1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
23 #include <sys/socket.h>
25 #include <openflow/openflow.h>
30 #include <sys/socket.h>
31 #include <sys/types.h>
37 #include "dynamic-string.h"
42 #include "mac-learning.h"
45 #include "ofp-print.h"
47 #include "ofproto/netflow.h"
48 #include "ofproto/ofproto.h"
49 #include "ovsdb-data.h"
51 #include "poll-loop.h"
52 #include "port-array.h"
53 #include "proc-net-compat.h"
57 #include "socket-util.h"
58 #include "stream-ssl.h"
64 #include "vswitchd/vswitch-idl.h"
65 #include "xenserver.h"
68 #include "sflow_api.h"
70 VLOG_DEFINE_THIS_MODULE(bridge)
78 /* These members are always valid. */
79 struct port *port; /* Containing port. */
80 size_t port_ifidx; /* Index within containing port. */
81 char *name; /* Host network device name. */
82 tag_type tag; /* Tag associated with this interface. */
83 long long delay_expires; /* Time after which 'enabled' may change. */
85 /* These members are valid only after bridge_reconfigure() causes them to
87 int dp_ifidx; /* Index within kernel datapath. */
88 struct netdev *netdev; /* Network device. */
89 bool enabled; /* May be chosen for flows? */
90 const struct ovsrec_interface *cfg;
93 #define BOND_MASK 0xff
95 int iface_idx; /* Index of assigned iface, or -1 if none. */
96 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
97 tag_type iface_tag; /* Tag associated with iface_idx. */
100 #define MAX_MIRRORS 32
101 typedef uint32_t mirror_mask_t;
102 #define MIRROR_MASK_C(X) UINT32_C(X)
103 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
105 struct bridge *bridge;
108 struct uuid uuid; /* UUID of this "mirror" record in database. */
110 /* Selection criteria. */
111 struct shash src_ports; /* Name is port name; data is always NULL. */
112 struct shash dst_ports; /* Name is port name; data is always NULL. */
117 struct port *out_port;
121 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
123 struct bridge *bridge;
125 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
126 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
127 * NULL if all VLANs are trunked. */
128 const struct ovsrec_port *cfg;
131 /* An ordinary bridge port has 1 interface.
132 * A bridge port for bonding has at least 2 interfaces. */
133 struct iface **ifaces;
134 size_t n_ifaces, allocated_ifaces;
137 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
138 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
139 tag_type active_iface_tag; /* Tag for bcast flows. */
140 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
141 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
142 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
143 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
144 long bond_next_fake_iface_update; /* Next update to fake bond stats. */
145 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
146 long long int bond_next_rebalance; /* Next rebalancing time. */
148 /* Port mirroring info. */
149 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
150 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
151 bool is_mirror_output_port; /* Does port mirroring send frames here? */
154 #define DP_MAX_PORTS 255
156 struct list node; /* Node in global list of bridges. */
157 char *name; /* User-specified arbitrary name. */
158 struct mac_learning *ml; /* MAC learning table. */
159 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
160 const struct ovsrec_bridge *cfg;
162 /* OpenFlow switch processing. */
163 struct ofproto *ofproto; /* OpenFlow switch. */
165 /* Kernel datapath information. */
166 struct dpif *dpif; /* Datapath. */
167 struct port_array ifaces; /* Indexed by kernel datapath port number. */
171 size_t n_ports, allocated_ports;
172 struct shash iface_by_name; /* "struct iface"s indexed by name. */
173 struct shash port_by_name; /* "struct port"s indexed by name. */
176 bool has_bonded_ports;
181 /* Port mirroring. */
182 struct mirror *mirrors[MAX_MIRRORS];
185 /* List of all bridges. */
186 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
188 /* OVSDB IDL used to obtain configuration. */
189 static struct ovsdb_idl *idl;
191 /* Each time this timer expires, the bridge fetches statistics for every
192 * interface and pushes them into the database. */
193 #define IFACE_STATS_INTERVAL (5 * 1000) /* In milliseconds. */
194 static long long int iface_stats_timer = LLONG_MIN;
196 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
197 static void bridge_destroy(struct bridge *);
198 static struct bridge *bridge_lookup(const char *name);
199 static unixctl_cb_func bridge_unixctl_dump_flows;
200 static unixctl_cb_func bridge_unixctl_reconnect;
201 static int bridge_run_one(struct bridge *);
202 static size_t bridge_get_controllers(const struct bridge *br,
203 struct ovsrec_controller ***controllersp);
204 static void bridge_reconfigure_one(struct bridge *);
205 static void bridge_reconfigure_remotes(struct bridge *,
206 const struct sockaddr_in *managers,
208 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
209 static void bridge_fetch_dp_ifaces(struct bridge *);
210 static void bridge_flush(struct bridge *);
211 static void bridge_pick_local_hw_addr(struct bridge *,
212 uint8_t ea[ETH_ADDR_LEN],
213 struct iface **hw_addr_iface);
214 static uint64_t bridge_pick_datapath_id(struct bridge *,
215 const uint8_t bridge_ea[ETH_ADDR_LEN],
216 struct iface *hw_addr_iface);
217 static struct iface *bridge_get_local_iface(struct bridge *);
218 static uint64_t dpid_from_hash(const void *, size_t nbytes);
220 static unixctl_cb_func bridge_unixctl_fdb_show;
222 static void bond_init(void);
223 static void bond_run(struct bridge *);
224 static void bond_wait(struct bridge *);
225 static void bond_rebalance_port(struct port *);
226 static void bond_send_learning_packets(struct port *);
227 static void bond_enable_slave(struct iface *iface, bool enable);
229 static struct port *port_create(struct bridge *, const char *name);
230 static void port_reconfigure(struct port *, const struct ovsrec_port *);
231 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
232 static void port_destroy(struct port *);
233 static struct port *port_lookup(const struct bridge *, const char *name);
234 static struct iface *port_lookup_iface(const struct port *, const char *name);
235 static struct port *port_from_dp_ifidx(const struct bridge *,
237 static void port_update_bond_compat(struct port *);
238 static void port_update_vlan_compat(struct port *);
239 static void port_update_bonding(struct port *);
241 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
242 static void mirror_destroy(struct mirror *);
243 static void mirror_reconfigure(struct bridge *);
244 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
245 static bool vlan_is_mirrored(const struct mirror *, int vlan);
247 static struct iface *iface_create(struct port *port,
248 const struct ovsrec_interface *if_cfg);
249 static void iface_destroy(struct iface *);
250 static struct iface *iface_lookup(const struct bridge *, const char *name);
251 static struct iface *iface_from_dp_ifidx(const struct bridge *,
253 static bool iface_is_internal(const struct bridge *, const char *name);
254 static void iface_set_mac(struct iface *);
255 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
257 /* Hooks into ofproto processing. */
258 static struct ofhooks bridge_ofhooks;
260 /* Public functions. */
262 /* Initializes the bridge module, configuring it to obtain its configuration
263 * from an OVSDB server accessed over 'remote', which should be a string in a
264 * form acceptable to ovsdb_idl_create(). */
266 bridge_init(const char *remote)
268 /* Create connection to database. */
269 idl = ovsdb_idl_create(remote, &ovsrec_idl_class);
271 /* Register unixctl commands. */
272 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
273 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
275 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
280 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
281 * but for which the ovs-vswitchd configuration 'cfg' is required. */
283 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
285 static bool already_configured_once;
286 struct svec bridge_names;
287 struct svec dpif_names, dpif_types;
290 /* Only do this once per ovs-vswitchd run. */
291 if (already_configured_once) {
294 already_configured_once = true;
296 iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
298 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
299 svec_init(&bridge_names);
300 for (i = 0; i < cfg->n_bridges; i++) {
301 svec_add(&bridge_names, cfg->bridges[i]->name);
303 svec_sort(&bridge_names);
305 /* Iterate over all system dpifs and delete any of them that do not appear
307 svec_init(&dpif_names);
308 svec_init(&dpif_types);
309 dp_enumerate_types(&dpif_types);
310 for (i = 0; i < dpif_types.n; i++) {
315 dp_enumerate_names(dpif_types.names[i], &dpif_names);
317 /* For each dpif... */
318 for (j = 0; j < dpif_names.n; j++) {
319 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
321 struct svec all_names;
324 /* ...check whether any of its names is in 'bridge_names'. */
325 svec_init(&all_names);
326 dpif_get_all_names(dpif, &all_names);
327 for (k = 0; k < all_names.n; k++) {
328 if (svec_contains(&bridge_names, all_names.names[k])) {
333 /* No. Delete the dpif. */
337 svec_destroy(&all_names);
342 svec_destroy(&bridge_names);
343 svec_destroy(&dpif_names);
344 svec_destroy(&dpif_types);
347 /* Attempt to create the network device 'iface_name' through the netdev
350 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
353 struct shash options;
357 shash_init(&options);
358 for (i = 0; i < iface_cfg->n_options; i++) {
359 shash_add(&options, iface_cfg->key_options[i],
360 xstrdup(iface_cfg->value_options[i]));
364 struct netdev_options netdev_options;
366 memset(&netdev_options, 0, sizeof netdev_options);
367 netdev_options.name = iface_cfg->name;
368 if (!strcmp(iface_cfg->type, "internal")) {
369 /* An "internal" config type maps to a netdev "system" type. */
370 netdev_options.type = "system";
372 netdev_options.type = iface_cfg->type;
374 netdev_options.args = &options;
375 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
377 error = netdev_open(&netdev_options, &iface->netdev);
380 netdev_get_carrier(iface->netdev, &iface->enabled);
382 } else if (iface->netdev) {
383 const char *netdev_type = netdev_get_type(iface->netdev);
384 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
385 ? iface_cfg->type : NULL;
387 /* An "internal" config type maps to a netdev "system" type. */
388 if (iface_type && !strcmp(iface_type, "internal")) {
389 iface_type = "system";
392 if (!iface_type || !strcmp(netdev_type, iface_type)) {
393 error = netdev_reconfigure(iface->netdev, &options);
395 VLOG_WARN("%s: attempting change device type from %s to %s",
396 iface_cfg->name, netdev_type, iface_type);
400 shash_destroy_free_data(&options);
406 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
408 return set_up_iface(iface_cfg, iface, false);
412 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
413 void *aux OVS_UNUSED)
415 if (!iface->netdev) {
416 int error = set_up_iface(iface->cfg, iface, true);
418 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
428 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
429 void *aux OVS_UNUSED)
431 if (iface->dp_ifidx >= 0) {
432 VLOG_DBG("%s has interface %s on port %d",
434 iface->name, iface->dp_ifidx);
437 VLOG_ERR("%s interface not in %s, dropping",
438 iface->name, dpif_name(br->dpif));
444 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
445 void *aux OVS_UNUSED)
447 /* Set policing attributes. */
448 netdev_set_policing(iface->netdev,
449 iface->cfg->ingress_policing_rate,
450 iface->cfg->ingress_policing_burst);
452 /* Set MAC address of internal interfaces other than the local
454 if (iface->dp_ifidx != ODPP_LOCAL
455 && iface_is_internal(br, iface->name)) {
456 iface_set_mac(iface);
462 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
463 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
464 * deletes from 'br' any ports that no longer have any interfaces. */
466 iterate_and_prune_ifaces(struct bridge *br,
467 bool (*cb)(struct bridge *, struct iface *,
473 for (i = 0; i < br->n_ports; ) {
474 struct port *port = br->ports[i];
475 for (j = 0; j < port->n_ifaces; ) {
476 struct iface *iface = port->ifaces[j];
477 if (cb(br, iface, aux)) {
480 iface_destroy(iface);
484 if (port->n_ifaces) {
487 VLOG_ERR("%s port has no interfaces, dropping", port->name);
493 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
494 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
495 * responsible for freeing '*managersp' (with free()).
497 * You may be asking yourself "why does ovs-vswitchd care?", because
498 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
499 * should not be and in fact is not directly involved in that. But
500 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
501 * it has to tell in-band control where the managers are to enable that.
504 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
505 struct sockaddr_in **managersp, size_t *n_managersp)
507 struct sockaddr_in *managers = NULL;
508 size_t n_managers = 0;
510 if (ovs_cfg->n_managers > 0) {
513 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
514 for (i = 0; i < ovs_cfg->n_managers; i++) {
515 const char *name = ovs_cfg->managers[i];
516 struct sockaddr_in *sin = &managers[i];
518 if ((!strncmp(name, "tcp:", 4)
519 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
520 (!strncmp(name, "ssl:", 4)
521 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
527 *managersp = managers;
528 *n_managersp = n_managers;
532 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
534 struct shash old_br, new_br;
535 struct shash_node *node;
536 struct bridge *br, *next;
537 struct sockaddr_in *managers;
540 int sflow_bridge_number;
542 COVERAGE_INC(bridge_reconfigure);
544 collect_managers(ovs_cfg, &managers, &n_managers);
546 /* Collect old and new bridges. */
549 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
550 shash_add(&old_br, br->name, br);
552 for (i = 0; i < ovs_cfg->n_bridges; i++) {
553 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
554 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
555 VLOG_WARN("more than one bridge named %s", br_cfg->name);
559 /* Get rid of deleted bridges and add new bridges. */
560 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
561 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
568 SHASH_FOR_EACH (node, &new_br) {
569 const char *br_name = node->name;
570 const struct ovsrec_bridge *br_cfg = node->data;
571 br = shash_find_data(&old_br, br_name);
573 /* If the bridge datapath type has changed, we need to tear it
574 * down and recreate. */
575 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
577 bridge_create(br_cfg);
580 bridge_create(br_cfg);
583 shash_destroy(&old_br);
584 shash_destroy(&new_br);
586 /* Reconfigure all bridges. */
587 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
588 bridge_reconfigure_one(br);
591 /* Add and delete ports on all datapaths.
593 * The kernel will reject any attempt to add a given port to a datapath if
594 * that port already belongs to a different datapath, so we must do all
595 * port deletions before any port additions. */
596 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
597 struct odp_port *dpif_ports;
599 struct shash want_ifaces;
601 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
602 bridge_get_all_ifaces(br, &want_ifaces);
603 for (i = 0; i < n_dpif_ports; i++) {
604 const struct odp_port *p = &dpif_ports[i];
605 if (!shash_find(&want_ifaces, p->devname)
606 && strcmp(p->devname, br->name)) {
607 int retval = dpif_port_del(br->dpif, p->port);
609 VLOG_ERR("failed to remove %s interface from %s: %s",
610 p->devname, dpif_name(br->dpif),
615 shash_destroy(&want_ifaces);
618 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
619 struct odp_port *dpif_ports;
621 struct shash cur_ifaces, want_ifaces;
622 struct shash_node *node;
624 /* Get the set of interfaces currently in this datapath. */
625 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
626 shash_init(&cur_ifaces);
627 for (i = 0; i < n_dpif_ports; i++) {
628 const char *name = dpif_ports[i].devname;
629 shash_add_once(&cur_ifaces, name, NULL);
633 /* Get the set of interfaces we want on this datapath. */
634 bridge_get_all_ifaces(br, &want_ifaces);
636 SHASH_FOR_EACH (node, &want_ifaces) {
637 const char *if_name = node->name;
638 struct iface *iface = node->data;
640 if (shash_find(&cur_ifaces, if_name)) {
641 /* Already exists, just reconfigure it. */
643 reconfigure_iface(iface->cfg, iface);
646 /* Need to add to datapath. */
650 /* Add to datapath. */
651 internal = iface_is_internal(br, if_name);
652 error = dpif_port_add(br->dpif, if_name,
653 internal ? ODP_PORT_INTERNAL : 0, NULL);
654 if (error == EFBIG) {
655 VLOG_ERR("ran out of valid port numbers on %s",
656 dpif_name(br->dpif));
659 VLOG_ERR("failed to add %s interface to %s: %s",
660 if_name, dpif_name(br->dpif), strerror(error));
664 shash_destroy(&cur_ifaces);
665 shash_destroy(&want_ifaces);
667 sflow_bridge_number = 0;
668 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
671 struct iface *local_iface;
672 struct iface *hw_addr_iface;
675 bridge_fetch_dp_ifaces(br);
677 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
678 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
680 /* Pick local port hardware address, datapath ID. */
681 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
682 local_iface = bridge_get_local_iface(br);
684 int error = netdev_set_etheraddr(local_iface->netdev, ea);
686 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
687 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
688 "Ethernet address: %s",
689 br->name, strerror(error));
693 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
694 ofproto_set_datapath_id(br->ofproto, dpid);
696 dpid_string = xasprintf("%016"PRIx64, dpid);
697 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
700 /* Set NetFlow configuration on this bridge. */
701 if (br->cfg->netflow) {
702 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
703 struct netflow_options opts;
705 memset(&opts, 0, sizeof opts);
707 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
708 if (nf_cfg->engine_type) {
709 opts.engine_type = *nf_cfg->engine_type;
711 if (nf_cfg->engine_id) {
712 opts.engine_id = *nf_cfg->engine_id;
715 opts.active_timeout = nf_cfg->active_timeout;
716 if (!opts.active_timeout) {
717 opts.active_timeout = -1;
718 } else if (opts.active_timeout < 0) {
719 VLOG_WARN("bridge %s: active timeout interval set to negative "
720 "value, using default instead (%d seconds)", br->name,
721 NF_ACTIVE_TIMEOUT_DEFAULT);
722 opts.active_timeout = -1;
725 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
726 if (opts.add_id_to_iface) {
727 if (opts.engine_id > 0x7f) {
728 VLOG_WARN("bridge %s: netflow port mangling may conflict "
729 "with another vswitch, choose an engine id less "
730 "than 128", br->name);
732 if (br->n_ports > 508) {
733 VLOG_WARN("bridge %s: netflow port mangling will conflict "
734 "with another port when more than 508 ports are "
739 opts.collectors.n = nf_cfg->n_targets;
740 opts.collectors.names = nf_cfg->targets;
741 if (ofproto_set_netflow(br->ofproto, &opts)) {
742 VLOG_ERR("bridge %s: problem setting netflow collectors",
746 ofproto_set_netflow(br->ofproto, NULL);
749 /* Set sFlow configuration on this bridge. */
750 if (br->cfg->sflow) {
751 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
752 struct ovsrec_controller **controllers;
753 struct ofproto_sflow_options oso;
754 size_t n_controllers;
757 memset(&oso, 0, sizeof oso);
759 oso.targets.n = sflow_cfg->n_targets;
760 oso.targets.names = sflow_cfg->targets;
762 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
763 if (sflow_cfg->sampling) {
764 oso.sampling_rate = *sflow_cfg->sampling;
767 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
768 if (sflow_cfg->polling) {
769 oso.polling_interval = *sflow_cfg->polling;
772 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
773 if (sflow_cfg->header) {
774 oso.header_len = *sflow_cfg->header;
777 oso.sub_id = sflow_bridge_number++;
778 oso.agent_device = sflow_cfg->agent;
780 oso.control_ip = NULL;
781 n_controllers = bridge_get_controllers(br, &controllers);
782 for (i = 0; i < n_controllers; i++) {
783 if (controllers[i]->local_ip) {
784 oso.control_ip = controllers[i]->local_ip;
788 ofproto_set_sflow(br->ofproto, &oso);
790 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
792 ofproto_set_sflow(br->ofproto, NULL);
795 /* Update the controller and related settings. It would be more
796 * straightforward to call this from bridge_reconfigure_one(), but we
797 * can't do it there for two reasons. First, and most importantly, at
798 * that point we don't know the dp_ifidx of any interfaces that have
799 * been added to the bridge (because we haven't actually added them to
800 * the datapath). Second, at that point we haven't set the datapath ID
801 * yet; when a controller is configured, resetting the datapath ID will
802 * immediately disconnect from the controller, so it's better to set
803 * the datapath ID before the controller. */
804 bridge_reconfigure_remotes(br, managers, n_managers);
806 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
807 for (i = 0; i < br->n_ports; i++) {
808 struct port *port = br->ports[i];
811 port_update_vlan_compat(port);
812 port_update_bonding(port);
814 for (j = 0; j < port->n_ifaces; j++) {
815 iface_update_qos(port->ifaces[j], port->cfg->qos);
819 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
820 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
827 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
828 const struct ovsdb_idl_column *column,
831 const struct ovsdb_datum *datum;
832 union ovsdb_atom atom;
835 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
836 atom.string = (char *) key;
837 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
838 return idx == UINT_MAX ? NULL : datum->values[idx].string;
842 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
844 return get_ovsrec_key_value(&br_cfg->header_,
845 &ovsrec_bridge_col_other_config, key);
849 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
850 struct iface **hw_addr_iface)
856 *hw_addr_iface = NULL;
858 /* Did the user request a particular MAC? */
859 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
860 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
861 if (eth_addr_is_multicast(ea)) {
862 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
863 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
864 } else if (eth_addr_is_zero(ea)) {
865 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
871 /* Otherwise choose the minimum non-local MAC address among all of the
873 memset(ea, 0xff, sizeof ea);
874 for (i = 0; i < br->n_ports; i++) {
875 struct port *port = br->ports[i];
876 uint8_t iface_ea[ETH_ADDR_LEN];
879 /* Mirror output ports don't participate. */
880 if (port->is_mirror_output_port) {
884 /* Choose the MAC address to represent the port. */
885 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
886 /* Find the interface with this Ethernet address (if any) so that
887 * we can provide the correct devname to the caller. */
889 for (j = 0; j < port->n_ifaces; j++) {
890 struct iface *candidate = port->ifaces[j];
891 uint8_t candidate_ea[ETH_ADDR_LEN];
892 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
893 && eth_addr_equals(iface_ea, candidate_ea)) {
898 /* Choose the interface whose MAC address will represent the port.
899 * The Linux kernel bonding code always chooses the MAC address of
900 * the first slave added to a bond, and the Fedora networking
901 * scripts always add slaves to a bond in alphabetical order, so
902 * for compatibility we choose the interface with the name that is
903 * first in alphabetical order. */
904 iface = port->ifaces[0];
905 for (j = 1; j < port->n_ifaces; j++) {
906 struct iface *candidate = port->ifaces[j];
907 if (strcmp(candidate->name, iface->name) < 0) {
912 /* The local port doesn't count (since we're trying to choose its
913 * MAC address anyway). */
914 if (iface->dp_ifidx == ODPP_LOCAL) {
919 error = netdev_get_etheraddr(iface->netdev, iface_ea);
921 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
922 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
923 iface->name, strerror(error));
928 /* Compare against our current choice. */
929 if (!eth_addr_is_multicast(iface_ea) &&
930 !eth_addr_is_local(iface_ea) &&
931 !eth_addr_is_reserved(iface_ea) &&
932 !eth_addr_is_zero(iface_ea) &&
933 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
935 memcpy(ea, iface_ea, ETH_ADDR_LEN);
936 *hw_addr_iface = iface;
939 if (eth_addr_is_multicast(ea)) {
940 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
941 *hw_addr_iface = NULL;
942 VLOG_WARN("bridge %s: using default bridge Ethernet "
943 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
945 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
946 br->name, ETH_ADDR_ARGS(ea));
950 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
951 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
952 * an interface on 'br', then that interface must be passed in as
953 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
954 * 'hw_addr_iface' must be passed in as a null pointer. */
956 bridge_pick_datapath_id(struct bridge *br,
957 const uint8_t bridge_ea[ETH_ADDR_LEN],
958 struct iface *hw_addr_iface)
961 * The procedure for choosing a bridge MAC address will, in the most
962 * ordinary case, also choose a unique MAC that we can use as a datapath
963 * ID. In some special cases, though, multiple bridges will end up with
964 * the same MAC address. This is OK for the bridges, but it will confuse
965 * the OpenFlow controller, because each datapath needs a unique datapath
968 * Datapath IDs must be unique. It is also very desirable that they be
969 * stable from one run to the next, so that policy set on a datapath
972 const char *datapath_id;
975 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
976 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
982 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
984 * A bridge whose MAC address is taken from a VLAN network device
985 * (that is, a network device created with vconfig(8) or similar
986 * tool) will have the same MAC address as a bridge on the VLAN
987 * device's physical network device.
989 * Handle this case by hashing the physical network device MAC
990 * along with the VLAN identifier.
992 uint8_t buf[ETH_ADDR_LEN + 2];
993 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
994 buf[ETH_ADDR_LEN] = vlan >> 8;
995 buf[ETH_ADDR_LEN + 1] = vlan;
996 return dpid_from_hash(buf, sizeof buf);
999 * Assume that this bridge's MAC address is unique, since it
1000 * doesn't fit any of the cases we handle specially.
1005 * A purely internal bridge, that is, one that has no non-virtual
1006 * network devices on it at all, is more difficult because it has no
1007 * natural unique identifier at all.
1009 * When the host is a XenServer, we handle this case by hashing the
1010 * host's UUID with the name of the bridge. Names of bridges are
1011 * persistent across XenServer reboots, although they can be reused if
1012 * an internal network is destroyed and then a new one is later
1013 * created, so this is fairly effective.
1015 * When the host is not a XenServer, we punt by using a random MAC
1016 * address on each run.
1018 const char *host_uuid = xenserver_get_host_uuid();
1020 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1021 dpid = dpid_from_hash(combined, strlen(combined));
1027 return eth_addr_to_uint64(bridge_ea);
1031 dpid_from_hash(const void *data, size_t n)
1033 uint8_t hash[SHA1_DIGEST_SIZE];
1035 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1036 sha1_bytes(data, n, hash);
1037 eth_addr_mark_random(hash);
1038 return eth_addr_to_uint64(hash);
1042 iface_refresh_stats(struct iface *iface)
1048 static const struct iface_stat iface_stats[] = {
1049 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1050 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1051 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1052 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1053 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1054 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1055 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1056 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1057 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1058 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1059 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1060 { "collisions", offsetof(struct netdev_stats, collisions) },
1062 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1063 const struct iface_stat *s;
1065 char *keys[N_STATS];
1066 int64_t values[N_STATS];
1069 struct netdev_stats stats;
1071 /* Intentionally ignore return value, since errors will set 'stats' to
1072 * all-1s, and we will deal with that correctly below. */
1073 netdev_get_stats(iface->netdev, &stats);
1076 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1077 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1078 if (value != UINT64_MAX) {
1085 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1091 const struct ovsrec_open_vswitch *cfg;
1093 bool datapath_destroyed;
1094 bool database_changed;
1097 /* Let each bridge do the work that it needs to do. */
1098 datapath_destroyed = false;
1099 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1100 int error = bridge_run_one(br);
1102 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1103 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1104 "forcing reconfiguration", br->name);
1105 datapath_destroyed = true;
1109 /* (Re)configure if necessary. */
1110 database_changed = ovsdb_idl_run(idl);
1111 cfg = ovsrec_open_vswitch_first(idl);
1112 if (database_changed || datapath_destroyed) {
1114 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1116 bridge_configure_once(cfg);
1117 bridge_reconfigure(cfg);
1119 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1120 ovsdb_idl_txn_commit(txn);
1121 ovsdb_idl_txn_destroy(txn); /* XXX */
1123 /* We still need to reconfigure to avoid dangling pointers to
1124 * now-destroyed ovsrec structures inside bridge data. */
1125 static const struct ovsrec_open_vswitch null_cfg;
1127 bridge_reconfigure(&null_cfg);
1132 /* Re-configure SSL. We do this on every trip through the main loop,
1133 * instead of just when the database changes, because the contents of the
1134 * key and certificate files can change without the database changing. */
1135 if (cfg && cfg->ssl) {
1136 const struct ovsrec_ssl *ssl = cfg->ssl;
1138 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1139 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1143 /* Refresh interface stats if necessary. */
1144 if (time_msec() >= iface_stats_timer) {
1145 struct ovsdb_idl_txn *txn;
1147 txn = ovsdb_idl_txn_create(idl);
1148 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1151 for (i = 0; i < br->n_ports; i++) {
1152 struct port *port = br->ports[i];
1155 for (j = 0; j < port->n_ifaces; j++) {
1156 struct iface *iface = port->ifaces[j];
1157 iface_refresh_stats(iface);
1161 ovsdb_idl_txn_commit(txn);
1162 ovsdb_idl_txn_destroy(txn); /* XXX */
1164 iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
1173 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1174 ofproto_wait(br->ofproto);
1175 if (ofproto_has_primary_controller(br->ofproto)) {
1179 mac_learning_wait(br->ml);
1182 ovsdb_idl_wait(idl);
1183 poll_timer_wait_until(iface_stats_timer);
1186 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1187 * configuration changes. */
1189 bridge_flush(struct bridge *br)
1191 COVERAGE_INC(bridge_flush);
1193 mac_learning_flush(br->ml);
1196 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1197 * such interface. */
1198 static struct iface *
1199 bridge_get_local_iface(struct bridge *br)
1203 for (i = 0; i < br->n_ports; i++) {
1204 struct port *port = br->ports[i];
1205 for (j = 0; j < port->n_ifaces; j++) {
1206 struct iface *iface = port->ifaces[j];
1207 if (iface->dp_ifidx == ODPP_LOCAL) {
1216 /* Bridge unixctl user interface functions. */
1218 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1219 const char *args, void *aux OVS_UNUSED)
1221 struct ds ds = DS_EMPTY_INITIALIZER;
1222 const struct bridge *br;
1223 const struct mac_entry *e;
1225 br = bridge_lookup(args);
1227 unixctl_command_reply(conn, 501, "no such bridge");
1231 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1232 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1233 if (e->port < 0 || e->port >= br->n_ports) {
1236 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1237 br->ports[e->port]->ifaces[0]->dp_ifidx,
1238 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1240 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1244 /* Bridge reconfiguration functions. */
1245 static struct bridge *
1246 bridge_create(const struct ovsrec_bridge *br_cfg)
1251 assert(!bridge_lookup(br_cfg->name));
1252 br = xzalloc(sizeof *br);
1254 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1260 dpif_flow_flush(br->dpif);
1262 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1265 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1267 dpif_delete(br->dpif);
1268 dpif_close(br->dpif);
1273 br->name = xstrdup(br_cfg->name);
1275 br->ml = mac_learning_create();
1276 eth_addr_nicira_random(br->default_ea);
1278 port_array_init(&br->ifaces);
1280 shash_init(&br->port_by_name);
1281 shash_init(&br->iface_by_name);
1285 list_push_back(&all_bridges, &br->node);
1287 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1293 bridge_destroy(struct bridge *br)
1298 while (br->n_ports > 0) {
1299 port_destroy(br->ports[br->n_ports - 1]);
1301 list_remove(&br->node);
1302 error = dpif_delete(br->dpif);
1303 if (error && error != ENOENT) {
1304 VLOG_ERR("failed to delete %s: %s",
1305 dpif_name(br->dpif), strerror(error));
1307 dpif_close(br->dpif);
1308 ofproto_destroy(br->ofproto);
1309 mac_learning_destroy(br->ml);
1310 port_array_destroy(&br->ifaces);
1311 shash_destroy(&br->port_by_name);
1312 shash_destroy(&br->iface_by_name);
1319 static struct bridge *
1320 bridge_lookup(const char *name)
1324 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1325 if (!strcmp(br->name, name)) {
1332 /* Handle requests for a listing of all flows known by the OpenFlow
1333 * stack, including those normally hidden. */
1335 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1336 const char *args, void *aux OVS_UNUSED)
1341 br = bridge_lookup(args);
1343 unixctl_command_reply(conn, 501, "Unknown bridge");
1348 ofproto_get_all_flows(br->ofproto, &results);
1350 unixctl_command_reply(conn, 200, ds_cstr(&results));
1351 ds_destroy(&results);
1354 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1355 * connections and reconnect. If BRIDGE is not specified, then all bridges
1356 * drop their controller connections and reconnect. */
1358 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1359 const char *args, void *aux OVS_UNUSED)
1362 if (args[0] != '\0') {
1363 br = bridge_lookup(args);
1365 unixctl_command_reply(conn, 501, "Unknown bridge");
1368 ofproto_reconnect_controllers(br->ofproto);
1370 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1371 ofproto_reconnect_controllers(br->ofproto);
1374 unixctl_command_reply(conn, 200, NULL);
1378 bridge_run_one(struct bridge *br)
1382 error = ofproto_run1(br->ofproto);
1387 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1390 error = ofproto_run2(br->ofproto, br->flush);
1397 bridge_get_controllers(const struct bridge *br,
1398 struct ovsrec_controller ***controllersp)
1400 struct ovsrec_controller **controllers;
1401 size_t n_controllers;
1403 controllers = br->cfg->controller;
1404 n_controllers = br->cfg->n_controller;
1406 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1412 *controllersp = controllers;
1414 return n_controllers;
1418 bridge_reconfigure_one(struct bridge *br)
1420 struct shash old_ports, new_ports;
1421 struct svec snoops, old_snoops;
1422 struct shash_node *node;
1423 enum ofproto_fail_mode fail_mode;
1426 /* Collect old ports. */
1427 shash_init(&old_ports);
1428 for (i = 0; i < br->n_ports; i++) {
1429 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1432 /* Collect new ports. */
1433 shash_init(&new_ports);
1434 for (i = 0; i < br->cfg->n_ports; i++) {
1435 const char *name = br->cfg->ports[i]->name;
1436 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1437 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1442 /* If we have a controller, then we need a local port. Complain if the
1443 * user didn't specify one.
1445 * XXX perhaps we should synthesize a port ourselves in this case. */
1446 if (bridge_get_controllers(br, NULL)) {
1447 char local_name[IF_NAMESIZE];
1450 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1451 local_name, sizeof local_name);
1452 if (!error && !shash_find(&new_ports, local_name)) {
1453 VLOG_WARN("bridge %s: controller specified but no local port "
1454 "(port named %s) defined",
1455 br->name, local_name);
1459 /* Get rid of deleted ports.
1460 * Get rid of deleted interfaces on ports that still exist. */
1461 SHASH_FOR_EACH (node, &old_ports) {
1462 struct port *port = node->data;
1463 const struct ovsrec_port *port_cfg;
1465 port_cfg = shash_find_data(&new_ports, node->name);
1469 port_del_ifaces(port, port_cfg);
1473 /* Create new ports.
1474 * Add new interfaces to existing ports.
1475 * Reconfigure existing ports. */
1476 SHASH_FOR_EACH (node, &new_ports) {
1477 struct port *port = shash_find_data(&old_ports, node->name);
1479 port = port_create(br, node->name);
1482 port_reconfigure(port, node->data);
1483 if (!port->n_ifaces) {
1484 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1485 br->name, port->name);
1489 shash_destroy(&old_ports);
1490 shash_destroy(&new_ports);
1492 /* Set the fail-mode */
1493 fail_mode = !br->cfg->fail_mode
1494 || !strcmp(br->cfg->fail_mode, "standalone")
1495 ? OFPROTO_FAIL_STANDALONE
1496 : OFPROTO_FAIL_SECURE;
1497 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1498 && !ofproto_has_primary_controller(br->ofproto)) {
1499 ofproto_flush_flows(br->ofproto);
1501 ofproto_set_fail_mode(br->ofproto, fail_mode);
1503 /* Delete all flows if we're switching from connected to standalone or vice
1504 * versa. (XXX Should we delete all flows if we are switching from one
1505 * controller to another?) */
1507 /* Configure OpenFlow controller connection snooping. */
1509 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1510 ovs_rundir, br->name));
1511 svec_init(&old_snoops);
1512 ofproto_get_snoops(br->ofproto, &old_snoops);
1513 if (!svec_equal(&snoops, &old_snoops)) {
1514 ofproto_set_snoops(br->ofproto, &snoops);
1516 svec_destroy(&snoops);
1517 svec_destroy(&old_snoops);
1519 mirror_reconfigure(br);
1522 /* Initializes 'oc' appropriately as a management service controller for
1525 * The caller must free oc->target when it is no longer needed. */
1527 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1528 struct ofproto_controller *oc)
1530 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir, br->name);
1531 oc->max_backoff = 0;
1532 oc->probe_interval = 60;
1533 oc->band = OFPROTO_OUT_OF_BAND;
1534 oc->accept_re = NULL;
1535 oc->update_resolv_conf = false;
1537 oc->burst_limit = 0;
1540 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1542 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1543 struct ofproto_controller *oc)
1545 oc->target = c->target;
1546 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1547 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1548 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1549 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1550 oc->accept_re = c->discover_accept_regex;
1551 oc->update_resolv_conf = c->discover_update_resolv_conf;
1552 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1553 oc->burst_limit = (c->controller_burst_limit
1554 ? *c->controller_burst_limit : 0);
1557 /* Configures the IP stack for 'br''s local interface properly according to the
1558 * configuration in 'c'. */
1560 bridge_configure_local_iface_netdev(struct bridge *br,
1561 struct ovsrec_controller *c)
1563 struct netdev *netdev;
1564 struct in_addr mask, gateway;
1566 struct iface *local_iface;
1569 /* Controller discovery does its own TCP/IP configuration later. */
1570 if (strcmp(c->target, "discover")) {
1574 /* If there's no local interface or no IP address, give up. */
1575 local_iface = bridge_get_local_iface(br);
1576 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1580 /* Bring up the local interface. */
1581 netdev = local_iface->netdev;
1582 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1584 /* Configure the IP address and netmask. */
1585 if (!c->local_netmask
1586 || !inet_aton(c->local_netmask, &mask)
1588 mask.s_addr = guess_netmask(ip.s_addr);
1590 if (!netdev_set_in4(netdev, ip, mask)) {
1591 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1592 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1595 /* Configure the default gateway. */
1596 if (c->local_gateway
1597 && inet_aton(c->local_gateway, &gateway)
1598 && gateway.s_addr) {
1599 if (!netdev_add_router(netdev, gateway)) {
1600 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1601 br->name, IP_ARGS(&gateway.s_addr));
1607 bridge_reconfigure_remotes(struct bridge *br,
1608 const struct sockaddr_in *managers,
1611 struct ovsrec_controller **controllers;
1612 size_t n_controllers;
1615 struct ofproto_controller *ocs;
1619 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1620 had_primary = ofproto_has_primary_controller(br->ofproto);
1622 n_controllers = bridge_get_controllers(br, &controllers);
1624 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1627 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1628 for (i = 0; i < n_controllers; i++) {
1629 struct ovsrec_controller *c = controllers[i];
1631 if (!strncmp(c->target, "punix:", 6)
1632 || !strncmp(c->target, "unix:", 5)) {
1633 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1635 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1636 * domain sockets and overwriting arbitrary local files. */
1637 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1638 "\"%s\" due to possibility for remote exploit",
1639 dpif_name(br->dpif), c->target);
1643 bridge_configure_local_iface_netdev(br, c);
1644 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs++]);
1647 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1648 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1651 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1652 ofproto_flush_flows(br->ofproto);
1655 /* If there are no controllers and the bridge is in standalone
1656 * mode, set up a flow that matches every packet and directs
1657 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1658 * switch is in secure mode and we won't pass any traffic until
1659 * a controller has been defined and it tells us to do so. */
1661 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1662 union ofp_action action;
1665 memset(&action, 0, sizeof action);
1666 action.type = htons(OFPAT_OUTPUT);
1667 action.output.len = htons(sizeof action);
1668 action.output.port = htons(OFPP_NORMAL);
1669 memset(&flow, 0, sizeof flow);
1670 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1675 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1680 for (i = 0; i < br->n_ports; i++) {
1681 struct port *port = br->ports[i];
1682 for (j = 0; j < port->n_ifaces; j++) {
1683 struct iface *iface = port->ifaces[j];
1684 shash_add_once(ifaces, iface->name, iface);
1686 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1687 shash_add_once(ifaces, port->name, NULL);
1692 /* For robustness, in case the administrator moves around datapath ports behind
1693 * our back, we re-check all the datapath port numbers here.
1695 * This function will set the 'dp_ifidx' members of interfaces that have
1696 * disappeared to -1, so only call this function from a context where those
1697 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1698 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1699 * datapath, which doesn't support UINT16_MAX+1 ports. */
1701 bridge_fetch_dp_ifaces(struct bridge *br)
1703 struct odp_port *dpif_ports;
1704 size_t n_dpif_ports;
1707 /* Reset all interface numbers. */
1708 for (i = 0; i < br->n_ports; i++) {
1709 struct port *port = br->ports[i];
1710 for (j = 0; j < port->n_ifaces; j++) {
1711 struct iface *iface = port->ifaces[j];
1712 iface->dp_ifidx = -1;
1715 port_array_clear(&br->ifaces);
1717 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1718 for (i = 0; i < n_dpif_ports; i++) {
1719 struct odp_port *p = &dpif_ports[i];
1720 struct iface *iface = iface_lookup(br, p->devname);
1722 if (iface->dp_ifidx >= 0) {
1723 VLOG_WARN("%s reported interface %s twice",
1724 dpif_name(br->dpif), p->devname);
1725 } else if (iface_from_dp_ifidx(br, p->port)) {
1726 VLOG_WARN("%s reported interface %"PRIu16" twice",
1727 dpif_name(br->dpif), p->port);
1729 port_array_set(&br->ifaces, p->port, iface);
1730 iface->dp_ifidx = p->port;
1734 int64_t ofport = (iface->dp_ifidx >= 0
1735 ? odp_port_to_ofp_port(iface->dp_ifidx)
1737 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1744 /* Bridge packet processing functions. */
1747 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1749 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1752 static struct bond_entry *
1753 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1755 return &port->bond_hash[bond_hash(mac)];
1759 bond_choose_iface(const struct port *port)
1761 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1762 size_t i, best_down_slave = -1;
1763 long long next_delay_expiration = LLONG_MAX;
1765 for (i = 0; i < port->n_ifaces; i++) {
1766 struct iface *iface = port->ifaces[i];
1768 if (iface->enabled) {
1770 } else if (iface->delay_expires < next_delay_expiration) {
1771 best_down_slave = i;
1772 next_delay_expiration = iface->delay_expires;
1776 if (best_down_slave != -1) {
1777 struct iface *iface = port->ifaces[best_down_slave];
1779 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1780 "since no other interface is up", iface->name,
1781 iface->delay_expires - time_msec());
1782 bond_enable_slave(iface, true);
1785 return best_down_slave;
1789 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1790 uint16_t *dp_ifidx, tag_type *tags)
1792 struct iface *iface;
1794 assert(port->n_ifaces);
1795 if (port->n_ifaces == 1) {
1796 iface = port->ifaces[0];
1798 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1799 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1800 || !port->ifaces[e->iface_idx]->enabled) {
1801 /* XXX select interface properly. The current interface selection
1802 * is only good for testing the rebalancing code. */
1803 e->iface_idx = bond_choose_iface(port);
1804 if (e->iface_idx < 0) {
1805 *tags |= port->no_ifaces_tag;
1808 e->iface_tag = tag_create_random();
1809 ((struct port *) port)->bond_compat_is_stale = true;
1811 *tags |= e->iface_tag;
1812 iface = port->ifaces[e->iface_idx];
1814 *dp_ifidx = iface->dp_ifidx;
1815 *tags |= iface->tag; /* Currently only used for bonding. */
1820 bond_link_status_update(struct iface *iface, bool carrier)
1822 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1823 struct port *port = iface->port;
1825 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1826 /* Nothing to do. */
1829 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1830 iface->name, carrier ? "detected" : "dropped");
1831 if (carrier == iface->enabled) {
1832 iface->delay_expires = LLONG_MAX;
1833 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1834 iface->name, carrier ? "disabled" : "enabled");
1835 } else if (carrier && port->active_iface < 0) {
1836 bond_enable_slave(iface, true);
1837 if (port->updelay) {
1838 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1839 "other interface is up", iface->name, port->updelay);
1842 int delay = carrier ? port->updelay : port->downdelay;
1843 iface->delay_expires = time_msec() + delay;
1846 "interface %s: will be %s if it stays %s for %d ms",
1848 carrier ? "enabled" : "disabled",
1849 carrier ? "up" : "down",
1856 bond_choose_active_iface(struct port *port)
1858 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1860 port->active_iface = bond_choose_iface(port);
1861 port->active_iface_tag = tag_create_random();
1862 if (port->active_iface >= 0) {
1863 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1864 port->name, port->ifaces[port->active_iface]->name);
1866 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1872 bond_enable_slave(struct iface *iface, bool enable)
1874 struct port *port = iface->port;
1875 struct bridge *br = port->bridge;
1877 /* This acts as a recursion check. If the act of disabling a slave
1878 * causes a different slave to be enabled, the flag will allow us to
1879 * skip redundant work when we reenter this function. It must be
1880 * cleared on exit to keep things safe with multiple bonds. */
1881 static bool moving_active_iface = false;
1883 iface->delay_expires = LLONG_MAX;
1884 if (enable == iface->enabled) {
1888 iface->enabled = enable;
1889 if (!iface->enabled) {
1890 VLOG_WARN("interface %s: disabled", iface->name);
1891 ofproto_revalidate(br->ofproto, iface->tag);
1892 if (iface->port_ifidx == port->active_iface) {
1893 ofproto_revalidate(br->ofproto,
1894 port->active_iface_tag);
1896 /* Disabling a slave can lead to another slave being immediately
1897 * enabled if there will be no active slaves but one is waiting
1898 * on an updelay. In this case we do not need to run most of the
1899 * code for the newly enabled slave since there was no period
1900 * without an active slave and it is redundant with the disabling
1902 moving_active_iface = true;
1903 bond_choose_active_iface(port);
1905 bond_send_learning_packets(port);
1907 VLOG_WARN("interface %s: enabled", iface->name);
1908 if (port->active_iface < 0 && !moving_active_iface) {
1909 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1910 bond_choose_active_iface(port);
1911 bond_send_learning_packets(port);
1913 iface->tag = tag_create_random();
1916 moving_active_iface = false;
1917 port->bond_compat_is_stale = true;
1920 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1921 * bond interface. */
1923 bond_update_fake_iface_stats(struct port *port)
1925 struct netdev_stats bond_stats;
1926 struct netdev *bond_dev;
1929 memset(&bond_stats, 0, sizeof bond_stats);
1931 for (i = 0; i < port->n_ifaces; i++) {
1932 struct netdev_stats slave_stats;
1934 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1935 /* XXX: We swap the stats here because they are swapped back when
1936 * reported by the internal device. The reason for this is
1937 * internal devices normally represent packets going into the system
1938 * but when used as fake bond device they represent packets leaving
1939 * the system. We really should do this in the internal device
1940 * itself because changing it here reverses the counts from the
1941 * perspective of the switch. However, the internal device doesn't
1942 * know what type of device it represents so we have to do it here
1944 bond_stats.tx_packets += slave_stats.rx_packets;
1945 bond_stats.tx_bytes += slave_stats.rx_bytes;
1946 bond_stats.rx_packets += slave_stats.tx_packets;
1947 bond_stats.rx_bytes += slave_stats.tx_bytes;
1951 if (!netdev_open_default(port->name, &bond_dev)) {
1952 netdev_set_stats(bond_dev, &bond_stats);
1953 netdev_close(bond_dev);
1958 bond_run(struct bridge *br)
1962 for (i = 0; i < br->n_ports; i++) {
1963 struct port *port = br->ports[i];
1965 if (port->n_ifaces >= 2) {
1966 for (j = 0; j < port->n_ifaces; j++) {
1967 struct iface *iface = port->ifaces[j];
1968 if (time_msec() >= iface->delay_expires) {
1969 bond_enable_slave(iface, !iface->enabled);
1973 if (port->bond_fake_iface
1974 && time_msec() >= port->bond_next_fake_iface_update) {
1975 bond_update_fake_iface_stats(port);
1976 port->bond_next_fake_iface_update = time_msec() + 1000;
1980 if (port->bond_compat_is_stale) {
1981 port->bond_compat_is_stale = false;
1982 port_update_bond_compat(port);
1988 bond_wait(struct bridge *br)
1992 for (i = 0; i < br->n_ports; i++) {
1993 struct port *port = br->ports[i];
1994 if (port->n_ifaces < 2) {
1997 for (j = 0; j < port->n_ifaces; j++) {
1998 struct iface *iface = port->ifaces[j];
1999 if (iface->delay_expires != LLONG_MAX) {
2000 poll_timer_wait_until(iface->delay_expires);
2003 if (port->bond_fake_iface) {
2004 poll_timer_wait_until(port->bond_next_fake_iface_update);
2010 set_dst(struct dst *p, const flow_t *flow,
2011 const struct port *in_port, const struct port *out_port,
2014 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2015 : in_port->vlan >= 0 ? in_port->vlan
2016 : ntohs(flow->dl_vlan));
2017 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
2021 swap_dst(struct dst *p, struct dst *q)
2023 struct dst tmp = *p;
2028 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2029 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2030 * that we push to the datapath. We could in fact fully sort the array by
2031 * vlan, but in most cases there are at most two different vlan tags so that's
2032 * possibly overkill.) */
2034 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2036 struct dst *first = dsts;
2037 struct dst *last = dsts + n_dsts;
2039 while (first != last) {
2041 * - All dsts < first have vlan == 'vlan'.
2042 * - All dsts >= last have vlan != 'vlan'.
2043 * - first < last. */
2044 while (first->vlan == vlan) {
2045 if (++first == last) {
2050 /* Same invariants, plus one additional:
2051 * - first->vlan != vlan.
2053 while (last[-1].vlan != vlan) {
2054 if (--last == first) {
2059 /* Same invariants, plus one additional:
2060 * - last[-1].vlan == vlan.*/
2061 swap_dst(first++, --last);
2066 mirror_mask_ffs(mirror_mask_t mask)
2068 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2073 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2074 const struct dst *test)
2077 for (i = 0; i < n_dsts; i++) {
2078 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2086 port_trunks_vlan(const struct port *port, uint16_t vlan)
2088 return (port->vlan < 0
2089 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2093 port_includes_vlan(const struct port *port, uint16_t vlan)
2095 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2099 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2100 const struct port *in_port, const struct port *out_port,
2101 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2103 mirror_mask_t mirrors = in_port->src_mirrors;
2104 struct dst *dst = dsts;
2107 if (out_port == FLOOD_PORT) {
2108 /* XXX use ODP_FLOOD if no vlans or bonding. */
2109 /* XXX even better, define each VLAN as a datapath port group */
2110 for (i = 0; i < br->n_ports; i++) {
2111 struct port *port = br->ports[i];
2112 if (port != in_port && port_includes_vlan(port, vlan)
2113 && !port->is_mirror_output_port
2114 && set_dst(dst, flow, in_port, port, tags)) {
2115 mirrors |= port->dst_mirrors;
2119 *nf_output_iface = NF_OUT_FLOOD;
2120 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2121 *nf_output_iface = dst->dp_ifidx;
2122 mirrors |= out_port->dst_mirrors;
2127 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2128 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2130 if (set_dst(dst, flow, in_port, m->out_port, tags)
2131 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2135 for (i = 0; i < br->n_ports; i++) {
2136 struct port *port = br->ports[i];
2137 if (port_includes_vlan(port, m->out_vlan)
2138 && set_dst(dst, flow, in_port, port, tags))
2142 if (port->vlan < 0) {
2143 dst->vlan = m->out_vlan;
2145 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2149 /* Use the vlan tag on the original flow instead of
2150 * the one passed in the vlan parameter. This ensures
2151 * that we compare the vlan from before any implicit
2152 * tagging tags place. This is necessary because
2153 * dst->vlan is the final vlan, after removing implicit
2155 flow_vlan = ntohs(flow->dl_vlan);
2156 if (flow_vlan == 0) {
2157 flow_vlan = OFP_VLAN_NONE;
2159 if (port == in_port && dst->vlan == flow_vlan) {
2160 /* Don't send out input port on same VLAN. */
2168 mirrors &= mirrors - 1;
2171 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2175 static void OVS_UNUSED
2176 print_dsts(const struct dst *dsts, size_t n)
2178 for (; n--; dsts++) {
2179 printf(">p%"PRIu16, dsts->dp_ifidx);
2180 if (dsts->vlan != OFP_VLAN_NONE) {
2181 printf("v%"PRIu16, dsts->vlan);
2187 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2188 const struct port *in_port, const struct port *out_port,
2189 tag_type *tags, struct odp_actions *actions,
2190 uint16_t *nf_output_iface)
2192 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2194 const struct dst *p;
2197 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2200 cur_vlan = ntohs(flow->dl_vlan);
2201 for (p = dsts; p < &dsts[n_dsts]; p++) {
2202 union odp_action *a;
2203 if (p->vlan != cur_vlan) {
2204 if (p->vlan == OFP_VLAN_NONE) {
2205 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2207 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2208 a->vlan_vid.vlan_vid = htons(p->vlan);
2212 a = odp_actions_add(actions, ODPAT_OUTPUT);
2213 a->output.port = p->dp_ifidx;
2217 /* Returns the effective vlan of a packet, taking into account both the
2218 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2219 * the packet is untagged and -1 indicates it has an invalid header and
2220 * should be dropped. */
2221 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2222 struct port *in_port, bool have_packet)
2224 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2225 * belongs to VLAN 0, so we should treat both cases identically. (In the
2226 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2227 * presumably to allow a priority to be specified. In the latter case, the
2228 * packet does not have any 802.1Q header.) */
2229 int vlan = ntohs(flow->dl_vlan);
2230 if (vlan == OFP_VLAN_NONE) {
2233 if (in_port->vlan >= 0) {
2235 /* XXX support double tagging? */
2237 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2238 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2239 "packet received on port %s configured with "
2240 "implicit VLAN %"PRIu16,
2241 br->name, ntohs(flow->dl_vlan),
2242 in_port->name, in_port->vlan);
2246 vlan = in_port->vlan;
2248 if (!port_includes_vlan(in_port, vlan)) {
2250 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2251 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2252 "packet received on port %s not configured for "
2254 br->name, vlan, in_port->name, vlan);
2263 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2264 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2265 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2267 is_gratuitous_arp(const flow_t *flow)
2269 return (flow->dl_type == htons(ETH_TYPE_ARP)
2270 && eth_addr_is_broadcast(flow->dl_dst)
2271 && (flow->nw_proto == ARP_OP_REPLY
2272 || (flow->nw_proto == ARP_OP_REQUEST
2273 && flow->nw_src == flow->nw_dst)));
2277 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2278 struct port *in_port)
2280 enum grat_arp_lock_type lock_type;
2283 /* We don't want to learn from gratuitous ARP packets that are reflected
2284 * back over bond slaves so we lock the learning table. */
2285 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2286 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2287 GRAT_ARP_LOCK_CHECK;
2289 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2292 /* The log messages here could actually be useful in debugging,
2293 * so keep the rate limit relatively high. */
2294 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2296 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2297 "on port %s in VLAN %d",
2298 br->name, ETH_ADDR_ARGS(flow->dl_src),
2299 in_port->name, vlan);
2300 ofproto_revalidate(br->ofproto, rev_tag);
2304 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2305 * dropped. Returns true if they may be forwarded, false if they should be
2308 * If 'have_packet' is true, it indicates that the caller is processing a
2309 * received packet. If 'have_packet' is false, then the caller is just
2310 * revalidating an existing flow because configuration has changed. Either
2311 * way, 'have_packet' only affects logging (there is no point in logging errors
2312 * during revalidation).
2314 * Sets '*in_portp' to the input port. This will be a null pointer if
2315 * flow->in_port does not designate a known input port (in which case
2316 * is_admissible() returns false).
2318 * When returning true, sets '*vlanp' to the effective VLAN of the input
2319 * packet, as returned by flow_get_vlan().
2321 * May also add tags to '*tags', although the current implementation only does
2322 * so in one special case.
2325 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2326 tag_type *tags, int *vlanp, struct port **in_portp)
2328 struct iface *in_iface;
2329 struct port *in_port;
2332 /* Find the interface and port structure for the received packet. */
2333 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2335 /* No interface? Something fishy... */
2337 /* Odd. A few possible reasons here:
2339 * - We deleted an interface but there are still a few packets
2340 * queued up from it.
2342 * - Someone externally added an interface (e.g. with "ovs-dpctl
2343 * add-if") that we don't know about.
2345 * - Packet arrived on the local port but the local port is not
2346 * one of our bridge ports.
2348 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2350 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2351 "interface %"PRIu16, br->name, flow->in_port);
2357 *in_portp = in_port = in_iface->port;
2358 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2363 /* Drop frames for reserved multicast addresses. */
2364 if (eth_addr_is_reserved(flow->dl_dst)) {
2368 /* Drop frames on ports reserved for mirroring. */
2369 if (in_port->is_mirror_output_port) {
2371 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2372 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2373 "%s, which is reserved exclusively for mirroring",
2374 br->name, in_port->name);
2379 /* Packets received on bonds need special attention to avoid duplicates. */
2380 if (in_port->n_ifaces > 1) {
2382 bool is_grat_arp_locked;
2384 if (eth_addr_is_multicast(flow->dl_dst)) {
2385 *tags |= in_port->active_iface_tag;
2386 if (in_port->active_iface != in_iface->port_ifidx) {
2387 /* Drop all multicast packets on inactive slaves. */
2392 /* Drop all packets for which we have learned a different input
2393 * port, because we probably sent the packet on one slave and got
2394 * it back on the other. Gratuitous ARP packets are an exception
2395 * to this rule: the host has moved to another switch. The exception
2396 * to the exception is if we locked the learning table to avoid
2397 * reflections on bond slaves. If this is the case, just drop the
2399 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2400 &is_grat_arp_locked);
2401 if (src_idx != -1 && src_idx != in_port->port_idx &&
2402 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2410 /* If the composed actions may be applied to any packet in the given 'flow',
2411 * returns true. Otherwise, the actions should only be applied to 'packet', or
2412 * not at all, if 'packet' was NULL. */
2414 process_flow(struct bridge *br, const flow_t *flow,
2415 const struct ofpbuf *packet, struct odp_actions *actions,
2416 tag_type *tags, uint16_t *nf_output_iface)
2418 struct port *in_port;
2419 struct port *out_port;
2423 /* Check whether we should drop packets in this flow. */
2424 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2429 /* Learn source MAC (but don't try to learn from revalidation). */
2431 update_learning_table(br, flow, vlan, in_port);
2434 /* Determine output port. */
2435 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2437 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2438 out_port = br->ports[out_port_idx];
2439 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2440 /* If we are revalidating but don't have a learning entry then
2441 * eject the flow. Installing a flow that floods packets opens
2442 * up a window of time where we could learn from a packet reflected
2443 * on a bond and blackhole packets before the learning table is
2444 * updated to reflect the correct port. */
2447 out_port = FLOOD_PORT;
2450 /* Don't send packets out their input ports. */
2451 if (in_port == out_port) {
2457 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2464 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2467 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2468 const struct ofp_phy_port *opp,
2471 struct bridge *br = br_;
2472 struct iface *iface;
2475 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2481 if (reason == OFPPR_DELETE) {
2482 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2483 br->name, iface->name);
2484 iface_destroy(iface);
2485 if (!port->n_ifaces) {
2486 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2487 br->name, port->name);
2493 if (port->n_ifaces > 1) {
2494 bool up = !(opp->state & OFPPS_LINK_DOWN);
2495 bond_link_status_update(iface, up);
2496 port_update_bond_compat(port);
2502 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2503 struct odp_actions *actions, tag_type *tags,
2504 uint16_t *nf_output_iface, void *br_)
2506 struct bridge *br = br_;
2508 COVERAGE_INC(bridge_process_flow);
2510 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2514 bridge_account_flow_ofhook_cb(const flow_t *flow, tag_type tags,
2515 const union odp_action *actions,
2516 size_t n_actions, unsigned long long int n_bytes,
2519 struct bridge *br = br_;
2520 const union odp_action *a;
2521 struct port *in_port;
2525 /* Feed information from the active flows back into the learning table to
2526 * ensure that table is always in sync with what is actually flowing
2527 * through the datapath.
2529 * We test that 'tags' is nonzero to ensure that only flows that include an
2530 * OFPP_NORMAL action are used for learning. This works because
2531 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2532 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2533 update_learning_table(br, flow, vlan, in_port);
2536 /* Account for bond slave utilization. */
2537 if (!br->has_bonded_ports) {
2540 for (a = actions; a < &actions[n_actions]; a++) {
2541 if (a->type == ODPAT_OUTPUT) {
2542 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2543 if (out_port && out_port->n_ifaces >= 2) {
2544 struct bond_entry *e = lookup_bond_entry(out_port,
2546 e->tx_bytes += n_bytes;
2553 bridge_account_checkpoint_ofhook_cb(void *br_)
2555 struct bridge *br = br_;
2559 if (!br->has_bonded_ports) {
2564 for (i = 0; i < br->n_ports; i++) {
2565 struct port *port = br->ports[i];
2566 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2567 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2568 bond_rebalance_port(port);
2573 static struct ofhooks bridge_ofhooks = {
2574 bridge_port_changed_ofhook_cb,
2575 bridge_normal_ofhook_cb,
2576 bridge_account_flow_ofhook_cb,
2577 bridge_account_checkpoint_ofhook_cb,
2580 /* Bonding functions. */
2582 /* Statistics for a single interface on a bonded port, used for load-based
2583 * bond rebalancing. */
2584 struct slave_balance {
2585 struct iface *iface; /* The interface. */
2586 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2588 /* All the "bond_entry"s that are assigned to this interface, in order of
2589 * increasing tx_bytes. */
2590 struct bond_entry **hashes;
2594 /* Sorts pointers to pointers to bond_entries in ascending order by the
2595 * interface to which they are assigned, and within a single interface in
2596 * ascending order of bytes transmitted. */
2598 compare_bond_entries(const void *a_, const void *b_)
2600 const struct bond_entry *const *ap = a_;
2601 const struct bond_entry *const *bp = b_;
2602 const struct bond_entry *a = *ap;
2603 const struct bond_entry *b = *bp;
2604 if (a->iface_idx != b->iface_idx) {
2605 return a->iface_idx > b->iface_idx ? 1 : -1;
2606 } else if (a->tx_bytes != b->tx_bytes) {
2607 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2613 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2614 * *descending* order by number of bytes transmitted. */
2616 compare_slave_balance(const void *a_, const void *b_)
2618 const struct slave_balance *a = a_;
2619 const struct slave_balance *b = b_;
2620 if (a->iface->enabled != b->iface->enabled) {
2621 return a->iface->enabled ? -1 : 1;
2622 } else if (a->tx_bytes != b->tx_bytes) {
2623 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2630 swap_bals(struct slave_balance *a, struct slave_balance *b)
2632 struct slave_balance tmp = *a;
2637 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2638 * given that 'p' (and only 'p') might be in the wrong location.
2640 * This function invalidates 'p', since it might now be in a different memory
2643 resort_bals(struct slave_balance *p,
2644 struct slave_balance bals[], size_t n_bals)
2647 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2648 swap_bals(p, p - 1);
2650 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2651 swap_bals(p, p + 1);
2657 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2659 if (VLOG_IS_DBG_ENABLED()) {
2660 struct ds ds = DS_EMPTY_INITIALIZER;
2661 const struct slave_balance *b;
2663 for (b = bals; b < bals + n_bals; b++) {
2667 ds_put_char(&ds, ',');
2669 ds_put_format(&ds, " %s %"PRIu64"kB",
2670 b->iface->name, b->tx_bytes / 1024);
2672 if (!b->iface->enabled) {
2673 ds_put_cstr(&ds, " (disabled)");
2675 if (b->n_hashes > 0) {
2676 ds_put_cstr(&ds, " (");
2677 for (i = 0; i < b->n_hashes; i++) {
2678 const struct bond_entry *e = b->hashes[i];
2680 ds_put_cstr(&ds, " + ");
2682 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2683 e - port->bond_hash, e->tx_bytes / 1024);
2685 ds_put_cstr(&ds, ")");
2688 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2693 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2695 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2698 struct bond_entry *hash = from->hashes[hash_idx];
2699 struct port *port = from->iface->port;
2700 uint64_t delta = hash->tx_bytes;
2702 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2703 "from %s to %s (now carrying %"PRIu64"kB and "
2704 "%"PRIu64"kB load, respectively)",
2705 port->name, delta / 1024, hash - port->bond_hash,
2706 from->iface->name, to->iface->name,
2707 (from->tx_bytes - delta) / 1024,
2708 (to->tx_bytes + delta) / 1024);
2710 /* Delete element from from->hashes.
2712 * We don't bother to add the element to to->hashes because not only would
2713 * it require more work, the only purpose it would be to allow that hash to
2714 * be migrated to another slave in this rebalancing run, and there is no
2715 * point in doing that. */
2716 if (hash_idx == 0) {
2719 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2720 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2724 /* Shift load away from 'from' to 'to'. */
2725 from->tx_bytes -= delta;
2726 to->tx_bytes += delta;
2728 /* Arrange for flows to be revalidated. */
2729 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2730 hash->iface_idx = to->iface->port_ifidx;
2731 hash->iface_tag = tag_create_random();
2735 bond_rebalance_port(struct port *port)
2737 struct slave_balance bals[DP_MAX_PORTS];
2739 struct bond_entry *hashes[BOND_MASK + 1];
2740 struct slave_balance *b, *from, *to;
2741 struct bond_entry *e;
2744 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2745 * descending order of tx_bytes, so that bals[0] represents the most
2746 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2749 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2750 * array for each slave_balance structure, we sort our local array of
2751 * hashes in order by slave, so that all of the hashes for a given slave
2752 * become contiguous in memory, and then we point each 'hashes' members of
2753 * a slave_balance structure to the start of a contiguous group. */
2754 n_bals = port->n_ifaces;
2755 for (b = bals; b < &bals[n_bals]; b++) {
2756 b->iface = port->ifaces[b - bals];
2761 for (i = 0; i <= BOND_MASK; i++) {
2762 hashes[i] = &port->bond_hash[i];
2764 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2765 for (i = 0; i <= BOND_MASK; i++) {
2767 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2768 b = &bals[e->iface_idx];
2769 b->tx_bytes += e->tx_bytes;
2771 b->hashes = &hashes[i];
2776 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2777 log_bals(bals, n_bals, port);
2779 /* Discard slaves that aren't enabled (which were sorted to the back of the
2780 * array earlier). */
2781 while (!bals[n_bals - 1].iface->enabled) {
2788 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2789 to = &bals[n_bals - 1];
2790 for (from = bals; from < to; ) {
2791 uint64_t overload = from->tx_bytes - to->tx_bytes;
2792 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2793 /* The extra load on 'from' (and all less-loaded slaves), compared
2794 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2795 * it is less than ~1Mbps. No point in rebalancing. */
2797 } else if (from->n_hashes == 1) {
2798 /* 'from' only carries a single MAC hash, so we can't shift any
2799 * load away from it, even though we want to. */
2802 /* 'from' is carrying significantly more load than 'to', and that
2803 * load is split across at least two different hashes. Pick a hash
2804 * to migrate to 'to' (the least-loaded slave), given that doing so
2805 * must decrease the ratio of the load on the two slaves by at
2808 * The sort order we use means that we prefer to shift away the
2809 * smallest hashes instead of the biggest ones. There is little
2810 * reason behind this decision; we could use the opposite sort
2811 * order to shift away big hashes ahead of small ones. */
2815 for (i = 0; i < from->n_hashes; i++) {
2816 double old_ratio, new_ratio;
2817 uint64_t delta = from->hashes[i]->tx_bytes;
2819 if (delta == 0 || from->tx_bytes - delta == 0) {
2820 /* Pointless move. */
2824 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2826 if (to->tx_bytes == 0) {
2827 /* Nothing on the new slave, move it. */
2831 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2832 new_ratio = (double)(from->tx_bytes - delta) /
2833 (to->tx_bytes + delta);
2835 if (new_ratio == 0) {
2836 /* Should already be covered but check to prevent division
2841 if (new_ratio < 1) {
2842 new_ratio = 1 / new_ratio;
2845 if (old_ratio - new_ratio > 0.1) {
2846 /* Would decrease the ratio, move it. */
2850 if (i < from->n_hashes) {
2851 bond_shift_load(from, to, i);
2852 port->bond_compat_is_stale = true;
2854 /* If the result of the migration changed the relative order of
2855 * 'from' and 'to' swap them back to maintain invariants. */
2856 if (order_swapped) {
2857 swap_bals(from, to);
2860 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2861 * point to different slave_balance structures. It is only
2862 * valid to do these two operations in a row at all because we
2863 * know that 'from' will not move past 'to' and vice versa. */
2864 resort_bals(from, bals, n_bals);
2865 resort_bals(to, bals, n_bals);
2872 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2873 * historical data to decay to <1% in 7 rebalancing runs. */
2874 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2880 bond_send_learning_packets(struct port *port)
2882 struct bridge *br = port->bridge;
2883 struct mac_entry *e;
2884 struct ofpbuf packet;
2885 int error, n_packets, n_errors;
2887 if (!port->n_ifaces || port->active_iface < 0) {
2891 ofpbuf_init(&packet, 128);
2892 error = n_packets = n_errors = 0;
2893 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2894 union ofp_action actions[2], *a;
2900 if (e->port == port->port_idx
2901 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2905 /* Compose actions. */
2906 memset(actions, 0, sizeof actions);
2909 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2910 a->vlan_vid.len = htons(sizeof *a);
2911 a->vlan_vid.vlan_vid = htons(e->vlan);
2914 a->output.type = htons(OFPAT_OUTPUT);
2915 a->output.len = htons(sizeof *a);
2916 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2921 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2923 flow_extract(&packet, 0, ODPP_NONE, &flow);
2924 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2931 ofpbuf_uninit(&packet);
2934 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2935 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2936 "packets, last error was: %s",
2937 port->name, n_errors, n_packets, strerror(error));
2939 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2940 port->name, n_packets);
2944 /* Bonding unixctl user interface functions. */
2947 bond_unixctl_list(struct unixctl_conn *conn,
2948 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2950 struct ds ds = DS_EMPTY_INITIALIZER;
2951 const struct bridge *br;
2953 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2955 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2958 for (i = 0; i < br->n_ports; i++) {
2959 const struct port *port = br->ports[i];
2960 if (port->n_ifaces > 1) {
2963 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2964 for (j = 0; j < port->n_ifaces; j++) {
2965 const struct iface *iface = port->ifaces[j];
2967 ds_put_cstr(&ds, ", ");
2969 ds_put_cstr(&ds, iface->name);
2971 ds_put_char(&ds, '\n');
2975 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2979 static struct port *
2980 bond_find(const char *name)
2982 const struct bridge *br;
2984 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2987 for (i = 0; i < br->n_ports; i++) {
2988 struct port *port = br->ports[i];
2989 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2998 bond_unixctl_show(struct unixctl_conn *conn,
2999 const char *args, void *aux OVS_UNUSED)
3001 struct ds ds = DS_EMPTY_INITIALIZER;
3002 const struct port *port;
3005 port = bond_find(args);
3007 unixctl_command_reply(conn, 501, "no such bond");
3011 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3012 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3013 ds_put_format(&ds, "next rebalance: %lld ms\n",
3014 port->bond_next_rebalance - time_msec());
3015 for (j = 0; j < port->n_ifaces; j++) {
3016 const struct iface *iface = port->ifaces[j];
3017 struct bond_entry *be;
3020 ds_put_format(&ds, "slave %s: %s\n",
3021 iface->name, iface->enabled ? "enabled" : "disabled");
3022 if (j == port->active_iface) {
3023 ds_put_cstr(&ds, "\tactive slave\n");
3025 if (iface->delay_expires != LLONG_MAX) {
3026 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3027 iface->enabled ? "downdelay" : "updelay",
3028 iface->delay_expires - time_msec());
3032 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3033 int hash = be - port->bond_hash;
3034 struct mac_entry *me;
3036 if (be->iface_idx != j) {
3040 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3041 hash, be->tx_bytes / 1024);
3044 LIST_FOR_EACH (me, struct mac_entry, lru_node,
3045 &port->bridge->ml->lrus) {
3048 if (bond_hash(me->mac) == hash
3049 && me->port != port->port_idx
3050 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3051 && dp_ifidx == iface->dp_ifidx)
3053 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3054 ETH_ADDR_ARGS(me->mac));
3059 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3064 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3065 void *aux OVS_UNUSED)
3067 char *args = (char *) args_;
3068 char *save_ptr = NULL;
3069 char *bond_s, *hash_s, *slave_s;
3070 uint8_t mac[ETH_ADDR_LEN];
3072 struct iface *iface;
3073 struct bond_entry *entry;
3076 bond_s = strtok_r(args, " ", &save_ptr);
3077 hash_s = strtok_r(NULL, " ", &save_ptr);
3078 slave_s = strtok_r(NULL, " ", &save_ptr);
3080 unixctl_command_reply(conn, 501,
3081 "usage: bond/migrate BOND HASH SLAVE");
3085 port = bond_find(bond_s);
3087 unixctl_command_reply(conn, 501, "no such bond");
3091 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3092 == ETH_ADDR_SCAN_COUNT) {
3093 hash = bond_hash(mac);
3094 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3095 hash = atoi(hash_s) & BOND_MASK;
3097 unixctl_command_reply(conn, 501, "bad hash");
3101 iface = port_lookup_iface(port, slave_s);
3103 unixctl_command_reply(conn, 501, "no such slave");
3107 if (!iface->enabled) {
3108 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3112 entry = &port->bond_hash[hash];
3113 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3114 entry->iface_idx = iface->port_ifidx;
3115 entry->iface_tag = tag_create_random();
3116 port->bond_compat_is_stale = true;
3117 unixctl_command_reply(conn, 200, "migrated");
3121 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3122 void *aux OVS_UNUSED)
3124 char *args = (char *) args_;
3125 char *save_ptr = NULL;
3126 char *bond_s, *slave_s;
3128 struct iface *iface;
3130 bond_s = strtok_r(args, " ", &save_ptr);
3131 slave_s = strtok_r(NULL, " ", &save_ptr);
3133 unixctl_command_reply(conn, 501,
3134 "usage: bond/set-active-slave BOND SLAVE");
3138 port = bond_find(bond_s);
3140 unixctl_command_reply(conn, 501, "no such bond");
3144 iface = port_lookup_iface(port, slave_s);
3146 unixctl_command_reply(conn, 501, "no such slave");
3150 if (!iface->enabled) {
3151 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3155 if (port->active_iface != iface->port_ifidx) {
3156 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3157 port->active_iface = iface->port_ifidx;
3158 port->active_iface_tag = tag_create_random();
3159 VLOG_INFO("port %s: active interface is now %s",
3160 port->name, iface->name);
3161 bond_send_learning_packets(port);
3162 unixctl_command_reply(conn, 200, "done");
3164 unixctl_command_reply(conn, 200, "no change");
3169 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3171 char *args = (char *) args_;
3172 char *save_ptr = NULL;
3173 char *bond_s, *slave_s;
3175 struct iface *iface;
3177 bond_s = strtok_r(args, " ", &save_ptr);
3178 slave_s = strtok_r(NULL, " ", &save_ptr);
3180 unixctl_command_reply(conn, 501,
3181 "usage: bond/enable/disable-slave BOND SLAVE");
3185 port = bond_find(bond_s);
3187 unixctl_command_reply(conn, 501, "no such bond");
3191 iface = port_lookup_iface(port, slave_s);
3193 unixctl_command_reply(conn, 501, "no such slave");
3197 bond_enable_slave(iface, enable);
3198 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3202 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3203 void *aux OVS_UNUSED)
3205 enable_slave(conn, args, true);
3209 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3210 void *aux OVS_UNUSED)
3212 enable_slave(conn, args, false);
3216 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3217 void *aux OVS_UNUSED)
3219 uint8_t mac[ETH_ADDR_LEN];
3223 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3224 == ETH_ADDR_SCAN_COUNT) {
3225 hash = bond_hash(mac);
3227 hash_cstr = xasprintf("%u", hash);
3228 unixctl_command_reply(conn, 200, hash_cstr);
3231 unixctl_command_reply(conn, 501, "invalid mac");
3238 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3239 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3240 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3241 unixctl_command_register("bond/set-active-slave",
3242 bond_unixctl_set_active_slave, NULL);
3243 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3245 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3247 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3250 /* Port functions. */
3252 static struct port *
3253 port_create(struct bridge *br, const char *name)
3257 port = xzalloc(sizeof *port);
3259 port->port_idx = br->n_ports;
3261 port->trunks = NULL;
3262 port->name = xstrdup(name);
3263 port->active_iface = -1;
3265 if (br->n_ports >= br->allocated_ports) {
3266 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3269 br->ports[br->n_ports++] = port;
3270 shash_add_assert(&br->port_by_name, port->name, port);
3272 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3279 get_port_other_config(const struct ovsrec_port *port, const char *key,
3280 const char *default_value)
3284 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3286 return value ? value : default_value;
3290 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3292 struct shash new_ifaces;
3295 /* Collect list of new interfaces. */
3296 shash_init(&new_ifaces);
3297 for (i = 0; i < cfg->n_interfaces; i++) {
3298 const char *name = cfg->interfaces[i]->name;
3299 shash_add_once(&new_ifaces, name, NULL);
3302 /* Get rid of deleted interfaces. */
3303 for (i = 0; i < port->n_ifaces; ) {
3304 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3305 iface_destroy(port->ifaces[i]);
3311 shash_destroy(&new_ifaces);
3315 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3317 struct shash new_ifaces;
3318 long long int next_rebalance;
3319 unsigned long *trunks;
3325 /* Update settings. */
3326 port->updelay = cfg->bond_updelay;
3327 if (port->updelay < 0) {
3330 port->updelay = cfg->bond_downdelay;
3331 if (port->downdelay < 0) {
3332 port->downdelay = 0;
3334 port->bond_rebalance_interval = atoi(
3335 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3336 if (port->bond_rebalance_interval < 1000) {
3337 port->bond_rebalance_interval = 1000;
3339 next_rebalance = time_msec() + port->bond_rebalance_interval;
3340 if (port->bond_next_rebalance > next_rebalance) {
3341 port->bond_next_rebalance = next_rebalance;
3344 /* Add new interfaces and update 'cfg' member of existing ones. */
3345 shash_init(&new_ifaces);
3346 for (i = 0; i < cfg->n_interfaces; i++) {
3347 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3348 struct iface *iface;
3350 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3351 VLOG_WARN("port %s: %s specified twice as port interface",
3352 port->name, if_cfg->name);
3356 iface = iface_lookup(port->bridge, if_cfg->name);
3358 if (iface->port != port) {
3359 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3361 port->bridge->name, if_cfg->name, iface->port->name);
3364 iface->cfg = if_cfg;
3366 iface_create(port, if_cfg);
3369 shash_destroy(&new_ifaces);
3374 if (port->n_ifaces < 2) {
3376 if (vlan >= 0 && vlan <= 4095) {
3377 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3382 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3383 * they even work as-is. But they have not been tested. */
3384 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3388 if (port->vlan != vlan) {
3390 bridge_flush(port->bridge);
3393 /* Get trunked VLANs. */
3395 if (vlan < 0 && cfg->n_trunks) {
3399 trunks = bitmap_allocate(4096);
3401 for (i = 0; i < cfg->n_trunks; i++) {
3402 int trunk = cfg->trunks[i];
3404 bitmap_set1(trunks, trunk);
3410 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3411 port->name, cfg->n_trunks);
3413 if (n_errors == cfg->n_trunks) {
3414 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3416 bitmap_free(trunks);
3419 } else if (vlan >= 0 && cfg->n_trunks) {
3420 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3424 ? port->trunks != NULL
3425 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3426 bridge_flush(port->bridge);
3428 bitmap_free(port->trunks);
3429 port->trunks = trunks;
3433 port_destroy(struct port *port)
3436 struct bridge *br = port->bridge;
3440 proc_net_compat_update_vlan(port->name, NULL, 0);
3441 proc_net_compat_update_bond(port->name, NULL);
3443 for (i = 0; i < MAX_MIRRORS; i++) {
3444 struct mirror *m = br->mirrors[i];
3445 if (m && m->out_port == port) {
3450 while (port->n_ifaces > 0) {
3451 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3454 shash_find_and_delete_assert(&br->port_by_name, port->name);
3456 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3457 del->port_idx = port->port_idx;
3460 bitmap_free(port->trunks);
3467 static struct port *
3468 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3470 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3471 return iface ? iface->port : NULL;
3474 static struct port *
3475 port_lookup(const struct bridge *br, const char *name)
3477 return shash_find_data(&br->port_by_name, name);
3480 static struct iface *
3481 port_lookup_iface(const struct port *port, const char *name)
3483 struct iface *iface = iface_lookup(port->bridge, name);
3484 return iface && iface->port == port ? iface : NULL;
3488 port_update_bonding(struct port *port)
3490 if (port->n_ifaces < 2) {
3491 /* Not a bonded port. */
3492 if (port->bond_hash) {
3493 free(port->bond_hash);
3494 port->bond_hash = NULL;
3495 port->bond_compat_is_stale = true;
3496 port->bond_fake_iface = false;
3499 if (!port->bond_hash) {
3502 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3503 for (i = 0; i <= BOND_MASK; i++) {
3504 struct bond_entry *e = &port->bond_hash[i];
3508 port->no_ifaces_tag = tag_create_random();
3509 bond_choose_active_iface(port);
3510 port->bond_next_rebalance
3511 = time_msec() + port->bond_rebalance_interval;
3513 if (port->cfg->bond_fake_iface) {
3514 port->bond_next_fake_iface_update = time_msec();
3517 port->bond_compat_is_stale = true;
3518 port->bond_fake_iface = port->cfg->bond_fake_iface;
3523 port_update_bond_compat(struct port *port)
3525 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3526 struct compat_bond bond;
3529 if (port->n_ifaces < 2) {
3530 proc_net_compat_update_bond(port->name, NULL);
3535 bond.updelay = port->updelay;
3536 bond.downdelay = port->downdelay;
3539 bond.hashes = compat_hashes;
3540 if (port->bond_hash) {
3541 const struct bond_entry *e;
3542 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3543 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3544 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3545 cbh->hash = e - port->bond_hash;
3546 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3551 bond.n_slaves = port->n_ifaces;
3552 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3553 for (i = 0; i < port->n_ifaces; i++) {
3554 struct iface *iface = port->ifaces[i];
3555 struct compat_bond_slave *slave = &bond.slaves[i];
3556 slave->name = iface->name;
3558 /* We need to make the same determination as the Linux bonding
3559 * code to determine whether a slave should be consider "up".
3560 * The Linux function bond_miimon_inspect() supports four
3561 * BOND_LINK_* states:
3563 * - BOND_LINK_UP: carrier detected, updelay has passed.
3564 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3565 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3566 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3568 * The function bond_info_show_slave() only considers BOND_LINK_UP
3569 * to be "up" and anything else to be "down".
3571 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3575 netdev_get_etheraddr(iface->netdev, slave->mac);
3578 if (port->bond_fake_iface) {
3579 struct netdev *bond_netdev;
3581 if (!netdev_open_default(port->name, &bond_netdev)) {
3583 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3585 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3587 netdev_close(bond_netdev);
3591 proc_net_compat_update_bond(port->name, &bond);
3596 port_update_vlan_compat(struct port *port)
3598 struct bridge *br = port->bridge;
3599 char *vlandev_name = NULL;
3601 if (port->vlan > 0) {
3602 /* Figure out the name that the VLAN device should actually have, if it
3603 * existed. This takes some work because the VLAN device would not
3604 * have port->name in its name; rather, it would have the trunk port's
3605 * name, and 'port' would be attached to a bridge that also had the
3606 * VLAN device one of its ports. So we need to find a trunk port that
3607 * includes port->vlan.
3609 * There might be more than one candidate. This doesn't happen on
3610 * XenServer, so if it happens we just pick the first choice in
3611 * alphabetical order instead of creating multiple VLAN devices. */
3613 for (i = 0; i < br->n_ports; i++) {
3614 struct port *p = br->ports[i];
3615 if (port_trunks_vlan(p, port->vlan)
3617 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3619 uint8_t ea[ETH_ADDR_LEN];
3620 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3621 if (!eth_addr_is_multicast(ea) &&
3622 !eth_addr_is_reserved(ea) &&
3623 !eth_addr_is_zero(ea)) {
3624 vlandev_name = p->name;
3629 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3632 /* Interface functions. */
3634 static struct iface *
3635 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3637 struct bridge *br = port->bridge;
3638 struct iface *iface;
3639 char *name = if_cfg->name;
3642 iface = xzalloc(sizeof *iface);
3644 iface->port_ifidx = port->n_ifaces;
3645 iface->name = xstrdup(name);
3646 iface->dp_ifidx = -1;
3647 iface->tag = tag_create_random();
3648 iface->delay_expires = LLONG_MAX;
3649 iface->netdev = NULL;
3650 iface->cfg = if_cfg;
3652 shash_add_assert(&br->iface_by_name, iface->name, iface);
3654 /* Attempt to create the network interface in case it doesn't exist yet. */
3655 if (!iface_is_internal(br, iface->name)) {
3656 error = set_up_iface(if_cfg, iface, true);
3658 VLOG_WARN("could not create iface %s: %s", iface->name,
3661 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3668 if (port->n_ifaces >= port->allocated_ifaces) {
3669 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3670 sizeof *port->ifaces);
3672 port->ifaces[port->n_ifaces++] = iface;
3673 if (port->n_ifaces > 1) {
3674 br->has_bonded_ports = true;
3677 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3685 iface_destroy(struct iface *iface)
3688 struct port *port = iface->port;
3689 struct bridge *br = port->bridge;
3690 bool del_active = port->active_iface == iface->port_ifidx;
3693 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3695 if (iface->dp_ifidx >= 0) {
3696 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3699 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3700 del->port_ifidx = iface->port_ifidx;
3702 netdev_close(iface->netdev);
3705 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3706 bond_choose_active_iface(port);
3707 bond_send_learning_packets(port);
3713 bridge_flush(port->bridge);
3717 static struct iface *
3718 iface_lookup(const struct bridge *br, const char *name)
3720 return shash_find_data(&br->iface_by_name, name);
3723 static struct iface *
3724 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3726 return port_array_get(&br->ifaces, dp_ifidx);
3729 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3730 * 'br', that is, an interface that is entirely simulated within the datapath.
3731 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3732 * interfaces are created by setting "iface.<iface>.internal = true".
3734 * In addition, we have a kluge-y feature that creates an internal port with
3735 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3736 * This feature needs to go away in the long term. Until then, this is one
3737 * reason why this function takes a name instead of a struct iface: the fake
3738 * interfaces created this way do not have a struct iface. */
3740 iface_is_internal(const struct bridge *br, const char *if_name)
3742 struct iface *iface;
3745 if (!strcmp(if_name, br->name)) {
3749 iface = iface_lookup(br, if_name);
3750 if (iface && !strcmp(iface->cfg->type, "internal")) {
3754 port = port_lookup(br, if_name);
3755 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3761 /* Set Ethernet address of 'iface', if one is specified in the configuration
3764 iface_set_mac(struct iface *iface)
3766 uint8_t ea[ETH_ADDR_LEN];
3768 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3769 if (eth_addr_is_multicast(ea)) {
3770 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3772 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3773 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3774 iface->name, iface->name);
3776 int error = netdev_set_etheraddr(iface->netdev, ea);
3778 VLOG_ERR("interface %s: setting MAC failed (%s)",
3779 iface->name, strerror(error));
3786 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3787 struct shash *shash)
3792 for (i = 0; i < n; i++) {
3793 shash_add(shash, keys[i], values[i]);
3797 struct iface_delete_queues_cbdata {
3798 struct netdev *netdev;
3799 const struct ovsdb_datum *queues;
3803 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
3805 union ovsdb_atom atom;
3807 atom.integer = target;
3808 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
3812 iface_delete_queues(unsigned int queue_id,
3813 const struct shash *details OVS_UNUSED, void *cbdata_)
3815 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3817 if (!queue_ids_include(cbdata->queues, queue_id)) {
3818 netdev_delete_queue(cbdata->netdev, queue_id);
3823 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3825 if (!qos || qos->type[0] == '\0') {
3826 netdev_set_qos(iface->netdev, NULL, NULL);
3828 struct iface_delete_queues_cbdata cbdata;
3829 struct shash details;
3832 /* Configure top-level Qos for 'iface'. */
3833 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3834 qos->n_other_config, &details);
3835 netdev_set_qos(iface->netdev, qos->type, &details);
3836 shash_destroy(&details);
3838 /* Deconfigure queues that were deleted. */
3839 cbdata.netdev = iface->netdev;
3840 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
3842 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3844 /* Configure queues for 'iface'. */
3845 for (i = 0; i < qos->n_queues; i++) {
3846 const struct ovsrec_queue *queue = qos->value_queues[i];
3847 unsigned int queue_id = qos->key_queues[i];
3849 shash_from_ovs_idl_map(queue->key_other_config,
3850 queue->value_other_config,
3851 queue->n_other_config, &details);
3852 netdev_set_queue(iface->netdev, queue_id, &details);
3853 shash_destroy(&details);
3858 /* Port mirroring. */
3860 static struct mirror *
3861 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
3865 for (i = 0; i < MAX_MIRRORS; i++) {
3866 struct mirror *m = br->mirrors[i];
3867 if (m && uuid_equals(uuid, &m->uuid)) {
3875 mirror_reconfigure(struct bridge *br)
3877 unsigned long *rspan_vlans;
3880 /* Get rid of deleted mirrors. */
3881 for (i = 0; i < MAX_MIRRORS; i++) {
3882 struct mirror *m = br->mirrors[i];
3884 const struct ovsdb_datum *mc;
3885 union ovsdb_atom atom;
3887 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
3888 atom.uuid = br->mirrors[i]->uuid;
3889 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
3895 /* Add new mirrors and reconfigure existing ones. */
3896 for (i = 0; i < br->cfg->n_mirrors; i++) {
3897 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3898 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
3900 mirror_reconfigure_one(m, cfg);
3902 mirror_create(br, cfg);
3906 /* Update port reserved status. */
3907 for (i = 0; i < br->n_ports; i++) {
3908 br->ports[i]->is_mirror_output_port = false;
3910 for (i = 0; i < MAX_MIRRORS; i++) {
3911 struct mirror *m = br->mirrors[i];
3912 if (m && m->out_port) {
3913 m->out_port->is_mirror_output_port = true;
3917 /* Update flooded vlans (for RSPAN). */
3919 if (br->cfg->n_flood_vlans) {
3920 rspan_vlans = bitmap_allocate(4096);
3922 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3923 int64_t vlan = br->cfg->flood_vlans[i];
3924 if (vlan >= 0 && vlan < 4096) {
3925 bitmap_set1(rspan_vlans, vlan);
3926 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3929 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3934 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3940 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
3945 for (i = 0; ; i++) {
3946 if (i >= MAX_MIRRORS) {
3947 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3948 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
3951 if (!br->mirrors[i]) {
3956 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
3959 br->mirrors[i] = m = xzalloc(sizeof *m);
3962 m->name = xstrdup(cfg->name);
3963 shash_init(&m->src_ports);
3964 shash_init(&m->dst_ports);
3970 mirror_reconfigure_one(m, cfg);
3974 mirror_destroy(struct mirror *m)
3977 struct bridge *br = m->bridge;
3980 for (i = 0; i < br->n_ports; i++) {
3981 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3982 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3985 shash_destroy(&m->src_ports);
3986 shash_destroy(&m->dst_ports);
3989 m->bridge->mirrors[m->idx] = NULL;
3998 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3999 struct shash *names)
4003 for (i = 0; i < n_ports; i++) {
4004 const char *name = ports[i]->name;
4005 if (port_lookup(m->bridge, name)) {
4006 shash_add_once(names, name, NULL);
4008 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4009 "port %s", m->bridge->name, m->name, name);
4015 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4021 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4023 for (i = 0; i < cfg->n_select_vlan; i++) {
4024 int64_t vlan = cfg->select_vlan[i];
4025 if (vlan < 0 || vlan > 4095) {
4026 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4027 m->bridge->name, m->name, vlan);
4029 (*vlans)[n_vlans++] = vlan;
4036 vlan_is_mirrored(const struct mirror *m, int vlan)
4040 for (i = 0; i < m->n_vlans; i++) {
4041 if (m->vlans[i] == vlan) {
4049 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4053 for (i = 0; i < m->n_vlans; i++) {
4054 if (port_trunks_vlan(p, m->vlans[i])) {
4062 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4064 struct shash src_ports, dst_ports;
4065 mirror_mask_t mirror_bit;
4066 struct port *out_port;
4073 if (strcmp(cfg->name, m->name)) {
4075 m->name = xstrdup(cfg->name);
4078 /* Get output port. */
4079 if (cfg->output_port) {
4080 out_port = port_lookup(m->bridge, cfg->output_port->name);
4082 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4083 m->bridge->name, m->name);
4089 if (cfg->output_vlan) {
4090 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4091 "output vlan; ignoring output vlan",
4092 m->bridge->name, m->name);
4094 } else if (cfg->output_vlan) {
4096 out_vlan = *cfg->output_vlan;
4098 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4099 m->bridge->name, m->name);
4104 shash_init(&src_ports);
4105 shash_init(&dst_ports);
4106 if (cfg->select_all) {
4107 for (i = 0; i < m->bridge->n_ports; i++) {
4108 const char *name = m->bridge->ports[i]->name;
4109 shash_add_once(&src_ports, name, NULL);
4110 shash_add_once(&dst_ports, name, NULL);
4115 /* Get ports, and drop duplicates and ports that don't exist. */
4116 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4118 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4121 /* Get all the vlans, and drop duplicate and invalid vlans. */
4122 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4125 /* Update mirror data. */
4126 if (!shash_equal_keys(&m->src_ports, &src_ports)
4127 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4128 || m->n_vlans != n_vlans
4129 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4130 || m->out_port != out_port
4131 || m->out_vlan != out_vlan) {
4132 bridge_flush(m->bridge);
4134 shash_swap(&m->src_ports, &src_ports);
4135 shash_swap(&m->dst_ports, &dst_ports);
4138 m->n_vlans = n_vlans;
4139 m->out_port = out_port;
4140 m->out_vlan = out_vlan;
4143 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4144 for (i = 0; i < m->bridge->n_ports; i++) {
4145 struct port *port = m->bridge->ports[i];
4147 if (shash_find(&m->src_ports, port->name)
4150 ? port_trunks_any_mirrored_vlan(m, port)
4151 : vlan_is_mirrored(m, port->vlan)))) {
4152 port->src_mirrors |= mirror_bit;
4154 port->src_mirrors &= ~mirror_bit;
4157 if (shash_find(&m->dst_ports, port->name)) {
4158 port->dst_mirrors |= mirror_bit;
4160 port->dst_mirrors &= ~mirror_bit;
4165 shash_destroy(&src_ports);
4166 shash_destroy(&dst_ports);