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 systems and interface
193 * statistics and pushes them into the database. */
194 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
195 static long long int 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 stats_timer = time_msec() + 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]));
377 /* Include 'other_config' keys in hash of netdev options. The
378 * namespace of 'other_config' and 'options' must be disjoint.
379 * Prefer 'options' keys over 'other_config' keys. */
380 for (i = 0; i < iface_cfg->n_other_config; i++) {
381 char *value = xstrdup(iface_cfg->value_other_config[i]);
382 if (!shash_add_once(&options, iface_cfg->key_other_config[i],
384 VLOG_WARN("%s: \"other_config\" key %s conflicts with existing "
385 "\"other_config\" or \"options\" entry...ignoring",
386 iface_cfg->name, iface_cfg->key_other_config[i]);
392 struct netdev_options netdev_options;
394 memset(&netdev_options, 0, sizeof netdev_options);
395 netdev_options.name = iface_cfg->name;
396 if (!strcmp(iface_cfg->type, "internal")) {
397 /* An "internal" config type maps to a netdev "system" type. */
398 netdev_options.type = "system";
400 netdev_options.type = iface_cfg->type;
402 netdev_options.args = &options;
403 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
405 error = netdev_open(&netdev_options, &iface->netdev);
408 netdev_get_carrier(iface->netdev, &iface->enabled);
410 } else if (iface->netdev) {
411 const char *netdev_type = netdev_get_type(iface->netdev);
412 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
413 ? iface_cfg->type : NULL;
415 /* An "internal" config type maps to a netdev "system" type. */
416 if (iface_type && !strcmp(iface_type, "internal")) {
417 iface_type = "system";
420 if (!iface_type || !strcmp(netdev_type, iface_type)) {
421 error = netdev_reconfigure(iface->netdev, &options);
423 VLOG_WARN("%s: attempting change device type from %s to %s",
424 iface_cfg->name, netdev_type, iface_type);
428 shash_destroy_free_data(&options);
434 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
436 return set_up_iface(iface_cfg, iface, false);
440 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
441 void *aux OVS_UNUSED)
443 if (!iface->netdev) {
444 int error = set_up_iface(iface->cfg, iface, true);
446 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
456 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
457 void *aux OVS_UNUSED)
459 if (iface->dp_ifidx >= 0) {
460 VLOG_DBG("%s has interface %s on port %d",
462 iface->name, iface->dp_ifidx);
465 VLOG_ERR("%s interface not in %s, dropping",
466 iface->name, dpif_name(br->dpif));
472 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
473 void *aux OVS_UNUSED)
475 /* Set policing attributes. */
476 netdev_set_policing(iface->netdev,
477 iface->cfg->ingress_policing_rate,
478 iface->cfg->ingress_policing_burst);
480 /* Set MAC address of internal interfaces other than the local
482 if (iface->dp_ifidx != ODPP_LOCAL
483 && iface_is_internal(br, iface->name)) {
484 iface_set_mac(iface);
490 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
491 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
492 * deletes from 'br' any ports that no longer have any interfaces. */
494 iterate_and_prune_ifaces(struct bridge *br,
495 bool (*cb)(struct bridge *, struct iface *,
501 for (i = 0; i < br->n_ports; ) {
502 struct port *port = br->ports[i];
503 for (j = 0; j < port->n_ifaces; ) {
504 struct iface *iface = port->ifaces[j];
505 if (cb(br, iface, aux)) {
508 iface_destroy(iface);
512 if (port->n_ifaces) {
515 VLOG_ERR("%s port has no interfaces, dropping", port->name);
521 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
522 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
523 * responsible for freeing '*managersp' (with free()).
525 * You may be asking yourself "why does ovs-vswitchd care?", because
526 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
527 * should not be and in fact is not directly involved in that. But
528 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
529 * it has to tell in-band control where the managers are to enable that.
532 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
533 struct sockaddr_in **managersp, size_t *n_managersp)
535 struct sockaddr_in *managers = NULL;
536 size_t n_managers = 0;
538 if (ovs_cfg->n_managers > 0) {
541 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
542 for (i = 0; i < ovs_cfg->n_managers; i++) {
543 const char *name = ovs_cfg->managers[i];
544 struct sockaddr_in *sin = &managers[i];
546 if ((!strncmp(name, "tcp:", 4)
547 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
548 (!strncmp(name, "ssl:", 4)
549 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
555 *managersp = managers;
556 *n_managersp = n_managers;
560 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
562 struct shash old_br, new_br;
563 struct shash_node *node;
564 struct bridge *br, *next;
565 struct sockaddr_in *managers;
568 int sflow_bridge_number;
570 COVERAGE_INC(bridge_reconfigure);
572 collect_managers(ovs_cfg, &managers, &n_managers);
574 /* Collect old and new bridges. */
577 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
578 shash_add(&old_br, br->name, br);
580 for (i = 0; i < ovs_cfg->n_bridges; i++) {
581 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
582 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
583 VLOG_WARN("more than one bridge named %s", br_cfg->name);
587 /* Get rid of deleted bridges and add new bridges. */
588 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
589 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
596 SHASH_FOR_EACH (node, &new_br) {
597 const char *br_name = node->name;
598 const struct ovsrec_bridge *br_cfg = node->data;
599 br = shash_find_data(&old_br, br_name);
601 /* If the bridge datapath type has changed, we need to tear it
602 * down and recreate. */
603 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
605 bridge_create(br_cfg);
608 bridge_create(br_cfg);
611 shash_destroy(&old_br);
612 shash_destroy(&new_br);
614 /* Reconfigure all bridges. */
615 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
616 bridge_reconfigure_one(br);
619 /* Add and delete ports on all datapaths.
621 * The kernel will reject any attempt to add a given port to a datapath if
622 * that port already belongs to a different datapath, so we must do all
623 * port deletions before any port additions. */
624 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
625 struct odp_port *dpif_ports;
627 struct shash want_ifaces;
629 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
630 bridge_get_all_ifaces(br, &want_ifaces);
631 for (i = 0; i < n_dpif_ports; i++) {
632 const struct odp_port *p = &dpif_ports[i];
633 if (!shash_find(&want_ifaces, p->devname)
634 && strcmp(p->devname, br->name)) {
635 int retval = dpif_port_del(br->dpif, p->port);
637 VLOG_ERR("failed to remove %s interface from %s: %s",
638 p->devname, dpif_name(br->dpif),
643 shash_destroy(&want_ifaces);
646 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
647 struct odp_port *dpif_ports;
649 struct shash cur_ifaces, want_ifaces;
651 /* Get the set of interfaces currently in this datapath. */
652 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
653 shash_init(&cur_ifaces);
654 for (i = 0; i < n_dpif_ports; i++) {
655 const char *name = dpif_ports[i].devname;
656 shash_add_once(&cur_ifaces, name, NULL);
660 /* Get the set of interfaces we want on this datapath. */
661 bridge_get_all_ifaces(br, &want_ifaces);
663 SHASH_FOR_EACH (node, &want_ifaces) {
664 const char *if_name = node->name;
665 struct iface *iface = node->data;
667 if (shash_find(&cur_ifaces, if_name)) {
668 /* Already exists, just reconfigure it. */
670 reconfigure_iface(iface->cfg, iface);
673 /* Need to add to datapath. */
677 /* Add to datapath. */
678 internal = iface_is_internal(br, if_name);
679 error = dpif_port_add(br->dpif, if_name,
680 internal ? ODP_PORT_INTERNAL : 0, NULL);
681 if (error == EFBIG) {
682 VLOG_ERR("ran out of valid port numbers on %s",
683 dpif_name(br->dpif));
686 VLOG_ERR("failed to add %s interface to %s: %s",
687 if_name, dpif_name(br->dpif), strerror(error));
691 shash_destroy(&cur_ifaces);
692 shash_destroy(&want_ifaces);
694 sflow_bridge_number = 0;
695 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
698 struct iface *local_iface;
699 struct iface *hw_addr_iface;
702 bridge_fetch_dp_ifaces(br);
704 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
705 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
707 /* Pick local port hardware address, datapath ID. */
708 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
709 local_iface = bridge_get_local_iface(br);
711 int error = netdev_set_etheraddr(local_iface->netdev, ea);
713 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
714 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
715 "Ethernet address: %s",
716 br->name, strerror(error));
720 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
721 ofproto_set_datapath_id(br->ofproto, dpid);
723 dpid_string = xasprintf("%016"PRIx64, dpid);
724 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
727 /* Set NetFlow configuration on this bridge. */
728 if (br->cfg->netflow) {
729 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
730 struct netflow_options opts;
732 memset(&opts, 0, sizeof opts);
734 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
735 if (nf_cfg->engine_type) {
736 opts.engine_type = *nf_cfg->engine_type;
738 if (nf_cfg->engine_id) {
739 opts.engine_id = *nf_cfg->engine_id;
742 opts.active_timeout = nf_cfg->active_timeout;
743 if (!opts.active_timeout) {
744 opts.active_timeout = -1;
745 } else if (opts.active_timeout < 0) {
746 VLOG_WARN("bridge %s: active timeout interval set to negative "
747 "value, using default instead (%d seconds)", br->name,
748 NF_ACTIVE_TIMEOUT_DEFAULT);
749 opts.active_timeout = -1;
752 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
753 if (opts.add_id_to_iface) {
754 if (opts.engine_id > 0x7f) {
755 VLOG_WARN("bridge %s: netflow port mangling may conflict "
756 "with another vswitch, choose an engine id less "
757 "than 128", br->name);
759 if (br->n_ports > 508) {
760 VLOG_WARN("bridge %s: netflow port mangling will conflict "
761 "with another port when more than 508 ports are "
766 opts.collectors.n = nf_cfg->n_targets;
767 opts.collectors.names = nf_cfg->targets;
768 if (ofproto_set_netflow(br->ofproto, &opts)) {
769 VLOG_ERR("bridge %s: problem setting netflow collectors",
773 ofproto_set_netflow(br->ofproto, NULL);
776 /* Set sFlow configuration on this bridge. */
777 if (br->cfg->sflow) {
778 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
779 struct ovsrec_controller **controllers;
780 struct ofproto_sflow_options oso;
781 size_t n_controllers;
783 memset(&oso, 0, sizeof oso);
785 oso.targets.n = sflow_cfg->n_targets;
786 oso.targets.names = sflow_cfg->targets;
788 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
789 if (sflow_cfg->sampling) {
790 oso.sampling_rate = *sflow_cfg->sampling;
793 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
794 if (sflow_cfg->polling) {
795 oso.polling_interval = *sflow_cfg->polling;
798 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
799 if (sflow_cfg->header) {
800 oso.header_len = *sflow_cfg->header;
803 oso.sub_id = sflow_bridge_number++;
804 oso.agent_device = sflow_cfg->agent;
806 oso.control_ip = NULL;
807 n_controllers = bridge_get_controllers(br, &controllers);
808 for (i = 0; i < n_controllers; i++) {
809 if (controllers[i]->local_ip) {
810 oso.control_ip = controllers[i]->local_ip;
814 ofproto_set_sflow(br->ofproto, &oso);
816 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
818 ofproto_set_sflow(br->ofproto, NULL);
821 /* Update the controller and related settings. It would be more
822 * straightforward to call this from bridge_reconfigure_one(), but we
823 * can't do it there for two reasons. First, and most importantly, at
824 * that point we don't know the dp_ifidx of any interfaces that have
825 * been added to the bridge (because we haven't actually added them to
826 * the datapath). Second, at that point we haven't set the datapath ID
827 * yet; when a controller is configured, resetting the datapath ID will
828 * immediately disconnect from the controller, so it's better to set
829 * the datapath ID before the controller. */
830 bridge_reconfigure_remotes(br, managers, n_managers);
832 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
833 for (i = 0; i < br->n_ports; i++) {
834 struct port *port = br->ports[i];
837 port_update_vlan_compat(port);
838 port_update_bonding(port);
840 for (j = 0; j < port->n_ifaces; j++) {
841 iface_update_qos(port->ifaces[j], port->cfg->qos);
845 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
846 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
853 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
854 const struct ovsdb_idl_column *column,
857 const struct ovsdb_datum *datum;
858 union ovsdb_atom atom;
861 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
862 atom.string = (char *) key;
863 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
864 return idx == UINT_MAX ? NULL : datum->values[idx].string;
868 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
870 return get_ovsrec_key_value(&br_cfg->header_,
871 &ovsrec_bridge_col_other_config, key);
875 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
876 struct iface **hw_addr_iface)
882 *hw_addr_iface = NULL;
884 /* Did the user request a particular MAC? */
885 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
886 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
887 if (eth_addr_is_multicast(ea)) {
888 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
889 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
890 } else if (eth_addr_is_zero(ea)) {
891 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
897 /* Otherwise choose the minimum non-local MAC address among all of the
899 memset(ea, 0xff, sizeof ea);
900 for (i = 0; i < br->n_ports; i++) {
901 struct port *port = br->ports[i];
902 uint8_t iface_ea[ETH_ADDR_LEN];
905 /* Mirror output ports don't participate. */
906 if (port->is_mirror_output_port) {
910 /* Choose the MAC address to represent the port. */
911 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
912 /* Find the interface with this Ethernet address (if any) so that
913 * we can provide the correct devname to the caller. */
915 for (j = 0; j < port->n_ifaces; j++) {
916 struct iface *candidate = port->ifaces[j];
917 uint8_t candidate_ea[ETH_ADDR_LEN];
918 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
919 && eth_addr_equals(iface_ea, candidate_ea)) {
924 /* Choose the interface whose MAC address will represent the port.
925 * The Linux kernel bonding code always chooses the MAC address of
926 * the first slave added to a bond, and the Fedora networking
927 * scripts always add slaves to a bond in alphabetical order, so
928 * for compatibility we choose the interface with the name that is
929 * first in alphabetical order. */
930 iface = port->ifaces[0];
931 for (j = 1; j < port->n_ifaces; j++) {
932 struct iface *candidate = port->ifaces[j];
933 if (strcmp(candidate->name, iface->name) < 0) {
938 /* The local port doesn't count (since we're trying to choose its
939 * MAC address anyway). */
940 if (iface->dp_ifidx == ODPP_LOCAL) {
945 error = netdev_get_etheraddr(iface->netdev, iface_ea);
947 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
948 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
949 iface->name, strerror(error));
954 /* Compare against our current choice. */
955 if (!eth_addr_is_multicast(iface_ea) &&
956 !eth_addr_is_local(iface_ea) &&
957 !eth_addr_is_reserved(iface_ea) &&
958 !eth_addr_is_zero(iface_ea) &&
959 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
961 memcpy(ea, iface_ea, ETH_ADDR_LEN);
962 *hw_addr_iface = iface;
965 if (eth_addr_is_multicast(ea)) {
966 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
967 *hw_addr_iface = NULL;
968 VLOG_WARN("bridge %s: using default bridge Ethernet "
969 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
971 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
972 br->name, ETH_ADDR_ARGS(ea));
976 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
977 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
978 * an interface on 'br', then that interface must be passed in as
979 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
980 * 'hw_addr_iface' must be passed in as a null pointer. */
982 bridge_pick_datapath_id(struct bridge *br,
983 const uint8_t bridge_ea[ETH_ADDR_LEN],
984 struct iface *hw_addr_iface)
987 * The procedure for choosing a bridge MAC address will, in the most
988 * ordinary case, also choose a unique MAC that we can use as a datapath
989 * ID. In some special cases, though, multiple bridges will end up with
990 * the same MAC address. This is OK for the bridges, but it will confuse
991 * the OpenFlow controller, because each datapath needs a unique datapath
994 * Datapath IDs must be unique. It is also very desirable that they be
995 * stable from one run to the next, so that policy set on a datapath
998 const char *datapath_id;
1001 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1002 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1006 if (hw_addr_iface) {
1008 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1010 * A bridge whose MAC address is taken from a VLAN network device
1011 * (that is, a network device created with vconfig(8) or similar
1012 * tool) will have the same MAC address as a bridge on the VLAN
1013 * device's physical network device.
1015 * Handle this case by hashing the physical network device MAC
1016 * along with the VLAN identifier.
1018 uint8_t buf[ETH_ADDR_LEN + 2];
1019 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1020 buf[ETH_ADDR_LEN] = vlan >> 8;
1021 buf[ETH_ADDR_LEN + 1] = vlan;
1022 return dpid_from_hash(buf, sizeof buf);
1025 * Assume that this bridge's MAC address is unique, since it
1026 * doesn't fit any of the cases we handle specially.
1031 * A purely internal bridge, that is, one that has no non-virtual
1032 * network devices on it at all, is more difficult because it has no
1033 * natural unique identifier at all.
1035 * When the host is a XenServer, we handle this case by hashing the
1036 * host's UUID with the name of the bridge. Names of bridges are
1037 * persistent across XenServer reboots, although they can be reused if
1038 * an internal network is destroyed and then a new one is later
1039 * created, so this is fairly effective.
1041 * When the host is not a XenServer, we punt by using a random MAC
1042 * address on each run.
1044 const char *host_uuid = xenserver_get_host_uuid();
1046 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1047 dpid = dpid_from_hash(combined, strlen(combined));
1053 return eth_addr_to_uint64(bridge_ea);
1057 dpid_from_hash(const void *data, size_t n)
1059 uint8_t hash[SHA1_DIGEST_SIZE];
1061 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1062 sha1_bytes(data, n, hash);
1063 eth_addr_mark_random(hash);
1064 return eth_addr_to_uint64(hash);
1068 iface_refresh_stats(struct iface *iface)
1074 static const struct iface_stat iface_stats[] = {
1075 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1076 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1077 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1078 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1079 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1080 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1081 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1082 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1083 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1084 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1085 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1086 { "collisions", offsetof(struct netdev_stats, collisions) },
1088 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1089 const struct iface_stat *s;
1091 char *keys[N_STATS];
1092 int64_t values[N_STATS];
1095 struct netdev_stats stats;
1097 /* Intentionally ignore return value, since errors will set 'stats' to
1098 * all-1s, and we will deal with that correctly below. */
1099 netdev_get_stats(iface->netdev, &stats);
1102 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1103 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1104 if (value != UINT64_MAX) {
1111 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1115 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1117 struct ovsdb_datum datum;
1121 get_system_stats(&stats);
1123 ovsdb_datum_from_shash(&datum, &stats);
1124 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1131 const struct ovsrec_open_vswitch *cfg;
1133 bool datapath_destroyed;
1134 bool database_changed;
1137 /* Let each bridge do the work that it needs to do. */
1138 datapath_destroyed = false;
1139 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1140 int error = bridge_run_one(br);
1142 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1143 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1144 "forcing reconfiguration", br->name);
1145 datapath_destroyed = true;
1149 /* (Re)configure if necessary. */
1150 database_changed = ovsdb_idl_run(idl);
1151 cfg = ovsrec_open_vswitch_first(idl);
1152 if (database_changed || datapath_destroyed) {
1154 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1156 bridge_configure_once(cfg);
1157 bridge_reconfigure(cfg);
1159 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1160 ovsdb_idl_txn_commit(txn);
1161 ovsdb_idl_txn_destroy(txn); /* XXX */
1163 /* We still need to reconfigure to avoid dangling pointers to
1164 * now-destroyed ovsrec structures inside bridge data. */
1165 static const struct ovsrec_open_vswitch null_cfg;
1167 bridge_reconfigure(&null_cfg);
1172 /* Re-configure SSL. We do this on every trip through the main loop,
1173 * instead of just when the database changes, because the contents of the
1174 * key and certificate files can change without the database changing. */
1175 if (cfg && cfg->ssl) {
1176 const struct ovsrec_ssl *ssl = cfg->ssl;
1178 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1179 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1183 /* Refresh system and interface stats if necessary. */
1184 if (time_msec() >= stats_timer) {
1186 struct ovsdb_idl_txn *txn;
1188 txn = ovsdb_idl_txn_create(idl);
1189 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1192 for (i = 0; i < br->n_ports; i++) {
1193 struct port *port = br->ports[i];
1196 for (j = 0; j < port->n_ifaces; j++) {
1197 struct iface *iface = port->ifaces[j];
1198 iface_refresh_stats(iface);
1202 refresh_system_stats(cfg);
1203 ovsdb_idl_txn_commit(txn);
1204 ovsdb_idl_txn_destroy(txn); /* XXX */
1207 stats_timer = time_msec() + STATS_INTERVAL;
1216 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1217 ofproto_wait(br->ofproto);
1218 if (ofproto_has_primary_controller(br->ofproto)) {
1222 mac_learning_wait(br->ml);
1225 ovsdb_idl_wait(idl);
1226 poll_timer_wait_until(stats_timer);
1229 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1230 * configuration changes. */
1232 bridge_flush(struct bridge *br)
1234 COVERAGE_INC(bridge_flush);
1236 mac_learning_flush(br->ml);
1239 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1240 * such interface. */
1241 static struct iface *
1242 bridge_get_local_iface(struct bridge *br)
1246 for (i = 0; i < br->n_ports; i++) {
1247 struct port *port = br->ports[i];
1248 for (j = 0; j < port->n_ifaces; j++) {
1249 struct iface *iface = port->ifaces[j];
1250 if (iface->dp_ifidx == ODPP_LOCAL) {
1259 /* Bridge unixctl user interface functions. */
1261 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1262 const char *args, void *aux OVS_UNUSED)
1264 struct ds ds = DS_EMPTY_INITIALIZER;
1265 const struct bridge *br;
1266 const struct mac_entry *e;
1268 br = bridge_lookup(args);
1270 unixctl_command_reply(conn, 501, "no such bridge");
1274 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1275 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1276 if (e->port < 0 || e->port >= br->n_ports) {
1279 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1280 br->ports[e->port]->ifaces[0]->dp_ifidx,
1281 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1283 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1287 /* Bridge reconfiguration functions. */
1288 static struct bridge *
1289 bridge_create(const struct ovsrec_bridge *br_cfg)
1294 assert(!bridge_lookup(br_cfg->name));
1295 br = xzalloc(sizeof *br);
1297 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1303 dpif_flow_flush(br->dpif);
1305 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1308 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1310 dpif_delete(br->dpif);
1311 dpif_close(br->dpif);
1316 br->name = xstrdup(br_cfg->name);
1318 br->ml = mac_learning_create();
1319 eth_addr_nicira_random(br->default_ea);
1321 port_array_init(&br->ifaces);
1323 shash_init(&br->port_by_name);
1324 shash_init(&br->iface_by_name);
1328 list_push_back(&all_bridges, &br->node);
1330 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1336 bridge_destroy(struct bridge *br)
1341 while (br->n_ports > 0) {
1342 port_destroy(br->ports[br->n_ports - 1]);
1344 list_remove(&br->node);
1345 error = dpif_delete(br->dpif);
1346 if (error && error != ENOENT) {
1347 VLOG_ERR("failed to delete %s: %s",
1348 dpif_name(br->dpif), strerror(error));
1350 dpif_close(br->dpif);
1351 ofproto_destroy(br->ofproto);
1352 mac_learning_destroy(br->ml);
1353 port_array_destroy(&br->ifaces);
1354 shash_destroy(&br->port_by_name);
1355 shash_destroy(&br->iface_by_name);
1362 static struct bridge *
1363 bridge_lookup(const char *name)
1367 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1368 if (!strcmp(br->name, name)) {
1375 /* Handle requests for a listing of all flows known by the OpenFlow
1376 * stack, including those normally hidden. */
1378 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1379 const char *args, void *aux OVS_UNUSED)
1384 br = bridge_lookup(args);
1386 unixctl_command_reply(conn, 501, "Unknown bridge");
1391 ofproto_get_all_flows(br->ofproto, &results);
1393 unixctl_command_reply(conn, 200, ds_cstr(&results));
1394 ds_destroy(&results);
1397 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1398 * connections and reconnect. If BRIDGE is not specified, then all bridges
1399 * drop their controller connections and reconnect. */
1401 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1402 const char *args, void *aux OVS_UNUSED)
1405 if (args[0] != '\0') {
1406 br = bridge_lookup(args);
1408 unixctl_command_reply(conn, 501, "Unknown bridge");
1411 ofproto_reconnect_controllers(br->ofproto);
1413 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1414 ofproto_reconnect_controllers(br->ofproto);
1417 unixctl_command_reply(conn, 200, NULL);
1421 bridge_run_one(struct bridge *br)
1425 error = ofproto_run1(br->ofproto);
1430 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1433 error = ofproto_run2(br->ofproto, br->flush);
1440 bridge_get_controllers(const struct bridge *br,
1441 struct ovsrec_controller ***controllersp)
1443 struct ovsrec_controller **controllers;
1444 size_t n_controllers;
1446 controllers = br->cfg->controller;
1447 n_controllers = br->cfg->n_controller;
1449 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1455 *controllersp = controllers;
1457 return n_controllers;
1461 bridge_reconfigure_one(struct bridge *br)
1463 struct shash old_ports, new_ports;
1464 struct svec snoops, old_snoops;
1465 struct shash_node *node;
1466 enum ofproto_fail_mode fail_mode;
1469 /* Collect old ports. */
1470 shash_init(&old_ports);
1471 for (i = 0; i < br->n_ports; i++) {
1472 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1475 /* Collect new ports. */
1476 shash_init(&new_ports);
1477 for (i = 0; i < br->cfg->n_ports; i++) {
1478 const char *name = br->cfg->ports[i]->name;
1479 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1480 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1485 /* If we have a controller, then we need a local port. Complain if the
1486 * user didn't specify one.
1488 * XXX perhaps we should synthesize a port ourselves in this case. */
1489 if (bridge_get_controllers(br, NULL)) {
1490 char local_name[IF_NAMESIZE];
1493 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1494 local_name, sizeof local_name);
1495 if (!error && !shash_find(&new_ports, local_name)) {
1496 VLOG_WARN("bridge %s: controller specified but no local port "
1497 "(port named %s) defined",
1498 br->name, local_name);
1502 /* Get rid of deleted ports.
1503 * Get rid of deleted interfaces on ports that still exist. */
1504 SHASH_FOR_EACH (node, &old_ports) {
1505 struct port *port = node->data;
1506 const struct ovsrec_port *port_cfg;
1508 port_cfg = shash_find_data(&new_ports, node->name);
1512 port_del_ifaces(port, port_cfg);
1516 /* Create new ports.
1517 * Add new interfaces to existing ports.
1518 * Reconfigure existing ports. */
1519 SHASH_FOR_EACH (node, &new_ports) {
1520 struct port *port = shash_find_data(&old_ports, node->name);
1522 port = port_create(br, node->name);
1525 port_reconfigure(port, node->data);
1526 if (!port->n_ifaces) {
1527 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1528 br->name, port->name);
1532 shash_destroy(&old_ports);
1533 shash_destroy(&new_ports);
1535 /* Set the fail-mode */
1536 fail_mode = !br->cfg->fail_mode
1537 || !strcmp(br->cfg->fail_mode, "standalone")
1538 ? OFPROTO_FAIL_STANDALONE
1539 : OFPROTO_FAIL_SECURE;
1540 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1541 && !ofproto_has_primary_controller(br->ofproto)) {
1542 ofproto_flush_flows(br->ofproto);
1544 ofproto_set_fail_mode(br->ofproto, fail_mode);
1546 /* Delete all flows if we're switching from connected to standalone or vice
1547 * versa. (XXX Should we delete all flows if we are switching from one
1548 * controller to another?) */
1550 /* Configure OpenFlow controller connection snooping. */
1552 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1553 ovs_rundir, br->name));
1554 svec_init(&old_snoops);
1555 ofproto_get_snoops(br->ofproto, &old_snoops);
1556 if (!svec_equal(&snoops, &old_snoops)) {
1557 ofproto_set_snoops(br->ofproto, &snoops);
1559 svec_destroy(&snoops);
1560 svec_destroy(&old_snoops);
1562 mirror_reconfigure(br);
1565 /* Initializes 'oc' appropriately as a management service controller for
1568 * The caller must free oc->target when it is no longer needed. */
1570 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1571 struct ofproto_controller *oc)
1573 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir, br->name);
1574 oc->max_backoff = 0;
1575 oc->probe_interval = 60;
1576 oc->band = OFPROTO_OUT_OF_BAND;
1577 oc->accept_re = NULL;
1578 oc->update_resolv_conf = false;
1580 oc->burst_limit = 0;
1583 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1585 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1586 struct ofproto_controller *oc)
1588 oc->target = c->target;
1589 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1590 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1591 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1592 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1593 oc->accept_re = c->discover_accept_regex;
1594 oc->update_resolv_conf = c->discover_update_resolv_conf;
1595 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1596 oc->burst_limit = (c->controller_burst_limit
1597 ? *c->controller_burst_limit : 0);
1600 /* Configures the IP stack for 'br''s local interface properly according to the
1601 * configuration in 'c'. */
1603 bridge_configure_local_iface_netdev(struct bridge *br,
1604 struct ovsrec_controller *c)
1606 struct netdev *netdev;
1607 struct in_addr mask, gateway;
1609 struct iface *local_iface;
1612 /* Controller discovery does its own TCP/IP configuration later. */
1613 if (strcmp(c->target, "discover")) {
1617 /* If there's no local interface or no IP address, give up. */
1618 local_iface = bridge_get_local_iface(br);
1619 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1623 /* Bring up the local interface. */
1624 netdev = local_iface->netdev;
1625 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1627 /* Configure the IP address and netmask. */
1628 if (!c->local_netmask
1629 || !inet_aton(c->local_netmask, &mask)
1631 mask.s_addr = guess_netmask(ip.s_addr);
1633 if (!netdev_set_in4(netdev, ip, mask)) {
1634 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1635 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1638 /* Configure the default gateway. */
1639 if (c->local_gateway
1640 && inet_aton(c->local_gateway, &gateway)
1641 && gateway.s_addr) {
1642 if (!netdev_add_router(netdev, gateway)) {
1643 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1644 br->name, IP_ARGS(&gateway.s_addr));
1650 bridge_reconfigure_remotes(struct bridge *br,
1651 const struct sockaddr_in *managers,
1654 struct ovsrec_controller **controllers;
1655 size_t n_controllers;
1658 struct ofproto_controller *ocs;
1662 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1663 had_primary = ofproto_has_primary_controller(br->ofproto);
1665 n_controllers = bridge_get_controllers(br, &controllers);
1667 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1670 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1671 for (i = 0; i < n_controllers; i++) {
1672 struct ovsrec_controller *c = controllers[i];
1674 if (!strncmp(c->target, "punix:", 6)
1675 || !strncmp(c->target, "unix:", 5)) {
1676 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1678 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1679 * domain sockets and overwriting arbitrary local files. */
1680 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1681 "\"%s\" due to possibility for remote exploit",
1682 dpif_name(br->dpif), c->target);
1686 bridge_configure_local_iface_netdev(br, c);
1687 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs++]);
1690 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1691 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1694 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1695 ofproto_flush_flows(br->ofproto);
1698 /* If there are no controllers and the bridge is in standalone
1699 * mode, set up a flow that matches every packet and directs
1700 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1701 * switch is in secure mode and we won't pass any traffic until
1702 * a controller has been defined and it tells us to do so. */
1704 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1705 union ofp_action action;
1708 memset(&action, 0, sizeof action);
1709 action.type = htons(OFPAT_OUTPUT);
1710 action.output.len = htons(sizeof action);
1711 action.output.port = htons(OFPP_NORMAL);
1712 memset(&flow, 0, sizeof flow);
1713 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1718 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1723 for (i = 0; i < br->n_ports; i++) {
1724 struct port *port = br->ports[i];
1725 for (j = 0; j < port->n_ifaces; j++) {
1726 struct iface *iface = port->ifaces[j];
1727 shash_add_once(ifaces, iface->name, iface);
1729 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1730 shash_add_once(ifaces, port->name, NULL);
1735 /* For robustness, in case the administrator moves around datapath ports behind
1736 * our back, we re-check all the datapath port numbers here.
1738 * This function will set the 'dp_ifidx' members of interfaces that have
1739 * disappeared to -1, so only call this function from a context where those
1740 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1741 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1742 * datapath, which doesn't support UINT16_MAX+1 ports. */
1744 bridge_fetch_dp_ifaces(struct bridge *br)
1746 struct odp_port *dpif_ports;
1747 size_t n_dpif_ports;
1750 /* Reset all interface numbers. */
1751 for (i = 0; i < br->n_ports; i++) {
1752 struct port *port = br->ports[i];
1753 for (j = 0; j < port->n_ifaces; j++) {
1754 struct iface *iface = port->ifaces[j];
1755 iface->dp_ifidx = -1;
1758 port_array_clear(&br->ifaces);
1760 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1761 for (i = 0; i < n_dpif_ports; i++) {
1762 struct odp_port *p = &dpif_ports[i];
1763 struct iface *iface = iface_lookup(br, p->devname);
1765 if (iface->dp_ifidx >= 0) {
1766 VLOG_WARN("%s reported interface %s twice",
1767 dpif_name(br->dpif), p->devname);
1768 } else if (iface_from_dp_ifidx(br, p->port)) {
1769 VLOG_WARN("%s reported interface %"PRIu16" twice",
1770 dpif_name(br->dpif), p->port);
1772 port_array_set(&br->ifaces, p->port, iface);
1773 iface->dp_ifidx = p->port;
1777 int64_t ofport = (iface->dp_ifidx >= 0
1778 ? odp_port_to_ofp_port(iface->dp_ifidx)
1780 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1787 /* Bridge packet processing functions. */
1790 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1792 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1795 static struct bond_entry *
1796 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1798 return &port->bond_hash[bond_hash(mac)];
1802 bond_choose_iface(const struct port *port)
1804 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1805 size_t i, best_down_slave = -1;
1806 long long next_delay_expiration = LLONG_MAX;
1808 for (i = 0; i < port->n_ifaces; i++) {
1809 struct iface *iface = port->ifaces[i];
1811 if (iface->enabled) {
1813 } else if (iface->delay_expires < next_delay_expiration) {
1814 best_down_slave = i;
1815 next_delay_expiration = iface->delay_expires;
1819 if (best_down_slave != -1) {
1820 struct iface *iface = port->ifaces[best_down_slave];
1822 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1823 "since no other interface is up", iface->name,
1824 iface->delay_expires - time_msec());
1825 bond_enable_slave(iface, true);
1828 return best_down_slave;
1832 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1833 uint16_t *dp_ifidx, tag_type *tags)
1835 struct iface *iface;
1837 assert(port->n_ifaces);
1838 if (port->n_ifaces == 1) {
1839 iface = port->ifaces[0];
1841 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1842 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1843 || !port->ifaces[e->iface_idx]->enabled) {
1844 /* XXX select interface properly. The current interface selection
1845 * is only good for testing the rebalancing code. */
1846 e->iface_idx = bond_choose_iface(port);
1847 if (e->iface_idx < 0) {
1848 *tags |= port->no_ifaces_tag;
1851 e->iface_tag = tag_create_random();
1852 ((struct port *) port)->bond_compat_is_stale = true;
1854 *tags |= e->iface_tag;
1855 iface = port->ifaces[e->iface_idx];
1857 *dp_ifidx = iface->dp_ifidx;
1858 *tags |= iface->tag; /* Currently only used for bonding. */
1863 bond_link_status_update(struct iface *iface, bool carrier)
1865 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1866 struct port *port = iface->port;
1868 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1869 /* Nothing to do. */
1872 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1873 iface->name, carrier ? "detected" : "dropped");
1874 if (carrier == iface->enabled) {
1875 iface->delay_expires = LLONG_MAX;
1876 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1877 iface->name, carrier ? "disabled" : "enabled");
1878 } else if (carrier && port->active_iface < 0) {
1879 bond_enable_slave(iface, true);
1880 if (port->updelay) {
1881 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1882 "other interface is up", iface->name, port->updelay);
1885 int delay = carrier ? port->updelay : port->downdelay;
1886 iface->delay_expires = time_msec() + delay;
1889 "interface %s: will be %s if it stays %s for %d ms",
1891 carrier ? "enabled" : "disabled",
1892 carrier ? "up" : "down",
1899 bond_choose_active_iface(struct port *port)
1901 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1903 port->active_iface = bond_choose_iface(port);
1904 port->active_iface_tag = tag_create_random();
1905 if (port->active_iface >= 0) {
1906 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1907 port->name, port->ifaces[port->active_iface]->name);
1909 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1915 bond_enable_slave(struct iface *iface, bool enable)
1917 struct port *port = iface->port;
1918 struct bridge *br = port->bridge;
1920 /* This acts as a recursion check. If the act of disabling a slave
1921 * causes a different slave to be enabled, the flag will allow us to
1922 * skip redundant work when we reenter this function. It must be
1923 * cleared on exit to keep things safe with multiple bonds. */
1924 static bool moving_active_iface = false;
1926 iface->delay_expires = LLONG_MAX;
1927 if (enable == iface->enabled) {
1931 iface->enabled = enable;
1932 if (!iface->enabled) {
1933 VLOG_WARN("interface %s: disabled", iface->name);
1934 ofproto_revalidate(br->ofproto, iface->tag);
1935 if (iface->port_ifidx == port->active_iface) {
1936 ofproto_revalidate(br->ofproto,
1937 port->active_iface_tag);
1939 /* Disabling a slave can lead to another slave being immediately
1940 * enabled if there will be no active slaves but one is waiting
1941 * on an updelay. In this case we do not need to run most of the
1942 * code for the newly enabled slave since there was no period
1943 * without an active slave and it is redundant with the disabling
1945 moving_active_iface = true;
1946 bond_choose_active_iface(port);
1948 bond_send_learning_packets(port);
1950 VLOG_WARN("interface %s: enabled", iface->name);
1951 if (port->active_iface < 0 && !moving_active_iface) {
1952 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1953 bond_choose_active_iface(port);
1954 bond_send_learning_packets(port);
1956 iface->tag = tag_create_random();
1959 moving_active_iface = false;
1960 port->bond_compat_is_stale = true;
1963 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1964 * bond interface. */
1966 bond_update_fake_iface_stats(struct port *port)
1968 struct netdev_stats bond_stats;
1969 struct netdev *bond_dev;
1972 memset(&bond_stats, 0, sizeof bond_stats);
1974 for (i = 0; i < port->n_ifaces; i++) {
1975 struct netdev_stats slave_stats;
1977 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1978 /* XXX: We swap the stats here because they are swapped back when
1979 * reported by the internal device. The reason for this is
1980 * internal devices normally represent packets going into the system
1981 * but when used as fake bond device they represent packets leaving
1982 * the system. We really should do this in the internal device
1983 * itself because changing it here reverses the counts from the
1984 * perspective of the switch. However, the internal device doesn't
1985 * know what type of device it represents so we have to do it here
1987 bond_stats.tx_packets += slave_stats.rx_packets;
1988 bond_stats.tx_bytes += slave_stats.rx_bytes;
1989 bond_stats.rx_packets += slave_stats.tx_packets;
1990 bond_stats.rx_bytes += slave_stats.tx_bytes;
1994 if (!netdev_open_default(port->name, &bond_dev)) {
1995 netdev_set_stats(bond_dev, &bond_stats);
1996 netdev_close(bond_dev);
2001 bond_run(struct bridge *br)
2005 for (i = 0; i < br->n_ports; i++) {
2006 struct port *port = br->ports[i];
2008 if (port->n_ifaces >= 2) {
2009 for (j = 0; j < port->n_ifaces; j++) {
2010 struct iface *iface = port->ifaces[j];
2011 if (time_msec() >= iface->delay_expires) {
2012 bond_enable_slave(iface, !iface->enabled);
2016 if (port->bond_fake_iface
2017 && time_msec() >= port->bond_next_fake_iface_update) {
2018 bond_update_fake_iface_stats(port);
2019 port->bond_next_fake_iface_update = time_msec() + 1000;
2023 if (port->bond_compat_is_stale) {
2024 port->bond_compat_is_stale = false;
2025 port_update_bond_compat(port);
2031 bond_wait(struct bridge *br)
2035 for (i = 0; i < br->n_ports; i++) {
2036 struct port *port = br->ports[i];
2037 if (port->n_ifaces < 2) {
2040 for (j = 0; j < port->n_ifaces; j++) {
2041 struct iface *iface = port->ifaces[j];
2042 if (iface->delay_expires != LLONG_MAX) {
2043 poll_timer_wait_until(iface->delay_expires);
2046 if (port->bond_fake_iface) {
2047 poll_timer_wait_until(port->bond_next_fake_iface_update);
2053 set_dst(struct dst *p, const flow_t *flow,
2054 const struct port *in_port, const struct port *out_port,
2057 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2058 : in_port->vlan >= 0 ? in_port->vlan
2059 : ntohs(flow->dl_vlan));
2060 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
2064 swap_dst(struct dst *p, struct dst *q)
2066 struct dst tmp = *p;
2071 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2072 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2073 * that we push to the datapath. We could in fact fully sort the array by
2074 * vlan, but in most cases there are at most two different vlan tags so that's
2075 * possibly overkill.) */
2077 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2079 struct dst *first = dsts;
2080 struct dst *last = dsts + n_dsts;
2082 while (first != last) {
2084 * - All dsts < first have vlan == 'vlan'.
2085 * - All dsts >= last have vlan != 'vlan'.
2086 * - first < last. */
2087 while (first->vlan == vlan) {
2088 if (++first == last) {
2093 /* Same invariants, plus one additional:
2094 * - first->vlan != vlan.
2096 while (last[-1].vlan != vlan) {
2097 if (--last == first) {
2102 /* Same invariants, plus one additional:
2103 * - last[-1].vlan == vlan.*/
2104 swap_dst(first++, --last);
2109 mirror_mask_ffs(mirror_mask_t mask)
2111 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2116 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2117 const struct dst *test)
2120 for (i = 0; i < n_dsts; i++) {
2121 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2129 port_trunks_vlan(const struct port *port, uint16_t vlan)
2131 return (port->vlan < 0
2132 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2136 port_includes_vlan(const struct port *port, uint16_t vlan)
2138 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2142 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2143 const struct port *in_port, const struct port *out_port,
2144 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2146 mirror_mask_t mirrors = in_port->src_mirrors;
2147 struct dst *dst = dsts;
2150 if (out_port == FLOOD_PORT) {
2151 /* XXX use ODP_FLOOD if no vlans or bonding. */
2152 /* XXX even better, define each VLAN as a datapath port group */
2153 for (i = 0; i < br->n_ports; i++) {
2154 struct port *port = br->ports[i];
2155 if (port != in_port && port_includes_vlan(port, vlan)
2156 && !port->is_mirror_output_port
2157 && set_dst(dst, flow, in_port, port, tags)) {
2158 mirrors |= port->dst_mirrors;
2162 *nf_output_iface = NF_OUT_FLOOD;
2163 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2164 *nf_output_iface = dst->dp_ifidx;
2165 mirrors |= out_port->dst_mirrors;
2170 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2171 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2173 if (set_dst(dst, flow, in_port, m->out_port, tags)
2174 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2178 for (i = 0; i < br->n_ports; i++) {
2179 struct port *port = br->ports[i];
2180 if (port_includes_vlan(port, m->out_vlan)
2181 && set_dst(dst, flow, in_port, port, tags))
2185 if (port->vlan < 0) {
2186 dst->vlan = m->out_vlan;
2188 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2192 /* Use the vlan tag on the original flow instead of
2193 * the one passed in the vlan parameter. This ensures
2194 * that we compare the vlan from before any implicit
2195 * tagging tags place. This is necessary because
2196 * dst->vlan is the final vlan, after removing implicit
2198 flow_vlan = ntohs(flow->dl_vlan);
2199 if (flow_vlan == 0) {
2200 flow_vlan = OFP_VLAN_NONE;
2202 if (port == in_port && dst->vlan == flow_vlan) {
2203 /* Don't send out input port on same VLAN. */
2211 mirrors &= mirrors - 1;
2214 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2218 static void OVS_UNUSED
2219 print_dsts(const struct dst *dsts, size_t n)
2221 for (; n--; dsts++) {
2222 printf(">p%"PRIu16, dsts->dp_ifidx);
2223 if (dsts->vlan != OFP_VLAN_NONE) {
2224 printf("v%"PRIu16, dsts->vlan);
2230 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2231 const struct port *in_port, const struct port *out_port,
2232 tag_type *tags, struct odp_actions *actions,
2233 uint16_t *nf_output_iface)
2235 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2237 const struct dst *p;
2240 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2243 cur_vlan = ntohs(flow->dl_vlan);
2244 for (p = dsts; p < &dsts[n_dsts]; p++) {
2245 union odp_action *a;
2246 if (p->vlan != cur_vlan) {
2247 if (p->vlan == OFP_VLAN_NONE) {
2248 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2250 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2251 a->vlan_vid.vlan_vid = htons(p->vlan);
2255 a = odp_actions_add(actions, ODPAT_OUTPUT);
2256 a->output.port = p->dp_ifidx;
2260 /* Returns the effective vlan of a packet, taking into account both the
2261 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2262 * the packet is untagged and -1 indicates it has an invalid header and
2263 * should be dropped. */
2264 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2265 struct port *in_port, bool have_packet)
2267 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2268 * belongs to VLAN 0, so we should treat both cases identically. (In the
2269 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2270 * presumably to allow a priority to be specified. In the latter case, the
2271 * packet does not have any 802.1Q header.) */
2272 int vlan = ntohs(flow->dl_vlan);
2273 if (vlan == OFP_VLAN_NONE) {
2276 if (in_port->vlan >= 0) {
2278 /* XXX support double tagging? */
2280 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2281 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2282 "packet received on port %s configured with "
2283 "implicit VLAN %"PRIu16,
2284 br->name, ntohs(flow->dl_vlan),
2285 in_port->name, in_port->vlan);
2289 vlan = in_port->vlan;
2291 if (!port_includes_vlan(in_port, vlan)) {
2293 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2294 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2295 "packet received on port %s not configured for "
2297 br->name, vlan, in_port->name, vlan);
2306 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2307 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2308 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2310 is_gratuitous_arp(const flow_t *flow)
2312 return (flow->dl_type == htons(ETH_TYPE_ARP)
2313 && eth_addr_is_broadcast(flow->dl_dst)
2314 && (flow->nw_proto == ARP_OP_REPLY
2315 || (flow->nw_proto == ARP_OP_REQUEST
2316 && flow->nw_src == flow->nw_dst)));
2320 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2321 struct port *in_port)
2323 enum grat_arp_lock_type lock_type;
2326 /* We don't want to learn from gratuitous ARP packets that are reflected
2327 * back over bond slaves so we lock the learning table. */
2328 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2329 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2330 GRAT_ARP_LOCK_CHECK;
2332 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2335 /* The log messages here could actually be useful in debugging,
2336 * so keep the rate limit relatively high. */
2337 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2339 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2340 "on port %s in VLAN %d",
2341 br->name, ETH_ADDR_ARGS(flow->dl_src),
2342 in_port->name, vlan);
2343 ofproto_revalidate(br->ofproto, rev_tag);
2347 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2348 * dropped. Returns true if they may be forwarded, false if they should be
2351 * If 'have_packet' is true, it indicates that the caller is processing a
2352 * received packet. If 'have_packet' is false, then the caller is just
2353 * revalidating an existing flow because configuration has changed. Either
2354 * way, 'have_packet' only affects logging (there is no point in logging errors
2355 * during revalidation).
2357 * Sets '*in_portp' to the input port. This will be a null pointer if
2358 * flow->in_port does not designate a known input port (in which case
2359 * is_admissible() returns false).
2361 * When returning true, sets '*vlanp' to the effective VLAN of the input
2362 * packet, as returned by flow_get_vlan().
2364 * May also add tags to '*tags', although the current implementation only does
2365 * so in one special case.
2368 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2369 tag_type *tags, int *vlanp, struct port **in_portp)
2371 struct iface *in_iface;
2372 struct port *in_port;
2375 /* Find the interface and port structure for the received packet. */
2376 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2378 /* No interface? Something fishy... */
2380 /* Odd. A few possible reasons here:
2382 * - We deleted an interface but there are still a few packets
2383 * queued up from it.
2385 * - Someone externally added an interface (e.g. with "ovs-dpctl
2386 * add-if") that we don't know about.
2388 * - Packet arrived on the local port but the local port is not
2389 * one of our bridge ports.
2391 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2393 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2394 "interface %"PRIu16, br->name, flow->in_port);
2400 *in_portp = in_port = in_iface->port;
2401 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2406 /* Drop frames for reserved multicast addresses. */
2407 if (eth_addr_is_reserved(flow->dl_dst)) {
2411 /* Drop frames on ports reserved for mirroring. */
2412 if (in_port->is_mirror_output_port) {
2414 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2415 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2416 "%s, which is reserved exclusively for mirroring",
2417 br->name, in_port->name);
2422 /* Packets received on bonds need special attention to avoid duplicates. */
2423 if (in_port->n_ifaces > 1) {
2425 bool is_grat_arp_locked;
2427 if (eth_addr_is_multicast(flow->dl_dst)) {
2428 *tags |= in_port->active_iface_tag;
2429 if (in_port->active_iface != in_iface->port_ifidx) {
2430 /* Drop all multicast packets on inactive slaves. */
2435 /* Drop all packets for which we have learned a different input
2436 * port, because we probably sent the packet on one slave and got
2437 * it back on the other. Gratuitous ARP packets are an exception
2438 * to this rule: the host has moved to another switch. The exception
2439 * to the exception is if we locked the learning table to avoid
2440 * reflections on bond slaves. If this is the case, just drop the
2442 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2443 &is_grat_arp_locked);
2444 if (src_idx != -1 && src_idx != in_port->port_idx &&
2445 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2453 /* If the composed actions may be applied to any packet in the given 'flow',
2454 * returns true. Otherwise, the actions should only be applied to 'packet', or
2455 * not at all, if 'packet' was NULL. */
2457 process_flow(struct bridge *br, const flow_t *flow,
2458 const struct ofpbuf *packet, struct odp_actions *actions,
2459 tag_type *tags, uint16_t *nf_output_iface)
2461 struct port *in_port;
2462 struct port *out_port;
2466 /* Check whether we should drop packets in this flow. */
2467 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2472 /* Learn source MAC (but don't try to learn from revalidation). */
2474 update_learning_table(br, flow, vlan, in_port);
2477 /* Determine output port. */
2478 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2480 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2481 out_port = br->ports[out_port_idx];
2482 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2483 /* If we are revalidating but don't have a learning entry then
2484 * eject the flow. Installing a flow that floods packets opens
2485 * up a window of time where we could learn from a packet reflected
2486 * on a bond and blackhole packets before the learning table is
2487 * updated to reflect the correct port. */
2490 out_port = FLOOD_PORT;
2493 /* Don't send packets out their input ports. */
2494 if (in_port == out_port) {
2500 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2507 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2510 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2511 const struct ofp_phy_port *opp,
2514 struct bridge *br = br_;
2515 struct iface *iface;
2518 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2524 if (reason == OFPPR_DELETE) {
2525 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2526 br->name, iface->name);
2527 iface_destroy(iface);
2528 if (!port->n_ifaces) {
2529 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2530 br->name, port->name);
2536 if (port->n_ifaces > 1) {
2537 bool up = !(opp->state & OFPPS_LINK_DOWN);
2538 bond_link_status_update(iface, up);
2539 port_update_bond_compat(port);
2545 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2546 struct odp_actions *actions, tag_type *tags,
2547 uint16_t *nf_output_iface, void *br_)
2549 struct bridge *br = br_;
2551 COVERAGE_INC(bridge_process_flow);
2553 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2557 bridge_account_flow_ofhook_cb(const flow_t *flow, tag_type tags,
2558 const union odp_action *actions,
2559 size_t n_actions, unsigned long long int n_bytes,
2562 struct bridge *br = br_;
2563 const union odp_action *a;
2564 struct port *in_port;
2568 /* Feed information from the active flows back into the learning table to
2569 * ensure that table is always in sync with what is actually flowing
2570 * through the datapath.
2572 * We test that 'tags' is nonzero to ensure that only flows that include an
2573 * OFPP_NORMAL action are used for learning. This works because
2574 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2575 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2576 update_learning_table(br, flow, vlan, in_port);
2579 /* Account for bond slave utilization. */
2580 if (!br->has_bonded_ports) {
2583 for (a = actions; a < &actions[n_actions]; a++) {
2584 if (a->type == ODPAT_OUTPUT) {
2585 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2586 if (out_port && out_port->n_ifaces >= 2) {
2587 struct bond_entry *e = lookup_bond_entry(out_port,
2589 e->tx_bytes += n_bytes;
2596 bridge_account_checkpoint_ofhook_cb(void *br_)
2598 struct bridge *br = br_;
2602 if (!br->has_bonded_ports) {
2607 for (i = 0; i < br->n_ports; i++) {
2608 struct port *port = br->ports[i];
2609 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2610 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2611 bond_rebalance_port(port);
2616 static struct ofhooks bridge_ofhooks = {
2617 bridge_port_changed_ofhook_cb,
2618 bridge_normal_ofhook_cb,
2619 bridge_account_flow_ofhook_cb,
2620 bridge_account_checkpoint_ofhook_cb,
2623 /* Bonding functions. */
2625 /* Statistics for a single interface on a bonded port, used for load-based
2626 * bond rebalancing. */
2627 struct slave_balance {
2628 struct iface *iface; /* The interface. */
2629 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2631 /* All the "bond_entry"s that are assigned to this interface, in order of
2632 * increasing tx_bytes. */
2633 struct bond_entry **hashes;
2637 /* Sorts pointers to pointers to bond_entries in ascending order by the
2638 * interface to which they are assigned, and within a single interface in
2639 * ascending order of bytes transmitted. */
2641 compare_bond_entries(const void *a_, const void *b_)
2643 const struct bond_entry *const *ap = a_;
2644 const struct bond_entry *const *bp = b_;
2645 const struct bond_entry *a = *ap;
2646 const struct bond_entry *b = *bp;
2647 if (a->iface_idx != b->iface_idx) {
2648 return a->iface_idx > b->iface_idx ? 1 : -1;
2649 } else if (a->tx_bytes != b->tx_bytes) {
2650 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2656 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2657 * *descending* order by number of bytes transmitted. */
2659 compare_slave_balance(const void *a_, const void *b_)
2661 const struct slave_balance *a = a_;
2662 const struct slave_balance *b = b_;
2663 if (a->iface->enabled != b->iface->enabled) {
2664 return a->iface->enabled ? -1 : 1;
2665 } else if (a->tx_bytes != b->tx_bytes) {
2666 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2673 swap_bals(struct slave_balance *a, struct slave_balance *b)
2675 struct slave_balance tmp = *a;
2680 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2681 * given that 'p' (and only 'p') might be in the wrong location.
2683 * This function invalidates 'p', since it might now be in a different memory
2686 resort_bals(struct slave_balance *p,
2687 struct slave_balance bals[], size_t n_bals)
2690 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2691 swap_bals(p, p - 1);
2693 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2694 swap_bals(p, p + 1);
2700 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2702 if (VLOG_IS_DBG_ENABLED()) {
2703 struct ds ds = DS_EMPTY_INITIALIZER;
2704 const struct slave_balance *b;
2706 for (b = bals; b < bals + n_bals; b++) {
2710 ds_put_char(&ds, ',');
2712 ds_put_format(&ds, " %s %"PRIu64"kB",
2713 b->iface->name, b->tx_bytes / 1024);
2715 if (!b->iface->enabled) {
2716 ds_put_cstr(&ds, " (disabled)");
2718 if (b->n_hashes > 0) {
2719 ds_put_cstr(&ds, " (");
2720 for (i = 0; i < b->n_hashes; i++) {
2721 const struct bond_entry *e = b->hashes[i];
2723 ds_put_cstr(&ds, " + ");
2725 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2726 e - port->bond_hash, e->tx_bytes / 1024);
2728 ds_put_cstr(&ds, ")");
2731 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2736 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2738 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2741 struct bond_entry *hash = from->hashes[hash_idx];
2742 struct port *port = from->iface->port;
2743 uint64_t delta = hash->tx_bytes;
2745 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2746 "from %s to %s (now carrying %"PRIu64"kB and "
2747 "%"PRIu64"kB load, respectively)",
2748 port->name, delta / 1024, hash - port->bond_hash,
2749 from->iface->name, to->iface->name,
2750 (from->tx_bytes - delta) / 1024,
2751 (to->tx_bytes + delta) / 1024);
2753 /* Delete element from from->hashes.
2755 * We don't bother to add the element to to->hashes because not only would
2756 * it require more work, the only purpose it would be to allow that hash to
2757 * be migrated to another slave in this rebalancing run, and there is no
2758 * point in doing that. */
2759 if (hash_idx == 0) {
2762 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2763 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2767 /* Shift load away from 'from' to 'to'. */
2768 from->tx_bytes -= delta;
2769 to->tx_bytes += delta;
2771 /* Arrange for flows to be revalidated. */
2772 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2773 hash->iface_idx = to->iface->port_ifidx;
2774 hash->iface_tag = tag_create_random();
2778 bond_rebalance_port(struct port *port)
2780 struct slave_balance bals[DP_MAX_PORTS];
2782 struct bond_entry *hashes[BOND_MASK + 1];
2783 struct slave_balance *b, *from, *to;
2784 struct bond_entry *e;
2787 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2788 * descending order of tx_bytes, so that bals[0] represents the most
2789 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2792 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2793 * array for each slave_balance structure, we sort our local array of
2794 * hashes in order by slave, so that all of the hashes for a given slave
2795 * become contiguous in memory, and then we point each 'hashes' members of
2796 * a slave_balance structure to the start of a contiguous group. */
2797 n_bals = port->n_ifaces;
2798 for (b = bals; b < &bals[n_bals]; b++) {
2799 b->iface = port->ifaces[b - bals];
2804 for (i = 0; i <= BOND_MASK; i++) {
2805 hashes[i] = &port->bond_hash[i];
2807 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2808 for (i = 0; i <= BOND_MASK; i++) {
2810 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2811 b = &bals[e->iface_idx];
2812 b->tx_bytes += e->tx_bytes;
2814 b->hashes = &hashes[i];
2819 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2820 log_bals(bals, n_bals, port);
2822 /* Discard slaves that aren't enabled (which were sorted to the back of the
2823 * array earlier). */
2824 while (!bals[n_bals - 1].iface->enabled) {
2831 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2832 to = &bals[n_bals - 1];
2833 for (from = bals; from < to; ) {
2834 uint64_t overload = from->tx_bytes - to->tx_bytes;
2835 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2836 /* The extra load on 'from' (and all less-loaded slaves), compared
2837 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2838 * it is less than ~1Mbps. No point in rebalancing. */
2840 } else if (from->n_hashes == 1) {
2841 /* 'from' only carries a single MAC hash, so we can't shift any
2842 * load away from it, even though we want to. */
2845 /* 'from' is carrying significantly more load than 'to', and that
2846 * load is split across at least two different hashes. Pick a hash
2847 * to migrate to 'to' (the least-loaded slave), given that doing so
2848 * must decrease the ratio of the load on the two slaves by at
2851 * The sort order we use means that we prefer to shift away the
2852 * smallest hashes instead of the biggest ones. There is little
2853 * reason behind this decision; we could use the opposite sort
2854 * order to shift away big hashes ahead of small ones. */
2857 for (i = 0; i < from->n_hashes; i++) {
2858 double old_ratio, new_ratio;
2859 uint64_t delta = from->hashes[i]->tx_bytes;
2861 if (delta == 0 || from->tx_bytes - delta == 0) {
2862 /* Pointless move. */
2866 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2868 if (to->tx_bytes == 0) {
2869 /* Nothing on the new slave, move it. */
2873 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2874 new_ratio = (double)(from->tx_bytes - delta) /
2875 (to->tx_bytes + delta);
2877 if (new_ratio == 0) {
2878 /* Should already be covered but check to prevent division
2883 if (new_ratio < 1) {
2884 new_ratio = 1 / new_ratio;
2887 if (old_ratio - new_ratio > 0.1) {
2888 /* Would decrease the ratio, move it. */
2892 if (i < from->n_hashes) {
2893 bond_shift_load(from, to, i);
2894 port->bond_compat_is_stale = true;
2896 /* If the result of the migration changed the relative order of
2897 * 'from' and 'to' swap them back to maintain invariants. */
2898 if (order_swapped) {
2899 swap_bals(from, to);
2902 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2903 * point to different slave_balance structures. It is only
2904 * valid to do these two operations in a row at all because we
2905 * know that 'from' will not move past 'to' and vice versa. */
2906 resort_bals(from, bals, n_bals);
2907 resort_bals(to, bals, n_bals);
2914 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2915 * historical data to decay to <1% in 7 rebalancing runs. */
2916 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2922 bond_send_learning_packets(struct port *port)
2924 struct bridge *br = port->bridge;
2925 struct mac_entry *e;
2926 struct ofpbuf packet;
2927 int error, n_packets, n_errors;
2929 if (!port->n_ifaces || port->active_iface < 0) {
2933 ofpbuf_init(&packet, 128);
2934 error = n_packets = n_errors = 0;
2935 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2936 union ofp_action actions[2], *a;
2942 if (e->port == port->port_idx
2943 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2947 /* Compose actions. */
2948 memset(actions, 0, sizeof actions);
2951 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2952 a->vlan_vid.len = htons(sizeof *a);
2953 a->vlan_vid.vlan_vid = htons(e->vlan);
2956 a->output.type = htons(OFPAT_OUTPUT);
2957 a->output.len = htons(sizeof *a);
2958 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2963 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2965 flow_extract(&packet, 0, ODPP_NONE, &flow);
2966 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2973 ofpbuf_uninit(&packet);
2976 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2977 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2978 "packets, last error was: %s",
2979 port->name, n_errors, n_packets, strerror(error));
2981 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2982 port->name, n_packets);
2986 /* Bonding unixctl user interface functions. */
2989 bond_unixctl_list(struct unixctl_conn *conn,
2990 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2992 struct ds ds = DS_EMPTY_INITIALIZER;
2993 const struct bridge *br;
2995 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2997 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
3000 for (i = 0; i < br->n_ports; i++) {
3001 const struct port *port = br->ports[i];
3002 if (port->n_ifaces > 1) {
3005 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
3006 for (j = 0; j < port->n_ifaces; j++) {
3007 const struct iface *iface = port->ifaces[j];
3009 ds_put_cstr(&ds, ", ");
3011 ds_put_cstr(&ds, iface->name);
3013 ds_put_char(&ds, '\n');
3017 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3021 static struct port *
3022 bond_find(const char *name)
3024 const struct bridge *br;
3026 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
3029 for (i = 0; i < br->n_ports; i++) {
3030 struct port *port = br->ports[i];
3031 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3040 bond_unixctl_show(struct unixctl_conn *conn,
3041 const char *args, void *aux OVS_UNUSED)
3043 struct ds ds = DS_EMPTY_INITIALIZER;
3044 const struct port *port;
3047 port = bond_find(args);
3049 unixctl_command_reply(conn, 501, "no such bond");
3053 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3054 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3055 ds_put_format(&ds, "next rebalance: %lld ms\n",
3056 port->bond_next_rebalance - time_msec());
3057 for (j = 0; j < port->n_ifaces; j++) {
3058 const struct iface *iface = port->ifaces[j];
3059 struct bond_entry *be;
3062 ds_put_format(&ds, "slave %s: %s\n",
3063 iface->name, iface->enabled ? "enabled" : "disabled");
3064 if (j == port->active_iface) {
3065 ds_put_cstr(&ds, "\tactive slave\n");
3067 if (iface->delay_expires != LLONG_MAX) {
3068 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3069 iface->enabled ? "downdelay" : "updelay",
3070 iface->delay_expires - time_msec());
3074 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3075 int hash = be - port->bond_hash;
3076 struct mac_entry *me;
3078 if (be->iface_idx != j) {
3082 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3083 hash, be->tx_bytes / 1024);
3086 LIST_FOR_EACH (me, struct mac_entry, lru_node,
3087 &port->bridge->ml->lrus) {
3090 if (bond_hash(me->mac) == hash
3091 && me->port != port->port_idx
3092 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3093 && dp_ifidx == iface->dp_ifidx)
3095 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3096 ETH_ADDR_ARGS(me->mac));
3101 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3106 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3107 void *aux OVS_UNUSED)
3109 char *args = (char *) args_;
3110 char *save_ptr = NULL;
3111 char *bond_s, *hash_s, *slave_s;
3112 uint8_t mac[ETH_ADDR_LEN];
3114 struct iface *iface;
3115 struct bond_entry *entry;
3118 bond_s = strtok_r(args, " ", &save_ptr);
3119 hash_s = strtok_r(NULL, " ", &save_ptr);
3120 slave_s = strtok_r(NULL, " ", &save_ptr);
3122 unixctl_command_reply(conn, 501,
3123 "usage: bond/migrate BOND HASH SLAVE");
3127 port = bond_find(bond_s);
3129 unixctl_command_reply(conn, 501, "no such bond");
3133 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3134 == ETH_ADDR_SCAN_COUNT) {
3135 hash = bond_hash(mac);
3136 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3137 hash = atoi(hash_s) & BOND_MASK;
3139 unixctl_command_reply(conn, 501, "bad hash");
3143 iface = port_lookup_iface(port, slave_s);
3145 unixctl_command_reply(conn, 501, "no such slave");
3149 if (!iface->enabled) {
3150 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3154 entry = &port->bond_hash[hash];
3155 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3156 entry->iface_idx = iface->port_ifidx;
3157 entry->iface_tag = tag_create_random();
3158 port->bond_compat_is_stale = true;
3159 unixctl_command_reply(conn, 200, "migrated");
3163 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3164 void *aux OVS_UNUSED)
3166 char *args = (char *) args_;
3167 char *save_ptr = NULL;
3168 char *bond_s, *slave_s;
3170 struct iface *iface;
3172 bond_s = strtok_r(args, " ", &save_ptr);
3173 slave_s = strtok_r(NULL, " ", &save_ptr);
3175 unixctl_command_reply(conn, 501,
3176 "usage: bond/set-active-slave BOND SLAVE");
3180 port = bond_find(bond_s);
3182 unixctl_command_reply(conn, 501, "no such bond");
3186 iface = port_lookup_iface(port, slave_s);
3188 unixctl_command_reply(conn, 501, "no such slave");
3192 if (!iface->enabled) {
3193 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3197 if (port->active_iface != iface->port_ifidx) {
3198 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3199 port->active_iface = iface->port_ifidx;
3200 port->active_iface_tag = tag_create_random();
3201 VLOG_INFO("port %s: active interface is now %s",
3202 port->name, iface->name);
3203 bond_send_learning_packets(port);
3204 unixctl_command_reply(conn, 200, "done");
3206 unixctl_command_reply(conn, 200, "no change");
3211 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3213 char *args = (char *) args_;
3214 char *save_ptr = NULL;
3215 char *bond_s, *slave_s;
3217 struct iface *iface;
3219 bond_s = strtok_r(args, " ", &save_ptr);
3220 slave_s = strtok_r(NULL, " ", &save_ptr);
3222 unixctl_command_reply(conn, 501,
3223 "usage: bond/enable/disable-slave BOND SLAVE");
3227 port = bond_find(bond_s);
3229 unixctl_command_reply(conn, 501, "no such bond");
3233 iface = port_lookup_iface(port, slave_s);
3235 unixctl_command_reply(conn, 501, "no such slave");
3239 bond_enable_slave(iface, enable);
3240 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3244 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3245 void *aux OVS_UNUSED)
3247 enable_slave(conn, args, true);
3251 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3252 void *aux OVS_UNUSED)
3254 enable_slave(conn, args, false);
3258 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3259 void *aux OVS_UNUSED)
3261 uint8_t mac[ETH_ADDR_LEN];
3265 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3266 == ETH_ADDR_SCAN_COUNT) {
3267 hash = bond_hash(mac);
3269 hash_cstr = xasprintf("%u", hash);
3270 unixctl_command_reply(conn, 200, hash_cstr);
3273 unixctl_command_reply(conn, 501, "invalid mac");
3280 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3281 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3282 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3283 unixctl_command_register("bond/set-active-slave",
3284 bond_unixctl_set_active_slave, NULL);
3285 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3287 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3289 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3292 /* Port functions. */
3294 static struct port *
3295 port_create(struct bridge *br, const char *name)
3299 port = xzalloc(sizeof *port);
3301 port->port_idx = br->n_ports;
3303 port->trunks = NULL;
3304 port->name = xstrdup(name);
3305 port->active_iface = -1;
3307 if (br->n_ports >= br->allocated_ports) {
3308 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3311 br->ports[br->n_ports++] = port;
3312 shash_add_assert(&br->port_by_name, port->name, port);
3314 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3321 get_port_other_config(const struct ovsrec_port *port, const char *key,
3322 const char *default_value)
3326 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3328 return value ? value : default_value;
3332 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3334 struct shash new_ifaces;
3337 /* Collect list of new interfaces. */
3338 shash_init(&new_ifaces);
3339 for (i = 0; i < cfg->n_interfaces; i++) {
3340 const char *name = cfg->interfaces[i]->name;
3341 shash_add_once(&new_ifaces, name, NULL);
3344 /* Get rid of deleted interfaces. */
3345 for (i = 0; i < port->n_ifaces; ) {
3346 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3347 iface_destroy(port->ifaces[i]);
3353 shash_destroy(&new_ifaces);
3357 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3359 struct shash new_ifaces;
3360 long long int next_rebalance;
3361 unsigned long *trunks;
3367 /* Update settings. */
3368 port->updelay = cfg->bond_updelay;
3369 if (port->updelay < 0) {
3372 port->downdelay = cfg->bond_downdelay;
3373 if (port->downdelay < 0) {
3374 port->downdelay = 0;
3376 port->bond_rebalance_interval = atoi(
3377 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3378 if (port->bond_rebalance_interval < 1000) {
3379 port->bond_rebalance_interval = 1000;
3381 next_rebalance = time_msec() + port->bond_rebalance_interval;
3382 if (port->bond_next_rebalance > next_rebalance) {
3383 port->bond_next_rebalance = next_rebalance;
3386 /* Add new interfaces and update 'cfg' member of existing ones. */
3387 shash_init(&new_ifaces);
3388 for (i = 0; i < cfg->n_interfaces; i++) {
3389 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3390 struct iface *iface;
3392 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3393 VLOG_WARN("port %s: %s specified twice as port interface",
3394 port->name, if_cfg->name);
3398 iface = iface_lookup(port->bridge, if_cfg->name);
3400 if (iface->port != port) {
3401 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3403 port->bridge->name, if_cfg->name, iface->port->name);
3406 iface->cfg = if_cfg;
3408 iface_create(port, if_cfg);
3411 shash_destroy(&new_ifaces);
3416 if (port->n_ifaces < 2) {
3418 if (vlan >= 0 && vlan <= 4095) {
3419 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3424 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3425 * they even work as-is. But they have not been tested. */
3426 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3430 if (port->vlan != vlan) {
3432 bridge_flush(port->bridge);
3435 /* Get trunked VLANs. */
3437 if (vlan < 0 && cfg->n_trunks) {
3440 trunks = bitmap_allocate(4096);
3442 for (i = 0; i < cfg->n_trunks; i++) {
3443 int trunk = cfg->trunks[i];
3445 bitmap_set1(trunks, trunk);
3451 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3452 port->name, cfg->n_trunks);
3454 if (n_errors == cfg->n_trunks) {
3455 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3457 bitmap_free(trunks);
3460 } else if (vlan >= 0 && cfg->n_trunks) {
3461 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3465 ? port->trunks != NULL
3466 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3467 bridge_flush(port->bridge);
3469 bitmap_free(port->trunks);
3470 port->trunks = trunks;
3474 port_destroy(struct port *port)
3477 struct bridge *br = port->bridge;
3481 proc_net_compat_update_vlan(port->name, NULL, 0);
3482 proc_net_compat_update_bond(port->name, NULL);
3484 for (i = 0; i < MAX_MIRRORS; i++) {
3485 struct mirror *m = br->mirrors[i];
3486 if (m && m->out_port == port) {
3491 while (port->n_ifaces > 0) {
3492 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3495 shash_find_and_delete_assert(&br->port_by_name, port->name);
3497 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3498 del->port_idx = port->port_idx;
3501 bitmap_free(port->trunks);
3508 static struct port *
3509 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3511 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3512 return iface ? iface->port : NULL;
3515 static struct port *
3516 port_lookup(const struct bridge *br, const char *name)
3518 return shash_find_data(&br->port_by_name, name);
3521 static struct iface *
3522 port_lookup_iface(const struct port *port, const char *name)
3524 struct iface *iface = iface_lookup(port->bridge, name);
3525 return iface && iface->port == port ? iface : NULL;
3529 port_update_bonding(struct port *port)
3531 if (port->n_ifaces < 2) {
3532 /* Not a bonded port. */
3533 if (port->bond_hash) {
3534 free(port->bond_hash);
3535 port->bond_hash = NULL;
3536 port->bond_compat_is_stale = true;
3537 port->bond_fake_iface = false;
3540 if (!port->bond_hash) {
3543 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3544 for (i = 0; i <= BOND_MASK; i++) {
3545 struct bond_entry *e = &port->bond_hash[i];
3549 port->no_ifaces_tag = tag_create_random();
3550 bond_choose_active_iface(port);
3551 port->bond_next_rebalance
3552 = time_msec() + port->bond_rebalance_interval;
3554 if (port->cfg->bond_fake_iface) {
3555 port->bond_next_fake_iface_update = time_msec();
3558 port->bond_compat_is_stale = true;
3559 port->bond_fake_iface = port->cfg->bond_fake_iface;
3564 port_update_bond_compat(struct port *port)
3566 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3567 struct compat_bond bond;
3570 if (port->n_ifaces < 2) {
3571 proc_net_compat_update_bond(port->name, NULL);
3576 bond.updelay = port->updelay;
3577 bond.downdelay = port->downdelay;
3580 bond.hashes = compat_hashes;
3581 if (port->bond_hash) {
3582 const struct bond_entry *e;
3583 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3584 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3585 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3586 cbh->hash = e - port->bond_hash;
3587 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3592 bond.n_slaves = port->n_ifaces;
3593 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3594 for (i = 0; i < port->n_ifaces; i++) {
3595 struct iface *iface = port->ifaces[i];
3596 struct compat_bond_slave *slave = &bond.slaves[i];
3597 slave->name = iface->name;
3599 /* We need to make the same determination as the Linux bonding
3600 * code to determine whether a slave should be consider "up".
3601 * The Linux function bond_miimon_inspect() supports four
3602 * BOND_LINK_* states:
3604 * - BOND_LINK_UP: carrier detected, updelay has passed.
3605 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3606 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3607 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3609 * The function bond_info_show_slave() only considers BOND_LINK_UP
3610 * to be "up" and anything else to be "down".
3612 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3616 netdev_get_etheraddr(iface->netdev, slave->mac);
3619 if (port->bond_fake_iface) {
3620 struct netdev *bond_netdev;
3622 if (!netdev_open_default(port->name, &bond_netdev)) {
3624 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3626 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3628 netdev_close(bond_netdev);
3632 proc_net_compat_update_bond(port->name, &bond);
3637 port_update_vlan_compat(struct port *port)
3639 struct bridge *br = port->bridge;
3640 char *vlandev_name = NULL;
3642 if (port->vlan > 0) {
3643 /* Figure out the name that the VLAN device should actually have, if it
3644 * existed. This takes some work because the VLAN device would not
3645 * have port->name in its name; rather, it would have the trunk port's
3646 * name, and 'port' would be attached to a bridge that also had the
3647 * VLAN device one of its ports. So we need to find a trunk port that
3648 * includes port->vlan.
3650 * There might be more than one candidate. This doesn't happen on
3651 * XenServer, so if it happens we just pick the first choice in
3652 * alphabetical order instead of creating multiple VLAN devices. */
3654 for (i = 0; i < br->n_ports; i++) {
3655 struct port *p = br->ports[i];
3656 if (port_trunks_vlan(p, port->vlan)
3658 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3660 uint8_t ea[ETH_ADDR_LEN];
3661 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3662 if (!eth_addr_is_multicast(ea) &&
3663 !eth_addr_is_reserved(ea) &&
3664 !eth_addr_is_zero(ea)) {
3665 vlandev_name = p->name;
3670 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3673 /* Interface functions. */
3675 static struct iface *
3676 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3678 struct bridge *br = port->bridge;
3679 struct iface *iface;
3680 char *name = if_cfg->name;
3683 iface = xzalloc(sizeof *iface);
3685 iface->port_ifidx = port->n_ifaces;
3686 iface->name = xstrdup(name);
3687 iface->dp_ifidx = -1;
3688 iface->tag = tag_create_random();
3689 iface->delay_expires = LLONG_MAX;
3690 iface->netdev = NULL;
3691 iface->cfg = if_cfg;
3693 shash_add_assert(&br->iface_by_name, iface->name, iface);
3695 /* Attempt to create the network interface in case it doesn't exist yet. */
3696 if (!iface_is_internal(br, iface->name)) {
3697 error = set_up_iface(if_cfg, iface, true);
3699 VLOG_WARN("could not create iface %s: %s", iface->name,
3702 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3709 if (port->n_ifaces >= port->allocated_ifaces) {
3710 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3711 sizeof *port->ifaces);
3713 port->ifaces[port->n_ifaces++] = iface;
3714 if (port->n_ifaces > 1) {
3715 br->has_bonded_ports = true;
3718 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3726 iface_destroy(struct iface *iface)
3729 struct port *port = iface->port;
3730 struct bridge *br = port->bridge;
3731 bool del_active = port->active_iface == iface->port_ifidx;
3734 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3736 if (iface->dp_ifidx >= 0) {
3737 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3740 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3741 del->port_ifidx = iface->port_ifidx;
3743 netdev_close(iface->netdev);
3746 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3747 bond_choose_active_iface(port);
3748 bond_send_learning_packets(port);
3754 bridge_flush(port->bridge);
3758 static struct iface *
3759 iface_lookup(const struct bridge *br, const char *name)
3761 return shash_find_data(&br->iface_by_name, name);
3764 static struct iface *
3765 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3767 return port_array_get(&br->ifaces, dp_ifidx);
3770 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3771 * 'br', that is, an interface that is entirely simulated within the datapath.
3772 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3773 * interfaces are created by setting "iface.<iface>.internal = true".
3775 * In addition, we have a kluge-y feature that creates an internal port with
3776 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3777 * This feature needs to go away in the long term. Until then, this is one
3778 * reason why this function takes a name instead of a struct iface: the fake
3779 * interfaces created this way do not have a struct iface. */
3781 iface_is_internal(const struct bridge *br, const char *if_name)
3783 struct iface *iface;
3786 if (!strcmp(if_name, br->name)) {
3790 iface = iface_lookup(br, if_name);
3791 if (iface && !strcmp(iface->cfg->type, "internal")) {
3795 port = port_lookup(br, if_name);
3796 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3802 /* Set Ethernet address of 'iface', if one is specified in the configuration
3805 iface_set_mac(struct iface *iface)
3807 uint8_t ea[ETH_ADDR_LEN];
3809 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3810 if (eth_addr_is_multicast(ea)) {
3811 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3813 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3814 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3815 iface->name, iface->name);
3817 int error = netdev_set_etheraddr(iface->netdev, ea);
3819 VLOG_ERR("interface %s: setting MAC failed (%s)",
3820 iface->name, strerror(error));
3827 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3828 struct shash *shash)
3833 for (i = 0; i < n; i++) {
3834 shash_add(shash, keys[i], values[i]);
3838 struct iface_delete_queues_cbdata {
3839 struct netdev *netdev;
3840 const struct ovsdb_datum *queues;
3844 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
3846 union ovsdb_atom atom;
3848 atom.integer = target;
3849 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
3853 iface_delete_queues(unsigned int queue_id,
3854 const struct shash *details OVS_UNUSED, void *cbdata_)
3856 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3858 if (!queue_ids_include(cbdata->queues, queue_id)) {
3859 netdev_delete_queue(cbdata->netdev, queue_id);
3864 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3866 if (!qos || qos->type[0] == '\0') {
3867 netdev_set_qos(iface->netdev, NULL, NULL);
3869 struct iface_delete_queues_cbdata cbdata;
3870 struct shash details;
3873 /* Configure top-level Qos for 'iface'. */
3874 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3875 qos->n_other_config, &details);
3876 netdev_set_qos(iface->netdev, qos->type, &details);
3877 shash_destroy(&details);
3879 /* Deconfigure queues that were deleted. */
3880 cbdata.netdev = iface->netdev;
3881 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
3883 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3885 /* Configure queues for 'iface'. */
3886 for (i = 0; i < qos->n_queues; i++) {
3887 const struct ovsrec_queue *queue = qos->value_queues[i];
3888 unsigned int queue_id = qos->key_queues[i];
3890 shash_from_ovs_idl_map(queue->key_other_config,
3891 queue->value_other_config,
3892 queue->n_other_config, &details);
3893 netdev_set_queue(iface->netdev, queue_id, &details);
3894 shash_destroy(&details);
3899 /* Port mirroring. */
3901 static struct mirror *
3902 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
3906 for (i = 0; i < MAX_MIRRORS; i++) {
3907 struct mirror *m = br->mirrors[i];
3908 if (m && uuid_equals(uuid, &m->uuid)) {
3916 mirror_reconfigure(struct bridge *br)
3918 unsigned long *rspan_vlans;
3921 /* Get rid of deleted mirrors. */
3922 for (i = 0; i < MAX_MIRRORS; i++) {
3923 struct mirror *m = br->mirrors[i];
3925 const struct ovsdb_datum *mc;
3926 union ovsdb_atom atom;
3928 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
3929 atom.uuid = br->mirrors[i]->uuid;
3930 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
3936 /* Add new mirrors and reconfigure existing ones. */
3937 for (i = 0; i < br->cfg->n_mirrors; i++) {
3938 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3939 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
3941 mirror_reconfigure_one(m, cfg);
3943 mirror_create(br, cfg);
3947 /* Update port reserved status. */
3948 for (i = 0; i < br->n_ports; i++) {
3949 br->ports[i]->is_mirror_output_port = false;
3951 for (i = 0; i < MAX_MIRRORS; i++) {
3952 struct mirror *m = br->mirrors[i];
3953 if (m && m->out_port) {
3954 m->out_port->is_mirror_output_port = true;
3958 /* Update flooded vlans (for RSPAN). */
3960 if (br->cfg->n_flood_vlans) {
3961 rspan_vlans = bitmap_allocate(4096);
3963 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3964 int64_t vlan = br->cfg->flood_vlans[i];
3965 if (vlan >= 0 && vlan < 4096) {
3966 bitmap_set1(rspan_vlans, vlan);
3967 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3970 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3975 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3981 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
3986 for (i = 0; ; i++) {
3987 if (i >= MAX_MIRRORS) {
3988 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3989 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
3992 if (!br->mirrors[i]) {
3997 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4000 br->mirrors[i] = m = xzalloc(sizeof *m);
4003 m->name = xstrdup(cfg->name);
4004 shash_init(&m->src_ports);
4005 shash_init(&m->dst_ports);
4011 mirror_reconfigure_one(m, cfg);
4015 mirror_destroy(struct mirror *m)
4018 struct bridge *br = m->bridge;
4021 for (i = 0; i < br->n_ports; i++) {
4022 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4023 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4026 shash_destroy(&m->src_ports);
4027 shash_destroy(&m->dst_ports);
4030 m->bridge->mirrors[m->idx] = NULL;
4039 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4040 struct shash *names)
4044 for (i = 0; i < n_ports; i++) {
4045 const char *name = ports[i]->name;
4046 if (port_lookup(m->bridge, name)) {
4047 shash_add_once(names, name, NULL);
4049 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4050 "port %s", m->bridge->name, m->name, name);
4056 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4062 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4064 for (i = 0; i < cfg->n_select_vlan; i++) {
4065 int64_t vlan = cfg->select_vlan[i];
4066 if (vlan < 0 || vlan > 4095) {
4067 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4068 m->bridge->name, m->name, vlan);
4070 (*vlans)[n_vlans++] = vlan;
4077 vlan_is_mirrored(const struct mirror *m, int vlan)
4081 for (i = 0; i < m->n_vlans; i++) {
4082 if (m->vlans[i] == vlan) {
4090 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4094 for (i = 0; i < m->n_vlans; i++) {
4095 if (port_trunks_vlan(p, m->vlans[i])) {
4103 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4105 struct shash src_ports, dst_ports;
4106 mirror_mask_t mirror_bit;
4107 struct port *out_port;
4114 if (strcmp(cfg->name, m->name)) {
4116 m->name = xstrdup(cfg->name);
4119 /* Get output port. */
4120 if (cfg->output_port) {
4121 out_port = port_lookup(m->bridge, cfg->output_port->name);
4123 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4124 m->bridge->name, m->name);
4130 if (cfg->output_vlan) {
4131 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4132 "output vlan; ignoring output vlan",
4133 m->bridge->name, m->name);
4135 } else if (cfg->output_vlan) {
4137 out_vlan = *cfg->output_vlan;
4139 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4140 m->bridge->name, m->name);
4145 shash_init(&src_ports);
4146 shash_init(&dst_ports);
4147 if (cfg->select_all) {
4148 for (i = 0; i < m->bridge->n_ports; i++) {
4149 const char *name = m->bridge->ports[i]->name;
4150 shash_add_once(&src_ports, name, NULL);
4151 shash_add_once(&dst_ports, name, NULL);
4156 /* Get ports, and drop duplicates and ports that don't exist. */
4157 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4159 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4162 /* Get all the vlans, and drop duplicate and invalid vlans. */
4163 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4166 /* Update mirror data. */
4167 if (!shash_equal_keys(&m->src_ports, &src_ports)
4168 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4169 || m->n_vlans != n_vlans
4170 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4171 || m->out_port != out_port
4172 || m->out_vlan != out_vlan) {
4173 bridge_flush(m->bridge);
4175 shash_swap(&m->src_ports, &src_ports);
4176 shash_swap(&m->dst_ports, &dst_ports);
4179 m->n_vlans = n_vlans;
4180 m->out_port = out_port;
4181 m->out_vlan = out_vlan;
4184 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4185 for (i = 0; i < m->bridge->n_ports; i++) {
4186 struct port *port = m->bridge->ports[i];
4188 if (shash_find(&m->src_ports, port->name)
4191 ? port_trunks_any_mirrored_vlan(m, port)
4192 : vlan_is_mirrored(m, port->vlan)))) {
4193 port->src_mirrors |= mirror_bit;
4195 port->src_mirrors &= ~mirror_bit;
4198 if (shash_find(&m->dst_ports, port->name)) {
4199 port->dst_mirrors |= mirror_bit;
4201 port->dst_mirrors &= ~mirror_bit;
4206 shash_destroy(&src_ports);
4207 shash_destroy(&dst_ports);