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.
18 #include "byte-order.h"
21 #include <arpa/inet.h>
24 #include <sys/socket.h>
26 #include <openflow/openflow.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
38 #include "dynamic-string.h"
44 #include "mac-learning.h"
47 #include "ofp-print.h"
49 #include "ofproto/netflow.h"
50 #include "ofproto/ofproto.h"
51 #include "ovsdb-data.h"
53 #include "poll-loop.h"
54 #include "proc-net-compat.h"
58 #include "socket-util.h"
59 #include "stream-ssl.h"
61 #include "system-stats.h"
66 #include "vswitchd/vswitch-idl.h"
67 #include "xenserver.h"
69 #include "sflow_api.h"
71 VLOG_DEFINE_THIS_MODULE(bridge);
79 /* These members are always valid. */
80 struct port *port; /* Containing port. */
81 size_t port_ifidx; /* Index within containing port. */
82 char *name; /* Host network device name. */
83 tag_type tag; /* Tag associated with this interface. */
84 long long delay_expires; /* Time after which 'enabled' may change. */
86 /* These members are valid only after bridge_reconfigure() causes them to
88 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
89 int dp_ifidx; /* Index within kernel datapath. */
90 struct netdev *netdev; /* Network device. */
91 bool enabled; /* May be chosen for flows? */
92 const char *type; /* Usually same as cfg->type. */
93 const struct ovsrec_interface *cfg;
96 #define BOND_MASK 0xff
98 int iface_idx; /* Index of assigned iface, or -1 if none. */
99 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
100 tag_type iface_tag; /* Tag associated with iface_idx. */
103 #define MAX_MIRRORS 32
104 typedef uint32_t mirror_mask_t;
105 #define MIRROR_MASK_C(X) UINT32_C(X)
106 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
108 struct bridge *bridge;
111 struct uuid uuid; /* UUID of this "mirror" record in database. */
113 /* Selection criteria. */
114 struct shash src_ports; /* Name is port name; data is always NULL. */
115 struct shash dst_ports; /* Name is port name; data is always NULL. */
120 struct port *out_port;
124 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
126 struct bridge *bridge;
128 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
129 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
130 * NULL if all VLANs are trunked. */
131 const struct ovsrec_port *cfg;
134 /* An ordinary bridge port has 1 interface.
135 * A bridge port for bonding has at least 2 interfaces. */
136 struct iface **ifaces;
137 size_t n_ifaces, allocated_ifaces;
140 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
141 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
142 tag_type active_iface_tag; /* Tag for bcast flows. */
143 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
144 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
145 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
146 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
147 long long int bond_next_fake_iface_update; /* Time of next update. */
148 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
149 long long int bond_next_rebalance; /* Next rebalancing time. */
150 struct netdev_monitor *monitor; /* Tracks carrier up/down status. */
152 /* Port mirroring info. */
153 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
154 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
155 bool is_mirror_output_port; /* Does port mirroring send frames here? */
158 #define DP_MAX_PORTS 255
160 struct list node; /* Node in global list of bridges. */
161 char *name; /* User-specified arbitrary name. */
162 struct mac_learning *ml; /* MAC learning table. */
163 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
164 const struct ovsrec_bridge *cfg;
166 /* OpenFlow switch processing. */
167 struct ofproto *ofproto; /* OpenFlow switch. */
169 /* Kernel datapath information. */
170 struct dpif *dpif; /* Datapath. */
171 struct hmap ifaces; /* Contains "struct iface"s. */
175 size_t n_ports, allocated_ports;
176 struct shash iface_by_name; /* "struct iface"s indexed by name. */
177 struct shash port_by_name; /* "struct port"s indexed by name. */
180 bool has_bonded_ports;
185 /* Port mirroring. */
186 struct mirror *mirrors[MAX_MIRRORS];
189 /* List of all bridges. */
190 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
192 /* OVSDB IDL used to obtain configuration. */
193 static struct ovsdb_idl *idl;
195 /* Each time this timer expires, the bridge fetches systems and interface
196 * statistics and pushes them into the database. */
197 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
198 static long long int stats_timer = LLONG_MIN;
200 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
201 static void bridge_destroy(struct bridge *);
202 static struct bridge *bridge_lookup(const char *name);
203 static unixctl_cb_func bridge_unixctl_dump_flows;
204 static unixctl_cb_func bridge_unixctl_reconnect;
205 static int bridge_run_one(struct bridge *);
206 static size_t bridge_get_controllers(const struct bridge *br,
207 struct ovsrec_controller ***controllersp);
208 static void bridge_reconfigure_one(struct bridge *);
209 static void bridge_reconfigure_remotes(struct bridge *,
210 const struct sockaddr_in *managers,
212 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
213 static void bridge_fetch_dp_ifaces(struct bridge *);
214 static void bridge_flush(struct bridge *);
215 static void bridge_pick_local_hw_addr(struct bridge *,
216 uint8_t ea[ETH_ADDR_LEN],
217 struct iface **hw_addr_iface);
218 static uint64_t bridge_pick_datapath_id(struct bridge *,
219 const uint8_t bridge_ea[ETH_ADDR_LEN],
220 struct iface *hw_addr_iface);
221 static struct iface *bridge_get_local_iface(struct bridge *);
222 static uint64_t dpid_from_hash(const void *, size_t nbytes);
224 static unixctl_cb_func bridge_unixctl_fdb_show;
226 static void bond_init(void);
227 static void bond_run(struct bridge *);
228 static void bond_wait(struct bridge *);
229 static void bond_rebalance_port(struct port *);
230 static void bond_send_learning_packets(struct port *);
231 static void bond_enable_slave(struct iface *iface, bool enable);
233 static struct port *port_create(struct bridge *, const char *name);
234 static void port_reconfigure(struct port *, const struct ovsrec_port *);
235 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
236 static void port_destroy(struct port *);
237 static struct port *port_lookup(const struct bridge *, const char *name);
238 static struct iface *port_lookup_iface(const struct port *, const char *name);
239 static struct port *port_from_dp_ifidx(const struct bridge *,
241 static void port_update_bond_compat(struct port *);
242 static void port_update_vlan_compat(struct port *);
243 static void port_update_bonding(struct port *);
245 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
246 static void mirror_destroy(struct mirror *);
247 static void mirror_reconfigure(struct bridge *);
248 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
249 static bool vlan_is_mirrored(const struct mirror *, int vlan);
251 static struct iface *iface_create(struct port *port,
252 const struct ovsrec_interface *if_cfg);
253 static void iface_destroy(struct iface *);
254 static struct iface *iface_lookup(const struct bridge *, const char *name);
255 static struct iface *iface_from_dp_ifidx(const struct bridge *,
257 static void iface_set_mac(struct iface *);
258 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
259 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
261 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
264 /* Hooks into ofproto processing. */
265 static struct ofhooks bridge_ofhooks;
267 /* Public functions. */
269 /* Initializes the bridge module, configuring it to obtain its configuration
270 * from an OVSDB server accessed over 'remote', which should be a string in a
271 * form acceptable to ovsdb_idl_create(). */
273 bridge_init(const char *remote)
275 /* Create connection to database. */
276 idl = ovsdb_idl_create(remote, &ovsrec_idl_class);
278 ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_cur_cfg);
279 ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_statistics);
280 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
282 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
284 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
285 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
287 ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_ofport);
288 ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_statistics);
289 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
291 /* Register unixctl commands. */
292 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
293 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
295 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
300 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
301 * but for which the ovs-vswitchd configuration 'cfg' is required. */
303 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
305 static bool already_configured_once;
306 struct svec bridge_names;
307 struct svec dpif_names, dpif_types;
310 /* Only do this once per ovs-vswitchd run. */
311 if (already_configured_once) {
314 already_configured_once = true;
316 stats_timer = time_msec() + STATS_INTERVAL;
318 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
319 svec_init(&bridge_names);
320 for (i = 0; i < cfg->n_bridges; i++) {
321 svec_add(&bridge_names, cfg->bridges[i]->name);
323 svec_sort(&bridge_names);
325 /* Iterate over all system dpifs and delete any of them that do not appear
327 svec_init(&dpif_names);
328 svec_init(&dpif_types);
329 dp_enumerate_types(&dpif_types);
330 for (i = 0; i < dpif_types.n; i++) {
335 dp_enumerate_names(dpif_types.names[i], &dpif_names);
337 /* For each dpif... */
338 for (j = 0; j < dpif_names.n; j++) {
339 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
341 struct svec all_names;
344 /* ...check whether any of its names is in 'bridge_names'. */
345 svec_init(&all_names);
346 dpif_get_all_names(dpif, &all_names);
347 for (k = 0; k < all_names.n; k++) {
348 if (svec_contains(&bridge_names, all_names.names[k])) {
353 /* No. Delete the dpif. */
357 svec_destroy(&all_names);
362 svec_destroy(&bridge_names);
363 svec_destroy(&dpif_names);
364 svec_destroy(&dpif_types);
367 /* Initializes 'options' and fills it with the options for 'if_cfg'. Merges
368 * keys from "options" and "other_config", preferring "options" keys over
369 * "other_config" keys.
371 * The value strings in '*options' are taken directly from if_cfg, not copied,
372 * so the caller should not modify or free them. */
374 iface_get_options(const struct ovsrec_interface *if_cfg, struct shash *options)
378 shash_from_ovs_idl_map(if_cfg->key_options, if_cfg->value_options,
379 if_cfg->n_options, options);
381 for (i = 0; i < if_cfg->n_other_config; i++) {
382 char *key = if_cfg->key_other_config[i];
383 char *value = if_cfg->value_other_config[i];
385 if (!shash_find_data(options, key)) {
386 shash_add(options, key, value);
388 VLOG_WARN("%s: ignoring \"other_config\" key %s that conflicts "
389 "with \"options\" key %s", if_cfg->name, key, key);
394 /* Returns the type of network device that 'iface' should have. (This is
395 * ordinarily the same type as the interface, but the network devices for
396 * "internal" ports have type "system".) */
398 iface_get_netdev_type(const struct iface *iface)
400 return !strcmp(iface->type, "internal") ? "system" : iface->type;
403 /* Attempt to create the network device for 'iface' through the netdev
406 create_iface_netdev(struct iface *iface)
408 struct netdev_options netdev_options;
409 struct shash options;
412 memset(&netdev_options, 0, sizeof netdev_options);
413 netdev_options.name = iface->cfg->name;
414 netdev_options.type = iface_get_netdev_type(iface);
415 netdev_options.args = &options;
416 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
418 iface_get_options(iface->cfg, &options);
420 error = netdev_open(&netdev_options, &iface->netdev);
423 iface->enabled = netdev_get_carrier(iface->netdev);
426 shash_destroy(&options);
432 reconfigure_iface_netdev(struct iface *iface)
434 const char *netdev_type, *iface_type;
435 struct shash options;
438 /* Skip reconfiguration if the device has the wrong type. This shouldn't
440 iface_type = iface_get_netdev_type(iface);
441 netdev_type = netdev_get_type(iface->netdev);
442 if (iface_type && strcmp(netdev_type, iface_type)) {
443 VLOG_WARN("%s: attempting change device type from %s to %s",
444 iface->cfg->name, netdev_type, iface_type);
448 /* Reconfigure device. */
449 iface_get_options(iface->cfg, &options);
450 error = netdev_reconfigure(iface->netdev, &options);
451 shash_destroy(&options);
456 /* Callback for iterate_and_prune_ifaces(). */
458 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
460 if (!iface->netdev) {
461 /* We already reported a related error, don't bother duplicating it. */
465 if (iface->dp_ifidx < 0) {
466 VLOG_ERR("%s interface not in %s, dropping",
467 iface->name, dpif_name(br->dpif));
471 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
472 iface->name, iface->dp_ifidx);
476 /* Callback for iterate_and_prune_ifaces(). */
478 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
479 void *aux OVS_UNUSED)
481 /* Set policing attributes. */
482 netdev_set_policing(iface->netdev,
483 iface->cfg->ingress_policing_rate,
484 iface->cfg->ingress_policing_burst);
486 /* Set MAC address of internal interfaces other than the local
488 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
489 iface_set_mac(iface);
495 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
496 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
497 * deletes from 'br' any ports that no longer have any interfaces. */
499 iterate_and_prune_ifaces(struct bridge *br,
500 bool (*cb)(struct bridge *, struct iface *,
506 for (i = 0; i < br->n_ports; ) {
507 struct port *port = br->ports[i];
508 for (j = 0; j < port->n_ifaces; ) {
509 struct iface *iface = port->ifaces[j];
510 if (cb(br, iface, aux)) {
513 iface_set_ofport(iface->cfg, -1);
514 iface_destroy(iface);
518 if (port->n_ifaces) {
521 VLOG_ERR("%s port has no interfaces, dropping", port->name);
527 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
528 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
529 * responsible for freeing '*managersp' (with free()).
531 * You may be asking yourself "why does ovs-vswitchd care?", because
532 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
533 * should not be and in fact is not directly involved in that. But
534 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
535 * it has to tell in-band control where the managers are to enable that.
536 * (Thus, only managers connected in-band are collected.)
539 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
540 struct sockaddr_in **managersp, size_t *n_managersp)
542 struct sockaddr_in *managers = NULL;
543 size_t n_managers = 0;
544 struct shash targets;
547 /* Collect all of the potential targets, as the union of the "managers"
548 * column and the "targets" columns of the rows pointed to by
549 * "manager_options", excluding any that are out-of-band. */
550 shash_init(&targets);
551 for (i = 0; i < ovs_cfg->n_managers; i++) {
552 shash_add_once(&targets, ovs_cfg->managers[i], NULL);
554 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
555 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
557 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
558 shash_find_and_delete(&targets, m->target);
560 shash_add_once(&targets, m->target, NULL);
564 /* Now extract the targets' IP addresses. */
565 if (!shash_is_empty(&targets)) {
566 struct shash_node *node;
568 managers = xmalloc(shash_count(&targets) * sizeof *managers);
569 SHASH_FOR_EACH (node, &targets) {
570 const char *target = node->name;
571 struct sockaddr_in *sin = &managers[n_managers];
573 if ((!strncmp(target, "tcp:", 4)
574 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
575 (!strncmp(target, "ssl:", 4)
576 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
581 shash_destroy(&targets);
583 *managersp = managers;
584 *n_managersp = n_managers;
588 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
590 struct shash old_br, new_br;
591 struct shash_node *node;
592 struct bridge *br, *next;
593 struct sockaddr_in *managers;
596 int sflow_bridge_number;
598 COVERAGE_INC(bridge_reconfigure);
600 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
602 /* Collect old and new bridges. */
605 LIST_FOR_EACH (br, node, &all_bridges) {
606 shash_add(&old_br, br->name, br);
608 for (i = 0; i < ovs_cfg->n_bridges; i++) {
609 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
610 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
611 VLOG_WARN("more than one bridge named %s", br_cfg->name);
615 /* Get rid of deleted bridges and add new bridges. */
616 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
617 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
624 SHASH_FOR_EACH (node, &new_br) {
625 const char *br_name = node->name;
626 const struct ovsrec_bridge *br_cfg = node->data;
627 br = shash_find_data(&old_br, br_name);
629 /* If the bridge datapath type has changed, we need to tear it
630 * down and recreate. */
631 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
633 bridge_create(br_cfg);
636 bridge_create(br_cfg);
639 shash_destroy(&old_br);
640 shash_destroy(&new_br);
642 /* Reconfigure all bridges. */
643 LIST_FOR_EACH (br, node, &all_bridges) {
644 bridge_reconfigure_one(br);
647 /* Add and delete ports on all datapaths.
649 * The kernel will reject any attempt to add a given port to a datapath if
650 * that port already belongs to a different datapath, so we must do all
651 * port deletions before any port additions. */
652 LIST_FOR_EACH (br, node, &all_bridges) {
653 struct odp_port *dpif_ports;
655 struct shash want_ifaces;
657 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
658 bridge_get_all_ifaces(br, &want_ifaces);
659 for (i = 0; i < n_dpif_ports; i++) {
660 const struct odp_port *p = &dpif_ports[i];
661 if (!shash_find(&want_ifaces, p->devname)
662 && strcmp(p->devname, br->name)) {
663 int retval = dpif_port_del(br->dpif, p->port);
665 VLOG_ERR("failed to remove %s interface from %s: %s",
666 p->devname, dpif_name(br->dpif),
671 shash_destroy(&want_ifaces);
674 LIST_FOR_EACH (br, node, &all_bridges) {
675 struct odp_port *dpif_ports;
677 struct shash cur_ifaces, want_ifaces;
679 /* Get the set of interfaces currently in this datapath. */
680 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
681 shash_init(&cur_ifaces);
682 for (i = 0; i < n_dpif_ports; i++) {
683 const char *name = dpif_ports[i].devname;
684 shash_add_once(&cur_ifaces, name, &dpif_ports[i]);
687 /* Get the set of interfaces we want on this datapath. */
688 bridge_get_all_ifaces(br, &want_ifaces);
690 hmap_clear(&br->ifaces);
691 SHASH_FOR_EACH (node, &want_ifaces) {
692 const char *if_name = node->name;
693 struct iface *iface = node->data;
694 bool internal = !iface || !strcmp(iface->type, "internal");
695 struct odp_port *dpif_port = shash_find_data(&cur_ifaces, if_name);
698 /* If we have a port or a netdev already, and it's not the type we
699 * want, then delete the port (if any) and close the netdev (if
702 ? dpif_port && !(dpif_port->flags & ODP_PORT_INTERNAL)
704 && strcmp(iface->type, netdev_get_type(iface->netdev))))
707 error = ofproto_port_del(br->ofproto, dpif_port->port);
714 netdev_close(iface->netdev);
715 iface->netdev = NULL;
719 /* If it's not an internal port, open (possibly create) the
722 if (!iface->netdev) {
723 error = create_iface_netdev(iface);
725 VLOG_WARN("could not create iface %s: %s", iface->name,
730 reconfigure_iface_netdev(iface);
734 /* If it's not part of the datapath, add it. */
736 error = dpif_port_add(br->dpif, if_name,
737 internal ? ODP_PORT_INTERNAL : 0, NULL);
738 if (error == EFBIG) {
739 VLOG_ERR("ran out of valid port numbers on %s",
740 dpif_name(br->dpif));
743 VLOG_ERR("failed to add %s interface to %s: %s",
744 if_name, dpif_name(br->dpif), strerror(error));
749 /* If it's an internal port, open the netdev. */
751 if (iface && !iface->netdev) {
752 error = create_iface_netdev(iface);
754 VLOG_WARN("could not create iface %s: %s", iface->name,
760 assert(iface->netdev != NULL);
764 shash_destroy(&cur_ifaces);
765 shash_destroy(&want_ifaces);
767 sflow_bridge_number = 0;
768 LIST_FOR_EACH (br, node, &all_bridges) {
771 struct iface *local_iface;
772 struct iface *hw_addr_iface;
775 bridge_fetch_dp_ifaces(br);
777 iterate_and_prune_ifaces(br, check_iface, NULL);
779 /* Pick local port hardware address, datapath ID. */
780 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
781 local_iface = bridge_get_local_iface(br);
783 int error = netdev_set_etheraddr(local_iface->netdev, ea);
785 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
786 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
787 "Ethernet address: %s",
788 br->name, strerror(error));
792 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
793 ofproto_set_datapath_id(br->ofproto, dpid);
795 dpid_string = xasprintf("%016"PRIx64, dpid);
796 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
799 /* Set NetFlow configuration on this bridge. */
800 if (br->cfg->netflow) {
801 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
802 struct netflow_options opts;
804 memset(&opts, 0, sizeof opts);
806 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
807 if (nf_cfg->engine_type) {
808 opts.engine_type = *nf_cfg->engine_type;
810 if (nf_cfg->engine_id) {
811 opts.engine_id = *nf_cfg->engine_id;
814 opts.active_timeout = nf_cfg->active_timeout;
815 if (!opts.active_timeout) {
816 opts.active_timeout = -1;
817 } else if (opts.active_timeout < 0) {
818 VLOG_WARN("bridge %s: active timeout interval set to negative "
819 "value, using default instead (%d seconds)", br->name,
820 NF_ACTIVE_TIMEOUT_DEFAULT);
821 opts.active_timeout = -1;
824 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
825 if (opts.add_id_to_iface) {
826 if (opts.engine_id > 0x7f) {
827 VLOG_WARN("bridge %s: netflow port mangling may conflict "
828 "with another vswitch, choose an engine id less "
829 "than 128", br->name);
831 if (br->n_ports > 508) {
832 VLOG_WARN("bridge %s: netflow port mangling will conflict "
833 "with another port when more than 508 ports are "
838 opts.collectors.n = nf_cfg->n_targets;
839 opts.collectors.names = nf_cfg->targets;
840 if (ofproto_set_netflow(br->ofproto, &opts)) {
841 VLOG_ERR("bridge %s: problem setting netflow collectors",
845 ofproto_set_netflow(br->ofproto, NULL);
848 /* Set sFlow configuration on this bridge. */
849 if (br->cfg->sflow) {
850 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
851 struct ovsrec_controller **controllers;
852 struct ofproto_sflow_options oso;
853 size_t n_controllers;
855 memset(&oso, 0, sizeof oso);
857 oso.targets.n = sflow_cfg->n_targets;
858 oso.targets.names = sflow_cfg->targets;
860 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
861 if (sflow_cfg->sampling) {
862 oso.sampling_rate = *sflow_cfg->sampling;
865 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
866 if (sflow_cfg->polling) {
867 oso.polling_interval = *sflow_cfg->polling;
870 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
871 if (sflow_cfg->header) {
872 oso.header_len = *sflow_cfg->header;
875 oso.sub_id = sflow_bridge_number++;
876 oso.agent_device = sflow_cfg->agent;
878 oso.control_ip = NULL;
879 n_controllers = bridge_get_controllers(br, &controllers);
880 for (i = 0; i < n_controllers; i++) {
881 if (controllers[i]->local_ip) {
882 oso.control_ip = controllers[i]->local_ip;
886 ofproto_set_sflow(br->ofproto, &oso);
888 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
890 ofproto_set_sflow(br->ofproto, NULL);
893 /* Update the controller and related settings. It would be more
894 * straightforward to call this from bridge_reconfigure_one(), but we
895 * can't do it there for two reasons. First, and most importantly, at
896 * that point we don't know the dp_ifidx of any interfaces that have
897 * been added to the bridge (because we haven't actually added them to
898 * the datapath). Second, at that point we haven't set the datapath ID
899 * yet; when a controller is configured, resetting the datapath ID will
900 * immediately disconnect from the controller, so it's better to set
901 * the datapath ID before the controller. */
902 bridge_reconfigure_remotes(br, managers, n_managers);
904 LIST_FOR_EACH (br, node, &all_bridges) {
905 for (i = 0; i < br->n_ports; i++) {
906 struct port *port = br->ports[i];
909 port_update_vlan_compat(port);
910 port_update_bonding(port);
912 for (j = 0; j < port->n_ifaces; j++) {
913 iface_update_qos(port->ifaces[j], port->cfg->qos);
917 LIST_FOR_EACH (br, node, &all_bridges) {
918 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
925 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
926 const struct ovsdb_idl_column *column,
929 const struct ovsdb_datum *datum;
930 union ovsdb_atom atom;
933 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
934 atom.string = (char *) key;
935 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
936 return idx == UINT_MAX ? NULL : datum->values[idx].string;
940 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
942 return get_ovsrec_key_value(&br_cfg->header_,
943 &ovsrec_bridge_col_other_config, key);
947 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
948 struct iface **hw_addr_iface)
954 *hw_addr_iface = NULL;
956 /* Did the user request a particular MAC? */
957 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
958 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
959 if (eth_addr_is_multicast(ea)) {
960 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
961 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
962 } else if (eth_addr_is_zero(ea)) {
963 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
969 /* Otherwise choose the minimum non-local MAC address among all of the
971 memset(ea, 0xff, sizeof ea);
972 for (i = 0; i < br->n_ports; i++) {
973 struct port *port = br->ports[i];
974 uint8_t iface_ea[ETH_ADDR_LEN];
977 /* Mirror output ports don't participate. */
978 if (port->is_mirror_output_port) {
982 /* Choose the MAC address to represent the port. */
983 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
984 /* Find the interface with this Ethernet address (if any) so that
985 * we can provide the correct devname to the caller. */
987 for (j = 0; j < port->n_ifaces; j++) {
988 struct iface *candidate = port->ifaces[j];
989 uint8_t candidate_ea[ETH_ADDR_LEN];
990 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
991 && eth_addr_equals(iface_ea, candidate_ea)) {
996 /* Choose the interface whose MAC address will represent the port.
997 * The Linux kernel bonding code always chooses the MAC address of
998 * the first slave added to a bond, and the Fedora networking
999 * scripts always add slaves to a bond in alphabetical order, so
1000 * for compatibility we choose the interface with the name that is
1001 * first in alphabetical order. */
1002 iface = port->ifaces[0];
1003 for (j = 1; j < port->n_ifaces; j++) {
1004 struct iface *candidate = port->ifaces[j];
1005 if (strcmp(candidate->name, iface->name) < 0) {
1010 /* The local port doesn't count (since we're trying to choose its
1011 * MAC address anyway). */
1012 if (iface->dp_ifidx == ODPP_LOCAL) {
1017 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1019 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1020 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1021 iface->name, strerror(error));
1026 /* Compare against our current choice. */
1027 if (!eth_addr_is_multicast(iface_ea) &&
1028 !eth_addr_is_local(iface_ea) &&
1029 !eth_addr_is_reserved(iface_ea) &&
1030 !eth_addr_is_zero(iface_ea) &&
1031 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
1033 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1034 *hw_addr_iface = iface;
1037 if (eth_addr_is_multicast(ea)) {
1038 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1039 *hw_addr_iface = NULL;
1040 VLOG_WARN("bridge %s: using default bridge Ethernet "
1041 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1043 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1044 br->name, ETH_ADDR_ARGS(ea));
1048 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1049 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1050 * an interface on 'br', then that interface must be passed in as
1051 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1052 * 'hw_addr_iface' must be passed in as a null pointer. */
1054 bridge_pick_datapath_id(struct bridge *br,
1055 const uint8_t bridge_ea[ETH_ADDR_LEN],
1056 struct iface *hw_addr_iface)
1059 * The procedure for choosing a bridge MAC address will, in the most
1060 * ordinary case, also choose a unique MAC that we can use as a datapath
1061 * ID. In some special cases, though, multiple bridges will end up with
1062 * the same MAC address. This is OK for the bridges, but it will confuse
1063 * the OpenFlow controller, because each datapath needs a unique datapath
1066 * Datapath IDs must be unique. It is also very desirable that they be
1067 * stable from one run to the next, so that policy set on a datapath
1070 const char *datapath_id;
1073 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1074 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1078 if (hw_addr_iface) {
1080 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1082 * A bridge whose MAC address is taken from a VLAN network device
1083 * (that is, a network device created with vconfig(8) or similar
1084 * tool) will have the same MAC address as a bridge on the VLAN
1085 * device's physical network device.
1087 * Handle this case by hashing the physical network device MAC
1088 * along with the VLAN identifier.
1090 uint8_t buf[ETH_ADDR_LEN + 2];
1091 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1092 buf[ETH_ADDR_LEN] = vlan >> 8;
1093 buf[ETH_ADDR_LEN + 1] = vlan;
1094 return dpid_from_hash(buf, sizeof buf);
1097 * Assume that this bridge's MAC address is unique, since it
1098 * doesn't fit any of the cases we handle specially.
1103 * A purely internal bridge, that is, one that has no non-virtual
1104 * network devices on it at all, is more difficult because it has no
1105 * natural unique identifier at all.
1107 * When the host is a XenServer, we handle this case by hashing the
1108 * host's UUID with the name of the bridge. Names of bridges are
1109 * persistent across XenServer reboots, although they can be reused if
1110 * an internal network is destroyed and then a new one is later
1111 * created, so this is fairly effective.
1113 * When the host is not a XenServer, we punt by using a random MAC
1114 * address on each run.
1116 const char *host_uuid = xenserver_get_host_uuid();
1118 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1119 dpid = dpid_from_hash(combined, strlen(combined));
1125 return eth_addr_to_uint64(bridge_ea);
1129 dpid_from_hash(const void *data, size_t n)
1131 uint8_t hash[SHA1_DIGEST_SIZE];
1133 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1134 sha1_bytes(data, n, hash);
1135 eth_addr_mark_random(hash);
1136 return eth_addr_to_uint64(hash);
1140 iface_refresh_stats(struct iface *iface)
1146 static const struct iface_stat iface_stats[] = {
1147 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1148 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1149 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1150 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1151 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1152 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1153 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1154 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1155 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1156 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1157 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1158 { "collisions", offsetof(struct netdev_stats, collisions) },
1160 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1161 const struct iface_stat *s;
1163 char *keys[N_STATS];
1164 int64_t values[N_STATS];
1167 struct netdev_stats stats;
1169 /* Intentionally ignore return value, since errors will set 'stats' to
1170 * all-1s, and we will deal with that correctly below. */
1171 netdev_get_stats(iface->netdev, &stats);
1174 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1175 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1176 if (value != UINT64_MAX) {
1183 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1187 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1189 struct ovsdb_datum datum;
1193 get_system_stats(&stats);
1195 ovsdb_datum_from_shash(&datum, &stats);
1196 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1203 const struct ovsrec_open_vswitch *cfg;
1205 bool datapath_destroyed;
1206 bool database_changed;
1209 /* Let each bridge do the work that it needs to do. */
1210 datapath_destroyed = false;
1211 LIST_FOR_EACH (br, node, &all_bridges) {
1212 int error = bridge_run_one(br);
1214 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1215 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1216 "forcing reconfiguration", br->name);
1217 datapath_destroyed = true;
1221 /* (Re)configure if necessary. */
1222 database_changed = ovsdb_idl_run(idl);
1223 cfg = ovsrec_open_vswitch_first(idl);
1224 if (database_changed || datapath_destroyed) {
1226 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1228 bridge_configure_once(cfg);
1229 bridge_reconfigure(cfg);
1231 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1232 ovsdb_idl_txn_commit(txn);
1233 ovsdb_idl_txn_destroy(txn); /* XXX */
1235 /* We still need to reconfigure to avoid dangling pointers to
1236 * now-destroyed ovsrec structures inside bridge data. */
1237 static const struct ovsrec_open_vswitch null_cfg;
1239 bridge_reconfigure(&null_cfg);
1244 /* Re-configure SSL. We do this on every trip through the main loop,
1245 * instead of just when the database changes, because the contents of the
1246 * key and certificate files can change without the database changing. */
1247 if (cfg && cfg->ssl) {
1248 const struct ovsrec_ssl *ssl = cfg->ssl;
1250 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1251 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1255 /* Refresh system and interface stats if necessary. */
1256 if (time_msec() >= stats_timer) {
1258 struct ovsdb_idl_txn *txn;
1260 txn = ovsdb_idl_txn_create(idl);
1261 LIST_FOR_EACH (br, node, &all_bridges) {
1264 for (i = 0; i < br->n_ports; i++) {
1265 struct port *port = br->ports[i];
1268 for (j = 0; j < port->n_ifaces; j++) {
1269 struct iface *iface = port->ifaces[j];
1270 iface_refresh_stats(iface);
1274 refresh_system_stats(cfg);
1275 ovsdb_idl_txn_commit(txn);
1276 ovsdb_idl_txn_destroy(txn); /* XXX */
1279 stats_timer = time_msec() + STATS_INTERVAL;
1288 LIST_FOR_EACH (br, node, &all_bridges) {
1289 ofproto_wait(br->ofproto);
1290 if (ofproto_has_primary_controller(br->ofproto)) {
1294 mac_learning_wait(br->ml);
1297 ovsdb_idl_wait(idl);
1298 poll_timer_wait_until(stats_timer);
1301 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1302 * configuration changes. */
1304 bridge_flush(struct bridge *br)
1306 COVERAGE_INC(bridge_flush);
1308 mac_learning_flush(br->ml);
1311 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1312 * such interface. */
1313 static struct iface *
1314 bridge_get_local_iface(struct bridge *br)
1318 for (i = 0; i < br->n_ports; i++) {
1319 struct port *port = br->ports[i];
1320 for (j = 0; j < port->n_ifaces; j++) {
1321 struct iface *iface = port->ifaces[j];
1322 if (iface->dp_ifidx == ODPP_LOCAL) {
1331 /* Bridge unixctl user interface functions. */
1333 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1334 const char *args, void *aux OVS_UNUSED)
1336 struct ds ds = DS_EMPTY_INITIALIZER;
1337 const struct bridge *br;
1338 const struct mac_entry *e;
1340 br = bridge_lookup(args);
1342 unixctl_command_reply(conn, 501, "no such bridge");
1346 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1347 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1348 if (e->port < 0 || e->port >= br->n_ports) {
1351 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1352 br->ports[e->port]->ifaces[0]->dp_ifidx,
1353 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1355 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1359 /* Bridge reconfiguration functions. */
1360 static struct bridge *
1361 bridge_create(const struct ovsrec_bridge *br_cfg)
1366 assert(!bridge_lookup(br_cfg->name));
1367 br = xzalloc(sizeof *br);
1369 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1375 dpif_flow_flush(br->dpif);
1377 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1380 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1382 dpif_delete(br->dpif);
1383 dpif_close(br->dpif);
1388 br->name = xstrdup(br_cfg->name);
1390 br->ml = mac_learning_create();
1391 eth_addr_nicira_random(br->default_ea);
1393 hmap_init(&br->ifaces);
1395 shash_init(&br->port_by_name);
1396 shash_init(&br->iface_by_name);
1400 list_push_back(&all_bridges, &br->node);
1402 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1408 bridge_destroy(struct bridge *br)
1413 while (br->n_ports > 0) {
1414 port_destroy(br->ports[br->n_ports - 1]);
1416 list_remove(&br->node);
1417 error = dpif_delete(br->dpif);
1418 if (error && error != ENOENT) {
1419 VLOG_ERR("failed to delete %s: %s",
1420 dpif_name(br->dpif), strerror(error));
1422 dpif_close(br->dpif);
1423 ofproto_destroy(br->ofproto);
1424 mac_learning_destroy(br->ml);
1425 hmap_destroy(&br->ifaces);
1426 shash_destroy(&br->port_by_name);
1427 shash_destroy(&br->iface_by_name);
1434 static struct bridge *
1435 bridge_lookup(const char *name)
1439 LIST_FOR_EACH (br, node, &all_bridges) {
1440 if (!strcmp(br->name, name)) {
1447 /* Handle requests for a listing of all flows known by the OpenFlow
1448 * stack, including those normally hidden. */
1450 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1451 const char *args, void *aux OVS_UNUSED)
1456 br = bridge_lookup(args);
1458 unixctl_command_reply(conn, 501, "Unknown bridge");
1463 ofproto_get_all_flows(br->ofproto, &results);
1465 unixctl_command_reply(conn, 200, ds_cstr(&results));
1466 ds_destroy(&results);
1469 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1470 * connections and reconnect. If BRIDGE is not specified, then all bridges
1471 * drop their controller connections and reconnect. */
1473 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1474 const char *args, void *aux OVS_UNUSED)
1477 if (args[0] != '\0') {
1478 br = bridge_lookup(args);
1480 unixctl_command_reply(conn, 501, "Unknown bridge");
1483 ofproto_reconnect_controllers(br->ofproto);
1485 LIST_FOR_EACH (br, node, &all_bridges) {
1486 ofproto_reconnect_controllers(br->ofproto);
1489 unixctl_command_reply(conn, 200, NULL);
1493 bridge_run_one(struct bridge *br)
1497 error = ofproto_run1(br->ofproto);
1502 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1505 error = ofproto_run2(br->ofproto, br->flush);
1512 bridge_get_controllers(const struct bridge *br,
1513 struct ovsrec_controller ***controllersp)
1515 struct ovsrec_controller **controllers;
1516 size_t n_controllers;
1518 controllers = br->cfg->controller;
1519 n_controllers = br->cfg->n_controller;
1521 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1527 *controllersp = controllers;
1529 return n_controllers;
1533 bridge_reconfigure_one(struct bridge *br)
1535 struct shash old_ports, new_ports;
1536 struct svec snoops, old_snoops;
1537 struct shash_node *node;
1538 enum ofproto_fail_mode fail_mode;
1541 /* Collect old ports. */
1542 shash_init(&old_ports);
1543 for (i = 0; i < br->n_ports; i++) {
1544 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1547 /* Collect new ports. */
1548 shash_init(&new_ports);
1549 for (i = 0; i < br->cfg->n_ports; i++) {
1550 const char *name = br->cfg->ports[i]->name;
1551 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1552 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1557 /* If we have a controller, then we need a local port. Complain if the
1558 * user didn't specify one.
1560 * XXX perhaps we should synthesize a port ourselves in this case. */
1561 if (bridge_get_controllers(br, NULL)) {
1562 char local_name[IF_NAMESIZE];
1565 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1566 local_name, sizeof local_name);
1567 if (!error && !shash_find(&new_ports, local_name)) {
1568 VLOG_WARN("bridge %s: controller specified but no local port "
1569 "(port named %s) defined",
1570 br->name, local_name);
1574 /* Get rid of deleted ports.
1575 * Get rid of deleted interfaces on ports that still exist. */
1576 SHASH_FOR_EACH (node, &old_ports) {
1577 struct port *port = node->data;
1578 const struct ovsrec_port *port_cfg;
1580 port_cfg = shash_find_data(&new_ports, node->name);
1584 port_del_ifaces(port, port_cfg);
1588 /* Create new ports.
1589 * Add new interfaces to existing ports.
1590 * Reconfigure existing ports. */
1591 SHASH_FOR_EACH (node, &new_ports) {
1592 struct port *port = shash_find_data(&old_ports, node->name);
1594 port = port_create(br, node->name);
1597 port_reconfigure(port, node->data);
1598 if (!port->n_ifaces) {
1599 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1600 br->name, port->name);
1604 shash_destroy(&old_ports);
1605 shash_destroy(&new_ports);
1607 /* Set the fail-mode */
1608 fail_mode = !br->cfg->fail_mode
1609 || !strcmp(br->cfg->fail_mode, "standalone")
1610 ? OFPROTO_FAIL_STANDALONE
1611 : OFPROTO_FAIL_SECURE;
1612 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1613 && !ofproto_has_primary_controller(br->ofproto)) {
1614 ofproto_flush_flows(br->ofproto);
1616 ofproto_set_fail_mode(br->ofproto, fail_mode);
1618 /* Delete all flows if we're switching from connected to standalone or vice
1619 * versa. (XXX Should we delete all flows if we are switching from one
1620 * controller to another?) */
1622 /* Configure OpenFlow controller connection snooping. */
1624 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1625 ovs_rundir, br->name));
1626 svec_init(&old_snoops);
1627 ofproto_get_snoops(br->ofproto, &old_snoops);
1628 if (!svec_equal(&snoops, &old_snoops)) {
1629 ofproto_set_snoops(br->ofproto, &snoops);
1631 svec_destroy(&snoops);
1632 svec_destroy(&old_snoops);
1634 mirror_reconfigure(br);
1637 /* Initializes 'oc' appropriately as a management service controller for
1640 * The caller must free oc->target when it is no longer needed. */
1642 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1643 struct ofproto_controller *oc)
1645 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir, br->name);
1646 oc->max_backoff = 0;
1647 oc->probe_interval = 60;
1648 oc->band = OFPROTO_OUT_OF_BAND;
1649 oc->accept_re = NULL;
1650 oc->update_resolv_conf = false;
1652 oc->burst_limit = 0;
1655 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1657 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1658 struct ofproto_controller *oc)
1660 oc->target = c->target;
1661 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1662 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1663 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1664 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1665 oc->accept_re = c->discover_accept_regex;
1666 oc->update_resolv_conf = c->discover_update_resolv_conf;
1667 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1668 oc->burst_limit = (c->controller_burst_limit
1669 ? *c->controller_burst_limit : 0);
1672 /* Configures the IP stack for 'br''s local interface properly according to the
1673 * configuration in 'c'. */
1675 bridge_configure_local_iface_netdev(struct bridge *br,
1676 struct ovsrec_controller *c)
1678 struct netdev *netdev;
1679 struct in_addr mask, gateway;
1681 struct iface *local_iface;
1684 /* Controller discovery does its own TCP/IP configuration later. */
1685 if (strcmp(c->target, "discover")) {
1689 /* If there's no local interface or no IP address, give up. */
1690 local_iface = bridge_get_local_iface(br);
1691 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1695 /* Bring up the local interface. */
1696 netdev = local_iface->netdev;
1697 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1699 /* Configure the IP address and netmask. */
1700 if (!c->local_netmask
1701 || !inet_aton(c->local_netmask, &mask)
1703 mask.s_addr = guess_netmask(ip.s_addr);
1705 if (!netdev_set_in4(netdev, ip, mask)) {
1706 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1707 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1710 /* Configure the default gateway. */
1711 if (c->local_gateway
1712 && inet_aton(c->local_gateway, &gateway)
1713 && gateway.s_addr) {
1714 if (!netdev_add_router(netdev, gateway)) {
1715 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1716 br->name, IP_ARGS(&gateway.s_addr));
1722 bridge_reconfigure_remotes(struct bridge *br,
1723 const struct sockaddr_in *managers,
1726 struct ovsrec_controller **controllers;
1727 size_t n_controllers;
1729 const char *disable_ib_str;
1730 bool disable_in_band = false;
1732 struct ofproto_controller *ocs;
1737 /* Check if we should disable in-band control on this bridge. */
1738 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1739 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1740 disable_in_band = true;
1743 if (disable_in_band) {
1744 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1746 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1748 had_primary = ofproto_has_primary_controller(br->ofproto);
1750 n_controllers = bridge_get_controllers(br, &controllers);
1752 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1755 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1756 for (i = 0; i < n_controllers; i++) {
1757 struct ovsrec_controller *c = controllers[i];
1759 if (!strncmp(c->target, "punix:", 6)
1760 || !strncmp(c->target, "unix:", 5)) {
1761 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1763 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1764 * domain sockets and overwriting arbitrary local files. */
1765 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1766 "\"%s\" due to possibility for remote exploit",
1767 dpif_name(br->dpif), c->target);
1771 bridge_configure_local_iface_netdev(br, c);
1772 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
1773 if (disable_in_band) {
1774 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
1779 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1780 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1783 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1784 ofproto_flush_flows(br->ofproto);
1787 /* If there are no controllers and the bridge is in standalone
1788 * mode, set up a flow that matches every packet and directs
1789 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1790 * switch is in secure mode and we won't pass any traffic until
1791 * a controller has been defined and it tells us to do so. */
1793 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1794 union ofp_action action;
1797 memset(&action, 0, sizeof action);
1798 action.type = htons(OFPAT_OUTPUT);
1799 action.output.len = htons(sizeof action);
1800 action.output.port = htons(OFPP_NORMAL);
1801 memset(&flow, 0, sizeof flow);
1802 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1807 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1812 for (i = 0; i < br->n_ports; i++) {
1813 struct port *port = br->ports[i];
1814 for (j = 0; j < port->n_ifaces; j++) {
1815 struct iface *iface = port->ifaces[j];
1816 shash_add_once(ifaces, iface->name, iface);
1818 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1819 shash_add_once(ifaces, port->name, NULL);
1824 /* For robustness, in case the administrator moves around datapath ports behind
1825 * our back, we re-check all the datapath port numbers here.
1827 * This function will set the 'dp_ifidx' members of interfaces that have
1828 * disappeared to -1, so only call this function from a context where those
1829 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1830 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1831 * datapath, which doesn't support UINT16_MAX+1 ports. */
1833 bridge_fetch_dp_ifaces(struct bridge *br)
1835 struct odp_port *dpif_ports;
1836 size_t n_dpif_ports;
1839 /* Reset all interface numbers. */
1840 for (i = 0; i < br->n_ports; i++) {
1841 struct port *port = br->ports[i];
1842 for (j = 0; j < port->n_ifaces; j++) {
1843 struct iface *iface = port->ifaces[j];
1844 iface->dp_ifidx = -1;
1847 hmap_clear(&br->ifaces);
1849 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1850 for (i = 0; i < n_dpif_ports; i++) {
1851 struct odp_port *p = &dpif_ports[i];
1852 struct iface *iface = iface_lookup(br, p->devname);
1854 if (iface->dp_ifidx >= 0) {
1855 VLOG_WARN("%s reported interface %s twice",
1856 dpif_name(br->dpif), p->devname);
1857 } else if (iface_from_dp_ifidx(br, p->port)) {
1858 VLOG_WARN("%s reported interface %"PRIu16" twice",
1859 dpif_name(br->dpif), p->port);
1861 iface->dp_ifidx = p->port;
1862 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
1863 hash_int(iface->dp_ifidx, 0));
1866 iface_set_ofport(iface->cfg,
1867 (iface->dp_ifidx >= 0
1868 ? odp_port_to_ofp_port(iface->dp_ifidx)
1875 /* Bridge packet processing functions. */
1878 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1880 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1883 static struct bond_entry *
1884 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1886 return &port->bond_hash[bond_hash(mac)];
1890 bond_choose_iface(const struct port *port)
1892 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1893 size_t i, best_down_slave = -1;
1894 long long next_delay_expiration = LLONG_MAX;
1896 for (i = 0; i < port->n_ifaces; i++) {
1897 struct iface *iface = port->ifaces[i];
1899 if (iface->enabled) {
1901 } else if (iface->delay_expires < next_delay_expiration) {
1902 best_down_slave = i;
1903 next_delay_expiration = iface->delay_expires;
1907 if (best_down_slave != -1) {
1908 struct iface *iface = port->ifaces[best_down_slave];
1910 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1911 "since no other interface is up", iface->name,
1912 iface->delay_expires - time_msec());
1913 bond_enable_slave(iface, true);
1916 return best_down_slave;
1920 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1921 uint16_t *dp_ifidx, tag_type *tags)
1923 struct iface *iface;
1925 assert(port->n_ifaces);
1926 if (port->n_ifaces == 1) {
1927 iface = port->ifaces[0];
1929 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1930 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1931 || !port->ifaces[e->iface_idx]->enabled) {
1932 /* XXX select interface properly. The current interface selection
1933 * is only good for testing the rebalancing code. */
1934 e->iface_idx = bond_choose_iface(port);
1935 if (e->iface_idx < 0) {
1936 *tags |= port->no_ifaces_tag;
1939 e->iface_tag = tag_create_random();
1940 ((struct port *) port)->bond_compat_is_stale = true;
1942 *tags |= e->iface_tag;
1943 iface = port->ifaces[e->iface_idx];
1945 *dp_ifidx = iface->dp_ifidx;
1946 *tags |= iface->tag; /* Currently only used for bonding. */
1951 bond_link_status_update(struct iface *iface, bool carrier)
1953 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1954 struct port *port = iface->port;
1956 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1957 /* Nothing to do. */
1960 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1961 iface->name, carrier ? "detected" : "dropped");
1962 if (carrier == iface->enabled) {
1963 iface->delay_expires = LLONG_MAX;
1964 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1965 iface->name, carrier ? "disabled" : "enabled");
1966 } else if (carrier && port->active_iface < 0) {
1967 bond_enable_slave(iface, true);
1968 if (port->updelay) {
1969 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1970 "other interface is up", iface->name, port->updelay);
1973 int delay = carrier ? port->updelay : port->downdelay;
1974 iface->delay_expires = time_msec() + delay;
1977 "interface %s: will be %s if it stays %s for %d ms",
1979 carrier ? "enabled" : "disabled",
1980 carrier ? "up" : "down",
1987 bond_choose_active_iface(struct port *port)
1989 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1991 port->active_iface = bond_choose_iface(port);
1992 port->active_iface_tag = tag_create_random();
1993 if (port->active_iface >= 0) {
1994 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1995 port->name, port->ifaces[port->active_iface]->name);
1997 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2003 bond_enable_slave(struct iface *iface, bool enable)
2005 struct port *port = iface->port;
2006 struct bridge *br = port->bridge;
2008 /* This acts as a recursion check. If the act of disabling a slave
2009 * causes a different slave to be enabled, the flag will allow us to
2010 * skip redundant work when we reenter this function. It must be
2011 * cleared on exit to keep things safe with multiple bonds. */
2012 static bool moving_active_iface = false;
2014 iface->delay_expires = LLONG_MAX;
2015 if (enable == iface->enabled) {
2019 iface->enabled = enable;
2020 if (!iface->enabled) {
2021 VLOG_WARN("interface %s: disabled", iface->name);
2022 ofproto_revalidate(br->ofproto, iface->tag);
2023 if (iface->port_ifidx == port->active_iface) {
2024 ofproto_revalidate(br->ofproto,
2025 port->active_iface_tag);
2027 /* Disabling a slave can lead to another slave being immediately
2028 * enabled if there will be no active slaves but one is waiting
2029 * on an updelay. In this case we do not need to run most of the
2030 * code for the newly enabled slave since there was no period
2031 * without an active slave and it is redundant with the disabling
2033 moving_active_iface = true;
2034 bond_choose_active_iface(port);
2036 bond_send_learning_packets(port);
2038 VLOG_WARN("interface %s: enabled", iface->name);
2039 if (port->active_iface < 0 && !moving_active_iface) {
2040 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2041 bond_choose_active_iface(port);
2042 bond_send_learning_packets(port);
2044 iface->tag = tag_create_random();
2047 moving_active_iface = false;
2048 port->bond_compat_is_stale = true;
2051 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2052 * bond interface. */
2054 bond_update_fake_iface_stats(struct port *port)
2056 struct netdev_stats bond_stats;
2057 struct netdev *bond_dev;
2060 memset(&bond_stats, 0, sizeof bond_stats);
2062 for (i = 0; i < port->n_ifaces; i++) {
2063 struct netdev_stats slave_stats;
2065 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2066 /* XXX: We swap the stats here because they are swapped back when
2067 * reported by the internal device. The reason for this is
2068 * internal devices normally represent packets going into the system
2069 * but when used as fake bond device they represent packets leaving
2070 * the system. We really should do this in the internal device
2071 * itself because changing it here reverses the counts from the
2072 * perspective of the switch. However, the internal device doesn't
2073 * know what type of device it represents so we have to do it here
2075 bond_stats.tx_packets += slave_stats.rx_packets;
2076 bond_stats.tx_bytes += slave_stats.rx_bytes;
2077 bond_stats.rx_packets += slave_stats.tx_packets;
2078 bond_stats.rx_bytes += slave_stats.tx_bytes;
2082 if (!netdev_open_default(port->name, &bond_dev)) {
2083 netdev_set_stats(bond_dev, &bond_stats);
2084 netdev_close(bond_dev);
2089 bond_run(struct bridge *br)
2093 for (i = 0; i < br->n_ports; i++) {
2094 struct port *port = br->ports[i];
2096 if (port->n_ifaces >= 2) {
2099 /* Track carrier going up and down on interfaces. */
2100 while (!netdev_monitor_poll(port->monitor, &devname)) {
2101 struct iface *iface;
2103 iface = port_lookup_iface(port, devname);
2105 bool carrier = netdev_get_carrier(iface->netdev);
2107 bond_link_status_update(iface, carrier);
2108 port_update_bond_compat(port);
2113 for (j = 0; j < port->n_ifaces; j++) {
2114 struct iface *iface = port->ifaces[j];
2115 if (time_msec() >= iface->delay_expires) {
2116 bond_enable_slave(iface, !iface->enabled);
2120 if (port->bond_fake_iface
2121 && time_msec() >= port->bond_next_fake_iface_update) {
2122 bond_update_fake_iface_stats(port);
2123 port->bond_next_fake_iface_update = time_msec() + 1000;
2127 if (port->bond_compat_is_stale) {
2128 port->bond_compat_is_stale = false;
2129 port_update_bond_compat(port);
2135 bond_wait(struct bridge *br)
2139 for (i = 0; i < br->n_ports; i++) {
2140 struct port *port = br->ports[i];
2141 if (port->n_ifaces < 2) {
2144 netdev_monitor_poll_wait(port->monitor);
2145 for (j = 0; j < port->n_ifaces; j++) {
2146 struct iface *iface = port->ifaces[j];
2147 if (iface->delay_expires != LLONG_MAX) {
2148 poll_timer_wait_until(iface->delay_expires);
2151 if (port->bond_fake_iface) {
2152 poll_timer_wait_until(port->bond_next_fake_iface_update);
2158 set_dst(struct dst *p, const struct flow *flow,
2159 const struct port *in_port, const struct port *out_port,
2162 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2163 : in_port->vlan >= 0 ? in_port->vlan
2164 : ntohs(flow->dl_vlan));
2165 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
2169 swap_dst(struct dst *p, struct dst *q)
2171 struct dst tmp = *p;
2176 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2177 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2178 * that we push to the datapath. We could in fact fully sort the array by
2179 * vlan, but in most cases there are at most two different vlan tags so that's
2180 * possibly overkill.) */
2182 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2184 struct dst *first = dsts;
2185 struct dst *last = dsts + n_dsts;
2187 while (first != last) {
2189 * - All dsts < first have vlan == 'vlan'.
2190 * - All dsts >= last have vlan != 'vlan'.
2191 * - first < last. */
2192 while (first->vlan == vlan) {
2193 if (++first == last) {
2198 /* Same invariants, plus one additional:
2199 * - first->vlan != vlan.
2201 while (last[-1].vlan != vlan) {
2202 if (--last == first) {
2207 /* Same invariants, plus one additional:
2208 * - last[-1].vlan == vlan.*/
2209 swap_dst(first++, --last);
2214 mirror_mask_ffs(mirror_mask_t mask)
2216 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2221 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2222 const struct dst *test)
2225 for (i = 0; i < n_dsts; i++) {
2226 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2234 port_trunks_vlan(const struct port *port, uint16_t vlan)
2236 return (port->vlan < 0
2237 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2241 port_includes_vlan(const struct port *port, uint16_t vlan)
2243 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2247 port_is_floodable(const struct port *port)
2251 for (i = 0; i < port->n_ifaces; i++) {
2252 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2253 port->ifaces[i]->dp_ifidx)) {
2261 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2262 const struct port *in_port, const struct port *out_port,
2263 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2265 mirror_mask_t mirrors = in_port->src_mirrors;
2266 struct dst *dst = dsts;
2269 if (out_port == FLOOD_PORT) {
2270 /* XXX use ODP_FLOOD if no vlans or bonding. */
2271 /* XXX even better, define each VLAN as a datapath port group */
2272 for (i = 0; i < br->n_ports; i++) {
2273 struct port *port = br->ports[i];
2275 && port_is_floodable(port)
2276 && port_includes_vlan(port, vlan)
2277 && !port->is_mirror_output_port
2278 && set_dst(dst, flow, in_port, port, tags)) {
2279 mirrors |= port->dst_mirrors;
2283 *nf_output_iface = NF_OUT_FLOOD;
2284 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2285 *nf_output_iface = dst->dp_ifidx;
2286 mirrors |= out_port->dst_mirrors;
2291 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2292 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2294 if (set_dst(dst, flow, in_port, m->out_port, tags)
2295 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2299 for (i = 0; i < br->n_ports; i++) {
2300 struct port *port = br->ports[i];
2301 if (port_includes_vlan(port, m->out_vlan)
2302 && set_dst(dst, flow, in_port, port, tags))
2306 if (port->vlan < 0) {
2307 dst->vlan = m->out_vlan;
2309 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2313 /* Use the vlan tag on the original flow instead of
2314 * the one passed in the vlan parameter. This ensures
2315 * that we compare the vlan from before any implicit
2316 * tagging tags place. This is necessary because
2317 * dst->vlan is the final vlan, after removing implicit
2319 flow_vlan = ntohs(flow->dl_vlan);
2320 if (flow_vlan == 0) {
2321 flow_vlan = OFP_VLAN_NONE;
2323 if (port == in_port && dst->vlan == flow_vlan) {
2324 /* Don't send out input port on same VLAN. */
2332 mirrors &= mirrors - 1;
2335 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2339 static void OVS_UNUSED
2340 print_dsts(const struct dst *dsts, size_t n)
2342 for (; n--; dsts++) {
2343 printf(">p%"PRIu16, dsts->dp_ifidx);
2344 if (dsts->vlan != OFP_VLAN_NONE) {
2345 printf("v%"PRIu16, dsts->vlan);
2351 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2352 const struct port *in_port, const struct port *out_port,
2353 tag_type *tags, struct odp_actions *actions,
2354 uint16_t *nf_output_iface)
2356 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2358 const struct dst *p;
2361 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2364 cur_vlan = ntohs(flow->dl_vlan);
2365 for (p = dsts; p < &dsts[n_dsts]; p++) {
2366 union odp_action *a;
2367 if (p->vlan != cur_vlan) {
2368 if (p->vlan == OFP_VLAN_NONE) {
2369 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2371 a = odp_actions_add(actions, ODPAT_SET_DL_TCI);
2372 a->dl_tci.tci = htons(p->vlan & VLAN_VID_MASK);
2373 a->dl_tci.tci |= htons(flow->dl_vlan_pcp << VLAN_PCP_SHIFT);
2377 a = odp_actions_add(actions, ODPAT_OUTPUT);
2378 a->output.port = p->dp_ifidx;
2382 /* Returns the effective vlan of a packet, taking into account both the
2383 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2384 * the packet is untagged and -1 indicates it has an invalid header and
2385 * should be dropped. */
2386 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2387 struct port *in_port, bool have_packet)
2389 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2390 * belongs to VLAN 0, so we should treat both cases identically. (In the
2391 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2392 * presumably to allow a priority to be specified. In the latter case, the
2393 * packet does not have any 802.1Q header.) */
2394 int vlan = ntohs(flow->dl_vlan);
2395 if (vlan == OFP_VLAN_NONE) {
2398 if (in_port->vlan >= 0) {
2400 /* XXX support double tagging? */
2402 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2403 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2404 "packet received on port %s configured with "
2405 "implicit VLAN %"PRIu16,
2406 br->name, ntohs(flow->dl_vlan),
2407 in_port->name, in_port->vlan);
2411 vlan = in_port->vlan;
2413 if (!port_includes_vlan(in_port, vlan)) {
2415 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2416 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2417 "packet received on port %s not configured for "
2419 br->name, vlan, in_port->name, vlan);
2428 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2429 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2430 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2432 is_gratuitous_arp(const struct flow *flow)
2434 return (flow->dl_type == htons(ETH_TYPE_ARP)
2435 && eth_addr_is_broadcast(flow->dl_dst)
2436 && (flow->nw_proto == ARP_OP_REPLY
2437 || (flow->nw_proto == ARP_OP_REQUEST
2438 && flow->nw_src == flow->nw_dst)));
2442 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2443 struct port *in_port)
2445 enum grat_arp_lock_type lock_type;
2448 /* We don't want to learn from gratuitous ARP packets that are reflected
2449 * back over bond slaves so we lock the learning table. */
2450 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2451 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2452 GRAT_ARP_LOCK_CHECK;
2454 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2457 /* The log messages here could actually be useful in debugging,
2458 * so keep the rate limit relatively high. */
2459 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2461 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2462 "on port %s in VLAN %d",
2463 br->name, ETH_ADDR_ARGS(flow->dl_src),
2464 in_port->name, vlan);
2465 ofproto_revalidate(br->ofproto, rev_tag);
2469 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2470 * dropped. Returns true if they may be forwarded, false if they should be
2473 * If 'have_packet' is true, it indicates that the caller is processing a
2474 * received packet. If 'have_packet' is false, then the caller is just
2475 * revalidating an existing flow because configuration has changed. Either
2476 * way, 'have_packet' only affects logging (there is no point in logging errors
2477 * during revalidation).
2479 * Sets '*in_portp' to the input port. This will be a null pointer if
2480 * flow->in_port does not designate a known input port (in which case
2481 * is_admissible() returns false).
2483 * When returning true, sets '*vlanp' to the effective VLAN of the input
2484 * packet, as returned by flow_get_vlan().
2486 * May also add tags to '*tags', although the current implementation only does
2487 * so in one special case.
2490 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2491 tag_type *tags, int *vlanp, struct port **in_portp)
2493 struct iface *in_iface;
2494 struct port *in_port;
2497 /* Find the interface and port structure for the received packet. */
2498 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2500 /* No interface? Something fishy... */
2502 /* Odd. A few possible reasons here:
2504 * - We deleted an interface but there are still a few packets
2505 * queued up from it.
2507 * - Someone externally added an interface (e.g. with "ovs-dpctl
2508 * add-if") that we don't know about.
2510 * - Packet arrived on the local port but the local port is not
2511 * one of our bridge ports.
2513 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2515 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2516 "interface %"PRIu16, br->name, flow->in_port);
2522 *in_portp = in_port = in_iface->port;
2523 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2528 /* Drop frames for reserved multicast addresses. */
2529 if (eth_addr_is_reserved(flow->dl_dst)) {
2533 /* Drop frames on ports reserved for mirroring. */
2534 if (in_port->is_mirror_output_port) {
2536 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2537 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2538 "%s, which is reserved exclusively for mirroring",
2539 br->name, in_port->name);
2544 /* Packets received on bonds need special attention to avoid duplicates. */
2545 if (in_port->n_ifaces > 1) {
2547 bool is_grat_arp_locked;
2549 if (eth_addr_is_multicast(flow->dl_dst)) {
2550 *tags |= in_port->active_iface_tag;
2551 if (in_port->active_iface != in_iface->port_ifidx) {
2552 /* Drop all multicast packets on inactive slaves. */
2557 /* Drop all packets for which we have learned a different input
2558 * port, because we probably sent the packet on one slave and got
2559 * it back on the other. Gratuitous ARP packets are an exception
2560 * to this rule: the host has moved to another switch. The exception
2561 * to the exception is if we locked the learning table to avoid
2562 * reflections on bond slaves. If this is the case, just drop the
2564 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2565 &is_grat_arp_locked);
2566 if (src_idx != -1 && src_idx != in_port->port_idx &&
2567 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2575 /* If the composed actions may be applied to any packet in the given 'flow',
2576 * returns true. Otherwise, the actions should only be applied to 'packet', or
2577 * not at all, if 'packet' was NULL. */
2579 process_flow(struct bridge *br, const struct flow *flow,
2580 const struct ofpbuf *packet, struct odp_actions *actions,
2581 tag_type *tags, uint16_t *nf_output_iface)
2583 struct port *in_port;
2584 struct port *out_port;
2588 /* Check whether we should drop packets in this flow. */
2589 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2594 /* Learn source MAC (but don't try to learn from revalidation). */
2596 update_learning_table(br, flow, vlan, in_port);
2599 /* Determine output port. */
2600 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2602 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2603 out_port = br->ports[out_port_idx];
2604 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2605 /* If we are revalidating but don't have a learning entry then
2606 * eject the flow. Installing a flow that floods packets opens
2607 * up a window of time where we could learn from a packet reflected
2608 * on a bond and blackhole packets before the learning table is
2609 * updated to reflect the correct port. */
2612 out_port = FLOOD_PORT;
2615 /* Don't send packets out their input ports. */
2616 if (in_port == out_port) {
2622 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2630 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2631 struct odp_actions *actions, tag_type *tags,
2632 uint16_t *nf_output_iface, void *br_)
2634 struct bridge *br = br_;
2636 COVERAGE_INC(bridge_process_flow);
2638 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2642 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
2643 const union odp_action *actions,
2644 size_t n_actions, unsigned long long int n_bytes,
2647 struct bridge *br = br_;
2648 const union odp_action *a;
2649 struct port *in_port;
2653 /* Feed information from the active flows back into the learning table to
2654 * ensure that table is always in sync with what is actually flowing
2655 * through the datapath.
2657 * We test that 'tags' is nonzero to ensure that only flows that include an
2658 * OFPP_NORMAL action are used for learning. This works because
2659 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2660 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2661 update_learning_table(br, flow, vlan, in_port);
2664 /* Account for bond slave utilization. */
2665 if (!br->has_bonded_ports) {
2668 for (a = actions; a < &actions[n_actions]; a++) {
2669 if (a->type == ODPAT_OUTPUT) {
2670 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2671 if (out_port && out_port->n_ifaces >= 2) {
2672 struct bond_entry *e = lookup_bond_entry(out_port,
2674 e->tx_bytes += n_bytes;
2681 bridge_account_checkpoint_ofhook_cb(void *br_)
2683 struct bridge *br = br_;
2687 if (!br->has_bonded_ports) {
2692 for (i = 0; i < br->n_ports; i++) {
2693 struct port *port = br->ports[i];
2694 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2695 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2696 bond_rebalance_port(port);
2701 static struct ofhooks bridge_ofhooks = {
2702 bridge_normal_ofhook_cb,
2703 bridge_account_flow_ofhook_cb,
2704 bridge_account_checkpoint_ofhook_cb,
2707 /* Bonding functions. */
2709 /* Statistics for a single interface on a bonded port, used for load-based
2710 * bond rebalancing. */
2711 struct slave_balance {
2712 struct iface *iface; /* The interface. */
2713 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2715 /* All the "bond_entry"s that are assigned to this interface, in order of
2716 * increasing tx_bytes. */
2717 struct bond_entry **hashes;
2721 /* Sorts pointers to pointers to bond_entries in ascending order by the
2722 * interface to which they are assigned, and within a single interface in
2723 * ascending order of bytes transmitted. */
2725 compare_bond_entries(const void *a_, const void *b_)
2727 const struct bond_entry *const *ap = a_;
2728 const struct bond_entry *const *bp = b_;
2729 const struct bond_entry *a = *ap;
2730 const struct bond_entry *b = *bp;
2731 if (a->iface_idx != b->iface_idx) {
2732 return a->iface_idx > b->iface_idx ? 1 : -1;
2733 } else if (a->tx_bytes != b->tx_bytes) {
2734 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2740 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2741 * *descending* order by number of bytes transmitted. */
2743 compare_slave_balance(const void *a_, const void *b_)
2745 const struct slave_balance *a = a_;
2746 const struct slave_balance *b = b_;
2747 if (a->iface->enabled != b->iface->enabled) {
2748 return a->iface->enabled ? -1 : 1;
2749 } else if (a->tx_bytes != b->tx_bytes) {
2750 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2757 swap_bals(struct slave_balance *a, struct slave_balance *b)
2759 struct slave_balance tmp = *a;
2764 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2765 * given that 'p' (and only 'p') might be in the wrong location.
2767 * This function invalidates 'p', since it might now be in a different memory
2770 resort_bals(struct slave_balance *p,
2771 struct slave_balance bals[], size_t n_bals)
2774 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2775 swap_bals(p, p - 1);
2777 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2778 swap_bals(p, p + 1);
2784 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2786 if (VLOG_IS_DBG_ENABLED()) {
2787 struct ds ds = DS_EMPTY_INITIALIZER;
2788 const struct slave_balance *b;
2790 for (b = bals; b < bals + n_bals; b++) {
2794 ds_put_char(&ds, ',');
2796 ds_put_format(&ds, " %s %"PRIu64"kB",
2797 b->iface->name, b->tx_bytes / 1024);
2799 if (!b->iface->enabled) {
2800 ds_put_cstr(&ds, " (disabled)");
2802 if (b->n_hashes > 0) {
2803 ds_put_cstr(&ds, " (");
2804 for (i = 0; i < b->n_hashes; i++) {
2805 const struct bond_entry *e = b->hashes[i];
2807 ds_put_cstr(&ds, " + ");
2809 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2810 e - port->bond_hash, e->tx_bytes / 1024);
2812 ds_put_cstr(&ds, ")");
2815 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2820 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2822 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2825 struct bond_entry *hash = from->hashes[hash_idx];
2826 struct port *port = from->iface->port;
2827 uint64_t delta = hash->tx_bytes;
2829 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2830 "from %s to %s (now carrying %"PRIu64"kB and "
2831 "%"PRIu64"kB load, respectively)",
2832 port->name, delta / 1024, hash - port->bond_hash,
2833 from->iface->name, to->iface->name,
2834 (from->tx_bytes - delta) / 1024,
2835 (to->tx_bytes + delta) / 1024);
2837 /* Delete element from from->hashes.
2839 * We don't bother to add the element to to->hashes because not only would
2840 * it require more work, the only purpose it would be to allow that hash to
2841 * be migrated to another slave in this rebalancing run, and there is no
2842 * point in doing that. */
2843 if (hash_idx == 0) {
2846 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2847 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2851 /* Shift load away from 'from' to 'to'. */
2852 from->tx_bytes -= delta;
2853 to->tx_bytes += delta;
2855 /* Arrange for flows to be revalidated. */
2856 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2857 hash->iface_idx = to->iface->port_ifidx;
2858 hash->iface_tag = tag_create_random();
2862 bond_rebalance_port(struct port *port)
2864 struct slave_balance bals[DP_MAX_PORTS];
2866 struct bond_entry *hashes[BOND_MASK + 1];
2867 struct slave_balance *b, *from, *to;
2868 struct bond_entry *e;
2871 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2872 * descending order of tx_bytes, so that bals[0] represents the most
2873 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2876 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2877 * array for each slave_balance structure, we sort our local array of
2878 * hashes in order by slave, so that all of the hashes for a given slave
2879 * become contiguous in memory, and then we point each 'hashes' members of
2880 * a slave_balance structure to the start of a contiguous group. */
2881 n_bals = port->n_ifaces;
2882 for (b = bals; b < &bals[n_bals]; b++) {
2883 b->iface = port->ifaces[b - bals];
2888 for (i = 0; i <= BOND_MASK; i++) {
2889 hashes[i] = &port->bond_hash[i];
2891 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2892 for (i = 0; i <= BOND_MASK; i++) {
2894 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2895 b = &bals[e->iface_idx];
2896 b->tx_bytes += e->tx_bytes;
2898 b->hashes = &hashes[i];
2903 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2904 log_bals(bals, n_bals, port);
2906 /* Discard slaves that aren't enabled (which were sorted to the back of the
2907 * array earlier). */
2908 while (!bals[n_bals - 1].iface->enabled) {
2915 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2916 to = &bals[n_bals - 1];
2917 for (from = bals; from < to; ) {
2918 uint64_t overload = from->tx_bytes - to->tx_bytes;
2919 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2920 /* The extra load on 'from' (and all less-loaded slaves), compared
2921 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2922 * it is less than ~1Mbps. No point in rebalancing. */
2924 } else if (from->n_hashes == 1) {
2925 /* 'from' only carries a single MAC hash, so we can't shift any
2926 * load away from it, even though we want to. */
2929 /* 'from' is carrying significantly more load than 'to', and that
2930 * load is split across at least two different hashes. Pick a hash
2931 * to migrate to 'to' (the least-loaded slave), given that doing so
2932 * must decrease the ratio of the load on the two slaves by at
2935 * The sort order we use means that we prefer to shift away the
2936 * smallest hashes instead of the biggest ones. There is little
2937 * reason behind this decision; we could use the opposite sort
2938 * order to shift away big hashes ahead of small ones. */
2941 for (i = 0; i < from->n_hashes; i++) {
2942 double old_ratio, new_ratio;
2943 uint64_t delta = from->hashes[i]->tx_bytes;
2945 if (delta == 0 || from->tx_bytes - delta == 0) {
2946 /* Pointless move. */
2950 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2952 if (to->tx_bytes == 0) {
2953 /* Nothing on the new slave, move it. */
2957 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2958 new_ratio = (double)(from->tx_bytes - delta) /
2959 (to->tx_bytes + delta);
2961 if (new_ratio == 0) {
2962 /* Should already be covered but check to prevent division
2967 if (new_ratio < 1) {
2968 new_ratio = 1 / new_ratio;
2971 if (old_ratio - new_ratio > 0.1) {
2972 /* Would decrease the ratio, move it. */
2976 if (i < from->n_hashes) {
2977 bond_shift_load(from, to, i);
2978 port->bond_compat_is_stale = true;
2980 /* If the result of the migration changed the relative order of
2981 * 'from' and 'to' swap them back to maintain invariants. */
2982 if (order_swapped) {
2983 swap_bals(from, to);
2986 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2987 * point to different slave_balance structures. It is only
2988 * valid to do these two operations in a row at all because we
2989 * know that 'from' will not move past 'to' and vice versa. */
2990 resort_bals(from, bals, n_bals);
2991 resort_bals(to, bals, n_bals);
2998 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2999 * historical data to decay to <1% in 7 rebalancing runs. */
3000 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3006 bond_send_learning_packets(struct port *port)
3008 struct bridge *br = port->bridge;
3009 struct mac_entry *e;
3010 struct ofpbuf packet;
3011 int error, n_packets, n_errors;
3013 if (!port->n_ifaces || port->active_iface < 0) {
3017 ofpbuf_init(&packet, 128);
3018 error = n_packets = n_errors = 0;
3019 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3020 union ofp_action actions[2], *a;
3026 if (e->port == port->port_idx
3027 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
3031 /* Compose actions. */
3032 memset(actions, 0, sizeof actions);
3035 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3036 a->vlan_vid.len = htons(sizeof *a);
3037 a->vlan_vid.vlan_vid = htons(e->vlan);
3040 a->output.type = htons(OFPAT_OUTPUT);
3041 a->output.len = htons(sizeof *a);
3042 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3047 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3049 flow_extract(&packet, 0, ODPP_NONE, &flow);
3050 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3057 ofpbuf_uninit(&packet);
3060 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3061 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3062 "packets, last error was: %s",
3063 port->name, n_errors, n_packets, strerror(error));
3065 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3066 port->name, n_packets);
3070 /* Bonding unixctl user interface functions. */
3073 bond_unixctl_list(struct unixctl_conn *conn,
3074 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3076 struct ds ds = DS_EMPTY_INITIALIZER;
3077 const struct bridge *br;
3079 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
3081 LIST_FOR_EACH (br, node, &all_bridges) {
3084 for (i = 0; i < br->n_ports; i++) {
3085 const struct port *port = br->ports[i];
3086 if (port->n_ifaces > 1) {
3089 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
3090 for (j = 0; j < port->n_ifaces; j++) {
3091 const struct iface *iface = port->ifaces[j];
3093 ds_put_cstr(&ds, ", ");
3095 ds_put_cstr(&ds, iface->name);
3097 ds_put_char(&ds, '\n');
3101 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3105 static struct port *
3106 bond_find(const char *name)
3108 const struct bridge *br;
3110 LIST_FOR_EACH (br, node, &all_bridges) {
3113 for (i = 0; i < br->n_ports; i++) {
3114 struct port *port = br->ports[i];
3115 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3124 bond_unixctl_show(struct unixctl_conn *conn,
3125 const char *args, void *aux OVS_UNUSED)
3127 struct ds ds = DS_EMPTY_INITIALIZER;
3128 const struct port *port;
3131 port = bond_find(args);
3133 unixctl_command_reply(conn, 501, "no such bond");
3137 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3138 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3139 ds_put_format(&ds, "next rebalance: %lld ms\n",
3140 port->bond_next_rebalance - time_msec());
3141 for (j = 0; j < port->n_ifaces; j++) {
3142 const struct iface *iface = port->ifaces[j];
3143 struct bond_entry *be;
3146 ds_put_format(&ds, "slave %s: %s\n",
3147 iface->name, iface->enabled ? "enabled" : "disabled");
3148 if (j == port->active_iface) {
3149 ds_put_cstr(&ds, "\tactive slave\n");
3151 if (iface->delay_expires != LLONG_MAX) {
3152 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3153 iface->enabled ? "downdelay" : "updelay",
3154 iface->delay_expires - time_msec());
3158 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3159 int hash = be - port->bond_hash;
3160 struct mac_entry *me;
3162 if (be->iface_idx != j) {
3166 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3167 hash, be->tx_bytes / 1024);
3170 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3173 if (bond_hash(me->mac) == hash
3174 && me->port != port->port_idx
3175 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3176 && dp_ifidx == iface->dp_ifidx)
3178 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3179 ETH_ADDR_ARGS(me->mac));
3184 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3189 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3190 void *aux OVS_UNUSED)
3192 char *args = (char *) args_;
3193 char *save_ptr = NULL;
3194 char *bond_s, *hash_s, *slave_s;
3195 uint8_t mac[ETH_ADDR_LEN];
3197 struct iface *iface;
3198 struct bond_entry *entry;
3201 bond_s = strtok_r(args, " ", &save_ptr);
3202 hash_s = strtok_r(NULL, " ", &save_ptr);
3203 slave_s = strtok_r(NULL, " ", &save_ptr);
3205 unixctl_command_reply(conn, 501,
3206 "usage: bond/migrate BOND HASH SLAVE");
3210 port = bond_find(bond_s);
3212 unixctl_command_reply(conn, 501, "no such bond");
3216 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3217 == ETH_ADDR_SCAN_COUNT) {
3218 hash = bond_hash(mac);
3219 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3220 hash = atoi(hash_s) & BOND_MASK;
3222 unixctl_command_reply(conn, 501, "bad hash");
3226 iface = port_lookup_iface(port, slave_s);
3228 unixctl_command_reply(conn, 501, "no such slave");
3232 if (!iface->enabled) {
3233 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3237 entry = &port->bond_hash[hash];
3238 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3239 entry->iface_idx = iface->port_ifidx;
3240 entry->iface_tag = tag_create_random();
3241 port->bond_compat_is_stale = true;
3242 unixctl_command_reply(conn, 200, "migrated");
3246 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3247 void *aux OVS_UNUSED)
3249 char *args = (char *) args_;
3250 char *save_ptr = NULL;
3251 char *bond_s, *slave_s;
3253 struct iface *iface;
3255 bond_s = strtok_r(args, " ", &save_ptr);
3256 slave_s = strtok_r(NULL, " ", &save_ptr);
3258 unixctl_command_reply(conn, 501,
3259 "usage: bond/set-active-slave BOND SLAVE");
3263 port = bond_find(bond_s);
3265 unixctl_command_reply(conn, 501, "no such bond");
3269 iface = port_lookup_iface(port, slave_s);
3271 unixctl_command_reply(conn, 501, "no such slave");
3275 if (!iface->enabled) {
3276 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3280 if (port->active_iface != iface->port_ifidx) {
3281 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3282 port->active_iface = iface->port_ifidx;
3283 port->active_iface_tag = tag_create_random();
3284 VLOG_INFO("port %s: active interface is now %s",
3285 port->name, iface->name);
3286 bond_send_learning_packets(port);
3287 unixctl_command_reply(conn, 200, "done");
3289 unixctl_command_reply(conn, 200, "no change");
3294 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3296 char *args = (char *) args_;
3297 char *save_ptr = NULL;
3298 char *bond_s, *slave_s;
3300 struct iface *iface;
3302 bond_s = strtok_r(args, " ", &save_ptr);
3303 slave_s = strtok_r(NULL, " ", &save_ptr);
3305 unixctl_command_reply(conn, 501,
3306 "usage: bond/enable/disable-slave BOND SLAVE");
3310 port = bond_find(bond_s);
3312 unixctl_command_reply(conn, 501, "no such bond");
3316 iface = port_lookup_iface(port, slave_s);
3318 unixctl_command_reply(conn, 501, "no such slave");
3322 bond_enable_slave(iface, enable);
3323 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3327 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3328 void *aux OVS_UNUSED)
3330 enable_slave(conn, args, true);
3334 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3335 void *aux OVS_UNUSED)
3337 enable_slave(conn, args, false);
3341 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3342 void *aux OVS_UNUSED)
3344 uint8_t mac[ETH_ADDR_LEN];
3348 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3349 == ETH_ADDR_SCAN_COUNT) {
3350 hash = bond_hash(mac);
3352 hash_cstr = xasprintf("%u", hash);
3353 unixctl_command_reply(conn, 200, hash_cstr);
3356 unixctl_command_reply(conn, 501, "invalid mac");
3363 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3364 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3365 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3366 unixctl_command_register("bond/set-active-slave",
3367 bond_unixctl_set_active_slave, NULL);
3368 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3370 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3372 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3375 /* Port functions. */
3377 static struct port *
3378 port_create(struct bridge *br, const char *name)
3382 port = xzalloc(sizeof *port);
3384 port->port_idx = br->n_ports;
3386 port->trunks = NULL;
3387 port->name = xstrdup(name);
3388 port->active_iface = -1;
3390 if (br->n_ports >= br->allocated_ports) {
3391 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3394 br->ports[br->n_ports++] = port;
3395 shash_add_assert(&br->port_by_name, port->name, port);
3397 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3404 get_port_other_config(const struct ovsrec_port *port, const char *key,
3405 const char *default_value)
3409 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3411 return value ? value : default_value;
3415 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3417 struct shash new_ifaces;
3420 /* Collect list of new interfaces. */
3421 shash_init(&new_ifaces);
3422 for (i = 0; i < cfg->n_interfaces; i++) {
3423 const char *name = cfg->interfaces[i]->name;
3424 shash_add_once(&new_ifaces, name, NULL);
3427 /* Get rid of deleted interfaces. */
3428 for (i = 0; i < port->n_ifaces; ) {
3429 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3430 iface_destroy(port->ifaces[i]);
3436 shash_destroy(&new_ifaces);
3440 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3442 struct shash new_ifaces;
3443 long long int next_rebalance;
3444 unsigned long *trunks;
3450 /* Update settings. */
3451 port->updelay = cfg->bond_updelay;
3452 if (port->updelay < 0) {
3455 port->downdelay = cfg->bond_downdelay;
3456 if (port->downdelay < 0) {
3457 port->downdelay = 0;
3459 port->bond_rebalance_interval = atoi(
3460 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3461 if (port->bond_rebalance_interval < 1000) {
3462 port->bond_rebalance_interval = 1000;
3464 next_rebalance = time_msec() + port->bond_rebalance_interval;
3465 if (port->bond_next_rebalance > next_rebalance) {
3466 port->bond_next_rebalance = next_rebalance;
3469 /* Add new interfaces and update 'cfg' member of existing ones. */
3470 shash_init(&new_ifaces);
3471 for (i = 0; i < cfg->n_interfaces; i++) {
3472 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3473 struct iface *iface;
3475 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3476 VLOG_WARN("port %s: %s specified twice as port interface",
3477 port->name, if_cfg->name);
3478 iface_set_ofport(if_cfg, -1);
3482 iface = iface_lookup(port->bridge, if_cfg->name);
3484 if (iface->port != port) {
3485 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3487 port->bridge->name, if_cfg->name, iface->port->name);
3490 iface->cfg = if_cfg;
3492 iface = iface_create(port, if_cfg);
3495 /* Determine interface type. The local port always has type
3496 * "internal". Other ports take their type from the database and
3497 * default to "system" if none is specified. */
3498 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
3499 : if_cfg->type[0] ? if_cfg->type
3502 shash_destroy(&new_ifaces);
3507 if (port->n_ifaces < 2) {
3509 if (vlan >= 0 && vlan <= 4095) {
3510 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3515 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3516 * they even work as-is. But they have not been tested. */
3517 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3521 if (port->vlan != vlan) {
3523 bridge_flush(port->bridge);
3526 /* Get trunked VLANs. */
3528 if (vlan < 0 && cfg->n_trunks) {
3531 trunks = bitmap_allocate(4096);
3533 for (i = 0; i < cfg->n_trunks; i++) {
3534 int trunk = cfg->trunks[i];
3536 bitmap_set1(trunks, trunk);
3542 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3543 port->name, cfg->n_trunks);
3545 if (n_errors == cfg->n_trunks) {
3546 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3548 bitmap_free(trunks);
3551 } else if (vlan >= 0 && cfg->n_trunks) {
3552 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3556 ? port->trunks != NULL
3557 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3558 bridge_flush(port->bridge);
3560 bitmap_free(port->trunks);
3561 port->trunks = trunks;
3565 port_destroy(struct port *port)
3568 struct bridge *br = port->bridge;
3572 proc_net_compat_update_vlan(port->name, NULL, 0);
3573 proc_net_compat_update_bond(port->name, NULL);
3575 for (i = 0; i < MAX_MIRRORS; i++) {
3576 struct mirror *m = br->mirrors[i];
3577 if (m && m->out_port == port) {
3582 while (port->n_ifaces > 0) {
3583 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3586 shash_find_and_delete_assert(&br->port_by_name, port->name);
3588 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3589 del->port_idx = port->port_idx;
3591 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
3593 netdev_monitor_destroy(port->monitor);
3595 bitmap_free(port->trunks);
3602 static struct port *
3603 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3605 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3606 return iface ? iface->port : NULL;
3609 static struct port *
3610 port_lookup(const struct bridge *br, const char *name)
3612 return shash_find_data(&br->port_by_name, name);
3615 static struct iface *
3616 port_lookup_iface(const struct port *port, const char *name)
3618 struct iface *iface = iface_lookup(port->bridge, name);
3619 return iface && iface->port == port ? iface : NULL;
3623 port_update_bonding(struct port *port)
3625 if (port->monitor) {
3626 netdev_monitor_destroy(port->monitor);
3627 port->monitor = NULL;
3629 if (port->n_ifaces < 2) {
3630 /* Not a bonded port. */
3631 if (port->bond_hash) {
3632 free(port->bond_hash);
3633 port->bond_hash = NULL;
3634 port->bond_compat_is_stale = true;
3635 port->bond_fake_iface = false;
3640 if (!port->bond_hash) {
3641 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3642 for (i = 0; i <= BOND_MASK; i++) {
3643 struct bond_entry *e = &port->bond_hash[i];
3647 port->no_ifaces_tag = tag_create_random();
3648 bond_choose_active_iface(port);
3649 port->bond_next_rebalance
3650 = time_msec() + port->bond_rebalance_interval;
3652 if (port->cfg->bond_fake_iface) {
3653 port->bond_next_fake_iface_update = time_msec();
3656 port->bond_compat_is_stale = true;
3657 port->bond_fake_iface = port->cfg->bond_fake_iface;
3659 port->monitor = netdev_monitor_create();
3660 for (i = 0; i < port->n_ifaces; i++) {
3661 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
3667 port_update_bond_compat(struct port *port)
3669 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3670 struct compat_bond bond;
3673 if (port->n_ifaces < 2) {
3674 proc_net_compat_update_bond(port->name, NULL);
3679 bond.updelay = port->updelay;
3680 bond.downdelay = port->downdelay;
3683 bond.hashes = compat_hashes;
3684 if (port->bond_hash) {
3685 const struct bond_entry *e;
3686 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3687 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3688 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3689 cbh->hash = e - port->bond_hash;
3690 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3695 bond.n_slaves = port->n_ifaces;
3696 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3697 for (i = 0; i < port->n_ifaces; i++) {
3698 struct iface *iface = port->ifaces[i];
3699 struct compat_bond_slave *slave = &bond.slaves[i];
3700 slave->name = iface->name;
3702 /* We need to make the same determination as the Linux bonding
3703 * code to determine whether a slave should be consider "up".
3704 * The Linux function bond_miimon_inspect() supports four
3705 * BOND_LINK_* states:
3707 * - BOND_LINK_UP: carrier detected, updelay has passed.
3708 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3709 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3710 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3712 * The function bond_info_show_slave() only considers BOND_LINK_UP
3713 * to be "up" and anything else to be "down".
3715 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3719 netdev_get_etheraddr(iface->netdev, slave->mac);
3722 if (port->bond_fake_iface) {
3723 struct netdev *bond_netdev;
3725 if (!netdev_open_default(port->name, &bond_netdev)) {
3727 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3729 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3731 netdev_close(bond_netdev);
3735 proc_net_compat_update_bond(port->name, &bond);
3740 port_update_vlan_compat(struct port *port)
3742 struct bridge *br = port->bridge;
3743 char *vlandev_name = NULL;
3745 if (port->vlan > 0) {
3746 /* Figure out the name that the VLAN device should actually have, if it
3747 * existed. This takes some work because the VLAN device would not
3748 * have port->name in its name; rather, it would have the trunk port's
3749 * name, and 'port' would be attached to a bridge that also had the
3750 * VLAN device one of its ports. So we need to find a trunk port that
3751 * includes port->vlan.
3753 * There might be more than one candidate. This doesn't happen on
3754 * XenServer, so if it happens we just pick the first choice in
3755 * alphabetical order instead of creating multiple VLAN devices. */
3757 for (i = 0; i < br->n_ports; i++) {
3758 struct port *p = br->ports[i];
3759 if (port_trunks_vlan(p, port->vlan)
3761 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3763 uint8_t ea[ETH_ADDR_LEN];
3764 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3765 if (!eth_addr_is_multicast(ea) &&
3766 !eth_addr_is_reserved(ea) &&
3767 !eth_addr_is_zero(ea)) {
3768 vlandev_name = p->name;
3773 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3776 /* Interface functions. */
3778 static struct iface *
3779 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3781 struct bridge *br = port->bridge;
3782 struct iface *iface;
3783 char *name = if_cfg->name;
3785 iface = xzalloc(sizeof *iface);
3787 iface->port_ifidx = port->n_ifaces;
3788 iface->name = xstrdup(name);
3789 iface->dp_ifidx = -1;
3790 iface->tag = tag_create_random();
3791 iface->delay_expires = LLONG_MAX;
3792 iface->netdev = NULL;
3793 iface->cfg = if_cfg;
3795 shash_add_assert(&br->iface_by_name, iface->name, iface);
3797 if (port->n_ifaces >= port->allocated_ifaces) {
3798 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3799 sizeof *port->ifaces);
3801 port->ifaces[port->n_ifaces++] = iface;
3802 if (port->n_ifaces > 1) {
3803 br->has_bonded_ports = true;
3806 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3814 iface_destroy(struct iface *iface)
3817 struct port *port = iface->port;
3818 struct bridge *br = port->bridge;
3819 bool del_active = port->active_iface == iface->port_ifidx;
3822 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3824 if (iface->dp_ifidx >= 0) {
3825 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
3828 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3829 del->port_ifidx = iface->port_ifidx;
3831 netdev_close(iface->netdev);
3834 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3835 bond_choose_active_iface(port);
3836 bond_send_learning_packets(port);
3842 bridge_flush(port->bridge);
3846 static struct iface *
3847 iface_lookup(const struct bridge *br, const char *name)
3849 return shash_find_data(&br->iface_by_name, name);
3852 static struct iface *
3853 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3855 struct iface *iface;
3857 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
3858 hash_int(dp_ifidx, 0), &br->ifaces) {
3859 if (iface->dp_ifidx == dp_ifidx) {
3866 /* Set Ethernet address of 'iface', if one is specified in the configuration
3869 iface_set_mac(struct iface *iface)
3871 uint8_t ea[ETH_ADDR_LEN];
3873 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3874 if (eth_addr_is_multicast(ea)) {
3875 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3877 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3878 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3879 iface->name, iface->name);
3881 int error = netdev_set_etheraddr(iface->netdev, ea);
3883 VLOG_ERR("interface %s: setting MAC failed (%s)",
3884 iface->name, strerror(error));
3890 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
3892 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
3895 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
3899 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
3901 * The value strings in '*shash' are taken directly from values[], not copied,
3902 * so the caller should not modify or free them. */
3904 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3905 struct shash *shash)
3910 for (i = 0; i < n; i++) {
3911 shash_add(shash, keys[i], values[i]);
3915 struct iface_delete_queues_cbdata {
3916 struct netdev *netdev;
3917 const struct ovsdb_datum *queues;
3921 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
3923 union ovsdb_atom atom;
3925 atom.integer = target;
3926 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
3930 iface_delete_queues(unsigned int queue_id,
3931 const struct shash *details OVS_UNUSED, void *cbdata_)
3933 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3935 if (!queue_ids_include(cbdata->queues, queue_id)) {
3936 netdev_delete_queue(cbdata->netdev, queue_id);
3941 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3943 if (!qos || qos->type[0] == '\0') {
3944 netdev_set_qos(iface->netdev, NULL, NULL);
3946 struct iface_delete_queues_cbdata cbdata;
3947 struct shash details;
3950 /* Configure top-level Qos for 'iface'. */
3951 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3952 qos->n_other_config, &details);
3953 netdev_set_qos(iface->netdev, qos->type, &details);
3954 shash_destroy(&details);
3956 /* Deconfigure queues that were deleted. */
3957 cbdata.netdev = iface->netdev;
3958 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
3960 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3962 /* Configure queues for 'iface'. */
3963 for (i = 0; i < qos->n_queues; i++) {
3964 const struct ovsrec_queue *queue = qos->value_queues[i];
3965 unsigned int queue_id = qos->key_queues[i];
3967 shash_from_ovs_idl_map(queue->key_other_config,
3968 queue->value_other_config,
3969 queue->n_other_config, &details);
3970 netdev_set_queue(iface->netdev, queue_id, &details);
3971 shash_destroy(&details);
3976 /* Port mirroring. */
3978 static struct mirror *
3979 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
3983 for (i = 0; i < MAX_MIRRORS; i++) {
3984 struct mirror *m = br->mirrors[i];
3985 if (m && uuid_equals(uuid, &m->uuid)) {
3993 mirror_reconfigure(struct bridge *br)
3995 unsigned long *rspan_vlans;
3998 /* Get rid of deleted mirrors. */
3999 for (i = 0; i < MAX_MIRRORS; i++) {
4000 struct mirror *m = br->mirrors[i];
4002 const struct ovsdb_datum *mc;
4003 union ovsdb_atom atom;
4005 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4006 atom.uuid = br->mirrors[i]->uuid;
4007 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4013 /* Add new mirrors and reconfigure existing ones. */
4014 for (i = 0; i < br->cfg->n_mirrors; i++) {
4015 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4016 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4018 mirror_reconfigure_one(m, cfg);
4020 mirror_create(br, cfg);
4024 /* Update port reserved status. */
4025 for (i = 0; i < br->n_ports; i++) {
4026 br->ports[i]->is_mirror_output_port = false;
4028 for (i = 0; i < MAX_MIRRORS; i++) {
4029 struct mirror *m = br->mirrors[i];
4030 if (m && m->out_port) {
4031 m->out_port->is_mirror_output_port = true;
4035 /* Update flooded vlans (for RSPAN). */
4037 if (br->cfg->n_flood_vlans) {
4038 rspan_vlans = bitmap_allocate(4096);
4040 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4041 int64_t vlan = br->cfg->flood_vlans[i];
4042 if (vlan >= 0 && vlan < 4096) {
4043 bitmap_set1(rspan_vlans, vlan);
4044 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4047 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4052 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4058 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4063 for (i = 0; ; i++) {
4064 if (i >= MAX_MIRRORS) {
4065 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4066 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4069 if (!br->mirrors[i]) {
4074 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4077 br->mirrors[i] = m = xzalloc(sizeof *m);
4080 m->name = xstrdup(cfg->name);
4081 shash_init(&m->src_ports);
4082 shash_init(&m->dst_ports);
4088 mirror_reconfigure_one(m, cfg);
4092 mirror_destroy(struct mirror *m)
4095 struct bridge *br = m->bridge;
4098 for (i = 0; i < br->n_ports; i++) {
4099 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4100 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4103 shash_destroy(&m->src_ports);
4104 shash_destroy(&m->dst_ports);
4107 m->bridge->mirrors[m->idx] = NULL;
4116 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4117 struct shash *names)
4121 for (i = 0; i < n_ports; i++) {
4122 const char *name = ports[i]->name;
4123 if (port_lookup(m->bridge, name)) {
4124 shash_add_once(names, name, NULL);
4126 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4127 "port %s", m->bridge->name, m->name, name);
4133 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4139 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4141 for (i = 0; i < cfg->n_select_vlan; i++) {
4142 int64_t vlan = cfg->select_vlan[i];
4143 if (vlan < 0 || vlan > 4095) {
4144 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4145 m->bridge->name, m->name, vlan);
4147 (*vlans)[n_vlans++] = vlan;
4154 vlan_is_mirrored(const struct mirror *m, int vlan)
4158 for (i = 0; i < m->n_vlans; i++) {
4159 if (m->vlans[i] == vlan) {
4167 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4171 for (i = 0; i < m->n_vlans; i++) {
4172 if (port_trunks_vlan(p, m->vlans[i])) {
4180 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4182 struct shash src_ports, dst_ports;
4183 mirror_mask_t mirror_bit;
4184 struct port *out_port;
4191 if (strcmp(cfg->name, m->name)) {
4193 m->name = xstrdup(cfg->name);
4196 /* Get output port. */
4197 if (cfg->output_port) {
4198 out_port = port_lookup(m->bridge, cfg->output_port->name);
4200 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4201 m->bridge->name, m->name);
4207 if (cfg->output_vlan) {
4208 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4209 "output vlan; ignoring output vlan",
4210 m->bridge->name, m->name);
4212 } else if (cfg->output_vlan) {
4214 out_vlan = *cfg->output_vlan;
4216 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4217 m->bridge->name, m->name);
4222 shash_init(&src_ports);
4223 shash_init(&dst_ports);
4224 if (cfg->select_all) {
4225 for (i = 0; i < m->bridge->n_ports; i++) {
4226 const char *name = m->bridge->ports[i]->name;
4227 shash_add_once(&src_ports, name, NULL);
4228 shash_add_once(&dst_ports, name, NULL);
4233 /* Get ports, and drop duplicates and ports that don't exist. */
4234 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4236 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4239 /* Get all the vlans, and drop duplicate and invalid vlans. */
4240 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4243 /* Update mirror data. */
4244 if (!shash_equal_keys(&m->src_ports, &src_ports)
4245 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4246 || m->n_vlans != n_vlans
4247 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4248 || m->out_port != out_port
4249 || m->out_vlan != out_vlan) {
4250 bridge_flush(m->bridge);
4252 shash_swap(&m->src_ports, &src_ports);
4253 shash_swap(&m->dst_ports, &dst_ports);
4256 m->n_vlans = n_vlans;
4257 m->out_port = out_port;
4258 m->out_vlan = out_vlan;
4261 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4262 for (i = 0; i < m->bridge->n_ports; i++) {
4263 struct port *port = m->bridge->ports[i];
4265 if (shash_find(&m->src_ports, port->name)
4268 ? port_trunks_any_mirrored_vlan(m, port)
4269 : vlan_is_mirrored(m, port->vlan)))) {
4270 port->src_mirrors |= mirror_bit;
4272 port->src_mirrors &= ~mirror_bit;
4275 if (shash_find(&m->dst_ports, port->name)) {
4276 port->dst_mirrors |= mirror_bit;
4278 port->dst_mirrors &= ~mirror_bit;
4283 shash_destroy(&src_ports);
4284 shash_destroy(&dst_ports);