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>
36 #include "dynamic-string.h"
41 #include "mac-learning.h"
43 #include "xflow-util.h"
44 #include "ofp-print.h"
46 #include "ofproto/netflow.h"
47 #include "ofproto/ofproto.h"
48 #include "ovsdb-data.h"
50 #include "poll-loop.h"
51 #include "port-array.h"
52 #include "proc-net-compat.h"
56 #include "socket-util.h"
57 #include "stream-ssl.h"
63 #include "vswitchd/vswitch-idl.h"
64 #include "xenserver.h"
68 #include "sflow_api.h"
70 VLOG_DEFINE_THIS_MODULE(bridge)
78 /* These members are always valid. */
79 struct port *port; /* Containing port. */
80 size_t port_ifidx; /* Index within containing port. */
81 char *name; /* Host network device name. */
82 tag_type tag; /* Tag associated with this interface. */
83 long long delay_expires; /* Time after which 'enabled' may change. */
85 /* These members are valid only after bridge_reconfigure() causes them to
87 int xf_ifidx; /* Index within kernel datapath. */
88 struct netdev *netdev; /* Network device. */
89 bool enabled; /* May be chosen for flows? */
90 const struct ovsrec_interface *cfg;
93 #define BOND_MASK 0xff
95 int iface_idx; /* Index of assigned iface, or -1 if none. */
96 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
97 tag_type iface_tag; /* Tag associated with iface_idx. */
100 #define MAX_MIRRORS 32
101 typedef uint32_t mirror_mask_t;
102 #define MIRROR_MASK_C(X) UINT32_C(X)
103 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
105 struct bridge *bridge;
108 struct uuid uuid; /* UUID of this "mirror" record in database. */
110 /* Selection criteria. */
111 struct shash src_ports; /* Name is port name; data is always NULL. */
112 struct shash dst_ports; /* Name is port name; data is always NULL. */
117 struct port *out_port;
121 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
123 struct bridge *bridge;
125 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
126 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
127 * NULL if all VLANs are trunked. */
128 const struct ovsrec_port *cfg;
131 /* An ordinary bridge port has 1 interface.
132 * A bridge port for bonding has at least 2 interfaces. */
133 struct iface **ifaces;
134 size_t n_ifaces, allocated_ifaces;
137 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
138 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
139 tag_type active_iface_tag; /* Tag for bcast flows. */
140 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
141 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
142 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
143 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
144 long long int bond_next_fake_iface_update; /* Time of next update. */
145 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
146 long long int bond_next_rebalance; /* Next rebalancing time. */
148 /* Port mirroring info. */
149 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
150 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
151 bool is_mirror_output_port; /* Does port mirroring send frames here? */
154 #define DP_MAX_PORTS 255
156 struct list node; /* Node in global list of bridges. */
157 char *name; /* User-specified arbitrary name. */
158 struct mac_learning *ml; /* MAC learning table. */
159 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
160 const struct ovsrec_bridge *cfg;
162 /* OpenFlow switch processing. */
163 struct ofproto *ofproto; /* OpenFlow switch. */
165 /* Kernel datapath information. */
166 struct xfif *xfif; /* Datapath. */
167 struct port_array ifaces; /* Indexed by kernel datapath port number. */
171 size_t n_ports, allocated_ports;
172 struct shash iface_by_name; /* "struct iface"s indexed by name. */
173 struct shash port_by_name; /* "struct port"s indexed by name. */
176 bool has_bonded_ports;
181 /* Port mirroring. */
182 struct mirror *mirrors[MAX_MIRRORS];
185 /* List of all bridges. */
186 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
188 /* OVSDB IDL used to obtain configuration. */
189 static struct ovsdb_idl *idl;
191 /* Each time this timer expires, the bridge fetches statistics for every
192 * interface and pushes them into the database. */
193 #define IFACE_STATS_INTERVAL (5 * 1000) /* In milliseconds. */
194 static long long int iface_stats_timer = LLONG_MIN;
196 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
197 static void bridge_destroy(struct bridge *);
198 static struct bridge *bridge_lookup(const char *name);
199 static unixctl_cb_func bridge_unixctl_dump_flows;
200 static unixctl_cb_func bridge_unixctl_reconnect;
201 static int bridge_run_one(struct bridge *);
202 static size_t bridge_get_controllers(const struct bridge *br,
203 struct ovsrec_controller ***controllersp);
204 static void bridge_reconfigure_one(struct bridge *);
205 static void bridge_reconfigure_remotes(struct bridge *,
206 const struct sockaddr_in *managers,
208 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
209 static void bridge_fetch_dp_ifaces(struct bridge *);
210 static void bridge_flush(struct bridge *);
211 static void bridge_pick_local_hw_addr(struct bridge *,
212 uint8_t ea[ETH_ADDR_LEN],
213 struct iface **hw_addr_iface);
214 static uint64_t bridge_pick_datapath_id(struct bridge *,
215 const uint8_t bridge_ea[ETH_ADDR_LEN],
216 struct iface *hw_addr_iface);
217 static struct iface *bridge_get_local_iface(struct bridge *);
218 static uint64_t dpid_from_hash(const void *, size_t nbytes);
220 static unixctl_cb_func bridge_unixctl_fdb_show;
222 static void bond_init(void);
223 static void bond_run(struct bridge *);
224 static void bond_wait(struct bridge *);
225 static void bond_rebalance_port(struct port *);
226 static void bond_send_learning_packets(struct port *);
227 static void bond_enable_slave(struct iface *iface, bool enable);
229 static struct port *port_create(struct bridge *, const char *name);
230 static void port_reconfigure(struct port *, const struct ovsrec_port *);
231 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
232 static void port_destroy(struct port *);
233 static struct port *port_lookup(const struct bridge *, const char *name);
234 static struct iface *port_lookup_iface(const struct port *, const char *name);
235 static struct port *port_from_xf_ifidx(const struct bridge *,
237 static void port_update_bond_compat(struct port *);
238 static void port_update_vlan_compat(struct port *);
239 static void port_update_bonding(struct port *);
241 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
242 static void mirror_destroy(struct mirror *);
243 static void mirror_reconfigure(struct bridge *);
244 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
245 static bool vlan_is_mirrored(const struct mirror *, int vlan);
247 static struct iface *iface_create(struct port *port,
248 const struct ovsrec_interface *if_cfg);
249 static void iface_destroy(struct iface *);
250 static struct iface *iface_lookup(const struct bridge *, const char *name);
251 static struct iface *iface_from_xf_ifidx(const struct bridge *,
253 static bool iface_is_internal(const struct bridge *, const char *name);
254 static void iface_set_mac(struct iface *);
255 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
257 /* Hooks into ofproto processing. */
258 static struct ofhooks bridge_ofhooks;
260 /* Public functions. */
262 /* Initializes the bridge module, configuring it to obtain its configuration
263 * from an OVSDB server accessed over 'remote', which should be a string in a
264 * form acceptable to ovsdb_idl_create(). */
266 bridge_init(const char *remote)
268 /* Create connection to database. */
269 idl = ovsdb_idl_create(remote, &ovsrec_idl_class);
271 ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_cur_cfg);
272 ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_statistics);
273 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
275 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
277 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
278 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
280 ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_ofport);
281 ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_statistics);
282 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
284 /* Register unixctl commands. */
285 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
286 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
288 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
293 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
294 * but for which the ovs-vswitchd configuration 'cfg' is required. */
296 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
298 static bool already_configured_once;
299 struct svec bridge_names;
300 struct svec xfif_names, xfif_types;
303 /* Only do this once per ovs-vswitchd run. */
304 if (already_configured_once) {
307 already_configured_once = true;
309 iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
311 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
312 svec_init(&bridge_names);
313 for (i = 0; i < cfg->n_bridges; i++) {
314 svec_add(&bridge_names, cfg->bridges[i]->name);
316 svec_sort(&bridge_names);
318 /* Iterate over all system xfifs and delete any of them that do not appear
320 svec_init(&xfif_names);
321 svec_init(&xfif_types);
322 xf_enumerate_types(&xfif_types);
323 for (i = 0; i < xfif_types.n; i++) {
328 xf_enumerate_names(xfif_types.names[i], &xfif_names);
330 /* For each xfif... */
331 for (j = 0; j < xfif_names.n; j++) {
332 retval = xfif_open(xfif_names.names[j], xfif_types.names[i], &xfif);
334 struct svec all_names;
337 /* ...check whether any of its names is in 'bridge_names'. */
338 svec_init(&all_names);
339 xfif_get_all_names(xfif, &all_names);
340 for (k = 0; k < all_names.n; k++) {
341 if (svec_contains(&bridge_names, all_names.names[k])) {
346 /* No. Delete the xfif. */
350 svec_destroy(&all_names);
355 svec_destroy(&bridge_names);
356 svec_destroy(&xfif_names);
357 svec_destroy(&xfif_types);
360 /* Attempt to create the network device 'iface_name' through the netdev
363 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
366 struct shash options;
370 shash_init(&options);
371 for (i = 0; i < iface_cfg->n_options; i++) {
372 shash_add(&options, iface_cfg->key_options[i],
373 xstrdup(iface_cfg->value_options[i]));
377 struct netdev_options netdev_options;
379 memset(&netdev_options, 0, sizeof netdev_options);
380 netdev_options.name = iface_cfg->name;
381 if (!strcmp(iface_cfg->type, "internal")) {
382 /* An "internal" config type maps to a netdev "system" type. */
383 netdev_options.type = "system";
385 netdev_options.type = iface_cfg->type;
387 netdev_options.args = &options;
388 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
389 error = netdev_open(&netdev_options, &iface->netdev);
392 netdev_get_carrier(iface->netdev, &iface->enabled);
394 } else if (iface->netdev) {
395 const char *netdev_type = netdev_get_type(iface->netdev);
396 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
397 ? iface_cfg->type : NULL;
399 /* An "internal" config type maps to a netdev "system" type. */
400 if (iface_type && !strcmp(iface_type, "internal")) {
401 iface_type = "system";
404 if (!iface_type || !strcmp(netdev_type, iface_type)) {
405 error = netdev_reconfigure(iface->netdev, &options);
407 VLOG_WARN("%s: attempting change device type from %s to %s",
408 iface_cfg->name, netdev_type, iface_type);
412 shash_destroy_free_data(&options);
418 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
420 return set_up_iface(iface_cfg, iface, false);
424 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
425 void *aux OVS_UNUSED)
427 if (!iface->netdev) {
428 int error = set_up_iface(iface->cfg, iface, true);
430 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
440 check_iface_xf_ifidx(struct bridge *br, struct iface *iface,
441 void *aux OVS_UNUSED)
443 if (iface->xf_ifidx >= 0) {
444 VLOG_DBG("%s has interface %s on port %d",
446 iface->name, iface->xf_ifidx);
449 VLOG_ERR("%s interface not in %s, dropping",
450 iface->name, xfif_name(br->xfif));
456 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
457 void *aux OVS_UNUSED)
459 /* Set policing attributes. */
460 netdev_set_policing(iface->netdev,
461 iface->cfg->ingress_policing_rate,
462 iface->cfg->ingress_policing_burst);
464 /* Set MAC address of internal interfaces other than the local
466 if (iface->xf_ifidx != XFLOWP_LOCAL
467 && iface_is_internal(br, iface->name)) {
468 iface_set_mac(iface);
474 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
475 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
476 * deletes from 'br' any ports that no longer have any interfaces. */
478 iterate_and_prune_ifaces(struct bridge *br,
479 bool (*cb)(struct bridge *, struct iface *,
485 for (i = 0; i < br->n_ports; ) {
486 struct port *port = br->ports[i];
487 for (j = 0; j < port->n_ifaces; ) {
488 struct iface *iface = port->ifaces[j];
489 if (cb(br, iface, aux)) {
492 iface_destroy(iface);
496 if (port->n_ifaces) {
499 VLOG_ERR("%s port has no interfaces, dropping", port->name);
505 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
506 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
507 * responsible for freeing '*managersp' (with free()).
509 * You may be asking yourself "why does ovs-vswitchd care?", because
510 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
511 * should not be and in fact is not directly involved in that. But
512 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
513 * it has to tell in-band control where the managers are to enable that.
516 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
517 struct sockaddr_in **managersp, size_t *n_managersp)
519 struct sockaddr_in *managers = NULL;
520 size_t n_managers = 0;
522 if (ovs_cfg->n_managers > 0) {
525 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
526 for (i = 0; i < ovs_cfg->n_managers; i++) {
527 const char *name = ovs_cfg->managers[i];
528 struct sockaddr_in *sin = &managers[i];
530 if ((!strncmp(name, "tcp:", 4)
531 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
532 (!strncmp(name, "ssl:", 4)
533 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
539 *managersp = managers;
540 *n_managersp = n_managers;
544 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
546 struct shash old_br, new_br;
547 struct shash_node *node;
548 struct bridge *br, *next;
549 struct sockaddr_in *managers;
552 int sflow_bridge_number;
554 COVERAGE_INC(bridge_reconfigure);
556 collect_managers(ovs_cfg, &managers, &n_managers);
558 /* Collect old and new bridges. */
561 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
562 shash_add(&old_br, br->name, br);
564 for (i = 0; i < ovs_cfg->n_bridges; i++) {
565 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
566 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
567 VLOG_WARN("more than one bridge named %s", br_cfg->name);
571 /* Get rid of deleted bridges and add new bridges. */
572 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
573 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
580 SHASH_FOR_EACH (node, &new_br) {
581 const char *br_name = node->name;
582 const struct ovsrec_bridge *br_cfg = node->data;
583 br = shash_find_data(&old_br, br_name);
585 /* If the bridge datapath type has changed, we need to tear it
586 * down and recreate. */
587 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
589 bridge_create(br_cfg);
592 bridge_create(br_cfg);
595 shash_destroy(&old_br);
596 shash_destroy(&new_br);
598 /* Reconfigure all bridges. */
599 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
600 bridge_reconfigure_one(br);
603 /* Add and delete ports on all datapaths.
605 * The kernel will reject any attempt to add a given port to a datapath if
606 * that port already belongs to a different datapath, so we must do all
607 * port deletions before any port additions. */
608 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
609 struct xflow_port *xfif_ports;
611 struct shash want_ifaces;
613 xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports);
614 bridge_get_all_ifaces(br, &want_ifaces);
615 for (i = 0; i < n_xfif_ports; i++) {
616 const struct xflow_port *p = &xfif_ports[i];
617 if (!shash_find(&want_ifaces, p->devname)
618 && strcmp(p->devname, br->name)) {
619 int retval = xfif_port_del(br->xfif, p->port);
621 VLOG_ERR("failed to remove %s interface from %s: %s",
622 p->devname, xfif_name(br->xfif),
627 shash_destroy(&want_ifaces);
630 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
631 struct xflow_port *xfif_ports;
633 struct shash cur_ifaces, want_ifaces;
634 struct shash_node *node;
636 /* Get the set of interfaces currently in this datapath. */
637 xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports);
638 shash_init(&cur_ifaces);
639 for (i = 0; i < n_xfif_ports; i++) {
640 const char *name = xfif_ports[i].devname;
641 shash_add_once(&cur_ifaces, name, NULL);
645 /* Get the set of interfaces we want on this datapath. */
646 bridge_get_all_ifaces(br, &want_ifaces);
648 SHASH_FOR_EACH (node, &want_ifaces) {
649 const char *if_name = node->name;
650 struct iface *iface = node->data;
652 if (shash_find(&cur_ifaces, if_name)) {
653 /* Already exists, just reconfigure it. */
655 reconfigure_iface(iface->cfg, iface);
658 /* Need to add to datapath. */
662 /* Add to datapath. */
663 internal = iface_is_internal(br, if_name);
664 error = xfif_port_add(br->xfif, if_name,
665 internal ? XFLOW_PORT_INTERNAL : 0, NULL);
666 if (error == EFBIG) {
667 VLOG_ERR("ran out of valid port numbers on %s",
668 xfif_name(br->xfif));
671 VLOG_ERR("failed to add %s interface to %s: %s",
672 if_name, xfif_name(br->xfif), strerror(error));
676 shash_destroy(&cur_ifaces);
677 shash_destroy(&want_ifaces);
679 sflow_bridge_number = 0;
680 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
683 struct iface *local_iface;
684 struct iface *hw_addr_iface;
687 bridge_fetch_dp_ifaces(br);
689 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
690 iterate_and_prune_ifaces(br, check_iface_xf_ifidx, NULL);
692 /* Pick local port hardware address, datapath ID. */
693 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
694 local_iface = bridge_get_local_iface(br);
696 int error = netdev_set_etheraddr(local_iface->netdev, ea);
698 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
699 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
700 "Ethernet address: %s",
701 br->name, strerror(error));
705 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
706 ofproto_set_datapath_id(br->ofproto, dpid);
708 dpid_string = xasprintf("%016"PRIx64, dpid);
709 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
712 /* Set NetFlow configuration on this bridge. */
713 if (br->cfg->netflow) {
714 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
715 struct netflow_options opts;
717 memset(&opts, 0, sizeof opts);
719 xfif_get_netflow_ids(br->xfif, &opts.engine_type, &opts.engine_id);
720 if (nf_cfg->engine_type) {
721 opts.engine_type = *nf_cfg->engine_type;
723 if (nf_cfg->engine_id) {
724 opts.engine_id = *nf_cfg->engine_id;
727 opts.active_timeout = nf_cfg->active_timeout;
728 if (!opts.active_timeout) {
729 opts.active_timeout = -1;
730 } else if (opts.active_timeout < 0) {
731 VLOG_WARN("bridge %s: active timeout interval set to negative "
732 "value, using default instead (%d seconds)", br->name,
733 NF_ACTIVE_TIMEOUT_DEFAULT);
734 opts.active_timeout = -1;
737 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
738 if (opts.add_id_to_iface) {
739 if (opts.engine_id > 0x7f) {
740 VLOG_WARN("bridge %s: netflow port mangling may conflict "
741 "with another vswitch, choose an engine id less "
742 "than 128", br->name);
744 if (br->n_ports > 508) {
745 VLOG_WARN("bridge %s: netflow port mangling will conflict "
746 "with another port when more than 508 ports are "
751 opts.collectors.n = nf_cfg->n_targets;
752 opts.collectors.names = nf_cfg->targets;
753 if (ofproto_set_netflow(br->ofproto, &opts)) {
754 VLOG_ERR("bridge %s: problem setting netflow collectors",
758 ofproto_set_netflow(br->ofproto, NULL);
761 /* Set sFlow configuration on this bridge. */
762 if (br->cfg->sflow) {
763 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
764 struct ovsrec_controller **controllers;
765 struct ofproto_sflow_options oso;
766 size_t n_controllers;
769 memset(&oso, 0, sizeof oso);
771 oso.targets.n = sflow_cfg->n_targets;
772 oso.targets.names = sflow_cfg->targets;
774 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
775 if (sflow_cfg->sampling) {
776 oso.sampling_rate = *sflow_cfg->sampling;
779 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
780 if (sflow_cfg->polling) {
781 oso.polling_interval = *sflow_cfg->polling;
784 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
785 if (sflow_cfg->header) {
786 oso.header_len = *sflow_cfg->header;
789 oso.sub_id = sflow_bridge_number++;
790 oso.agent_device = sflow_cfg->agent;
792 oso.control_ip = NULL;
793 n_controllers = bridge_get_controllers(br, &controllers);
794 for (i = 0; i < n_controllers; i++) {
795 if (controllers[i]->local_ip) {
796 oso.control_ip = controllers[i]->local_ip;
800 ofproto_set_sflow(br->ofproto, &oso);
802 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
804 ofproto_set_sflow(br->ofproto, NULL);
807 /* Update the controller and related settings. It would be more
808 * straightforward to call this from bridge_reconfigure_one(), but we
809 * can't do it there for two reasons. First, and most importantly, at
810 * that point we don't know the xf_ifidx of any interfaces that have
811 * been added to the bridge (because we haven't actually added them to
812 * the datapath). Second, at that point we haven't set the datapath ID
813 * yet; when a controller is configured, resetting the datapath ID will
814 * immediately disconnect from the controller, so it's better to set
815 * the datapath ID before the controller. */
816 bridge_reconfigure_remotes(br, managers, n_managers);
818 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
819 for (i = 0; i < br->n_ports; i++) {
820 struct port *port = br->ports[i];
823 port_update_vlan_compat(port);
824 port_update_bonding(port);
826 for (j = 0; j < port->n_ifaces; j++) {
827 iface_update_qos(port->ifaces[j], port->cfg->qos);
831 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
832 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
839 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
840 const struct ovsdb_idl_column *column,
843 const struct ovsdb_datum *datum;
844 union ovsdb_atom atom;
847 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
848 atom.string = (char *) key;
849 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
850 return idx == UINT_MAX ? NULL : datum->values[idx].string;
854 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
856 return get_ovsrec_key_value(&br_cfg->header_,
857 &ovsrec_bridge_col_other_config, key);
861 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
862 struct iface **hw_addr_iface)
868 *hw_addr_iface = NULL;
870 /* Did the user request a particular MAC? */
871 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
872 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
873 if (eth_addr_is_multicast(ea)) {
874 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
875 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
876 } else if (eth_addr_is_zero(ea)) {
877 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
883 /* Otherwise choose the minimum non-local MAC address among all of the
885 memset(ea, 0xff, sizeof ea);
886 for (i = 0; i < br->n_ports; i++) {
887 struct port *port = br->ports[i];
888 uint8_t iface_ea[ETH_ADDR_LEN];
891 /* Mirror output ports don't participate. */
892 if (port->is_mirror_output_port) {
896 /* Choose the MAC address to represent the port. */
897 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
898 /* Find the interface with this Ethernet address (if any) so that
899 * we can provide the correct devname to the caller. */
901 for (j = 0; j < port->n_ifaces; j++) {
902 struct iface *candidate = port->ifaces[j];
903 uint8_t candidate_ea[ETH_ADDR_LEN];
904 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
905 && eth_addr_equals(iface_ea, candidate_ea)) {
910 /* Choose the interface whose MAC address will represent the port.
911 * The Linux kernel bonding code always chooses the MAC address of
912 * the first slave added to a bond, and the Fedora networking
913 * scripts always add slaves to a bond in alphabetical order, so
914 * for compatibility we choose the interface with the name that is
915 * first in alphabetical order. */
916 iface = port->ifaces[0];
917 for (j = 1; j < port->n_ifaces; j++) {
918 struct iface *candidate = port->ifaces[j];
919 if (strcmp(candidate->name, iface->name) < 0) {
924 /* The local port doesn't count (since we're trying to choose its
925 * MAC address anyway). */
926 if (iface->xf_ifidx == XFLOWP_LOCAL) {
931 error = netdev_get_etheraddr(iface->netdev, iface_ea);
933 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
934 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
935 iface->name, strerror(error));
940 /* Compare against our current choice. */
941 if (!eth_addr_is_multicast(iface_ea) &&
942 !eth_addr_is_local(iface_ea) &&
943 !eth_addr_is_reserved(iface_ea) &&
944 !eth_addr_is_zero(iface_ea) &&
945 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
947 memcpy(ea, iface_ea, ETH_ADDR_LEN);
948 *hw_addr_iface = iface;
951 if (eth_addr_is_multicast(ea)) {
952 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
953 *hw_addr_iface = NULL;
954 VLOG_WARN("bridge %s: using default bridge Ethernet "
955 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
957 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
958 br->name, ETH_ADDR_ARGS(ea));
962 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
963 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
964 * an interface on 'br', then that interface must be passed in as
965 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
966 * 'hw_addr_iface' must be passed in as a null pointer. */
968 bridge_pick_datapath_id(struct bridge *br,
969 const uint8_t bridge_ea[ETH_ADDR_LEN],
970 struct iface *hw_addr_iface)
973 * The procedure for choosing a bridge MAC address will, in the most
974 * ordinary case, also choose a unique MAC that we can use as a datapath
975 * ID. In some special cases, though, multiple bridges will end up with
976 * the same MAC address. This is OK for the bridges, but it will confuse
977 * the OpenFlow controller, because each datapath needs a unique datapath
980 * Datapath IDs must be unique. It is also very desirable that they be
981 * stable from one run to the next, so that policy set on a datapath
984 const char *datapath_id;
987 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
988 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
994 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
996 * A bridge whose MAC address is taken from a VLAN network device
997 * (that is, a network device created with vconfig(8) or similar
998 * tool) will have the same MAC address as a bridge on the VLAN
999 * device's physical network device.
1001 * Handle this case by hashing the physical network device MAC
1002 * along with the VLAN identifier.
1004 uint8_t buf[ETH_ADDR_LEN + 2];
1005 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1006 buf[ETH_ADDR_LEN] = vlan >> 8;
1007 buf[ETH_ADDR_LEN + 1] = vlan;
1008 return dpid_from_hash(buf, sizeof buf);
1011 * Assume that this bridge's MAC address is unique, since it
1012 * doesn't fit any of the cases we handle specially.
1017 * A purely internal bridge, that is, one that has no non-virtual
1018 * network devices on it at all, is more difficult because it has no
1019 * natural unique identifier at all.
1021 * When the host is a XenServer, we handle this case by hashing the
1022 * host's UUID with the name of the bridge. Names of bridges are
1023 * persistent across XenServer reboots, although they can be reused if
1024 * an internal network is destroyed and then a new one is later
1025 * created, so this is fairly effective.
1027 * When the host is not a XenServer, we punt by using a random MAC
1028 * address on each run.
1030 const char *host_uuid = xenserver_get_host_uuid();
1032 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1033 dpid = dpid_from_hash(combined, strlen(combined));
1039 return eth_addr_to_uint64(bridge_ea);
1043 dpid_from_hash(const void *data, size_t n)
1045 uint8_t hash[SHA1_DIGEST_SIZE];
1047 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1048 sha1_bytes(data, n, hash);
1049 eth_addr_mark_random(hash);
1050 return eth_addr_to_uint64(hash);
1054 iface_refresh_stats(struct iface *iface)
1060 static const struct iface_stat iface_stats[] = {
1061 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1062 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1063 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1064 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1065 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1066 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1067 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1068 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1069 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1070 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1071 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1072 { "collisions", offsetof(struct netdev_stats, collisions) },
1074 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1075 const struct iface_stat *s;
1077 char *keys[N_STATS];
1078 int64_t values[N_STATS];
1081 struct netdev_stats stats;
1083 /* Intentionally ignore return value, since errors will set 'stats' to
1084 * all-1s, and we will deal with that correctly below. */
1085 netdev_get_stats(iface->netdev, &stats);
1088 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1089 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1090 if (value != UINT64_MAX) {
1097 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1103 const struct ovsrec_open_vswitch *cfg;
1105 bool datapath_destroyed;
1106 bool database_changed;
1109 /* Let each bridge do the work that it needs to do. */
1110 datapath_destroyed = false;
1111 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1112 int error = bridge_run_one(br);
1114 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1115 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1116 "forcing reconfiguration", br->name);
1117 datapath_destroyed = true;
1121 /* (Re)configure if necessary. */
1122 database_changed = ovsdb_idl_run(idl);
1123 cfg = ovsrec_open_vswitch_first(idl);
1124 if (database_changed || datapath_destroyed) {
1126 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1128 bridge_configure_once(cfg);
1129 bridge_reconfigure(cfg);
1131 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1132 ovsdb_idl_txn_commit(txn);
1133 ovsdb_idl_txn_destroy(txn); /* XXX */
1135 /* We still need to reconfigure to avoid dangling pointers to
1136 * now-destroyed ovsrec structures inside bridge data. */
1137 static const struct ovsrec_open_vswitch null_cfg;
1139 bridge_reconfigure(&null_cfg);
1144 /* Re-configure SSL. We do this on every trip through the main loop,
1145 * instead of just when the database changes, because the contents of the
1146 * key and certificate files can change without the database changing. */
1147 if (cfg && cfg->ssl) {
1148 const struct ovsrec_ssl *ssl = cfg->ssl;
1150 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1151 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1155 /* Refresh interface stats if necessary. */
1156 if (time_msec() >= iface_stats_timer) {
1157 struct ovsdb_idl_txn *txn;
1159 txn = ovsdb_idl_txn_create(idl);
1160 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1163 for (i = 0; i < br->n_ports; i++) {
1164 struct port *port = br->ports[i];
1167 for (j = 0; j < port->n_ifaces; j++) {
1168 struct iface *iface = port->ifaces[j];
1169 iface_refresh_stats(iface);
1173 ovsdb_idl_txn_commit(txn);
1174 ovsdb_idl_txn_destroy(txn); /* XXX */
1176 iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
1185 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1186 ofproto_wait(br->ofproto);
1187 if (ofproto_has_primary_controller(br->ofproto)) {
1191 mac_learning_wait(br->ml);
1194 ovsdb_idl_wait(idl);
1195 poll_timer_wait_until(iface_stats_timer);
1198 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1199 * configuration changes. */
1201 bridge_flush(struct bridge *br)
1203 COVERAGE_INC(bridge_flush);
1204 ofproto_revalidate_all(br->ofproto);
1205 mac_learning_flush(br->ml);
1208 /* Returns the 'br' interface for the XFLOWP_LOCAL port, or null if 'br' has no
1209 * such interface. */
1210 static struct iface *
1211 bridge_get_local_iface(struct bridge *br)
1215 for (i = 0; i < br->n_ports; i++) {
1216 struct port *port = br->ports[i];
1217 for (j = 0; j < port->n_ifaces; j++) {
1218 struct iface *iface = port->ifaces[j];
1219 if (iface->xf_ifidx == XFLOWP_LOCAL) {
1228 /* Bridge unixctl user interface functions. */
1230 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1231 const char *args, void *aux OVS_UNUSED)
1233 struct ds ds = DS_EMPTY_INITIALIZER;
1234 const struct bridge *br;
1235 const struct mac_entry *e;
1237 br = bridge_lookup(args);
1239 unixctl_command_reply(conn, 501, "no such bridge");
1243 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1244 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1245 if (e->port < 0 || e->port >= br->n_ports) {
1248 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1249 br->ports[e->port]->ifaces[0]->xf_ifidx,
1250 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1252 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1256 /* Bridge reconfiguration functions. */
1257 static struct bridge *
1258 bridge_create(const struct ovsrec_bridge *br_cfg)
1263 assert(!bridge_lookup(br_cfg->name));
1264 br = xzalloc(sizeof *br);
1266 error = xfif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1272 xfif_flow_flush(br->xfif);
1274 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1277 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1279 xfif_delete(br->xfif);
1280 xfif_close(br->xfif);
1285 br->name = xstrdup(br_cfg->name);
1287 br->ml = mac_learning_create();
1288 eth_addr_nicira_random(br->default_ea);
1290 port_array_init(&br->ifaces);
1292 shash_init(&br->port_by_name);
1293 shash_init(&br->iface_by_name);
1295 list_push_back(&all_bridges, &br->node);
1297 VLOG_INFO("created bridge %s on %s", br->name, xfif_name(br->xfif));
1303 bridge_destroy(struct bridge *br)
1308 while (br->n_ports > 0) {
1309 port_destroy(br->ports[br->n_ports - 1]);
1311 list_remove(&br->node);
1312 error = xfif_delete(br->xfif);
1313 if (error && error != ENOENT) {
1314 VLOG_ERR("failed to delete %s: %s",
1315 xfif_name(br->xfif), strerror(error));
1317 xfif_close(br->xfif);
1318 ofproto_destroy(br->ofproto);
1319 mac_learning_destroy(br->ml);
1320 port_array_destroy(&br->ifaces);
1321 shash_destroy(&br->port_by_name);
1322 shash_destroy(&br->iface_by_name);
1329 static struct bridge *
1330 bridge_lookup(const char *name)
1334 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1335 if (!strcmp(br->name, name)) {
1342 /* Handle requests for a listing of all flows known by the OpenFlow
1343 * stack, including those normally hidden. */
1345 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1346 const char *args, void *aux OVS_UNUSED)
1351 br = bridge_lookup(args);
1353 unixctl_command_reply(conn, 501, "Unknown bridge");
1358 ofproto_get_all_flows(br->ofproto, &results);
1360 unixctl_command_reply(conn, 200, ds_cstr(&results));
1361 ds_destroy(&results);
1364 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1365 * connections and reconnect. If BRIDGE is not specified, then all bridges
1366 * drop their controller connections and reconnect. */
1368 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1369 const char *args, void *aux OVS_UNUSED)
1372 if (args[0] != '\0') {
1373 br = bridge_lookup(args);
1375 unixctl_command_reply(conn, 501, "Unknown bridge");
1378 ofproto_reconnect_controllers(br->ofproto);
1380 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1381 ofproto_reconnect_controllers(br->ofproto);
1384 unixctl_command_reply(conn, 200, NULL);
1388 bridge_run_one(struct bridge *br)
1390 ofproto_revalidate(br->ofproto, mac_learning_run(br->ml));
1392 return ofproto_run(br->ofproto);
1396 bridge_get_controllers(const struct bridge *br,
1397 struct ovsrec_controller ***controllersp)
1399 struct ovsrec_controller **controllers;
1400 size_t n_controllers;
1402 controllers = br->cfg->controller;
1403 n_controllers = br->cfg->n_controller;
1405 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1411 *controllersp = controllers;
1413 return n_controllers;
1417 bridge_reconfigure_one(struct bridge *br)
1419 struct shash old_ports, new_ports;
1420 struct svec snoops, old_snoops;
1421 struct shash_node *node;
1422 enum ofproto_fail_mode fail_mode;
1425 /* Collect old ports. */
1426 shash_init(&old_ports);
1427 for (i = 0; i < br->n_ports; i++) {
1428 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1431 /* Collect new ports. */
1432 shash_init(&new_ports);
1433 for (i = 0; i < br->cfg->n_ports; i++) {
1434 const char *name = br->cfg->ports[i]->name;
1435 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1436 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1441 /* If we have a controller, then we need a local port. Complain if the
1442 * user didn't specify one.
1444 * XXX perhaps we should synthesize a port ourselves in this case. */
1445 if (bridge_get_controllers(br, NULL)) {
1446 char local_name[IF_NAMESIZE];
1449 error = xfif_port_get_name(br->xfif, XFLOWP_LOCAL,
1450 local_name, sizeof local_name);
1451 if (!error && !shash_find(&new_ports, local_name)) {
1452 VLOG_WARN("bridge %s: controller specified but no local port "
1453 "(port named %s) defined",
1454 br->name, local_name);
1458 /* Get rid of deleted ports.
1459 * Get rid of deleted interfaces on ports that still exist. */
1460 SHASH_FOR_EACH (node, &old_ports) {
1461 struct port *port = node->data;
1462 const struct ovsrec_port *port_cfg;
1464 port_cfg = shash_find_data(&new_ports, node->name);
1468 port_del_ifaces(port, port_cfg);
1472 /* Create new ports.
1473 * Add new interfaces to existing ports.
1474 * Reconfigure existing ports. */
1475 SHASH_FOR_EACH (node, &new_ports) {
1476 struct port *port = shash_find_data(&old_ports, node->name);
1478 port = port_create(br, node->name);
1481 port_reconfigure(port, node->data);
1482 if (!port->n_ifaces) {
1483 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1484 br->name, port->name);
1488 shash_destroy(&old_ports);
1489 shash_destroy(&new_ports);
1491 /* Set the fail-mode */
1492 fail_mode = !br->cfg->fail_mode
1493 || !strcmp(br->cfg->fail_mode, "standalone")
1494 ? OFPROTO_FAIL_STANDALONE
1495 : OFPROTO_FAIL_SECURE;
1496 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1497 && !ofproto_has_primary_controller(br->ofproto)) {
1498 ofproto_flush_flows(br->ofproto);
1500 ofproto_set_fail_mode(br->ofproto, fail_mode);
1502 /* Delete all flows if we're switching from connected to standalone or vice
1503 * versa. (XXX Should we delete all flows if we are switching from one
1504 * controller to another?) */
1506 /* Configure OpenFlow controller connection snooping. */
1508 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1509 ovs_rundir, br->name));
1510 svec_init(&old_snoops);
1511 ofproto_get_snoops(br->ofproto, &old_snoops);
1512 if (!svec_equal(&snoops, &old_snoops)) {
1513 ofproto_set_snoops(br->ofproto, &snoops);
1515 svec_destroy(&snoops);
1516 svec_destroy(&old_snoops);
1518 mirror_reconfigure(br);
1521 /* Initializes 'oc' appropriately as a management service controller for
1524 * The caller must free oc->target when it is no longer needed. */
1526 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1527 struct ofproto_controller *oc)
1529 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir, br->name);
1530 oc->max_backoff = 0;
1531 oc->probe_interval = 60;
1532 oc->band = OFPROTO_OUT_OF_BAND;
1533 oc->accept_re = NULL;
1534 oc->update_resolv_conf = false;
1536 oc->burst_limit = 0;
1539 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1541 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1542 struct ofproto_controller *oc)
1544 oc->target = c->target;
1545 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1546 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1547 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1548 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1549 oc->accept_re = c->discover_accept_regex;
1550 oc->update_resolv_conf = c->discover_update_resolv_conf;
1551 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1552 oc->burst_limit = (c->controller_burst_limit
1553 ? *c->controller_burst_limit : 0);
1556 /* Configures the IP stack for 'br''s local interface properly according to the
1557 * configuration in 'c'. */
1559 bridge_configure_local_iface_netdev(struct bridge *br,
1560 struct ovsrec_controller *c)
1562 struct netdev *netdev;
1563 struct in_addr mask, gateway;
1565 struct iface *local_iface;
1568 /* Controller discovery does its own TCP/IP configuration later. */
1569 if (strcmp(c->target, "discover")) {
1573 /* If there's no local interface or no IP address, give up. */
1574 local_iface = bridge_get_local_iface(br);
1575 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1579 /* Bring up the local interface. */
1580 netdev = local_iface->netdev;
1581 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1583 /* Configure the IP address and netmask. */
1584 if (!c->local_netmask
1585 || !inet_aton(c->local_netmask, &mask)
1587 mask.s_addr = guess_netmask(ip.s_addr);
1589 if (!netdev_set_in4(netdev, ip, mask)) {
1590 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1591 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1594 /* Configure the default gateway. */
1595 if (c->local_gateway
1596 && inet_aton(c->local_gateway, &gateway)
1597 && gateway.s_addr) {
1598 if (!netdev_add_router(netdev, gateway)) {
1599 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1600 br->name, IP_ARGS(&gateway.s_addr));
1606 bridge_reconfigure_remotes(struct bridge *br,
1607 const struct sockaddr_in *managers,
1610 struct ovsrec_controller **controllers;
1611 size_t n_controllers;
1614 struct ofproto_controller *ocs;
1618 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1619 had_primary = ofproto_has_primary_controller(br->ofproto);
1621 n_controllers = bridge_get_controllers(br, &controllers);
1623 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1626 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1627 for (i = 0; i < n_controllers; i++) {
1628 struct ovsrec_controller *c = controllers[i];
1630 if (!strncmp(c->target, "punix:", 6)
1631 || !strncmp(c->target, "unix:", 5)) {
1632 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1634 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1635 * domain sockets and overwriting arbitrary local files. */
1636 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1637 "\"%s\" due to possibility for remote exploit",
1638 br->name, c->target);
1642 bridge_configure_local_iface_netdev(br, c);
1643 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs++]);
1646 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1647 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1650 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1651 ofproto_flush_flows(br->ofproto);
1654 /* If there are no controllers and the bridge is in standalone
1655 * mode, set up a flow that matches every packet and directs
1656 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1657 * switch is in secure mode and we won't pass any traffic until
1658 * a controller has been defined and it tells us to do so. */
1660 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1661 union ofp_action action;
1664 memset(&action, 0, sizeof action);
1665 action.type = htons(OFPAT_OUTPUT);
1666 action.output.len = htons(sizeof action);
1667 action.output.port = htons(OFPP_NORMAL);
1668 memset(&flow, 0, sizeof flow);
1669 flow.wildcards = OVSFW_ALL;
1670 ofproto_add_flow(br->ofproto, &flow, &action, 1, 0);
1675 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1680 for (i = 0; i < br->n_ports; i++) {
1681 struct port *port = br->ports[i];
1682 for (j = 0; j < port->n_ifaces; j++) {
1683 struct iface *iface = port->ifaces[j];
1684 shash_add_once(ifaces, iface->name, iface);
1686 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1687 shash_add_once(ifaces, port->name, NULL);
1692 /* For robustness, in case the administrator moves around datapath ports behind
1693 * our back, we re-check all the datapath port numbers here.
1695 * This function will set the 'xf_ifidx' members of interfaces that have
1696 * disappeared to -1, so only call this function from a context where those
1697 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1698 * 'xf_ifidx'es will cause trouble later when we try to send them to the
1699 * datapath, which doesn't support UINT16_MAX+1 ports. */
1701 bridge_fetch_dp_ifaces(struct bridge *br)
1703 struct xflow_port *xfif_ports;
1704 size_t n_xfif_ports;
1707 /* Reset all interface numbers. */
1708 for (i = 0; i < br->n_ports; i++) {
1709 struct port *port = br->ports[i];
1710 for (j = 0; j < port->n_ifaces; j++) {
1711 struct iface *iface = port->ifaces[j];
1712 iface->xf_ifidx = -1;
1715 port_array_clear(&br->ifaces);
1717 xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports);
1718 for (i = 0; i < n_xfif_ports; i++) {
1719 struct xflow_port *p = &xfif_ports[i];
1720 struct iface *iface = iface_lookup(br, p->devname);
1722 if (iface->xf_ifidx >= 0) {
1723 VLOG_WARN("%s reported interface %s twice",
1724 xfif_name(br->xfif), p->devname);
1725 } else if (iface_from_xf_ifidx(br, p->port)) {
1726 VLOG_WARN("%s reported interface %"PRIu16" twice",
1727 xfif_name(br->xfif), p->port);
1729 port_array_set(&br->ifaces, p->port, iface);
1730 iface->xf_ifidx = p->port;
1734 int64_t ofport = (iface->xf_ifidx >= 0
1735 ? xflow_port_to_ofp_port(iface->xf_ifidx)
1737 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1744 /* Bridge packet processing functions. */
1747 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1749 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1752 static struct bond_entry *
1753 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1755 return &port->bond_hash[bond_hash(mac)];
1759 bond_choose_iface(const struct port *port)
1761 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1762 size_t i, best_down_slave = -1;
1763 long long next_delay_expiration = LLONG_MAX;
1765 for (i = 0; i < port->n_ifaces; i++) {
1766 struct iface *iface = port->ifaces[i];
1768 if (iface->enabled) {
1770 } else if (iface->delay_expires < next_delay_expiration) {
1771 best_down_slave = i;
1772 next_delay_expiration = iface->delay_expires;
1776 if (best_down_slave != -1) {
1777 struct iface *iface = port->ifaces[best_down_slave];
1779 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1780 "since no other interface is up", iface->name,
1781 iface->delay_expires - time_msec());
1782 bond_enable_slave(iface, true);
1785 return best_down_slave;
1789 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1790 uint16_t *xf_ifidx, tag_type *tags)
1792 struct iface *iface;
1794 assert(port->n_ifaces);
1795 if (port->n_ifaces == 1) {
1796 iface = port->ifaces[0];
1798 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1799 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1800 || !port->ifaces[e->iface_idx]->enabled) {
1801 /* XXX select interface properly. The current interface selection
1802 * is only good for testing the rebalancing code. */
1803 e->iface_idx = bond_choose_iface(port);
1804 if (e->iface_idx < 0) {
1805 *tags |= port->no_ifaces_tag;
1808 e->iface_tag = tag_create_random();
1809 ((struct port *) port)->bond_compat_is_stale = true;
1811 *tags |= e->iface_tag;
1812 iface = port->ifaces[e->iface_idx];
1814 *xf_ifidx = iface->xf_ifidx;
1815 *tags |= iface->tag; /* Currently only used for bonding. */
1820 bond_link_status_update(struct iface *iface, bool carrier)
1822 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1823 struct port *port = iface->port;
1825 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1826 /* Nothing to do. */
1829 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1830 iface->name, carrier ? "detected" : "dropped");
1831 if (carrier == iface->enabled) {
1832 iface->delay_expires = LLONG_MAX;
1833 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1834 iface->name, carrier ? "disabled" : "enabled");
1835 } else if (carrier && port->active_iface < 0) {
1836 bond_enable_slave(iface, true);
1837 if (port->updelay) {
1838 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1839 "other interface is up", iface->name, port->updelay);
1842 int delay = carrier ? port->updelay : port->downdelay;
1843 iface->delay_expires = time_msec() + delay;
1846 "interface %s: will be %s if it stays %s for %d ms",
1848 carrier ? "enabled" : "disabled",
1849 carrier ? "up" : "down",
1856 bond_choose_active_iface(struct port *port)
1858 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1860 port->active_iface = bond_choose_iface(port);
1861 port->active_iface_tag = tag_create_random();
1862 if (port->active_iface >= 0) {
1863 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1864 port->name, port->ifaces[port->active_iface]->name);
1866 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1872 bond_enable_slave(struct iface *iface, bool enable)
1874 struct port *port = iface->port;
1875 struct bridge *br = port->bridge;
1877 /* This acts as a recursion check. If the act of disabling a slave
1878 * causes a different slave to be enabled, the flag will allow us to
1879 * skip redundant work when we reenter this function. It must be
1880 * cleared on exit to keep things safe with multiple bonds. */
1881 static bool moving_active_iface = false;
1883 iface->delay_expires = LLONG_MAX;
1884 if (enable == iface->enabled) {
1888 iface->enabled = enable;
1889 if (!iface->enabled) {
1890 VLOG_WARN("interface %s: disabled", iface->name);
1891 ofproto_revalidate(br->ofproto, iface->tag);
1892 if (iface->port_ifidx == port->active_iface) {
1893 ofproto_revalidate(br->ofproto,
1894 port->active_iface_tag);
1896 /* Disabling a slave can lead to another slave being immediately
1897 * enabled if there will be no active slaves but one is waiting
1898 * on an updelay. In this case we do not need to run most of the
1899 * code for the newly enabled slave since there was no period
1900 * without an active slave and it is redundant with the disabling
1902 moving_active_iface = true;
1903 bond_choose_active_iface(port);
1905 bond_send_learning_packets(port);
1907 VLOG_WARN("interface %s: enabled", iface->name);
1908 if (port->active_iface < 0 && !moving_active_iface) {
1909 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1910 bond_choose_active_iface(port);
1911 bond_send_learning_packets(port);
1913 iface->tag = tag_create_random();
1916 moving_active_iface = false;
1917 port->bond_compat_is_stale = true;
1920 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1921 * bond interface. */
1923 bond_update_fake_iface_stats(struct port *port)
1925 struct netdev_stats bond_stats;
1926 struct netdev *bond_dev;
1929 memset(&bond_stats, 0, sizeof bond_stats);
1931 for (i = 0; i < port->n_ifaces; i++) {
1932 struct netdev_stats slave_stats;
1934 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1935 /* XXX: We swap the stats here because they are swapped back when
1936 * reported by the internal device. The reason for this is
1937 * internal devices normally represent packets going into the system
1938 * but when used as fake bond device they represent packets leaving
1939 * the system. We really should do this in the internal device
1940 * itself because changing it here reverses the counts from the
1941 * perspective of the switch. However, the internal device doesn't
1942 * know what type of device it represents so we have to do it here
1944 bond_stats.tx_packets += slave_stats.rx_packets;
1945 bond_stats.tx_bytes += slave_stats.rx_bytes;
1946 bond_stats.rx_packets += slave_stats.tx_packets;
1947 bond_stats.rx_bytes += slave_stats.tx_bytes;
1951 if (!netdev_open_default(port->name, &bond_dev)) {
1952 netdev_set_stats(bond_dev, &bond_stats);
1953 netdev_close(bond_dev);
1958 bond_run(struct bridge *br)
1962 for (i = 0; i < br->n_ports; i++) {
1963 struct port *port = br->ports[i];
1965 if (port->n_ifaces >= 2) {
1966 for (j = 0; j < port->n_ifaces; j++) {
1967 struct iface *iface = port->ifaces[j];
1968 if (time_msec() >= iface->delay_expires) {
1969 bond_enable_slave(iface, !iface->enabled);
1973 if (port->bond_fake_iface
1974 && time_msec() >= port->bond_next_fake_iface_update) {
1975 bond_update_fake_iface_stats(port);
1976 port->bond_next_fake_iface_update = time_msec() + 1000;
1980 if (port->bond_compat_is_stale) {
1981 port->bond_compat_is_stale = false;
1982 port_update_bond_compat(port);
1988 bond_wait(struct bridge *br)
1992 for (i = 0; i < br->n_ports; i++) {
1993 struct port *port = br->ports[i];
1994 if (port->n_ifaces < 2) {
1997 for (j = 0; j < port->n_ifaces; j++) {
1998 struct iface *iface = port->ifaces[j];
1999 if (iface->delay_expires != LLONG_MAX) {
2000 poll_timer_wait_until(iface->delay_expires);
2003 if (port->bond_fake_iface) {
2004 poll_timer_wait_until(port->bond_next_fake_iface_update);
2010 set_dst(struct dst *p, const flow_t *flow,
2011 const struct port *in_port, const struct port *out_port,
2014 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2015 : in_port->vlan >= 0 ? in_port->vlan
2016 : ntohs(flow->dl_vlan));
2017 return choose_output_iface(out_port, flow->dl_src, &p->xf_ifidx, tags);
2021 swap_dst(struct dst *p, struct dst *q)
2023 struct dst tmp = *p;
2028 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2029 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2030 * that we push to the datapath. We could in fact fully sort the array by
2031 * vlan, but in most cases there are at most two different vlan tags so that's
2032 * possibly overkill.) */
2034 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2036 struct dst *first = dsts;
2037 struct dst *last = dsts + n_dsts;
2039 while (first != last) {
2041 * - All dsts < first have vlan == 'vlan'.
2042 * - All dsts >= last have vlan != 'vlan'.
2043 * - first < last. */
2044 while (first->vlan == vlan) {
2045 if (++first == last) {
2050 /* Same invariants, plus one additional:
2051 * - first->vlan != vlan.
2053 while (last[-1].vlan != vlan) {
2054 if (--last == first) {
2059 /* Same invariants, plus one additional:
2060 * - last[-1].vlan == vlan.*/
2061 swap_dst(first++, --last);
2066 mirror_mask_ffs(mirror_mask_t mask)
2068 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2073 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2074 const struct dst *test)
2077 for (i = 0; i < n_dsts; i++) {
2078 if (dsts[i].vlan == test->vlan && dsts[i].xf_ifidx == test->xf_ifidx) {
2086 port_trunks_vlan(const struct port *port, uint16_t vlan)
2088 return (port->vlan < 0
2089 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2093 port_includes_vlan(const struct port *port, uint16_t vlan)
2095 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2099 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2100 const struct port *in_port, const struct port *out_port,
2101 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2103 mirror_mask_t mirrors = in_port->src_mirrors;
2104 struct dst *dst = dsts;
2107 if (out_port == FLOOD_PORT) {
2108 /* XXX use XFLOW_FLOOD if no vlans or bonding. */
2109 /* XXX even better, define each VLAN as a datapath port group */
2110 for (i = 0; i < br->n_ports; i++) {
2111 struct port *port = br->ports[i];
2112 if (port != in_port && port_includes_vlan(port, vlan)
2113 && !port->is_mirror_output_port
2114 && set_dst(dst, flow, in_port, port, tags)) {
2115 mirrors |= port->dst_mirrors;
2119 *nf_output_iface = NF_OUT_FLOOD;
2120 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2121 *nf_output_iface = dst->xf_ifidx;
2122 mirrors |= out_port->dst_mirrors;
2127 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2128 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2130 if (set_dst(dst, flow, in_port, m->out_port, tags)
2131 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2135 for (i = 0; i < br->n_ports; i++) {
2136 struct port *port = br->ports[i];
2137 if (port_includes_vlan(port, m->out_vlan)
2138 && set_dst(dst, flow, in_port, port, tags))
2142 if (port->vlan < 0) {
2143 dst->vlan = m->out_vlan;
2145 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2149 /* Use the vlan tag on the original flow instead of
2150 * the one passed in the vlan parameter. This ensures
2151 * that we compare the vlan from before any implicit
2152 * tagging tags place. This is necessary because
2153 * dst->vlan is the final vlan, after removing implicit
2155 flow_vlan = ntohs(flow->dl_vlan);
2156 if (flow_vlan == 0) {
2157 flow_vlan = OFP_VLAN_NONE;
2159 if (port == in_port && dst->vlan == flow_vlan) {
2160 /* Don't send out input port on same VLAN. */
2168 mirrors &= mirrors - 1;
2171 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2175 static void OVS_UNUSED
2176 print_dsts(const struct dst *dsts, size_t n)
2178 for (; n--; dsts++) {
2179 printf(">p%"PRIu16, dsts->xf_ifidx);
2180 if (dsts->vlan != OFP_VLAN_NONE) {
2181 printf("v%"PRIu16, dsts->vlan);
2187 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2188 const struct port *in_port, const struct port *out_port,
2189 tag_type *tags, struct xflow_actions *actions,
2190 uint16_t *nf_output_iface)
2192 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2194 const struct dst *p;
2197 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2200 cur_vlan = ntohs(flow->dl_vlan);
2201 for (p = dsts; p < &dsts[n_dsts]; p++) {
2202 union xflow_action *a;
2203 if (p->vlan != cur_vlan) {
2204 if (p->vlan == OFP_VLAN_NONE) {
2205 xflow_actions_add(actions, XFLOWAT_STRIP_VLAN);
2207 a = xflow_actions_add(actions, XFLOWAT_SET_DL_TCI);
2208 a->dl_tci.tci = htons(p->vlan & VLAN_VID_MASK);
2209 a->dl_tci.mask = htons(VLAN_VID_MASK);
2213 a = xflow_actions_add(actions, XFLOWAT_OUTPUT);
2214 a->output.port = p->xf_ifidx;
2218 /* Returns the effective vlan of a packet, taking into account both the
2219 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2220 * the packet is untagged and -1 indicates it has an invalid header and
2221 * should be dropped. */
2222 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2223 struct port *in_port, bool have_packet)
2225 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2226 * belongs to VLAN 0, so we should treat both cases identically. (In the
2227 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2228 * presumably to allow a priority to be specified. In the latter case, the
2229 * packet does not have any 802.1Q header.) */
2230 int vlan = ntohs(flow->dl_vlan);
2231 if (vlan == OFP_VLAN_NONE) {
2234 if (in_port->vlan >= 0) {
2236 /* XXX support double tagging? */
2238 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2239 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2240 "packet received on port %s configured with "
2241 "implicit VLAN %"PRIu16,
2242 br->name, ntohs(flow->dl_vlan),
2243 in_port->name, in_port->vlan);
2247 vlan = in_port->vlan;
2249 if (!port_includes_vlan(in_port, vlan)) {
2251 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2252 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2253 "packet received on port %s not configured for "
2255 br->name, vlan, in_port->name, vlan);
2264 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2265 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2266 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2268 is_gratuitous_arp(const flow_t *flow)
2270 return (flow->dl_type == htons(ETH_TYPE_ARP)
2271 && eth_addr_is_broadcast(flow->dl_dst)
2272 && (flow->nw_proto == ARP_OP_REPLY
2273 || (flow->nw_proto == ARP_OP_REQUEST
2274 && flow->nw_src == flow->nw_dst)));
2278 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2279 struct port *in_port)
2281 enum grat_arp_lock_type lock_type;
2284 /* We don't want to learn from gratuitous ARP packets that are reflected
2285 * back over bond slaves so we lock the learning table. */
2286 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2287 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2288 GRAT_ARP_LOCK_CHECK;
2290 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2293 /* The log messages here could actually be useful in debugging,
2294 * so keep the rate limit relatively high. */
2295 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2297 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2298 "on port %s in VLAN %d",
2299 br->name, ETH_ADDR_ARGS(flow->dl_src),
2300 in_port->name, vlan);
2301 ofproto_revalidate(br->ofproto, rev_tag);
2305 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2306 * dropped. Returns true if they may be forwarded, false if they should be
2309 * If 'have_packet' is true, it indicates that the caller is processing a
2310 * received packet. If 'have_packet' is false, then the caller is just
2311 * revalidating an existing flow because configuration has changed. Either
2312 * way, 'have_packet' only affects logging (there is no point in logging errors
2313 * during revalidation).
2315 * Sets '*in_portp' to the input port. This will be a null pointer if
2316 * flow->in_port does not designate a known input port (in which case
2317 * is_admissible() returns false).
2319 * When returning true, sets '*vlanp' to the effective VLAN of the input
2320 * packet, as returned by flow_get_vlan().
2322 * May also add tags to '*tags', although the current implementation only does
2323 * so in one special case.
2326 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2327 tag_type *tags, int *vlanp, struct port **in_portp)
2329 struct iface *in_iface;
2330 struct port *in_port;
2333 /* Find the interface and port structure for the received packet. */
2334 in_iface = iface_from_xf_ifidx(br, ofp_port_to_xflow_port(flow->in_port));
2336 /* No interface? Something fishy... */
2338 /* Odd. A few possible reasons here:
2340 * - We deleted an interface but there are still a few packets
2341 * queued up from it.
2343 * - Someone externally added an interface (e.g. with "ovs-dpctl
2344 * add-if") that we don't know about.
2346 * - Packet arrived on the local port but the local port is not
2347 * one of our bridge ports.
2349 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2351 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2352 "interface %"PRIu16, br->name, flow->in_port);
2358 *in_portp = in_port = in_iface->port;
2359 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2364 /* Drop frames for reserved multicast addresses. */
2365 if (eth_addr_is_reserved(flow->dl_dst)) {
2369 /* Drop frames on ports reserved for mirroring. */
2370 if (in_port->is_mirror_output_port) {
2372 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2373 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2374 "%s, which is reserved exclusively for mirroring",
2375 br->name, in_port->name);
2380 /* Packets received on bonds need special attention to avoid duplicates. */
2381 if (in_port->n_ifaces > 1) {
2383 bool is_grat_arp_locked;
2385 if (eth_addr_is_multicast(flow->dl_dst)) {
2386 *tags |= in_port->active_iface_tag;
2387 if (in_port->active_iface != in_iface->port_ifidx) {
2388 /* Drop all multicast packets on inactive slaves. */
2393 /* Drop all packets for which we have learned a different input
2394 * port, because we probably sent the packet on one slave and got
2395 * it back on the other. Gratuitous ARP packets are an exception
2396 * to this rule: the host has moved to another switch. The exception
2397 * to the exception is if we locked the learning table to avoid
2398 * reflections on bond slaves. If this is the case, just drop the
2400 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2401 &is_grat_arp_locked);
2402 if (src_idx != -1 && src_idx != in_port->port_idx &&
2403 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2411 /* If the composed actions may be applied to any packet in the given 'flow',
2412 * returns true. Otherwise, the actions should only be applied to 'packet', or
2413 * not at all, if 'packet' was NULL. */
2415 process_flow(struct bridge *br, const flow_t *flow,
2416 const struct ofpbuf *packet, struct xflow_actions *actions,
2417 tag_type *tags, uint16_t *nf_output_iface)
2419 struct port *in_port;
2420 struct port *out_port;
2424 /* Check whether we should drop packets in this flow. */
2425 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2430 /* Learn source MAC (but don't try to learn from revalidation). */
2432 update_learning_table(br, flow, vlan, in_port);
2435 /* Determine output port. */
2436 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2438 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2439 out_port = br->ports[out_port_idx];
2440 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2441 /* If we are revalidating but don't have a learning entry then
2442 * eject the flow. Installing a flow that floods packets opens
2443 * up a window of time where we could learn from a packet reflected
2444 * on a bond and blackhole packets before the learning table is
2445 * updated to reflect the correct port. */
2448 out_port = FLOOD_PORT;
2451 /* Don't send packets out their input ports. */
2452 if (in_port == out_port) {
2458 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2465 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2468 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2469 const struct ofp_phy_port *opp,
2472 struct bridge *br = br_;
2473 struct iface *iface;
2476 iface = iface_from_xf_ifidx(br, ofp_port_to_xflow_port(opp->port_no));
2482 if (reason == OFPPR_DELETE) {
2483 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2484 br->name, iface->name);
2485 iface_destroy(iface);
2486 if (!port->n_ifaces) {
2487 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2488 br->name, port->name);
2494 if (port->n_ifaces > 1) {
2495 bool up = !(opp->state & OFPPS_LINK_DOWN);
2496 bond_link_status_update(iface, up);
2497 port_update_bond_compat(port);
2503 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2504 struct xflow_actions *actions, tag_type *tags,
2505 uint16_t *nf_output_iface, void *br_)
2507 struct bridge *br = br_;
2509 COVERAGE_INC(bridge_process_flow);
2511 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2515 bridge_account_flow_ofhook_cb(const flow_t *flow, tag_type tags,
2516 const union xflow_action *actions,
2517 size_t n_actions, unsigned long long int n_bytes,
2520 struct bridge *br = br_;
2521 const union xflow_action *a;
2522 struct port *in_port;
2526 /* Feed information from the active flows back into the learning table to
2527 * ensure that table is always in sync with what is actually flowing
2528 * through the datapath.
2530 * We test that 'tags' is nonzero to ensure that only flows that include an
2531 * OFPP_NORMAL action are used for learning. This works because
2532 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2533 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2534 update_learning_table(br, flow, vlan, in_port);
2537 /* Account for bond slave utilization. */
2538 if (!br->has_bonded_ports) {
2541 for (a = actions; a < &actions[n_actions]; a++) {
2542 if (a->type == XFLOWAT_OUTPUT) {
2543 struct port *out_port = port_from_xf_ifidx(br, a->output.port);
2544 if (out_port && out_port->n_ifaces >= 2) {
2545 struct bond_entry *e = lookup_bond_entry(out_port,
2547 e->tx_bytes += n_bytes;
2554 bridge_account_checkpoint_ofhook_cb(void *br_)
2556 struct bridge *br = br_;
2560 if (!br->has_bonded_ports) {
2565 for (i = 0; i < br->n_ports; i++) {
2566 struct port *port = br->ports[i];
2567 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2568 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2569 bond_rebalance_port(port);
2574 static struct ofhooks bridge_ofhooks = {
2575 bridge_port_changed_ofhook_cb,
2576 bridge_normal_ofhook_cb,
2577 bridge_account_flow_ofhook_cb,
2578 bridge_account_checkpoint_ofhook_cb,
2581 /* Bonding functions. */
2583 /* Statistics for a single interface on a bonded port, used for load-based
2584 * bond rebalancing. */
2585 struct slave_balance {
2586 struct iface *iface; /* The interface. */
2587 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2589 /* All the "bond_entry"s that are assigned to this interface, in order of
2590 * increasing tx_bytes. */
2591 struct bond_entry **hashes;
2595 /* Sorts pointers to pointers to bond_entries in ascending order by the
2596 * interface to which they are assigned, and within a single interface in
2597 * ascending order of bytes transmitted. */
2599 compare_bond_entries(const void *a_, const void *b_)
2601 const struct bond_entry *const *ap = a_;
2602 const struct bond_entry *const *bp = b_;
2603 const struct bond_entry *a = *ap;
2604 const struct bond_entry *b = *bp;
2605 if (a->iface_idx != b->iface_idx) {
2606 return a->iface_idx > b->iface_idx ? 1 : -1;
2607 } else if (a->tx_bytes != b->tx_bytes) {
2608 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2614 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2615 * *descending* order by number of bytes transmitted. */
2617 compare_slave_balance(const void *a_, const void *b_)
2619 const struct slave_balance *a = a_;
2620 const struct slave_balance *b = b_;
2621 if (a->iface->enabled != b->iface->enabled) {
2622 return a->iface->enabled ? -1 : 1;
2623 } else if (a->tx_bytes != b->tx_bytes) {
2624 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2631 swap_bals(struct slave_balance *a, struct slave_balance *b)
2633 struct slave_balance tmp = *a;
2638 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2639 * given that 'p' (and only 'p') might be in the wrong location.
2641 * This function invalidates 'p', since it might now be in a different memory
2644 resort_bals(struct slave_balance *p,
2645 struct slave_balance bals[], size_t n_bals)
2648 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2649 swap_bals(p, p - 1);
2651 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2652 swap_bals(p, p + 1);
2658 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2660 if (VLOG_IS_DBG_ENABLED()) {
2661 struct ds ds = DS_EMPTY_INITIALIZER;
2662 const struct slave_balance *b;
2664 for (b = bals; b < bals + n_bals; b++) {
2668 ds_put_char(&ds, ',');
2670 ds_put_format(&ds, " %s %"PRIu64"kB",
2671 b->iface->name, b->tx_bytes / 1024);
2673 if (!b->iface->enabled) {
2674 ds_put_cstr(&ds, " (disabled)");
2676 if (b->n_hashes > 0) {
2677 ds_put_cstr(&ds, " (");
2678 for (i = 0; i < b->n_hashes; i++) {
2679 const struct bond_entry *e = b->hashes[i];
2681 ds_put_cstr(&ds, " + ");
2683 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2684 e - port->bond_hash, e->tx_bytes / 1024);
2686 ds_put_cstr(&ds, ")");
2689 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2694 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2696 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2699 struct bond_entry *hash = from->hashes[hash_idx];
2700 struct port *port = from->iface->port;
2701 uint64_t delta = hash->tx_bytes;
2703 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2704 "from %s to %s (now carrying %"PRIu64"kB and "
2705 "%"PRIu64"kB load, respectively)",
2706 port->name, delta / 1024, hash - port->bond_hash,
2707 from->iface->name, to->iface->name,
2708 (from->tx_bytes - delta) / 1024,
2709 (to->tx_bytes + delta) / 1024);
2711 /* Delete element from from->hashes.
2713 * We don't bother to add the element to to->hashes because not only would
2714 * it require more work, the only purpose it would be to allow that hash to
2715 * be migrated to another slave in this rebalancing run, and there is no
2716 * point in doing that. */
2717 if (hash_idx == 0) {
2720 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2721 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2725 /* Shift load away from 'from' to 'to'. */
2726 from->tx_bytes -= delta;
2727 to->tx_bytes += delta;
2729 /* Arrange for flows to be revalidated. */
2730 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2731 hash->iface_idx = to->iface->port_ifidx;
2732 hash->iface_tag = tag_create_random();
2736 bond_rebalance_port(struct port *port)
2738 struct slave_balance bals[DP_MAX_PORTS];
2740 struct bond_entry *hashes[BOND_MASK + 1];
2741 struct slave_balance *b, *from, *to;
2742 struct bond_entry *e;
2745 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2746 * descending order of tx_bytes, so that bals[0] represents the most
2747 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2750 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2751 * array for each slave_balance structure, we sort our local array of
2752 * hashes in order by slave, so that all of the hashes for a given slave
2753 * become contiguous in memory, and then we point each 'hashes' members of
2754 * a slave_balance structure to the start of a contiguous group. */
2755 n_bals = port->n_ifaces;
2756 for (b = bals; b < &bals[n_bals]; b++) {
2757 b->iface = port->ifaces[b - bals];
2762 for (i = 0; i <= BOND_MASK; i++) {
2763 hashes[i] = &port->bond_hash[i];
2765 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2766 for (i = 0; i <= BOND_MASK; i++) {
2768 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2769 b = &bals[e->iface_idx];
2770 b->tx_bytes += e->tx_bytes;
2772 b->hashes = &hashes[i];
2777 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2778 log_bals(bals, n_bals, port);
2780 /* Discard slaves that aren't enabled (which were sorted to the back of the
2781 * array earlier). */
2782 while (!bals[n_bals - 1].iface->enabled) {
2789 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2790 to = &bals[n_bals - 1];
2791 for (from = bals; from < to; ) {
2792 uint64_t overload = from->tx_bytes - to->tx_bytes;
2793 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2794 /* The extra load on 'from' (and all less-loaded slaves), compared
2795 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2796 * it is less than ~1Mbps. No point in rebalancing. */
2798 } else if (from->n_hashes == 1) {
2799 /* 'from' only carries a single MAC hash, so we can't shift any
2800 * load away from it, even though we want to. */
2803 /* 'from' is carrying significantly more load than 'to', and that
2804 * load is split across at least two different hashes. Pick a hash
2805 * to migrate to 'to' (the least-loaded slave), given that doing so
2806 * must decrease the ratio of the load on the two slaves by at
2809 * The sort order we use means that we prefer to shift away the
2810 * smallest hashes instead of the biggest ones. There is little
2811 * reason behind this decision; we could use the opposite sort
2812 * order to shift away big hashes ahead of small ones. */
2816 for (i = 0; i < from->n_hashes; i++) {
2817 double old_ratio, new_ratio;
2818 uint64_t delta = from->hashes[i]->tx_bytes;
2820 if (delta == 0 || from->tx_bytes - delta == 0) {
2821 /* Pointless move. */
2825 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2827 if (to->tx_bytes == 0) {
2828 /* Nothing on the new slave, move it. */
2832 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2833 new_ratio = (double)(from->tx_bytes - delta) /
2834 (to->tx_bytes + delta);
2836 if (new_ratio == 0) {
2837 /* Should already be covered but check to prevent division
2842 if (new_ratio < 1) {
2843 new_ratio = 1 / new_ratio;
2846 if (old_ratio - new_ratio > 0.1) {
2847 /* Would decrease the ratio, move it. */
2851 if (i < from->n_hashes) {
2852 bond_shift_load(from, to, i);
2853 port->bond_compat_is_stale = true;
2855 /* If the result of the migration changed the relative order of
2856 * 'from' and 'to' swap them back to maintain invariants. */
2857 if (order_swapped) {
2858 swap_bals(from, to);
2861 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2862 * point to different slave_balance structures. It is only
2863 * valid to do these two operations in a row at all because we
2864 * know that 'from' will not move past 'to' and vice versa. */
2865 resort_bals(from, bals, n_bals);
2866 resort_bals(to, bals, n_bals);
2873 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2874 * historical data to decay to <1% in 7 rebalancing runs. */
2875 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2881 bond_send_learning_packets(struct port *port)
2883 struct bridge *br = port->bridge;
2884 struct mac_entry *e;
2885 struct ofpbuf packet;
2886 int error, n_packets, n_errors;
2888 if (!port->n_ifaces || port->active_iface < 0) {
2892 ofpbuf_init(&packet, 128);
2893 error = n_packets = n_errors = 0;
2894 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2895 union ofp_action actions[2], *a;
2901 if (e->port == port->port_idx
2902 || !choose_output_iface(port, e->mac, &xf_ifidx, &tags)) {
2906 /* Compose actions. */
2907 memset(actions, 0, sizeof actions);
2910 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2911 a->vlan_vid.len = htons(sizeof *a);
2912 a->vlan_vid.vlan_vid = htons(e->vlan);
2915 a->output.type = htons(OFPAT_OUTPUT);
2916 a->output.len = htons(sizeof *a);
2917 a->output.port = htons(xflow_port_to_ofp_port(xf_ifidx));
2922 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2924 flow_extract(&packet, 0, OFPP_NONE, &flow);
2925 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2932 ofpbuf_uninit(&packet);
2935 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2936 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2937 "packets, last error was: %s",
2938 port->name, n_errors, n_packets, strerror(error));
2940 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2941 port->name, n_packets);
2945 /* Bonding unixctl user interface functions. */
2948 bond_unixctl_list(struct unixctl_conn *conn,
2949 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2951 struct ds ds = DS_EMPTY_INITIALIZER;
2952 const struct bridge *br;
2954 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2956 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2959 for (i = 0; i < br->n_ports; i++) {
2960 const struct port *port = br->ports[i];
2961 if (port->n_ifaces > 1) {
2964 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2965 for (j = 0; j < port->n_ifaces; j++) {
2966 const struct iface *iface = port->ifaces[j];
2968 ds_put_cstr(&ds, ", ");
2970 ds_put_cstr(&ds, iface->name);
2972 ds_put_char(&ds, '\n');
2976 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2980 static struct port *
2981 bond_find(const char *name)
2983 const struct bridge *br;
2985 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2988 for (i = 0; i < br->n_ports; i++) {
2989 struct port *port = br->ports[i];
2990 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2999 bond_unixctl_show(struct unixctl_conn *conn,
3000 const char *args, void *aux OVS_UNUSED)
3002 struct ds ds = DS_EMPTY_INITIALIZER;
3003 const struct port *port;
3006 port = bond_find(args);
3008 unixctl_command_reply(conn, 501, "no such bond");
3012 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3013 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3014 ds_put_format(&ds, "next rebalance: %lld ms\n",
3015 port->bond_next_rebalance - time_msec());
3016 for (j = 0; j < port->n_ifaces; j++) {
3017 const struct iface *iface = port->ifaces[j];
3018 struct bond_entry *be;
3021 ds_put_format(&ds, "slave %s: %s\n",
3022 iface->name, iface->enabled ? "enabled" : "disabled");
3023 if (j == port->active_iface) {
3024 ds_put_cstr(&ds, "\tactive slave\n");
3026 if (iface->delay_expires != LLONG_MAX) {
3027 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3028 iface->enabled ? "downdelay" : "updelay",
3029 iface->delay_expires - time_msec());
3033 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3034 int hash = be - port->bond_hash;
3035 struct mac_entry *me;
3037 if (be->iface_idx != j) {
3041 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3042 hash, be->tx_bytes / 1024);
3045 LIST_FOR_EACH (me, struct mac_entry, lru_node,
3046 &port->bridge->ml->lrus) {
3049 if (bond_hash(me->mac) == hash
3050 && me->port != port->port_idx
3051 && choose_output_iface(port, me->mac, &xf_ifidx, &tags)
3052 && xf_ifidx == iface->xf_ifidx)
3054 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3055 ETH_ADDR_ARGS(me->mac));
3060 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3065 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3066 void *aux OVS_UNUSED)
3068 char *args = (char *) args_;
3069 char *save_ptr = NULL;
3070 char *bond_s, *hash_s, *slave_s;
3071 uint8_t mac[ETH_ADDR_LEN];
3073 struct iface *iface;
3074 struct bond_entry *entry;
3077 bond_s = strtok_r(args, " ", &save_ptr);
3078 hash_s = strtok_r(NULL, " ", &save_ptr);
3079 slave_s = strtok_r(NULL, " ", &save_ptr);
3081 unixctl_command_reply(conn, 501,
3082 "usage: bond/migrate BOND HASH SLAVE");
3086 port = bond_find(bond_s);
3088 unixctl_command_reply(conn, 501, "no such bond");
3092 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3093 == ETH_ADDR_SCAN_COUNT) {
3094 hash = bond_hash(mac);
3095 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3096 hash = atoi(hash_s) & BOND_MASK;
3098 unixctl_command_reply(conn, 501, "bad hash");
3102 iface = port_lookup_iface(port, slave_s);
3104 unixctl_command_reply(conn, 501, "no such slave");
3108 if (!iface->enabled) {
3109 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3113 entry = &port->bond_hash[hash];
3114 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3115 entry->iface_idx = iface->port_ifidx;
3116 entry->iface_tag = tag_create_random();
3117 port->bond_compat_is_stale = true;
3118 unixctl_command_reply(conn, 200, "migrated");
3122 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3123 void *aux OVS_UNUSED)
3125 char *args = (char *) args_;
3126 char *save_ptr = NULL;
3127 char *bond_s, *slave_s;
3129 struct iface *iface;
3131 bond_s = strtok_r(args, " ", &save_ptr);
3132 slave_s = strtok_r(NULL, " ", &save_ptr);
3134 unixctl_command_reply(conn, 501,
3135 "usage: bond/set-active-slave BOND SLAVE");
3139 port = bond_find(bond_s);
3141 unixctl_command_reply(conn, 501, "no such bond");
3145 iface = port_lookup_iface(port, slave_s);
3147 unixctl_command_reply(conn, 501, "no such slave");
3151 if (!iface->enabled) {
3152 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3156 if (port->active_iface != iface->port_ifidx) {
3157 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3158 port->active_iface = iface->port_ifidx;
3159 port->active_iface_tag = tag_create_random();
3160 VLOG_INFO("port %s: active interface is now %s",
3161 port->name, iface->name);
3162 bond_send_learning_packets(port);
3163 unixctl_command_reply(conn, 200, "done");
3165 unixctl_command_reply(conn, 200, "no change");
3170 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3172 char *args = (char *) args_;
3173 char *save_ptr = NULL;
3174 char *bond_s, *slave_s;
3176 struct iface *iface;
3178 bond_s = strtok_r(args, " ", &save_ptr);
3179 slave_s = strtok_r(NULL, " ", &save_ptr);
3181 unixctl_command_reply(conn, 501,
3182 "usage: bond/enable/disable-slave BOND SLAVE");
3186 port = bond_find(bond_s);
3188 unixctl_command_reply(conn, 501, "no such bond");
3192 iface = port_lookup_iface(port, slave_s);
3194 unixctl_command_reply(conn, 501, "no such slave");
3198 bond_enable_slave(iface, enable);
3199 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3203 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3204 void *aux OVS_UNUSED)
3206 enable_slave(conn, args, true);
3210 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3211 void *aux OVS_UNUSED)
3213 enable_slave(conn, args, false);
3217 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3218 void *aux OVS_UNUSED)
3220 uint8_t mac[ETH_ADDR_LEN];
3224 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3225 == ETH_ADDR_SCAN_COUNT) {
3226 hash = bond_hash(mac);
3228 hash_cstr = xasprintf("%u", hash);
3229 unixctl_command_reply(conn, 200, hash_cstr);
3232 unixctl_command_reply(conn, 501, "invalid mac");
3239 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3240 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3241 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3242 unixctl_command_register("bond/set-active-slave",
3243 bond_unixctl_set_active_slave, NULL);
3244 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3246 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3248 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3251 /* Port functions. */
3253 static struct port *
3254 port_create(struct bridge *br, const char *name)
3258 port = xzalloc(sizeof *port);
3260 port->port_idx = br->n_ports;
3262 port->trunks = NULL;
3263 port->name = xstrdup(name);
3264 port->active_iface = -1;
3266 if (br->n_ports >= br->allocated_ports) {
3267 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3270 br->ports[br->n_ports++] = port;
3271 shash_add_assert(&br->port_by_name, port->name, port);
3273 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3280 get_port_other_config(const struct ovsrec_port *port, const char *key,
3281 const char *default_value)
3285 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3287 return value ? value : default_value;
3291 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3293 struct shash new_ifaces;
3296 /* Collect list of new interfaces. */
3297 shash_init(&new_ifaces);
3298 for (i = 0; i < cfg->n_interfaces; i++) {
3299 const char *name = cfg->interfaces[i]->name;
3300 shash_add_once(&new_ifaces, name, NULL);
3303 /* Get rid of deleted interfaces. */
3304 for (i = 0; i < port->n_ifaces; ) {
3305 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3306 iface_destroy(port->ifaces[i]);
3312 shash_destroy(&new_ifaces);
3316 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3318 struct shash new_ifaces;
3319 long long int next_rebalance;
3320 unsigned long *trunks;
3326 /* Update settings. */
3327 port->updelay = cfg->bond_updelay;
3328 if (port->updelay < 0) {
3331 port->downdelay = cfg->bond_downdelay;
3332 if (port->downdelay < 0) {
3333 port->downdelay = 0;
3335 port->bond_rebalance_interval = atoi(
3336 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3337 if (port->bond_rebalance_interval < 1000) {
3338 port->bond_rebalance_interval = 1000;
3340 next_rebalance = time_msec() + port->bond_rebalance_interval;
3341 if (port->bond_next_rebalance > next_rebalance) {
3342 port->bond_next_rebalance = next_rebalance;
3345 /* Add new interfaces and update 'cfg' member of existing ones. */
3346 shash_init(&new_ifaces);
3347 for (i = 0; i < cfg->n_interfaces; i++) {
3348 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3349 struct iface *iface;
3351 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3352 VLOG_WARN("port %s: %s specified twice as port interface",
3353 port->name, if_cfg->name);
3357 iface = iface_lookup(port->bridge, if_cfg->name);
3359 if (iface->port != port) {
3360 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3362 port->bridge->name, if_cfg->name, iface->port->name);
3365 iface->cfg = if_cfg;
3367 iface_create(port, if_cfg);
3370 shash_destroy(&new_ifaces);
3375 if (port->n_ifaces < 2) {
3377 if (vlan >= 0 && vlan <= 4095) {
3378 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3383 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3384 * they even work as-is. But they have not been tested. */
3385 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3389 if (port->vlan != vlan) {
3391 bridge_flush(port->bridge);
3394 /* Get trunked VLANs. */
3396 if (vlan < 0 && cfg->n_trunks) {
3400 trunks = bitmap_allocate(4096);
3402 for (i = 0; i < cfg->n_trunks; i++) {
3403 int trunk = cfg->trunks[i];
3405 bitmap_set1(trunks, trunk);
3411 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3412 port->name, cfg->n_trunks);
3414 if (n_errors == cfg->n_trunks) {
3415 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3417 bitmap_free(trunks);
3420 } else if (vlan >= 0 && cfg->n_trunks) {
3421 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3425 ? port->trunks != NULL
3426 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3427 bridge_flush(port->bridge);
3429 bitmap_free(port->trunks);
3430 port->trunks = trunks;
3434 port_destroy(struct port *port)
3437 struct bridge *br = port->bridge;
3441 proc_net_compat_update_vlan(port->name, NULL, 0);
3442 proc_net_compat_update_bond(port->name, NULL);
3444 for (i = 0; i < MAX_MIRRORS; i++) {
3445 struct mirror *m = br->mirrors[i];
3446 if (m && m->out_port == port) {
3451 while (port->n_ifaces > 0) {
3452 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3455 shash_find_and_delete_assert(&br->port_by_name, port->name);
3457 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3458 del->port_idx = port->port_idx;
3461 bitmap_free(port->trunks);
3468 static struct port *
3469 port_from_xf_ifidx(const struct bridge *br, uint16_t xf_ifidx)
3471 struct iface *iface = iface_from_xf_ifidx(br, xf_ifidx);
3472 return iface ? iface->port : NULL;
3475 static struct port *
3476 port_lookup(const struct bridge *br, const char *name)
3478 return shash_find_data(&br->port_by_name, name);
3481 static struct iface *
3482 port_lookup_iface(const struct port *port, const char *name)
3484 struct iface *iface = iface_lookup(port->bridge, name);
3485 return iface && iface->port == port ? iface : NULL;
3489 port_update_bonding(struct port *port)
3491 if (port->n_ifaces < 2) {
3492 /* Not a bonded port. */
3493 if (port->bond_hash) {
3494 free(port->bond_hash);
3495 port->bond_hash = NULL;
3496 port->bond_compat_is_stale = true;
3497 port->bond_fake_iface = false;
3500 if (!port->bond_hash) {
3503 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3504 for (i = 0; i <= BOND_MASK; i++) {
3505 struct bond_entry *e = &port->bond_hash[i];
3509 port->no_ifaces_tag = tag_create_random();
3510 bond_choose_active_iface(port);
3511 port->bond_next_rebalance
3512 = time_msec() + port->bond_rebalance_interval;
3514 if (port->cfg->bond_fake_iface) {
3515 port->bond_next_fake_iface_update = time_msec();
3518 port->bond_compat_is_stale = true;
3519 port->bond_fake_iface = port->cfg->bond_fake_iface;
3524 port_update_bond_compat(struct port *port)
3526 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3527 struct compat_bond bond;
3530 if (port->n_ifaces < 2) {
3531 proc_net_compat_update_bond(port->name, NULL);
3536 bond.updelay = port->updelay;
3537 bond.downdelay = port->downdelay;
3540 bond.hashes = compat_hashes;
3541 if (port->bond_hash) {
3542 const struct bond_entry *e;
3543 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3544 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3545 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3546 cbh->hash = e - port->bond_hash;
3547 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3552 bond.n_slaves = port->n_ifaces;
3553 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3554 for (i = 0; i < port->n_ifaces; i++) {
3555 struct iface *iface = port->ifaces[i];
3556 struct compat_bond_slave *slave = &bond.slaves[i];
3557 slave->name = iface->name;
3559 /* We need to make the same determination as the Linux bonding
3560 * code to determine whether a slave should be consider "up".
3561 * The Linux function bond_miimon_inspect() supports four
3562 * BOND_LINK_* states:
3564 * - BOND_LINK_UP: carrier detected, updelay has passed.
3565 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3566 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3567 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3569 * The function bond_info_show_slave() only considers BOND_LINK_UP
3570 * to be "up" and anything else to be "down".
3572 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3576 netdev_get_etheraddr(iface->netdev, slave->mac);
3579 if (port->bond_fake_iface) {
3580 struct netdev *bond_netdev;
3582 if (!netdev_open_default(port->name, &bond_netdev)) {
3584 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3586 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3588 netdev_close(bond_netdev);
3592 proc_net_compat_update_bond(port->name, &bond);
3597 port_update_vlan_compat(struct port *port)
3599 struct bridge *br = port->bridge;
3600 char *vlandev_name = NULL;
3602 if (port->vlan > 0) {
3603 /* Figure out the name that the VLAN device should actually have, if it
3604 * existed. This takes some work because the VLAN device would not
3605 * have port->name in its name; rather, it would have the trunk port's
3606 * name, and 'port' would be attached to a bridge that also had the
3607 * VLAN device one of its ports. So we need to find a trunk port that
3608 * includes port->vlan.
3610 * There might be more than one candidate. This doesn't happen on
3611 * XenServer, so if it happens we just pick the first choice in
3612 * alphabetical order instead of creating multiple VLAN devices. */
3614 for (i = 0; i < br->n_ports; i++) {
3615 struct port *p = br->ports[i];
3616 if (port_trunks_vlan(p, port->vlan)
3618 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3620 uint8_t ea[ETH_ADDR_LEN];
3621 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3622 if (!eth_addr_is_multicast(ea) &&
3623 !eth_addr_is_reserved(ea) &&
3624 !eth_addr_is_zero(ea)) {
3625 vlandev_name = p->name;
3630 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3633 /* Interface functions. */
3635 static struct iface *
3636 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3638 struct bridge *br = port->bridge;
3639 struct iface *iface;
3640 char *name = if_cfg->name;
3643 iface = xzalloc(sizeof *iface);
3645 iface->port_ifidx = port->n_ifaces;
3646 iface->name = xstrdup(name);
3647 iface->xf_ifidx = -1;
3648 iface->tag = tag_create_random();
3649 iface->delay_expires = LLONG_MAX;
3650 iface->netdev = NULL;
3651 iface->cfg = if_cfg;
3653 shash_add_assert(&br->iface_by_name, iface->name, iface);
3655 /* Attempt to create the network interface in case it doesn't exist yet. */
3656 if (!iface_is_internal(br, iface->name)) {
3657 error = set_up_iface(if_cfg, iface, true);
3659 VLOG_WARN("could not create iface %s: %s", iface->name,
3662 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3669 if (port->n_ifaces >= port->allocated_ifaces) {
3670 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3671 sizeof *port->ifaces);
3673 port->ifaces[port->n_ifaces++] = iface;
3674 if (port->n_ifaces > 1) {
3675 br->has_bonded_ports = true;
3678 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3686 iface_destroy(struct iface *iface)
3689 struct port *port = iface->port;
3690 struct bridge *br = port->bridge;
3691 bool del_active = port->active_iface == iface->port_ifidx;
3694 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3696 if (iface->xf_ifidx >= 0) {
3697 port_array_set(&br->ifaces, iface->xf_ifidx, NULL);
3700 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3701 del->port_ifidx = iface->port_ifidx;
3703 netdev_close(iface->netdev);
3706 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3707 bond_choose_active_iface(port);
3708 bond_send_learning_packets(port);
3714 bridge_flush(port->bridge);
3718 static struct iface *
3719 iface_lookup(const struct bridge *br, const char *name)
3721 return shash_find_data(&br->iface_by_name, name);
3724 static struct iface *
3725 iface_from_xf_ifidx(const struct bridge *br, uint16_t xf_ifidx)
3727 return port_array_get(&br->ifaces, xf_ifidx);
3730 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3731 * 'br', that is, an interface that is entirely simulated within the datapath.
3732 * The local port (XFLOWP_LOCAL) is always an internal interface. Other local
3733 * interfaces are created by setting "iface.<iface>.internal = true".
3735 * In addition, we have a kluge-y feature that creates an internal port with
3736 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3737 * This feature needs to go away in the long term. Until then, this is one
3738 * reason why this function takes a name instead of a struct iface: the fake
3739 * interfaces created this way do not have a struct iface. */
3741 iface_is_internal(const struct bridge *br, const char *if_name)
3743 struct iface *iface;
3746 if (!strcmp(if_name, br->name)) {
3750 iface = iface_lookup(br, if_name);
3751 if (iface && !strcmp(iface->cfg->type, "internal")) {
3755 port = port_lookup(br, if_name);
3756 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3762 /* Set Ethernet address of 'iface', if one is specified in the configuration
3765 iface_set_mac(struct iface *iface)
3767 uint8_t ea[ETH_ADDR_LEN];
3769 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3770 if (eth_addr_is_multicast(ea)) {
3771 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3773 } else if (iface->xf_ifidx == XFLOWP_LOCAL) {
3774 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3775 iface->name, iface->name);
3777 int error = netdev_set_etheraddr(iface->netdev, ea);
3779 VLOG_ERR("interface %s: setting MAC failed (%s)",
3780 iface->name, strerror(error));
3787 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3788 struct shash *shash)
3793 for (i = 0; i < n; i++) {
3794 shash_add(shash, keys[i], values[i]);
3798 struct iface_delete_queues_cbdata {
3799 struct netdev *netdev;
3800 const struct ovsdb_datum *queues;
3804 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
3806 union ovsdb_atom atom;
3808 atom.integer = target;
3809 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
3813 iface_delete_queues(unsigned int queue_id,
3814 const struct shash *details OVS_UNUSED, void *cbdata_)
3816 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3818 if (!queue_ids_include(cbdata->queues, queue_id)) {
3819 netdev_delete_queue(cbdata->netdev, queue_id);
3824 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3826 if (!qos || qos->type[0] == '\0') {
3827 netdev_set_qos(iface->netdev, NULL, NULL);
3829 struct iface_delete_queues_cbdata cbdata;
3830 struct shash details;
3833 /* Configure top-level Qos for 'iface'. */
3834 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3835 qos->n_other_config, &details);
3836 netdev_set_qos(iface->netdev, qos->type, &details);
3837 shash_destroy(&details);
3839 /* Deconfigure queues that were deleted. */
3840 cbdata.netdev = iface->netdev;
3841 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
3843 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3845 /* Configure queues for 'iface'. */
3846 for (i = 0; i < qos->n_queues; i++) {
3847 const struct ovsrec_queue *queue = qos->value_queues[i];
3848 unsigned int queue_id = qos->key_queues[i];
3850 shash_from_ovs_idl_map(queue->key_other_config,
3851 queue->value_other_config,
3852 queue->n_other_config, &details);
3853 netdev_set_queue(iface->netdev, queue_id, &details);
3854 shash_destroy(&details);
3859 /* Port mirroring. */
3861 static struct mirror *
3862 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
3866 for (i = 0; i < MAX_MIRRORS; i++) {
3867 struct mirror *m = br->mirrors[i];
3868 if (m && uuid_equals(uuid, &m->uuid)) {
3876 mirror_reconfigure(struct bridge *br)
3878 unsigned long *rspan_vlans;
3881 /* Get rid of deleted mirrors. */
3882 for (i = 0; i < MAX_MIRRORS; i++) {
3883 struct mirror *m = br->mirrors[i];
3885 const struct ovsdb_datum *mc;
3886 union ovsdb_atom atom;
3888 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
3889 atom.uuid = br->mirrors[i]->uuid;
3890 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
3896 /* Add new mirrors and reconfigure existing ones. */
3897 for (i = 0; i < br->cfg->n_mirrors; i++) {
3898 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3899 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
3901 mirror_reconfigure_one(m, cfg);
3903 mirror_create(br, cfg);
3907 /* Update port reserved status. */
3908 for (i = 0; i < br->n_ports; i++) {
3909 br->ports[i]->is_mirror_output_port = false;
3911 for (i = 0; i < MAX_MIRRORS; i++) {
3912 struct mirror *m = br->mirrors[i];
3913 if (m && m->out_port) {
3914 m->out_port->is_mirror_output_port = true;
3918 /* Update flooded vlans (for RSPAN). */
3920 if (br->cfg->n_flood_vlans) {
3921 rspan_vlans = bitmap_allocate(4096);
3923 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3924 int64_t vlan = br->cfg->flood_vlans[i];
3925 if (vlan >= 0 && vlan < 4096) {
3926 bitmap_set1(rspan_vlans, vlan);
3927 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3930 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3935 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3941 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
3946 for (i = 0; ; i++) {
3947 if (i >= MAX_MIRRORS) {
3948 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3949 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
3952 if (!br->mirrors[i]) {
3957 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
3960 br->mirrors[i] = m = xzalloc(sizeof *m);
3963 m->name = xstrdup(cfg->name);
3964 shash_init(&m->src_ports);
3965 shash_init(&m->dst_ports);
3971 mirror_reconfigure_one(m, cfg);
3975 mirror_destroy(struct mirror *m)
3978 struct bridge *br = m->bridge;
3981 for (i = 0; i < br->n_ports; i++) {
3982 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3983 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3986 shash_destroy(&m->src_ports);
3987 shash_destroy(&m->dst_ports);
3990 m->bridge->mirrors[m->idx] = NULL;
3999 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4000 struct shash *names)
4004 for (i = 0; i < n_ports; i++) {
4005 const char *name = ports[i]->name;
4006 if (port_lookup(m->bridge, name)) {
4007 shash_add_once(names, name, NULL);
4009 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4010 "port %s", m->bridge->name, m->name, name);
4016 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4022 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4024 for (i = 0; i < cfg->n_select_vlan; i++) {
4025 int64_t vlan = cfg->select_vlan[i];
4026 if (vlan < 0 || vlan > 4095) {
4027 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4028 m->bridge->name, m->name, vlan);
4030 (*vlans)[n_vlans++] = vlan;
4037 vlan_is_mirrored(const struct mirror *m, int vlan)
4041 for (i = 0; i < m->n_vlans; i++) {
4042 if (m->vlans[i] == vlan) {
4050 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4054 for (i = 0; i < m->n_vlans; i++) {
4055 if (port_trunks_vlan(p, m->vlans[i])) {
4063 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4065 struct shash src_ports, dst_ports;
4066 mirror_mask_t mirror_bit;
4067 struct port *out_port;
4074 if (strcmp(cfg->name, m->name)) {
4076 m->name = xstrdup(cfg->name);
4079 /* Get output port. */
4080 if (cfg->output_port) {
4081 out_port = port_lookup(m->bridge, cfg->output_port->name);
4083 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4084 m->bridge->name, m->name);
4090 if (cfg->output_vlan) {
4091 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4092 "output vlan; ignoring output vlan",
4093 m->bridge->name, m->name);
4095 } else if (cfg->output_vlan) {
4097 out_vlan = *cfg->output_vlan;
4099 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4100 m->bridge->name, m->name);
4105 shash_init(&src_ports);
4106 shash_init(&dst_ports);
4107 if (cfg->select_all) {
4108 for (i = 0; i < m->bridge->n_ports; i++) {
4109 const char *name = m->bridge->ports[i]->name;
4110 shash_add_once(&src_ports, name, NULL);
4111 shash_add_once(&dst_ports, name, NULL);
4116 /* Get ports, and drop duplicates and ports that don't exist. */
4117 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4119 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4122 /* Get all the vlans, and drop duplicate and invalid vlans. */
4123 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4126 /* Update mirror data. */
4127 if (!shash_equal_keys(&m->src_ports, &src_ports)
4128 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4129 || m->n_vlans != n_vlans
4130 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4131 || m->out_port != out_port
4132 || m->out_vlan != out_vlan) {
4133 bridge_flush(m->bridge);
4135 shash_swap(&m->src_ports, &src_ports);
4136 shash_swap(&m->dst_ports, &dst_ports);
4139 m->n_vlans = n_vlans;
4140 m->out_port = out_port;
4141 m->out_vlan = out_vlan;
4144 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4145 for (i = 0; i < m->bridge->n_ports; i++) {
4146 struct port *port = m->bridge->ports[i];
4148 if (shash_find(&m->src_ports, port->name)
4151 ? port_trunks_any_mirrored_vlan(m, port)
4152 : vlan_is_mirrored(m, port->vlan)))) {
4153 port->src_mirrors |= mirror_bit;
4155 port->src_mirrors &= ~mirror_bit;
4158 if (shash_find(&m->dst_ports, port->name)) {
4159 port->dst_mirrors |= mirror_bit;
4161 port->dst_mirrors &= ~mirror_bit;
4166 shash_destroy(&src_ports);
4167 shash_destroy(&dst_ports);