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"
49 #include "ofp-print.h"
51 #include "ofproto/netflow.h"
52 #include "ofproto/ofproto.h"
53 #include "ovsdb-data.h"
55 #include "poll-loop.h"
56 #include "proc-net-compat.h"
60 #include "socket-util.h"
61 #include "stream-ssl.h"
63 #include "system-stats.h"
68 #include "vswitchd/vswitch-idl.h"
69 #include "xenserver.h"
71 #include "sflow_api.h"
73 VLOG_DEFINE_THIS_MODULE(bridge);
75 COVERAGE_DEFINE(bridge_flush);
76 COVERAGE_DEFINE(bridge_process_flow);
77 COVERAGE_DEFINE(bridge_reconfigure);
85 /* These members are always valid. */
86 struct port *port; /* Containing port. */
87 size_t port_ifidx; /* Index within containing port. */
88 char *name; /* Host network device name. */
89 tag_type tag; /* Tag associated with this interface. */
90 long long delay_expires; /* Time after which 'enabled' may change. */
92 /* These members are valid only after bridge_reconfigure() causes them to
94 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
95 int dp_ifidx; /* Index within kernel datapath. */
96 struct netdev *netdev; /* Network device. */
97 bool enabled; /* May be chosen for flows? */
98 const char *type; /* Usually same as cfg->type. */
99 struct cfm *cfm; /* Connectivity Fault Management */
100 const struct ovsrec_interface *cfg;
103 #define BOND_MASK 0xff
105 int iface_idx; /* Index of assigned iface, or -1 if none. */
106 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
107 tag_type iface_tag; /* Tag associated with iface_idx. */
110 #define MAX_MIRRORS 32
111 typedef uint32_t mirror_mask_t;
112 #define MIRROR_MASK_C(X) UINT32_C(X)
113 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
115 struct bridge *bridge;
118 struct uuid uuid; /* UUID of this "mirror" record in database. */
120 /* Selection criteria. */
121 struct shash src_ports; /* Name is port name; data is always NULL. */
122 struct shash dst_ports; /* Name is port name; data is always NULL. */
127 struct port *out_port;
131 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
133 struct bridge *bridge;
135 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
136 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
137 * NULL if all VLANs are trunked. */
138 const struct ovsrec_port *cfg;
141 /* An ordinary bridge port has 1 interface.
142 * A bridge port for bonding has at least 2 interfaces. */
143 struct iface **ifaces;
144 size_t n_ifaces, allocated_ifaces;
147 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
148 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
149 tag_type active_iface_tag; /* Tag for bcast flows. */
150 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
151 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
152 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
153 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
154 long long int bond_next_fake_iface_update; /* Time of next update. */
155 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
156 long long int bond_next_rebalance; /* Next rebalancing time. */
157 struct netdev_monitor *monitor; /* Tracks carrier up/down status. */
159 /* Port mirroring info. */
160 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
161 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
162 bool is_mirror_output_port; /* Does port mirroring send frames here? */
165 #define DP_MAX_PORTS 255
167 struct list node; /* Node in global list of bridges. */
168 char *name; /* User-specified arbitrary name. */
169 struct mac_learning *ml; /* MAC learning table. */
170 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
171 const struct ovsrec_bridge *cfg;
173 /* OpenFlow switch processing. */
174 struct ofproto *ofproto; /* OpenFlow switch. */
176 /* Kernel datapath information. */
177 struct dpif *dpif; /* Datapath. */
178 struct hmap ifaces; /* Contains "struct iface"s. */
182 size_t n_ports, allocated_ports;
183 struct shash iface_by_name; /* "struct iface"s indexed by name. */
184 struct shash port_by_name; /* "struct port"s indexed by name. */
187 bool has_bonded_ports;
192 /* Port mirroring. */
193 struct mirror *mirrors[MAX_MIRRORS];
196 /* List of all bridges. */
197 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
199 /* OVSDB IDL used to obtain configuration. */
200 static struct ovsdb_idl *idl;
202 /* Each time this timer expires, the bridge fetches systems and interface
203 * statistics and pushes them into the database. */
204 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
205 static long long int stats_timer = LLONG_MIN;
207 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
208 static void bridge_destroy(struct bridge *);
209 static struct bridge *bridge_lookup(const char *name);
210 static unixctl_cb_func bridge_unixctl_dump_flows;
211 static unixctl_cb_func bridge_unixctl_reconnect;
212 static int bridge_run_one(struct bridge *);
213 static size_t bridge_get_controllers(const struct bridge *br,
214 struct ovsrec_controller ***controllersp);
215 static void bridge_reconfigure_one(struct bridge *);
216 static void bridge_reconfigure_remotes(struct bridge *,
217 const struct sockaddr_in *managers,
219 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
220 static void bridge_fetch_dp_ifaces(struct bridge *);
221 static void bridge_flush(struct bridge *);
222 static void bridge_pick_local_hw_addr(struct bridge *,
223 uint8_t ea[ETH_ADDR_LEN],
224 struct iface **hw_addr_iface);
225 static uint64_t bridge_pick_datapath_id(struct bridge *,
226 const uint8_t bridge_ea[ETH_ADDR_LEN],
227 struct iface *hw_addr_iface);
228 static struct iface *bridge_get_local_iface(struct bridge *);
229 static uint64_t dpid_from_hash(const void *, size_t nbytes);
231 static unixctl_cb_func bridge_unixctl_fdb_show;
233 static void bond_init(void);
234 static void bond_run(struct bridge *);
235 static void bond_wait(struct bridge *);
236 static void bond_rebalance_port(struct port *);
237 static void bond_send_learning_packets(struct port *);
238 static void bond_enable_slave(struct iface *iface, bool enable);
240 static struct port *port_create(struct bridge *, const char *name);
241 static void port_reconfigure(struct port *, const struct ovsrec_port *);
242 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
243 static void port_destroy(struct port *);
244 static struct port *port_lookup(const struct bridge *, const char *name);
245 static struct iface *port_lookup_iface(const struct port *, const char *name);
246 static struct port *port_from_dp_ifidx(const struct bridge *,
248 static void port_update_bond_compat(struct port *);
249 static void port_update_vlan_compat(struct port *);
250 static void port_update_bonding(struct port *);
252 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
253 static void mirror_destroy(struct mirror *);
254 static void mirror_reconfigure(struct bridge *);
255 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
256 static bool vlan_is_mirrored(const struct mirror *, int vlan);
258 static struct iface *iface_create(struct port *port,
259 const struct ovsrec_interface *if_cfg);
260 static void iface_destroy(struct iface *);
261 static struct iface *iface_lookup(const struct bridge *, const char *name);
262 static struct iface *iface_from_dp_ifidx(const struct bridge *,
264 static void iface_set_mac(struct iface *);
265 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
266 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
267 static void iface_update_cfm(struct iface *);
268 static void iface_refresh_cfm_stats(struct iface *iface);
269 static void iface_send_packet(struct iface *, struct ofpbuf *packet);
271 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
274 /* Hooks into ofproto processing. */
275 static struct ofhooks bridge_ofhooks;
277 /* Public functions. */
279 /* Initializes the bridge module, configuring it to obtain its configuration
280 * from an OVSDB server accessed over 'remote', which should be a string in a
281 * form acceptable to ovsdb_idl_create(). */
283 bridge_init(const char *remote)
285 /* Create connection to database. */
286 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
288 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
289 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
290 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
292 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
294 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
295 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
297 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
298 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
299 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
301 /* Register unixctl commands. */
302 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
303 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
305 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
310 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
311 * but for which the ovs-vswitchd configuration 'cfg' is required. */
313 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
315 static bool already_configured_once;
316 struct svec bridge_names;
317 struct svec dpif_names, dpif_types;
320 /* Only do this once per ovs-vswitchd run. */
321 if (already_configured_once) {
324 already_configured_once = true;
326 stats_timer = time_msec() + STATS_INTERVAL;
328 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
329 svec_init(&bridge_names);
330 for (i = 0; i < cfg->n_bridges; i++) {
331 svec_add(&bridge_names, cfg->bridges[i]->name);
333 svec_sort(&bridge_names);
335 /* Iterate over all system dpifs and delete any of them that do not appear
337 svec_init(&dpif_names);
338 svec_init(&dpif_types);
339 dp_enumerate_types(&dpif_types);
340 for (i = 0; i < dpif_types.n; i++) {
345 dp_enumerate_names(dpif_types.names[i], &dpif_names);
347 /* For each dpif... */
348 for (j = 0; j < dpif_names.n; j++) {
349 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
351 struct svec all_names;
354 /* ...check whether any of its names is in 'bridge_names'. */
355 svec_init(&all_names);
356 dpif_get_all_names(dpif, &all_names);
357 for (k = 0; k < all_names.n; k++) {
358 if (svec_contains(&bridge_names, all_names.names[k])) {
363 /* No. Delete the dpif. */
367 svec_destroy(&all_names);
372 svec_destroy(&bridge_names);
373 svec_destroy(&dpif_names);
374 svec_destroy(&dpif_types);
377 /* Initializes 'options' and fills it with the options for 'if_cfg'. Merges
378 * keys from "options" and "other_config", preferring "options" keys over
379 * "other_config" keys. */
381 iface_get_options(const struct ovsrec_interface *if_cfg, struct shash *options)
385 shash_from_ovs_idl_map(if_cfg->key_options, if_cfg->value_options,
386 if_cfg->n_options, options);
388 for (i = 0; i < if_cfg->n_other_config; i++) {
389 char *key = if_cfg->key_other_config[i];
390 char *value = if_cfg->value_other_config[i];
392 if (!shash_find_data(options, key)) {
393 shash_add(options, key, value);
395 VLOG_WARN("%s: ignoring \"other_config\" key %s that conflicts "
396 "with \"options\" key %s", if_cfg->name, key, key);
401 /* Callback for iterate_and_prune_ifaces(). */
403 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
405 if (!iface->netdev) {
406 /* We already reported a related error, don't bother duplicating it. */
410 if (iface->dp_ifidx < 0) {
411 VLOG_ERR("%s interface not in %s, dropping",
412 iface->name, dpif_name(br->dpif));
416 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
417 iface->name, iface->dp_ifidx);
421 /* Callback for iterate_and_prune_ifaces(). */
423 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
424 void *aux OVS_UNUSED)
426 /* Set policing attributes. */
427 netdev_set_policing(iface->netdev,
428 iface->cfg->ingress_policing_rate,
429 iface->cfg->ingress_policing_burst);
431 /* Set MAC address of internal interfaces other than the local
433 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
434 iface_set_mac(iface);
440 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
441 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
442 * deletes from 'br' any ports that no longer have any interfaces. */
444 iterate_and_prune_ifaces(struct bridge *br,
445 bool (*cb)(struct bridge *, struct iface *,
451 for (i = 0; i < br->n_ports; ) {
452 struct port *port = br->ports[i];
453 for (j = 0; j < port->n_ifaces; ) {
454 struct iface *iface = port->ifaces[j];
455 if (cb(br, iface, aux)) {
458 iface_set_ofport(iface->cfg, -1);
459 iface_destroy(iface);
463 if (port->n_ifaces) {
466 VLOG_ERR("%s port has no interfaces, dropping", port->name);
472 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
473 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
474 * responsible for freeing '*managersp' (with free()).
476 * You may be asking yourself "why does ovs-vswitchd care?", because
477 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
478 * should not be and in fact is not directly involved in that. But
479 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
480 * it has to tell in-band control where the managers are to enable that.
481 * (Thus, only managers connected in-band are collected.)
484 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
485 struct sockaddr_in **managersp, size_t *n_managersp)
487 struct sockaddr_in *managers = NULL;
488 size_t n_managers = 0;
489 struct shash targets;
492 /* Collect all of the potential targets, as the union of the "managers"
493 * column and the "targets" columns of the rows pointed to by
494 * "manager_options", excluding any that are out-of-band. */
495 shash_init(&targets);
496 for (i = 0; i < ovs_cfg->n_managers; i++) {
497 shash_add_once(&targets, ovs_cfg->managers[i], NULL);
499 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
500 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
502 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
503 shash_find_and_delete(&targets, m->target);
505 shash_add_once(&targets, m->target, NULL);
509 /* Now extract the targets' IP addresses. */
510 if (!shash_is_empty(&targets)) {
511 struct shash_node *node;
513 managers = xmalloc(shash_count(&targets) * sizeof *managers);
514 SHASH_FOR_EACH (node, &targets) {
515 const char *target = node->name;
516 struct sockaddr_in *sin = &managers[n_managers];
518 if ((!strncmp(target, "tcp:", 4)
519 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
520 (!strncmp(target, "ssl:", 4)
521 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
526 shash_destroy(&targets);
528 *managersp = managers;
529 *n_managersp = n_managers;
533 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
535 struct shash old_br, new_br;
536 struct shash_node *node;
537 struct bridge *br, *next;
538 struct sockaddr_in *managers;
541 int sflow_bridge_number;
543 COVERAGE_INC(bridge_reconfigure);
545 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
547 /* Collect old and new bridges. */
550 LIST_FOR_EACH (br, node, &all_bridges) {
551 shash_add(&old_br, br->name, br);
553 for (i = 0; i < ovs_cfg->n_bridges; i++) {
554 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
555 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
556 VLOG_WARN("more than one bridge named %s", br_cfg->name);
560 /* Get rid of deleted bridges and add new bridges. */
561 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
562 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
569 SHASH_FOR_EACH (node, &new_br) {
570 const char *br_name = node->name;
571 const struct ovsrec_bridge *br_cfg = node->data;
572 br = shash_find_data(&old_br, br_name);
574 /* If the bridge datapath type has changed, we need to tear it
575 * down and recreate. */
576 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
578 bridge_create(br_cfg);
581 bridge_create(br_cfg);
584 shash_destroy(&old_br);
585 shash_destroy(&new_br);
587 /* Reconfigure all bridges. */
588 LIST_FOR_EACH (br, node, &all_bridges) {
589 bridge_reconfigure_one(br);
592 /* Add and delete ports on all datapaths.
594 * The kernel will reject any attempt to add a given port to a datapath if
595 * that port already belongs to a different datapath, so we must do all
596 * port deletions before any port additions. */
597 LIST_FOR_EACH (br, node, &all_bridges) {
598 struct odp_port *dpif_ports;
600 struct shash want_ifaces;
602 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
603 bridge_get_all_ifaces(br, &want_ifaces);
604 for (i = 0; i < n_dpif_ports; i++) {
605 const struct odp_port *p = &dpif_ports[i];
606 if (!shash_find(&want_ifaces, p->devname)
607 && strcmp(p->devname, br->name)) {
608 int retval = dpif_port_del(br->dpif, p->port);
610 VLOG_ERR("failed to remove %s interface from %s: %s",
611 p->devname, dpif_name(br->dpif),
616 shash_destroy(&want_ifaces);
619 LIST_FOR_EACH (br, node, &all_bridges) {
620 struct odp_port *dpif_ports;
622 struct shash cur_ifaces, want_ifaces;
624 /* Get the set of interfaces currently in this datapath. */
625 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
626 shash_init(&cur_ifaces);
627 for (i = 0; i < n_dpif_ports; i++) {
628 const char *name = dpif_ports[i].devname;
629 shash_add_once(&cur_ifaces, name, &dpif_ports[i]);
632 /* Get the set of interfaces we want on this datapath. */
633 bridge_get_all_ifaces(br, &want_ifaces);
635 hmap_clear(&br->ifaces);
636 SHASH_FOR_EACH (node, &want_ifaces) {
637 const char *if_name = node->name;
638 struct iface *iface = node->data;
639 struct odp_port *dpif_port = shash_find_data(&cur_ifaces, if_name);
640 const char *type = iface ? iface->type : "internal";
643 /* If we have a port or a netdev already, and it's not the type we
644 * want, then delete the port (if any) and close the netdev (if
646 if ((dpif_port && strcmp(dpif_port->type, type))
647 || (iface && iface->netdev
648 && strcmp(type, netdev_get_type(iface->netdev)))) {
650 error = ofproto_port_del(br->ofproto, dpif_port->port);
657 netdev_close(iface->netdev);
658 iface->netdev = NULL;
662 /* If the port doesn't exist or we don't have the netdev open,
663 * we need to do more work. */
664 if (!dpif_port || (iface && !iface->netdev)) {
665 struct netdev_options options;
666 struct netdev *netdev;
669 /* First open the network device. */
670 options.name = if_name;
672 options.args = &args;
673 options.ethertype = NETDEV_ETH_TYPE_NONE;
677 iface_get_options(iface->cfg, &args);
679 error = netdev_open(&options, &netdev);
680 shash_destroy(&args);
683 VLOG_WARN("could not open network device %s (%s)",
684 if_name, strerror(error));
688 /* Then add the port if we haven't already. */
690 error = dpif_port_add(br->dpif, netdev, NULL);
692 netdev_close(netdev);
693 if (error == EFBIG) {
694 VLOG_ERR("ran out of valid port numbers on %s",
695 dpif_name(br->dpif));
698 VLOG_ERR("failed to add %s interface to %s: %s",
699 if_name, dpif_name(br->dpif),
706 /* Update 'iface'. */
708 iface->netdev = netdev;
709 iface->enabled = netdev_get_carrier(iface->netdev);
711 } else if (iface && iface->netdev) {
715 iface_get_options(iface->cfg, &args);
716 netdev_reconfigure(iface->netdev, &args);
717 shash_destroy(&args);
721 shash_destroy(&cur_ifaces);
722 shash_destroy(&want_ifaces);
724 sflow_bridge_number = 0;
725 LIST_FOR_EACH (br, node, &all_bridges) {
728 struct iface *local_iface;
729 struct iface *hw_addr_iface;
732 bridge_fetch_dp_ifaces(br);
734 iterate_and_prune_ifaces(br, check_iface, NULL);
736 /* Pick local port hardware address, datapath ID. */
737 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
738 local_iface = bridge_get_local_iface(br);
740 int error = netdev_set_etheraddr(local_iface->netdev, ea);
742 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
743 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
744 "Ethernet address: %s",
745 br->name, strerror(error));
749 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
750 ofproto_set_datapath_id(br->ofproto, dpid);
752 dpid_string = xasprintf("%016"PRIx64, dpid);
753 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
756 /* Set NetFlow configuration on this bridge. */
757 if (br->cfg->netflow) {
758 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
759 struct netflow_options opts;
761 memset(&opts, 0, sizeof opts);
763 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
764 if (nf_cfg->engine_type) {
765 opts.engine_type = *nf_cfg->engine_type;
767 if (nf_cfg->engine_id) {
768 opts.engine_id = *nf_cfg->engine_id;
771 opts.active_timeout = nf_cfg->active_timeout;
772 if (!opts.active_timeout) {
773 opts.active_timeout = -1;
774 } else if (opts.active_timeout < 0) {
775 VLOG_WARN("bridge %s: active timeout interval set to negative "
776 "value, using default instead (%d seconds)", br->name,
777 NF_ACTIVE_TIMEOUT_DEFAULT);
778 opts.active_timeout = -1;
781 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
782 if (opts.add_id_to_iface) {
783 if (opts.engine_id > 0x7f) {
784 VLOG_WARN("bridge %s: netflow port mangling may conflict "
785 "with another vswitch, choose an engine id less "
786 "than 128", br->name);
788 if (br->n_ports > 508) {
789 VLOG_WARN("bridge %s: netflow port mangling will conflict "
790 "with another port when more than 508 ports are "
795 opts.collectors.n = nf_cfg->n_targets;
796 opts.collectors.names = nf_cfg->targets;
797 if (ofproto_set_netflow(br->ofproto, &opts)) {
798 VLOG_ERR("bridge %s: problem setting netflow collectors",
802 ofproto_set_netflow(br->ofproto, NULL);
805 /* Set sFlow configuration on this bridge. */
806 if (br->cfg->sflow) {
807 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
808 struct ovsrec_controller **controllers;
809 struct ofproto_sflow_options oso;
810 size_t n_controllers;
812 memset(&oso, 0, sizeof oso);
814 oso.targets.n = sflow_cfg->n_targets;
815 oso.targets.names = sflow_cfg->targets;
817 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
818 if (sflow_cfg->sampling) {
819 oso.sampling_rate = *sflow_cfg->sampling;
822 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
823 if (sflow_cfg->polling) {
824 oso.polling_interval = *sflow_cfg->polling;
827 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
828 if (sflow_cfg->header) {
829 oso.header_len = *sflow_cfg->header;
832 oso.sub_id = sflow_bridge_number++;
833 oso.agent_device = sflow_cfg->agent;
835 oso.control_ip = NULL;
836 n_controllers = bridge_get_controllers(br, &controllers);
837 for (i = 0; i < n_controllers; i++) {
838 if (controllers[i]->local_ip) {
839 oso.control_ip = controllers[i]->local_ip;
843 ofproto_set_sflow(br->ofproto, &oso);
845 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
847 ofproto_set_sflow(br->ofproto, NULL);
850 /* Update the controller and related settings. It would be more
851 * straightforward to call this from bridge_reconfigure_one(), but we
852 * can't do it there for two reasons. First, and most importantly, at
853 * that point we don't know the dp_ifidx of any interfaces that have
854 * been added to the bridge (because we haven't actually added them to
855 * the datapath). Second, at that point we haven't set the datapath ID
856 * yet; when a controller is configured, resetting the datapath ID will
857 * immediately disconnect from the controller, so it's better to set
858 * the datapath ID before the controller. */
859 bridge_reconfigure_remotes(br, managers, n_managers);
861 LIST_FOR_EACH (br, node, &all_bridges) {
862 for (i = 0; i < br->n_ports; i++) {
863 struct port *port = br->ports[i];
866 port_update_vlan_compat(port);
867 port_update_bonding(port);
869 for (j = 0; j < port->n_ifaces; j++) {
870 iface_update_qos(port->ifaces[j], port->cfg->qos);
874 LIST_FOR_EACH (br, node, &all_bridges) {
875 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
878 LIST_FOR_EACH (br, node, &all_bridges) {
880 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
881 iface_update_cfm(iface);
889 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
890 const struct ovsdb_idl_column *column,
893 const struct ovsdb_datum *datum;
894 union ovsdb_atom atom;
897 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
898 atom.string = (char *) key;
899 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
900 return idx == UINT_MAX ? NULL : datum->values[idx].string;
904 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
906 return get_ovsrec_key_value(&br_cfg->header_,
907 &ovsrec_bridge_col_other_config, key);
911 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
912 struct iface **hw_addr_iface)
918 *hw_addr_iface = NULL;
920 /* Did the user request a particular MAC? */
921 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
922 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
923 if (eth_addr_is_multicast(ea)) {
924 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
925 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
926 } else if (eth_addr_is_zero(ea)) {
927 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
933 /* Otherwise choose the minimum non-local MAC address among all of the
935 memset(ea, 0xff, sizeof ea);
936 for (i = 0; i < br->n_ports; i++) {
937 struct port *port = br->ports[i];
938 uint8_t iface_ea[ETH_ADDR_LEN];
941 /* Mirror output ports don't participate. */
942 if (port->is_mirror_output_port) {
946 /* Choose the MAC address to represent the port. */
947 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
948 /* Find the interface with this Ethernet address (if any) so that
949 * we can provide the correct devname to the caller. */
951 for (j = 0; j < port->n_ifaces; j++) {
952 struct iface *candidate = port->ifaces[j];
953 uint8_t candidate_ea[ETH_ADDR_LEN];
954 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
955 && eth_addr_equals(iface_ea, candidate_ea)) {
960 /* Choose the interface whose MAC address will represent the port.
961 * The Linux kernel bonding code always chooses the MAC address of
962 * the first slave added to a bond, and the Fedora networking
963 * scripts always add slaves to a bond in alphabetical order, so
964 * for compatibility we choose the interface with the name that is
965 * first in alphabetical order. */
966 iface = port->ifaces[0];
967 for (j = 1; j < port->n_ifaces; j++) {
968 struct iface *candidate = port->ifaces[j];
969 if (strcmp(candidate->name, iface->name) < 0) {
974 /* The local port doesn't count (since we're trying to choose its
975 * MAC address anyway). */
976 if (iface->dp_ifidx == ODPP_LOCAL) {
981 error = netdev_get_etheraddr(iface->netdev, iface_ea);
983 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
984 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
985 iface->name, strerror(error));
990 /* Compare against our current choice. */
991 if (!eth_addr_is_multicast(iface_ea) &&
992 !eth_addr_is_local(iface_ea) &&
993 !eth_addr_is_reserved(iface_ea) &&
994 !eth_addr_is_zero(iface_ea) &&
995 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
997 memcpy(ea, iface_ea, ETH_ADDR_LEN);
998 *hw_addr_iface = iface;
1001 if (eth_addr_is_multicast(ea)) {
1002 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1003 *hw_addr_iface = NULL;
1004 VLOG_WARN("bridge %s: using default bridge Ethernet "
1005 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1007 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1008 br->name, ETH_ADDR_ARGS(ea));
1012 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1013 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1014 * an interface on 'br', then that interface must be passed in as
1015 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1016 * 'hw_addr_iface' must be passed in as a null pointer. */
1018 bridge_pick_datapath_id(struct bridge *br,
1019 const uint8_t bridge_ea[ETH_ADDR_LEN],
1020 struct iface *hw_addr_iface)
1023 * The procedure for choosing a bridge MAC address will, in the most
1024 * ordinary case, also choose a unique MAC that we can use as a datapath
1025 * ID. In some special cases, though, multiple bridges will end up with
1026 * the same MAC address. This is OK for the bridges, but it will confuse
1027 * the OpenFlow controller, because each datapath needs a unique datapath
1030 * Datapath IDs must be unique. It is also very desirable that they be
1031 * stable from one run to the next, so that policy set on a datapath
1034 const char *datapath_id;
1037 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1038 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1042 if (hw_addr_iface) {
1044 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1046 * A bridge whose MAC address is taken from a VLAN network device
1047 * (that is, a network device created with vconfig(8) or similar
1048 * tool) will have the same MAC address as a bridge on the VLAN
1049 * device's physical network device.
1051 * Handle this case by hashing the physical network device MAC
1052 * along with the VLAN identifier.
1054 uint8_t buf[ETH_ADDR_LEN + 2];
1055 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1056 buf[ETH_ADDR_LEN] = vlan >> 8;
1057 buf[ETH_ADDR_LEN + 1] = vlan;
1058 return dpid_from_hash(buf, sizeof buf);
1061 * Assume that this bridge's MAC address is unique, since it
1062 * doesn't fit any of the cases we handle specially.
1067 * A purely internal bridge, that is, one that has no non-virtual
1068 * network devices on it at all, is more difficult because it has no
1069 * natural unique identifier at all.
1071 * When the host is a XenServer, we handle this case by hashing the
1072 * host's UUID with the name of the bridge. Names of bridges are
1073 * persistent across XenServer reboots, although they can be reused if
1074 * an internal network is destroyed and then a new one is later
1075 * created, so this is fairly effective.
1077 * When the host is not a XenServer, we punt by using a random MAC
1078 * address on each run.
1080 const char *host_uuid = xenserver_get_host_uuid();
1082 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1083 dpid = dpid_from_hash(combined, strlen(combined));
1089 return eth_addr_to_uint64(bridge_ea);
1093 dpid_from_hash(const void *data, size_t n)
1095 uint8_t hash[SHA1_DIGEST_SIZE];
1097 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1098 sha1_bytes(data, n, hash);
1099 eth_addr_mark_random(hash);
1100 return eth_addr_to_uint64(hash);
1104 iface_refresh_cfm_stats(struct iface *iface)
1108 const struct ovsrec_monitor *mon;
1110 mon = iface->cfg->monitor;
1117 for (i = 0; i < mon->n_remote_mps; i++) {
1118 const struct ovsrec_maintenance_point *mp;
1119 const struct remote_mp *rmp;
1121 mp = mon->remote_mps[i];
1122 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1124 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1127 if (hmap_is_empty(&cfm->x_remote_mps)) {
1128 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1131 struct remote_mp *rmp;
1132 int64_t *x_remote_mps;
1134 length = hmap_count(&cfm->x_remote_mps);
1135 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1138 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1139 x_remote_mps[i++] = rmp->mpid;
1142 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1146 if (hmap_is_empty(&cfm->x_remote_maids)) {
1147 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1150 char **x_remote_maids;
1151 struct remote_maid *rmaid;
1153 length = hmap_count(&cfm->x_remote_maids);
1154 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1157 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1160 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1162 for (j = 0; j < CCM_MAID_LEN; j++) {
1163 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1168 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1170 for (i = 0; i < length; i++) {
1171 free(x_remote_maids[i]);
1173 free(x_remote_maids);
1176 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1180 iface_refresh_stats(struct iface *iface)
1186 static const struct iface_stat iface_stats[] = {
1187 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1188 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1189 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1190 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1191 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1192 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1193 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1194 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1195 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1196 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1197 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1198 { "collisions", offsetof(struct netdev_stats, collisions) },
1200 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1201 const struct iface_stat *s;
1203 char *keys[N_STATS];
1204 int64_t values[N_STATS];
1207 struct netdev_stats stats;
1209 /* Intentionally ignore return value, since errors will set 'stats' to
1210 * all-1s, and we will deal with that correctly below. */
1211 netdev_get_stats(iface->netdev, &stats);
1214 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1215 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1216 if (value != UINT64_MAX) {
1223 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1227 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1229 struct ovsdb_datum datum;
1233 get_system_stats(&stats);
1235 ovsdb_datum_from_shash(&datum, &stats);
1236 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1243 const struct ovsrec_open_vswitch *cfg;
1245 bool datapath_destroyed;
1246 bool database_changed;
1249 /* Let each bridge do the work that it needs to do. */
1250 datapath_destroyed = false;
1251 LIST_FOR_EACH (br, node, &all_bridges) {
1252 int error = bridge_run_one(br);
1254 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1255 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1256 "forcing reconfiguration", br->name);
1257 datapath_destroyed = true;
1261 /* (Re)configure if necessary. */
1262 database_changed = ovsdb_idl_run(idl);
1263 cfg = ovsrec_open_vswitch_first(idl);
1264 if (database_changed || datapath_destroyed) {
1266 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1268 bridge_configure_once(cfg);
1269 bridge_reconfigure(cfg);
1271 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1272 ovsdb_idl_txn_commit(txn);
1273 ovsdb_idl_txn_destroy(txn); /* XXX */
1275 /* We still need to reconfigure to avoid dangling pointers to
1276 * now-destroyed ovsrec structures inside bridge data. */
1277 static const struct ovsrec_open_vswitch null_cfg;
1279 bridge_reconfigure(&null_cfg);
1284 /* Re-configure SSL. We do this on every trip through the main loop,
1285 * instead of just when the database changes, because the contents of the
1286 * key and certificate files can change without the database changing. */
1287 if (cfg && cfg->ssl) {
1288 const struct ovsrec_ssl *ssl = cfg->ssl;
1290 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1291 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1295 /* Refresh system and interface stats if necessary. */
1296 if (time_msec() >= stats_timer) {
1298 struct ovsdb_idl_txn *txn;
1300 txn = ovsdb_idl_txn_create(idl);
1301 LIST_FOR_EACH (br, node, &all_bridges) {
1304 for (i = 0; i < br->n_ports; i++) {
1305 struct port *port = br->ports[i];
1308 for (j = 0; j < port->n_ifaces; j++) {
1309 struct iface *iface = port->ifaces[j];
1310 iface_refresh_stats(iface);
1311 iface_refresh_cfm_stats(iface);
1315 refresh_system_stats(cfg);
1316 ovsdb_idl_txn_commit(txn);
1317 ovsdb_idl_txn_destroy(txn); /* XXX */
1320 stats_timer = time_msec() + STATS_INTERVAL;
1328 struct iface *iface;
1330 LIST_FOR_EACH (br, node, &all_bridges) {
1331 ofproto_wait(br->ofproto);
1332 if (ofproto_has_primary_controller(br->ofproto)) {
1336 mac_learning_wait(br->ml);
1339 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1341 cfm_wait(iface->cfm);
1345 ovsdb_idl_wait(idl);
1346 poll_timer_wait_until(stats_timer);
1349 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1350 * configuration changes. */
1352 bridge_flush(struct bridge *br)
1354 COVERAGE_INC(bridge_flush);
1356 mac_learning_flush(br->ml);
1359 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1360 * such interface. */
1361 static struct iface *
1362 bridge_get_local_iface(struct bridge *br)
1366 for (i = 0; i < br->n_ports; i++) {
1367 struct port *port = br->ports[i];
1368 for (j = 0; j < port->n_ifaces; j++) {
1369 struct iface *iface = port->ifaces[j];
1370 if (iface->dp_ifidx == ODPP_LOCAL) {
1379 /* Bridge unixctl user interface functions. */
1381 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1382 const char *args, void *aux OVS_UNUSED)
1384 struct ds ds = DS_EMPTY_INITIALIZER;
1385 const struct bridge *br;
1386 const struct mac_entry *e;
1388 br = bridge_lookup(args);
1390 unixctl_command_reply(conn, 501, "no such bridge");
1394 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1395 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1396 if (e->port < 0 || e->port >= br->n_ports) {
1399 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1400 br->ports[e->port]->ifaces[0]->dp_ifidx,
1401 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1403 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1407 /* Bridge reconfiguration functions. */
1408 static struct bridge *
1409 bridge_create(const struct ovsrec_bridge *br_cfg)
1414 assert(!bridge_lookup(br_cfg->name));
1415 br = xzalloc(sizeof *br);
1417 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1423 dpif_flow_flush(br->dpif);
1425 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1428 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1430 dpif_delete(br->dpif);
1431 dpif_close(br->dpif);
1436 br->name = xstrdup(br_cfg->name);
1438 br->ml = mac_learning_create();
1439 eth_addr_nicira_random(br->default_ea);
1441 hmap_init(&br->ifaces);
1443 shash_init(&br->port_by_name);
1444 shash_init(&br->iface_by_name);
1448 list_push_back(&all_bridges, &br->node);
1450 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1456 bridge_destroy(struct bridge *br)
1461 while (br->n_ports > 0) {
1462 port_destroy(br->ports[br->n_ports - 1]);
1464 list_remove(&br->node);
1465 error = dpif_delete(br->dpif);
1466 if (error && error != ENOENT) {
1467 VLOG_ERR("failed to delete %s: %s",
1468 dpif_name(br->dpif), strerror(error));
1470 dpif_close(br->dpif);
1471 ofproto_destroy(br->ofproto);
1472 mac_learning_destroy(br->ml);
1473 hmap_destroy(&br->ifaces);
1474 shash_destroy(&br->port_by_name);
1475 shash_destroy(&br->iface_by_name);
1482 static struct bridge *
1483 bridge_lookup(const char *name)
1487 LIST_FOR_EACH (br, node, &all_bridges) {
1488 if (!strcmp(br->name, name)) {
1495 /* Handle requests for a listing of all flows known by the OpenFlow
1496 * stack, including those normally hidden. */
1498 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1499 const char *args, void *aux OVS_UNUSED)
1504 br = bridge_lookup(args);
1506 unixctl_command_reply(conn, 501, "Unknown bridge");
1511 ofproto_get_all_flows(br->ofproto, &results);
1513 unixctl_command_reply(conn, 200, ds_cstr(&results));
1514 ds_destroy(&results);
1517 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1518 * connections and reconnect. If BRIDGE is not specified, then all bridges
1519 * drop their controller connections and reconnect. */
1521 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1522 const char *args, void *aux OVS_UNUSED)
1525 if (args[0] != '\0') {
1526 br = bridge_lookup(args);
1528 unixctl_command_reply(conn, 501, "Unknown bridge");
1531 ofproto_reconnect_controllers(br->ofproto);
1533 LIST_FOR_EACH (br, node, &all_bridges) {
1534 ofproto_reconnect_controllers(br->ofproto);
1537 unixctl_command_reply(conn, 200, NULL);
1541 bridge_run_one(struct bridge *br)
1544 struct iface *iface;
1546 error = ofproto_run1(br->ofproto);
1551 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1554 error = ofproto_run2(br->ofproto, br->flush);
1557 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1558 struct ofpbuf *packet;
1564 packet = cfm_run(iface->cfm);
1566 iface_send_packet(iface, packet);
1567 ofpbuf_uninit(packet);
1576 bridge_get_controllers(const struct bridge *br,
1577 struct ovsrec_controller ***controllersp)
1579 struct ovsrec_controller **controllers;
1580 size_t n_controllers;
1582 controllers = br->cfg->controller;
1583 n_controllers = br->cfg->n_controller;
1585 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1591 *controllersp = controllers;
1593 return n_controllers;
1597 bridge_reconfigure_one(struct bridge *br)
1599 struct shash old_ports, new_ports;
1600 struct svec snoops, old_snoops;
1601 struct shash_node *node;
1602 enum ofproto_fail_mode fail_mode;
1605 /* Collect old ports. */
1606 shash_init(&old_ports);
1607 for (i = 0; i < br->n_ports; i++) {
1608 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1611 /* Collect new ports. */
1612 shash_init(&new_ports);
1613 for (i = 0; i < br->cfg->n_ports; i++) {
1614 const char *name = br->cfg->ports[i]->name;
1615 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1616 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1621 /* If we have a controller, then we need a local port. Complain if the
1622 * user didn't specify one.
1624 * XXX perhaps we should synthesize a port ourselves in this case. */
1625 if (bridge_get_controllers(br, NULL)) {
1626 char local_name[IF_NAMESIZE];
1629 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1630 local_name, sizeof local_name);
1631 if (!error && !shash_find(&new_ports, local_name)) {
1632 VLOG_WARN("bridge %s: controller specified but no local port "
1633 "(port named %s) defined",
1634 br->name, local_name);
1638 /* Get rid of deleted ports.
1639 * Get rid of deleted interfaces on ports that still exist. */
1640 SHASH_FOR_EACH (node, &old_ports) {
1641 struct port *port = node->data;
1642 const struct ovsrec_port *port_cfg;
1644 port_cfg = shash_find_data(&new_ports, node->name);
1648 port_del_ifaces(port, port_cfg);
1652 /* Create new ports.
1653 * Add new interfaces to existing ports.
1654 * Reconfigure existing ports. */
1655 SHASH_FOR_EACH (node, &new_ports) {
1656 struct port *port = shash_find_data(&old_ports, node->name);
1658 port = port_create(br, node->name);
1661 port_reconfigure(port, node->data);
1662 if (!port->n_ifaces) {
1663 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1664 br->name, port->name);
1668 shash_destroy(&old_ports);
1669 shash_destroy(&new_ports);
1671 /* Set the fail-mode */
1672 fail_mode = !br->cfg->fail_mode
1673 || !strcmp(br->cfg->fail_mode, "standalone")
1674 ? OFPROTO_FAIL_STANDALONE
1675 : OFPROTO_FAIL_SECURE;
1676 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1677 && !ofproto_has_primary_controller(br->ofproto)) {
1678 ofproto_flush_flows(br->ofproto);
1680 ofproto_set_fail_mode(br->ofproto, fail_mode);
1682 /* Delete all flows if we're switching from connected to standalone or vice
1683 * versa. (XXX Should we delete all flows if we are switching from one
1684 * controller to another?) */
1686 /* Configure OpenFlow controller connection snooping. */
1688 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1689 ovs_rundir(), br->name));
1690 svec_init(&old_snoops);
1691 ofproto_get_snoops(br->ofproto, &old_snoops);
1692 if (!svec_equal(&snoops, &old_snoops)) {
1693 ofproto_set_snoops(br->ofproto, &snoops);
1695 svec_destroy(&snoops);
1696 svec_destroy(&old_snoops);
1698 mirror_reconfigure(br);
1701 /* Initializes 'oc' appropriately as a management service controller for
1704 * The caller must free oc->target when it is no longer needed. */
1706 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1707 struct ofproto_controller *oc)
1709 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1710 oc->max_backoff = 0;
1711 oc->probe_interval = 60;
1712 oc->band = OFPROTO_OUT_OF_BAND;
1713 oc->accept_re = NULL;
1714 oc->update_resolv_conf = false;
1716 oc->burst_limit = 0;
1719 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1721 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1722 struct ofproto_controller *oc)
1724 oc->target = c->target;
1725 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1726 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1727 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1728 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1729 oc->accept_re = c->discover_accept_regex;
1730 oc->update_resolv_conf = c->discover_update_resolv_conf;
1731 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1732 oc->burst_limit = (c->controller_burst_limit
1733 ? *c->controller_burst_limit : 0);
1736 /* Configures the IP stack for 'br''s local interface properly according to the
1737 * configuration in 'c'. */
1739 bridge_configure_local_iface_netdev(struct bridge *br,
1740 struct ovsrec_controller *c)
1742 struct netdev *netdev;
1743 struct in_addr mask, gateway;
1745 struct iface *local_iface;
1748 /* Controller discovery does its own TCP/IP configuration later. */
1749 if (strcmp(c->target, "discover")) {
1753 /* If there's no local interface or no IP address, give up. */
1754 local_iface = bridge_get_local_iface(br);
1755 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1759 /* Bring up the local interface. */
1760 netdev = local_iface->netdev;
1761 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1763 /* Configure the IP address and netmask. */
1764 if (!c->local_netmask
1765 || !inet_aton(c->local_netmask, &mask)
1767 mask.s_addr = guess_netmask(ip.s_addr);
1769 if (!netdev_set_in4(netdev, ip, mask)) {
1770 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1771 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1774 /* Configure the default gateway. */
1775 if (c->local_gateway
1776 && inet_aton(c->local_gateway, &gateway)
1777 && gateway.s_addr) {
1778 if (!netdev_add_router(netdev, gateway)) {
1779 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1780 br->name, IP_ARGS(&gateway.s_addr));
1786 bridge_reconfigure_remotes(struct bridge *br,
1787 const struct sockaddr_in *managers,
1790 const char *disable_ib_str, *queue_id_str;
1791 bool disable_in_band = false;
1794 struct ovsrec_controller **controllers;
1795 size_t n_controllers;
1798 struct ofproto_controller *ocs;
1802 /* Check if we should disable in-band control on this bridge. */
1803 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1804 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1805 disable_in_band = true;
1808 /* Set OpenFlow queue ID for in-band control. */
1809 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
1810 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
1811 ofproto_set_in_band_queue(br->ofproto, queue_id);
1813 if (disable_in_band) {
1814 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1816 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1818 had_primary = ofproto_has_primary_controller(br->ofproto);
1820 n_controllers = bridge_get_controllers(br, &controllers);
1822 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1825 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1826 for (i = 0; i < n_controllers; i++) {
1827 struct ovsrec_controller *c = controllers[i];
1829 if (!strncmp(c->target, "punix:", 6)
1830 || !strncmp(c->target, "unix:", 5)) {
1831 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1833 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1834 * domain sockets and overwriting arbitrary local files. */
1835 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1836 "\"%s\" due to possibility for remote exploit",
1837 dpif_name(br->dpif), c->target);
1841 bridge_configure_local_iface_netdev(br, c);
1842 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
1843 if (disable_in_band) {
1844 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
1849 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1850 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1853 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1854 ofproto_flush_flows(br->ofproto);
1857 /* If there are no controllers and the bridge is in standalone
1858 * mode, set up a flow that matches every packet and directs
1859 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1860 * switch is in secure mode and we won't pass any traffic until
1861 * a controller has been defined and it tells us to do so. */
1863 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1864 union ofp_action action;
1865 struct cls_rule rule;
1867 memset(&action, 0, sizeof action);
1868 action.type = htons(OFPAT_OUTPUT);
1869 action.output.len = htons(sizeof action);
1870 action.output.port = htons(OFPP_NORMAL);
1871 cls_rule_init_catchall(&rule, 0);
1872 ofproto_add_flow(br->ofproto, &rule, &action, 1);
1877 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1882 for (i = 0; i < br->n_ports; i++) {
1883 struct port *port = br->ports[i];
1884 for (j = 0; j < port->n_ifaces; j++) {
1885 struct iface *iface = port->ifaces[j];
1886 shash_add_once(ifaces, iface->name, iface);
1888 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1889 shash_add_once(ifaces, port->name, NULL);
1894 /* For robustness, in case the administrator moves around datapath ports behind
1895 * our back, we re-check all the datapath port numbers here.
1897 * This function will set the 'dp_ifidx' members of interfaces that have
1898 * disappeared to -1, so only call this function from a context where those
1899 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1900 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1901 * datapath, which doesn't support UINT16_MAX+1 ports. */
1903 bridge_fetch_dp_ifaces(struct bridge *br)
1905 struct odp_port *dpif_ports;
1906 size_t n_dpif_ports;
1909 /* Reset all interface numbers. */
1910 for (i = 0; i < br->n_ports; i++) {
1911 struct port *port = br->ports[i];
1912 for (j = 0; j < port->n_ifaces; j++) {
1913 struct iface *iface = port->ifaces[j];
1914 iface->dp_ifidx = -1;
1917 hmap_clear(&br->ifaces);
1919 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1920 for (i = 0; i < n_dpif_ports; i++) {
1921 struct odp_port *p = &dpif_ports[i];
1922 struct iface *iface = iface_lookup(br, p->devname);
1924 if (iface->dp_ifidx >= 0) {
1925 VLOG_WARN("%s reported interface %s twice",
1926 dpif_name(br->dpif), p->devname);
1927 } else if (iface_from_dp_ifidx(br, p->port)) {
1928 VLOG_WARN("%s reported interface %"PRIu16" twice",
1929 dpif_name(br->dpif), p->port);
1931 iface->dp_ifidx = p->port;
1932 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
1933 hash_int(iface->dp_ifidx, 0));
1936 iface_set_ofport(iface->cfg,
1937 (iface->dp_ifidx >= 0
1938 ? odp_port_to_ofp_port(iface->dp_ifidx)
1945 /* Bridge packet processing functions. */
1948 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1950 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1953 static struct bond_entry *
1954 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1956 return &port->bond_hash[bond_hash(mac)];
1960 bond_choose_iface(const struct port *port)
1962 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1963 size_t i, best_down_slave = -1;
1964 long long next_delay_expiration = LLONG_MAX;
1966 for (i = 0; i < port->n_ifaces; i++) {
1967 struct iface *iface = port->ifaces[i];
1969 if (iface->enabled) {
1971 } else if (iface->delay_expires < next_delay_expiration) {
1972 best_down_slave = i;
1973 next_delay_expiration = iface->delay_expires;
1977 if (best_down_slave != -1) {
1978 struct iface *iface = port->ifaces[best_down_slave];
1980 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1981 "since no other interface is up", iface->name,
1982 iface->delay_expires - time_msec());
1983 bond_enable_slave(iface, true);
1986 return best_down_slave;
1990 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1991 uint16_t *dp_ifidx, tag_type *tags)
1993 struct iface *iface;
1995 assert(port->n_ifaces);
1996 if (port->n_ifaces == 1) {
1997 iface = port->ifaces[0];
1999 struct bond_entry *e = lookup_bond_entry(port, dl_src);
2000 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2001 || !port->ifaces[e->iface_idx]->enabled) {
2002 /* XXX select interface properly. The current interface selection
2003 * is only good for testing the rebalancing code. */
2004 e->iface_idx = bond_choose_iface(port);
2005 if (e->iface_idx < 0) {
2006 *tags |= port->no_ifaces_tag;
2009 e->iface_tag = tag_create_random();
2010 ((struct port *) port)->bond_compat_is_stale = true;
2012 *tags |= e->iface_tag;
2013 iface = port->ifaces[e->iface_idx];
2015 *dp_ifidx = iface->dp_ifidx;
2016 *tags |= iface->tag; /* Currently only used for bonding. */
2021 bond_link_status_update(struct iface *iface, bool carrier)
2023 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2024 struct port *port = iface->port;
2026 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2027 /* Nothing to do. */
2030 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
2031 iface->name, carrier ? "detected" : "dropped");
2032 if (carrier == iface->enabled) {
2033 iface->delay_expires = LLONG_MAX;
2034 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2035 iface->name, carrier ? "disabled" : "enabled");
2036 } else if (carrier && port->active_iface < 0) {
2037 bond_enable_slave(iface, true);
2038 if (port->updelay) {
2039 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2040 "other interface is up", iface->name, port->updelay);
2043 int delay = carrier ? port->updelay : port->downdelay;
2044 iface->delay_expires = time_msec() + delay;
2047 "interface %s: will be %s if it stays %s for %d ms",
2049 carrier ? "enabled" : "disabled",
2050 carrier ? "up" : "down",
2057 bond_choose_active_iface(struct port *port)
2059 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2061 port->active_iface = bond_choose_iface(port);
2062 port->active_iface_tag = tag_create_random();
2063 if (port->active_iface >= 0) {
2064 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2065 port->name, port->ifaces[port->active_iface]->name);
2067 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2073 bond_enable_slave(struct iface *iface, bool enable)
2075 struct port *port = iface->port;
2076 struct bridge *br = port->bridge;
2078 /* This acts as a recursion check. If the act of disabling a slave
2079 * causes a different slave to be enabled, the flag will allow us to
2080 * skip redundant work when we reenter this function. It must be
2081 * cleared on exit to keep things safe with multiple bonds. */
2082 static bool moving_active_iface = false;
2084 iface->delay_expires = LLONG_MAX;
2085 if (enable == iface->enabled) {
2089 iface->enabled = enable;
2090 if (!iface->enabled) {
2091 VLOG_WARN("interface %s: disabled", iface->name);
2092 ofproto_revalidate(br->ofproto, iface->tag);
2093 if (iface->port_ifidx == port->active_iface) {
2094 ofproto_revalidate(br->ofproto,
2095 port->active_iface_tag);
2097 /* Disabling a slave can lead to another slave being immediately
2098 * enabled if there will be no active slaves but one is waiting
2099 * on an updelay. In this case we do not need to run most of the
2100 * code for the newly enabled slave since there was no period
2101 * without an active slave and it is redundant with the disabling
2103 moving_active_iface = true;
2104 bond_choose_active_iface(port);
2106 bond_send_learning_packets(port);
2108 VLOG_WARN("interface %s: enabled", iface->name);
2109 if (port->active_iface < 0 && !moving_active_iface) {
2110 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2111 bond_choose_active_iface(port);
2112 bond_send_learning_packets(port);
2114 iface->tag = tag_create_random();
2117 moving_active_iface = false;
2118 port->bond_compat_is_stale = true;
2121 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2122 * bond interface. */
2124 bond_update_fake_iface_stats(struct port *port)
2126 struct netdev_stats bond_stats;
2127 struct netdev *bond_dev;
2130 memset(&bond_stats, 0, sizeof bond_stats);
2132 for (i = 0; i < port->n_ifaces; i++) {
2133 struct netdev_stats slave_stats;
2135 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2136 /* XXX: We swap the stats here because they are swapped back when
2137 * reported by the internal device. The reason for this is
2138 * internal devices normally represent packets going into the system
2139 * but when used as fake bond device they represent packets leaving
2140 * the system. We really should do this in the internal device
2141 * itself because changing it here reverses the counts from the
2142 * perspective of the switch. However, the internal device doesn't
2143 * know what type of device it represents so we have to do it here
2145 bond_stats.tx_packets += slave_stats.rx_packets;
2146 bond_stats.tx_bytes += slave_stats.rx_bytes;
2147 bond_stats.rx_packets += slave_stats.tx_packets;
2148 bond_stats.rx_bytes += slave_stats.tx_bytes;
2152 if (!netdev_open_default(port->name, &bond_dev)) {
2153 netdev_set_stats(bond_dev, &bond_stats);
2154 netdev_close(bond_dev);
2159 bond_run(struct bridge *br)
2163 for (i = 0; i < br->n_ports; i++) {
2164 struct port *port = br->ports[i];
2166 if (port->n_ifaces >= 2) {
2169 /* Track carrier going up and down on interfaces. */
2170 while (!netdev_monitor_poll(port->monitor, &devname)) {
2171 struct iface *iface;
2173 iface = port_lookup_iface(port, devname);
2175 bool carrier = netdev_get_carrier(iface->netdev);
2177 bond_link_status_update(iface, carrier);
2178 port_update_bond_compat(port);
2183 for (j = 0; j < port->n_ifaces; j++) {
2184 struct iface *iface = port->ifaces[j];
2185 if (time_msec() >= iface->delay_expires) {
2186 bond_enable_slave(iface, !iface->enabled);
2190 if (port->bond_fake_iface
2191 && time_msec() >= port->bond_next_fake_iface_update) {
2192 bond_update_fake_iface_stats(port);
2193 port->bond_next_fake_iface_update = time_msec() + 1000;
2197 if (port->bond_compat_is_stale) {
2198 port->bond_compat_is_stale = false;
2199 port_update_bond_compat(port);
2205 bond_wait(struct bridge *br)
2209 for (i = 0; i < br->n_ports; i++) {
2210 struct port *port = br->ports[i];
2211 if (port->n_ifaces < 2) {
2214 netdev_monitor_poll_wait(port->monitor);
2215 for (j = 0; j < port->n_ifaces; j++) {
2216 struct iface *iface = port->ifaces[j];
2217 if (iface->delay_expires != LLONG_MAX) {
2218 poll_timer_wait_until(iface->delay_expires);
2221 if (port->bond_fake_iface) {
2222 poll_timer_wait_until(port->bond_next_fake_iface_update);
2228 set_dst(struct dst *p, const struct flow *flow,
2229 const struct port *in_port, const struct port *out_port,
2232 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2233 : in_port->vlan >= 0 ? in_port->vlan
2234 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2235 : vlan_tci_to_vid(flow->vlan_tci));
2236 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
2240 swap_dst(struct dst *p, struct dst *q)
2242 struct dst tmp = *p;
2247 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2248 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2249 * that we push to the datapath. We could in fact fully sort the array by
2250 * vlan, but in most cases there are at most two different vlan tags so that's
2251 * possibly overkill.) */
2253 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2255 struct dst *first = dsts;
2256 struct dst *last = dsts + n_dsts;
2258 while (first != last) {
2260 * - All dsts < first have vlan == 'vlan'.
2261 * - All dsts >= last have vlan != 'vlan'.
2262 * - first < last. */
2263 while (first->vlan == vlan) {
2264 if (++first == last) {
2269 /* Same invariants, plus one additional:
2270 * - first->vlan != vlan.
2272 while (last[-1].vlan != vlan) {
2273 if (--last == first) {
2278 /* Same invariants, plus one additional:
2279 * - last[-1].vlan == vlan.*/
2280 swap_dst(first++, --last);
2285 mirror_mask_ffs(mirror_mask_t mask)
2287 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2292 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2293 const struct dst *test)
2296 for (i = 0; i < n_dsts; i++) {
2297 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2305 port_trunks_vlan(const struct port *port, uint16_t vlan)
2307 return (port->vlan < 0
2308 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2312 port_includes_vlan(const struct port *port, uint16_t vlan)
2314 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2318 port_is_floodable(const struct port *port)
2322 for (i = 0; i < port->n_ifaces; i++) {
2323 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2324 port->ifaces[i]->dp_ifidx)) {
2332 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2333 const struct port *in_port, const struct port *out_port,
2334 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2336 mirror_mask_t mirrors = in_port->src_mirrors;
2338 struct dst *dst = dsts;
2341 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2342 if (flow_vlan == 0) {
2343 flow_vlan = OFP_VLAN_NONE;
2346 if (out_port == FLOOD_PORT) {
2347 /* XXX use ODP_FLOOD if no vlans or bonding. */
2348 /* XXX even better, define each VLAN as a datapath port group */
2349 for (i = 0; i < br->n_ports; i++) {
2350 struct port *port = br->ports[i];
2352 && port_is_floodable(port)
2353 && port_includes_vlan(port, vlan)
2354 && !port->is_mirror_output_port
2355 && set_dst(dst, flow, in_port, port, tags)) {
2356 mirrors |= port->dst_mirrors;
2360 *nf_output_iface = NF_OUT_FLOOD;
2361 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2362 *nf_output_iface = dst->dp_ifidx;
2363 mirrors |= out_port->dst_mirrors;
2368 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2369 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2371 if (set_dst(dst, flow, in_port, m->out_port, tags)
2372 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2376 for (i = 0; i < br->n_ports; i++) {
2377 struct port *port = br->ports[i];
2378 if (port_includes_vlan(port, m->out_vlan)
2379 && set_dst(dst, flow, in_port, port, tags))
2382 if (port->vlan < 0) {
2383 dst->vlan = m->out_vlan;
2385 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2389 /* Use the vlan tag on the original flow instead of
2390 * the one passed in the vlan parameter. This ensures
2391 * that we compare the vlan from before any implicit
2392 * tagging tags place. This is necessary because
2393 * dst->vlan is the final vlan, after removing implicit
2395 if (port == in_port && dst->vlan == flow_vlan) {
2396 /* Don't send out input port on same VLAN. */
2404 mirrors &= mirrors - 1;
2407 partition_dsts(dsts, dst - dsts, flow_vlan);
2411 static void OVS_UNUSED
2412 print_dsts(const struct dst *dsts, size_t n)
2414 for (; n--; dsts++) {
2415 printf(">p%"PRIu16, dsts->dp_ifidx);
2416 if (dsts->vlan != OFP_VLAN_NONE) {
2417 printf("v%"PRIu16, dsts->vlan);
2423 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2424 const struct port *in_port, const struct port *out_port,
2425 tag_type *tags, struct odp_actions *actions,
2426 uint16_t *nf_output_iface)
2428 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2430 const struct dst *p;
2433 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2436 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2437 if (cur_vlan == 0) {
2438 cur_vlan = OFP_VLAN_NONE;
2440 for (p = dsts; p < &dsts[n_dsts]; p++) {
2441 union odp_action *a;
2442 if (p->vlan != cur_vlan) {
2443 if (p->vlan == OFP_VLAN_NONE) {
2444 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2446 a = odp_actions_add(actions, ODPAT_SET_DL_TCI);
2447 a->dl_tci.tci = htons(p->vlan & VLAN_VID_MASK);
2448 a->dl_tci.tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2452 a = odp_actions_add(actions, ODPAT_OUTPUT);
2453 a->output.port = p->dp_ifidx;
2457 /* Returns the effective vlan of a packet, taking into account both the
2458 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2459 * the packet is untagged and -1 indicates it has an invalid header and
2460 * should be dropped. */
2461 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2462 struct port *in_port, bool have_packet)
2464 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2465 if (in_port->vlan >= 0) {
2467 /* XXX support double tagging? */
2469 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2470 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2471 "packet received on port %s configured with "
2472 "implicit VLAN %"PRIu16,
2473 br->name, vlan, in_port->name, in_port->vlan);
2477 vlan = in_port->vlan;
2479 if (!port_includes_vlan(in_port, vlan)) {
2481 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2482 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2483 "packet received on port %s not configured for "
2485 br->name, vlan, in_port->name, vlan);
2494 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2495 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2496 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2498 is_gratuitous_arp(const struct flow *flow)
2500 return (flow->dl_type == htons(ETH_TYPE_ARP)
2501 && eth_addr_is_broadcast(flow->dl_dst)
2502 && (flow->nw_proto == ARP_OP_REPLY
2503 || (flow->nw_proto == ARP_OP_REQUEST
2504 && flow->nw_src == flow->nw_dst)));
2508 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2509 struct port *in_port)
2511 enum grat_arp_lock_type lock_type;
2514 /* We don't want to learn from gratuitous ARP packets that are reflected
2515 * back over bond slaves so we lock the learning table. */
2516 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2517 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2518 GRAT_ARP_LOCK_CHECK;
2520 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2523 /* The log messages here could actually be useful in debugging,
2524 * so keep the rate limit relatively high. */
2525 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2527 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2528 "on port %s in VLAN %d",
2529 br->name, ETH_ADDR_ARGS(flow->dl_src),
2530 in_port->name, vlan);
2531 ofproto_revalidate(br->ofproto, rev_tag);
2535 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2536 * dropped. Returns true if they may be forwarded, false if they should be
2539 * If 'have_packet' is true, it indicates that the caller is processing a
2540 * received packet. If 'have_packet' is false, then the caller is just
2541 * revalidating an existing flow because configuration has changed. Either
2542 * way, 'have_packet' only affects logging (there is no point in logging errors
2543 * during revalidation).
2545 * Sets '*in_portp' to the input port. This will be a null pointer if
2546 * flow->in_port does not designate a known input port (in which case
2547 * is_admissible() returns false).
2549 * When returning true, sets '*vlanp' to the effective VLAN of the input
2550 * packet, as returned by flow_get_vlan().
2552 * May also add tags to '*tags', although the current implementation only does
2553 * so in one special case.
2556 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2557 tag_type *tags, int *vlanp, struct port **in_portp)
2559 struct iface *in_iface;
2560 struct port *in_port;
2563 /* Find the interface and port structure for the received packet. */
2564 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2566 /* No interface? Something fishy... */
2568 /* Odd. A few possible reasons here:
2570 * - We deleted an interface but there are still a few packets
2571 * queued up from it.
2573 * - Someone externally added an interface (e.g. with "ovs-dpctl
2574 * add-if") that we don't know about.
2576 * - Packet arrived on the local port but the local port is not
2577 * one of our bridge ports.
2579 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2581 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2582 "interface %"PRIu16, br->name, flow->in_port);
2588 *in_portp = in_port = in_iface->port;
2589 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2594 /* Drop frames for reserved multicast addresses. */
2595 if (eth_addr_is_reserved(flow->dl_dst)) {
2599 /* Drop frames on ports reserved for mirroring. */
2600 if (in_port->is_mirror_output_port) {
2602 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2603 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2604 "%s, which is reserved exclusively for mirroring",
2605 br->name, in_port->name);
2610 /* Packets received on bonds need special attention to avoid duplicates. */
2611 if (in_port->n_ifaces > 1) {
2613 bool is_grat_arp_locked;
2615 if (eth_addr_is_multicast(flow->dl_dst)) {
2616 *tags |= in_port->active_iface_tag;
2617 if (in_port->active_iface != in_iface->port_ifidx) {
2618 /* Drop all multicast packets on inactive slaves. */
2623 /* Drop all packets for which we have learned a different input
2624 * port, because we probably sent the packet on one slave and got
2625 * it back on the other. Gratuitous ARP packets are an exception
2626 * to this rule: the host has moved to another switch. The exception
2627 * to the exception is if we locked the learning table to avoid
2628 * reflections on bond slaves. If this is the case, just drop the
2630 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2631 &is_grat_arp_locked);
2632 if (src_idx != -1 && src_idx != in_port->port_idx &&
2633 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2641 /* If the composed actions may be applied to any packet in the given 'flow',
2642 * returns true. Otherwise, the actions should only be applied to 'packet', or
2643 * not at all, if 'packet' was NULL. */
2645 process_flow(struct bridge *br, const struct flow *flow,
2646 const struct ofpbuf *packet, struct odp_actions *actions,
2647 tag_type *tags, uint16_t *nf_output_iface)
2649 struct port *in_port;
2650 struct port *out_port;
2654 /* Check whether we should drop packets in this flow. */
2655 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2660 /* Learn source MAC (but don't try to learn from revalidation). */
2662 update_learning_table(br, flow, vlan, in_port);
2665 /* Determine output port. */
2666 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2668 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2669 out_port = br->ports[out_port_idx];
2670 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2671 /* If we are revalidating but don't have a learning entry then
2672 * eject the flow. Installing a flow that floods packets opens
2673 * up a window of time where we could learn from a packet reflected
2674 * on a bond and blackhole packets before the learning table is
2675 * updated to reflect the correct port. */
2678 out_port = FLOOD_PORT;
2681 /* Don't send packets out their input ports. */
2682 if (in_port == out_port) {
2688 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2696 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2697 struct odp_actions *actions, tag_type *tags,
2698 uint16_t *nf_output_iface, void *br_)
2700 struct iface *iface;
2701 struct bridge *br = br_;
2703 COVERAGE_INC(bridge_process_flow);
2705 iface = iface_from_dp_ifidx(br, flow->in_port);
2707 if (cfm_should_process_flow(flow)) {
2708 if (packet && iface->cfm) {
2709 cfm_process_heartbeat(iface->cfm, packet);
2714 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2718 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
2719 const union odp_action *actions,
2720 size_t n_actions, unsigned long long int n_bytes,
2723 struct bridge *br = br_;
2724 const union odp_action *a;
2725 struct port *in_port;
2729 /* Feed information from the active flows back into the learning table to
2730 * ensure that table is always in sync with what is actually flowing
2731 * through the datapath.
2733 * We test that 'tags' is nonzero to ensure that only flows that include an
2734 * OFPP_NORMAL action are used for learning. This works because
2735 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2736 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2737 update_learning_table(br, flow, vlan, in_port);
2740 /* Account for bond slave utilization. */
2741 if (!br->has_bonded_ports) {
2744 for (a = actions; a < &actions[n_actions]; a++) {
2745 if (a->type == ODPAT_OUTPUT) {
2746 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2747 if (out_port && out_port->n_ifaces >= 2) {
2748 struct bond_entry *e = lookup_bond_entry(out_port,
2750 e->tx_bytes += n_bytes;
2757 bridge_account_checkpoint_ofhook_cb(void *br_)
2759 struct bridge *br = br_;
2763 if (!br->has_bonded_ports) {
2768 for (i = 0; i < br->n_ports; i++) {
2769 struct port *port = br->ports[i];
2770 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2771 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2772 bond_rebalance_port(port);
2777 static struct ofhooks bridge_ofhooks = {
2778 bridge_normal_ofhook_cb,
2779 bridge_account_flow_ofhook_cb,
2780 bridge_account_checkpoint_ofhook_cb,
2783 /* Bonding functions. */
2785 /* Statistics for a single interface on a bonded port, used for load-based
2786 * bond rebalancing. */
2787 struct slave_balance {
2788 struct iface *iface; /* The interface. */
2789 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2791 /* All the "bond_entry"s that are assigned to this interface, in order of
2792 * increasing tx_bytes. */
2793 struct bond_entry **hashes;
2797 /* Sorts pointers to pointers to bond_entries in ascending order by the
2798 * interface to which they are assigned, and within a single interface in
2799 * ascending order of bytes transmitted. */
2801 compare_bond_entries(const void *a_, const void *b_)
2803 const struct bond_entry *const *ap = a_;
2804 const struct bond_entry *const *bp = b_;
2805 const struct bond_entry *a = *ap;
2806 const struct bond_entry *b = *bp;
2807 if (a->iface_idx != b->iface_idx) {
2808 return a->iface_idx > b->iface_idx ? 1 : -1;
2809 } else if (a->tx_bytes != b->tx_bytes) {
2810 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2816 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2817 * *descending* order by number of bytes transmitted. */
2819 compare_slave_balance(const void *a_, const void *b_)
2821 const struct slave_balance *a = a_;
2822 const struct slave_balance *b = b_;
2823 if (a->iface->enabled != b->iface->enabled) {
2824 return a->iface->enabled ? -1 : 1;
2825 } else if (a->tx_bytes != b->tx_bytes) {
2826 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2833 swap_bals(struct slave_balance *a, struct slave_balance *b)
2835 struct slave_balance tmp = *a;
2840 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2841 * given that 'p' (and only 'p') might be in the wrong location.
2843 * This function invalidates 'p', since it might now be in a different memory
2846 resort_bals(struct slave_balance *p,
2847 struct slave_balance bals[], size_t n_bals)
2850 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2851 swap_bals(p, p - 1);
2853 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2854 swap_bals(p, p + 1);
2860 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2862 if (VLOG_IS_DBG_ENABLED()) {
2863 struct ds ds = DS_EMPTY_INITIALIZER;
2864 const struct slave_balance *b;
2866 for (b = bals; b < bals + n_bals; b++) {
2870 ds_put_char(&ds, ',');
2872 ds_put_format(&ds, " %s %"PRIu64"kB",
2873 b->iface->name, b->tx_bytes / 1024);
2875 if (!b->iface->enabled) {
2876 ds_put_cstr(&ds, " (disabled)");
2878 if (b->n_hashes > 0) {
2879 ds_put_cstr(&ds, " (");
2880 for (i = 0; i < b->n_hashes; i++) {
2881 const struct bond_entry *e = b->hashes[i];
2883 ds_put_cstr(&ds, " + ");
2885 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2886 e - port->bond_hash, e->tx_bytes / 1024);
2888 ds_put_cstr(&ds, ")");
2891 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2896 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2898 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2901 struct bond_entry *hash = from->hashes[hash_idx];
2902 struct port *port = from->iface->port;
2903 uint64_t delta = hash->tx_bytes;
2905 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2906 "from %s to %s (now carrying %"PRIu64"kB and "
2907 "%"PRIu64"kB load, respectively)",
2908 port->name, delta / 1024, hash - port->bond_hash,
2909 from->iface->name, to->iface->name,
2910 (from->tx_bytes - delta) / 1024,
2911 (to->tx_bytes + delta) / 1024);
2913 /* Delete element from from->hashes.
2915 * We don't bother to add the element to to->hashes because not only would
2916 * it require more work, the only purpose it would be to allow that hash to
2917 * be migrated to another slave in this rebalancing run, and there is no
2918 * point in doing that. */
2919 if (hash_idx == 0) {
2922 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2923 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2927 /* Shift load away from 'from' to 'to'. */
2928 from->tx_bytes -= delta;
2929 to->tx_bytes += delta;
2931 /* Arrange for flows to be revalidated. */
2932 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2933 hash->iface_idx = to->iface->port_ifidx;
2934 hash->iface_tag = tag_create_random();
2938 bond_rebalance_port(struct port *port)
2940 struct slave_balance bals[DP_MAX_PORTS];
2942 struct bond_entry *hashes[BOND_MASK + 1];
2943 struct slave_balance *b, *from, *to;
2944 struct bond_entry *e;
2947 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2948 * descending order of tx_bytes, so that bals[0] represents the most
2949 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2952 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2953 * array for each slave_balance structure, we sort our local array of
2954 * hashes in order by slave, so that all of the hashes for a given slave
2955 * become contiguous in memory, and then we point each 'hashes' members of
2956 * a slave_balance structure to the start of a contiguous group. */
2957 n_bals = port->n_ifaces;
2958 for (b = bals; b < &bals[n_bals]; b++) {
2959 b->iface = port->ifaces[b - bals];
2964 for (i = 0; i <= BOND_MASK; i++) {
2965 hashes[i] = &port->bond_hash[i];
2967 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2968 for (i = 0; i <= BOND_MASK; i++) {
2970 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2971 b = &bals[e->iface_idx];
2972 b->tx_bytes += e->tx_bytes;
2974 b->hashes = &hashes[i];
2979 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2980 log_bals(bals, n_bals, port);
2982 /* Discard slaves that aren't enabled (which were sorted to the back of the
2983 * array earlier). */
2984 while (!bals[n_bals - 1].iface->enabled) {
2991 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2992 to = &bals[n_bals - 1];
2993 for (from = bals; from < to; ) {
2994 uint64_t overload = from->tx_bytes - to->tx_bytes;
2995 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2996 /* The extra load on 'from' (and all less-loaded slaves), compared
2997 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2998 * it is less than ~1Mbps. No point in rebalancing. */
3000 } else if (from->n_hashes == 1) {
3001 /* 'from' only carries a single MAC hash, so we can't shift any
3002 * load away from it, even though we want to. */
3005 /* 'from' is carrying significantly more load than 'to', and that
3006 * load is split across at least two different hashes. Pick a hash
3007 * to migrate to 'to' (the least-loaded slave), given that doing so
3008 * must decrease the ratio of the load on the two slaves by at
3011 * The sort order we use means that we prefer to shift away the
3012 * smallest hashes instead of the biggest ones. There is little
3013 * reason behind this decision; we could use the opposite sort
3014 * order to shift away big hashes ahead of small ones. */
3017 for (i = 0; i < from->n_hashes; i++) {
3018 double old_ratio, new_ratio;
3019 uint64_t delta = from->hashes[i]->tx_bytes;
3021 if (delta == 0 || from->tx_bytes - delta == 0) {
3022 /* Pointless move. */
3026 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3028 if (to->tx_bytes == 0) {
3029 /* Nothing on the new slave, move it. */
3033 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3034 new_ratio = (double)(from->tx_bytes - delta) /
3035 (to->tx_bytes + delta);
3037 if (new_ratio == 0) {
3038 /* Should already be covered but check to prevent division
3043 if (new_ratio < 1) {
3044 new_ratio = 1 / new_ratio;
3047 if (old_ratio - new_ratio > 0.1) {
3048 /* Would decrease the ratio, move it. */
3052 if (i < from->n_hashes) {
3053 bond_shift_load(from, to, i);
3054 port->bond_compat_is_stale = true;
3056 /* If the result of the migration changed the relative order of
3057 * 'from' and 'to' swap them back to maintain invariants. */
3058 if (order_swapped) {
3059 swap_bals(from, to);
3062 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3063 * point to different slave_balance structures. It is only
3064 * valid to do these two operations in a row at all because we
3065 * know that 'from' will not move past 'to' and vice versa. */
3066 resort_bals(from, bals, n_bals);
3067 resort_bals(to, bals, n_bals);
3074 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3075 * historical data to decay to <1% in 7 rebalancing runs. */
3076 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3082 bond_send_learning_packets(struct port *port)
3084 struct bridge *br = port->bridge;
3085 struct mac_entry *e;
3086 struct ofpbuf packet;
3087 int error, n_packets, n_errors;
3089 if (!port->n_ifaces || port->active_iface < 0) {
3093 ofpbuf_init(&packet, 128);
3094 error = n_packets = n_errors = 0;
3095 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3096 union ofp_action actions[2], *a;
3102 if (e->port == port->port_idx
3103 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
3107 /* Compose actions. */
3108 memset(actions, 0, sizeof actions);
3111 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3112 a->vlan_vid.len = htons(sizeof *a);
3113 a->vlan_vid.vlan_vid = htons(e->vlan);
3116 a->output.type = htons(OFPAT_OUTPUT);
3117 a->output.len = htons(sizeof *a);
3118 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3123 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3125 flow_extract(&packet, 0, ODPP_NONE, &flow);
3126 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3133 ofpbuf_uninit(&packet);
3136 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3137 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3138 "packets, last error was: %s",
3139 port->name, n_errors, n_packets, strerror(error));
3141 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3142 port->name, n_packets);
3146 /* Bonding unixctl user interface functions. */
3149 bond_unixctl_list(struct unixctl_conn *conn,
3150 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3152 struct ds ds = DS_EMPTY_INITIALIZER;
3153 const struct bridge *br;
3155 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
3157 LIST_FOR_EACH (br, node, &all_bridges) {
3160 for (i = 0; i < br->n_ports; i++) {
3161 const struct port *port = br->ports[i];
3162 if (port->n_ifaces > 1) {
3165 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
3166 for (j = 0; j < port->n_ifaces; j++) {
3167 const struct iface *iface = port->ifaces[j];
3169 ds_put_cstr(&ds, ", ");
3171 ds_put_cstr(&ds, iface->name);
3173 ds_put_char(&ds, '\n');
3177 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3181 static struct port *
3182 bond_find(const char *name)
3184 const struct bridge *br;
3186 LIST_FOR_EACH (br, node, &all_bridges) {
3189 for (i = 0; i < br->n_ports; i++) {
3190 struct port *port = br->ports[i];
3191 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3200 bond_unixctl_show(struct unixctl_conn *conn,
3201 const char *args, void *aux OVS_UNUSED)
3203 struct ds ds = DS_EMPTY_INITIALIZER;
3204 const struct port *port;
3207 port = bond_find(args);
3209 unixctl_command_reply(conn, 501, "no such bond");
3213 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3214 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3215 ds_put_format(&ds, "next rebalance: %lld ms\n",
3216 port->bond_next_rebalance - time_msec());
3217 for (j = 0; j < port->n_ifaces; j++) {
3218 const struct iface *iface = port->ifaces[j];
3219 struct bond_entry *be;
3222 ds_put_format(&ds, "slave %s: %s\n",
3223 iface->name, iface->enabled ? "enabled" : "disabled");
3224 if (j == port->active_iface) {
3225 ds_put_cstr(&ds, "\tactive slave\n");
3227 if (iface->delay_expires != LLONG_MAX) {
3228 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3229 iface->enabled ? "downdelay" : "updelay",
3230 iface->delay_expires - time_msec());
3234 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3235 int hash = be - port->bond_hash;
3236 struct mac_entry *me;
3238 if (be->iface_idx != j) {
3242 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3243 hash, be->tx_bytes / 1024);
3246 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3249 if (bond_hash(me->mac) == hash
3250 && me->port != port->port_idx
3251 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3252 && dp_ifidx == iface->dp_ifidx)
3254 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3255 ETH_ADDR_ARGS(me->mac));
3260 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3265 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3266 void *aux OVS_UNUSED)
3268 char *args = (char *) args_;
3269 char *save_ptr = NULL;
3270 char *bond_s, *hash_s, *slave_s;
3271 uint8_t mac[ETH_ADDR_LEN];
3273 struct iface *iface;
3274 struct bond_entry *entry;
3277 bond_s = strtok_r(args, " ", &save_ptr);
3278 hash_s = strtok_r(NULL, " ", &save_ptr);
3279 slave_s = strtok_r(NULL, " ", &save_ptr);
3281 unixctl_command_reply(conn, 501,
3282 "usage: bond/migrate BOND HASH SLAVE");
3286 port = bond_find(bond_s);
3288 unixctl_command_reply(conn, 501, "no such bond");
3292 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3293 == ETH_ADDR_SCAN_COUNT) {
3294 hash = bond_hash(mac);
3295 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3296 hash = atoi(hash_s) & BOND_MASK;
3298 unixctl_command_reply(conn, 501, "bad hash");
3302 iface = port_lookup_iface(port, slave_s);
3304 unixctl_command_reply(conn, 501, "no such slave");
3308 if (!iface->enabled) {
3309 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3313 entry = &port->bond_hash[hash];
3314 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3315 entry->iface_idx = iface->port_ifidx;
3316 entry->iface_tag = tag_create_random();
3317 port->bond_compat_is_stale = true;
3318 unixctl_command_reply(conn, 200, "migrated");
3322 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3323 void *aux OVS_UNUSED)
3325 char *args = (char *) args_;
3326 char *save_ptr = NULL;
3327 char *bond_s, *slave_s;
3329 struct iface *iface;
3331 bond_s = strtok_r(args, " ", &save_ptr);
3332 slave_s = strtok_r(NULL, " ", &save_ptr);
3334 unixctl_command_reply(conn, 501,
3335 "usage: bond/set-active-slave BOND SLAVE");
3339 port = bond_find(bond_s);
3341 unixctl_command_reply(conn, 501, "no such bond");
3345 iface = port_lookup_iface(port, slave_s);
3347 unixctl_command_reply(conn, 501, "no such slave");
3351 if (!iface->enabled) {
3352 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3356 if (port->active_iface != iface->port_ifidx) {
3357 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3358 port->active_iface = iface->port_ifidx;
3359 port->active_iface_tag = tag_create_random();
3360 VLOG_INFO("port %s: active interface is now %s",
3361 port->name, iface->name);
3362 bond_send_learning_packets(port);
3363 unixctl_command_reply(conn, 200, "done");
3365 unixctl_command_reply(conn, 200, "no change");
3370 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3372 char *args = (char *) args_;
3373 char *save_ptr = NULL;
3374 char *bond_s, *slave_s;
3376 struct iface *iface;
3378 bond_s = strtok_r(args, " ", &save_ptr);
3379 slave_s = strtok_r(NULL, " ", &save_ptr);
3381 unixctl_command_reply(conn, 501,
3382 "usage: bond/enable/disable-slave BOND SLAVE");
3386 port = bond_find(bond_s);
3388 unixctl_command_reply(conn, 501, "no such bond");
3392 iface = port_lookup_iface(port, slave_s);
3394 unixctl_command_reply(conn, 501, "no such slave");
3398 bond_enable_slave(iface, enable);
3399 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3403 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3404 void *aux OVS_UNUSED)
3406 enable_slave(conn, args, true);
3410 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3411 void *aux OVS_UNUSED)
3413 enable_slave(conn, args, false);
3417 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3418 void *aux OVS_UNUSED)
3420 uint8_t mac[ETH_ADDR_LEN];
3424 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3425 == ETH_ADDR_SCAN_COUNT) {
3426 hash = bond_hash(mac);
3428 hash_cstr = xasprintf("%u", hash);
3429 unixctl_command_reply(conn, 200, hash_cstr);
3432 unixctl_command_reply(conn, 501, "invalid mac");
3439 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3440 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3441 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3442 unixctl_command_register("bond/set-active-slave",
3443 bond_unixctl_set_active_slave, NULL);
3444 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3446 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3448 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3451 /* Port functions. */
3453 static struct port *
3454 port_create(struct bridge *br, const char *name)
3458 port = xzalloc(sizeof *port);
3460 port->port_idx = br->n_ports;
3462 port->trunks = NULL;
3463 port->name = xstrdup(name);
3464 port->active_iface = -1;
3466 if (br->n_ports >= br->allocated_ports) {
3467 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3470 br->ports[br->n_ports++] = port;
3471 shash_add_assert(&br->port_by_name, port->name, port);
3473 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3480 get_port_other_config(const struct ovsrec_port *port, const char *key,
3481 const char *default_value)
3485 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3487 return value ? value : default_value;
3491 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3493 struct shash new_ifaces;
3496 /* Collect list of new interfaces. */
3497 shash_init(&new_ifaces);
3498 for (i = 0; i < cfg->n_interfaces; i++) {
3499 const char *name = cfg->interfaces[i]->name;
3500 shash_add_once(&new_ifaces, name, NULL);
3503 /* Get rid of deleted interfaces. */
3504 for (i = 0; i < port->n_ifaces; ) {
3505 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3506 iface_destroy(port->ifaces[i]);
3512 shash_destroy(&new_ifaces);
3516 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3518 struct shash new_ifaces;
3519 long long int next_rebalance;
3520 unsigned long *trunks;
3526 /* Update settings. */
3527 port->updelay = cfg->bond_updelay;
3528 if (port->updelay < 0) {
3531 port->downdelay = cfg->bond_downdelay;
3532 if (port->downdelay < 0) {
3533 port->downdelay = 0;
3535 port->bond_rebalance_interval = atoi(
3536 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3537 if (port->bond_rebalance_interval < 1000) {
3538 port->bond_rebalance_interval = 1000;
3540 next_rebalance = time_msec() + port->bond_rebalance_interval;
3541 if (port->bond_next_rebalance > next_rebalance) {
3542 port->bond_next_rebalance = next_rebalance;
3545 /* Add new interfaces and update 'cfg' member of existing ones. */
3546 shash_init(&new_ifaces);
3547 for (i = 0; i < cfg->n_interfaces; i++) {
3548 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3549 struct iface *iface;
3551 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3552 VLOG_WARN("port %s: %s specified twice as port interface",
3553 port->name, if_cfg->name);
3554 iface_set_ofport(if_cfg, -1);
3558 iface = iface_lookup(port->bridge, if_cfg->name);
3560 if (iface->port != port) {
3561 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3563 port->bridge->name, if_cfg->name, iface->port->name);
3566 iface->cfg = if_cfg;
3568 iface = iface_create(port, if_cfg);
3571 /* Determine interface type. The local port always has type
3572 * "internal". Other ports take their type from the database and
3573 * default to "system" if none is specified. */
3574 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
3575 : if_cfg->type[0] ? if_cfg->type
3578 shash_destroy(&new_ifaces);
3583 if (port->n_ifaces < 2) {
3585 if (vlan >= 0 && vlan <= 4095) {
3586 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3591 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3592 * they even work as-is. But they have not been tested. */
3593 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3597 if (port->vlan != vlan) {
3599 bridge_flush(port->bridge);
3602 /* Get trunked VLANs. */
3604 if (vlan < 0 && cfg->n_trunks) {
3607 trunks = bitmap_allocate(4096);
3609 for (i = 0; i < cfg->n_trunks; i++) {
3610 int trunk = cfg->trunks[i];
3612 bitmap_set1(trunks, trunk);
3618 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3619 port->name, cfg->n_trunks);
3621 if (n_errors == cfg->n_trunks) {
3622 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3624 bitmap_free(trunks);
3627 } else if (vlan >= 0 && cfg->n_trunks) {
3628 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3632 ? port->trunks != NULL
3633 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3634 bridge_flush(port->bridge);
3636 bitmap_free(port->trunks);
3637 port->trunks = trunks;
3641 port_destroy(struct port *port)
3644 struct bridge *br = port->bridge;
3648 proc_net_compat_update_vlan(port->name, NULL, 0);
3649 proc_net_compat_update_bond(port->name, NULL);
3651 for (i = 0; i < MAX_MIRRORS; i++) {
3652 struct mirror *m = br->mirrors[i];
3653 if (m && m->out_port == port) {
3658 while (port->n_ifaces > 0) {
3659 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3662 shash_find_and_delete_assert(&br->port_by_name, port->name);
3664 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3665 del->port_idx = port->port_idx;
3667 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
3669 netdev_monitor_destroy(port->monitor);
3671 bitmap_free(port->trunks);
3678 static struct port *
3679 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3681 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3682 return iface ? iface->port : NULL;
3685 static struct port *
3686 port_lookup(const struct bridge *br, const char *name)
3688 return shash_find_data(&br->port_by_name, name);
3691 static struct iface *
3692 port_lookup_iface(const struct port *port, const char *name)
3694 struct iface *iface = iface_lookup(port->bridge, name);
3695 return iface && iface->port == port ? iface : NULL;
3699 port_update_bonding(struct port *port)
3701 if (port->monitor) {
3702 netdev_monitor_destroy(port->monitor);
3703 port->monitor = NULL;
3705 if (port->n_ifaces < 2) {
3706 /* Not a bonded port. */
3707 if (port->bond_hash) {
3708 free(port->bond_hash);
3709 port->bond_hash = NULL;
3710 port->bond_compat_is_stale = true;
3711 port->bond_fake_iface = false;
3716 if (!port->bond_hash) {
3717 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3718 for (i = 0; i <= BOND_MASK; i++) {
3719 struct bond_entry *e = &port->bond_hash[i];
3723 port->no_ifaces_tag = tag_create_random();
3724 bond_choose_active_iface(port);
3725 port->bond_next_rebalance
3726 = time_msec() + port->bond_rebalance_interval;
3728 if (port->cfg->bond_fake_iface) {
3729 port->bond_next_fake_iface_update = time_msec();
3732 port->bond_compat_is_stale = true;
3733 port->bond_fake_iface = port->cfg->bond_fake_iface;
3735 port->monitor = netdev_monitor_create();
3736 for (i = 0; i < port->n_ifaces; i++) {
3737 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
3743 port_update_bond_compat(struct port *port)
3745 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3746 struct compat_bond bond;
3749 if (port->n_ifaces < 2) {
3750 proc_net_compat_update_bond(port->name, NULL);
3755 bond.updelay = port->updelay;
3756 bond.downdelay = port->downdelay;
3759 bond.hashes = compat_hashes;
3760 if (port->bond_hash) {
3761 const struct bond_entry *e;
3762 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3763 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3764 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3765 cbh->hash = e - port->bond_hash;
3766 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3771 bond.n_slaves = port->n_ifaces;
3772 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3773 for (i = 0; i < port->n_ifaces; i++) {
3774 struct iface *iface = port->ifaces[i];
3775 struct compat_bond_slave *slave = &bond.slaves[i];
3776 slave->name = iface->name;
3778 /* We need to make the same determination as the Linux bonding
3779 * code to determine whether a slave should be consider "up".
3780 * The Linux function bond_miimon_inspect() supports four
3781 * BOND_LINK_* states:
3783 * - BOND_LINK_UP: carrier detected, updelay has passed.
3784 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3785 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3786 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3788 * The function bond_info_show_slave() only considers BOND_LINK_UP
3789 * to be "up" and anything else to be "down".
3791 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3795 netdev_get_etheraddr(iface->netdev, slave->mac);
3798 if (port->bond_fake_iface) {
3799 struct netdev *bond_netdev;
3801 if (!netdev_open_default(port->name, &bond_netdev)) {
3803 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3805 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3807 netdev_close(bond_netdev);
3811 proc_net_compat_update_bond(port->name, &bond);
3816 port_update_vlan_compat(struct port *port)
3818 struct bridge *br = port->bridge;
3819 char *vlandev_name = NULL;
3821 if (port->vlan > 0) {
3822 /* Figure out the name that the VLAN device should actually have, if it
3823 * existed. This takes some work because the VLAN device would not
3824 * have port->name in its name; rather, it would have the trunk port's
3825 * name, and 'port' would be attached to a bridge that also had the
3826 * VLAN device one of its ports. So we need to find a trunk port that
3827 * includes port->vlan.
3829 * There might be more than one candidate. This doesn't happen on
3830 * XenServer, so if it happens we just pick the first choice in
3831 * alphabetical order instead of creating multiple VLAN devices. */
3833 for (i = 0; i < br->n_ports; i++) {
3834 struct port *p = br->ports[i];
3835 if (port_trunks_vlan(p, port->vlan)
3837 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3839 uint8_t ea[ETH_ADDR_LEN];
3840 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3841 if (!eth_addr_is_multicast(ea) &&
3842 !eth_addr_is_reserved(ea) &&
3843 !eth_addr_is_zero(ea)) {
3844 vlandev_name = p->name;
3849 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3852 /* Interface functions. */
3855 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
3858 union ofp_action action;
3860 memset(&action, 0, sizeof action);
3861 action.output.type = htons(OFPAT_OUTPUT);
3862 action.output.len = htons(sizeof action);
3863 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
3865 flow_extract(packet, 0, ODPP_NONE, &flow);
3867 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
3869 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3870 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
3874 static struct iface *
3875 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3877 struct bridge *br = port->bridge;
3878 struct iface *iface;
3879 char *name = if_cfg->name;
3881 iface = xzalloc(sizeof *iface);
3883 iface->port_ifidx = port->n_ifaces;
3884 iface->name = xstrdup(name);
3885 iface->dp_ifidx = -1;
3886 iface->tag = tag_create_random();
3887 iface->delay_expires = LLONG_MAX;
3888 iface->netdev = NULL;
3889 iface->cfg = if_cfg;
3891 shash_add_assert(&br->iface_by_name, iface->name, iface);
3893 if (port->n_ifaces >= port->allocated_ifaces) {
3894 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3895 sizeof *port->ifaces);
3897 port->ifaces[port->n_ifaces++] = iface;
3898 if (port->n_ifaces > 1) {
3899 br->has_bonded_ports = true;
3902 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3910 iface_destroy(struct iface *iface)
3913 struct port *port = iface->port;
3914 struct bridge *br = port->bridge;
3915 bool del_active = port->active_iface == iface->port_ifidx;
3918 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3920 if (iface->dp_ifidx >= 0) {
3921 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
3924 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3925 del->port_ifidx = iface->port_ifidx;
3927 netdev_close(iface->netdev);
3930 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3931 bond_choose_active_iface(port);
3932 bond_send_learning_packets(port);
3935 cfm_destroy(iface->cfm);
3940 bridge_flush(port->bridge);
3944 static struct iface *
3945 iface_lookup(const struct bridge *br, const char *name)
3947 return shash_find_data(&br->iface_by_name, name);
3950 static struct iface *
3951 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3953 struct iface *iface;
3955 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
3956 hash_int(dp_ifidx, 0), &br->ifaces) {
3957 if (iface->dp_ifidx == dp_ifidx) {
3964 /* Set Ethernet address of 'iface', if one is specified in the configuration
3967 iface_set_mac(struct iface *iface)
3969 uint8_t ea[ETH_ADDR_LEN];
3971 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3972 if (eth_addr_is_multicast(ea)) {
3973 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3975 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3976 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3977 iface->name, iface->name);
3979 int error = netdev_set_etheraddr(iface->netdev, ea);
3981 VLOG_ERR("interface %s: setting MAC failed (%s)",
3982 iface->name, strerror(error));
3988 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
3990 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
3993 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
3997 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
3999 * The value strings in '*shash' are taken directly from values[], not copied,
4000 * so the caller should not modify or free them. */
4002 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4003 struct shash *shash)
4008 for (i = 0; i < n; i++) {
4009 shash_add(shash, keys[i], values[i]);
4013 struct iface_delete_queues_cbdata {
4014 struct netdev *netdev;
4015 const struct ovsdb_datum *queues;
4019 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4021 union ovsdb_atom atom;
4023 atom.integer = target;
4024 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4028 iface_delete_queues(unsigned int queue_id,
4029 const struct shash *details OVS_UNUSED, void *cbdata_)
4031 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4033 if (!queue_ids_include(cbdata->queues, queue_id)) {
4034 netdev_delete_queue(cbdata->netdev, queue_id);
4039 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4041 if (!qos || qos->type[0] == '\0') {
4042 netdev_set_qos(iface->netdev, NULL, NULL);
4044 struct iface_delete_queues_cbdata cbdata;
4045 struct shash details;
4048 /* Configure top-level Qos for 'iface'. */
4049 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4050 qos->n_other_config, &details);
4051 netdev_set_qos(iface->netdev, qos->type, &details);
4052 shash_destroy(&details);
4054 /* Deconfigure queues that were deleted. */
4055 cbdata.netdev = iface->netdev;
4056 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4058 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4060 /* Configure queues for 'iface'. */
4061 for (i = 0; i < qos->n_queues; i++) {
4062 const struct ovsrec_queue *queue = qos->value_queues[i];
4063 unsigned int queue_id = qos->key_queues[i];
4065 shash_from_ovs_idl_map(queue->key_other_config,
4066 queue->value_other_config,
4067 queue->n_other_config, &details);
4068 netdev_set_queue(iface->netdev, queue_id, &details);
4069 shash_destroy(&details);
4075 iface_update_cfm(struct iface *iface)
4079 uint16_t *remote_mps;
4080 struct ovsrec_monitor *mon;
4081 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
4083 mon = iface->cfg->monitor;
4089 if (netdev_get_etheraddr(iface->netdev, ea)) {
4090 VLOG_WARN("interface %s: Failed to get ethernet address. "
4091 "Skipping Monitor.", iface->name);
4095 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4096 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4101 iface->cfm = cfm_create();
4105 cfm->mpid = mon->mpid;
4106 cfm->interval = mon->interval ? *mon->interval : 1000;
4108 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
4109 memcpy(cfm->maid, maid, sizeof cfm->maid);
4111 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4112 for(i = 0; i < mon->n_remote_mps; i++) {
4113 remote_mps[i] = mon->remote_mps[i]->mpid;
4115 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
4118 if (!cfm_configure(iface->cfm)) {
4119 cfm_destroy(iface->cfm);
4124 /* Port mirroring. */
4126 static struct mirror *
4127 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4131 for (i = 0; i < MAX_MIRRORS; i++) {
4132 struct mirror *m = br->mirrors[i];
4133 if (m && uuid_equals(uuid, &m->uuid)) {
4141 mirror_reconfigure(struct bridge *br)
4143 unsigned long *rspan_vlans;
4146 /* Get rid of deleted mirrors. */
4147 for (i = 0; i < MAX_MIRRORS; i++) {
4148 struct mirror *m = br->mirrors[i];
4150 const struct ovsdb_datum *mc;
4151 union ovsdb_atom atom;
4153 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4154 atom.uuid = br->mirrors[i]->uuid;
4155 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4161 /* Add new mirrors and reconfigure existing ones. */
4162 for (i = 0; i < br->cfg->n_mirrors; i++) {
4163 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4164 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4166 mirror_reconfigure_one(m, cfg);
4168 mirror_create(br, cfg);
4172 /* Update port reserved status. */
4173 for (i = 0; i < br->n_ports; i++) {
4174 br->ports[i]->is_mirror_output_port = false;
4176 for (i = 0; i < MAX_MIRRORS; i++) {
4177 struct mirror *m = br->mirrors[i];
4178 if (m && m->out_port) {
4179 m->out_port->is_mirror_output_port = true;
4183 /* Update flooded vlans (for RSPAN). */
4185 if (br->cfg->n_flood_vlans) {
4186 rspan_vlans = bitmap_allocate(4096);
4188 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4189 int64_t vlan = br->cfg->flood_vlans[i];
4190 if (vlan >= 0 && vlan < 4096) {
4191 bitmap_set1(rspan_vlans, vlan);
4192 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4195 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4200 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4206 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4211 for (i = 0; ; i++) {
4212 if (i >= MAX_MIRRORS) {
4213 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4214 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4217 if (!br->mirrors[i]) {
4222 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4225 br->mirrors[i] = m = xzalloc(sizeof *m);
4228 m->name = xstrdup(cfg->name);
4229 shash_init(&m->src_ports);
4230 shash_init(&m->dst_ports);
4236 mirror_reconfigure_one(m, cfg);
4240 mirror_destroy(struct mirror *m)
4243 struct bridge *br = m->bridge;
4246 for (i = 0; i < br->n_ports; i++) {
4247 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4248 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4251 shash_destroy(&m->src_ports);
4252 shash_destroy(&m->dst_ports);
4255 m->bridge->mirrors[m->idx] = NULL;
4264 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4265 struct shash *names)
4269 for (i = 0; i < n_ports; i++) {
4270 const char *name = ports[i]->name;
4271 if (port_lookup(m->bridge, name)) {
4272 shash_add_once(names, name, NULL);
4274 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4275 "port %s", m->bridge->name, m->name, name);
4281 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4287 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4289 for (i = 0; i < cfg->n_select_vlan; i++) {
4290 int64_t vlan = cfg->select_vlan[i];
4291 if (vlan < 0 || vlan > 4095) {
4292 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4293 m->bridge->name, m->name, vlan);
4295 (*vlans)[n_vlans++] = vlan;
4302 vlan_is_mirrored(const struct mirror *m, int vlan)
4306 for (i = 0; i < m->n_vlans; i++) {
4307 if (m->vlans[i] == vlan) {
4315 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4319 for (i = 0; i < m->n_vlans; i++) {
4320 if (port_trunks_vlan(p, m->vlans[i])) {
4328 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4330 struct shash src_ports, dst_ports;
4331 mirror_mask_t mirror_bit;
4332 struct port *out_port;
4339 if (strcmp(cfg->name, m->name)) {
4341 m->name = xstrdup(cfg->name);
4344 /* Get output port. */
4345 if (cfg->output_port) {
4346 out_port = port_lookup(m->bridge, cfg->output_port->name);
4348 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4349 m->bridge->name, m->name);
4355 if (cfg->output_vlan) {
4356 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4357 "output vlan; ignoring output vlan",
4358 m->bridge->name, m->name);
4360 } else if (cfg->output_vlan) {
4362 out_vlan = *cfg->output_vlan;
4364 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4365 m->bridge->name, m->name);
4370 shash_init(&src_ports);
4371 shash_init(&dst_ports);
4372 if (cfg->select_all) {
4373 for (i = 0; i < m->bridge->n_ports; i++) {
4374 const char *name = m->bridge->ports[i]->name;
4375 shash_add_once(&src_ports, name, NULL);
4376 shash_add_once(&dst_ports, name, NULL);
4381 /* Get ports, and drop duplicates and ports that don't exist. */
4382 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4384 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4387 /* Get all the vlans, and drop duplicate and invalid vlans. */
4388 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4391 /* Update mirror data. */
4392 if (!shash_equal_keys(&m->src_ports, &src_ports)
4393 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4394 || m->n_vlans != n_vlans
4395 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4396 || m->out_port != out_port
4397 || m->out_vlan != out_vlan) {
4398 bridge_flush(m->bridge);
4400 shash_swap(&m->src_ports, &src_ports);
4401 shash_swap(&m->dst_ports, &dst_ports);
4404 m->n_vlans = n_vlans;
4405 m->out_port = out_port;
4406 m->out_vlan = out_vlan;
4409 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4410 for (i = 0; i < m->bridge->n_ports; i++) {
4411 struct port *port = m->bridge->ports[i];
4413 if (shash_find(&m->src_ports, port->name)
4416 ? port_trunks_any_mirrored_vlan(m, port)
4417 : vlan_is_mirrored(m, port->vlan)))) {
4418 port->src_mirrors |= mirror_bit;
4420 port->src_mirrors &= ~mirror_bit;
4423 if (shash_find(&m->dst_ports, port->name)) {
4424 port->dst_mirrors |= mirror_bit;
4426 port->dst_mirrors &= ~mirror_bit;
4431 shash_destroy(&src_ports);
4432 shash_destroy(&dst_ports);