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>
36 #include "classifier.h"
40 #include "dynamic-string.h"
46 #include "mac-learning.h"
50 #include "ofp-print.h"
52 #include "ofproto/netflow.h"
53 #include "ofproto/ofproto.h"
54 #include "ovsdb-data.h"
56 #include "poll-loop.h"
57 #include "proc-net-compat.h"
61 #include "socket-util.h"
62 #include "stream-ssl.h"
64 #include "system-stats.h"
69 #include "vswitchd/vswitch-idl.h"
70 #include "xenserver.h"
72 #include "sflow_api.h"
74 VLOG_DEFINE_THIS_MODULE(bridge);
76 COVERAGE_DEFINE(bridge_flush);
77 COVERAGE_DEFINE(bridge_process_flow);
78 COVERAGE_DEFINE(bridge_reconfigure);
86 struct dst builtin[32];
91 static void dst_set_init(struct dst_set *);
92 static void dst_set_add(struct dst_set *, const struct dst *);
93 static void dst_set_free(struct dst_set *);
96 /* These members are always valid. */
97 struct port *port; /* Containing port. */
98 size_t port_ifidx; /* Index within containing port. */
99 char *name; /* Host network device name. */
100 tag_type tag; /* Tag associated with this interface. */
101 long long delay_expires; /* Time after which 'enabled' may change. */
103 /* These members are valid only after bridge_reconfigure() causes them to
105 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
106 int dp_ifidx; /* Index within kernel datapath. */
107 struct netdev *netdev; /* Network device. */
108 bool enabled; /* May be chosen for flows? */
109 const char *type; /* Usually same as cfg->type. */
110 struct cfm *cfm; /* Connectivity Fault Management */
111 const struct ovsrec_interface *cfg;
114 #define BOND_MASK 0xff
116 int iface_idx; /* Index of assigned iface, or -1 if none. */
117 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
118 tag_type iface_tag; /* Tag associated with iface_idx. */
122 BT_SLB, /* Source Load Balance (Default). */
123 BT_AB /* Active Backup. */
126 #define MAX_MIRRORS 32
127 typedef uint32_t mirror_mask_t;
128 #define MIRROR_MASK_C(X) UINT32_C(X)
129 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
131 struct bridge *bridge;
134 struct uuid uuid; /* UUID of this "mirror" record in database. */
136 /* Selection criteria. */
137 struct shash src_ports; /* Name is port name; data is always NULL. */
138 struct shash dst_ports; /* Name is port name; data is always NULL. */
143 struct port *out_port;
147 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
149 struct bridge *bridge;
151 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
152 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
153 * NULL if all VLANs are trunked. */
154 const struct ovsrec_port *cfg;
157 /* An ordinary bridge port has 1 interface.
158 * A bridge port for bonding has at least 2 interfaces. */
159 struct iface **ifaces;
160 size_t n_ifaces, allocated_ifaces;
163 enum bond_type bond_type; /* Type of the bond. BT_SLB is the default. */
164 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
165 tag_type active_iface_tag; /* Tag for bcast flows. */
166 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
167 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
168 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
169 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
170 long long int bond_next_fake_iface_update; /* Time of next update. */
171 struct netdev_monitor *monitor; /* Tracks carrier up/down status. */
173 /* SLB specific bonding info. */
174 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
175 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
176 long long int bond_next_rebalance; /* Next rebalancing time. */
178 /* Port mirroring info. */
179 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
180 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
181 bool is_mirror_output_port; /* Does port mirroring send frames here? */
185 struct list node; /* Node in global list of bridges. */
186 char *name; /* User-specified arbitrary name. */
187 struct mac_learning *ml; /* MAC learning table. */
188 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
189 const struct ovsrec_bridge *cfg;
191 /* OpenFlow switch processing. */
192 struct ofproto *ofproto; /* OpenFlow switch. */
194 /* Kernel datapath information. */
195 struct dpif *dpif; /* Datapath. */
196 struct hmap ifaces; /* Contains "struct iface"s. */
200 size_t n_ports, allocated_ports;
201 struct shash iface_by_name; /* "struct iface"s indexed by name. */
202 struct shash port_by_name; /* "struct port"s indexed by name. */
205 bool has_bonded_ports;
210 /* Port mirroring. */
211 struct mirror *mirrors[MAX_MIRRORS];
214 /* List of all bridges. */
215 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
217 /* OVSDB IDL used to obtain configuration. */
218 static struct ovsdb_idl *idl;
220 /* Each time this timer expires, the bridge fetches systems and interface
221 * statistics and pushes them into the database. */
222 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
223 static long long int stats_timer = LLONG_MIN;
225 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
226 static void bridge_destroy(struct bridge *);
227 static struct bridge *bridge_lookup(const char *name);
228 static unixctl_cb_func bridge_unixctl_dump_flows;
229 static unixctl_cb_func bridge_unixctl_reconnect;
230 static int bridge_run_one(struct bridge *);
231 static size_t bridge_get_controllers(const struct bridge *br,
232 struct ovsrec_controller ***controllersp);
233 static void bridge_reconfigure_one(struct bridge *);
234 static void bridge_reconfigure_remotes(struct bridge *,
235 const struct sockaddr_in *managers,
237 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
238 static void bridge_fetch_dp_ifaces(struct bridge *);
239 static void bridge_flush(struct bridge *);
240 static void bridge_pick_local_hw_addr(struct bridge *,
241 uint8_t ea[ETH_ADDR_LEN],
242 struct iface **hw_addr_iface);
243 static uint64_t bridge_pick_datapath_id(struct bridge *,
244 const uint8_t bridge_ea[ETH_ADDR_LEN],
245 struct iface *hw_addr_iface);
246 static struct iface *bridge_get_local_iface(struct bridge *);
247 static uint64_t dpid_from_hash(const void *, size_t nbytes);
249 static unixctl_cb_func bridge_unixctl_fdb_show;
251 static void bond_init(void);
252 static void bond_run(struct bridge *);
253 static void bond_wait(struct bridge *);
254 static void bond_rebalance_port(struct port *);
255 static void bond_send_learning_packets(struct port *);
256 static void bond_enable_slave(struct iface *iface, bool enable);
258 static struct port *port_create(struct bridge *, const char *name);
259 static void port_reconfigure(struct port *, const struct ovsrec_port *);
260 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
261 static void port_destroy(struct port *);
262 static struct port *port_lookup(const struct bridge *, const char *name);
263 static struct iface *port_lookup_iface(const struct port *, const char *name);
264 static struct port *port_from_dp_ifidx(const struct bridge *,
266 static void port_update_bond_compat(struct port *);
267 static void port_update_vlan_compat(struct port *);
268 static void port_update_bonding(struct port *);
270 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
271 static void mirror_destroy(struct mirror *);
272 static void mirror_reconfigure(struct bridge *);
273 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
274 static bool vlan_is_mirrored(const struct mirror *, int vlan);
276 static struct iface *iface_create(struct port *port,
277 const struct ovsrec_interface *if_cfg);
278 static void iface_destroy(struct iface *);
279 static struct iface *iface_lookup(const struct bridge *, const char *name);
280 static struct iface *iface_from_dp_ifidx(const struct bridge *,
282 static void iface_set_mac(struct iface *);
283 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
284 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
285 static void iface_update_cfm(struct iface *);
286 static void iface_refresh_cfm_stats(struct iface *iface);
287 static void iface_send_packet(struct iface *, struct ofpbuf *packet);
289 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
292 /* Hooks into ofproto processing. */
293 static struct ofhooks bridge_ofhooks;
295 /* Public functions. */
297 /* Initializes the bridge module, configuring it to obtain its configuration
298 * from an OVSDB server accessed over 'remote', which should be a string in a
299 * form acceptable to ovsdb_idl_create(). */
301 bridge_init(const char *remote)
303 /* Create connection to database. */
304 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
306 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
307 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
308 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
310 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
312 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
313 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
315 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
316 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
317 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
319 /* Register unixctl commands. */
320 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
321 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
323 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
331 struct bridge *br, *next_br;
333 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
336 ovsdb_idl_destroy(idl);
339 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
340 * but for which the ovs-vswitchd configuration 'cfg' is required. */
342 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
344 static bool already_configured_once;
345 struct svec bridge_names;
346 struct svec dpif_names, dpif_types;
349 /* Only do this once per ovs-vswitchd run. */
350 if (already_configured_once) {
353 already_configured_once = true;
355 stats_timer = time_msec() + STATS_INTERVAL;
357 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
358 svec_init(&bridge_names);
359 for (i = 0; i < cfg->n_bridges; i++) {
360 svec_add(&bridge_names, cfg->bridges[i]->name);
362 svec_sort(&bridge_names);
364 /* Iterate over all system dpifs and delete any of them that do not appear
366 svec_init(&dpif_names);
367 svec_init(&dpif_types);
368 dp_enumerate_types(&dpif_types);
369 for (i = 0; i < dpif_types.n; i++) {
374 dp_enumerate_names(dpif_types.names[i], &dpif_names);
376 /* For each dpif... */
377 for (j = 0; j < dpif_names.n; j++) {
378 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
380 struct svec all_names;
383 /* ...check whether any of its names is in 'bridge_names'. */
384 svec_init(&all_names);
385 dpif_get_all_names(dpif, &all_names);
386 for (k = 0; k < all_names.n; k++) {
387 if (svec_contains(&bridge_names, all_names.names[k])) {
392 /* No. Delete the dpif. */
396 svec_destroy(&all_names);
401 svec_destroy(&bridge_names);
402 svec_destroy(&dpif_names);
403 svec_destroy(&dpif_types);
406 /* Callback for iterate_and_prune_ifaces(). */
408 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
410 if (!iface->netdev) {
411 /* We already reported a related error, don't bother duplicating it. */
415 if (iface->dp_ifidx < 0) {
416 VLOG_ERR("%s interface not in %s, dropping",
417 iface->name, dpif_name(br->dpif));
421 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
422 iface->name, iface->dp_ifidx);
426 /* Callback for iterate_and_prune_ifaces(). */
428 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
429 void *aux OVS_UNUSED)
431 /* Set policing attributes. */
432 netdev_set_policing(iface->netdev,
433 iface->cfg->ingress_policing_rate,
434 iface->cfg->ingress_policing_burst);
436 /* Set MAC address of internal interfaces other than the local
438 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
439 iface_set_mac(iface);
445 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
446 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
447 * deletes from 'br' any ports that no longer have any interfaces. */
449 iterate_and_prune_ifaces(struct bridge *br,
450 bool (*cb)(struct bridge *, struct iface *,
456 for (i = 0; i < br->n_ports; ) {
457 struct port *port = br->ports[i];
458 for (j = 0; j < port->n_ifaces; ) {
459 struct iface *iface = port->ifaces[j];
460 if (cb(br, iface, aux)) {
463 iface_set_ofport(iface->cfg, -1);
464 iface_destroy(iface);
468 if (port->n_ifaces) {
471 VLOG_ERR("%s port has no interfaces, dropping", port->name);
477 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
478 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
479 * responsible for freeing '*managersp' (with free()).
481 * You may be asking yourself "why does ovs-vswitchd care?", because
482 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
483 * should not be and in fact is not directly involved in that. But
484 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
485 * it has to tell in-band control where the managers are to enable that.
486 * (Thus, only managers connected in-band are collected.)
489 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
490 struct sockaddr_in **managersp, size_t *n_managersp)
492 struct sockaddr_in *managers = NULL;
493 size_t n_managers = 0;
494 struct shash targets;
497 /* Collect all of the potential targets, as the union of the "managers"
498 * column and the "targets" columns of the rows pointed to by
499 * "manager_options", excluding any that are out-of-band. */
500 shash_init(&targets);
501 for (i = 0; i < ovs_cfg->n_managers; i++) {
502 shash_add_once(&targets, ovs_cfg->managers[i], NULL);
504 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
505 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
507 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
508 shash_find_and_delete(&targets, m->target);
510 shash_add_once(&targets, m->target, NULL);
514 /* Now extract the targets' IP addresses. */
515 if (!shash_is_empty(&targets)) {
516 struct shash_node *node;
518 managers = xmalloc(shash_count(&targets) * sizeof *managers);
519 SHASH_FOR_EACH (node, &targets) {
520 const char *target = node->name;
521 struct sockaddr_in *sin = &managers[n_managers];
523 if ((!strncmp(target, "tcp:", 4)
524 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
525 (!strncmp(target, "ssl:", 4)
526 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
531 shash_destroy(&targets);
533 *managersp = managers;
534 *n_managersp = n_managers;
538 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
540 struct shash old_br, new_br;
541 struct shash_node *node;
542 struct bridge *br, *next;
543 struct sockaddr_in *managers;
546 int sflow_bridge_number;
548 COVERAGE_INC(bridge_reconfigure);
550 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
552 /* Collect old and new bridges. */
555 LIST_FOR_EACH (br, node, &all_bridges) {
556 shash_add(&old_br, br->name, br);
558 for (i = 0; i < ovs_cfg->n_bridges; i++) {
559 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
560 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
561 VLOG_WARN("more than one bridge named %s", br_cfg->name);
565 /* Get rid of deleted bridges and add new bridges. */
566 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
567 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
574 SHASH_FOR_EACH (node, &new_br) {
575 const char *br_name = node->name;
576 const struct ovsrec_bridge *br_cfg = node->data;
577 br = shash_find_data(&old_br, br_name);
579 /* If the bridge datapath type has changed, we need to tear it
580 * down and recreate. */
581 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
583 bridge_create(br_cfg);
586 bridge_create(br_cfg);
589 shash_destroy(&old_br);
590 shash_destroy(&new_br);
592 /* Reconfigure all bridges. */
593 LIST_FOR_EACH (br, node, &all_bridges) {
594 bridge_reconfigure_one(br);
597 /* Add and delete ports on all datapaths.
599 * The kernel will reject any attempt to add a given port to a datapath if
600 * that port already belongs to a different datapath, so we must do all
601 * port deletions before any port additions. */
602 LIST_FOR_EACH (br, node, &all_bridges) {
603 struct odp_port *dpif_ports;
605 struct shash want_ifaces;
607 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
608 bridge_get_all_ifaces(br, &want_ifaces);
609 for (i = 0; i < n_dpif_ports; i++) {
610 const struct odp_port *p = &dpif_ports[i];
611 if (!shash_find(&want_ifaces, p->devname)
612 && strcmp(p->devname, br->name)) {
613 int retval = dpif_port_del(br->dpif, p->port);
615 VLOG_ERR("failed to remove %s interface from %s: %s",
616 p->devname, dpif_name(br->dpif),
621 shash_destroy(&want_ifaces);
624 LIST_FOR_EACH (br, node, &all_bridges) {
625 struct odp_port *dpif_ports;
627 struct shash cur_ifaces, want_ifaces;
629 /* Get the set of interfaces currently in this datapath. */
630 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
631 shash_init(&cur_ifaces);
632 for (i = 0; i < n_dpif_ports; i++) {
633 const char *name = dpif_ports[i].devname;
634 shash_add_once(&cur_ifaces, name, &dpif_ports[i]);
637 /* Get the set of interfaces we want on this datapath. */
638 bridge_get_all_ifaces(br, &want_ifaces);
640 hmap_clear(&br->ifaces);
641 SHASH_FOR_EACH (node, &want_ifaces) {
642 const char *if_name = node->name;
643 struct iface *iface = node->data;
644 struct odp_port *dpif_port = shash_find_data(&cur_ifaces, if_name);
645 const char *type = iface ? iface->type : "internal";
648 /* If we have a port or a netdev already, and it's not the type we
649 * want, then delete the port (if any) and close the netdev (if
651 if ((dpif_port && strcmp(dpif_port->type, type))
652 || (iface && iface->netdev
653 && strcmp(type, netdev_get_type(iface->netdev)))) {
655 error = ofproto_port_del(br->ofproto, dpif_port->port);
662 netdev_close(iface->netdev);
663 iface->netdev = NULL;
667 /* If the port doesn't exist or we don't have the netdev open,
668 * we need to do more work. */
669 if (!dpif_port || (iface && !iface->netdev)) {
670 struct netdev_options options;
671 struct netdev *netdev;
674 /* First open the network device. */
675 options.name = if_name;
677 options.args = &args;
678 options.ethertype = NETDEV_ETH_TYPE_NONE;
682 shash_from_ovs_idl_map(iface->cfg->key_options,
683 iface->cfg->value_options,
684 iface->cfg->n_options, &args);
686 error = netdev_open(&options, &netdev);
687 shash_destroy(&args);
690 VLOG_WARN("could not open network device %s (%s)",
691 if_name, strerror(error));
695 /* Then add the port if we haven't already. */
697 error = dpif_port_add(br->dpif, netdev, NULL);
699 netdev_close(netdev);
700 if (error == EFBIG) {
701 VLOG_ERR("ran out of valid port numbers on %s",
702 dpif_name(br->dpif));
705 VLOG_ERR("failed to add %s interface to %s: %s",
706 if_name, dpif_name(br->dpif),
713 /* Update 'iface'. */
715 iface->netdev = netdev;
716 iface->enabled = netdev_get_carrier(iface->netdev);
718 } else if (iface && iface->netdev) {
722 shash_from_ovs_idl_map(iface->cfg->key_options,
723 iface->cfg->value_options,
724 iface->cfg->n_options, &args);
725 netdev_reconfigure(iface->netdev, &args);
726 shash_destroy(&args);
730 shash_destroy(&cur_ifaces);
731 shash_destroy(&want_ifaces);
733 sflow_bridge_number = 0;
734 LIST_FOR_EACH (br, node, &all_bridges) {
737 struct iface *local_iface;
738 struct iface *hw_addr_iface;
741 bridge_fetch_dp_ifaces(br);
743 iterate_and_prune_ifaces(br, check_iface, NULL);
745 /* Pick local port hardware address, datapath ID. */
746 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
747 local_iface = bridge_get_local_iface(br);
749 int error = netdev_set_etheraddr(local_iface->netdev, ea);
751 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
752 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
753 "Ethernet address: %s",
754 br->name, strerror(error));
758 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
759 ofproto_set_datapath_id(br->ofproto, dpid);
761 dpid_string = xasprintf("%016"PRIx64, dpid);
762 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
765 /* Set NetFlow configuration on this bridge. */
766 if (br->cfg->netflow) {
767 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
768 struct netflow_options opts;
770 memset(&opts, 0, sizeof opts);
772 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
773 if (nf_cfg->engine_type) {
774 opts.engine_type = *nf_cfg->engine_type;
776 if (nf_cfg->engine_id) {
777 opts.engine_id = *nf_cfg->engine_id;
780 opts.active_timeout = nf_cfg->active_timeout;
781 if (!opts.active_timeout) {
782 opts.active_timeout = -1;
783 } else if (opts.active_timeout < 0) {
784 VLOG_WARN("bridge %s: active timeout interval set to negative "
785 "value, using default instead (%d seconds)", br->name,
786 NF_ACTIVE_TIMEOUT_DEFAULT);
787 opts.active_timeout = -1;
790 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
791 if (opts.add_id_to_iface) {
792 if (opts.engine_id > 0x7f) {
793 VLOG_WARN("bridge %s: netflow port mangling may conflict "
794 "with another vswitch, choose an engine id less "
795 "than 128", br->name);
797 if (br->n_ports > 508) {
798 VLOG_WARN("bridge %s: netflow port mangling will conflict "
799 "with another port when more than 508 ports are "
804 opts.collectors.n = nf_cfg->n_targets;
805 opts.collectors.names = nf_cfg->targets;
806 if (ofproto_set_netflow(br->ofproto, &opts)) {
807 VLOG_ERR("bridge %s: problem setting netflow collectors",
811 ofproto_set_netflow(br->ofproto, NULL);
814 /* Set sFlow configuration on this bridge. */
815 if (br->cfg->sflow) {
816 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
817 struct ovsrec_controller **controllers;
818 struct ofproto_sflow_options oso;
819 size_t n_controllers;
821 memset(&oso, 0, sizeof oso);
823 oso.targets.n = sflow_cfg->n_targets;
824 oso.targets.names = sflow_cfg->targets;
826 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
827 if (sflow_cfg->sampling) {
828 oso.sampling_rate = *sflow_cfg->sampling;
831 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
832 if (sflow_cfg->polling) {
833 oso.polling_interval = *sflow_cfg->polling;
836 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
837 if (sflow_cfg->header) {
838 oso.header_len = *sflow_cfg->header;
841 oso.sub_id = sflow_bridge_number++;
842 oso.agent_device = sflow_cfg->agent;
844 oso.control_ip = NULL;
845 n_controllers = bridge_get_controllers(br, &controllers);
846 for (i = 0; i < n_controllers; i++) {
847 if (controllers[i]->local_ip) {
848 oso.control_ip = controllers[i]->local_ip;
852 ofproto_set_sflow(br->ofproto, &oso);
854 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
856 ofproto_set_sflow(br->ofproto, NULL);
859 /* Update the controller and related settings. It would be more
860 * straightforward to call this from bridge_reconfigure_one(), but we
861 * can't do it there for two reasons. First, and most importantly, at
862 * that point we don't know the dp_ifidx of any interfaces that have
863 * been added to the bridge (because we haven't actually added them to
864 * the datapath). Second, at that point we haven't set the datapath ID
865 * yet; when a controller is configured, resetting the datapath ID will
866 * immediately disconnect from the controller, so it's better to set
867 * the datapath ID before the controller. */
868 bridge_reconfigure_remotes(br, managers, n_managers);
870 LIST_FOR_EACH (br, node, &all_bridges) {
871 for (i = 0; i < br->n_ports; i++) {
872 struct port *port = br->ports[i];
875 port_update_vlan_compat(port);
876 port_update_bonding(port);
878 for (j = 0; j < port->n_ifaces; j++) {
879 iface_update_qos(port->ifaces[j], port->cfg->qos);
883 LIST_FOR_EACH (br, node, &all_bridges) {
884 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
887 LIST_FOR_EACH (br, node, &all_bridges) {
889 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
890 iface_update_cfm(iface);
898 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
899 const struct ovsdb_idl_column *column,
902 const struct ovsdb_datum *datum;
903 union ovsdb_atom atom;
906 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
907 atom.string = (char *) key;
908 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
909 return idx == UINT_MAX ? NULL : datum->values[idx].string;
913 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
915 return get_ovsrec_key_value(&br_cfg->header_,
916 &ovsrec_bridge_col_other_config, key);
920 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
921 struct iface **hw_addr_iface)
927 *hw_addr_iface = NULL;
929 /* Did the user request a particular MAC? */
930 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
931 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
932 if (eth_addr_is_multicast(ea)) {
933 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
934 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
935 } else if (eth_addr_is_zero(ea)) {
936 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
942 /* Otherwise choose the minimum non-local MAC address among all of the
944 memset(ea, 0xff, sizeof ea);
945 for (i = 0; i < br->n_ports; i++) {
946 struct port *port = br->ports[i];
947 uint8_t iface_ea[ETH_ADDR_LEN];
950 /* Mirror output ports don't participate. */
951 if (port->is_mirror_output_port) {
955 /* Choose the MAC address to represent the port. */
956 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
957 /* Find the interface with this Ethernet address (if any) so that
958 * we can provide the correct devname to the caller. */
960 for (j = 0; j < port->n_ifaces; j++) {
961 struct iface *candidate = port->ifaces[j];
962 uint8_t candidate_ea[ETH_ADDR_LEN];
963 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
964 && eth_addr_equals(iface_ea, candidate_ea)) {
969 /* Choose the interface whose MAC address will represent the port.
970 * The Linux kernel bonding code always chooses the MAC address of
971 * the first slave added to a bond, and the Fedora networking
972 * scripts always add slaves to a bond in alphabetical order, so
973 * for compatibility we choose the interface with the name that is
974 * first in alphabetical order. */
975 iface = port->ifaces[0];
976 for (j = 1; j < port->n_ifaces; j++) {
977 struct iface *candidate = port->ifaces[j];
978 if (strcmp(candidate->name, iface->name) < 0) {
983 /* The local port doesn't count (since we're trying to choose its
984 * MAC address anyway). */
985 if (iface->dp_ifidx == ODPP_LOCAL) {
990 error = netdev_get_etheraddr(iface->netdev, iface_ea);
992 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
993 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
994 iface->name, strerror(error));
999 /* Compare against our current choice. */
1000 if (!eth_addr_is_multicast(iface_ea) &&
1001 !eth_addr_is_local(iface_ea) &&
1002 !eth_addr_is_reserved(iface_ea) &&
1003 !eth_addr_is_zero(iface_ea) &&
1004 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
1006 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1007 *hw_addr_iface = iface;
1010 if (eth_addr_is_multicast(ea)) {
1011 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1012 *hw_addr_iface = NULL;
1013 VLOG_WARN("bridge %s: using default bridge Ethernet "
1014 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1016 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1017 br->name, ETH_ADDR_ARGS(ea));
1021 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1022 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1023 * an interface on 'br', then that interface must be passed in as
1024 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1025 * 'hw_addr_iface' must be passed in as a null pointer. */
1027 bridge_pick_datapath_id(struct bridge *br,
1028 const uint8_t bridge_ea[ETH_ADDR_LEN],
1029 struct iface *hw_addr_iface)
1032 * The procedure for choosing a bridge MAC address will, in the most
1033 * ordinary case, also choose a unique MAC that we can use as a datapath
1034 * ID. In some special cases, though, multiple bridges will end up with
1035 * the same MAC address. This is OK for the bridges, but it will confuse
1036 * the OpenFlow controller, because each datapath needs a unique datapath
1039 * Datapath IDs must be unique. It is also very desirable that they be
1040 * stable from one run to the next, so that policy set on a datapath
1043 const char *datapath_id;
1046 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1047 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1051 if (hw_addr_iface) {
1053 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1055 * A bridge whose MAC address is taken from a VLAN network device
1056 * (that is, a network device created with vconfig(8) or similar
1057 * tool) will have the same MAC address as a bridge on the VLAN
1058 * device's physical network device.
1060 * Handle this case by hashing the physical network device MAC
1061 * along with the VLAN identifier.
1063 uint8_t buf[ETH_ADDR_LEN + 2];
1064 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1065 buf[ETH_ADDR_LEN] = vlan >> 8;
1066 buf[ETH_ADDR_LEN + 1] = vlan;
1067 return dpid_from_hash(buf, sizeof buf);
1070 * Assume that this bridge's MAC address is unique, since it
1071 * doesn't fit any of the cases we handle specially.
1076 * A purely internal bridge, that is, one that has no non-virtual
1077 * network devices on it at all, is more difficult because it has no
1078 * natural unique identifier at all.
1080 * When the host is a XenServer, we handle this case by hashing the
1081 * host's UUID with the name of the bridge. Names of bridges are
1082 * persistent across XenServer reboots, although they can be reused if
1083 * an internal network is destroyed and then a new one is later
1084 * created, so this is fairly effective.
1086 * When the host is not a XenServer, we punt by using a random MAC
1087 * address on each run.
1089 const char *host_uuid = xenserver_get_host_uuid();
1091 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1092 dpid = dpid_from_hash(combined, strlen(combined));
1098 return eth_addr_to_uint64(bridge_ea);
1102 dpid_from_hash(const void *data, size_t n)
1104 uint8_t hash[SHA1_DIGEST_SIZE];
1106 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1107 sha1_bytes(data, n, hash);
1108 eth_addr_mark_random(hash);
1109 return eth_addr_to_uint64(hash);
1113 iface_refresh_tunnel_egress(struct iface *iface)
1115 const char *name = netdev_get_tnl_iface(iface->netdev);
1117 ovsrec_interface_set_tunnel_egress_iface(iface->cfg, name);
1121 iface_refresh_cfm_stats(struct iface *iface)
1125 const struct ovsrec_monitor *mon;
1127 mon = iface->cfg->monitor;
1134 for (i = 0; i < mon->n_remote_mps; i++) {
1135 const struct ovsrec_maintenance_point *mp;
1136 const struct remote_mp *rmp;
1138 mp = mon->remote_mps[i];
1139 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1141 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1144 if (hmap_is_empty(&cfm->x_remote_mps)) {
1145 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1148 struct remote_mp *rmp;
1149 int64_t *x_remote_mps;
1151 length = hmap_count(&cfm->x_remote_mps);
1152 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1155 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1156 x_remote_mps[i++] = rmp->mpid;
1159 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1163 if (hmap_is_empty(&cfm->x_remote_maids)) {
1164 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1167 char **x_remote_maids;
1168 struct remote_maid *rmaid;
1170 length = hmap_count(&cfm->x_remote_maids);
1171 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1174 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1177 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1179 for (j = 0; j < CCM_MAID_LEN; j++) {
1180 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1185 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1187 for (i = 0; i < length; i++) {
1188 free(x_remote_maids[i]);
1190 free(x_remote_maids);
1193 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1197 iface_refresh_stats(struct iface *iface)
1203 static const struct iface_stat iface_stats[] = {
1204 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1205 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1206 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1207 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1208 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1209 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1210 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1211 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1212 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1213 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1214 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1215 { "collisions", offsetof(struct netdev_stats, collisions) },
1217 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1218 const struct iface_stat *s;
1220 char *keys[N_STATS];
1221 int64_t values[N_STATS];
1224 struct netdev_stats stats;
1226 /* Intentionally ignore return value, since errors will set 'stats' to
1227 * all-1s, and we will deal with that correctly below. */
1228 netdev_get_stats(iface->netdev, &stats);
1231 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1232 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1233 if (value != UINT64_MAX) {
1240 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1244 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1246 struct ovsdb_datum datum;
1250 get_system_stats(&stats);
1252 ovsdb_datum_from_shash(&datum, &stats);
1253 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1260 const struct ovsrec_open_vswitch *cfg;
1262 bool datapath_destroyed;
1263 bool database_changed;
1266 /* Let each bridge do the work that it needs to do. */
1267 datapath_destroyed = false;
1268 LIST_FOR_EACH (br, node, &all_bridges) {
1269 int error = bridge_run_one(br);
1271 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1272 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1273 "forcing reconfiguration", br->name);
1274 datapath_destroyed = true;
1278 /* (Re)configure if necessary. */
1279 database_changed = ovsdb_idl_run(idl);
1280 cfg = ovsrec_open_vswitch_first(idl);
1281 if (database_changed || datapath_destroyed) {
1283 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1285 bridge_configure_once(cfg);
1286 bridge_reconfigure(cfg);
1288 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1289 ovsdb_idl_txn_commit(txn);
1290 ovsdb_idl_txn_destroy(txn); /* XXX */
1292 /* We still need to reconfigure to avoid dangling pointers to
1293 * now-destroyed ovsrec structures inside bridge data. */
1294 static const struct ovsrec_open_vswitch null_cfg;
1296 bridge_reconfigure(&null_cfg);
1301 /* Re-configure SSL. We do this on every trip through the main loop,
1302 * instead of just when the database changes, because the contents of the
1303 * key and certificate files can change without the database changing. */
1304 if (cfg && cfg->ssl) {
1305 const struct ovsrec_ssl *ssl = cfg->ssl;
1307 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1308 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1312 /* Refresh system and interface stats if necessary. */
1313 if (time_msec() >= stats_timer) {
1315 struct ovsdb_idl_txn *txn;
1317 txn = ovsdb_idl_txn_create(idl);
1318 LIST_FOR_EACH (br, node, &all_bridges) {
1321 for (i = 0; i < br->n_ports; i++) {
1322 struct port *port = br->ports[i];
1325 for (j = 0; j < port->n_ifaces; j++) {
1326 struct iface *iface = port->ifaces[j];
1327 iface_refresh_stats(iface);
1328 iface_refresh_cfm_stats(iface);
1329 iface_refresh_tunnel_egress(iface);
1333 refresh_system_stats(cfg);
1334 ovsdb_idl_txn_commit(txn);
1335 ovsdb_idl_txn_destroy(txn); /* XXX */
1338 stats_timer = time_msec() + STATS_INTERVAL;
1346 struct iface *iface;
1348 LIST_FOR_EACH (br, node, &all_bridges) {
1349 ofproto_wait(br->ofproto);
1350 if (ofproto_has_primary_controller(br->ofproto)) {
1354 mac_learning_wait(br->ml);
1357 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1359 cfm_wait(iface->cfm);
1363 ovsdb_idl_wait(idl);
1364 poll_timer_wait_until(stats_timer);
1367 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1368 * configuration changes. */
1370 bridge_flush(struct bridge *br)
1372 COVERAGE_INC(bridge_flush);
1374 mac_learning_flush(br->ml);
1377 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1378 * such interface. */
1379 static struct iface *
1380 bridge_get_local_iface(struct bridge *br)
1384 for (i = 0; i < br->n_ports; i++) {
1385 struct port *port = br->ports[i];
1386 for (j = 0; j < port->n_ifaces; j++) {
1387 struct iface *iface = port->ifaces[j];
1388 if (iface->dp_ifidx == ODPP_LOCAL) {
1397 /* Bridge unixctl user interface functions. */
1399 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1400 const char *args, void *aux OVS_UNUSED)
1402 struct ds ds = DS_EMPTY_INITIALIZER;
1403 const struct bridge *br;
1404 const struct mac_entry *e;
1406 br = bridge_lookup(args);
1408 unixctl_command_reply(conn, 501, "no such bridge");
1412 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1413 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1414 if (e->port < 0 || e->port >= br->n_ports) {
1417 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1418 br->ports[e->port]->ifaces[0]->dp_ifidx,
1419 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1421 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1425 /* Bridge reconfiguration functions. */
1426 static struct bridge *
1427 bridge_create(const struct ovsrec_bridge *br_cfg)
1432 assert(!bridge_lookup(br_cfg->name));
1433 br = xzalloc(sizeof *br);
1435 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1441 dpif_flow_flush(br->dpif);
1443 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1446 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1448 dpif_delete(br->dpif);
1449 dpif_close(br->dpif);
1454 br->name = xstrdup(br_cfg->name);
1456 br->ml = mac_learning_create();
1457 eth_addr_nicira_random(br->default_ea);
1459 hmap_init(&br->ifaces);
1461 shash_init(&br->port_by_name);
1462 shash_init(&br->iface_by_name);
1466 list_push_back(&all_bridges, &br->node);
1468 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1474 bridge_destroy(struct bridge *br)
1479 while (br->n_ports > 0) {
1480 port_destroy(br->ports[br->n_ports - 1]);
1482 list_remove(&br->node);
1483 error = dpif_delete(br->dpif);
1484 if (error && error != ENOENT) {
1485 VLOG_ERR("failed to delete %s: %s",
1486 dpif_name(br->dpif), strerror(error));
1488 dpif_close(br->dpif);
1489 ofproto_destroy(br->ofproto);
1490 mac_learning_destroy(br->ml);
1491 hmap_destroy(&br->ifaces);
1492 shash_destroy(&br->port_by_name);
1493 shash_destroy(&br->iface_by_name);
1500 static struct bridge *
1501 bridge_lookup(const char *name)
1505 LIST_FOR_EACH (br, node, &all_bridges) {
1506 if (!strcmp(br->name, name)) {
1513 /* Handle requests for a listing of all flows known by the OpenFlow
1514 * stack, including those normally hidden. */
1516 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1517 const char *args, void *aux OVS_UNUSED)
1522 br = bridge_lookup(args);
1524 unixctl_command_reply(conn, 501, "Unknown bridge");
1529 ofproto_get_all_flows(br->ofproto, &results);
1531 unixctl_command_reply(conn, 200, ds_cstr(&results));
1532 ds_destroy(&results);
1535 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1536 * connections and reconnect. If BRIDGE is not specified, then all bridges
1537 * drop their controller connections and reconnect. */
1539 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1540 const char *args, void *aux OVS_UNUSED)
1543 if (args[0] != '\0') {
1544 br = bridge_lookup(args);
1546 unixctl_command_reply(conn, 501, "Unknown bridge");
1549 ofproto_reconnect_controllers(br->ofproto);
1551 LIST_FOR_EACH (br, node, &all_bridges) {
1552 ofproto_reconnect_controllers(br->ofproto);
1555 unixctl_command_reply(conn, 200, NULL);
1559 bridge_run_one(struct bridge *br)
1562 struct iface *iface;
1564 error = ofproto_run1(br->ofproto);
1569 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1572 error = ofproto_run2(br->ofproto, br->flush);
1575 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1576 struct ofpbuf *packet;
1582 packet = cfm_run(iface->cfm);
1584 iface_send_packet(iface, packet);
1585 ofpbuf_uninit(packet);
1594 bridge_get_controllers(const struct bridge *br,
1595 struct ovsrec_controller ***controllersp)
1597 struct ovsrec_controller **controllers;
1598 size_t n_controllers;
1600 controllers = br->cfg->controller;
1601 n_controllers = br->cfg->n_controller;
1603 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1609 *controllersp = controllers;
1611 return n_controllers;
1615 bridge_reconfigure_one(struct bridge *br)
1617 struct shash old_ports, new_ports;
1618 struct svec snoops, old_snoops;
1619 struct shash_node *node;
1620 enum ofproto_fail_mode fail_mode;
1623 /* Collect old ports. */
1624 shash_init(&old_ports);
1625 for (i = 0; i < br->n_ports; i++) {
1626 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1629 /* Collect new ports. */
1630 shash_init(&new_ports);
1631 for (i = 0; i < br->cfg->n_ports; i++) {
1632 const char *name = br->cfg->ports[i]->name;
1633 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1634 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1639 /* If we have a controller, then we need a local port. Complain if the
1640 * user didn't specify one.
1642 * XXX perhaps we should synthesize a port ourselves in this case. */
1643 if (bridge_get_controllers(br, NULL)) {
1644 char local_name[IF_NAMESIZE];
1647 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1648 local_name, sizeof local_name);
1649 if (!error && !shash_find(&new_ports, local_name)) {
1650 VLOG_WARN("bridge %s: controller specified but no local port "
1651 "(port named %s) defined",
1652 br->name, local_name);
1656 /* Get rid of deleted ports.
1657 * Get rid of deleted interfaces on ports that still exist. */
1658 SHASH_FOR_EACH (node, &old_ports) {
1659 struct port *port = node->data;
1660 const struct ovsrec_port *port_cfg;
1662 port_cfg = shash_find_data(&new_ports, node->name);
1666 port_del_ifaces(port, port_cfg);
1670 /* Create new ports.
1671 * Add new interfaces to existing ports.
1672 * Reconfigure existing ports. */
1673 SHASH_FOR_EACH (node, &new_ports) {
1674 struct port *port = shash_find_data(&old_ports, node->name);
1676 port = port_create(br, node->name);
1679 port_reconfigure(port, node->data);
1680 if (!port->n_ifaces) {
1681 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1682 br->name, port->name);
1686 shash_destroy(&old_ports);
1687 shash_destroy(&new_ports);
1689 /* Set the fail-mode */
1690 fail_mode = !br->cfg->fail_mode
1691 || !strcmp(br->cfg->fail_mode, "standalone")
1692 ? OFPROTO_FAIL_STANDALONE
1693 : OFPROTO_FAIL_SECURE;
1694 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1695 && !ofproto_has_primary_controller(br->ofproto)) {
1696 ofproto_flush_flows(br->ofproto);
1698 ofproto_set_fail_mode(br->ofproto, fail_mode);
1700 /* Delete all flows if we're switching from connected to standalone or vice
1701 * versa. (XXX Should we delete all flows if we are switching from one
1702 * controller to another?) */
1704 /* Configure OpenFlow controller connection snooping. */
1706 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1707 ovs_rundir(), br->name));
1708 svec_init(&old_snoops);
1709 ofproto_get_snoops(br->ofproto, &old_snoops);
1710 if (!svec_equal(&snoops, &old_snoops)) {
1711 ofproto_set_snoops(br->ofproto, &snoops);
1713 svec_destroy(&snoops);
1714 svec_destroy(&old_snoops);
1716 mirror_reconfigure(br);
1719 /* Initializes 'oc' appropriately as a management service controller for
1722 * The caller must free oc->target when it is no longer needed. */
1724 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1725 struct ofproto_controller *oc)
1727 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1728 oc->max_backoff = 0;
1729 oc->probe_interval = 60;
1730 oc->band = OFPROTO_OUT_OF_BAND;
1731 oc->accept_re = NULL;
1732 oc->update_resolv_conf = false;
1734 oc->burst_limit = 0;
1737 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1739 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1740 struct ofproto_controller *oc)
1742 oc->target = c->target;
1743 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1744 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1745 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1746 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1747 oc->accept_re = c->discover_accept_regex;
1748 oc->update_resolv_conf = c->discover_update_resolv_conf;
1749 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1750 oc->burst_limit = (c->controller_burst_limit
1751 ? *c->controller_burst_limit : 0);
1754 /* Configures the IP stack for 'br''s local interface properly according to the
1755 * configuration in 'c'. */
1757 bridge_configure_local_iface_netdev(struct bridge *br,
1758 struct ovsrec_controller *c)
1760 struct netdev *netdev;
1761 struct in_addr mask, gateway;
1763 struct iface *local_iface;
1766 /* Controller discovery does its own TCP/IP configuration later. */
1767 if (strcmp(c->target, "discover")) {
1771 /* If there's no local interface or no IP address, give up. */
1772 local_iface = bridge_get_local_iface(br);
1773 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1777 /* Bring up the local interface. */
1778 netdev = local_iface->netdev;
1779 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1781 /* Configure the IP address and netmask. */
1782 if (!c->local_netmask
1783 || !inet_aton(c->local_netmask, &mask)
1785 mask.s_addr = guess_netmask(ip.s_addr);
1787 if (!netdev_set_in4(netdev, ip, mask)) {
1788 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1789 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1792 /* Configure the default gateway. */
1793 if (c->local_gateway
1794 && inet_aton(c->local_gateway, &gateway)
1795 && gateway.s_addr) {
1796 if (!netdev_add_router(netdev, gateway)) {
1797 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1798 br->name, IP_ARGS(&gateway.s_addr));
1804 bridge_reconfigure_remotes(struct bridge *br,
1805 const struct sockaddr_in *managers,
1808 const char *disable_ib_str, *queue_id_str;
1809 bool disable_in_band = false;
1812 struct ovsrec_controller **controllers;
1813 size_t n_controllers;
1816 struct ofproto_controller *ocs;
1820 /* Check if we should disable in-band control on this bridge. */
1821 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1822 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1823 disable_in_band = true;
1826 /* Set OpenFlow queue ID for in-band control. */
1827 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
1828 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
1829 ofproto_set_in_band_queue(br->ofproto, queue_id);
1831 if (disable_in_band) {
1832 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1834 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1836 had_primary = ofproto_has_primary_controller(br->ofproto);
1838 n_controllers = bridge_get_controllers(br, &controllers);
1840 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1843 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1844 for (i = 0; i < n_controllers; i++) {
1845 struct ovsrec_controller *c = controllers[i];
1847 if (!strncmp(c->target, "punix:", 6)
1848 || !strncmp(c->target, "unix:", 5)) {
1849 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1851 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1852 * domain sockets and overwriting arbitrary local files. */
1853 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1854 "\"%s\" due to possibility for remote exploit",
1855 dpif_name(br->dpif), c->target);
1859 bridge_configure_local_iface_netdev(br, c);
1860 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
1861 if (disable_in_band) {
1862 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
1867 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1868 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1871 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1872 ofproto_flush_flows(br->ofproto);
1875 /* If there are no controllers and the bridge is in standalone
1876 * mode, set up a flow that matches every packet and directs
1877 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1878 * switch is in secure mode and we won't pass any traffic until
1879 * a controller has been defined and it tells us to do so. */
1881 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1882 union ofp_action action;
1883 struct cls_rule rule;
1885 memset(&action, 0, sizeof action);
1886 action.type = htons(OFPAT_OUTPUT);
1887 action.output.len = htons(sizeof action);
1888 action.output.port = htons(OFPP_NORMAL);
1889 cls_rule_init_catchall(&rule, 0);
1890 ofproto_add_flow(br->ofproto, &rule, &action, 1);
1895 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1900 for (i = 0; i < br->n_ports; i++) {
1901 struct port *port = br->ports[i];
1902 for (j = 0; j < port->n_ifaces; j++) {
1903 struct iface *iface = port->ifaces[j];
1904 shash_add_once(ifaces, iface->name, iface);
1906 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1907 shash_add_once(ifaces, port->name, NULL);
1912 /* For robustness, in case the administrator moves around datapath ports behind
1913 * our back, we re-check all the datapath port numbers here.
1915 * This function will set the 'dp_ifidx' members of interfaces that have
1916 * disappeared to -1, so only call this function from a context where those
1917 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1918 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1919 * datapath, which doesn't support UINT16_MAX+1 ports. */
1921 bridge_fetch_dp_ifaces(struct bridge *br)
1923 struct odp_port *dpif_ports;
1924 size_t n_dpif_ports;
1927 /* Reset all interface numbers. */
1928 for (i = 0; i < br->n_ports; i++) {
1929 struct port *port = br->ports[i];
1930 for (j = 0; j < port->n_ifaces; j++) {
1931 struct iface *iface = port->ifaces[j];
1932 iface->dp_ifidx = -1;
1935 hmap_clear(&br->ifaces);
1937 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1938 for (i = 0; i < n_dpif_ports; i++) {
1939 struct odp_port *p = &dpif_ports[i];
1940 struct iface *iface = iface_lookup(br, p->devname);
1942 if (iface->dp_ifidx >= 0) {
1943 VLOG_WARN("%s reported interface %s twice",
1944 dpif_name(br->dpif), p->devname);
1945 } else if (iface_from_dp_ifidx(br, p->port)) {
1946 VLOG_WARN("%s reported interface %"PRIu16" twice",
1947 dpif_name(br->dpif), p->port);
1949 iface->dp_ifidx = p->port;
1950 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
1951 hash_int(iface->dp_ifidx, 0));
1954 iface_set_ofport(iface->cfg,
1955 (iface->dp_ifidx >= 0
1956 ? odp_port_to_ofp_port(iface->dp_ifidx)
1963 /* Bridge packet processing functions. */
1966 bond_hash(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
1968 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
1971 static struct bond_entry *
1972 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN],
1975 assert(port->bond_type == BT_SLB);
1976 return &port->bond_hash[bond_hash(mac, vlan)];
1980 bond_choose_iface(const struct port *port)
1982 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1983 size_t i, best_down_slave = -1;
1984 long long next_delay_expiration = LLONG_MAX;
1986 for (i = 0; i < port->n_ifaces; i++) {
1987 struct iface *iface = port->ifaces[i];
1989 if (iface->enabled) {
1991 } else if (iface->delay_expires < next_delay_expiration) {
1992 best_down_slave = i;
1993 next_delay_expiration = iface->delay_expires;
1997 if (best_down_slave != -1) {
1998 struct iface *iface = port->ifaces[best_down_slave];
2000 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2001 "since no other interface is up", iface->name,
2002 iface->delay_expires - time_msec());
2003 bond_enable_slave(iface, true);
2006 return best_down_slave;
2010 choose_output_iface(const struct port *port, const uint8_t *dl_src,
2011 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2013 struct iface *iface;
2015 assert(port->n_ifaces);
2016 if (port->n_ifaces == 1) {
2017 iface = port->ifaces[0];
2018 } else if (port->bond_type == BT_AB) {
2019 if (port->active_iface < 0) {
2020 *tags |= port->no_ifaces_tag;
2023 iface = port->ifaces[port->active_iface];
2024 } else if (port->bond_type == BT_SLB){
2025 struct bond_entry *e = lookup_bond_entry(port, dl_src, vlan);
2026 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2027 || !port->ifaces[e->iface_idx]->enabled) {
2028 /* XXX select interface properly. The current interface selection
2029 * is only good for testing the rebalancing code. */
2030 e->iface_idx = bond_choose_iface(port);
2031 if (e->iface_idx < 0) {
2032 *tags |= port->no_ifaces_tag;
2035 e->iface_tag = tag_create_random();
2036 ((struct port *) port)->bond_compat_is_stale = true;
2038 *tags |= e->iface_tag;
2039 iface = port->ifaces[e->iface_idx];
2043 *dp_ifidx = iface->dp_ifidx;
2044 *tags |= iface->tag; /* Currently only used for bonding. */
2049 bond_link_status_update(struct iface *iface, bool carrier)
2051 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2052 struct port *port = iface->port;
2054 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2055 /* Nothing to do. */
2058 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
2059 iface->name, carrier ? "detected" : "dropped");
2060 if (carrier == iface->enabled) {
2061 iface->delay_expires = LLONG_MAX;
2062 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2063 iface->name, carrier ? "disabled" : "enabled");
2064 } else if (carrier && port->active_iface < 0) {
2065 bond_enable_slave(iface, true);
2066 if (port->updelay) {
2067 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2068 "other interface is up", iface->name, port->updelay);
2071 int delay = carrier ? port->updelay : port->downdelay;
2072 iface->delay_expires = time_msec() + delay;
2075 "interface %s: will be %s if it stays %s for %d ms",
2077 carrier ? "enabled" : "disabled",
2078 carrier ? "up" : "down",
2085 bond_choose_active_iface(struct port *port)
2087 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2089 port->active_iface = bond_choose_iface(port);
2090 port->active_iface_tag = tag_create_random();
2091 if (port->active_iface >= 0) {
2092 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2093 port->name, port->ifaces[port->active_iface]->name);
2095 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2101 bond_enable_slave(struct iface *iface, bool enable)
2103 struct port *port = iface->port;
2104 struct bridge *br = port->bridge;
2106 /* This acts as a recursion check. If the act of disabling a slave
2107 * causes a different slave to be enabled, the flag will allow us to
2108 * skip redundant work when we reenter this function. It must be
2109 * cleared on exit to keep things safe with multiple bonds. */
2110 static bool moving_active_iface = false;
2112 iface->delay_expires = LLONG_MAX;
2113 if (enable == iface->enabled) {
2117 iface->enabled = enable;
2118 if (!iface->enabled) {
2119 VLOG_WARN("interface %s: disabled", iface->name);
2120 ofproto_revalidate(br->ofproto, iface->tag);
2121 if (iface->port_ifidx == port->active_iface) {
2122 ofproto_revalidate(br->ofproto,
2123 port->active_iface_tag);
2125 /* Disabling a slave can lead to another slave being immediately
2126 * enabled if there will be no active slaves but one is waiting
2127 * on an updelay. In this case we do not need to run most of the
2128 * code for the newly enabled slave since there was no period
2129 * without an active slave and it is redundant with the disabling
2131 moving_active_iface = true;
2132 bond_choose_active_iface(port);
2134 bond_send_learning_packets(port);
2136 VLOG_WARN("interface %s: enabled", iface->name);
2137 if (port->active_iface < 0 && !moving_active_iface) {
2138 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2139 bond_choose_active_iface(port);
2140 bond_send_learning_packets(port);
2142 iface->tag = tag_create_random();
2145 moving_active_iface = false;
2146 port->bond_compat_is_stale = true;
2149 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2150 * bond interface. */
2152 bond_update_fake_iface_stats(struct port *port)
2154 struct netdev_stats bond_stats;
2155 struct netdev *bond_dev;
2158 memset(&bond_stats, 0, sizeof bond_stats);
2160 for (i = 0; i < port->n_ifaces; i++) {
2161 struct netdev_stats slave_stats;
2163 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2164 /* XXX: We swap the stats here because they are swapped back when
2165 * reported by the internal device. The reason for this is
2166 * internal devices normally represent packets going into the system
2167 * but when used as fake bond device they represent packets leaving
2168 * the system. We really should do this in the internal device
2169 * itself because changing it here reverses the counts from the
2170 * perspective of the switch. However, the internal device doesn't
2171 * know what type of device it represents so we have to do it here
2173 bond_stats.tx_packets += slave_stats.rx_packets;
2174 bond_stats.tx_bytes += slave_stats.rx_bytes;
2175 bond_stats.rx_packets += slave_stats.tx_packets;
2176 bond_stats.rx_bytes += slave_stats.tx_bytes;
2180 if (!netdev_open_default(port->name, &bond_dev)) {
2181 netdev_set_stats(bond_dev, &bond_stats);
2182 netdev_close(bond_dev);
2187 bond_run(struct bridge *br)
2191 for (i = 0; i < br->n_ports; i++) {
2192 struct port *port = br->ports[i];
2194 if (port->n_ifaces >= 2) {
2197 /* Track carrier going up and down on interfaces. */
2198 while (!netdev_monitor_poll(port->monitor, &devname)) {
2199 struct iface *iface;
2201 iface = port_lookup_iface(port, devname);
2203 bool carrier = netdev_get_carrier(iface->netdev);
2205 bond_link_status_update(iface, carrier);
2206 port_update_bond_compat(port);
2211 for (j = 0; j < port->n_ifaces; j++) {
2212 struct iface *iface = port->ifaces[j];
2213 if (time_msec() >= iface->delay_expires) {
2214 bond_enable_slave(iface, !iface->enabled);
2218 if (port->bond_fake_iface
2219 && time_msec() >= port->bond_next_fake_iface_update) {
2220 bond_update_fake_iface_stats(port);
2221 port->bond_next_fake_iface_update = time_msec() + 1000;
2225 if (port->bond_compat_is_stale) {
2226 port->bond_compat_is_stale = false;
2227 port_update_bond_compat(port);
2233 bond_wait(struct bridge *br)
2237 for (i = 0; i < br->n_ports; i++) {
2238 struct port *port = br->ports[i];
2239 if (port->n_ifaces < 2) {
2242 netdev_monitor_poll_wait(port->monitor);
2243 for (j = 0; j < port->n_ifaces; j++) {
2244 struct iface *iface = port->ifaces[j];
2245 if (iface->delay_expires != LLONG_MAX) {
2246 poll_timer_wait_until(iface->delay_expires);
2249 if (port->bond_fake_iface) {
2250 poll_timer_wait_until(port->bond_next_fake_iface_update);
2256 set_dst(struct dst *dst, const struct flow *flow,
2257 const struct port *in_port, const struct port *out_port,
2260 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2261 : in_port->vlan >= 0 ? in_port->vlan
2262 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2263 : vlan_tci_to_vid(flow->vlan_tci));
2264 return choose_output_iface(out_port, flow->dl_src, dst->vlan,
2265 &dst->dp_ifidx, tags);
2269 swap_dst(struct dst *p, struct dst *q)
2271 struct dst tmp = *p;
2276 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2277 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2278 * that we push to the datapath. We could in fact fully sort the array by
2279 * vlan, but in most cases there are at most two different vlan tags so that's
2280 * possibly overkill.) */
2282 partition_dsts(struct dst_set *set, int vlan)
2284 struct dst *first = set->dsts;
2285 struct dst *last = set->dsts + set->n;
2287 while (first != last) {
2289 * - All dsts < first have vlan == 'vlan'.
2290 * - All dsts >= last have vlan != 'vlan'.
2291 * - first < last. */
2292 while (first->vlan == vlan) {
2293 if (++first == last) {
2298 /* Same invariants, plus one additional:
2299 * - first->vlan != vlan.
2301 while (last[-1].vlan != vlan) {
2302 if (--last == first) {
2307 /* Same invariants, plus one additional:
2308 * - last[-1].vlan == vlan.*/
2309 swap_dst(first++, --last);
2314 mirror_mask_ffs(mirror_mask_t mask)
2316 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2321 dst_set_init(struct dst_set *set)
2323 set->dsts = set->builtin;
2325 set->allocated = ARRAY_SIZE(set->builtin);
2329 dst_set_add(struct dst_set *set, const struct dst *dst)
2331 if (set->n >= set->allocated) {
2332 size_t new_allocated;
2333 struct dst *new_dsts;
2335 new_allocated = set->allocated * 2;
2336 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2337 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2341 set->dsts = new_dsts;
2342 set->allocated = new_allocated;
2344 set->dsts[set->n++] = *dst;
2348 dst_set_free(struct dst_set *set)
2350 if (set->dsts != set->builtin) {
2356 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2359 for (i = 0; i < set->n; i++) {
2360 if (set->dsts[i].vlan == test->vlan
2361 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2369 port_trunks_vlan(const struct port *port, uint16_t vlan)
2371 return (port->vlan < 0
2372 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2376 port_includes_vlan(const struct port *port, uint16_t vlan)
2378 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2382 port_is_floodable(const struct port *port)
2386 for (i = 0; i < port->n_ifaces; i++) {
2387 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2388 port->ifaces[i]->dp_ifidx)) {
2396 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2397 const struct port *in_port, const struct port *out_port,
2398 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2400 mirror_mask_t mirrors = in_port->src_mirrors;
2405 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2406 if (flow_vlan == 0) {
2407 flow_vlan = OFP_VLAN_NONE;
2410 if (out_port == FLOOD_PORT) {
2411 for (i = 0; i < br->n_ports; i++) {
2412 struct port *port = br->ports[i];
2414 && port_is_floodable(port)
2415 && port_includes_vlan(port, vlan)
2416 && !port->is_mirror_output_port
2417 && set_dst(&dst, flow, in_port, port, tags)) {
2418 mirrors |= port->dst_mirrors;
2419 dst_set_add(set, &dst);
2422 *nf_output_iface = NF_OUT_FLOOD;
2423 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2424 dst_set_add(set, &dst);
2425 *nf_output_iface = dst.dp_ifidx;
2426 mirrors |= out_port->dst_mirrors;
2430 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2431 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2433 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2434 && !dst_is_duplicate(set, &dst)) {
2435 dst_set_add(set, &dst);
2438 for (i = 0; i < br->n_ports; i++) {
2439 struct port *port = br->ports[i];
2440 if (port_includes_vlan(port, m->out_vlan)
2441 && set_dst(&dst, flow, in_port, port, tags))
2443 if (port->vlan < 0) {
2444 dst.vlan = m->out_vlan;
2446 if (dst_is_duplicate(set, &dst)) {
2450 /* Use the vlan tag on the original flow instead of
2451 * the one passed in the vlan parameter. This ensures
2452 * that we compare the vlan from before any implicit
2453 * tagging tags place. This is necessary because
2454 * dst->vlan is the final vlan, after removing implicit
2456 if (port == in_port && dst.vlan == flow_vlan) {
2457 /* Don't send out input port on same VLAN. */
2460 dst_set_add(set, &dst);
2465 mirrors &= mirrors - 1;
2468 partition_dsts(set, flow_vlan);
2471 static void OVS_UNUSED
2472 print_dsts(const struct dst_set *set)
2476 for (i = 0; i < set->n; i++) {
2477 const struct dst *dst = &set->dsts[i];
2479 printf(">p%"PRIu16, dst->dp_ifidx);
2480 if (dst->vlan != OFP_VLAN_NONE) {
2481 printf("v%"PRIu16, dst->vlan);
2487 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2488 const struct port *in_port, const struct port *out_port,
2489 tag_type *tags, struct ofpbuf *actions,
2490 uint16_t *nf_output_iface)
2497 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2500 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2501 if (cur_vlan == 0) {
2502 cur_vlan = OFP_VLAN_NONE;
2504 for (i = 0; i < set.n; i++) {
2505 const struct dst *dst = &set.dsts[i];
2506 if (dst->vlan != cur_vlan) {
2507 if (dst->vlan == OFP_VLAN_NONE) {
2508 nl_msg_put_flag(actions, ODPAT_STRIP_VLAN);
2511 tci = htons(dst->vlan & VLAN_VID_MASK);
2512 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2513 nl_msg_put_be16(actions, ODPAT_SET_DL_TCI, tci);
2515 cur_vlan = dst->vlan;
2517 nl_msg_put_u32(actions, ODPAT_OUTPUT, dst->dp_ifidx);
2522 /* Returns the effective vlan of a packet, taking into account both the
2523 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2524 * the packet is untagged and -1 indicates it has an invalid header and
2525 * should be dropped. */
2526 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2527 struct port *in_port, bool have_packet)
2529 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2530 if (in_port->vlan >= 0) {
2532 /* XXX support double tagging? */
2534 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2535 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2536 "packet received on port %s configured with "
2537 "implicit VLAN %"PRIu16,
2538 br->name, vlan, in_port->name, in_port->vlan);
2542 vlan = in_port->vlan;
2544 if (!port_includes_vlan(in_port, vlan)) {
2546 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2547 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2548 "packet received on port %s not configured for "
2550 br->name, vlan, in_port->name, vlan);
2559 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2560 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2561 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2563 is_gratuitous_arp(const struct flow *flow)
2565 return (flow->dl_type == htons(ETH_TYPE_ARP)
2566 && eth_addr_is_broadcast(flow->dl_dst)
2567 && (flow->nw_proto == ARP_OP_REPLY
2568 || (flow->nw_proto == ARP_OP_REQUEST
2569 && flow->nw_src == flow->nw_dst)));
2573 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2574 struct port *in_port)
2576 enum grat_arp_lock_type lock_type;
2579 /* We don't want to learn from gratuitous ARP packets that are reflected
2580 * back over bond slaves so we lock the learning table. */
2581 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2582 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2583 GRAT_ARP_LOCK_CHECK;
2585 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2588 /* The log messages here could actually be useful in debugging,
2589 * so keep the rate limit relatively high. */
2590 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2592 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2593 "on port %s in VLAN %d",
2594 br->name, ETH_ADDR_ARGS(flow->dl_src),
2595 in_port->name, vlan);
2596 ofproto_revalidate(br->ofproto, rev_tag);
2600 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2601 * dropped. Returns true if they may be forwarded, false if they should be
2604 * If 'have_packet' is true, it indicates that the caller is processing a
2605 * received packet. If 'have_packet' is false, then the caller is just
2606 * revalidating an existing flow because configuration has changed. Either
2607 * way, 'have_packet' only affects logging (there is no point in logging errors
2608 * during revalidation).
2610 * Sets '*in_portp' to the input port. This will be a null pointer if
2611 * flow->in_port does not designate a known input port (in which case
2612 * is_admissible() returns false).
2614 * When returning true, sets '*vlanp' to the effective VLAN of the input
2615 * packet, as returned by flow_get_vlan().
2617 * May also add tags to '*tags', although the current implementation only does
2618 * so in one special case.
2621 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2622 tag_type *tags, int *vlanp, struct port **in_portp)
2624 struct iface *in_iface;
2625 struct port *in_port;
2628 /* Find the interface and port structure for the received packet. */
2629 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2631 /* No interface? Something fishy... */
2633 /* Odd. A few possible reasons here:
2635 * - We deleted an interface but there are still a few packets
2636 * queued up from it.
2638 * - Someone externally added an interface (e.g. with "ovs-dpctl
2639 * add-if") that we don't know about.
2641 * - Packet arrived on the local port but the local port is not
2642 * one of our bridge ports.
2644 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2646 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2647 "interface %"PRIu16, br->name, flow->in_port);
2653 *in_portp = in_port = in_iface->port;
2654 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2659 /* Drop frames for reserved multicast addresses. */
2660 if (eth_addr_is_reserved(flow->dl_dst)) {
2664 /* Drop frames on ports reserved for mirroring. */
2665 if (in_port->is_mirror_output_port) {
2667 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2668 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2669 "%s, which is reserved exclusively for mirroring",
2670 br->name, in_port->name);
2675 /* Packets received on bonds need special attention to avoid duplicates. */
2676 if (in_port->n_ifaces > 1) {
2678 bool is_grat_arp_locked;
2680 if (eth_addr_is_multicast(flow->dl_dst)) {
2681 *tags |= in_port->active_iface_tag;
2682 if (in_port->active_iface != in_iface->port_ifidx) {
2683 /* Drop all multicast packets on inactive slaves. */
2688 /* Drop all packets for which we have learned a different input
2689 * port, because we probably sent the packet on one slave and got
2690 * it back on the other. Gratuitous ARP packets are an exception
2691 * to this rule: the host has moved to another switch. The exception
2692 * to the exception is if we locked the learning table to avoid
2693 * reflections on bond slaves. If this is the case, just drop the
2695 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2696 &is_grat_arp_locked);
2697 if (src_idx != -1 && src_idx != in_port->port_idx &&
2698 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2706 /* If the composed actions may be applied to any packet in the given 'flow',
2707 * returns true. Otherwise, the actions should only be applied to 'packet', or
2708 * not at all, if 'packet' was NULL. */
2710 process_flow(struct bridge *br, const struct flow *flow,
2711 const struct ofpbuf *packet, struct ofpbuf *actions,
2712 tag_type *tags, uint16_t *nf_output_iface)
2714 struct port *in_port;
2715 struct port *out_port;
2719 /* Check whether we should drop packets in this flow. */
2720 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2725 /* Learn source MAC (but don't try to learn from revalidation). */
2727 update_learning_table(br, flow, vlan, in_port);
2730 /* Determine output port. */
2731 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2733 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2734 out_port = br->ports[out_port_idx];
2735 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2736 /* If we are revalidating but don't have a learning entry then
2737 * eject the flow. Installing a flow that floods packets opens
2738 * up a window of time where we could learn from a packet reflected
2739 * on a bond and blackhole packets before the learning table is
2740 * updated to reflect the correct port. */
2743 out_port = FLOOD_PORT;
2746 /* Don't send packets out their input ports. */
2747 if (in_port == out_port) {
2753 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2761 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2762 struct ofpbuf *actions, tag_type *tags,
2763 uint16_t *nf_output_iface, void *br_)
2765 struct iface *iface;
2766 struct bridge *br = br_;
2768 COVERAGE_INC(bridge_process_flow);
2770 iface = iface_from_dp_ifidx(br, flow->in_port);
2772 if (cfm_should_process_flow(flow)) {
2773 if (packet && iface->cfm) {
2774 cfm_process_heartbeat(iface->cfm, packet);
2779 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2783 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
2784 const struct nlattr *actions,
2786 unsigned long long int n_bytes, void *br_)
2788 struct bridge *br = br_;
2789 const struct nlattr *a;
2790 struct port *in_port;
2795 /* Feed information from the active flows back into the learning table to
2796 * ensure that table is always in sync with what is actually flowing
2797 * through the datapath.
2799 * We test that 'tags' is nonzero to ensure that only flows that include an
2800 * OFPP_NORMAL action are used for learning. This works because
2801 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2802 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2803 update_learning_table(br, flow, vlan, in_port);
2806 /* Account for bond slave utilization. */
2807 if (!br->has_bonded_ports) {
2810 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
2811 if (nl_attr_type(a) == ODPAT_OUTPUT) {
2812 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
2813 if (out_port && out_port->n_ifaces >= 2 &&
2814 out_port->bond_type == BT_SLB) {
2815 uint16_t vlan = (flow->vlan_tci
2816 ? vlan_tci_to_vid(flow->vlan_tci)
2818 struct bond_entry *e = lookup_bond_entry(out_port,
2819 flow->dl_src, vlan);
2820 e->tx_bytes += n_bytes;
2827 bridge_account_checkpoint_ofhook_cb(void *br_)
2829 struct bridge *br = br_;
2833 if (!br->has_bonded_ports) {
2838 for (i = 0; i < br->n_ports; i++) {
2839 struct port *port = br->ports[i];
2840 if (port->n_ifaces > 1 && port->bond_type == BT_SLB
2841 && now >= port->bond_next_rebalance) {
2842 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2843 bond_rebalance_port(port);
2848 static struct ofhooks bridge_ofhooks = {
2849 bridge_normal_ofhook_cb,
2850 bridge_account_flow_ofhook_cb,
2851 bridge_account_checkpoint_ofhook_cb,
2854 /* Bonding functions. */
2856 /* Statistics for a single interface on a bonded port, used for load-based
2857 * bond rebalancing. */
2858 struct slave_balance {
2859 struct iface *iface; /* The interface. */
2860 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2862 /* All the "bond_entry"s that are assigned to this interface, in order of
2863 * increasing tx_bytes. */
2864 struct bond_entry **hashes;
2869 bond_type_to_string(enum bond_type bt) {
2870 static char *bt_slb = "slb";
2871 static char *bt_ab = "active-backup";
2874 case BT_SLB: return bt_slb;
2875 case BT_AB: return bt_ab;
2882 /* Sorts pointers to pointers to bond_entries in ascending order by the
2883 * interface to which they are assigned, and within a single interface in
2884 * ascending order of bytes transmitted. */
2886 compare_bond_entries(const void *a_, const void *b_)
2888 const struct bond_entry *const *ap = a_;
2889 const struct bond_entry *const *bp = b_;
2890 const struct bond_entry *a = *ap;
2891 const struct bond_entry *b = *bp;
2892 if (a->iface_idx != b->iface_idx) {
2893 return a->iface_idx > b->iface_idx ? 1 : -1;
2894 } else if (a->tx_bytes != b->tx_bytes) {
2895 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2901 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2902 * *descending* order by number of bytes transmitted. */
2904 compare_slave_balance(const void *a_, const void *b_)
2906 const struct slave_balance *a = a_;
2907 const struct slave_balance *b = b_;
2908 if (a->iface->enabled != b->iface->enabled) {
2909 return a->iface->enabled ? -1 : 1;
2910 } else if (a->tx_bytes != b->tx_bytes) {
2911 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2918 swap_bals(struct slave_balance *a, struct slave_balance *b)
2920 struct slave_balance tmp = *a;
2925 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2926 * given that 'p' (and only 'p') might be in the wrong location.
2928 * This function invalidates 'p', since it might now be in a different memory
2931 resort_bals(struct slave_balance *p,
2932 struct slave_balance bals[], size_t n_bals)
2935 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2936 swap_bals(p, p - 1);
2938 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2939 swap_bals(p, p + 1);
2945 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2947 if (VLOG_IS_DBG_ENABLED()) {
2948 struct ds ds = DS_EMPTY_INITIALIZER;
2949 const struct slave_balance *b;
2951 for (b = bals; b < bals + n_bals; b++) {
2955 ds_put_char(&ds, ',');
2957 ds_put_format(&ds, " %s %"PRIu64"kB",
2958 b->iface->name, b->tx_bytes / 1024);
2960 if (!b->iface->enabled) {
2961 ds_put_cstr(&ds, " (disabled)");
2963 if (b->n_hashes > 0) {
2964 ds_put_cstr(&ds, " (");
2965 for (i = 0; i < b->n_hashes; i++) {
2966 const struct bond_entry *e = b->hashes[i];
2968 ds_put_cstr(&ds, " + ");
2970 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2971 e - port->bond_hash, e->tx_bytes / 1024);
2973 ds_put_cstr(&ds, ")");
2976 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2981 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2983 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2986 struct bond_entry *hash = from->hashes[hash_idx];
2987 struct port *port = from->iface->port;
2988 uint64_t delta = hash->tx_bytes;
2990 assert(port->bond_type == BT_SLB);
2992 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2993 "from %s to %s (now carrying %"PRIu64"kB and "
2994 "%"PRIu64"kB load, respectively)",
2995 port->name, delta / 1024, hash - port->bond_hash,
2996 from->iface->name, to->iface->name,
2997 (from->tx_bytes - delta) / 1024,
2998 (to->tx_bytes + delta) / 1024);
3000 /* Delete element from from->hashes.
3002 * We don't bother to add the element to to->hashes because not only would
3003 * it require more work, the only purpose it would be to allow that hash to
3004 * be migrated to another slave in this rebalancing run, and there is no
3005 * point in doing that. */
3006 if (hash_idx == 0) {
3009 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3010 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3014 /* Shift load away from 'from' to 'to'. */
3015 from->tx_bytes -= delta;
3016 to->tx_bytes += delta;
3018 /* Arrange for flows to be revalidated. */
3019 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
3020 hash->iface_idx = to->iface->port_ifidx;
3021 hash->iface_tag = tag_create_random();
3025 bond_rebalance_port(struct port *port)
3027 struct slave_balance *bals;
3029 struct bond_entry *hashes[BOND_MASK + 1];
3030 struct slave_balance *b, *from, *to;
3031 struct bond_entry *e;
3034 assert(port->bond_type == BT_SLB);
3036 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3037 * descending order of tx_bytes, so that bals[0] represents the most
3038 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3041 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3042 * array for each slave_balance structure, we sort our local array of
3043 * hashes in order by slave, so that all of the hashes for a given slave
3044 * become contiguous in memory, and then we point each 'hashes' members of
3045 * a slave_balance structure to the start of a contiguous group. */
3046 n_bals = port->n_ifaces;
3047 bals = xmalloc(n_bals * sizeof *bals);
3048 for (b = bals; b < &bals[n_bals]; b++) {
3049 b->iface = port->ifaces[b - bals];
3054 for (i = 0; i <= BOND_MASK; i++) {
3055 hashes[i] = &port->bond_hash[i];
3057 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3058 for (i = 0; i <= BOND_MASK; i++) {
3060 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3061 b = &bals[e->iface_idx];
3062 b->tx_bytes += e->tx_bytes;
3064 b->hashes = &hashes[i];
3069 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3070 log_bals(bals, n_bals, port);
3072 /* Discard slaves that aren't enabled (which were sorted to the back of the
3073 * array earlier). */
3074 while (!bals[n_bals - 1].iface->enabled) {
3081 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3082 to = &bals[n_bals - 1];
3083 for (from = bals; from < to; ) {
3084 uint64_t overload = from->tx_bytes - to->tx_bytes;
3085 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3086 /* The extra load on 'from' (and all less-loaded slaves), compared
3087 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3088 * it is less than ~1Mbps. No point in rebalancing. */
3090 } else if (from->n_hashes == 1) {
3091 /* 'from' only carries a single MAC hash, so we can't shift any
3092 * load away from it, even though we want to. */
3095 /* 'from' is carrying significantly more load than 'to', and that
3096 * load is split across at least two different hashes. Pick a hash
3097 * to migrate to 'to' (the least-loaded slave), given that doing so
3098 * must decrease the ratio of the load on the two slaves by at
3101 * The sort order we use means that we prefer to shift away the
3102 * smallest hashes instead of the biggest ones. There is little
3103 * reason behind this decision; we could use the opposite sort
3104 * order to shift away big hashes ahead of small ones. */
3107 for (i = 0; i < from->n_hashes; i++) {
3108 double old_ratio, new_ratio;
3109 uint64_t delta = from->hashes[i]->tx_bytes;
3111 if (delta == 0 || from->tx_bytes - delta == 0) {
3112 /* Pointless move. */
3116 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3118 if (to->tx_bytes == 0) {
3119 /* Nothing on the new slave, move it. */
3123 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3124 new_ratio = (double)(from->tx_bytes - delta) /
3125 (to->tx_bytes + delta);
3127 if (new_ratio == 0) {
3128 /* Should already be covered but check to prevent division
3133 if (new_ratio < 1) {
3134 new_ratio = 1 / new_ratio;
3137 if (old_ratio - new_ratio > 0.1) {
3138 /* Would decrease the ratio, move it. */
3142 if (i < from->n_hashes) {
3143 bond_shift_load(from, to, i);
3144 port->bond_compat_is_stale = true;
3146 /* If the result of the migration changed the relative order of
3147 * 'from' and 'to' swap them back to maintain invariants. */
3148 if (order_swapped) {
3149 swap_bals(from, to);
3152 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3153 * point to different slave_balance structures. It is only
3154 * valid to do these two operations in a row at all because we
3155 * know that 'from' will not move past 'to' and vice versa. */
3156 resort_bals(from, bals, n_bals);
3157 resort_bals(to, bals, n_bals);
3164 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3165 * historical data to decay to <1% in 7 rebalancing runs. */
3166 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3175 bond_send_learning_packets(struct port *port)
3177 struct bridge *br = port->bridge;
3178 struct mac_entry *e;
3179 struct ofpbuf packet;
3180 int error, n_packets, n_errors;
3182 if (!port->n_ifaces || port->active_iface < 0) {
3186 ofpbuf_init(&packet, 128);
3187 error = n_packets = n_errors = 0;
3188 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3189 union ofp_action actions[2], *a;
3195 if (e->port == port->port_idx
3196 || !choose_output_iface(port, e->mac, e->vlan, &dp_ifidx, &tags)) {
3200 /* Compose actions. */
3201 memset(actions, 0, sizeof actions);
3204 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3205 a->vlan_vid.len = htons(sizeof *a);
3206 a->vlan_vid.vlan_vid = htons(e->vlan);
3209 a->output.type = htons(OFPAT_OUTPUT);
3210 a->output.len = htons(sizeof *a);
3211 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3216 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3218 flow_extract(&packet, 0, ODPP_NONE, &flow);
3219 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3226 ofpbuf_uninit(&packet);
3229 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3230 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3231 "packets, last error was: %s",
3232 port->name, n_errors, n_packets, strerror(error));
3234 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3235 port->name, n_packets);
3239 /* Bonding unixctl user interface functions. */
3242 bond_unixctl_list(struct unixctl_conn *conn,
3243 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3245 struct ds ds = DS_EMPTY_INITIALIZER;
3246 const struct bridge *br;
3248 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3250 LIST_FOR_EACH (br, node, &all_bridges) {
3253 for (i = 0; i < br->n_ports; i++) {
3254 const struct port *port = br->ports[i];
3255 if (port->n_ifaces > 1) {
3258 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3259 bond_type_to_string(port->bond_type));
3260 for (j = 0; j < port->n_ifaces; j++) {
3261 const struct iface *iface = port->ifaces[j];
3263 ds_put_cstr(&ds, ", ");
3265 ds_put_cstr(&ds, iface->name);
3267 ds_put_char(&ds, '\n');
3271 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3275 static struct port *
3276 bond_find(const char *name)
3278 const struct bridge *br;
3280 LIST_FOR_EACH (br, node, &all_bridges) {
3283 for (i = 0; i < br->n_ports; i++) {
3284 struct port *port = br->ports[i];
3285 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3294 bond_unixctl_show(struct unixctl_conn *conn,
3295 const char *args, void *aux OVS_UNUSED)
3297 struct ds ds = DS_EMPTY_INITIALIZER;
3298 const struct port *port;
3301 port = bond_find(args);
3303 unixctl_command_reply(conn, 501, "no such bond");
3307 ds_put_format(&ds, "bond_type: %s\n",
3308 bond_type_to_string(port->bond_type));
3309 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3310 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3312 if (port->bond_type == BT_SLB) {
3313 ds_put_format(&ds, "next rebalance: %lld ms\n",
3314 port->bond_next_rebalance - time_msec());
3317 for (j = 0; j < port->n_ifaces; j++) {
3318 const struct iface *iface = port->ifaces[j];
3319 struct bond_entry *be;
3322 ds_put_format(&ds, "slave %s: %s\n",
3323 iface->name, iface->enabled ? "enabled" : "disabled");
3324 if (j == port->active_iface) {
3325 ds_put_cstr(&ds, "\tactive slave\n");
3327 if (iface->delay_expires != LLONG_MAX) {
3328 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3329 iface->enabled ? "downdelay" : "updelay",
3330 iface->delay_expires - time_msec());
3333 if (port->bond_type != BT_SLB) {
3338 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3339 int hash = be - port->bond_hash;
3340 struct mac_entry *me;
3342 if (be->iface_idx != j) {
3346 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3347 hash, be->tx_bytes / 1024);
3350 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3353 if (bond_hash(me->mac, me->vlan) == hash
3354 && me->port != port->port_idx
3355 && choose_output_iface(port, me->mac, me->vlan,
3357 && dp_ifidx == iface->dp_ifidx)
3359 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3360 ETH_ADDR_ARGS(me->mac));
3365 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3370 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3371 void *aux OVS_UNUSED)
3373 char *args = (char *) args_;
3374 char *save_ptr = NULL;
3375 char *bond_s, *hash_s, *slave_s;
3377 struct iface *iface;
3378 struct bond_entry *entry;
3381 bond_s = strtok_r(args, " ", &save_ptr);
3382 hash_s = strtok_r(NULL, " ", &save_ptr);
3383 slave_s = strtok_r(NULL, " ", &save_ptr);
3385 unixctl_command_reply(conn, 501,
3386 "usage: bond/migrate BOND HASH SLAVE");
3390 port = bond_find(bond_s);
3392 unixctl_command_reply(conn, 501, "no such bond");
3396 if (port->bond_type != BT_SLB) {
3397 unixctl_command_reply(conn, 501, "not an SLB bond");
3401 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3402 hash = atoi(hash_s) & BOND_MASK;
3404 unixctl_command_reply(conn, 501, "bad hash");
3408 iface = port_lookup_iface(port, slave_s);
3410 unixctl_command_reply(conn, 501, "no such slave");
3414 if (!iface->enabled) {
3415 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3419 entry = &port->bond_hash[hash];
3420 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3421 entry->iface_idx = iface->port_ifidx;
3422 entry->iface_tag = tag_create_random();
3423 port->bond_compat_is_stale = true;
3424 unixctl_command_reply(conn, 200, "migrated");
3428 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3429 void *aux OVS_UNUSED)
3431 char *args = (char *) args_;
3432 char *save_ptr = NULL;
3433 char *bond_s, *slave_s;
3435 struct iface *iface;
3437 bond_s = strtok_r(args, " ", &save_ptr);
3438 slave_s = strtok_r(NULL, " ", &save_ptr);
3440 unixctl_command_reply(conn, 501,
3441 "usage: bond/set-active-slave BOND SLAVE");
3445 port = bond_find(bond_s);
3447 unixctl_command_reply(conn, 501, "no such bond");
3451 iface = port_lookup_iface(port, slave_s);
3453 unixctl_command_reply(conn, 501, "no such slave");
3457 if (!iface->enabled) {
3458 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3462 if (port->active_iface != iface->port_ifidx) {
3463 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3464 port->active_iface = iface->port_ifidx;
3465 port->active_iface_tag = tag_create_random();
3466 VLOG_INFO("port %s: active interface is now %s",
3467 port->name, iface->name);
3468 bond_send_learning_packets(port);
3469 unixctl_command_reply(conn, 200, "done");
3471 unixctl_command_reply(conn, 200, "no change");
3476 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3478 char *args = (char *) args_;
3479 char *save_ptr = NULL;
3480 char *bond_s, *slave_s;
3482 struct iface *iface;
3484 bond_s = strtok_r(args, " ", &save_ptr);
3485 slave_s = strtok_r(NULL, " ", &save_ptr);
3487 unixctl_command_reply(conn, 501,
3488 "usage: bond/enable/disable-slave BOND SLAVE");
3492 port = bond_find(bond_s);
3494 unixctl_command_reply(conn, 501, "no such bond");
3498 iface = port_lookup_iface(port, slave_s);
3500 unixctl_command_reply(conn, 501, "no such slave");
3504 bond_enable_slave(iface, enable);
3505 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3509 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3510 void *aux OVS_UNUSED)
3512 enable_slave(conn, args, true);
3516 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3517 void *aux OVS_UNUSED)
3519 enable_slave(conn, args, false);
3523 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3524 void *aux OVS_UNUSED)
3526 char *args = (char *) args_;
3527 uint8_t mac[ETH_ADDR_LEN];
3531 char *mac_s, *vlan_s;
3532 char *save_ptr = NULL;
3534 mac_s = strtok_r(args, " ", &save_ptr);
3535 vlan_s = strtok_r(NULL, " ", &save_ptr);
3538 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3539 unixctl_command_reply(conn, 501, "invalid vlan");
3543 vlan = OFP_VLAN_NONE;
3546 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3547 == ETH_ADDR_SCAN_COUNT) {
3548 hash = bond_hash(mac, vlan);
3550 hash_cstr = xasprintf("%u", hash);
3551 unixctl_command_reply(conn, 200, hash_cstr);
3554 unixctl_command_reply(conn, 501, "invalid mac");
3561 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3562 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3563 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3564 unixctl_command_register("bond/set-active-slave",
3565 bond_unixctl_set_active_slave, NULL);
3566 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3568 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3570 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3573 /* Port functions. */
3575 static struct port *
3576 port_create(struct bridge *br, const char *name)
3580 port = xzalloc(sizeof *port);
3582 port->port_idx = br->n_ports;
3584 port->trunks = NULL;
3585 port->name = xstrdup(name);
3586 port->active_iface = -1;
3588 if (br->n_ports >= br->allocated_ports) {
3589 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3592 br->ports[br->n_ports++] = port;
3593 shash_add_assert(&br->port_by_name, port->name, port);
3595 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3602 get_port_other_config(const struct ovsrec_port *port, const char *key,
3603 const char *default_value)
3607 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3609 return value ? value : default_value;
3613 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3615 struct shash new_ifaces;
3618 /* Collect list of new interfaces. */
3619 shash_init(&new_ifaces);
3620 for (i = 0; i < cfg->n_interfaces; i++) {
3621 const char *name = cfg->interfaces[i]->name;
3622 shash_add_once(&new_ifaces, name, NULL);
3625 /* Get rid of deleted interfaces. */
3626 for (i = 0; i < port->n_ifaces; ) {
3627 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3628 iface_destroy(port->ifaces[i]);
3634 shash_destroy(&new_ifaces);
3638 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3640 struct shash new_ifaces;
3641 long long int next_rebalance;
3642 unsigned long *trunks;
3648 /* Update settings. */
3649 port->updelay = cfg->bond_updelay;
3650 if (port->updelay < 0) {
3653 port->downdelay = cfg->bond_downdelay;
3654 if (port->downdelay < 0) {
3655 port->downdelay = 0;
3657 port->bond_rebalance_interval = atoi(
3658 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3659 if (port->bond_rebalance_interval < 1000) {
3660 port->bond_rebalance_interval = 1000;
3662 next_rebalance = time_msec() + port->bond_rebalance_interval;
3663 if (port->bond_next_rebalance > next_rebalance) {
3664 port->bond_next_rebalance = next_rebalance;
3667 if (!port->cfg->bond_type ||
3668 !strcmp(port->cfg->bond_type, bond_type_to_string(BT_SLB))) {
3669 port->bond_type = BT_SLB;
3670 } else if (!strcmp(port->cfg->bond_type, bond_type_to_string(BT_AB))) {
3671 port->bond_type = BT_AB;
3673 port->bond_type = BT_SLB;
3674 VLOG_WARN("port %s: unknown bond_type %s, defaulting to %s",
3675 port->name, port->cfg->bond_type,
3676 bond_type_to_string(port->bond_type));
3679 /* Add new interfaces and update 'cfg' member of existing ones. */
3680 shash_init(&new_ifaces);
3681 for (i = 0; i < cfg->n_interfaces; i++) {
3682 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3683 struct iface *iface;
3685 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3686 VLOG_WARN("port %s: %s specified twice as port interface",
3687 port->name, if_cfg->name);
3688 iface_set_ofport(if_cfg, -1);
3692 iface = iface_lookup(port->bridge, if_cfg->name);
3694 if (iface->port != port) {
3695 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3697 port->bridge->name, if_cfg->name, iface->port->name);
3700 iface->cfg = if_cfg;
3702 iface = iface_create(port, if_cfg);
3705 /* Determine interface type. The local port always has type
3706 * "internal". Other ports take their type from the database and
3707 * default to "system" if none is specified. */
3708 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
3709 : if_cfg->type[0] ? if_cfg->type
3712 shash_destroy(&new_ifaces);
3717 if (port->n_ifaces < 2) {
3719 if (vlan >= 0 && vlan <= 4095) {
3720 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3725 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3726 * they even work as-is. But they have not been tested. */
3727 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3731 if (port->vlan != vlan) {
3733 bridge_flush(port->bridge);
3736 /* Get trunked VLANs. */
3738 if (vlan < 0 && cfg->n_trunks) {
3741 trunks = bitmap_allocate(4096);
3743 for (i = 0; i < cfg->n_trunks; i++) {
3744 int trunk = cfg->trunks[i];
3746 bitmap_set1(trunks, trunk);
3752 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3753 port->name, cfg->n_trunks);
3755 if (n_errors == cfg->n_trunks) {
3756 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3758 bitmap_free(trunks);
3761 } else if (vlan >= 0 && cfg->n_trunks) {
3762 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3766 ? port->trunks != NULL
3767 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3768 bridge_flush(port->bridge);
3770 bitmap_free(port->trunks);
3771 port->trunks = trunks;
3775 port_destroy(struct port *port)
3778 struct bridge *br = port->bridge;
3782 proc_net_compat_update_vlan(port->name, NULL, 0);
3783 proc_net_compat_update_bond(port->name, NULL);
3785 for (i = 0; i < MAX_MIRRORS; i++) {
3786 struct mirror *m = br->mirrors[i];
3787 if (m && m->out_port == port) {
3792 while (port->n_ifaces > 0) {
3793 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3796 shash_find_and_delete_assert(&br->port_by_name, port->name);
3798 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3799 del->port_idx = port->port_idx;
3801 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
3803 netdev_monitor_destroy(port->monitor);
3805 bitmap_free(port->trunks);
3812 static struct port *
3813 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3815 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3816 return iface ? iface->port : NULL;
3819 static struct port *
3820 port_lookup(const struct bridge *br, const char *name)
3822 return shash_find_data(&br->port_by_name, name);
3825 static struct iface *
3826 port_lookup_iface(const struct port *port, const char *name)
3828 struct iface *iface = iface_lookup(port->bridge, name);
3829 return iface && iface->port == port ? iface : NULL;
3833 port_update_bonding(struct port *port)
3835 if (port->monitor) {
3836 netdev_monitor_destroy(port->monitor);
3837 port->monitor = NULL;
3839 if (port->n_ifaces < 2) {
3840 /* Not a bonded port. */
3841 if (port->bond_hash) {
3842 free(port->bond_hash);
3843 port->bond_hash = NULL;
3844 port->bond_compat_is_stale = true;
3847 port->bond_fake_iface = false;
3851 if (port->bond_type == BT_SLB && !port->bond_hash) {
3852 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3853 for (i = 0; i <= BOND_MASK; i++) {
3854 struct bond_entry *e = &port->bond_hash[i];
3858 port->no_ifaces_tag = tag_create_random();
3859 bond_choose_active_iface(port);
3860 port->bond_next_rebalance
3861 = time_msec() + port->bond_rebalance_interval;
3863 if (port->cfg->bond_fake_iface) {
3864 port->bond_next_fake_iface_update = time_msec();
3866 } else if (port->bond_type != BT_SLB) {
3867 free(port->bond_hash);
3868 port->bond_hash = NULL;
3870 port->bond_compat_is_stale = true;
3871 port->bond_fake_iface = port->cfg->bond_fake_iface;
3873 port->monitor = netdev_monitor_create();
3874 for (i = 0; i < port->n_ifaces; i++) {
3875 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
3881 port_update_bond_compat(struct port *port)
3883 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3884 struct compat_bond bond;
3887 if (port->n_ifaces < 2 || port->bond_type != BT_SLB) {
3888 proc_net_compat_update_bond(port->name, NULL);
3893 bond.updelay = port->updelay;
3894 bond.downdelay = port->downdelay;
3897 bond.hashes = compat_hashes;
3898 if (port->bond_hash) {
3899 const struct bond_entry *e;
3900 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3901 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3902 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3903 cbh->hash = e - port->bond_hash;
3904 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3909 bond.n_slaves = port->n_ifaces;
3910 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3911 for (i = 0; i < port->n_ifaces; i++) {
3912 struct iface *iface = port->ifaces[i];
3913 struct compat_bond_slave *slave = &bond.slaves[i];
3914 slave->name = iface->name;
3916 /* We need to make the same determination as the Linux bonding
3917 * code to determine whether a slave should be consider "up".
3918 * The Linux function bond_miimon_inspect() supports four
3919 * BOND_LINK_* states:
3921 * - BOND_LINK_UP: carrier detected, updelay has passed.
3922 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3923 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3924 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3926 * The function bond_info_show_slave() only considers BOND_LINK_UP
3927 * to be "up" and anything else to be "down".
3929 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3933 netdev_get_etheraddr(iface->netdev, slave->mac);
3936 if (port->bond_fake_iface) {
3937 struct netdev *bond_netdev;
3939 if (!netdev_open_default(port->name, &bond_netdev)) {
3941 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3943 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3945 netdev_close(bond_netdev);
3949 proc_net_compat_update_bond(port->name, &bond);
3954 port_update_vlan_compat(struct port *port)
3956 struct bridge *br = port->bridge;
3957 char *vlandev_name = NULL;
3959 if (port->vlan > 0) {
3960 /* Figure out the name that the VLAN device should actually have, if it
3961 * existed. This takes some work because the VLAN device would not
3962 * have port->name in its name; rather, it would have the trunk port's
3963 * name, and 'port' would be attached to a bridge that also had the
3964 * VLAN device one of its ports. So we need to find a trunk port that
3965 * includes port->vlan.
3967 * There might be more than one candidate. This doesn't happen on
3968 * XenServer, so if it happens we just pick the first choice in
3969 * alphabetical order instead of creating multiple VLAN devices. */
3971 for (i = 0; i < br->n_ports; i++) {
3972 struct port *p = br->ports[i];
3973 if (port_trunks_vlan(p, port->vlan)
3975 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3977 uint8_t ea[ETH_ADDR_LEN];
3978 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3979 if (!eth_addr_is_multicast(ea) &&
3980 !eth_addr_is_reserved(ea) &&
3981 !eth_addr_is_zero(ea)) {
3982 vlandev_name = p->name;
3987 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3990 /* Interface functions. */
3993 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
3996 union ofp_action action;
3998 memset(&action, 0, sizeof action);
3999 action.output.type = htons(OFPAT_OUTPUT);
4000 action.output.len = htons(sizeof action);
4001 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
4003 flow_extract(packet, 0, ODPP_NONE, &flow);
4005 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
4007 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4008 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
4012 static struct iface *
4013 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4015 struct bridge *br = port->bridge;
4016 struct iface *iface;
4017 char *name = if_cfg->name;
4019 iface = xzalloc(sizeof *iface);
4021 iface->port_ifidx = port->n_ifaces;
4022 iface->name = xstrdup(name);
4023 iface->dp_ifidx = -1;
4024 iface->tag = tag_create_random();
4025 iface->delay_expires = LLONG_MAX;
4026 iface->netdev = NULL;
4027 iface->cfg = if_cfg;
4029 shash_add_assert(&br->iface_by_name, iface->name, iface);
4031 if (port->n_ifaces >= port->allocated_ifaces) {
4032 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
4033 sizeof *port->ifaces);
4035 port->ifaces[port->n_ifaces++] = iface;
4036 if (port->n_ifaces > 1) {
4037 br->has_bonded_ports = true;
4040 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4048 iface_destroy(struct iface *iface)
4051 struct port *port = iface->port;
4052 struct bridge *br = port->bridge;
4053 bool del_active = port->active_iface == iface->port_ifidx;
4056 if (port->monitor) {
4057 netdev_monitor_remove(port->monitor, iface->netdev);
4060 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4062 if (iface->dp_ifidx >= 0) {
4063 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4066 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
4067 del->port_ifidx = iface->port_ifidx;
4069 netdev_close(iface->netdev);
4072 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
4073 bond_choose_active_iface(port);
4074 bond_send_learning_packets(port);
4077 cfm_destroy(iface->cfm);
4082 bridge_flush(port->bridge);
4086 static struct iface *
4087 iface_lookup(const struct bridge *br, const char *name)
4089 return shash_find_data(&br->iface_by_name, name);
4092 static struct iface *
4093 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4095 struct iface *iface;
4097 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4098 hash_int(dp_ifidx, 0), &br->ifaces) {
4099 if (iface->dp_ifidx == dp_ifidx) {
4106 /* Set Ethernet address of 'iface', if one is specified in the configuration
4109 iface_set_mac(struct iface *iface)
4111 uint8_t ea[ETH_ADDR_LEN];
4113 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4114 if (eth_addr_is_multicast(ea)) {
4115 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4117 } else if (iface->dp_ifidx == ODPP_LOCAL) {
4118 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
4119 iface->name, iface->name);
4121 int error = netdev_set_etheraddr(iface->netdev, ea);
4123 VLOG_ERR("interface %s: setting MAC failed (%s)",
4124 iface->name, strerror(error));
4130 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4132 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4135 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4139 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4141 * The value strings in '*shash' are taken directly from values[], not copied,
4142 * so the caller should not modify or free them. */
4144 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4145 struct shash *shash)
4150 for (i = 0; i < n; i++) {
4151 shash_add(shash, keys[i], values[i]);
4155 struct iface_delete_queues_cbdata {
4156 struct netdev *netdev;
4157 const struct ovsdb_datum *queues;
4161 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4163 union ovsdb_atom atom;
4165 atom.integer = target;
4166 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4170 iface_delete_queues(unsigned int queue_id,
4171 const struct shash *details OVS_UNUSED, void *cbdata_)
4173 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4175 if (!queue_ids_include(cbdata->queues, queue_id)) {
4176 netdev_delete_queue(cbdata->netdev, queue_id);
4181 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4183 if (!qos || qos->type[0] == '\0') {
4184 netdev_set_qos(iface->netdev, NULL, NULL);
4186 struct iface_delete_queues_cbdata cbdata;
4187 struct shash details;
4190 /* Configure top-level Qos for 'iface'. */
4191 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4192 qos->n_other_config, &details);
4193 netdev_set_qos(iface->netdev, qos->type, &details);
4194 shash_destroy(&details);
4196 /* Deconfigure queues that were deleted. */
4197 cbdata.netdev = iface->netdev;
4198 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4200 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4202 /* Configure queues for 'iface'. */
4203 for (i = 0; i < qos->n_queues; i++) {
4204 const struct ovsrec_queue *queue = qos->value_queues[i];
4205 unsigned int queue_id = qos->key_queues[i];
4207 shash_from_ovs_idl_map(queue->key_other_config,
4208 queue->value_other_config,
4209 queue->n_other_config, &details);
4210 netdev_set_queue(iface->netdev, queue_id, &details);
4211 shash_destroy(&details);
4217 iface_update_cfm(struct iface *iface)
4221 uint16_t *remote_mps;
4222 struct ovsrec_monitor *mon;
4223 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
4225 mon = iface->cfg->monitor;
4231 if (netdev_get_etheraddr(iface->netdev, ea)) {
4232 VLOG_WARN("interface %s: Failed to get ethernet address. "
4233 "Skipping Monitor.", iface->name);
4237 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4238 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4243 iface->cfm = cfm_create();
4247 cfm->mpid = mon->mpid;
4248 cfm->interval = mon->interval ? *mon->interval : 1000;
4250 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
4251 memcpy(cfm->maid, maid, sizeof cfm->maid);
4253 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4254 for(i = 0; i < mon->n_remote_mps; i++) {
4255 remote_mps[i] = mon->remote_mps[i]->mpid;
4257 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
4260 if (!cfm_configure(iface->cfm)) {
4261 cfm_destroy(iface->cfm);
4266 /* Port mirroring. */
4268 static struct mirror *
4269 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4273 for (i = 0; i < MAX_MIRRORS; i++) {
4274 struct mirror *m = br->mirrors[i];
4275 if (m && uuid_equals(uuid, &m->uuid)) {
4283 mirror_reconfigure(struct bridge *br)
4285 unsigned long *rspan_vlans;
4288 /* Get rid of deleted mirrors. */
4289 for (i = 0; i < MAX_MIRRORS; i++) {
4290 struct mirror *m = br->mirrors[i];
4292 const struct ovsdb_datum *mc;
4293 union ovsdb_atom atom;
4295 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4296 atom.uuid = br->mirrors[i]->uuid;
4297 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4303 /* Add new mirrors and reconfigure existing ones. */
4304 for (i = 0; i < br->cfg->n_mirrors; i++) {
4305 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4306 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4308 mirror_reconfigure_one(m, cfg);
4310 mirror_create(br, cfg);
4314 /* Update port reserved status. */
4315 for (i = 0; i < br->n_ports; i++) {
4316 br->ports[i]->is_mirror_output_port = false;
4318 for (i = 0; i < MAX_MIRRORS; i++) {
4319 struct mirror *m = br->mirrors[i];
4320 if (m && m->out_port) {
4321 m->out_port->is_mirror_output_port = true;
4325 /* Update flooded vlans (for RSPAN). */
4327 if (br->cfg->n_flood_vlans) {
4328 rspan_vlans = bitmap_allocate(4096);
4330 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4331 int64_t vlan = br->cfg->flood_vlans[i];
4332 if (vlan >= 0 && vlan < 4096) {
4333 bitmap_set1(rspan_vlans, vlan);
4334 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4337 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4342 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4348 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4353 for (i = 0; ; i++) {
4354 if (i >= MAX_MIRRORS) {
4355 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4356 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4359 if (!br->mirrors[i]) {
4364 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4367 br->mirrors[i] = m = xzalloc(sizeof *m);
4370 m->name = xstrdup(cfg->name);
4371 shash_init(&m->src_ports);
4372 shash_init(&m->dst_ports);
4378 mirror_reconfigure_one(m, cfg);
4382 mirror_destroy(struct mirror *m)
4385 struct bridge *br = m->bridge;
4388 for (i = 0; i < br->n_ports; i++) {
4389 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4390 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4393 shash_destroy(&m->src_ports);
4394 shash_destroy(&m->dst_ports);
4397 m->bridge->mirrors[m->idx] = NULL;
4406 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4407 struct shash *names)
4411 for (i = 0; i < n_ports; i++) {
4412 const char *name = ports[i]->name;
4413 if (port_lookup(m->bridge, name)) {
4414 shash_add_once(names, name, NULL);
4416 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4417 "port %s", m->bridge->name, m->name, name);
4423 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4429 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4431 for (i = 0; i < cfg->n_select_vlan; i++) {
4432 int64_t vlan = cfg->select_vlan[i];
4433 if (vlan < 0 || vlan > 4095) {
4434 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4435 m->bridge->name, m->name, vlan);
4437 (*vlans)[n_vlans++] = vlan;
4444 vlan_is_mirrored(const struct mirror *m, int vlan)
4448 for (i = 0; i < m->n_vlans; i++) {
4449 if (m->vlans[i] == vlan) {
4457 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4461 for (i = 0; i < m->n_vlans; i++) {
4462 if (port_trunks_vlan(p, m->vlans[i])) {
4470 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4472 struct shash src_ports, dst_ports;
4473 mirror_mask_t mirror_bit;
4474 struct port *out_port;
4481 if (strcmp(cfg->name, m->name)) {
4483 m->name = xstrdup(cfg->name);
4486 /* Get output port. */
4487 if (cfg->output_port) {
4488 out_port = port_lookup(m->bridge, cfg->output_port->name);
4490 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4491 m->bridge->name, m->name);
4497 if (cfg->output_vlan) {
4498 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4499 "output vlan; ignoring output vlan",
4500 m->bridge->name, m->name);
4502 } else if (cfg->output_vlan) {
4504 out_vlan = *cfg->output_vlan;
4506 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4507 m->bridge->name, m->name);
4512 shash_init(&src_ports);
4513 shash_init(&dst_ports);
4514 if (cfg->select_all) {
4515 for (i = 0; i < m->bridge->n_ports; i++) {
4516 const char *name = m->bridge->ports[i]->name;
4517 shash_add_once(&src_ports, name, NULL);
4518 shash_add_once(&dst_ports, name, NULL);
4523 /* Get ports, and drop duplicates and ports that don't exist. */
4524 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4526 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4529 /* Get all the vlans, and drop duplicate and invalid vlans. */
4530 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4533 /* Update mirror data. */
4534 if (!shash_equal_keys(&m->src_ports, &src_ports)
4535 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4536 || m->n_vlans != n_vlans
4537 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4538 || m->out_port != out_port
4539 || m->out_vlan != out_vlan) {
4540 bridge_flush(m->bridge);
4542 shash_swap(&m->src_ports, &src_ports);
4543 shash_swap(&m->dst_ports, &dst_ports);
4546 m->n_vlans = n_vlans;
4547 m->out_port = out_port;
4548 m->out_vlan = out_vlan;
4551 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4552 for (i = 0; i < m->bridge->n_ports; i++) {
4553 struct port *port = m->bridge->ports[i];
4555 if (shash_find(&m->src_ports, port->name)
4558 ? port_trunks_any_mirrored_vlan(m, port)
4559 : vlan_is_mirrored(m, port->vlan)))) {
4560 port->src_mirrors |= mirror_bit;
4562 port->src_mirrors &= ~mirror_bit;
4565 if (shash_find(&m->dst_ports, port->name)) {
4566 port->dst_mirrors |= mirror_bit;
4568 port->dst_mirrors &= ~mirror_bit;
4573 shash_destroy(&src_ports);
4574 shash_destroy(&dst_ports);