), strerror(error));
1336 dpif_close(br->dpif);
1337 ofproto_destroy(br->ofproto);
1338 mac_learning_destroy(br->ml);
1339 port_array_destroy(&br->ifaces);
1340 shash_destroy(&br->port_by_name);
1341 shash_destroy(&br->iface_by_name);
1348 static struct bridge *
1349 bridge_lookup(const char *name)
1353 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1354 if (!strcmp(br->name, name)) {
1361 /* Handle requests for a listing of all flows known by the OpenFlow
1362 * stack, including those normally hidden. */
1364 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1365 const char *args, void *aux OVS_UNUSED)
1370 br = bridge_lookup(args);
1372 unixctl_command_reply(conn, 501, "Unknown bridge");
1377 ofproto_get_all_flows(br->ofproto, &results);
1379 unixctl_command_reply(conn, 200, ds_cstr(&results));
1380 ds_destroy(&results);
1383 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1384 * connections and reconnect. If BRIDGE is not specified, then all bridges
1385 * drop their controller connections and reconnect. */
1387 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1388 const char *args, void *aux OVS_UNUSED)
1391 if (args[0] != '\0') {
1392 br = bridge_lookup(args);
1394 unixctl_command_reply(conn, 501, "Unknown bridge");
1397 ofproto_reconnect_controllers(br->ofproto);
1399 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1400 ofproto_reconnect_controllers(br->ofproto);
1403 unixctl_command_reply(conn, 200, NULL);
1407 bridge_run_one(struct bridge *br)
1411 error = ofproto_run1(br->ofproto);
1416 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1419 error = ofproto_run2(br->ofproto, br->flush);
1426 bridge_get_controllers(const struct bridge *br,
1427 struct ovsrec_controller ***controllersp)
1429 struct ovsrec_controller **controllers;
1430 size_t n_controllers;
1432 controllers = br->cfg->controller;
1433 n_controllers = br->cfg->n_controller;
1435 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1441 *controllersp = controllers;
1443 return n_controllers;
1447 bridge_reconfigure_one(struct bridge *br)
1449 struct shash old_ports, new_ports;
1450 struct svec snoops, old_snoops;
1451 struct shash_node *node;
1452 enum ofproto_fail_mode fail_mode;
1455 /* Collect old ports. */
1456 shash_init(&old_ports);
1457 for (i = 0; i < br->n_ports; i++) {
1458 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1461 /* Collect new ports. */
1462 shash_init(&new_ports);
1463 for (i = 0; i < br->cfg->n_ports; i++) {
1464 const char *name = br->cfg->ports[i]->name;
1465 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1466 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1471 /* If we have a controller, then we need a local port. Complain if the
1472 * user didn't specify one.
1474 * XXX perhaps we should synthesize a port ourselves in this case. */
1475 if (bridge_get_controllers(br, NULL)) {
1476 char local_name[IF_NAMESIZE];
1479 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1480 local_name, sizeof local_name);
1481 if (!error && !shash_find(&new_ports, local_name)) {
1482 VLOG_WARN("bridge %s: controller specified but no local port "
1483 "(port named %s) defined",
1484 br->name, local_name);
1488 /* Get rid of deleted ports.
1489 * Get rid of deleted interfaces on ports that still exist. */
1490 SHASH_FOR_EACH (node, &old_ports) {
1491 struct port *port = node->data;
1492 const struct ovsrec_port *port_cfg;
1494 port_cfg = shash_find_data(&new_ports, node->name);
1498 port_del_ifaces(port, port_cfg);
1502 /* Create new ports.
1503 * Add new interfaces to existing ports.
1504 * Reconfigure existing ports. */
1505 SHASH_FOR_EACH (node, &new_ports) {
1506 struct port *port = shash_find_data(&old_ports, node->name);
1508 port = port_create(br, node->name);
1511 port_reconfigure(port, node->data);
1512 if (!port->n_ifaces) {
1513 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1514 br->name, port->name);
1518 shash_destroy(&old_ports);
1519 shash_destroy(&new_ports);
1521 /* Set the fail-mode */
1522 fail_mode = !br->cfg->fail_mode
1523 || !strcmp(br->cfg->fail_mode, "standalone")
1524 ? OFPROTO_FAIL_STANDALONE
1525 : OFPROTO_FAIL_SECURE;
1526 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1527 && !ofproto_has_primary_controller(br->ofproto)) {
1528 ofproto_flush_flows(br->ofproto);
1530 ofproto_set_fail_mode(br->ofproto, fail_mode);
1532 /* Delete all flows if we're switching from connected to standalone or vice
1533 * versa. (XXX Should we delete all flows if we are switching from one
1534 * controller to another?) */
1536 /* Configure OpenFlow controller connection snooping. */
1538 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1539 ovs_rundir, br->name));
1540 svec_init(&old_snoops);
1541 ofproto_get_snoops(br->ofproto, &old_snoops);
1542 if (!svec_equal(&snoops, &old_snoops)) {
1543 ofproto_set_snoops(br->ofproto, &snoops);
1545 svec_destroy(&snoops);
1546 svec_destroy(&old_snoops);
1548 mirror_reconfigure(br);
1551 /* Initializes 'oc' appropriately as a management service controller for
1554 * The caller must free oc->target when it is no longer needed. */
1556 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1557 struct ofproto_controller *oc)
1559 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir, br->name);
1560 oc->max_backoff = 0;
1561 oc->probe_interval = 60;
1562 oc->band = OFPROTO_OUT_OF_BAND;
1563 oc->accept_re = NULL;
1564 oc->update_resolv_conf = false;
1566 oc->burst_limit = 0;
1569 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1571 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1572 struct ofproto_controller *oc)
1574 oc->target = c->target;
1575 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1576 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1577 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1578 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1579 oc->accept_re = c->discover_accept_regex;
1580 oc->update_resolv_conf = c->discover_update_resolv_conf;
1581 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1582 oc->burst_limit = (c->controller_burst_limit
1583 ? *c->controller_burst_limit : 0);
1586 /* Configures the IP stack for 'br''s local interface properly according to the
1587 * configuration in 'c'. */
1589 bridge_configure_local_iface_netdev(struct bridge *br,
1590 struct ovsrec_controller *c)
1592 struct netdev *netdev;
1593 struct in_addr mask, gateway;
1595 struct iface *local_iface;
1598 /* Controller discovery does its own TCP/IP configuration later. */
1599 if (strcmp(c->target, "discover")) {
1603 /* If there's no local interface or no IP address, give up. */
1604 local_iface = bridge_get_local_iface(br);
1605 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1609 /* Bring up the local interface. */
1610 netdev = local_iface->netdev;
1611 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1613 /* Configure the IP address and netmask. */
1614 if (!c->local_netmask
1615 || !inet_aton(c->local_netmask, &mask)
1617 mask.s_addr = guess_netmask(ip.s_addr);
1619 if (!netdev_set_in4(netdev, ip, mask)) {
1620 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1621 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1624 /* Configure the default gateway. */
1625 if (c->local_gateway
1626 && inet_aton(c->local_gateway, &gateway)
1627 && gateway.s_addr) {
1628 if (!netdev_add_router(netdev, gateway)) {
1629 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1630 br->name, IP_ARGS(&gateway.s_addr));
1636 bridge_reconfigure_remotes(struct bridge *br,
1637 const struct sockaddr_in *managers,
1640 struct ovsrec_controller **controllers;
1641 size_t n_controllers;
1644 struct ofproto_controller *ocs;
1648 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1649 had_primary = ofproto_has_primary_controller(br->ofproto);
1651 n_controllers = bridge_get_controllers(br, &controllers);
1653 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1656 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1657 for (i = 0; i < n_controllers; i++) {
1658 struct ovsrec_controller *c = controllers[i];
1660 if (!strncmp(c->target, "punix:", 6)
1661 || !strncmp(c->target, "unix:", 5)) {
1662 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1664 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1665 * domain sockets and overwriting arbitrary local files. */
1666 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1667 "\"%s\" due to possibility for remote exploit",
1668 dpif_name(br->dpif), c->target);
1672 bridge_configure_local_iface_netdev(br, c);
1673 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs++]);
1676 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1677 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1680 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1681 ofproto_flush_flows(br->ofproto);
1684 /* If there are no controllers and the bridge is in standalone
1685 * mode, set up a flow that matches every packet and directs
1686 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1687 * switch is in secure mode and we won't pass any traffic until
1688 * a controller has been defined and it tells us to do so. */
1690 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1691 union ofp_action action;
1694 memset(&action, 0, sizeof action);
1695 action.type = htons(OFPAT_OUTPUT);
1696 action.output.len = htons(sizeof action);
1697 action.output.port = htons(OFPP_NORMAL);
1698 memset(&flow, 0, sizeof flow);
1699 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1704 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1709 for (i = 0; i < br->n_ports; i++) {
1710 struct port *port = br->ports[i];
1711 for (j = 0; j < port->n_ifaces; j++) {
1712 struct iface *iface = port->ifaces[j];
1713 shash_add_once(ifaces, iface->name, iface);
1715 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1716 shash_add_once(ifaces, port->name, NULL);
1721 /* For robustness, in case the administrator moves around datapath ports behind
1722 * our back, we re-check all the datapath port numbers here.
1724 * This function will set the 'dp_ifidx' members of interfaces that have
1725 * disappeared to -1, so only call this function from a context where those
1726 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1727 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1728 * datapath, which doesn't support UINT16_MAX+1 ports. */
1730 bridge_fetch_dp_ifaces(struct bridge *br)
1732 struct odp_port *dpif_ports;
1733 size_t n_dpif_ports;
1736 /* Reset all interface numbers. */
1737 for (i = 0; i < br->n_ports; i++) {
1738 struct port *port = br->ports[i];
1739 for (j = 0; j < port->n_ifaces; j++) {
1740 struct iface *iface = port->ifaces[j];
1741 iface->dp_ifidx = -1;
1744 port_array_clear(&br->ifaces);
1746 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1747 for (i = 0; i < n_dpif_ports; i++) {
1748 struct odp_port *p = &dpif_ports[i];
1749 struct iface *iface = iface_lookup(br, p->devname);
1751 if (iface->dp_ifidx >= 0) {
1752 VLOG_WARN("%s reported interface %s twice",
1753 dpif_name(br->dpif), p->devname);
1754 } else if (iface_from_dp_ifidx(br, p->port)) {
1755 VLOG_WARN("%s reported interface %"PRIu16" twice",
1756 dpif_name(br->dpif), p->port);
1758 port_array_set(&br->ifaces, p->port, iface);
1759 iface->dp_ifidx = p->port;
1763 int64_t ofport = (iface->dp_ifidx >= 0
1764 ? odp_port_to_ofp_port(iface->dp_ifidx)
1766 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1773 /* Bridge packet processing functions. */
1776 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1778 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1781 static struct bond_entry *
1782 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1784 return &port->bond_hash[bond_hash(mac)];
1788 bond_choose_iface(const struct port *port)
1790 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1791 size_t i, best_down_slave = -1;
1792 long long next_delay_expiration = LLONG_MAX;
1794 for (i = 0; i < port->n_ifaces; i++) {
1795 struct iface *iface = port->ifaces[i];
1797 if (iface->enabled) {
1799 } else if (iface->delay_expires < next_delay_expiration) {
1800 best_down_slave = i;
1801 next_delay_expiration = iface->delay_expires;
1805 if (best_down_slave != -1) {
1806 struct iface *iface = port->ifaces[best_down_slave];
1808 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1809 "since no other interface is up", iface->name,
1810 iface->delay_expires - time_msec());
1811 bond_enable_slave(iface, true);
1814 return best_down_slave;
1818 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1819 uint16_t *dp_ifidx, tag_type *tags)
1821 struct iface *iface;
1823 assert(port->n_ifaces);
1824 if (port->n_ifaces == 1) {
1825 iface = port->ifaces[0];
1827 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1828 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1829 || !port->ifaces[e->iface_idx]->enabled) {
1830 /* XXX select interface properly. The current interface selection
1831 * is only good for testing the rebalancing code. */
1832 e->iface_idx = bond_choose_iface(port);
1833 if (e->iface_idx < 0) {
1834 *tags |= port->no_ifaces_tag;
1837 e->iface_tag = tag_create_random();
1838 ((struct port *) port)->bond_compat_is_stale = true;
1840 *tags |= e->iface_tag;
1841 iface = port->ifaces[e->iface_idx];
1843 *dp_ifidx = iface->dp_ifidx;
1844 *tags |= iface->tag; /* Currently only used for bonding. */
1849 bond_link_status_update(struct iface *iface, bool carrier)
1851 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1852 struct port *port = iface->port;
1854 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1855 /* Nothing to do. */
1858 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1859 iface->name, carrier ? "detected" : "dropped");
1860 if (carrier == iface->enabled) {
1861 iface->delay_expires = LLONG_MAX;
1862 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1863 iface->name, carrier ? "disabled" : "enabled");
1864 } else if (carrier && port->active_iface < 0) {
1865 bond_enable_slave(iface, true);
1866 if (port->updelay) {
1867 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1868 "other interface is up", iface->name, port->updelay);
1871 int delay = carrier ? port->updelay : port->downdelay;
1872 iface->delay_expires = time_msec() + delay;
1875 "interface %s: will be %s if it stays %s for %d ms",
1877 carrier ? "enabled" : "disabled",
1878 carrier ? "up" : "down",
1885 bond_choose_active_iface(struct port *port)
1887 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1889 port->active_iface = bond_choose_iface(port);
1890 port->active_iface_tag = tag_create_random();
1891 if (port->active_iface >= 0) {
1892 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1893 port->name, port->ifaces[port->active_iface]->name);
1895 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1901 bond_enable_slave(struct iface *iface, bool enable)
1903 struct port *port = iface->port;
1904 struct bridge *br = port->bridge;
1906 /* This acts as a recursion check. If the act of disabling a slave
1907 * causes a different slave to be enabled, the flag will allow us to
1908 * skip redundant work when we reenter this function. It must be
1909 * cleared on exit to keep things safe with multiple bonds. */
1910 static bool moving_active_iface = false;
1912 iface->delay_expires = LLONG_MAX;
1913 if (enable == iface->enabled) {
1917 iface->enabled = enable;
1918 if (!iface->enabled) {
1919 VLOG_WARN("interface %s: disabled", iface->name);
1920 ofproto_revalidate(br->ofproto, iface->tag);
1921 if (iface->port_ifidx == port->active_iface) {
1922 ofproto_revalidate(br->ofproto,
1923 port->active_iface_tag);
1925 /* Disabling a slave can lead to another slave being immediately
1926 * enabled if there will be no active slaves but one is waiting
1927 * on an updelay. In this case we do not need to run most of the
1928 * code for the newly enabled slave since there was no period
1929 * without an active slave and it is redundant with the disabling
1931 moving_active_iface = true;
1932 bond_choose_active_iface(port);
1934 bond_send_learning_packets(port);
1936 VLOG_WARN("interface %s: enabled", iface->name);
1937 if (port->active_iface < 0 && !moving_active_iface) {
1938 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1939 bond_choose_active_iface(port);
1940 bond_send_learning_packets(port);
1942 iface->tag = tag_create_random();
1945 moving_active_iface = false;
1946 port->bond_compat_is_stale = true;
1949 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1950 * bond interface. */
1952 bond_update_fake_iface_stats(struct port *port)
1954 struct netdev_stats bond_stats;
1955 struct netdev *bond_dev;
1958 memset(&bond_stats, 0, sizeof bond_stats);
1960 for (i = 0; i < port->n_ifaces; i++) {
1961 struct netdev_stats slave_stats;
1963 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1964 /* XXX: We swap the stats here because they are swapped back when
1965 * reported by the internal device. The reason for this is
1966 * internal devices normally represent packets going into the system
1967 * but when used as fake bond device they represent packets leaving
1968 * the system. We really should do this in the internal device
1969 * itself because changing it here reverses the counts from the
1970 * perspective of the switch. However, the internal device doesn't
1971 * know what type of device it represents so we have to do it here
1973 bond_stats.tx_packets += slave_stats.rx_packets;
1974 bond_stats.tx_bytes += slave_stats.rx_bytes;
1975 bond_stats.rx_packets += slave_stats.tx_packets;
1976 bond_stats.rx_bytes += slave_stats.tx_bytes;
1980 if (!netdev_open_default(port->name, &bond_dev)) {
1981 netdev_set_stats(bond_dev, &bond_stats);
1982 netdev_close(bond_dev);
1987 bond_run(struct bridge *br)
1991 for (i = 0; i < br->n_ports; i++) {
1992 struct port *port = br->ports[i];
1994 if (port->n_ifaces >= 2) {
1995 for (j = 0; j < port->n_ifaces; j++) {
1996 struct iface *iface = port->ifaces[j];
1997 if (time_msec() >= iface->delay_expires) {
1998 bond_enable_slave(iface, !iface->enabled);
2002 if (port->bond_fake_iface
2003 && time_msec() >= port->bond_next_fake_iface_update) {
2004 bond_update_fake_iface_stats(port);
2005 port->bond_next_fake_iface_update = time_msec() + 1000;
2009 if (port->bond_compat_is_stale) {
2010 port->bond_compat_is_stale = false;
2011 port_update_bond_compat(port);
2017 bond_wait(struct bridge *br)
2021 for (i = 0; i < br->n_ports; i++) {
2022 struct port *port = br->ports[i];
2023 if (port->n_ifaces < 2) {
2026 for (j = 0; j < port->n_ifaces; j++) {
2027 struct iface *iface = port->ifaces[j];
2028 if (iface->delay_expires != LLONG_MAX) {
2029 poll_timer_wait_until(iface->delay_expires);
2032 if (port->bond_fake_iface) {
2033 poll_timer_wait_until(port->bond_next_fake_iface_update);
2039 set_dst(struct dst *p, const flow_t *flow,
2040 const struct port *in_port, const struct port *out_port,
2043 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2044 : in_port->vlan >= 0 ? in_port->vlan
2045 : ntohs(flow->dl_vlan));
2046 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
2050 swap_dst(struct dst *p, struct dst *q)
2052 struct dst tmp = *p;
2057 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2058 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2059 * that we push to the datapath. We could in fact fully sort the array by
2060 * vlan, but in most cases there are at most two different vlan tags so that's
2061 * possibly overkill.) */
2063 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2065 struct dst *first = dsts;
2066 struct dst *last = dsts + n_dsts;
2068 while (first != last) {
2070 * - All dsts < first have vlan == 'vlan'.
2071 * - All dsts >= last have vlan != 'vlan'.
2072 * - first < last. */
2073 while (first->vlan == vlan) {
2074 if (++first == last) {
2079 /* Same invariants, plus one additional:
2080 * - first->vlan != vlan.
2082 while (last[-1].vlan != vlan) {
2083 if (--last == first) {
2088 /* Same invariants, plus one additional:
2089 * - last[-1].vlan == vlan.*/
2090 swap_dst(first++, --last);
2095 mirror_mask_ffs(mirror_mask_t mask)
2097 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2102 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2103 const struct dst *test)
2106 for (i = 0; i < n_dsts; i++) {
2107 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2115 port_trunks_vlan(const struct port *port, uint16_t vlan)
2117 return (port->vlan < 0
2118 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2122 port_includes_vlan(const struct port *port, uint16_t vlan)
2124 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2128 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2129 const struct port *in_port, const struct port *out_port,
2130 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2132 mirror_mask_t mirrors = in_port->src_mirrors;
2133 struct dst *dst = dsts;
2136 if (out_port == FLOOD_PORT) {
2137 /* XXX use ODP_FLOOD if no vlans or bonding. */
2138 /* XXX even better, define each VLAN as a datapath port group */
2139 for (i = 0; i < br->n_ports; i++) {
2140 struct port *port = br->ports[i];
2141 if (port != in_port && port_includes_vlan(port, vlan)
2142 && !port->is_mirror_output_port
2143 && set_dst(dst, flow, in_port, port, tags)) {
2144 mirrors |= port->dst_mirrors;
2148 *nf_output_iface = NF_OUT_FLOOD;
2149 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2150 *nf_output_iface = dst->dp_ifidx;
2151 mirrors |= out_port->dst_mirrors;
2156 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2157 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2159 if (set_dst(dst, flow, in_port, m->out_port, tags)
2160 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2164 for (i = 0; i < br->n_ports; i++) {
2165 struct port *port = br->ports[i];
2166 if (port_includes_vlan(port, m->out_vlan)
2167 && set_dst(dst, flow, in_port, port, tags))
2171 if (port->vlan < 0) {
2172 dst->vlan = m->out_vlan;
2174 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2178 /* Use the vlan tag on the original flow instead of
2179 * the one passed in the vlan parameter. This ensures
2180 * that we compare the vlan from before any implicit
2181 * tagging tags place. This is necessary because
2182 * dst->vlan is the final vlan, after removing implicit
2184 flow_vlan = ntohs(flow->dl_vlan);
2185 if (flow_vlan == 0) {
2186 flow_vlan = OFP_VLAN_NONE;
2188 if (port == in_port && dst->vlan == flow_vlan) {
2189 /* Don't send out input port on same VLAN. */
2197 mirrors &= mirrors - 1;
2200 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2204 static void OVS_UNUSED
2205 print_dsts(const struct dst *dsts, size_t n)
2207 for (; n--; dsts++) {
2208 printf(">p%"PRIu16, dsts->dp_ifidx);
2209 if (dsts->vlan != OFP_VLAN_NONE) {
2210 printf("v%"PRIu16, dsts->vlan);
2216 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2217 const struct port *in_port, const struct port *out_port,
2218 tag_type *tags, struct odp_actions *actions,
2219 uint16_t *nf_output_iface)
2221 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2223 const struct dst *p;
2226 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2229 cur_vlan = ntohs(flow->dl_vlan);
2230 for (p = dsts; p < &dsts[n_dsts]; p++) {
2231 union odp_action *a;
2232 if (p->vlan != cur_vlan) {
2233 if (p->vlan == OFP_VLAN_NONE) {
2234 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2236 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2237 a->vlan_vid.vlan_vid = htons(p->vlan);
2241 a = odp_actions_add(actions, ODPAT_OUTPUT);
2242 a->output.port = p->dp_ifidx;
2246 /* Returns the effective vlan of a packet, taking into account both the
2247 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2248 * the packet is untagged and -1 indicates it has an invalid header and
2249 * should be dropped. */
2250 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2251 struct port *in_port, bool have_packet)
2253 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2254 * belongs to VLAN 0, so we should treat both cases identically. (In the
2255 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2256 * presumably to allow a priority to be specified. In the latter case, the
2257 * packet does not have any 802.1Q header.) */
2258 int vlan = ntohs(flow->dl_vlan);
2259 if (vlan == OFP_VLAN_NONE) {
2262 if (in_port->vlan >= 0) {
2264 /* XXX support double tagging? */
2266 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2267 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2268 "packet received on port %s configured with "
2269 "implicit VLAN %"PRIu16,
2270 br->name, ntohs(flow->dl_vlan),
2271 in_port->name, in_port->vlan);
2275 vlan = in_port->vlan;
2277 if (!port_includes_vlan(in_port, vlan)) {
2279 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2280 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2281 "packet received on port %s not configured for "
2283 br->name, vlan, in_port->name, vlan);
2292 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2293 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2294 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2296 is_gratuitous_arp(const flow_t *flow)
2298 return (flow->dl_type == htons(ETH_TYPE_ARP)
2299 && eth_addr_is_broadcast(flow->dl_dst)
2300 && (flow->nw_proto == ARP_OP_REPLY
2301 || (flow->nw_proto == ARP_OP_REQUEST
2302 && flow->nw_src == flow->nw_dst)));
2306 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2307 struct port *in_port)
2309 enum grat_arp_lock_type lock_type;
2312 /* We don't want to learn from gratuitous ARP packets that are reflected
2313 * back over bond slaves so we lock the learning table. */
2314 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2315 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2316 GRAT_ARP_LOCK_CHECK;
2318 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2321 /* The log messages here could actually be useful in debugging,
2322 * so keep the rate limit relatively high. */
2323 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2325 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2326 "on port %s in VLAN %d",
2327 br->name, ETH_ADDR_ARGS(flow->dl_src),
2328 in_port->name, vlan);
2329 ofproto_revalidate(br->ofproto, rev_tag);
2333 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2334 * dropped. Returns true if they may be forwarded, false if they should be
2337 * If 'have_packet' is true, it indicates that the caller is processing a
2338 * received packet. If 'have_packet' is false, then the caller is just
2339 * revalidating an existing flow because configuration has changed. Either
2340 * way, 'have_packet' only affects logging (there is no point in logging errors
2341 * during revalidation).
2343 * Sets '*in_portp' to the input port. This will be a null pointer if
2344 * flow->in_port does not designate a known input port (in which case
2345 * is_admissible() returns false).
2347 * When returning true, sets '*vlanp' to the effective VLAN of the input
2348 * packet, as returned by flow_get_vlan().
2350 * May also add tags to '*tags', although the current implementation only does
2351 * so in one special case.
2354 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2355 tag_type *tags, int *vlanp, struct port **in_portp)
2357 struct iface *in_iface;
2358 struct port *in_port;
2361 /* Find the interface and port structure for the received packet. */
2362 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2364 /* No interface? Something fishy... */
2366 /* Odd. A few possible reasons here:
2368 * - We deleted an interface but there are still a few packets
2369 * queued up from it.
2371 * - Someone externally added an interface (e.g. with "ovs-dpctl
2372 * add-if") that we don't know about.
2374 * - Packet arrived on the local port but the local port is not
2375 * one of our bridge ports.
2377 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2379 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2380 "interface %"PRIu16, br->name, flow->in_port);
2386 *in_portp = in_port = in_iface->port;
2387 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2392 /* Drop frames for reserved multicast addresses. */
2393 if (eth_addr_is_reserved(flow->dl_dst)) {
2397 /* Drop frames on ports reserved for mirroring. */
2398 if (in_port->is_mirror_output_port) {
2400 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2401 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2402 "%s, which is reserved exclusively for mirroring",
2403 br->name, in_port->name);
2408 /* Packets received on bonds need special attention to avoid duplicates. */
2409 if (in_port->n_ifaces > 1) {
2411 bool is_grat_arp_locked;
2413 if (eth_addr_is_multicast(flow->dl_dst)) {
2414 *tags |= in_port->active_iface_tag;
2415 if (in_port->active_iface != in_iface->port_ifidx) {
2416 /* Drop all multicast packets on inactive slaves. */
2421 /* Drop all packets for which we have learned a different input
2422 * port, because we probably sent the packet on one slave and got
2423 * it back on the other. Gratuitous ARP packets are an exception
2424 * to this rule: the host has moved to another switch. The exception
2425 * to the exception is if we locked the learning table to avoid
2426 * reflections on bond slaves. If this is the case, just drop the
2428 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2429 &is_grat_arp_locked);
2430 if (src_idx != -1 && src_idx != in_port->port_idx &&
2431 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2439 /* If the composed actions may be applied to any packet in the given 'flow',
2440 * returns true. Otherwise, the actions should only be applied to 'packet', or
2441 * not at all, if 'packet' was NULL. */
2443 process_flow(struct bridge *br, const flow_t *flow,
2444 const struct ofpbuf *packet, struct odp_actions *actions,
2445 tag_type *tags, uint16_t *nf_output_iface)
2447 struct port *in_port;
2448 struct port *out_port;
2452 /* Check whether we should drop packets in this flow. */
2453 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2458 /* Learn source MAC (but don't try to learn from revalidation). */
2460 update_learning_table(br, flow, vlan, in_port);
2463 /* Determine output port. */
2464 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2466 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2467 out_port = br->ports[out_port_idx];
2468 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2469 /* If we are revalidating but don't have a learning entry then
2470 * eject the flow. Installing a flow that floods packets opens
2471 * up a window of time where we could learn from a packet reflected
2472 * on a bond and blackhole packets before the learning table is
2473 * updated to reflect the correct port. */
2476 out_port = FLOOD_PORT;
2479 /* Don't send packets out their input ports. */
2480 if (in_port == out_port) {
2486 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2493 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2496 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2497 const struct ofp_phy_port *opp,
2500 struct bridge *br = br_;
2501 struct iface *iface;
2504 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2510 if (reason == OFPPR_DELETE) {
2511 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2512 br->name, iface->name);
2513 iface_destroy(iface);
2514 if (!port->n_ifaces) {
2515 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2516 br->name, port->name);
2522 if (port->n_ifaces > 1) {
2523 bool up = !(opp->state & OFPPS_LINK_DOWN);
2524 bond_link_status_update(iface, up);
2525 port_update_bond_compat(port);
2531 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2532 struct odp_actions *actions, tag_type *tags,
2533 uint16_t *nf_output_iface, void *br_)
2535 struct bridge *br = br_;
2537 COVERAGE_INC(bridge_process_flow);
2539 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2543 bridge_account_flow_ofhook_cb(const flow_t *flow, tag_type tags,
2544 const union odp_action *actions,
2545 size_t n_actions, unsigned long long int n_bytes,
2548 struct bridge *br = br_;
2549 const union odp_action *a;
2550 struct port *in_port;
2554 /* Feed information from the active flows back into the learning table to
2555 * ensure that table is always in sync with what is actually flowing
2556 * through the datapath.
2558 * We test that 'tags' is nonzero to ensure that only flows that include an
2559 * OFPP_NORMAL action are used for learning. This works because
2560 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2561 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2562 update_learning_table(br, flow, vlan, in_port);
2565 /* Account for bond slave utilization. */
2566 if (!br->has_bonded_ports) {
2569 for (a = actions; a < &actions[n_actions]; a++) {
2570 if (a->type == ODPAT_OUTPUT) {
2571 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2572 if (out_port && out_port->n_ifaces >= 2) {
2573 struct bond_entry *e = lookup_bond_entry(out_port,
2575 e->tx_bytes += n_bytes;
2582 bridge_account_checkpoint_ofhook_cb(void *br_)
2584 struct bridge *br = br_;
2588 if (!br->has_bonded_ports) {
2593 for (i = 0; i < br->n_ports; i++) {
2594 struct port *port = br->ports[i];
2595 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2596 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2597 bond_rebalance_port(port);
2602 static struct ofhooks bridge_ofhooks = {
2603 bridge_port_changed_ofhook_cb,
2604 bridge_normal_ofhook_cb,
2605 bridge_account_flow_ofhook_cb,
2606 bridge_account_checkpoint_ofhook_cb,
2609 /* Bonding functions. */
2611 /* Statistics for a single interface on a bonded port, used for load-based
2612 * bond rebalancing. */
2613 struct slave_balance {
2614 struct iface *iface; /* The interface. */
2615 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2617 /* All the "bond_entry"s that are assigned to this interface, in order of
2618 * increasing tx_bytes. */
2619 struct bond_entry **hashes;
2623 /* Sorts pointers to pointers to bond_entries in ascending order by the
2624 * interface to which they are assigned, and within a single interface in
2625 * ascending order of bytes transmitted. */
2627 compare_bond_entries(const void *a_, const void *b_)
2629 const struct bond_entry *const *ap = a_;
2630 const struct bond_entry *const *bp = b_;
2631 const struct bond_entry *a = *ap;
2632 const struct bond_entry *b = *bp;
2633 if (a->iface_idx != b->iface_idx) {
2634 return a->iface_idx > b->iface_idx ? 1 : -1;
2635 } else if (a->tx_bytes != b->tx_bytes) {
2636 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2642 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2643 * *descending* order by number of bytes transmitted. */
2645 compare_slave_balance(const void *a_, const void *b_)
2647 const struct slave_balance *a = a_;
2648 const struct slave_balance *b = b_;
2649 if (a->iface->enabled != b->iface->enabled) {
2650 return a->iface->enabled ? -1 : 1;
2651 } else if (a->tx_bytes != b->tx_bytes) {
2652 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2659 swap_bals(struct slave_balance *a, struct slave_balance *b)
2661 struct slave_balance tmp = *a;
2666 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2667 * given that 'p' (and only 'p') might be in the wrong location.
2669 * This function invalidates 'p', since it might now be in a different memory
2672 resort_bals(struct slave_balance *p,
2673 struct slave_balance bals[], size_t n_bals)
2676 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2677 swap_bals(p, p - 1);
2679 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2680 swap_bals(p, p + 1);
2686 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2688 if (VLOG_IS_DBG_ENABLED()) {
2689 struct ds ds = DS_EMPTY_INITIALIZER;
2690 const struct slave_balance *b;
2692 for (b = bals; b < bals + n_bals; b++) {
2696 ds_put_char(&ds, ',');
2698 ds_put_format(&ds, " %s %"PRIu64"kB",
2699 b->iface->name, b->tx_bytes / 1024);
2701 if (!b->iface->enabled) {
2702 ds_put_cstr(&ds, " (disabled)");
2704 if (b->n_hashes > 0) {
2705 ds_put_cstr(&ds, " (");
2706 for (i = 0; i < b->n_hashes; i++) {
2707 const struct bond_entry *e = b->hashes[i];
2709 ds_put_cstr(&ds, " + ");
2711 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2712 e - port->bond_hash, e->tx_bytes / 1024);
2714 ds_put_cstr(&ds, ")");
2717 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2722 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2724 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2727 struct bond_entry *hash = from->hashes[hash_idx];
2728 struct port *port = from->iface->port;
2729 uint64_t delta = hash->tx_bytes;
2731 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2732 "from %s to %s (now carrying %"PRIu64"kB and "
2733 "%"PRIu64"kB load, respectively)",
2734 port->name, delta / 1024, hash - port->bond_hash,
2735 from->iface->name, to->iface->name,
2736 (from->tx_bytes - delta) / 1024,
2737 (to->tx_bytes + delta) / 1024);
2739 /* Delete element from from->hashes.
2741 * We don't bother to add the element to to->hashes because not only would
2742 * it require more work, the only purpose it would be to allow that hash to
2743 * be migrated to another slave in this rebalancing run, and there is no
2744 * point in doing that. */
2745 if (hash_idx == 0) {
2748 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2749 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2753 /* Shift load away from 'from' to 'to'. */
2754 from->tx_bytes -= delta;
2755 to->tx_bytes += delta;
2757 /* Arrange for flows to be revalidated. */
2758 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2759 hash->iface_idx = to->iface->port_ifidx;
2760 hash->iface_tag = tag_create_random();
2764 bond_rebalance_port(struct port *port)
2766 struct slave_balance bals[DP_MAX_PORTS];
2768 struct bond_entry *hashes[BOND_MASK + 1];
2769 struct slave_balance *b, *from, *to;
2770 struct bond_entry *e;
2773 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2774 * descending order of tx_bytes, so that bals[0] represents the most
2775 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2778 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2779 * array for each slave_balance structure, we sort our local array of
2780 * hashes in order by slave, so that all of the hashes for a given slave
2781 * become contiguous in memory, and then we point each 'hashes' members of
2782 * a slave_balance structure to the start of a contiguous group. */
2783 n_bals = port->n_ifaces;
2784 for (b = bals; b < &bals[n_bals]; b++) {
2785 b->iface = port->ifaces[b - bals];
2790 for (i = 0; i <= BOND_MASK; i++) {
2791 hashes[i] = &port->bond_hash[i];
2793 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2794 for (i = 0; i <= BOND_MASK; i++) {
2796 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2797 b = &bals[e->iface_idx];
2798 b->tx_bytes += e->tx_bytes;
2800 b->hashes = &hashes[i];
2805 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2806 log_bals(bals, n_bals, port);
2808 /* Discard slaves that aren't enabled (which were sorted to the back of the
2809 * array earlier). */
2810 while (!bals[n_bals - 1].iface->enabled) {
2817 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2818 to = &bals[n_bals - 1];
2819 for (from = bals; from < to; ) {
2820 uint64_t overload = from->tx_bytes - to->tx_bytes;
2821 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2822 /* The extra load on 'from' (and all less-loaded slaves), compared
2823 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2824 * it is less than ~1Mbps. No point in rebalancing. */
2826 } else if (from->n_hashes == 1) {
2827 /* 'from' only carries a single MAC hash, so we can't shift any
2828 * load away from it, even though we want to. */
2831 /* 'from' is carrying significantly more load than 'to', and that
2832 * load is split across at least two different hashes. Pick a hash
2833 * to migrate to 'to' (the least-loaded slave), given that doing so
2834 * must decrease the ratio of the load on the two slaves by at
2837 * The sort order we use means that we prefer to shift away the
2838 * smallest hashes instead of the biggest ones. There is little
2839 * reason behind this decision; we could use the opposite sort
2840 * order to shift away big hashes ahead of small ones. */
2843 for (i = 0; i < from->n_hashes; i++) {
2844 double old_ratio, new_ratio;
2845 uint64_t delta = from->hashes[i]->tx_bytes;
2847 if (delta == 0 || from->tx_bytes - delta == 0) {
2848 /* Pointless move. */
2852 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2854 if (to->tx_bytes == 0) {
2855 /* Nothing on the new slave, move it. */
2859 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2860 new_ratio = (double)(from->tx_bytes - delta) /
2861 (to->tx_bytes + delta);
2863 if (new_ratio == 0) {
2864 /* Should already be covered but check to prevent division
2869 if (new_ratio < 1) {
2870 new_ratio = 1 / new_ratio;
2873 if (old_ratio - new_ratio > 0.1) {
2874 /* Would decrease the ratio, move it. */
2878 if (i < from->n_hashes) {
2879 bond_shift_load(from, to, i);
2880 port->bond_compat_is_stale = true;
2882 /* If the result of the migration changed the relative order of
2883 * 'from' and 'to' swap them back to maintain invariants. */
2884 if (order_swapped) {
2885 swap_bals(from, to);
2888 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2889 * point to different slave_balance structures. It is only
2890 * valid to do these two operations in a row at all because we
2891 * know that 'from' will not move past 'to' and vice versa. */
2892 resort_bals(from, bals, n_bals);
2893 resort_bals(to, bals, n_bals);
2900 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2901 * historical data to decay to <1% in 7 rebalancing runs. */
2902 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2908 bond_send_learning_packets(struct port *port)
2910 struct bridge *br = port->bridge;
2911 struct mac_entry *e;
2912 struct ofpbuf packet;
2913 int error, n_packets, n_errors;
2915 if (!port->n_ifaces || port->active_iface < 0) {
2919 ofpbuf_init(&packet, 128);
2920 error = n_packets = n_errors = 0;
2921 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2922 union ofp_action actions[2], *a;
2928 if (e->port == port->port_idx
2929 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2933 /* Compose actions. */
2934 memset(actions, 0, sizeof actions);
2937 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2938 a->vlan_vid.len = htons(sizeof *a);
2939 a->vlan_vid.vlan_vid = htons(e->vlan);
2942 a->output.type = htons(OFPAT_OUTPUT);
2943 a->output.len = htons(sizeof *a);
2944 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2949 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2951 flow_extract(&packet, 0, ODPP_NONE, &flow);
2952 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2959 ofpbuf_uninit(&packet);
2962 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2963 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2964 "packets, last error was: %s",
2965 port->name, n_errors, n_packets, strerror(error));
2967 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2968 port->name, n_packets);
2972 /* Bonding unixctl user interface functions. */
2975 bond_unixctl_list(struct unixctl_conn *conn,
2976 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2978 struct ds ds = DS_EMPTY_INITIALIZER;
2979 const struct bridge *br;
2981 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2983 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2986 for (i = 0; i < br->n_ports; i++) {
2987 const struct port *port = br->ports[i];
2988 if (port->n_ifaces > 1) {
2991 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2992 for (j = 0; j < port->n_ifaces; j++) {
2993 const struct iface *iface = port->ifaces[j];
2995 ds_put_cstr(&ds, ", ");
2997 ds_put_cstr(&ds, iface->name);
2999 ds_put_char(&ds, '\n');
3003 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3007 static struct port *
3008 bond_find(const char *name)
3010 const struct bridge *br;
3012 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
3015 for (i = 0; i < br->n_ports; i++) {
3016 struct port *port = br->ports[i];
3017 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3026 bond_unixctl_show(struct unixctl_conn *conn,
3027 const char *args, void *aux OVS_UNUSED)
3029 struct ds ds = DS_EMPTY_INITIALIZER;
3030 const struct port *port;
3033 port = bond_find(args);
3035 unixctl_command_reply(conn, 501, "no such bond");
3039 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3040 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3041 ds_put_format(&ds, "next rebalance: %lld ms\n",
3042 port->bond_next_rebalance - time_msec());
3043 for (j = 0; j < port->n_ifaces; j++) {
3044 const struct iface *iface = port->ifaces[j];
3045 struct bond_entry *be;
3048 ds_put_format(&ds, "slave %s: %s\n",
3049 iface->name, iface->enabled ? "enabled" : "disabled");
3050 if (j == port->active_iface) {
3051 ds_put_cstr(&ds, "\tactive slave\n");
3053 if (iface->delay_expires != LLONG_MAX) {
3054 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3055 iface->enabled ? "downdelay" : "updelay",
3056 iface->delay_expires - time_msec());
3060 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3061 int hash = be - port->bond_hash;
3062 struct mac_entry *me;
3064 if (be->iface_idx != j) {
3068 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3069 hash, be->tx_bytes / 1024);
3072 LIST_FOR_EACH (me, struct mac_entry, lru_node,
3073 &port->bridge->ml->lrus) {
3076 if (bond_hash(me->mac) == hash
3077 && me->port != port->port_idx
3078 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3079 && dp_ifidx == iface->dp_ifidx)
3081 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3082 ETH_ADDR_ARGS(me->mac));
3087 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3092 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3093 void *aux OVS_UNUSED)
3095 char *args = (char *) args_;
3096 char *save_ptr = NULL;
3097 char *bond_s, *hash_s, *slave_s;
3098 uint8_t mac[ETH_ADDR_LEN];
3100 struct iface *iface;
3101 struct bond_entry *entry;
3104 bond_s = strtok_r(args, " ", &save_ptr);
3105 hash_s = strtok_r(NULL, " ", &save_ptr);
3106 slave_s = strtok_r(NULL, " ", &save_ptr);
3108 unixctl_command_reply(conn, 501,
3109 "usage: bond/migrate BOND HASH SLAVE");
3113 port = bond_find(bond_s);
3115 unixctl_command_reply(conn, 501, "no such bond");
3119 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3120 == ETH_ADDR_SCAN_COUNT) {
3121 hash = bond_hash(mac);
3122 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3123 hash = atoi(hash_s) & BOND_MASK;
3125 unixctl_command_reply(conn, 501, "bad hash");
3129 iface = port_lookup_iface(port, slave_s);
3131 unixctl_command_reply(conn, 501, "no such slave");
3135 if (!iface->enabled) {
3136 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3140 entry = &port->bond_hash[hash];
3141 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3142 entry->iface_idx = iface->port_ifidx;
3143 entry->iface_tag = tag_create_random();
3144 port->bond_compat_is_stale = true;
3145 unixctl_command_reply(conn, 200, "migrated");
3149 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3150 void *aux OVS_UNUSED)
3152 char *args = (char *) args_;
3153 char *save_ptr = NULL;
3154 char *bond_s, *slave_s;
3156 struct iface *iface;
3158 bond_s = strtok_r(args, " ", &save_ptr);
3159 slave_s = strtok_r(NULL, " ", &save_ptr);
3161 unixctl_command_reply(conn, 501,
3162 "usage: bond/set-active-slave BOND SLAVE");
3166 port = bond_find(bond_s);
3168 unixctl_command_reply(conn, 501, "no such bond");
3172 iface = port_lookup_iface(port, slave_s);
3174 unixctl_command_reply(conn, 501, "no such slave");
3178 if (!iface->enabled) {
3179 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3183 if (port->active_iface != iface->port_ifidx) {
3184 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3185 port->active_iface = iface->port_ifidx;
3186 port->active_iface_tag = tag_create_random();
3187 VLOG_INFO("port %s: active interface is now %s",
3188 port->name, iface->name);
3189 bond_send_learning_packets(port);
3190 unixctl_command_reply(conn, 200, "done");
3192 unixctl_command_reply(conn, 200, "no change");
3197 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3199 char *args = (char *) args_;
3200 char *save_ptr = NULL;
3201 char *bond_s, *slave_s;
3203 struct iface *iface;
3205 bond_s = strtok_r(args, " ", &save_ptr);
3206 slave_s = strtok_r(NULL, " ", &save_ptr);
3208 unixctl_command_reply(conn, 501,
3209 "usage: bond/enable/disable-slave BOND SLAVE");
3213 port = bond_find(bond_s);
3215 unixctl_command_reply(conn, 501, "no such bond");
3219 iface = port_lookup_iface(port, slave_s);
3221 unixctl_command_reply(conn, 501, "no such slave");
3225 bond_enable_slave(iface, enable);
3226 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3230 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3231 void *aux OVS_UNUSED)
3233 enable_slave(conn, args, true);
3237 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3238 void *aux OVS_UNUSED)
3240 enable_slave(conn, args, false);
3244 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3245 void *aux OVS_UNUSED)
3247 uint8_t mac[ETH_ADDR_LEN];
3251 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3252 == ETH_ADDR_SCAN_COUNT) {
3253 hash = bond_hash(mac);
3255 hash_cstr = xasprintf("%u", hash);
3256 unixctl_command_reply(conn, 200, hash_cstr);
3259 unixctl_command_reply(conn, 501, "invalid mac");
3266 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3267 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3268 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3269 unixctl_command_register("bond/set-active-slave",
3270 bond_unixctl_set_active_slave, NULL);
3271 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3273 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3275 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3278 /* Port functions. */
3280 static struct port *
3281 port_create(struct bridge *br, const char *name)
3285 port = xzalloc(sizeof *port);
3287 port->port_idx = br->n_ports;
3289 port->trunks = NULL;
3290 port->name = xstrdup(name);
3291 port->active_iface = -1;
3293 if (br->n_ports >= br->allocated_ports) {
3294 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3297 br->ports[br->n_ports++] = port;
3298 shash_add_assert(&br->port_by_name, port->name, port);
3300 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3307 get_port_other_config(const struct ovsrec_port *port, const char *key,
3308 const char *default_value)
3312 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3314 return value ? value : default_value;
3318 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3320 struct shash new_ifaces;
3323 /* Collect list of new interfaces. */
3324 shash_init(&new_ifaces);
3325 for (i = 0; i < cfg->n_interfaces; i++) {
3326 const char *name = cfg->interfaces[i]->name;
3327 shash_add_once(&new_ifaces, name, NULL);
3330 /* Get rid of deleted interfaces. */
3331 for (i = 0; i < port->n_ifaces; ) {
3332 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3333 iface_destroy(port->ifaces[i]);
3339 shash_destroy(&new_ifaces);
3343 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3345 struct shash new_ifaces;
3346 long long int next_rebalance;
3347 unsigned long *trunks;
3353 /* Update settings. */
3354 port->updelay = cfg->bond_updelay;
3355 if (port->updelay < 0) {
3358 port->downdelay = cfg->bond_downdelay;
3359 if (port->downdelay < 0) {
3360 port->downdelay = 0;
3362 port->bond_rebalance_interval = atoi(
3363 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3364 if (port->bond_rebalance_interval < 1000) {
3365 port->bond_rebalance_interval = 1000;
3367 next_rebalance = time_msec() + port->bond_rebalance_interval;
3368 if (port->bond_next_rebalance > next_rebalance) {
3369 port->bond_next_rebalance = next_rebalance;
3372 /* Add new interfaces and update 'cfg' member of existing ones. */
3373 shash_init(&new_ifaces);
3374 for (i = 0; i < cfg->n_interfaces; i++) {
3375 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3376 struct iface *iface;
3378 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3379 VLOG_WARN("port %s: %s specified twice as port interface",
3380 port->name, if_cfg->name);
3384 iface = iface_lookup(port->bridge, if_cfg->name);
3386 if (iface->port != port) {
3387 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3389 port->bridge->name, if_cfg->name, iface->port->name);
3392 iface->cfg = if_cfg;
3394 iface_create(port, if_cfg);
3397 shash_destroy(&new_ifaces);
3402 if (port->n_ifaces < 2) {
3404 if (vlan >= 0 && vlan <= 4095) {
3405 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3410 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3411 * they even work as-is. But they have not been tested. */
3412 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3416 if (port->vlan != vlan) {
3418 bridge_flush(port->bridge);
3421 /* Get trunked VLANs. */
3423 if (vlan < 0 && cfg->n_trunks) {
3426 trunks = bitmap_allocate(4096);
3428 for (i = 0; i < cfg->n_trunks; i++) {
3429 int trunk = cfg->trunks[i];
3431 bitmap_set1(trunks, trunk);
3437 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3438 port->name, cfg->n_trunks);
3440 if (n_errors == cfg->n_trunks) {
3441 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3443 bitmap_free(trunks);
3446 } else if (vlan >= 0 && cfg->n_trunks) {
3447 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3451 ? port->trunks != NULL
3452 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3453 bridge_flush(port->bridge);
3455 bitmap_free(port->trunks);
3456 port->trunks = trunks;
3460 port_destroy(struct port *port)
3463 struct bridge *br = port->bridge;
3467 proc_net_compat_update_vlan(port->name, NULL, 0);
3468 proc_net_compat_update_bond(port->name, NULL);
3470 for (i = 0; i < MAX_MIRRORS; i++) {
3471 struct mirror *m = br->mirrors[i];
3472 if (m && m->out_port == port) {
3477 while (port->n_ifaces > 0) {
3478 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3481 shash_find_and_delete_assert(&br->port_by_name, port->name);
3483 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3484 del->port_idx = port->port_idx;
3487 bitmap_free(port->trunks);
3494 static struct port *
3495 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3497 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3498 return iface ? iface->port : NULL;
3501 static struct port *
3502 port_lookup(const struct bridge *br, const char *name)
3504 return shash_find_data(&br->port_by_name, name);
3507 static struct iface *
3508 port_lookup_iface(const struct port *port, const char *name)
3510 struct iface *iface = iface_lookup(port->bridge, name);
3511 return iface && iface->port == port ? iface : NULL;
3515 port_update_bonding(struct port *port)
3517 if (port->n_ifaces < 2) {
3518 /* Not a bonded port. */
3519 if (port->bond_hash) {
3520 free(port->bond_hash);
3521 port->bond_hash = NULL;
3522 port->bond_compat_is_stale = true;
3523 port->bond_fake_iface = false;
3526 if (!port->bond_hash) {
3529 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3530 for (i = 0; i <= BOND_MASK; i++) {
3531 struct bond_entry *e = &port->bond_hash[i];
3535 port->no_ifaces_tag = tag_create_random();
3536 bond_choose_active_iface(port);
3537 port->bond_next_rebalance
3538 = time_msec() + port->bond_rebalance_interval;
3540 if (port->cfg->bond_fake_iface) {
3541 port->bond_next_fake_iface_update = time_msec();
3544 port->bond_compat_is_stale = true;
3545 port->bond_fake_iface = port->cfg->bond_fake_iface;
3550 port_update_bond_compat(struct port *port)
3552 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3553 struct compat_bond bond;
3556 if (port->n_ifaces < 2) {
3557 proc_net_compat_update_bond(port->name, NULL);
3562 bond.updelay = port->updelay;
3563 bond.downdelay = port->downdelay;
3566 bond.hashes = compat_hashes;
3567 if (port->bond_hash) {
3568 const struct bond_entry *e;
3569 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3570 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3571 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3572 cbh->hash = e - port->bond_hash;
3573 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3578 bond.n_slaves = port->n_ifaces;
3579 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3580 for (i = 0; i < port->n_ifaces; i++) {
3581 struct iface *iface = port->ifaces[i];
3582 struct compat_bond_slave *slave = &bond.slaves[i];
3583 slave->name = iface->name;
3585 /* We need to make the same determination as the Linux bonding
3586 * code to determine whether a slave should be consider "up".
3587 * The Linux function bond_miimon_inspect() supports four
3588 * BOND_LINK_* states:
3590 * - BOND_LINK_UP: carrier detected, updelay has passed.
3591 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3592 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3593 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3595 * The function bond_info_show_slave() only considers BOND_LINK_UP
3596 * to be "up" and anything else to be "down".
3598 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3602 netdev_get_etheraddr(iface->netdev, slave->mac);
3605 if (port->bond_fake_iface) {
3606 struct netdev *bond_netdev;
3608 if (!netdev_open_default(port->name, &bond_netdev)) {
3610 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3612 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3614 netdev_close(bond_netdev);
3618 proc_net_compat_update_bond(port->name, &bond);
3623 port_update_vlan_compat(struct port *port)
3625 struct bridge *br = port->bridge;
3626 char *vlandev_name = NULL;
3628 if (port->vlan > 0) {
3629 /* Figure out the name that the VLAN device should actually have, if it
3630 * existed. This takes some work because the VLAN device would not
3631 * have port->name in its name; rather, it would have the trunk port's
3632 * name, and 'port' would be attached to a bridge that also had the
3633 * VLAN device one of its ports. So we need to find a trunk port that
3634 * includes port->vlan.
3636 * There might be more than one candidate. This doesn't happen on
3637 * XenServer, so if it happens we just pick the first choice in
3638 * alphabetical order instead of creating multiple VLAN devices. */
3640 for (i = 0; i < br->n_ports; i++) {
3641 struct port *p = br->ports[i];
3642 if (port_trunks_vlan(p, port->vlan)
3644 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3646 uint8_t ea[ETH_ADDR_LEN];
3647 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3648 if (!eth_addr_is_multicast(ea) &&
3649 !eth_addr_is_reserved(ea) &&
3650 !eth_addr_is_zero(ea)) {
3651 vlandev_name = p->name;
3656 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3659 /* Interface functions. */
3661 static struct iface *
3662 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3664 struct bridge *br = port->bridge;
3665 struct iface *iface;
3666 char *name = if_cfg->name;
3669 iface = xzalloc(sizeof *iface);
3671 iface->port_ifidx = port->n_ifaces;
3672 iface->name = xstrdup(name);
3673 iface->dp_ifidx = -1;
3674 iface->tag = tag_create_random();
3675 iface->delay_expires = LLONG_MAX;
3676 iface->netdev = NULL;
3677 iface->cfg = if_cfg;
3679 shash_add_assert(&br->iface_by_name, iface->name, iface);
3681 /* Attempt to create the network interface in case it doesn't exist yet. */
3682 if (!iface_is_internal(br, iface->name)) {
3683 error = set_up_iface(if_cfg, iface, true);
3685 VLOG_WARN("could not create iface %s: %s", iface->name,
3688 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3695 if (port->n_ifaces >= port->allocated_ifaces) {
3696 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3697 sizeof *port->ifaces);
3699 port->ifaces[port->n_ifaces++] = iface;
3700 if (port->n_ifaces > 1) {
3701 br->has_bonded_ports = true;
3704 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3712 iface_destroy(struct iface *iface)
3715 struct port *port = iface->port;
3716 struct bridge *br = port->bridge;
3717 bool del_active = port->active_iface == iface->port_ifidx;
3720 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3722 if (iface->dp_ifidx >= 0) {
3723 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3726 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3727 del->port_ifidx = iface->port_ifidx;
3729 netdev_close(iface->netdev);
3732 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3733 bond_choose_active_iface(port);
3734 bond_send_learning_packets(port);
3740 bridge_flush(port->bridge);
3744 static struct iface *
3745 iface_lookup(const struct bridge *br, const char *name)
3747 return shash_find_data(&br->iface_by_name, name);
3750 static struct iface *
3751 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3753 return port_array_get(&br->ifaces, dp_ifidx);
3756 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3757 * 'br', that is, an interface that is entirely simulated within the datapath.
3758 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3759 * interfaces are created by setting "iface.<iface>.internal = true".
3761 * In addition, we have a kluge-y feature that creates an internal port with
3762 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3763 * This feature needs to go away in the long term. Until then, this is one
3764 * reason why this function takes a name instead of a struct iface: the fake
3765 * interfaces created this way do not have a struct iface. */
3767 iface_is_internal(const struct bridge *br, const char *if_name)
3769 struct iface *iface;
3772 if (!strcmp(if_name, br->name)) {
3776 iface = iface_lookup(br, if_name);
3777 if (iface && !strcmp(iface->cfg->type, "internal")) {
3781 port = port_lookup(br, if_name);
3782 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3788 /* Set Ethernet address of 'iface', if one is specified in the configuration
3791 iface_set_mac(struct iface *iface)
3793 uint8_t ea[ETH_ADDR_LEN];
3795 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3796 if (eth_addr_is_multicast(ea)) {
3797 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3799 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3800 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3801 iface->name, iface->name);
3803 int error = netdev_set_etheraddr(iface->netdev, ea);
3805 VLOG_ERR("interface %s: setting MAC failed (%s)",
3806 iface->name, strerror(error));
3813 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3814 struct shash *shash)
3819 for (i = 0; i < n; i++) {
3820 shash_add(shash, keys[i], values[i]);
3824 struct iface_delete_queues_cbdata {
3825 struct netdev *netdev;
3826 const struct ovsdb_datum *queues;
3830 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
3832 union ovsdb_atom atom;
3834 atom.integer = target;
3835 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
3839 iface_delete_queues(unsigned int queue_id,
3840 const struct shash *details OVS_UNUSED, void *cbdata_)
3842 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3844 if (!queue_ids_include(cbdata->queues, queue_id)) {
3845 netdev_delete_queue(cbdata->netdev, queue_id);
3850 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3852 if (!qos || qos->type[0] == '\0') {
3853 netdev_set_qos(iface->netdev, NULL, NULL);
3855 struct iface_delete_queues_cbdata cbdata;
3856 struct shash details;
3859 /* Configure top-level Qos for 'iface'. */
3860 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3861 qos->n_other_config, &details);
3862 netdev_set_qos(iface->netdev, qos->type, &details);
3863 shash_destroy(&details);
3865 /* Deconfigure queues that were deleted. */
3866 cbdata.netdev = iface->netdev;
3867 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
3869 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3871 /* Configure queues for 'iface'. */
3872 for (i = 0; i < qos->n_queues; i++) {
3873 const struct ovsrec_queue *queue = qos->value_queues[i];
3874 unsigned int queue_id = qos->key_queues[i];
3876 shash_from_ovs_idl_map(queue->key_other_config,
3877 queue->value_other_config,
3878 queue->n_other_config, &details);
3879 netdev_set_queue(iface->netdev, queue_id, &details);
3880 shash_destroy(&details);
3885 /* Port mirroring. */
3887 static struct mirror *
3888 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
3892 for (i = 0; i < MAX_MIRRORS; i++) {
3893 struct mirror *m = br->mirrors[i];
3894 if (m && uuid_equals(uuid, &m->uuid)) {
3902 mirror_reconfigure(struct bridge *br)
3904 unsigned long *rspan_vlans;
3907 /* Get rid of deleted mirrors. */
3908 for (i = 0; i < MAX_MIRRORS; i++) {
3909 struct mirror *m = br->mirrors[i];
3911 const struct ovsdb_datum *mc;
3912 union ovsdb_atom atom;
3914 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
3915 atom.uuid = br->mirrors[i]->uuid;
3916 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
3922 /* Add new mirrors and reconfigure existing ones. */
3923 for (i = 0; i < br->cfg->n_mirrors; i++) {
3924 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3925 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
3927 mirror_reconfigure_one(m, cfg);
3929 mirror_create(br, cfg);
3933 /* Update port reserved status. */
3934 for (i = 0; i < br->n_ports; i++) {
3935 br->ports[i]->is_mirror_output_port = false;
3937 for (i = 0; i < MAX_MIRRORS; i++) {
3938 struct mirror *m = br->mirrors[i];
3939 if (m && m->out_port) {
3940 m->out_port->is_mirror_output_port = true;
3944 /* Update flooded vlans (for RSPAN). */
3946 if (br->cfg->n_flood_vlans) {
3947 rspan_vlans = bitmap_allocate(4096);
3949 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3950 int64_t vlan = br->cfg->flood_vlans[i];
3951 if (vlan >= 0 && vlan < 4096) {
3952 bitmap_set1(rspan_vlans, vlan);
3953 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3956 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3961 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3967 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
3972 for (i = 0; ; i++) {
3973 if (i >= MAX_MIRRORS) {
3974 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3975 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
3978 if (!br->mirrors[i]) {
3983 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
3986 br->mirrors[i] = m = xzalloc(sizeof *m);
3989 m->name = xstrdup(cfg->name);
3990 shash_init(&m->src_ports);
3991 shash_init(&m->dst_ports);
3997 mirror_reconfigure_one(m, cfg);
4001 mirror_destroy(struct mirror *m)
4004 struct bridge *br = m->bridge;
4007 for (i = 0; i < br->n_ports; i++) {
4008 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4009 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4012 shash_destroy(&m->src_ports);
4013 shash_destroy(&m->dst_ports);
4016 m->bridge->mirrors[m->idx] = NULL;
4025 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4026 struct shash *names)
4030 for (i = 0; i < n_ports; i++) {
4031 const char *name = ports[i]->name;
4032 if (port_lookup(m->bridge, name)) {
4033 shash_add_once(names, name, NULL);
4035 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4036 "port %s", m->bridge->name, m->name, name);
4042 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4048 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4050 for (i = 0; i < cfg->n_select_vlan; i++) {
4051 int64_t vlan = cfg->select_vlan[i];
4052 if (vlan < 0 || vlan > 4095) {
4053 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4054 m->bridge->name, m->name, vlan);
4056 (*vlans)[n_vlans++] = vlan;
4063 vlan_is_mirrored(const struct mirror *m, int vlan)
4067 for (i = 0; i < m->n_vlans; i++) {
4068 if (m->vlans[i] == vlan) {
4076 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4080 for (i = 0; i < m->n_vlans; i++) {
4081 if (port_trunks_vlan(p, m->vlans[i])) {
4089 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4091 struct shash src_ports, dst_ports;
4092 mirror_mask_t mirror_bit;
4093 struct port *out_port;
4100 if (strcmp(cfg->name, m->name)) {
4102 m->name = xstrdup(cfg->name);
4105 /* Get output port. */
4106 if (cfg->output_port) {
4107 out_port = port_lookup(m->bridge, cfg->output_port->name);
4109 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4110 m->bridge->name, m->name);
4116 if (cfg->output_vlan) {
4117 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4118 "output vlan; ignoring output vlan",
4119 m->bridge->name, m->name);
4121 } else if (cfg->output_vlan) {
4123 out_vlan = *cfg->output_vlan;
4125 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4126 m->bridge->name, m->name);
4131 shash_init(&src_ports);
4132 shash_init(&dst_ports);
4133 if (cfg->select_all) {
4134 for (i = 0; i < m->bridge->n_ports; i++) {
4135 const char *name = m->bridge->ports[i]->name;
4136 shash_add_once(&src_ports, name, NULL);
4137 shash_add_once(&dst_ports, name, NULL);
4142 /* Get ports, and drop duplicates and ports that don't exist. */
4143 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4145 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4148 /* Get all the vlans, and drop duplicate and invalid vlans. */
4149 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4152 /* Update mirror data. */
4153 if (!shash_equal_keys(&m->src_ports, &src_ports)
4154 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4155 || m->n_vlans != n_vlans
4156 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4157 || m->out_port != out_port
4158 || m->out_vlan != out_vlan) {
4159 bridge_flush(m->bridge);
4161 shash_swap(&m->src_ports, &src_ports);
4162 shash_swap(&m->dst_ports, &dst_ports);
4165 m->n_vlans = n_vlans;
4166 m->out_port = out_port;
4167 m->out_vlan = out_vlan;
4170 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4171 for (i = 0; i < m->bridge->n_ports; i++) {
4172 struct port *port = m->bridge->ports[i];
4174 if (shash_find(&m->src_ports, port->name)
4177 ? port_trunks_any_mirrored_vlan(m, port)
4178 : vlan_is_mirrored(m, port->vlan)))) {
4179 port->src_mirrors |= mirror_bit;
4181 port->src_mirrors &= ~mirror_bit;
4184 if (shash_find(&m->dst_ports, port->name)) {
4185 port->dst_mirrors |= mirror_bit;
4187 port->dst_mirrors &= ~mirror_bit;
4192 shash_destroy(&src_ports);
4193 shash_destroy(&dst_ports);