1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
23 #include <sys/socket.h>
25 #include <openflow/openflow.h>
30 #include <sys/socket.h>
31 #include <sys/types.h>
37 #include "dynamic-string.h"
43 #include "mac-learning.h"
46 #include "ofp-print.h"
48 #include "ofproto/netflow.h"
49 #include "ofproto/ofproto.h"
50 #include "ovsdb-data.h"
52 #include "poll-loop.h"
53 #include "proc-net-compat.h"
57 #include "socket-util.h"
58 #include "stream-ssl.h"
60 #include "system-stats.h"
65 #include "vswitchd/vswitch-idl.h"
66 #include "xenserver.h"
69 #include "sflow_api.h"
71 VLOG_DEFINE_THIS_MODULE(bridge)
79 /* These members are always valid. */
80 struct port *port; /* Containing port. */
81 size_t port_ifidx; /* Index within containing port. */
82 char *name; /* Host network device name. */
83 tag_type tag; /* Tag associated with this interface. */
84 long long delay_expires; /* Time after which 'enabled' may change. */
86 /* These members are valid only after bridge_reconfigure() causes them to
88 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
89 int dp_ifidx; /* Index within kernel datapath. */
90 struct netdev *netdev; /* Network device. */
91 bool enabled; /* May be chosen for flows? */
92 const struct ovsrec_interface *cfg;
95 #define BOND_MASK 0xff
97 int iface_idx; /* Index of assigned iface, or -1 if none. */
98 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
99 tag_type iface_tag; /* Tag associated with iface_idx. */
102 #define MAX_MIRRORS 32
103 typedef uint32_t mirror_mask_t;
104 #define MIRROR_MASK_C(X) UINT32_C(X)
105 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
107 struct bridge *bridge;
110 struct uuid uuid; /* UUID of this "mirror" record in database. */
112 /* Selection criteria. */
113 struct shash src_ports; /* Name is port name; data is always NULL. */
114 struct shash dst_ports; /* Name is port name; data is always NULL. */
119 struct port *out_port;
123 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
125 struct bridge *bridge;
127 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
128 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
129 * NULL if all VLANs are trunked. */
130 const struct ovsrec_port *cfg;
133 /* An ordinary bridge port has 1 interface.
134 * A bridge port for bonding has at least 2 interfaces. */
135 struct iface **ifaces;
136 size_t n_ifaces, allocated_ifaces;
139 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
140 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
141 tag_type active_iface_tag; /* Tag for bcast flows. */
142 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
143 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
144 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
145 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
146 long long int bond_next_fake_iface_update; /* Time of next update. */
147 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
148 long long int bond_next_rebalance; /* Next rebalancing time. */
150 /* Port mirroring info. */
151 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
152 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
153 bool is_mirror_output_port; /* Does port mirroring send frames here? */
156 #define DP_MAX_PORTS 255
158 struct list node; /* Node in global list of bridges. */
159 char *name; /* User-specified arbitrary name. */
160 struct mac_learning *ml; /* MAC learning table. */
161 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
162 const struct ovsrec_bridge *cfg;
164 /* OpenFlow switch processing. */
165 struct ofproto *ofproto; /* OpenFlow switch. */
167 /* Kernel datapath information. */
168 struct dpif *dpif; /* Datapath. */
169 struct hmap ifaces; /* Contains "struct iface"s. */
173 size_t n_ports, allocated_ports;
174 struct shash iface_by_name; /* "struct iface"s indexed by name. */
175 struct shash port_by_name; /* "struct port"s indexed by name. */
178 bool has_bonded_ports;
183 /* Port mirroring. */
184 struct mirror *mirrors[MAX_MIRRORS];
187 /* List of all bridges. */
188 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
190 /* OVSDB IDL used to obtain configuration. */
191 static struct ovsdb_idl *idl;
193 /* Each time this timer expires, the bridge fetches systems and interface
194 * statistics and pushes them into the database. */
195 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
196 static long long int stats_timer = LLONG_MIN;
198 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
199 static void bridge_destroy(struct bridge *);
200 static struct bridge *bridge_lookup(const char *name);
201 static unixctl_cb_func bridge_unixctl_dump_flows;
202 static unixctl_cb_func bridge_unixctl_reconnect;
203 static int bridge_run_one(struct bridge *);
204 static size_t bridge_get_controllers(const struct bridge *br,
205 struct ovsrec_controller ***controllersp);
206 static void bridge_reconfigure_one(struct bridge *);
207 static void bridge_reconfigure_remotes(struct bridge *,
208 const struct sockaddr_in *managers,
210 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
211 static void bridge_fetch_dp_ifaces(struct bridge *);
212 static void bridge_flush(struct bridge *);
213 static void bridge_pick_local_hw_addr(struct bridge *,
214 uint8_t ea[ETH_ADDR_LEN],
215 struct iface **hw_addr_iface);
216 static uint64_t bridge_pick_datapath_id(struct bridge *,
217 const uint8_t bridge_ea[ETH_ADDR_LEN],
218 struct iface *hw_addr_iface);
219 static struct iface *bridge_get_local_iface(struct bridge *);
220 static uint64_t dpid_from_hash(const void *, size_t nbytes);
222 static unixctl_cb_func bridge_unixctl_fdb_show;
224 static void bond_init(void);
225 static void bond_run(struct bridge *);
226 static void bond_wait(struct bridge *);
227 static void bond_rebalance_port(struct port *);
228 static void bond_send_learning_packets(struct port *);
229 static void bond_enable_slave(struct iface *iface, bool enable);
231 static struct port *port_create(struct bridge *, const char *name);
232 static void port_reconfigure(struct port *, const struct ovsrec_port *);
233 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
234 static void port_destroy(struct port *);
235 static struct port *port_lookup(const struct bridge *, const char *name);
236 static struct iface *port_lookup_iface(const struct port *, const char *name);
237 static struct port *port_from_dp_ifidx(const struct bridge *,
239 static void port_update_bond_compat(struct port *);
240 static void port_update_vlan_compat(struct port *);
241 static void port_update_bonding(struct port *);
243 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
244 static void mirror_destroy(struct mirror *);
245 static void mirror_reconfigure(struct bridge *);
246 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
247 static bool vlan_is_mirrored(const struct mirror *, int vlan);
249 static struct iface *iface_create(struct port *port,
250 const struct ovsrec_interface *if_cfg);
251 static void iface_destroy(struct iface *);
252 static struct iface *iface_lookup(const struct bridge *, const char *name);
253 static struct iface *iface_from_dp_ifidx(const struct bridge *,
255 static bool iface_is_internal(const struct bridge *, const char *name);
256 static void iface_set_mac(struct iface *);
257 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
259 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
262 /* Hooks into ofproto processing. */
263 static struct ofhooks bridge_ofhooks;
265 /* Public functions. */
267 /* Initializes the bridge module, configuring it to obtain its configuration
268 * from an OVSDB server accessed over 'remote', which should be a string in a
269 * form acceptable to ovsdb_idl_create(). */
271 bridge_init(const char *remote)
273 /* Create connection to database. */
274 idl = ovsdb_idl_create(remote, &ovsrec_idl_class);
276 ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_cur_cfg);
277 ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_statistics);
278 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
280 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
282 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
283 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
285 ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_ofport);
286 ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_statistics);
287 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
289 /* Register unixctl commands. */
290 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
291 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
293 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
298 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
299 * but for which the ovs-vswitchd configuration 'cfg' is required. */
301 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
303 static bool already_configured_once;
304 struct svec bridge_names;
305 struct svec dpif_names, dpif_types;
308 /* Only do this once per ovs-vswitchd run. */
309 if (already_configured_once) {
312 already_configured_once = true;
314 stats_timer = time_msec() + STATS_INTERVAL;
316 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
317 svec_init(&bridge_names);
318 for (i = 0; i < cfg->n_bridges; i++) {
319 svec_add(&bridge_names, cfg->bridges[i]->name);
321 svec_sort(&bridge_names);
323 /* Iterate over all system dpifs and delete any of them that do not appear
325 svec_init(&dpif_names);
326 svec_init(&dpif_types);
327 dp_enumerate_types(&dpif_types);
328 for (i = 0; i < dpif_types.n; i++) {
333 dp_enumerate_names(dpif_types.names[i], &dpif_names);
335 /* For each dpif... */
336 for (j = 0; j < dpif_names.n; j++) {
337 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
339 struct svec all_names;
342 /* ...check whether any of its names is in 'bridge_names'. */
343 svec_init(&all_names);
344 dpif_get_all_names(dpif, &all_names);
345 for (k = 0; k < all_names.n; k++) {
346 if (svec_contains(&bridge_names, all_names.names[k])) {
351 /* No. Delete the dpif. */
355 svec_destroy(&all_names);
360 svec_destroy(&bridge_names);
361 svec_destroy(&dpif_names);
362 svec_destroy(&dpif_types);
365 /* Initializes 'options' and fills it with the options for 'if_cfg'. Merges
366 * keys from "options" and "other_config", preferring "options" keys over
367 * "other_config" keys.
369 * The value strings in '*options' are taken directly from if_cfg, not copied,
370 * so the caller should not modify or free them. */
372 iface_get_options(const struct ovsrec_interface *if_cfg, struct shash *options)
376 shash_from_ovs_idl_map(if_cfg->key_options, if_cfg->value_options,
377 if_cfg->n_options, options);
379 for (i = 0; i < if_cfg->n_other_config; i++) {
380 char *key = if_cfg->key_other_config[i];
381 char *value = if_cfg->value_other_config[i];
383 if (!shash_find_data(options, key)) {
384 shash_add(options, key, value);
386 VLOG_WARN("%s: ignoring \"other_config\" key %s that conflicts "
387 "with \"options\" key %s", if_cfg->name, key, key);
392 /* Attempt to create the network device for 'iface' through the netdev
395 create_iface_netdev(struct iface *iface)
397 struct netdev_options netdev_options;
398 struct shash options;
401 memset(&netdev_options, 0, sizeof netdev_options);
402 netdev_options.name = iface->cfg->name;
403 if (!strcmp(iface->cfg->type, "internal")) {
404 /* An "internal" config type maps to a netdev "system" type. */
405 netdev_options.type = "system";
407 netdev_options.type = iface->cfg->type;
409 netdev_options.args = &options;
410 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
412 iface_get_options(iface->cfg, &options);
414 error = netdev_open(&netdev_options, &iface->netdev);
417 netdev_get_carrier(iface->netdev, &iface->enabled);
420 shash_destroy(&options);
426 reconfigure_iface_netdev(struct iface *iface)
428 const char *netdev_type, *iface_type;
429 struct shash options;
432 /* Skip reconfiguration if the device has the wrong type. This shouldn't
434 iface_type = (!iface->cfg->type[0] ? NULL
435 : !strcmp(iface->cfg->type, "internal") ? "system"
437 netdev_type = netdev_get_type(iface->netdev);
438 if (iface_type && strcmp(netdev_type, iface_type)) {
439 VLOG_WARN("%s: attempting change device type from %s to %s",
440 iface->cfg->name, netdev_type, iface_type);
444 /* Reconfigure device. */
445 iface_get_options(iface->cfg, &options);
446 error = netdev_reconfigure(iface->netdev, &options);
447 shash_destroy(&options);
453 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
454 void *aux OVS_UNUSED)
456 if (!iface->netdev) {
457 int error = create_iface_netdev(iface);
459 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
469 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
470 void *aux OVS_UNUSED)
472 if (iface->dp_ifidx >= 0) {
473 VLOG_DBG("%s has interface %s on port %d",
475 iface->name, iface->dp_ifidx);
478 VLOG_ERR("%s interface not in %s, dropping",
479 iface->name, dpif_name(br->dpif));
485 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
486 void *aux OVS_UNUSED)
488 /* Set policing attributes. */
489 netdev_set_policing(iface->netdev,
490 iface->cfg->ingress_policing_rate,
491 iface->cfg->ingress_policing_burst);
493 /* Set MAC address of internal interfaces other than the local
495 if (iface->dp_ifidx != ODPP_LOCAL
496 && iface_is_internal(br, iface->name)) {
497 iface_set_mac(iface);
503 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
504 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
505 * deletes from 'br' any ports that no longer have any interfaces. */
507 iterate_and_prune_ifaces(struct bridge *br,
508 bool (*cb)(struct bridge *, struct iface *,
514 for (i = 0; i < br->n_ports; ) {
515 struct port *port = br->ports[i];
516 for (j = 0; j < port->n_ifaces; ) {
517 struct iface *iface = port->ifaces[j];
518 if (cb(br, iface, aux)) {
521 iface_destroy(iface);
525 if (port->n_ifaces) {
528 VLOG_ERR("%s port has no interfaces, dropping", port->name);
534 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
535 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
536 * responsible for freeing '*managersp' (with free()).
538 * You may be asking yourself "why does ovs-vswitchd care?", because
539 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
540 * should not be and in fact is not directly involved in that. But
541 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
542 * it has to tell in-band control where the managers are to enable that.
545 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
546 struct sockaddr_in **managersp, size_t *n_managersp)
548 struct sockaddr_in *managers = NULL;
549 size_t n_managers = 0;
551 if (ovs_cfg->n_managers > 0) {
554 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
555 for (i = 0; i < ovs_cfg->n_managers; i++) {
556 const char *name = ovs_cfg->managers[i];
557 struct sockaddr_in *sin = &managers[i];
559 if ((!strncmp(name, "tcp:", 4)
560 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
561 (!strncmp(name, "ssl:", 4)
562 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
568 *managersp = managers;
569 *n_managersp = n_managers;
573 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
575 struct shash old_br, new_br;
576 struct shash_node *node;
577 struct bridge *br, *next;
578 struct sockaddr_in *managers;
581 int sflow_bridge_number;
583 COVERAGE_INC(bridge_reconfigure);
585 collect_managers(ovs_cfg, &managers, &n_managers);
587 /* Collect old and new bridges. */
590 LIST_FOR_EACH (br, node, &all_bridges) {
591 shash_add(&old_br, br->name, br);
593 for (i = 0; i < ovs_cfg->n_bridges; i++) {
594 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
595 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
596 VLOG_WARN("more than one bridge named %s", br_cfg->name);
600 /* Get rid of deleted bridges and add new bridges. */
601 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
602 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
609 SHASH_FOR_EACH (node, &new_br) {
610 const char *br_name = node->name;
611 const struct ovsrec_bridge *br_cfg = node->data;
612 br = shash_find_data(&old_br, br_name);
614 /* If the bridge datapath type has changed, we need to tear it
615 * down and recreate. */
616 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
618 bridge_create(br_cfg);
621 bridge_create(br_cfg);
624 shash_destroy(&old_br);
625 shash_destroy(&new_br);
627 /* Reconfigure all bridges. */
628 LIST_FOR_EACH (br, node, &all_bridges) {
629 bridge_reconfigure_one(br);
632 /* Add and delete ports on all datapaths.
634 * The kernel will reject any attempt to add a given port to a datapath if
635 * that port already belongs to a different datapath, so we must do all
636 * port deletions before any port additions. */
637 LIST_FOR_EACH (br, node, &all_bridges) {
638 struct odp_port *dpif_ports;
640 struct shash want_ifaces;
642 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
643 bridge_get_all_ifaces(br, &want_ifaces);
644 for (i = 0; i < n_dpif_ports; i++) {
645 const struct odp_port *p = &dpif_ports[i];
646 if (!shash_find(&want_ifaces, p->devname)
647 && strcmp(p->devname, br->name)) {
648 int retval = dpif_port_del(br->dpif, p->port);
650 VLOG_ERR("failed to remove %s interface from %s: %s",
651 p->devname, dpif_name(br->dpif),
656 shash_destroy(&want_ifaces);
659 LIST_FOR_EACH (br, node, &all_bridges) {
660 struct odp_port *dpif_ports;
662 struct shash cur_ifaces, want_ifaces;
664 /* Get the set of interfaces currently in this datapath. */
665 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
666 shash_init(&cur_ifaces);
667 for (i = 0; i < n_dpif_ports; i++) {
668 const char *name = dpif_ports[i].devname;
669 shash_add_once(&cur_ifaces, name, NULL);
673 /* Get the set of interfaces we want on this datapath. */
674 bridge_get_all_ifaces(br, &want_ifaces);
676 SHASH_FOR_EACH (node, &want_ifaces) {
677 const char *if_name = node->name;
678 struct iface *iface = node->data;
680 if (shash_find(&cur_ifaces, if_name)) {
681 /* Already exists on the datapath. If we have it open,
682 * reconfigure it; otherwise we'll open it later. */
683 if (iface && iface->netdev) {
684 reconfigure_iface_netdev(iface);
690 /* Create interface if it doesn't already exist. */
691 internal = iface_is_internal(br, if_name);
693 error = create_iface_netdev(iface);
695 VLOG_WARN("could not create iface %s: %s", iface->name,
701 /* Add to datapath. */
702 error = dpif_port_add(br->dpif, if_name,
703 internal ? ODP_PORT_INTERNAL : 0, NULL);
704 if (error == EFBIG) {
705 VLOG_ERR("ran out of valid port numbers on %s",
706 dpif_name(br->dpif));
709 VLOG_ERR("failed to add %s interface to %s: %s",
710 if_name, dpif_name(br->dpif), strerror(error));
714 shash_destroy(&cur_ifaces);
715 shash_destroy(&want_ifaces);
717 sflow_bridge_number = 0;
718 LIST_FOR_EACH (br, node, &all_bridges) {
721 struct iface *local_iface;
722 struct iface *hw_addr_iface;
725 bridge_fetch_dp_ifaces(br);
727 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
728 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
730 /* Pick local port hardware address, datapath ID. */
731 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
732 local_iface = bridge_get_local_iface(br);
734 int error = netdev_set_etheraddr(local_iface->netdev, ea);
736 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
737 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
738 "Ethernet address: %s",
739 br->name, strerror(error));
743 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
744 ofproto_set_datapath_id(br->ofproto, dpid);
746 dpid_string = xasprintf("%016"PRIx64, dpid);
747 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
750 /* Set NetFlow configuration on this bridge. */
751 if (br->cfg->netflow) {
752 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
753 struct netflow_options opts;
755 memset(&opts, 0, sizeof opts);
757 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
758 if (nf_cfg->engine_type) {
759 opts.engine_type = *nf_cfg->engine_type;
761 if (nf_cfg->engine_id) {
762 opts.engine_id = *nf_cfg->engine_id;
765 opts.active_timeout = nf_cfg->active_timeout;
766 if (!opts.active_timeout) {
767 opts.active_timeout = -1;
768 } else if (opts.active_timeout < 0) {
769 VLOG_WARN("bridge %s: active timeout interval set to negative "
770 "value, using default instead (%d seconds)", br->name,
771 NF_ACTIVE_TIMEOUT_DEFAULT);
772 opts.active_timeout = -1;
775 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
776 if (opts.add_id_to_iface) {
777 if (opts.engine_id > 0x7f) {
778 VLOG_WARN("bridge %s: netflow port mangling may conflict "
779 "with another vswitch, choose an engine id less "
780 "than 128", br->name);
782 if (br->n_ports > 508) {
783 VLOG_WARN("bridge %s: netflow port mangling will conflict "
784 "with another port when more than 508 ports are "
789 opts.collectors.n = nf_cfg->n_targets;
790 opts.collectors.names = nf_cfg->targets;
791 if (ofproto_set_netflow(br->ofproto, &opts)) {
792 VLOG_ERR("bridge %s: problem setting netflow collectors",
796 ofproto_set_netflow(br->ofproto, NULL);
799 /* Set sFlow configuration on this bridge. */
800 if (br->cfg->sflow) {
801 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
802 struct ovsrec_controller **controllers;
803 struct ofproto_sflow_options oso;
804 size_t n_controllers;
806 memset(&oso, 0, sizeof oso);
808 oso.targets.n = sflow_cfg->n_targets;
809 oso.targets.names = sflow_cfg->targets;
811 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
812 if (sflow_cfg->sampling) {
813 oso.sampling_rate = *sflow_cfg->sampling;
816 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
817 if (sflow_cfg->polling) {
818 oso.polling_interval = *sflow_cfg->polling;
821 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
822 if (sflow_cfg->header) {
823 oso.header_len = *sflow_cfg->header;
826 oso.sub_id = sflow_bridge_number++;
827 oso.agent_device = sflow_cfg->agent;
829 oso.control_ip = NULL;
830 n_controllers = bridge_get_controllers(br, &controllers);
831 for (i = 0; i < n_controllers; i++) {
832 if (controllers[i]->local_ip) {
833 oso.control_ip = controllers[i]->local_ip;
837 ofproto_set_sflow(br->ofproto, &oso);
839 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
841 ofproto_set_sflow(br->ofproto, NULL);
844 /* Update the controller and related settings. It would be more
845 * straightforward to call this from bridge_reconfigure_one(), but we
846 * can't do it there for two reasons. First, and most importantly, at
847 * that point we don't know the dp_ifidx of any interfaces that have
848 * been added to the bridge (because we haven't actually added them to
849 * the datapath). Second, at that point we haven't set the datapath ID
850 * yet; when a controller is configured, resetting the datapath ID will
851 * immediately disconnect from the controller, so it's better to set
852 * the datapath ID before the controller. */
853 bridge_reconfigure_remotes(br, managers, n_managers);
855 LIST_FOR_EACH (br, node, &all_bridges) {
856 for (i = 0; i < br->n_ports; i++) {
857 struct port *port = br->ports[i];
860 port_update_vlan_compat(port);
861 port_update_bonding(port);
863 for (j = 0; j < port->n_ifaces; j++) {
864 iface_update_qos(port->ifaces[j], port->cfg->qos);
868 LIST_FOR_EACH (br, node, &all_bridges) {
869 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
876 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
877 const struct ovsdb_idl_column *column,
880 const struct ovsdb_datum *datum;
881 union ovsdb_atom atom;
884 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
885 atom.string = (char *) key;
886 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
887 return idx == UINT_MAX ? NULL : datum->values[idx].string;
891 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
893 return get_ovsrec_key_value(&br_cfg->header_,
894 &ovsrec_bridge_col_other_config, key);
898 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
899 struct iface **hw_addr_iface)
905 *hw_addr_iface = NULL;
907 /* Did the user request a particular MAC? */
908 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
909 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
910 if (eth_addr_is_multicast(ea)) {
911 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
912 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
913 } else if (eth_addr_is_zero(ea)) {
914 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
920 /* Otherwise choose the minimum non-local MAC address among all of the
922 memset(ea, 0xff, sizeof ea);
923 for (i = 0; i < br->n_ports; i++) {
924 struct port *port = br->ports[i];
925 uint8_t iface_ea[ETH_ADDR_LEN];
928 /* Mirror output ports don't participate. */
929 if (port->is_mirror_output_port) {
933 /* Choose the MAC address to represent the port. */
934 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
935 /* Find the interface with this Ethernet address (if any) so that
936 * we can provide the correct devname to the caller. */
938 for (j = 0; j < port->n_ifaces; j++) {
939 struct iface *candidate = port->ifaces[j];
940 uint8_t candidate_ea[ETH_ADDR_LEN];
941 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
942 && eth_addr_equals(iface_ea, candidate_ea)) {
947 /* Choose the interface whose MAC address will represent the port.
948 * The Linux kernel bonding code always chooses the MAC address of
949 * the first slave added to a bond, and the Fedora networking
950 * scripts always add slaves to a bond in alphabetical order, so
951 * for compatibility we choose the interface with the name that is
952 * first in alphabetical order. */
953 iface = port->ifaces[0];
954 for (j = 1; j < port->n_ifaces; j++) {
955 struct iface *candidate = port->ifaces[j];
956 if (strcmp(candidate->name, iface->name) < 0) {
961 /* The local port doesn't count (since we're trying to choose its
962 * MAC address anyway). */
963 if (iface->dp_ifidx == ODPP_LOCAL) {
968 error = netdev_get_etheraddr(iface->netdev, iface_ea);
970 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
971 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
972 iface->name, strerror(error));
977 /* Compare against our current choice. */
978 if (!eth_addr_is_multicast(iface_ea) &&
979 !eth_addr_is_local(iface_ea) &&
980 !eth_addr_is_reserved(iface_ea) &&
981 !eth_addr_is_zero(iface_ea) &&
982 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
984 memcpy(ea, iface_ea, ETH_ADDR_LEN);
985 *hw_addr_iface = iface;
988 if (eth_addr_is_multicast(ea)) {
989 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
990 *hw_addr_iface = NULL;
991 VLOG_WARN("bridge %s: using default bridge Ethernet "
992 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
994 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
995 br->name, ETH_ADDR_ARGS(ea));
999 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1000 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1001 * an interface on 'br', then that interface must be passed in as
1002 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1003 * 'hw_addr_iface' must be passed in as a null pointer. */
1005 bridge_pick_datapath_id(struct bridge *br,
1006 const uint8_t bridge_ea[ETH_ADDR_LEN],
1007 struct iface *hw_addr_iface)
1010 * The procedure for choosing a bridge MAC address will, in the most
1011 * ordinary case, also choose a unique MAC that we can use as a datapath
1012 * ID. In some special cases, though, multiple bridges will end up with
1013 * the same MAC address. This is OK for the bridges, but it will confuse
1014 * the OpenFlow controller, because each datapath needs a unique datapath
1017 * Datapath IDs must be unique. It is also very desirable that they be
1018 * stable from one run to the next, so that policy set on a datapath
1021 const char *datapath_id;
1024 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1025 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1029 if (hw_addr_iface) {
1031 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1033 * A bridge whose MAC address is taken from a VLAN network device
1034 * (that is, a network device created with vconfig(8) or similar
1035 * tool) will have the same MAC address as a bridge on the VLAN
1036 * device's physical network device.
1038 * Handle this case by hashing the physical network device MAC
1039 * along with the VLAN identifier.
1041 uint8_t buf[ETH_ADDR_LEN + 2];
1042 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1043 buf[ETH_ADDR_LEN] = vlan >> 8;
1044 buf[ETH_ADDR_LEN + 1] = vlan;
1045 return dpid_from_hash(buf, sizeof buf);
1048 * Assume that this bridge's MAC address is unique, since it
1049 * doesn't fit any of the cases we handle specially.
1054 * A purely internal bridge, that is, one that has no non-virtual
1055 * network devices on it at all, is more difficult because it has no
1056 * natural unique identifier at all.
1058 * When the host is a XenServer, we handle this case by hashing the
1059 * host's UUID with the name of the bridge. Names of bridges are
1060 * persistent across XenServer reboots, although they can be reused if
1061 * an internal network is destroyed and then a new one is later
1062 * created, so this is fairly effective.
1064 * When the host is not a XenServer, we punt by using a random MAC
1065 * address on each run.
1067 const char *host_uuid = xenserver_get_host_uuid();
1069 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1070 dpid = dpid_from_hash(combined, strlen(combined));
1076 return eth_addr_to_uint64(bridge_ea);
1080 dpid_from_hash(const void *data, size_t n)
1082 uint8_t hash[SHA1_DIGEST_SIZE];
1084 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1085 sha1_bytes(data, n, hash);
1086 eth_addr_mark_random(hash);
1087 return eth_addr_to_uint64(hash);
1091 iface_refresh_stats(struct iface *iface)
1097 static const struct iface_stat iface_stats[] = {
1098 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1099 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1100 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1101 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1102 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1103 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1104 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1105 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1106 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1107 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1108 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1109 { "collisions", offsetof(struct netdev_stats, collisions) },
1111 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1112 const struct iface_stat *s;
1114 char *keys[N_STATS];
1115 int64_t values[N_STATS];
1118 struct netdev_stats stats;
1120 /* Intentionally ignore return value, since errors will set 'stats' to
1121 * all-1s, and we will deal with that correctly below. */
1122 netdev_get_stats(iface->netdev, &stats);
1125 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1126 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1127 if (value != UINT64_MAX) {
1134 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1138 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1140 struct ovsdb_datum datum;
1144 get_system_stats(&stats);
1146 ovsdb_datum_from_shash(&datum, &stats);
1147 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1154 const struct ovsrec_open_vswitch *cfg;
1156 bool datapath_destroyed;
1157 bool database_changed;
1160 /* Let each bridge do the work that it needs to do. */
1161 datapath_destroyed = false;
1162 LIST_FOR_EACH (br, node, &all_bridges) {
1163 int error = bridge_run_one(br);
1165 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1166 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1167 "forcing reconfiguration", br->name);
1168 datapath_destroyed = true;
1172 /* (Re)configure if necessary. */
1173 database_changed = ovsdb_idl_run(idl);
1174 cfg = ovsrec_open_vswitch_first(idl);
1175 if (database_changed || datapath_destroyed) {
1177 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1179 bridge_configure_once(cfg);
1180 bridge_reconfigure(cfg);
1182 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1183 ovsdb_idl_txn_commit(txn);
1184 ovsdb_idl_txn_destroy(txn); /* XXX */
1186 /* We still need to reconfigure to avoid dangling pointers to
1187 * now-destroyed ovsrec structures inside bridge data. */
1188 static const struct ovsrec_open_vswitch null_cfg;
1190 bridge_reconfigure(&null_cfg);
1195 /* Re-configure SSL. We do this on every trip through the main loop,
1196 * instead of just when the database changes, because the contents of the
1197 * key and certificate files can change without the database changing. */
1198 if (cfg && cfg->ssl) {
1199 const struct ovsrec_ssl *ssl = cfg->ssl;
1201 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1202 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1206 /* Refresh system and interface stats if necessary. */
1207 if (time_msec() >= stats_timer) {
1209 struct ovsdb_idl_txn *txn;
1211 txn = ovsdb_idl_txn_create(idl);
1212 LIST_FOR_EACH (br, node, &all_bridges) {
1215 for (i = 0; i < br->n_ports; i++) {
1216 struct port *port = br->ports[i];
1219 for (j = 0; j < port->n_ifaces; j++) {
1220 struct iface *iface = port->ifaces[j];
1221 iface_refresh_stats(iface);
1225 refresh_system_stats(cfg);
1226 ovsdb_idl_txn_commit(txn);
1227 ovsdb_idl_txn_destroy(txn); /* XXX */
1230 stats_timer = time_msec() + STATS_INTERVAL;
1239 LIST_FOR_EACH (br, node, &all_bridges) {
1240 ofproto_wait(br->ofproto);
1241 if (ofproto_has_primary_controller(br->ofproto)) {
1245 mac_learning_wait(br->ml);
1248 ovsdb_idl_wait(idl);
1249 poll_timer_wait_until(stats_timer);
1252 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1253 * configuration changes. */
1255 bridge_flush(struct bridge *br)
1257 COVERAGE_INC(bridge_flush);
1259 mac_learning_flush(br->ml);
1262 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1263 * such interface. */
1264 static struct iface *
1265 bridge_get_local_iface(struct bridge *br)
1269 for (i = 0; i < br->n_ports; i++) {
1270 struct port *port = br->ports[i];
1271 for (j = 0; j < port->n_ifaces; j++) {
1272 struct iface *iface = port->ifaces[j];
1273 if (iface->dp_ifidx == ODPP_LOCAL) {
1282 /* Bridge unixctl user interface functions. */
1284 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1285 const char *args, void *aux OVS_UNUSED)
1287 struct ds ds = DS_EMPTY_INITIALIZER;
1288 const struct bridge *br;
1289 const struct mac_entry *e;
1291 br = bridge_lookup(args);
1293 unixctl_command_reply(conn, 501, "no such bridge");
1297 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1298 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1299 if (e->port < 0 || e->port >= br->n_ports) {
1302 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1303 br->ports[e->port]->ifaces[0]->dp_ifidx,
1304 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1306 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1310 /* Bridge reconfiguration functions. */
1311 static struct bridge *
1312 bridge_create(const struct ovsrec_bridge *br_cfg)
1317 assert(!bridge_lookup(br_cfg->name));
1318 br = xzalloc(sizeof *br);
1320 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1326 dpif_flow_flush(br->dpif);
1328 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1331 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1333 dpif_delete(br->dpif);
1334 dpif_close(br->dpif);
1339 br->name = xstrdup(br_cfg->name);
1341 br->ml = mac_learning_create();
1342 eth_addr_nicira_random(br->default_ea);
1344 hmap_init(&br->ifaces);
1346 shash_init(&br->port_by_name);
1347 shash_init(&br->iface_by_name);
1351 list_push_back(&all_bridges, &br->node);
1353 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1359 bridge_destroy(struct bridge *br)
1364 while (br->n_ports > 0) {
1365 port_destroy(br->ports[br->n_ports - 1]);
1367 list_remove(&br->node);
1368 error = dpif_delete(br->dpif);
1369 if (error && error != ENOENT) {
1370 VLOG_ERR("failed to delete %s: %s",
1371 dpif_name(br->dpif), strerror(error));
1373 dpif_close(br->dpif);
1374 ofproto_destroy(br->ofproto);
1375 mac_learning_destroy(br->ml);
1376 hmap_destroy(&br->ifaces);
1377 shash_destroy(&br->port_by_name);
1378 shash_destroy(&br->iface_by_name);
1385 static struct bridge *
1386 bridge_lookup(const char *name)
1390 LIST_FOR_EACH (br, node, &all_bridges) {
1391 if (!strcmp(br->name, name)) {
1398 /* Handle requests for a listing of all flows known by the OpenFlow
1399 * stack, including those normally hidden. */
1401 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1402 const char *args, void *aux OVS_UNUSED)
1407 br = bridge_lookup(args);
1409 unixctl_command_reply(conn, 501, "Unknown bridge");
1414 ofproto_get_all_flows(br->ofproto, &results);
1416 unixctl_command_reply(conn, 200, ds_cstr(&results));
1417 ds_destroy(&results);
1420 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1421 * connections and reconnect. If BRIDGE is not specified, then all bridges
1422 * drop their controller connections and reconnect. */
1424 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1425 const char *args, void *aux OVS_UNUSED)
1428 if (args[0] != '\0') {
1429 br = bridge_lookup(args);
1431 unixctl_command_reply(conn, 501, "Unknown bridge");
1434 ofproto_reconnect_controllers(br->ofproto);
1436 LIST_FOR_EACH (br, node, &all_bridges) {
1437 ofproto_reconnect_controllers(br->ofproto);
1440 unixctl_command_reply(conn, 200, NULL);
1444 bridge_run_one(struct bridge *br)
1448 error = ofproto_run1(br->ofproto);
1453 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1456 error = ofproto_run2(br->ofproto, br->flush);
1463 bridge_get_controllers(const struct bridge *br,
1464 struct ovsrec_controller ***controllersp)
1466 struct ovsrec_controller **controllers;
1467 size_t n_controllers;
1469 controllers = br->cfg->controller;
1470 n_controllers = br->cfg->n_controller;
1472 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1478 *controllersp = controllers;
1480 return n_controllers;
1484 bridge_reconfigure_one(struct bridge *br)
1486 struct shash old_ports, new_ports;
1487 struct svec snoops, old_snoops;
1488 struct shash_node *node;
1489 enum ofproto_fail_mode fail_mode;
1492 /* Collect old ports. */
1493 shash_init(&old_ports);
1494 for (i = 0; i < br->n_ports; i++) {
1495 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1498 /* Collect new ports. */
1499 shash_init(&new_ports);
1500 for (i = 0; i < br->cfg->n_ports; i++) {
1501 const char *name = br->cfg->ports[i]->name;
1502 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1503 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1508 /* If we have a controller, then we need a local port. Complain if the
1509 * user didn't specify one.
1511 * XXX perhaps we should synthesize a port ourselves in this case. */
1512 if (bridge_get_controllers(br, NULL)) {
1513 char local_name[IF_NAMESIZE];
1516 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1517 local_name, sizeof local_name);
1518 if (!error && !shash_find(&new_ports, local_name)) {
1519 VLOG_WARN("bridge %s: controller specified but no local port "
1520 "(port named %s) defined",
1521 br->name, local_name);
1525 /* Get rid of deleted ports.
1526 * Get rid of deleted interfaces on ports that still exist. */
1527 SHASH_FOR_EACH (node, &old_ports) {
1528 struct port *port = node->data;
1529 const struct ovsrec_port *port_cfg;
1531 port_cfg = shash_find_data(&new_ports, node->name);
1535 port_del_ifaces(port, port_cfg);
1539 /* Create new ports.
1540 * Add new interfaces to existing ports.
1541 * Reconfigure existing ports. */
1542 SHASH_FOR_EACH (node, &new_ports) {
1543 struct port *port = shash_find_data(&old_ports, node->name);
1545 port = port_create(br, node->name);
1548 port_reconfigure(port, node->data);
1549 if (!port->n_ifaces) {
1550 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1551 br->name, port->name);
1555 shash_destroy(&old_ports);
1556 shash_destroy(&new_ports);
1558 /* Set the fail-mode */
1559 fail_mode = !br->cfg->fail_mode
1560 || !strcmp(br->cfg->fail_mode, "standalone")
1561 ? OFPROTO_FAIL_STANDALONE
1562 : OFPROTO_FAIL_SECURE;
1563 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1564 && !ofproto_has_primary_controller(br->ofproto)) {
1565 ofproto_flush_flows(br->ofproto);
1567 ofproto_set_fail_mode(br->ofproto, fail_mode);
1569 /* Delete all flows if we're switching from connected to standalone or vice
1570 * versa. (XXX Should we delete all flows if we are switching from one
1571 * controller to another?) */
1573 /* Configure OpenFlow controller connection snooping. */
1575 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1576 ovs_rundir, br->name));
1577 svec_init(&old_snoops);
1578 ofproto_get_snoops(br->ofproto, &old_snoops);
1579 if (!svec_equal(&snoops, &old_snoops)) {
1580 ofproto_set_snoops(br->ofproto, &snoops);
1582 svec_destroy(&snoops);
1583 svec_destroy(&old_snoops);
1585 mirror_reconfigure(br);
1588 /* Initializes 'oc' appropriately as a management service controller for
1591 * The caller must free oc->target when it is no longer needed. */
1593 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1594 struct ofproto_controller *oc)
1596 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir, br->name);
1597 oc->max_backoff = 0;
1598 oc->probe_interval = 60;
1599 oc->band = OFPROTO_OUT_OF_BAND;
1600 oc->accept_re = NULL;
1601 oc->update_resolv_conf = false;
1603 oc->burst_limit = 0;
1606 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1608 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1609 struct ofproto_controller *oc)
1611 oc->target = c->target;
1612 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1613 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1614 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1615 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1616 oc->accept_re = c->discover_accept_regex;
1617 oc->update_resolv_conf = c->discover_update_resolv_conf;
1618 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1619 oc->burst_limit = (c->controller_burst_limit
1620 ? *c->controller_burst_limit : 0);
1623 /* Configures the IP stack for 'br''s local interface properly according to the
1624 * configuration in 'c'. */
1626 bridge_configure_local_iface_netdev(struct bridge *br,
1627 struct ovsrec_controller *c)
1629 struct netdev *netdev;
1630 struct in_addr mask, gateway;
1632 struct iface *local_iface;
1635 /* Controller discovery does its own TCP/IP configuration later. */
1636 if (strcmp(c->target, "discover")) {
1640 /* If there's no local interface or no IP address, give up. */
1641 local_iface = bridge_get_local_iface(br);
1642 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1646 /* Bring up the local interface. */
1647 netdev = local_iface->netdev;
1648 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1650 /* Configure the IP address and netmask. */
1651 if (!c->local_netmask
1652 || !inet_aton(c->local_netmask, &mask)
1654 mask.s_addr = guess_netmask(ip.s_addr);
1656 if (!netdev_set_in4(netdev, ip, mask)) {
1657 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1658 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1661 /* Configure the default gateway. */
1662 if (c->local_gateway
1663 && inet_aton(c->local_gateway, &gateway)
1664 && gateway.s_addr) {
1665 if (!netdev_add_router(netdev, gateway)) {
1666 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1667 br->name, IP_ARGS(&gateway.s_addr));
1673 bridge_reconfigure_remotes(struct bridge *br,
1674 const struct sockaddr_in *managers,
1677 struct ovsrec_controller **controllers;
1678 size_t n_controllers;
1681 struct ofproto_controller *ocs;
1685 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1686 had_primary = ofproto_has_primary_controller(br->ofproto);
1688 n_controllers = bridge_get_controllers(br, &controllers);
1690 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1693 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1694 for (i = 0; i < n_controllers; i++) {
1695 struct ovsrec_controller *c = controllers[i];
1697 if (!strncmp(c->target, "punix:", 6)
1698 || !strncmp(c->target, "unix:", 5)) {
1699 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1701 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1702 * domain sockets and overwriting arbitrary local files. */
1703 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1704 "\"%s\" due to possibility for remote exploit",
1705 dpif_name(br->dpif), c->target);
1709 bridge_configure_local_iface_netdev(br, c);
1710 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs++]);
1713 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1714 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1717 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1718 ofproto_flush_flows(br->ofproto);
1721 /* If there are no controllers and the bridge is in standalone
1722 * mode, set up a flow that matches every packet and directs
1723 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1724 * switch is in secure mode and we won't pass any traffic until
1725 * a controller has been defined and it tells us to do so. */
1727 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1728 union ofp_action action;
1731 memset(&action, 0, sizeof action);
1732 action.type = htons(OFPAT_OUTPUT);
1733 action.output.len = htons(sizeof action);
1734 action.output.port = htons(OFPP_NORMAL);
1735 memset(&flow, 0, sizeof flow);
1736 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1741 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1746 for (i = 0; i < br->n_ports; i++) {
1747 struct port *port = br->ports[i];
1748 for (j = 0; j < port->n_ifaces; j++) {
1749 struct iface *iface = port->ifaces[j];
1750 shash_add_once(ifaces, iface->name, iface);
1752 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1753 shash_add_once(ifaces, port->name, NULL);
1758 /* For robustness, in case the administrator moves around datapath ports behind
1759 * our back, we re-check all the datapath port numbers here.
1761 * This function will set the 'dp_ifidx' members of interfaces that have
1762 * disappeared to -1, so only call this function from a context where those
1763 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1764 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1765 * datapath, which doesn't support UINT16_MAX+1 ports. */
1767 bridge_fetch_dp_ifaces(struct bridge *br)
1769 struct odp_port *dpif_ports;
1770 size_t n_dpif_ports;
1773 /* Reset all interface numbers. */
1774 for (i = 0; i < br->n_ports; i++) {
1775 struct port *port = br->ports[i];
1776 for (j = 0; j < port->n_ifaces; j++) {
1777 struct iface *iface = port->ifaces[j];
1778 iface->dp_ifidx = -1;
1781 hmap_clear(&br->ifaces);
1783 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1784 for (i = 0; i < n_dpif_ports; i++) {
1785 struct odp_port *p = &dpif_ports[i];
1786 struct iface *iface = iface_lookup(br, p->devname);
1788 if (iface->dp_ifidx >= 0) {
1789 VLOG_WARN("%s reported interface %s twice",
1790 dpif_name(br->dpif), p->devname);
1791 } else if (iface_from_dp_ifidx(br, p->port)) {
1792 VLOG_WARN("%s reported interface %"PRIu16" twice",
1793 dpif_name(br->dpif), p->port);
1795 iface->dp_ifidx = p->port;
1796 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
1797 hash_int(iface->dp_ifidx, 0));
1801 int64_t ofport = (iface->dp_ifidx >= 0
1802 ? odp_port_to_ofp_port(iface->dp_ifidx)
1804 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1811 /* Bridge packet processing functions. */
1814 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1816 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1819 static struct bond_entry *
1820 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1822 return &port->bond_hash[bond_hash(mac)];
1826 bond_choose_iface(const struct port *port)
1828 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1829 size_t i, best_down_slave = -1;
1830 long long next_delay_expiration = LLONG_MAX;
1832 for (i = 0; i < port->n_ifaces; i++) {
1833 struct iface *iface = port->ifaces[i];
1835 if (iface->enabled) {
1837 } else if (iface->delay_expires < next_delay_expiration) {
1838 best_down_slave = i;
1839 next_delay_expiration = iface->delay_expires;
1843 if (best_down_slave != -1) {
1844 struct iface *iface = port->ifaces[best_down_slave];
1846 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1847 "since no other interface is up", iface->name,
1848 iface->delay_expires - time_msec());
1849 bond_enable_slave(iface, true);
1852 return best_down_slave;
1856 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1857 uint16_t *dp_ifidx, tag_type *tags)
1859 struct iface *iface;
1861 assert(port->n_ifaces);
1862 if (port->n_ifaces == 1) {
1863 iface = port->ifaces[0];
1865 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1866 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1867 || !port->ifaces[e->iface_idx]->enabled) {
1868 /* XXX select interface properly. The current interface selection
1869 * is only good for testing the rebalancing code. */
1870 e->iface_idx = bond_choose_iface(port);
1871 if (e->iface_idx < 0) {
1872 *tags |= port->no_ifaces_tag;
1875 e->iface_tag = tag_create_random();
1876 ((struct port *) port)->bond_compat_is_stale = true;
1878 *tags |= e->iface_tag;
1879 iface = port->ifaces[e->iface_idx];
1881 *dp_ifidx = iface->dp_ifidx;
1882 *tags |= iface->tag; /* Currently only used for bonding. */
1887 bond_link_status_update(struct iface *iface, bool carrier)
1889 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1890 struct port *port = iface->port;
1892 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1893 /* Nothing to do. */
1896 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1897 iface->name, carrier ? "detected" : "dropped");
1898 if (carrier == iface->enabled) {
1899 iface->delay_expires = LLONG_MAX;
1900 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1901 iface->name, carrier ? "disabled" : "enabled");
1902 } else if (carrier && port->active_iface < 0) {
1903 bond_enable_slave(iface, true);
1904 if (port->updelay) {
1905 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1906 "other interface is up", iface->name, port->updelay);
1909 int delay = carrier ? port->updelay : port->downdelay;
1910 iface->delay_expires = time_msec() + delay;
1913 "interface %s: will be %s if it stays %s for %d ms",
1915 carrier ? "enabled" : "disabled",
1916 carrier ? "up" : "down",
1923 bond_choose_active_iface(struct port *port)
1925 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1927 port->active_iface = bond_choose_iface(port);
1928 port->active_iface_tag = tag_create_random();
1929 if (port->active_iface >= 0) {
1930 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1931 port->name, port->ifaces[port->active_iface]->name);
1933 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1939 bond_enable_slave(struct iface *iface, bool enable)
1941 struct port *port = iface->port;
1942 struct bridge *br = port->bridge;
1944 /* This acts as a recursion check. If the act of disabling a slave
1945 * causes a different slave to be enabled, the flag will allow us to
1946 * skip redundant work when we reenter this function. It must be
1947 * cleared on exit to keep things safe with multiple bonds. */
1948 static bool moving_active_iface = false;
1950 iface->delay_expires = LLONG_MAX;
1951 if (enable == iface->enabled) {
1955 iface->enabled = enable;
1956 if (!iface->enabled) {
1957 VLOG_WARN("interface %s: disabled", iface->name);
1958 ofproto_revalidate(br->ofproto, iface->tag);
1959 if (iface->port_ifidx == port->active_iface) {
1960 ofproto_revalidate(br->ofproto,
1961 port->active_iface_tag);
1963 /* Disabling a slave can lead to another slave being immediately
1964 * enabled if there will be no active slaves but one is waiting
1965 * on an updelay. In this case we do not need to run most of the
1966 * code for the newly enabled slave since there was no period
1967 * without an active slave and it is redundant with the disabling
1969 moving_active_iface = true;
1970 bond_choose_active_iface(port);
1972 bond_send_learning_packets(port);
1974 VLOG_WARN("interface %s: enabled", iface->name);
1975 if (port->active_iface < 0 && !moving_active_iface) {
1976 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1977 bond_choose_active_iface(port);
1978 bond_send_learning_packets(port);
1980 iface->tag = tag_create_random();
1983 moving_active_iface = false;
1984 port->bond_compat_is_stale = true;
1987 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1988 * bond interface. */
1990 bond_update_fake_iface_stats(struct port *port)
1992 struct netdev_stats bond_stats;
1993 struct netdev *bond_dev;
1996 memset(&bond_stats, 0, sizeof bond_stats);
1998 for (i = 0; i < port->n_ifaces; i++) {
1999 struct netdev_stats slave_stats;
2001 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2002 /* XXX: We swap the stats here because they are swapped back when
2003 * reported by the internal device. The reason for this is
2004 * internal devices normally represent packets going into the system
2005 * but when used as fake bond device they represent packets leaving
2006 * the system. We really should do this in the internal device
2007 * itself because changing it here reverses the counts from the
2008 * perspective of the switch. However, the internal device doesn't
2009 * know what type of device it represents so we have to do it here
2011 bond_stats.tx_packets += slave_stats.rx_packets;
2012 bond_stats.tx_bytes += slave_stats.rx_bytes;
2013 bond_stats.rx_packets += slave_stats.tx_packets;
2014 bond_stats.rx_bytes += slave_stats.tx_bytes;
2018 if (!netdev_open_default(port->name, &bond_dev)) {
2019 netdev_set_stats(bond_dev, &bond_stats);
2020 netdev_close(bond_dev);
2025 bond_run(struct bridge *br)
2029 for (i = 0; i < br->n_ports; i++) {
2030 struct port *port = br->ports[i];
2032 if (port->n_ifaces >= 2) {
2033 for (j = 0; j < port->n_ifaces; j++) {
2034 struct iface *iface = port->ifaces[j];
2035 if (time_msec() >= iface->delay_expires) {
2036 bond_enable_slave(iface, !iface->enabled);
2040 if (port->bond_fake_iface
2041 && time_msec() >= port->bond_next_fake_iface_update) {
2042 bond_update_fake_iface_stats(port);
2043 port->bond_next_fake_iface_update = time_msec() + 1000;
2047 if (port->bond_compat_is_stale) {
2048 port->bond_compat_is_stale = false;
2049 port_update_bond_compat(port);
2055 bond_wait(struct bridge *br)
2059 for (i = 0; i < br->n_ports; i++) {
2060 struct port *port = br->ports[i];
2061 if (port->n_ifaces < 2) {
2064 for (j = 0; j < port->n_ifaces; j++) {
2065 struct iface *iface = port->ifaces[j];
2066 if (iface->delay_expires != LLONG_MAX) {
2067 poll_timer_wait_until(iface->delay_expires);
2070 if (port->bond_fake_iface) {
2071 poll_timer_wait_until(port->bond_next_fake_iface_update);
2077 set_dst(struct dst *p, const flow_t *flow,
2078 const struct port *in_port, const struct port *out_port,
2081 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2082 : in_port->vlan >= 0 ? in_port->vlan
2083 : ntohs(flow->dl_vlan));
2084 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
2088 swap_dst(struct dst *p, struct dst *q)
2090 struct dst tmp = *p;
2095 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2096 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2097 * that we push to the datapath. We could in fact fully sort the array by
2098 * vlan, but in most cases there are at most two different vlan tags so that's
2099 * possibly overkill.) */
2101 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2103 struct dst *first = dsts;
2104 struct dst *last = dsts + n_dsts;
2106 while (first != last) {
2108 * - All dsts < first have vlan == 'vlan'.
2109 * - All dsts >= last have vlan != 'vlan'.
2110 * - first < last. */
2111 while (first->vlan == vlan) {
2112 if (++first == last) {
2117 /* Same invariants, plus one additional:
2118 * - first->vlan != vlan.
2120 while (last[-1].vlan != vlan) {
2121 if (--last == first) {
2126 /* Same invariants, plus one additional:
2127 * - last[-1].vlan == vlan.*/
2128 swap_dst(first++, --last);
2133 mirror_mask_ffs(mirror_mask_t mask)
2135 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2140 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2141 const struct dst *test)
2144 for (i = 0; i < n_dsts; i++) {
2145 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2153 port_trunks_vlan(const struct port *port, uint16_t vlan)
2155 return (port->vlan < 0
2156 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2160 port_includes_vlan(const struct port *port, uint16_t vlan)
2162 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2166 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2167 const struct port *in_port, const struct port *out_port,
2168 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2170 mirror_mask_t mirrors = in_port->src_mirrors;
2171 struct dst *dst = dsts;
2174 if (out_port == FLOOD_PORT) {
2175 /* XXX use ODP_FLOOD if no vlans or bonding. */
2176 /* XXX even better, define each VLAN as a datapath port group */
2177 for (i = 0; i < br->n_ports; i++) {
2178 struct port *port = br->ports[i];
2179 if (port != in_port && port_includes_vlan(port, vlan)
2180 && !port->is_mirror_output_port
2181 && set_dst(dst, flow, in_port, port, tags)) {
2182 mirrors |= port->dst_mirrors;
2186 *nf_output_iface = NF_OUT_FLOOD;
2187 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2188 *nf_output_iface = dst->dp_ifidx;
2189 mirrors |= out_port->dst_mirrors;
2194 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2195 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2197 if (set_dst(dst, flow, in_port, m->out_port, tags)
2198 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2202 for (i = 0; i < br->n_ports; i++) {
2203 struct port *port = br->ports[i];
2204 if (port_includes_vlan(port, m->out_vlan)
2205 && set_dst(dst, flow, in_port, port, tags))
2209 if (port->vlan < 0) {
2210 dst->vlan = m->out_vlan;
2212 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2216 /* Use the vlan tag on the original flow instead of
2217 * the one passed in the vlan parameter. This ensures
2218 * that we compare the vlan from before any implicit
2219 * tagging tags place. This is necessary because
2220 * dst->vlan is the final vlan, after removing implicit
2222 flow_vlan = ntohs(flow->dl_vlan);
2223 if (flow_vlan == 0) {
2224 flow_vlan = OFP_VLAN_NONE;
2226 if (port == in_port && dst->vlan == flow_vlan) {
2227 /* Don't send out input port on same VLAN. */
2235 mirrors &= mirrors - 1;
2238 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2242 static void OVS_UNUSED
2243 print_dsts(const struct dst *dsts, size_t n)
2245 for (; n--; dsts++) {
2246 printf(">p%"PRIu16, dsts->dp_ifidx);
2247 if (dsts->vlan != OFP_VLAN_NONE) {
2248 printf("v%"PRIu16, dsts->vlan);
2254 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2255 const struct port *in_port, const struct port *out_port,
2256 tag_type *tags, struct odp_actions *actions,
2257 uint16_t *nf_output_iface)
2259 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2261 const struct dst *p;
2264 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2267 cur_vlan = ntohs(flow->dl_vlan);
2268 for (p = dsts; p < &dsts[n_dsts]; p++) {
2269 union odp_action *a;
2270 if (p->vlan != cur_vlan) {
2271 if (p->vlan == OFP_VLAN_NONE) {
2272 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2274 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2275 a->vlan_vid.vlan_vid = htons(p->vlan);
2279 a = odp_actions_add(actions, ODPAT_OUTPUT);
2280 a->output.port = p->dp_ifidx;
2284 /* Returns the effective vlan of a packet, taking into account both the
2285 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2286 * the packet is untagged and -1 indicates it has an invalid header and
2287 * should be dropped. */
2288 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2289 struct port *in_port, bool have_packet)
2291 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2292 * belongs to VLAN 0, so we should treat both cases identically. (In the
2293 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2294 * presumably to allow a priority to be specified. In the latter case, the
2295 * packet does not have any 802.1Q header.) */
2296 int vlan = ntohs(flow->dl_vlan);
2297 if (vlan == OFP_VLAN_NONE) {
2300 if (in_port->vlan >= 0) {
2302 /* XXX support double tagging? */
2304 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2305 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2306 "packet received on port %s configured with "
2307 "implicit VLAN %"PRIu16,
2308 br->name, ntohs(flow->dl_vlan),
2309 in_port->name, in_port->vlan);
2313 vlan = in_port->vlan;
2315 if (!port_includes_vlan(in_port, vlan)) {
2317 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2318 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2319 "packet received on port %s not configured for "
2321 br->name, vlan, in_port->name, vlan);
2330 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2331 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2332 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2334 is_gratuitous_arp(const flow_t *flow)
2336 return (flow->dl_type == htons(ETH_TYPE_ARP)
2337 && eth_addr_is_broadcast(flow->dl_dst)
2338 && (flow->nw_proto == ARP_OP_REPLY
2339 || (flow->nw_proto == ARP_OP_REQUEST
2340 && flow->nw_src == flow->nw_dst)));
2344 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2345 struct port *in_port)
2347 enum grat_arp_lock_type lock_type;
2350 /* We don't want to learn from gratuitous ARP packets that are reflected
2351 * back over bond slaves so we lock the learning table. */
2352 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2353 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2354 GRAT_ARP_LOCK_CHECK;
2356 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2359 /* The log messages here could actually be useful in debugging,
2360 * so keep the rate limit relatively high. */
2361 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2363 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2364 "on port %s in VLAN %d",
2365 br->name, ETH_ADDR_ARGS(flow->dl_src),
2366 in_port->name, vlan);
2367 ofproto_revalidate(br->ofproto, rev_tag);
2371 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2372 * dropped. Returns true if they may be forwarded, false if they should be
2375 * If 'have_packet' is true, it indicates that the caller is processing a
2376 * received packet. If 'have_packet' is false, then the caller is just
2377 * revalidating an existing flow because configuration has changed. Either
2378 * way, 'have_packet' only affects logging (there is no point in logging errors
2379 * during revalidation).
2381 * Sets '*in_portp' to the input port. This will be a null pointer if
2382 * flow->in_port does not designate a known input port (in which case
2383 * is_admissible() returns false).
2385 * When returning true, sets '*vlanp' to the effective VLAN of the input
2386 * packet, as returned by flow_get_vlan().
2388 * May also add tags to '*tags', although the current implementation only does
2389 * so in one special case.
2392 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2393 tag_type *tags, int *vlanp, struct port **in_portp)
2395 struct iface *in_iface;
2396 struct port *in_port;
2399 /* Find the interface and port structure for the received packet. */
2400 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2402 /* No interface? Something fishy... */
2404 /* Odd. A few possible reasons here:
2406 * - We deleted an interface but there are still a few packets
2407 * queued up from it.
2409 * - Someone externally added an interface (e.g. with "ovs-dpctl
2410 * add-if") that we don't know about.
2412 * - Packet arrived on the local port but the local port is not
2413 * one of our bridge ports.
2415 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2417 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2418 "interface %"PRIu16, br->name, flow->in_port);
2424 *in_portp = in_port = in_iface->port;
2425 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2430 /* Drop frames for reserved multicast addresses. */
2431 if (eth_addr_is_reserved(flow->dl_dst)) {
2435 /* Drop frames on ports reserved for mirroring. */
2436 if (in_port->is_mirror_output_port) {
2438 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2439 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2440 "%s, which is reserved exclusively for mirroring",
2441 br->name, in_port->name);
2446 /* Packets received on bonds need special attention to avoid duplicates. */
2447 if (in_port->n_ifaces > 1) {
2449 bool is_grat_arp_locked;
2451 if (eth_addr_is_multicast(flow->dl_dst)) {
2452 *tags |= in_port->active_iface_tag;
2453 if (in_port->active_iface != in_iface->port_ifidx) {
2454 /* Drop all multicast packets on inactive slaves. */
2459 /* Drop all packets for which we have learned a different input
2460 * port, because we probably sent the packet on one slave and got
2461 * it back on the other. Gratuitous ARP packets are an exception
2462 * to this rule: the host has moved to another switch. The exception
2463 * to the exception is if we locked the learning table to avoid
2464 * reflections on bond slaves. If this is the case, just drop the
2466 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2467 &is_grat_arp_locked);
2468 if (src_idx != -1 && src_idx != in_port->port_idx &&
2469 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2477 /* If the composed actions may be applied to any packet in the given 'flow',
2478 * returns true. Otherwise, the actions should only be applied to 'packet', or
2479 * not at all, if 'packet' was NULL. */
2481 process_flow(struct bridge *br, const flow_t *flow,
2482 const struct ofpbuf *packet, struct odp_actions *actions,
2483 tag_type *tags, uint16_t *nf_output_iface)
2485 struct port *in_port;
2486 struct port *out_port;
2490 /* Check whether we should drop packets in this flow. */
2491 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2496 /* Learn source MAC (but don't try to learn from revalidation). */
2498 update_learning_table(br, flow, vlan, in_port);
2501 /* Determine output port. */
2502 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2504 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2505 out_port = br->ports[out_port_idx];
2506 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2507 /* If we are revalidating but don't have a learning entry then
2508 * eject the flow. Installing a flow that floods packets opens
2509 * up a window of time where we could learn from a packet reflected
2510 * on a bond and blackhole packets before the learning table is
2511 * updated to reflect the correct port. */
2514 out_port = FLOOD_PORT;
2517 /* Don't send packets out their input ports. */
2518 if (in_port == out_port) {
2524 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2531 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2534 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2535 const struct ofp_phy_port *opp,
2538 struct bridge *br = br_;
2539 struct iface *iface;
2542 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2548 if (reason == OFPPR_DELETE) {
2549 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2550 br->name, iface->name);
2551 iface_destroy(iface);
2552 if (!port->n_ifaces) {
2553 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2554 br->name, port->name);
2560 if (port->n_ifaces > 1) {
2561 bool up = !(opp->state & OFPPS_LINK_DOWN);
2562 bond_link_status_update(iface, up);
2563 port_update_bond_compat(port);
2569 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2570 struct odp_actions *actions, tag_type *tags,
2571 uint16_t *nf_output_iface, void *br_)
2573 struct bridge *br = br_;
2575 COVERAGE_INC(bridge_process_flow);
2577 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2581 bridge_account_flow_ofhook_cb(const flow_t *flow, tag_type tags,
2582 const union odp_action *actions,
2583 size_t n_actions, unsigned long long int n_bytes,
2586 struct bridge *br = br_;
2587 const union odp_action *a;
2588 struct port *in_port;
2592 /* Feed information from the active flows back into the learning table to
2593 * ensure that table is always in sync with what is actually flowing
2594 * through the datapath.
2596 * We test that 'tags' is nonzero to ensure that only flows that include an
2597 * OFPP_NORMAL action are used for learning. This works because
2598 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2599 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2600 update_learning_table(br, flow, vlan, in_port);
2603 /* Account for bond slave utilization. */
2604 if (!br->has_bonded_ports) {
2607 for (a = actions; a < &actions[n_actions]; a++) {
2608 if (a->type == ODPAT_OUTPUT) {
2609 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2610 if (out_port && out_port->n_ifaces >= 2) {
2611 struct bond_entry *e = lookup_bond_entry(out_port,
2613 e->tx_bytes += n_bytes;
2620 bridge_account_checkpoint_ofhook_cb(void *br_)
2622 struct bridge *br = br_;
2626 if (!br->has_bonded_ports) {
2631 for (i = 0; i < br->n_ports; i++) {
2632 struct port *port = br->ports[i];
2633 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2634 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2635 bond_rebalance_port(port);
2640 static struct ofhooks bridge_ofhooks = {
2641 bridge_port_changed_ofhook_cb,
2642 bridge_normal_ofhook_cb,
2643 bridge_account_flow_ofhook_cb,
2644 bridge_account_checkpoint_ofhook_cb,
2647 /* Bonding functions. */
2649 /* Statistics for a single interface on a bonded port, used for load-based
2650 * bond rebalancing. */
2651 struct slave_balance {
2652 struct iface *iface; /* The interface. */
2653 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2655 /* All the "bond_entry"s that are assigned to this interface, in order of
2656 * increasing tx_bytes. */
2657 struct bond_entry **hashes;
2661 /* Sorts pointers to pointers to bond_entries in ascending order by the
2662 * interface to which they are assigned, and within a single interface in
2663 * ascending order of bytes transmitted. */
2665 compare_bond_entries(const void *a_, const void *b_)
2667 const struct bond_entry *const *ap = a_;
2668 const struct bond_entry *const *bp = b_;
2669 const struct bond_entry *a = *ap;
2670 const struct bond_entry *b = *bp;
2671 if (a->iface_idx != b->iface_idx) {
2672 return a->iface_idx > b->iface_idx ? 1 : -1;
2673 } else if (a->tx_bytes != b->tx_bytes) {
2674 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2680 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2681 * *descending* order by number of bytes transmitted. */
2683 compare_slave_balance(const void *a_, const void *b_)
2685 const struct slave_balance *a = a_;
2686 const struct slave_balance *b = b_;
2687 if (a->iface->enabled != b->iface->enabled) {
2688 return a->iface->enabled ? -1 : 1;
2689 } else if (a->tx_bytes != b->tx_bytes) {
2690 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2697 swap_bals(struct slave_balance *a, struct slave_balance *b)
2699 struct slave_balance tmp = *a;
2704 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2705 * given that 'p' (and only 'p') might be in the wrong location.
2707 * This function invalidates 'p', since it might now be in a different memory
2710 resort_bals(struct slave_balance *p,
2711 struct slave_balance bals[], size_t n_bals)
2714 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2715 swap_bals(p, p - 1);
2717 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2718 swap_bals(p, p + 1);
2724 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2726 if (VLOG_IS_DBG_ENABLED()) {
2727 struct ds ds = DS_EMPTY_INITIALIZER;
2728 const struct slave_balance *b;
2730 for (b = bals; b < bals + n_bals; b++) {
2734 ds_put_char(&ds, ',');
2736 ds_put_format(&ds, " %s %"PRIu64"kB",
2737 b->iface->name, b->tx_bytes / 1024);
2739 if (!b->iface->enabled) {
2740 ds_put_cstr(&ds, " (disabled)");
2742 if (b->n_hashes > 0) {
2743 ds_put_cstr(&ds, " (");
2744 for (i = 0; i < b->n_hashes; i++) {
2745 const struct bond_entry *e = b->hashes[i];
2747 ds_put_cstr(&ds, " + ");
2749 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2750 e - port->bond_hash, e->tx_bytes / 1024);
2752 ds_put_cstr(&ds, ")");
2755 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2760 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2762 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2765 struct bond_entry *hash = from->hashes[hash_idx];
2766 struct port *port = from->iface->port;
2767 uint64_t delta = hash->tx_bytes;
2769 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2770 "from %s to %s (now carrying %"PRIu64"kB and "
2771 "%"PRIu64"kB load, respectively)",
2772 port->name, delta / 1024, hash - port->bond_hash,
2773 from->iface->name, to->iface->name,
2774 (from->tx_bytes - delta) / 1024,
2775 (to->tx_bytes + delta) / 1024);
2777 /* Delete element from from->hashes.
2779 * We don't bother to add the element to to->hashes because not only would
2780 * it require more work, the only purpose it would be to allow that hash to
2781 * be migrated to another slave in this rebalancing run, and there is no
2782 * point in doing that. */
2783 if (hash_idx == 0) {
2786 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2787 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2791 /* Shift load away from 'from' to 'to'. */
2792 from->tx_bytes -= delta;
2793 to->tx_bytes += delta;
2795 /* Arrange for flows to be revalidated. */
2796 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2797 hash->iface_idx = to->iface->port_ifidx;
2798 hash->iface_tag = tag_create_random();
2802 bond_rebalance_port(struct port *port)
2804 struct slave_balance bals[DP_MAX_PORTS];
2806 struct bond_entry *hashes[BOND_MASK + 1];
2807 struct slave_balance *b, *from, *to;
2808 struct bond_entry *e;
2811 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2812 * descending order of tx_bytes, so that bals[0] represents the most
2813 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2816 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2817 * array for each slave_balance structure, we sort our local array of
2818 * hashes in order by slave, so that all of the hashes for a given slave
2819 * become contiguous in memory, and then we point each 'hashes' members of
2820 * a slave_balance structure to the start of a contiguous group. */
2821 n_bals = port->n_ifaces;
2822 for (b = bals; b < &bals[n_bals]; b++) {
2823 b->iface = port->ifaces[b - bals];
2828 for (i = 0; i <= BOND_MASK; i++) {
2829 hashes[i] = &port->bond_hash[i];
2831 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2832 for (i = 0; i <= BOND_MASK; i++) {
2834 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2835 b = &bals[e->iface_idx];
2836 b->tx_bytes += e->tx_bytes;
2838 b->hashes = &hashes[i];
2843 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2844 log_bals(bals, n_bals, port);
2846 /* Discard slaves that aren't enabled (which were sorted to the back of the
2847 * array earlier). */
2848 while (!bals[n_bals - 1].iface->enabled) {
2855 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2856 to = &bals[n_bals - 1];
2857 for (from = bals; from < to; ) {
2858 uint64_t overload = from->tx_bytes - to->tx_bytes;
2859 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2860 /* The extra load on 'from' (and all less-loaded slaves), compared
2861 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2862 * it is less than ~1Mbps. No point in rebalancing. */
2864 } else if (from->n_hashes == 1) {
2865 /* 'from' only carries a single MAC hash, so we can't shift any
2866 * load away from it, even though we want to. */
2869 /* 'from' is carrying significantly more load than 'to', and that
2870 * load is split across at least two different hashes. Pick a hash
2871 * to migrate to 'to' (the least-loaded slave), given that doing so
2872 * must decrease the ratio of the load on the two slaves by at
2875 * The sort order we use means that we prefer to shift away the
2876 * smallest hashes instead of the biggest ones. There is little
2877 * reason behind this decision; we could use the opposite sort
2878 * order to shift away big hashes ahead of small ones. */
2881 for (i = 0; i < from->n_hashes; i++) {
2882 double old_ratio, new_ratio;
2883 uint64_t delta = from->hashes[i]->tx_bytes;
2885 if (delta == 0 || from->tx_bytes - delta == 0) {
2886 /* Pointless move. */
2890 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2892 if (to->tx_bytes == 0) {
2893 /* Nothing on the new slave, move it. */
2897 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2898 new_ratio = (double)(from->tx_bytes - delta) /
2899 (to->tx_bytes + delta);
2901 if (new_ratio == 0) {
2902 /* Should already be covered but check to prevent division
2907 if (new_ratio < 1) {
2908 new_ratio = 1 / new_ratio;
2911 if (old_ratio - new_ratio > 0.1) {
2912 /* Would decrease the ratio, move it. */
2916 if (i < from->n_hashes) {
2917 bond_shift_load(from, to, i);
2918 port->bond_compat_is_stale = true;
2920 /* If the result of the migration changed the relative order of
2921 * 'from' and 'to' swap them back to maintain invariants. */
2922 if (order_swapped) {
2923 swap_bals(from, to);
2926 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2927 * point to different slave_balance structures. It is only
2928 * valid to do these two operations in a row at all because we
2929 * know that 'from' will not move past 'to' and vice versa. */
2930 resort_bals(from, bals, n_bals);
2931 resort_bals(to, bals, n_bals);
2938 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2939 * historical data to decay to <1% in 7 rebalancing runs. */
2940 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2946 bond_send_learning_packets(struct port *port)
2948 struct bridge *br = port->bridge;
2949 struct mac_entry *e;
2950 struct ofpbuf packet;
2951 int error, n_packets, n_errors;
2953 if (!port->n_ifaces || port->active_iface < 0) {
2957 ofpbuf_init(&packet, 128);
2958 error = n_packets = n_errors = 0;
2959 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
2960 union ofp_action actions[2], *a;
2966 if (e->port == port->port_idx
2967 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2971 /* Compose actions. */
2972 memset(actions, 0, sizeof actions);
2975 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2976 a->vlan_vid.len = htons(sizeof *a);
2977 a->vlan_vid.vlan_vid = htons(e->vlan);
2980 a->output.type = htons(OFPAT_OUTPUT);
2981 a->output.len = htons(sizeof *a);
2982 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2987 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2989 flow_extract(&packet, 0, ODPP_NONE, &flow);
2990 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2997 ofpbuf_uninit(&packet);
3000 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3001 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3002 "packets, last error was: %s",
3003 port->name, n_errors, n_packets, strerror(error));
3005 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3006 port->name, n_packets);
3010 /* Bonding unixctl user interface functions. */
3013 bond_unixctl_list(struct unixctl_conn *conn,
3014 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3016 struct ds ds = DS_EMPTY_INITIALIZER;
3017 const struct bridge *br;
3019 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
3021 LIST_FOR_EACH (br, node, &all_bridges) {
3024 for (i = 0; i < br->n_ports; i++) {
3025 const struct port *port = br->ports[i];
3026 if (port->n_ifaces > 1) {
3029 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
3030 for (j = 0; j < port->n_ifaces; j++) {
3031 const struct iface *iface = port->ifaces[j];
3033 ds_put_cstr(&ds, ", ");
3035 ds_put_cstr(&ds, iface->name);
3037 ds_put_char(&ds, '\n');
3041 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3045 static struct port *
3046 bond_find(const char *name)
3048 const struct bridge *br;
3050 LIST_FOR_EACH (br, node, &all_bridges) {
3053 for (i = 0; i < br->n_ports; i++) {
3054 struct port *port = br->ports[i];
3055 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3064 bond_unixctl_show(struct unixctl_conn *conn,
3065 const char *args, void *aux OVS_UNUSED)
3067 struct ds ds = DS_EMPTY_INITIALIZER;
3068 const struct port *port;
3071 port = bond_find(args);
3073 unixctl_command_reply(conn, 501, "no such bond");
3077 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3078 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3079 ds_put_format(&ds, "next rebalance: %lld ms\n",
3080 port->bond_next_rebalance - time_msec());
3081 for (j = 0; j < port->n_ifaces; j++) {
3082 const struct iface *iface = port->ifaces[j];
3083 struct bond_entry *be;
3086 ds_put_format(&ds, "slave %s: %s\n",
3087 iface->name, iface->enabled ? "enabled" : "disabled");
3088 if (j == port->active_iface) {
3089 ds_put_cstr(&ds, "\tactive slave\n");
3091 if (iface->delay_expires != LLONG_MAX) {
3092 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3093 iface->enabled ? "downdelay" : "updelay",
3094 iface->delay_expires - time_msec());
3098 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3099 int hash = be - port->bond_hash;
3100 struct mac_entry *me;
3102 if (be->iface_idx != j) {
3106 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3107 hash, be->tx_bytes / 1024);
3110 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3113 if (bond_hash(me->mac) == hash
3114 && me->port != port->port_idx
3115 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3116 && dp_ifidx == iface->dp_ifidx)
3118 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3119 ETH_ADDR_ARGS(me->mac));
3124 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3129 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3130 void *aux OVS_UNUSED)
3132 char *args = (char *) args_;
3133 char *save_ptr = NULL;
3134 char *bond_s, *hash_s, *slave_s;
3135 uint8_t mac[ETH_ADDR_LEN];
3137 struct iface *iface;
3138 struct bond_entry *entry;
3141 bond_s = strtok_r(args, " ", &save_ptr);
3142 hash_s = strtok_r(NULL, " ", &save_ptr);
3143 slave_s = strtok_r(NULL, " ", &save_ptr);
3145 unixctl_command_reply(conn, 501,
3146 "usage: bond/migrate BOND HASH SLAVE");
3150 port = bond_find(bond_s);
3152 unixctl_command_reply(conn, 501, "no such bond");
3156 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3157 == ETH_ADDR_SCAN_COUNT) {
3158 hash = bond_hash(mac);
3159 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3160 hash = atoi(hash_s) & BOND_MASK;
3162 unixctl_command_reply(conn, 501, "bad hash");
3166 iface = port_lookup_iface(port, slave_s);
3168 unixctl_command_reply(conn, 501, "no such slave");
3172 if (!iface->enabled) {
3173 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3177 entry = &port->bond_hash[hash];
3178 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3179 entry->iface_idx = iface->port_ifidx;
3180 entry->iface_tag = tag_create_random();
3181 port->bond_compat_is_stale = true;
3182 unixctl_command_reply(conn, 200, "migrated");
3186 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3187 void *aux OVS_UNUSED)
3189 char *args = (char *) args_;
3190 char *save_ptr = NULL;
3191 char *bond_s, *slave_s;
3193 struct iface *iface;
3195 bond_s = strtok_r(args, " ", &save_ptr);
3196 slave_s = strtok_r(NULL, " ", &save_ptr);
3198 unixctl_command_reply(conn, 501,
3199 "usage: bond/set-active-slave BOND SLAVE");
3203 port = bond_find(bond_s);
3205 unixctl_command_reply(conn, 501, "no such bond");
3209 iface = port_lookup_iface(port, slave_s);
3211 unixctl_command_reply(conn, 501, "no such slave");
3215 if (!iface->enabled) {
3216 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3220 if (port->active_iface != iface->port_ifidx) {
3221 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3222 port->active_iface = iface->port_ifidx;
3223 port->active_iface_tag = tag_create_random();
3224 VLOG_INFO("port %s: active interface is now %s",
3225 port->name, iface->name);
3226 bond_send_learning_packets(port);
3227 unixctl_command_reply(conn, 200, "done");
3229 unixctl_command_reply(conn, 200, "no change");
3234 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3236 char *args = (char *) args_;
3237 char *save_ptr = NULL;
3238 char *bond_s, *slave_s;
3240 struct iface *iface;
3242 bond_s = strtok_r(args, " ", &save_ptr);
3243 slave_s = strtok_r(NULL, " ", &save_ptr);
3245 unixctl_command_reply(conn, 501,
3246 "usage: bond/enable/disable-slave BOND SLAVE");
3250 port = bond_find(bond_s);
3252 unixctl_command_reply(conn, 501, "no such bond");
3256 iface = port_lookup_iface(port, slave_s);
3258 unixctl_command_reply(conn, 501, "no such slave");
3262 bond_enable_slave(iface, enable);
3263 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3267 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3268 void *aux OVS_UNUSED)
3270 enable_slave(conn, args, true);
3274 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3275 void *aux OVS_UNUSED)
3277 enable_slave(conn, args, false);
3281 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3282 void *aux OVS_UNUSED)
3284 uint8_t mac[ETH_ADDR_LEN];
3288 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3289 == ETH_ADDR_SCAN_COUNT) {
3290 hash = bond_hash(mac);
3292 hash_cstr = xasprintf("%u", hash);
3293 unixctl_command_reply(conn, 200, hash_cstr);
3296 unixctl_command_reply(conn, 501, "invalid mac");
3303 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3304 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3305 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3306 unixctl_command_register("bond/set-active-slave",
3307 bond_unixctl_set_active_slave, NULL);
3308 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3310 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3312 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3315 /* Port functions. */
3317 static struct port *
3318 port_create(struct bridge *br, const char *name)
3322 port = xzalloc(sizeof *port);
3324 port->port_idx = br->n_ports;
3326 port->trunks = NULL;
3327 port->name = xstrdup(name);
3328 port->active_iface = -1;
3330 if (br->n_ports >= br->allocated_ports) {
3331 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3334 br->ports[br->n_ports++] = port;
3335 shash_add_assert(&br->port_by_name, port->name, port);
3337 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3344 get_port_other_config(const struct ovsrec_port *port, const char *key,
3345 const char *default_value)
3349 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3351 return value ? value : default_value;
3355 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3357 struct shash new_ifaces;
3360 /* Collect list of new interfaces. */
3361 shash_init(&new_ifaces);
3362 for (i = 0; i < cfg->n_interfaces; i++) {
3363 const char *name = cfg->interfaces[i]->name;
3364 shash_add_once(&new_ifaces, name, NULL);
3367 /* Get rid of deleted interfaces. */
3368 for (i = 0; i < port->n_ifaces; ) {
3369 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3370 iface_destroy(port->ifaces[i]);
3376 shash_destroy(&new_ifaces);
3380 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3382 struct shash new_ifaces;
3383 long long int next_rebalance;
3384 unsigned long *trunks;
3390 /* Update settings. */
3391 port->updelay = cfg->bond_updelay;
3392 if (port->updelay < 0) {
3395 port->downdelay = cfg->bond_downdelay;
3396 if (port->downdelay < 0) {
3397 port->downdelay = 0;
3399 port->bond_rebalance_interval = atoi(
3400 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3401 if (port->bond_rebalance_interval < 1000) {
3402 port->bond_rebalance_interval = 1000;
3404 next_rebalance = time_msec() + port->bond_rebalance_interval;
3405 if (port->bond_next_rebalance > next_rebalance) {
3406 port->bond_next_rebalance = next_rebalance;
3409 /* Add new interfaces and update 'cfg' member of existing ones. */
3410 shash_init(&new_ifaces);
3411 for (i = 0; i < cfg->n_interfaces; i++) {
3412 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3413 struct iface *iface;
3415 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3416 VLOG_WARN("port %s: %s specified twice as port interface",
3417 port->name, if_cfg->name);
3421 iface = iface_lookup(port->bridge, if_cfg->name);
3423 if (iface->port != port) {
3424 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3426 port->bridge->name, if_cfg->name, iface->port->name);
3429 iface->cfg = if_cfg;
3431 iface_create(port, if_cfg);
3434 shash_destroy(&new_ifaces);
3439 if (port->n_ifaces < 2) {
3441 if (vlan >= 0 && vlan <= 4095) {
3442 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3447 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3448 * they even work as-is. But they have not been tested. */
3449 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3453 if (port->vlan != vlan) {
3455 bridge_flush(port->bridge);
3458 /* Get trunked VLANs. */
3460 if (vlan < 0 && cfg->n_trunks) {
3463 trunks = bitmap_allocate(4096);
3465 for (i = 0; i < cfg->n_trunks; i++) {
3466 int trunk = cfg->trunks[i];
3468 bitmap_set1(trunks, trunk);
3474 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3475 port->name, cfg->n_trunks);
3477 if (n_errors == cfg->n_trunks) {
3478 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3480 bitmap_free(trunks);
3483 } else if (vlan >= 0 && cfg->n_trunks) {
3484 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3488 ? port->trunks != NULL
3489 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3490 bridge_flush(port->bridge);
3492 bitmap_free(port->trunks);
3493 port->trunks = trunks;
3497 port_destroy(struct port *port)
3500 struct bridge *br = port->bridge;
3504 proc_net_compat_update_vlan(port->name, NULL, 0);
3505 proc_net_compat_update_bond(port->name, NULL);
3507 for (i = 0; i < MAX_MIRRORS; i++) {
3508 struct mirror *m = br->mirrors[i];
3509 if (m && m->out_port == port) {
3514 while (port->n_ifaces > 0) {
3515 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3518 shash_find_and_delete_assert(&br->port_by_name, port->name);
3520 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3521 del->port_idx = port->port_idx;
3524 bitmap_free(port->trunks);
3531 static struct port *
3532 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3534 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3535 return iface ? iface->port : NULL;
3538 static struct port *
3539 port_lookup(const struct bridge *br, const char *name)
3541 return shash_find_data(&br->port_by_name, name);
3544 static struct iface *
3545 port_lookup_iface(const struct port *port, const char *name)
3547 struct iface *iface = iface_lookup(port->bridge, name);
3548 return iface && iface->port == port ? iface : NULL;
3552 port_update_bonding(struct port *port)
3554 if (port->n_ifaces < 2) {
3555 /* Not a bonded port. */
3556 if (port->bond_hash) {
3557 free(port->bond_hash);
3558 port->bond_hash = NULL;
3559 port->bond_compat_is_stale = true;
3560 port->bond_fake_iface = false;
3563 if (!port->bond_hash) {
3566 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3567 for (i = 0; i <= BOND_MASK; i++) {
3568 struct bond_entry *e = &port->bond_hash[i];
3572 port->no_ifaces_tag = tag_create_random();
3573 bond_choose_active_iface(port);
3574 port->bond_next_rebalance
3575 = time_msec() + port->bond_rebalance_interval;
3577 if (port->cfg->bond_fake_iface) {
3578 port->bond_next_fake_iface_update = time_msec();
3581 port->bond_compat_is_stale = true;
3582 port->bond_fake_iface = port->cfg->bond_fake_iface;
3587 port_update_bond_compat(struct port *port)
3589 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3590 struct compat_bond bond;
3593 if (port->n_ifaces < 2) {
3594 proc_net_compat_update_bond(port->name, NULL);
3599 bond.updelay = port->updelay;
3600 bond.downdelay = port->downdelay;
3603 bond.hashes = compat_hashes;
3604 if (port->bond_hash) {
3605 const struct bond_entry *e;
3606 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3607 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3608 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3609 cbh->hash = e - port->bond_hash;
3610 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3615 bond.n_slaves = port->n_ifaces;
3616 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3617 for (i = 0; i < port->n_ifaces; i++) {
3618 struct iface *iface = port->ifaces[i];
3619 struct compat_bond_slave *slave = &bond.slaves[i];
3620 slave->name = iface->name;
3622 /* We need to make the same determination as the Linux bonding
3623 * code to determine whether a slave should be consider "up".
3624 * The Linux function bond_miimon_inspect() supports four
3625 * BOND_LINK_* states:
3627 * - BOND_LINK_UP: carrier detected, updelay has passed.
3628 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3629 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3630 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3632 * The function bond_info_show_slave() only considers BOND_LINK_UP
3633 * to be "up" and anything else to be "down".
3635 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3639 netdev_get_etheraddr(iface->netdev, slave->mac);
3642 if (port->bond_fake_iface) {
3643 struct netdev *bond_netdev;
3645 if (!netdev_open_default(port->name, &bond_netdev)) {
3647 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3649 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3651 netdev_close(bond_netdev);
3655 proc_net_compat_update_bond(port->name, &bond);
3660 port_update_vlan_compat(struct port *port)
3662 struct bridge *br = port->bridge;
3663 char *vlandev_name = NULL;
3665 if (port->vlan > 0) {
3666 /* Figure out the name that the VLAN device should actually have, if it
3667 * existed. This takes some work because the VLAN device would not
3668 * have port->name in its name; rather, it would have the trunk port's
3669 * name, and 'port' would be attached to a bridge that also had the
3670 * VLAN device one of its ports. So we need to find a trunk port that
3671 * includes port->vlan.
3673 * There might be more than one candidate. This doesn't happen on
3674 * XenServer, so if it happens we just pick the first choice in
3675 * alphabetical order instead of creating multiple VLAN devices. */
3677 for (i = 0; i < br->n_ports; i++) {
3678 struct port *p = br->ports[i];
3679 if (port_trunks_vlan(p, port->vlan)
3681 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3683 uint8_t ea[ETH_ADDR_LEN];
3684 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3685 if (!eth_addr_is_multicast(ea) &&
3686 !eth_addr_is_reserved(ea) &&
3687 !eth_addr_is_zero(ea)) {
3688 vlandev_name = p->name;
3693 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3696 /* Interface functions. */
3698 static struct iface *
3699 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3701 struct bridge *br = port->bridge;
3702 struct iface *iface;
3703 char *name = if_cfg->name;
3705 iface = xzalloc(sizeof *iface);
3707 iface->port_ifidx = port->n_ifaces;
3708 iface->name = xstrdup(name);
3709 iface->dp_ifidx = -1;
3710 iface->tag = tag_create_random();
3711 iface->delay_expires = LLONG_MAX;
3712 iface->netdev = NULL;
3713 iface->cfg = if_cfg;
3715 shash_add_assert(&br->iface_by_name, iface->name, iface);
3717 if (port->n_ifaces >= port->allocated_ifaces) {
3718 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3719 sizeof *port->ifaces);
3721 port->ifaces[port->n_ifaces++] = iface;
3722 if (port->n_ifaces > 1) {
3723 br->has_bonded_ports = true;
3726 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3734 iface_destroy(struct iface *iface)
3737 struct port *port = iface->port;
3738 struct bridge *br = port->bridge;
3739 bool del_active = port->active_iface == iface->port_ifidx;
3742 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3744 if (iface->dp_ifidx >= 0) {
3745 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
3748 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3749 del->port_ifidx = iface->port_ifidx;
3751 netdev_close(iface->netdev);
3754 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3755 bond_choose_active_iface(port);
3756 bond_send_learning_packets(port);
3762 bridge_flush(port->bridge);
3766 static struct iface *
3767 iface_lookup(const struct bridge *br, const char *name)
3769 return shash_find_data(&br->iface_by_name, name);
3772 static struct iface *
3773 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3775 struct iface *iface;
3777 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
3778 hash_int(dp_ifidx, 0), &br->ifaces) {
3779 if (iface->dp_ifidx == dp_ifidx) {
3786 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3787 * 'br', that is, an interface that is entirely simulated within the datapath.
3788 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3789 * interfaces are created by setting "iface.<iface>.internal = true".
3791 * In addition, we have a kluge-y feature that creates an internal port with
3792 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3793 * This feature needs to go away in the long term. Until then, this is one
3794 * reason why this function takes a name instead of a struct iface: the fake
3795 * interfaces created this way do not have a struct iface. */
3797 iface_is_internal(const struct bridge *br, const char *if_name)
3799 struct iface *iface;
3802 if (!strcmp(if_name, br->name)) {
3806 iface = iface_lookup(br, if_name);
3807 if (iface && !strcmp(iface->cfg->type, "internal")) {
3811 port = port_lookup(br, if_name);
3812 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3818 /* Set Ethernet address of 'iface', if one is specified in the configuration
3821 iface_set_mac(struct iface *iface)
3823 uint8_t ea[ETH_ADDR_LEN];
3825 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3826 if (eth_addr_is_multicast(ea)) {
3827 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3829 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3830 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3831 iface->name, iface->name);
3833 int error = netdev_set_etheraddr(iface->netdev, ea);
3835 VLOG_ERR("interface %s: setting MAC failed (%s)",
3836 iface->name, strerror(error));
3842 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
3844 * The value strings in '*shash' are taken directly from values[], not copied,
3845 * so the caller should not modify or free them. */
3847 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3848 struct shash *shash)
3853 for (i = 0; i < n; i++) {
3854 shash_add(shash, keys[i], values[i]);
3858 struct iface_delete_queues_cbdata {
3859 struct netdev *netdev;
3860 const struct ovsdb_datum *queues;
3864 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
3866 union ovsdb_atom atom;
3868 atom.integer = target;
3869 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
3873 iface_delete_queues(unsigned int queue_id,
3874 const struct shash *details OVS_UNUSED, void *cbdata_)
3876 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3878 if (!queue_ids_include(cbdata->queues, queue_id)) {
3879 netdev_delete_queue(cbdata->netdev, queue_id);
3884 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3886 if (!qos || qos->type[0] == '\0') {
3887 netdev_set_qos(iface->netdev, NULL, NULL);
3889 struct iface_delete_queues_cbdata cbdata;
3890 struct shash details;
3893 /* Configure top-level Qos for 'iface'. */
3894 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3895 qos->n_other_config, &details);
3896 netdev_set_qos(iface->netdev, qos->type, &details);
3897 shash_destroy(&details);
3899 /* Deconfigure queues that were deleted. */
3900 cbdata.netdev = iface->netdev;
3901 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
3903 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3905 /* Configure queues for 'iface'. */
3906 for (i = 0; i < qos->n_queues; i++) {
3907 const struct ovsrec_queue *queue = qos->value_queues[i];
3908 unsigned int queue_id = qos->key_queues[i];
3910 shash_from_ovs_idl_map(queue->key_other_config,
3911 queue->value_other_config,
3912 queue->n_other_config, &details);
3913 netdev_set_queue(iface->netdev, queue_id, &details);
3914 shash_destroy(&details);
3919 /* Port mirroring. */
3921 static struct mirror *
3922 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
3926 for (i = 0; i < MAX_MIRRORS; i++) {
3927 struct mirror *m = br->mirrors[i];
3928 if (m && uuid_equals(uuid, &m->uuid)) {
3936 mirror_reconfigure(struct bridge *br)
3938 unsigned long *rspan_vlans;
3941 /* Get rid of deleted mirrors. */
3942 for (i = 0; i < MAX_MIRRORS; i++) {
3943 struct mirror *m = br->mirrors[i];
3945 const struct ovsdb_datum *mc;
3946 union ovsdb_atom atom;
3948 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
3949 atom.uuid = br->mirrors[i]->uuid;
3950 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
3956 /* Add new mirrors and reconfigure existing ones. */
3957 for (i = 0; i < br->cfg->n_mirrors; i++) {
3958 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3959 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
3961 mirror_reconfigure_one(m, cfg);
3963 mirror_create(br, cfg);
3967 /* Update port reserved status. */
3968 for (i = 0; i < br->n_ports; i++) {
3969 br->ports[i]->is_mirror_output_port = false;
3971 for (i = 0; i < MAX_MIRRORS; i++) {
3972 struct mirror *m = br->mirrors[i];
3973 if (m && m->out_port) {
3974 m->out_port->is_mirror_output_port = true;
3978 /* Update flooded vlans (for RSPAN). */
3980 if (br->cfg->n_flood_vlans) {
3981 rspan_vlans = bitmap_allocate(4096);
3983 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3984 int64_t vlan = br->cfg->flood_vlans[i];
3985 if (vlan >= 0 && vlan < 4096) {
3986 bitmap_set1(rspan_vlans, vlan);
3987 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3990 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3995 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4001 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4006 for (i = 0; ; i++) {
4007 if (i >= MAX_MIRRORS) {
4008 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4009 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4012 if (!br->mirrors[i]) {
4017 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4020 br->mirrors[i] = m = xzalloc(sizeof *m);
4023 m->name = xstrdup(cfg->name);
4024 shash_init(&m->src_ports);
4025 shash_init(&m->dst_ports);
4031 mirror_reconfigure_one(m, cfg);
4035 mirror_destroy(struct mirror *m)
4038 struct bridge *br = m->bridge;
4041 for (i = 0; i < br->n_ports; i++) {
4042 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4043 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4046 shash_destroy(&m->src_ports);
4047 shash_destroy(&m->dst_ports);
4050 m->bridge->mirrors[m->idx] = NULL;
4059 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4060 struct shash *names)
4064 for (i = 0; i < n_ports; i++) {
4065 const char *name = ports[i]->name;
4066 if (port_lookup(m->bridge, name)) {
4067 shash_add_once(names, name, NULL);
4069 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4070 "port %s", m->bridge->name, m->name, name);
4076 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4082 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4084 for (i = 0; i < cfg->n_select_vlan; i++) {
4085 int64_t vlan = cfg->select_vlan[i];
4086 if (vlan < 0 || vlan > 4095) {
4087 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4088 m->bridge->name, m->name, vlan);
4090 (*vlans)[n_vlans++] = vlan;
4097 vlan_is_mirrored(const struct mirror *m, int vlan)
4101 for (i = 0; i < m->n_vlans; i++) {
4102 if (m->vlans[i] == vlan) {
4110 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4114 for (i = 0; i < m->n_vlans; i++) {
4115 if (port_trunks_vlan(p, m->vlans[i])) {
4123 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4125 struct shash src_ports, dst_ports;
4126 mirror_mask_t mirror_bit;
4127 struct port *out_port;
4134 if (strcmp(cfg->name, m->name)) {
4136 m->name = xstrdup(cfg->name);
4139 /* Get output port. */
4140 if (cfg->output_port) {
4141 out_port = port_lookup(m->bridge, cfg->output_port->name);
4143 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4144 m->bridge->name, m->name);
4150 if (cfg->output_vlan) {
4151 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4152 "output vlan; ignoring output vlan",
4153 m->bridge->name, m->name);
4155 } else if (cfg->output_vlan) {
4157 out_vlan = *cfg->output_vlan;
4159 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4160 m->bridge->name, m->name);
4165 shash_init(&src_ports);
4166 shash_init(&dst_ports);
4167 if (cfg->select_all) {
4168 for (i = 0; i < m->bridge->n_ports; i++) {
4169 const char *name = m->bridge->ports[i]->name;
4170 shash_add_once(&src_ports, name, NULL);
4171 shash_add_once(&dst_ports, name, NULL);
4176 /* Get ports, and drop duplicates and ports that don't exist. */
4177 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4179 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4182 /* Get all the vlans, and drop duplicate and invalid vlans. */
4183 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4186 /* Update mirror data. */
4187 if (!shash_equal_keys(&m->src_ports, &src_ports)
4188 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4189 || m->n_vlans != n_vlans
4190 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4191 || m->out_port != out_port
4192 || m->out_vlan != out_vlan) {
4193 bridge_flush(m->bridge);
4195 shash_swap(&m->src_ports, &src_ports);
4196 shash_swap(&m->dst_ports, &dst_ports);
4199 m->n_vlans = n_vlans;
4200 m->out_port = out_port;
4201 m->out_vlan = out_vlan;
4204 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4205 for (i = 0; i < m->bridge->n_ports; i++) {
4206 struct port *port = m->bridge->ports[i];
4208 if (shash_find(&m->src_ports, port->name)
4211 ? port_trunks_any_mirrored_vlan(m, port)
4212 : vlan_is_mirrored(m, port->vlan)))) {
4213 port->src_mirrors |= mirror_bit;
4215 port->src_mirrors &= ~mirror_bit;
4218 if (shash_find(&m->dst_ports, port->name)) {
4219 port->dst_mirrors |= mirror_bit;
4221 port->dst_mirrors &= ~mirror_bit;
4226 shash_destroy(&src_ports);
4227 shash_destroy(&dst_ports);