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 bond_next_fake_iface_update; /* Next update to fake bond stats. */
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 /* Register unixctl commands. */
272 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
273 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
275 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
280 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
281 * but for which the ovs-vswitchd configuration 'cfg' is required. */
283 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
285 static bool already_configured_once;
286 struct svec bridge_names;
287 struct svec xfif_names, xfif_types;
290 /* Only do this once per ovs-vswitchd run. */
291 if (already_configured_once) {
294 already_configured_once = true;
296 iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
298 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
299 svec_init(&bridge_names);
300 for (i = 0; i < cfg->n_bridges; i++) {
301 svec_add(&bridge_names, cfg->bridges[i]->name);
303 svec_sort(&bridge_names);
305 /* Iterate over all system xfifs and delete any of them that do not appear
307 svec_init(&xfif_names);
308 svec_init(&xfif_types);
309 xf_enumerate_types(&xfif_types);
310 for (i = 0; i < xfif_types.n; i++) {
315 xf_enumerate_names(xfif_types.names[i], &xfif_names);
317 /* For each xfif... */
318 for (j = 0; j < xfif_names.n; j++) {
319 retval = xfif_open(xfif_names.names[j], xfif_types.names[i], &xfif);
321 struct svec all_names;
324 /* ...check whether any of its names is in 'bridge_names'. */
325 svec_init(&all_names);
326 xfif_get_all_names(xfif, &all_names);
327 for (k = 0; k < all_names.n; k++) {
328 if (svec_contains(&bridge_names, all_names.names[k])) {
333 /* No. Delete the xfif. */
337 svec_destroy(&all_names);
342 svec_destroy(&bridge_names);
343 svec_destroy(&xfif_names);
344 svec_destroy(&xfif_types);
347 /* Attempt to create the network device 'iface_name' through the netdev
350 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
353 struct shash options;
357 shash_init(&options);
358 for (i = 0; i < iface_cfg->n_options; i++) {
359 shash_add(&options, iface_cfg->key_options[i],
360 xstrdup(iface_cfg->value_options[i]));
364 struct netdev_options netdev_options;
366 memset(&netdev_options, 0, sizeof netdev_options);
367 netdev_options.name = iface_cfg->name;
368 if (!strcmp(iface_cfg->type, "internal")) {
369 /* An "internal" config type maps to a netdev "system" type. */
370 netdev_options.type = "system";
372 netdev_options.type = iface_cfg->type;
374 netdev_options.args = &options;
375 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
376 error = netdev_open(&netdev_options, &iface->netdev);
379 netdev_get_carrier(iface->netdev, &iface->enabled);
381 } else if (iface->netdev) {
382 const char *netdev_type = netdev_get_type(iface->netdev);
383 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
384 ? iface_cfg->type : NULL;
386 /* An "internal" config type maps to a netdev "system" type. */
387 if (iface_type && !strcmp(iface_type, "internal")) {
388 iface_type = "system";
391 if (!iface_type || !strcmp(netdev_type, iface_type)) {
392 error = netdev_reconfigure(iface->netdev, &options);
394 VLOG_WARN("%s: attempting change device type from %s to %s",
395 iface_cfg->name, netdev_type, iface_type);
399 shash_destroy_free_data(&options);
405 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
407 return set_up_iface(iface_cfg, iface, false);
411 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
412 void *aux OVS_UNUSED)
414 if (!iface->netdev) {
415 int error = set_up_iface(iface->cfg, iface, true);
417 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
427 check_iface_xf_ifidx(struct bridge *br, struct iface *iface,
428 void *aux OVS_UNUSED)
430 if (iface->xf_ifidx >= 0) {
431 VLOG_DBG("%s has interface %s on port %d",
433 iface->name, iface->xf_ifidx);
436 VLOG_ERR("%s interface not in %s, dropping",
437 iface->name, xfif_name(br->xfif));
443 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
444 void *aux OVS_UNUSED)
446 /* Set policing attributes. */
447 netdev_set_policing(iface->netdev,
448 iface->cfg->ingress_policing_rate,
449 iface->cfg->ingress_policing_burst);
451 /* Set MAC address of internal interfaces other than the local
453 if (iface->xf_ifidx != XFLOWP_LOCAL
454 && iface_is_internal(br, iface->name)) {
455 iface_set_mac(iface);
461 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
462 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
463 * deletes from 'br' any ports that no longer have any interfaces. */
465 iterate_and_prune_ifaces(struct bridge *br,
466 bool (*cb)(struct bridge *, struct iface *,
472 for (i = 0; i < br->n_ports; ) {
473 struct port *port = br->ports[i];
474 for (j = 0; j < port->n_ifaces; ) {
475 struct iface *iface = port->ifaces[j];
476 if (cb(br, iface, aux)) {
479 iface_destroy(iface);
483 if (port->n_ifaces) {
486 VLOG_ERR("%s port has no interfaces, dropping", port->name);
492 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
493 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
494 * responsible for freeing '*managersp' (with free()).
496 * You may be asking yourself "why does ovs-vswitchd care?", because
497 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
498 * should not be and in fact is not directly involved in that. But
499 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
500 * it has to tell in-band control where the managers are to enable that.
503 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
504 struct sockaddr_in **managersp, size_t *n_managersp)
506 struct sockaddr_in *managers = NULL;
507 size_t n_managers = 0;
509 if (ovs_cfg->n_managers > 0) {
512 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
513 for (i = 0; i < ovs_cfg->n_managers; i++) {
514 const char *name = ovs_cfg->managers[i];
515 struct sockaddr_in *sin = &managers[i];
517 if ((!strncmp(name, "tcp:", 4)
518 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
519 (!strncmp(name, "ssl:", 4)
520 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
526 *managersp = managers;
527 *n_managersp = n_managers;
531 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
533 struct shash old_br, new_br;
534 struct shash_node *node;
535 struct bridge *br, *next;
536 struct sockaddr_in *managers;
539 int sflow_bridge_number;
541 COVERAGE_INC(bridge_reconfigure);
543 collect_managers(ovs_cfg, &managers, &n_managers);
545 /* Collect old and new bridges. */
548 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
549 shash_add(&old_br, br->name, br);
551 for (i = 0; i < ovs_cfg->n_bridges; i++) {
552 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
553 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
554 VLOG_WARN("more than one bridge named %s", br_cfg->name);
558 /* Get rid of deleted bridges and add new bridges. */
559 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
560 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
567 SHASH_FOR_EACH (node, &new_br) {
568 const char *br_name = node->name;
569 const struct ovsrec_bridge *br_cfg = node->data;
570 br = shash_find_data(&old_br, br_name);
572 /* If the bridge datapath type has changed, we need to tear it
573 * down and recreate. */
574 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
576 bridge_create(br_cfg);
579 bridge_create(br_cfg);
582 shash_destroy(&old_br);
583 shash_destroy(&new_br);
585 /* Reconfigure all bridges. */
586 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
587 bridge_reconfigure_one(br);
590 /* Add and delete ports on all datapaths.
592 * The kernel will reject any attempt to add a given port to a datapath if
593 * that port already belongs to a different datapath, so we must do all
594 * port deletions before any port additions. */
595 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
596 struct xflow_port *xfif_ports;
598 struct shash want_ifaces;
600 xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports);
601 bridge_get_all_ifaces(br, &want_ifaces);
602 for (i = 0; i < n_xfif_ports; i++) {
603 const struct xflow_port *p = &xfif_ports[i];
604 if (!shash_find(&want_ifaces, p->devname)
605 && strcmp(p->devname, br->name)) {
606 int retval = xfif_port_del(br->xfif, p->port);
608 VLOG_ERR("failed to remove %s interface from %s: %s",
609 p->devname, xfif_name(br->xfif),
614 shash_destroy(&want_ifaces);
617 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
618 struct xflow_port *xfif_ports;
620 struct shash cur_ifaces, want_ifaces;
621 struct shash_node *node;
623 /* Get the set of interfaces currently in this datapath. */
624 xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports);
625 shash_init(&cur_ifaces);
626 for (i = 0; i < n_xfif_ports; i++) {
627 const char *name = xfif_ports[i].devname;
628 shash_add_once(&cur_ifaces, name, NULL);
632 /* Get the set of interfaces we want on this datapath. */
633 bridge_get_all_ifaces(br, &want_ifaces);
635 SHASH_FOR_EACH (node, &want_ifaces) {
636 const char *if_name = node->name;
637 struct iface *iface = node->data;
639 if (shash_find(&cur_ifaces, if_name)) {
640 /* Already exists, just reconfigure it. */
642 reconfigure_iface(iface->cfg, iface);
645 /* Need to add to datapath. */
649 /* Add to datapath. */
650 internal = iface_is_internal(br, if_name);
651 error = xfif_port_add(br->xfif, if_name,
652 internal ? XFLOW_PORT_INTERNAL : 0, NULL);
653 if (error == EFBIG) {
654 VLOG_ERR("ran out of valid port numbers on %s",
655 xfif_name(br->xfif));
658 VLOG_ERR("failed to add %s interface to %s: %s",
659 if_name, xfif_name(br->xfif), strerror(error));
663 shash_destroy(&cur_ifaces);
664 shash_destroy(&want_ifaces);
666 sflow_bridge_number = 0;
667 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
670 struct iface *local_iface;
671 struct iface *hw_addr_iface;
674 bridge_fetch_dp_ifaces(br);
676 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
677 iterate_and_prune_ifaces(br, check_iface_xf_ifidx, NULL);
679 /* Pick local port hardware address, datapath ID. */
680 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
681 local_iface = bridge_get_local_iface(br);
683 int error = netdev_set_etheraddr(local_iface->netdev, ea);
685 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
686 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
687 "Ethernet address: %s",
688 br->name, strerror(error));
692 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
693 ofproto_set_datapath_id(br->ofproto, dpid);
695 dpid_string = xasprintf("%016"PRIx64, dpid);
696 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
699 /* Set NetFlow configuration on this bridge. */
700 if (br->cfg->netflow) {
701 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
702 struct netflow_options opts;
704 memset(&opts, 0, sizeof opts);
706 xfif_get_netflow_ids(br->xfif, &opts.engine_type, &opts.engine_id);
707 if (nf_cfg->engine_type) {
708 opts.engine_type = *nf_cfg->engine_type;
710 if (nf_cfg->engine_id) {
711 opts.engine_id = *nf_cfg->engine_id;
714 opts.active_timeout = nf_cfg->active_timeout;
715 if (!opts.active_timeout) {
716 opts.active_timeout = -1;
717 } else if (opts.active_timeout < 0) {
718 VLOG_WARN("bridge %s: active timeout interval set to negative "
719 "value, using default instead (%d seconds)", br->name,
720 NF_ACTIVE_TIMEOUT_DEFAULT);
721 opts.active_timeout = -1;
724 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
725 if (opts.add_id_to_iface) {
726 if (opts.engine_id > 0x7f) {
727 VLOG_WARN("bridge %s: netflow port mangling may conflict "
728 "with another vswitch, choose an engine id less "
729 "than 128", br->name);
731 if (br->n_ports > 508) {
732 VLOG_WARN("bridge %s: netflow port mangling will conflict "
733 "with another port when more than 508 ports are "
738 opts.collectors.n = nf_cfg->n_targets;
739 opts.collectors.names = nf_cfg->targets;
740 if (ofproto_set_netflow(br->ofproto, &opts)) {
741 VLOG_ERR("bridge %s: problem setting netflow collectors",
745 ofproto_set_netflow(br->ofproto, NULL);
748 /* Set sFlow configuration on this bridge. */
749 if (br->cfg->sflow) {
750 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
751 struct ovsrec_controller **controllers;
752 struct ofproto_sflow_options oso;
753 size_t n_controllers;
756 memset(&oso, 0, sizeof oso);
758 oso.targets.n = sflow_cfg->n_targets;
759 oso.targets.names = sflow_cfg->targets;
761 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
762 if (sflow_cfg->sampling) {
763 oso.sampling_rate = *sflow_cfg->sampling;
766 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
767 if (sflow_cfg->polling) {
768 oso.polling_interval = *sflow_cfg->polling;
771 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
772 if (sflow_cfg->header) {
773 oso.header_len = *sflow_cfg->header;
776 oso.sub_id = sflow_bridge_number++;
777 oso.agent_device = sflow_cfg->agent;
779 oso.control_ip = NULL;
780 n_controllers = bridge_get_controllers(br, &controllers);
781 for (i = 0; i < n_controllers; i++) {
782 if (controllers[i]->local_ip) {
783 oso.control_ip = controllers[i]->local_ip;
787 ofproto_set_sflow(br->ofproto, &oso);
789 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
791 ofproto_set_sflow(br->ofproto, NULL);
794 /* Update the controller and related settings. It would be more
795 * straightforward to call this from bridge_reconfigure_one(), but we
796 * can't do it there for two reasons. First, and most importantly, at
797 * that point we don't know the xf_ifidx of any interfaces that have
798 * been added to the bridge (because we haven't actually added them to
799 * the datapath). Second, at that point we haven't set the datapath ID
800 * yet; when a controller is configured, resetting the datapath ID will
801 * immediately disconnect from the controller, so it's better to set
802 * the datapath ID before the controller. */
803 bridge_reconfigure_remotes(br, managers, n_managers);
805 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
806 for (i = 0; i < br->n_ports; i++) {
807 struct port *port = br->ports[i];
810 port_update_vlan_compat(port);
811 port_update_bonding(port);
813 for (j = 0; j < port->n_ifaces; j++) {
814 iface_update_qos(port->ifaces[j], port->cfg->qos);
818 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
819 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
826 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
827 const struct ovsdb_idl_column *column,
830 const struct ovsdb_datum *datum;
831 union ovsdb_atom atom;
834 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
835 atom.string = (char *) key;
836 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
837 return idx == UINT_MAX ? NULL : datum->values[idx].string;
841 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
843 return get_ovsrec_key_value(&br_cfg->header_,
844 &ovsrec_bridge_col_other_config, key);
848 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
849 struct iface **hw_addr_iface)
855 *hw_addr_iface = NULL;
857 /* Did the user request a particular MAC? */
858 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
859 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
860 if (eth_addr_is_multicast(ea)) {
861 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
862 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
863 } else if (eth_addr_is_zero(ea)) {
864 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
870 /* Otherwise choose the minimum non-local MAC address among all of the
872 memset(ea, 0xff, sizeof ea);
873 for (i = 0; i < br->n_ports; i++) {
874 struct port *port = br->ports[i];
875 uint8_t iface_ea[ETH_ADDR_LEN];
878 /* Mirror output ports don't participate. */
879 if (port->is_mirror_output_port) {
883 /* Choose the MAC address to represent the port. */
884 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
885 /* Find the interface with this Ethernet address (if any) so that
886 * we can provide the correct devname to the caller. */
888 for (j = 0; j < port->n_ifaces; j++) {
889 struct iface *candidate = port->ifaces[j];
890 uint8_t candidate_ea[ETH_ADDR_LEN];
891 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
892 && eth_addr_equals(iface_ea, candidate_ea)) {
897 /* Choose the interface whose MAC address will represent the port.
898 * The Linux kernel bonding code always chooses the MAC address of
899 * the first slave added to a bond, and the Fedora networking
900 * scripts always add slaves to a bond in alphabetical order, so
901 * for compatibility we choose the interface with the name that is
902 * first in alphabetical order. */
903 iface = port->ifaces[0];
904 for (j = 1; j < port->n_ifaces; j++) {
905 struct iface *candidate = port->ifaces[j];
906 if (strcmp(candidate->name, iface->name) < 0) {
911 /* The local port doesn't count (since we're trying to choose its
912 * MAC address anyway). */
913 if (iface->xf_ifidx == XFLOWP_LOCAL) {
918 error = netdev_get_etheraddr(iface->netdev, iface_ea);
920 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
921 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
922 iface->name, strerror(error));
927 /* Compare against our current choice. */
928 if (!eth_addr_is_multicast(iface_ea) &&
929 !eth_addr_is_local(iface_ea) &&
930 !eth_addr_is_reserved(iface_ea) &&
931 !eth_addr_is_zero(iface_ea) &&
932 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
934 memcpy(ea, iface_ea, ETH_ADDR_LEN);
935 *hw_addr_iface = iface;
938 if (eth_addr_is_multicast(ea)) {
939 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
940 *hw_addr_iface = NULL;
941 VLOG_WARN("bridge %s: using default bridge Ethernet "
942 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
944 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
945 br->name, ETH_ADDR_ARGS(ea));
949 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
950 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
951 * an interface on 'br', then that interface must be passed in as
952 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
953 * 'hw_addr_iface' must be passed in as a null pointer. */
955 bridge_pick_datapath_id(struct bridge *br,
956 const uint8_t bridge_ea[ETH_ADDR_LEN],
957 struct iface *hw_addr_iface)
960 * The procedure for choosing a bridge MAC address will, in the most
961 * ordinary case, also choose a unique MAC that we can use as a datapath
962 * ID. In some special cases, though, multiple bridges will end up with
963 * the same MAC address. This is OK for the bridges, but it will confuse
964 * the OpenFlow controller, because each datapath needs a unique datapath
967 * Datapath IDs must be unique. It is also very desirable that they be
968 * stable from one run to the next, so that policy set on a datapath
971 const char *datapath_id;
974 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
975 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
981 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
983 * A bridge whose MAC address is taken from a VLAN network device
984 * (that is, a network device created with vconfig(8) or similar
985 * tool) will have the same MAC address as a bridge on the VLAN
986 * device's physical network device.
988 * Handle this case by hashing the physical network device MAC
989 * along with the VLAN identifier.
991 uint8_t buf[ETH_ADDR_LEN + 2];
992 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
993 buf[ETH_ADDR_LEN] = vlan >> 8;
994 buf[ETH_ADDR_LEN + 1] = vlan;
995 return dpid_from_hash(buf, sizeof buf);
998 * Assume that this bridge's MAC address is unique, since it
999 * doesn't fit any of the cases we handle specially.
1004 * A purely internal bridge, that is, one that has no non-virtual
1005 * network devices on it at all, is more difficult because it has no
1006 * natural unique identifier at all.
1008 * When the host is a XenServer, we handle this case by hashing the
1009 * host's UUID with the name of the bridge. Names of bridges are
1010 * persistent across XenServer reboots, although they can be reused if
1011 * an internal network is destroyed and then a new one is later
1012 * created, so this is fairly effective.
1014 * When the host is not a XenServer, we punt by using a random MAC
1015 * address on each run.
1017 const char *host_uuid = xenserver_get_host_uuid();
1019 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1020 dpid = dpid_from_hash(combined, strlen(combined));
1026 return eth_addr_to_uint64(bridge_ea);
1030 dpid_from_hash(const void *data, size_t n)
1032 uint8_t hash[SHA1_DIGEST_SIZE];
1034 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1035 sha1_bytes(data, n, hash);
1036 eth_addr_mark_random(hash);
1037 return eth_addr_to_uint64(hash);
1041 iface_refresh_stats(struct iface *iface)
1047 static const struct iface_stat iface_stats[] = {
1048 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1049 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1050 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1051 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1052 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1053 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1054 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1055 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1056 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1057 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1058 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1059 { "collisions", offsetof(struct netdev_stats, collisions) },
1061 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1062 const struct iface_stat *s;
1064 char *keys[N_STATS];
1065 int64_t values[N_STATS];
1068 struct netdev_stats stats;
1070 /* Intentionally ignore return value, since errors will set 'stats' to
1071 * all-1s, and we will deal with that correctly below. */
1072 netdev_get_stats(iface->netdev, &stats);
1075 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1076 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1077 if (value != UINT64_MAX) {
1084 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1090 const struct ovsrec_open_vswitch *cfg;
1092 bool datapath_destroyed;
1093 bool database_changed;
1096 /* Let each bridge do the work that it needs to do. */
1097 datapath_destroyed = false;
1098 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1099 int error = bridge_run_one(br);
1101 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1102 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1103 "forcing reconfiguration", br->name);
1104 datapath_destroyed = true;
1108 /* (Re)configure if necessary. */
1109 database_changed = ovsdb_idl_run(idl);
1110 cfg = ovsrec_open_vswitch_first(idl);
1111 if (database_changed || datapath_destroyed) {
1113 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1115 bridge_configure_once(cfg);
1116 bridge_reconfigure(cfg);
1118 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1119 ovsdb_idl_txn_commit(txn);
1120 ovsdb_idl_txn_destroy(txn); /* XXX */
1122 /* We still need to reconfigure to avoid dangling pointers to
1123 * now-destroyed ovsrec structures inside bridge data. */
1124 static const struct ovsrec_open_vswitch null_cfg;
1126 bridge_reconfigure(&null_cfg);
1131 /* Re-configure SSL. We do this on every trip through the main loop,
1132 * instead of just when the database changes, because the contents of the
1133 * key and certificate files can change without the database changing. */
1134 if (cfg && cfg->ssl) {
1135 const struct ovsrec_ssl *ssl = cfg->ssl;
1137 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1138 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1142 /* Refresh interface stats if necessary. */
1143 if (time_msec() >= iface_stats_timer) {
1144 struct ovsdb_idl_txn *txn;
1146 txn = ovsdb_idl_txn_create(idl);
1147 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1150 for (i = 0; i < br->n_ports; i++) {
1151 struct port *port = br->ports[i];
1154 for (j = 0; j < port->n_ifaces; j++) {
1155 struct iface *iface = port->ifaces[j];
1156 iface_refresh_stats(iface);
1160 ovsdb_idl_txn_commit(txn);
1161 ovsdb_idl_txn_destroy(txn); /* XXX */
1163 iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
1172 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1173 ofproto_wait(br->ofproto);
1174 if (ofproto_has_primary_controller(br->ofproto)) {
1178 mac_learning_wait(br->ml);
1181 ovsdb_idl_wait(idl);
1182 poll_timer_wait_until(iface_stats_timer);
1185 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1186 * configuration changes. */
1188 bridge_flush(struct bridge *br)
1190 COVERAGE_INC(bridge_flush);
1191 ofproto_revalidate_all(br->ofproto);
1192 mac_learning_flush(br->ml);
1195 /* Returns the 'br' interface for the XFLOWP_LOCAL port, or null if 'br' has no
1196 * such interface. */
1197 static struct iface *
1198 bridge_get_local_iface(struct bridge *br)
1202 for (i = 0; i < br->n_ports; i++) {
1203 struct port *port = br->ports[i];
1204 for (j = 0; j < port->n_ifaces; j++) {
1205 struct iface *iface = port->ifaces[j];
1206 if (iface->xf_ifidx == XFLOWP_LOCAL) {
1215 /* Bridge unixctl user interface functions. */
1217 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1218 const char *args, void *aux OVS_UNUSED)
1220 struct ds ds = DS_EMPTY_INITIALIZER;
1221 const struct bridge *br;
1222 const struct mac_entry *e;
1224 br = bridge_lookup(args);
1226 unixctl_command_reply(conn, 501, "no such bridge");
1230 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1231 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1232 if (e->port < 0 || e->port >= br->n_ports) {
1235 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1236 br->ports[e->port]->ifaces[0]->xf_ifidx,
1237 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1239 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1243 /* Bridge reconfiguration functions. */
1244 static struct bridge *
1245 bridge_create(const struct ovsrec_bridge *br_cfg)
1250 assert(!bridge_lookup(br_cfg->name));
1251 br = xzalloc(sizeof *br);
1253 error = xfif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1259 xfif_flow_flush(br->xfif);
1261 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1264 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1266 xfif_delete(br->xfif);
1267 xfif_close(br->xfif);
1272 br->name = xstrdup(br_cfg->name);
1274 br->ml = mac_learning_create();
1275 eth_addr_nicira_random(br->default_ea);
1277 port_array_init(&br->ifaces);
1279 shash_init(&br->port_by_name);
1280 shash_init(&br->iface_by_name);
1282 list_push_back(&all_bridges, &br->node);
1284 VLOG_INFO("created bridge %s on %s", br->name, xfif_name(br->xfif));
1290 bridge_destroy(struct bridge *br)
1295 while (br->n_ports > 0) {
1296 port_destroy(br->ports[br->n_ports - 1]);
1298 list_remove(&br->node);
1299 error = xfif_delete(br->xfif);
1300 if (error && error != ENOENT) {
1301 VLOG_ERR("failed to delete %s: %s",
1302 xfif_name(br->xfif), strerror(error));
1304 xfif_close(br->xfif);
1305 ofproto_destroy(br->ofproto);
1306 mac_learning_destroy(br->ml);
1307 port_array_destroy(&br->ifaces);
1308 shash_destroy(&br->port_by_name);
1309 shash_destroy(&br->iface_by_name);
1316 static struct bridge *
1317 bridge_lookup(const char *name)
1321 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1322 if (!strcmp(br->name, name)) {
1329 /* Handle requests for a listing of all flows known by the OpenFlow
1330 * stack, including those normally hidden. */
1332 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1333 const char *args, void *aux OVS_UNUSED)
1338 br = bridge_lookup(args);
1340 unixctl_command_reply(conn, 501, "Unknown bridge");
1345 ofproto_get_all_flows(br->ofproto, &results);
1347 unixctl_command_reply(conn, 200, ds_cstr(&results));
1348 ds_destroy(&results);
1351 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1352 * connections and reconnect. If BRIDGE is not specified, then all bridges
1353 * drop their controller connections and reconnect. */
1355 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1356 const char *args, void *aux OVS_UNUSED)
1359 if (args[0] != '\0') {
1360 br = bridge_lookup(args);
1362 unixctl_command_reply(conn, 501, "Unknown bridge");
1365 ofproto_reconnect_controllers(br->ofproto);
1367 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1368 ofproto_reconnect_controllers(br->ofproto);
1371 unixctl_command_reply(conn, 200, NULL);
1375 bridge_run_one(struct bridge *br)
1377 ofproto_revalidate(br->ofproto, mac_learning_run(br->ml));
1379 return ofproto_run(br->ofproto);
1383 bridge_get_controllers(const struct bridge *br,
1384 struct ovsrec_controller ***controllersp)
1386 struct ovsrec_controller **controllers;
1387 size_t n_controllers;
1389 controllers = br->cfg->controller;
1390 n_controllers = br->cfg->n_controller;
1392 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1398 *controllersp = controllers;
1400 return n_controllers;
1404 bridge_reconfigure_one(struct bridge *br)
1406 struct shash old_ports, new_ports;
1407 struct svec snoops, old_snoops;
1408 struct shash_node *node;
1409 enum ofproto_fail_mode fail_mode;
1412 /* Collect old ports. */
1413 shash_init(&old_ports);
1414 for (i = 0; i < br->n_ports; i++) {
1415 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1418 /* Collect new ports. */
1419 shash_init(&new_ports);
1420 for (i = 0; i < br->cfg->n_ports; i++) {
1421 const char *name = br->cfg->ports[i]->name;
1422 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1423 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1428 /* If we have a controller, then we need a local port. Complain if the
1429 * user didn't specify one.
1431 * XXX perhaps we should synthesize a port ourselves in this case. */
1432 if (bridge_get_controllers(br, NULL)) {
1433 char local_name[IF_NAMESIZE];
1436 error = xfif_port_get_name(br->xfif, XFLOWP_LOCAL,
1437 local_name, sizeof local_name);
1438 if (!error && !shash_find(&new_ports, local_name)) {
1439 VLOG_WARN("bridge %s: controller specified but no local port "
1440 "(port named %s) defined",
1441 br->name, local_name);
1445 /* Get rid of deleted ports.
1446 * Get rid of deleted interfaces on ports that still exist. */
1447 SHASH_FOR_EACH (node, &old_ports) {
1448 struct port *port = node->data;
1449 const struct ovsrec_port *port_cfg;
1451 port_cfg = shash_find_data(&new_ports, node->name);
1455 port_del_ifaces(port, port_cfg);
1459 /* Create new ports.
1460 * Add new interfaces to existing ports.
1461 * Reconfigure existing ports. */
1462 SHASH_FOR_EACH (node, &new_ports) {
1463 struct port *port = shash_find_data(&old_ports, node->name);
1465 port = port_create(br, node->name);
1468 port_reconfigure(port, node->data);
1469 if (!port->n_ifaces) {
1470 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1471 br->name, port->name);
1475 shash_destroy(&old_ports);
1476 shash_destroy(&new_ports);
1478 /* Set the fail-mode */
1479 fail_mode = !br->cfg->fail_mode
1480 || !strcmp(br->cfg->fail_mode, "standalone")
1481 ? OFPROTO_FAIL_STANDALONE
1482 : OFPROTO_FAIL_SECURE;
1483 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1484 && !ofproto_has_primary_controller(br->ofproto)) {
1485 ofproto_flush_flows(br->ofproto);
1487 ofproto_set_fail_mode(br->ofproto, fail_mode);
1489 /* Delete all flows if we're switching from connected to standalone or vice
1490 * versa. (XXX Should we delete all flows if we are switching from one
1491 * controller to another?) */
1493 /* Configure OpenFlow controller connection snooping. */
1495 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1496 ovs_rundir, br->name));
1497 svec_init(&old_snoops);
1498 ofproto_get_snoops(br->ofproto, &old_snoops);
1499 if (!svec_equal(&snoops, &old_snoops)) {
1500 ofproto_set_snoops(br->ofproto, &snoops);
1502 svec_destroy(&snoops);
1503 svec_destroy(&old_snoops);
1505 mirror_reconfigure(br);
1508 /* Initializes 'oc' appropriately as a management service controller for
1511 * The caller must free oc->target when it is no longer needed. */
1513 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1514 struct ofproto_controller *oc)
1516 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir, br->name);
1517 oc->max_backoff = 0;
1518 oc->probe_interval = 60;
1519 oc->band = OFPROTO_OUT_OF_BAND;
1520 oc->accept_re = NULL;
1521 oc->update_resolv_conf = false;
1523 oc->burst_limit = 0;
1526 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1528 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1529 struct ofproto_controller *oc)
1531 oc->target = c->target;
1532 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1533 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1534 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1535 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1536 oc->accept_re = c->discover_accept_regex;
1537 oc->update_resolv_conf = c->discover_update_resolv_conf;
1538 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1539 oc->burst_limit = (c->controller_burst_limit
1540 ? *c->controller_burst_limit : 0);
1543 /* Configures the IP stack for 'br''s local interface properly according to the
1544 * configuration in 'c'. */
1546 bridge_configure_local_iface_netdev(struct bridge *br,
1547 struct ovsrec_controller *c)
1549 struct netdev *netdev;
1550 struct in_addr mask, gateway;
1552 struct iface *local_iface;
1555 /* Controller discovery does its own TCP/IP configuration later. */
1556 if (strcmp(c->target, "discover")) {
1560 /* If there's no local interface or no IP address, give up. */
1561 local_iface = bridge_get_local_iface(br);
1562 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1566 /* Bring up the local interface. */
1567 netdev = local_iface->netdev;
1568 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1570 /* Configure the IP address and netmask. */
1571 if (!c->local_netmask
1572 || !inet_aton(c->local_netmask, &mask)
1574 mask.s_addr = guess_netmask(ip.s_addr);
1576 if (!netdev_set_in4(netdev, ip, mask)) {
1577 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1578 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1581 /* Configure the default gateway. */
1582 if (c->local_gateway
1583 && inet_aton(c->local_gateway, &gateway)
1584 && gateway.s_addr) {
1585 if (!netdev_add_router(netdev, gateway)) {
1586 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1587 br->name, IP_ARGS(&gateway.s_addr));
1593 bridge_reconfigure_remotes(struct bridge *br,
1594 const struct sockaddr_in *managers,
1597 struct ovsrec_controller **controllers;
1598 size_t n_controllers;
1601 struct ofproto_controller *ocs;
1605 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1606 had_primary = ofproto_has_primary_controller(br->ofproto);
1608 n_controllers = bridge_get_controllers(br, &controllers);
1610 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1613 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1614 for (i = 0; i < n_controllers; i++) {
1615 struct ovsrec_controller *c = controllers[i];
1617 if (!strncmp(c->target, "punix:", 6)
1618 || !strncmp(c->target, "unix:", 5)) {
1619 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1621 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1622 * domain sockets and overwriting arbitrary local files. */
1623 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1624 "\"%s\" due to possibility for remote exploit",
1625 br->name, c->target);
1629 bridge_configure_local_iface_netdev(br, c);
1630 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs++]);
1633 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1634 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1637 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1638 ofproto_flush_flows(br->ofproto);
1641 /* If there are no controllers and the bridge is in standalone
1642 * mode, set up a flow that matches every packet and directs
1643 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1644 * switch is in secure mode and we won't pass any traffic until
1645 * a controller has been defined and it tells us to do so. */
1647 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1648 union ofp_action action;
1651 memset(&action, 0, sizeof action);
1652 action.type = htons(OFPAT_OUTPUT);
1653 action.output.len = htons(sizeof action);
1654 action.output.port = htons(OFPP_NORMAL);
1655 memset(&flow, 0, sizeof flow);
1656 flow.wildcards = OVSFW_ALL;
1657 ofproto_add_flow(br->ofproto, &flow, &action, 1, 0);
1662 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1667 for (i = 0; i < br->n_ports; i++) {
1668 struct port *port = br->ports[i];
1669 for (j = 0; j < port->n_ifaces; j++) {
1670 struct iface *iface = port->ifaces[j];
1671 shash_add_once(ifaces, iface->name, iface);
1673 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1674 shash_add_once(ifaces, port->name, NULL);
1679 /* For robustness, in case the administrator moves around datapath ports behind
1680 * our back, we re-check all the datapath port numbers here.
1682 * This function will set the 'xf_ifidx' members of interfaces that have
1683 * disappeared to -1, so only call this function from a context where those
1684 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1685 * 'xf_ifidx'es will cause trouble later when we try to send them to the
1686 * datapath, which doesn't support UINT16_MAX+1 ports. */
1688 bridge_fetch_dp_ifaces(struct bridge *br)
1690 struct xflow_port *xfif_ports;
1691 size_t n_xfif_ports;
1694 /* Reset all interface numbers. */
1695 for (i = 0; i < br->n_ports; i++) {
1696 struct port *port = br->ports[i];
1697 for (j = 0; j < port->n_ifaces; j++) {
1698 struct iface *iface = port->ifaces[j];
1699 iface->xf_ifidx = -1;
1702 port_array_clear(&br->ifaces);
1704 xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports);
1705 for (i = 0; i < n_xfif_ports; i++) {
1706 struct xflow_port *p = &xfif_ports[i];
1707 struct iface *iface = iface_lookup(br, p->devname);
1709 if (iface->xf_ifidx >= 0) {
1710 VLOG_WARN("%s reported interface %s twice",
1711 xfif_name(br->xfif), p->devname);
1712 } else if (iface_from_xf_ifidx(br, p->port)) {
1713 VLOG_WARN("%s reported interface %"PRIu16" twice",
1714 xfif_name(br->xfif), p->port);
1716 port_array_set(&br->ifaces, p->port, iface);
1717 iface->xf_ifidx = p->port;
1721 int64_t ofport = (iface->xf_ifidx >= 0
1722 ? xflow_port_to_ofp_port(iface->xf_ifidx)
1724 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1731 /* Bridge packet processing functions. */
1734 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1736 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1739 static struct bond_entry *
1740 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1742 return &port->bond_hash[bond_hash(mac)];
1746 bond_choose_iface(const struct port *port)
1748 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1749 size_t i, best_down_slave = -1;
1750 long long next_delay_expiration = LLONG_MAX;
1752 for (i = 0; i < port->n_ifaces; i++) {
1753 struct iface *iface = port->ifaces[i];
1755 if (iface->enabled) {
1757 } else if (iface->delay_expires < next_delay_expiration) {
1758 best_down_slave = i;
1759 next_delay_expiration = iface->delay_expires;
1763 if (best_down_slave != -1) {
1764 struct iface *iface = port->ifaces[best_down_slave];
1766 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1767 "since no other interface is up", iface->name,
1768 iface->delay_expires - time_msec());
1769 bond_enable_slave(iface, true);
1772 return best_down_slave;
1776 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1777 uint16_t *xf_ifidx, tag_type *tags)
1779 struct iface *iface;
1781 assert(port->n_ifaces);
1782 if (port->n_ifaces == 1) {
1783 iface = port->ifaces[0];
1785 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1786 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1787 || !port->ifaces[e->iface_idx]->enabled) {
1788 /* XXX select interface properly. The current interface selection
1789 * is only good for testing the rebalancing code. */
1790 e->iface_idx = bond_choose_iface(port);
1791 if (e->iface_idx < 0) {
1792 *tags |= port->no_ifaces_tag;
1795 e->iface_tag = tag_create_random();
1796 ((struct port *) port)->bond_compat_is_stale = true;
1798 *tags |= e->iface_tag;
1799 iface = port->ifaces[e->iface_idx];
1801 *xf_ifidx = iface->xf_ifidx;
1802 *tags |= iface->tag; /* Currently only used for bonding. */
1807 bond_link_status_update(struct iface *iface, bool carrier)
1809 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1810 struct port *port = iface->port;
1812 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1813 /* Nothing to do. */
1816 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1817 iface->name, carrier ? "detected" : "dropped");
1818 if (carrier == iface->enabled) {
1819 iface->delay_expires = LLONG_MAX;
1820 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1821 iface->name, carrier ? "disabled" : "enabled");
1822 } else if (carrier && port->active_iface < 0) {
1823 bond_enable_slave(iface, true);
1824 if (port->updelay) {
1825 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1826 "other interface is up", iface->name, port->updelay);
1829 int delay = carrier ? port->updelay : port->downdelay;
1830 iface->delay_expires = time_msec() + delay;
1833 "interface %s: will be %s if it stays %s for %d ms",
1835 carrier ? "enabled" : "disabled",
1836 carrier ? "up" : "down",
1843 bond_choose_active_iface(struct port *port)
1845 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1847 port->active_iface = bond_choose_iface(port);
1848 port->active_iface_tag = tag_create_random();
1849 if (port->active_iface >= 0) {
1850 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1851 port->name, port->ifaces[port->active_iface]->name);
1853 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1859 bond_enable_slave(struct iface *iface, bool enable)
1861 struct port *port = iface->port;
1862 struct bridge *br = port->bridge;
1864 /* This acts as a recursion check. If the act of disabling a slave
1865 * causes a different slave to be enabled, the flag will allow us to
1866 * skip redundant work when we reenter this function. It must be
1867 * cleared on exit to keep things safe with multiple bonds. */
1868 static bool moving_active_iface = false;
1870 iface->delay_expires = LLONG_MAX;
1871 if (enable == iface->enabled) {
1875 iface->enabled = enable;
1876 if (!iface->enabled) {
1877 VLOG_WARN("interface %s: disabled", iface->name);
1878 ofproto_revalidate(br->ofproto, iface->tag);
1879 if (iface->port_ifidx == port->active_iface) {
1880 ofproto_revalidate(br->ofproto,
1881 port->active_iface_tag);
1883 /* Disabling a slave can lead to another slave being immediately
1884 * enabled if there will be no active slaves but one is waiting
1885 * on an updelay. In this case we do not need to run most of the
1886 * code for the newly enabled slave since there was no period
1887 * without an active slave and it is redundant with the disabling
1889 moving_active_iface = true;
1890 bond_choose_active_iface(port);
1892 bond_send_learning_packets(port);
1894 VLOG_WARN("interface %s: enabled", iface->name);
1895 if (port->active_iface < 0 && !moving_active_iface) {
1896 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1897 bond_choose_active_iface(port);
1898 bond_send_learning_packets(port);
1900 iface->tag = tag_create_random();
1903 moving_active_iface = false;
1904 port->bond_compat_is_stale = true;
1907 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1908 * bond interface. */
1910 bond_update_fake_iface_stats(struct port *port)
1912 struct netdev_stats bond_stats;
1913 struct netdev *bond_dev;
1916 memset(&bond_stats, 0, sizeof bond_stats);
1918 for (i = 0; i < port->n_ifaces; i++) {
1919 struct netdev_stats slave_stats;
1921 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1922 /* XXX: We swap the stats here because they are swapped back when
1923 * reported by the internal device. The reason for this is
1924 * internal devices normally represent packets going into the system
1925 * but when used as fake bond device they represent packets leaving
1926 * the system. We really should do this in the internal device
1927 * itself because changing it here reverses the counts from the
1928 * perspective of the switch. However, the internal device doesn't
1929 * know what type of device it represents so we have to do it here
1931 bond_stats.tx_packets += slave_stats.rx_packets;
1932 bond_stats.tx_bytes += slave_stats.rx_bytes;
1933 bond_stats.rx_packets += slave_stats.tx_packets;
1934 bond_stats.rx_bytes += slave_stats.tx_bytes;
1938 if (!netdev_open_default(port->name, &bond_dev)) {
1939 netdev_set_stats(bond_dev, &bond_stats);
1940 netdev_close(bond_dev);
1945 bond_run(struct bridge *br)
1949 for (i = 0; i < br->n_ports; i++) {
1950 struct port *port = br->ports[i];
1952 if (port->n_ifaces >= 2) {
1953 for (j = 0; j < port->n_ifaces; j++) {
1954 struct iface *iface = port->ifaces[j];
1955 if (time_msec() >= iface->delay_expires) {
1956 bond_enable_slave(iface, !iface->enabled);
1960 if (port->bond_fake_iface
1961 && time_msec() >= port->bond_next_fake_iface_update) {
1962 bond_update_fake_iface_stats(port);
1963 port->bond_next_fake_iface_update = time_msec() + 1000;
1967 if (port->bond_compat_is_stale) {
1968 port->bond_compat_is_stale = false;
1969 port_update_bond_compat(port);
1975 bond_wait(struct bridge *br)
1979 for (i = 0; i < br->n_ports; i++) {
1980 struct port *port = br->ports[i];
1981 if (port->n_ifaces < 2) {
1984 for (j = 0; j < port->n_ifaces; j++) {
1985 struct iface *iface = port->ifaces[j];
1986 if (iface->delay_expires != LLONG_MAX) {
1987 poll_timer_wait_until(iface->delay_expires);
1990 if (port->bond_fake_iface) {
1991 poll_timer_wait_until(port->bond_next_fake_iface_update);
1997 set_dst(struct dst *p, const flow_t *flow,
1998 const struct port *in_port, const struct port *out_port,
2001 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2002 : in_port->vlan >= 0 ? in_port->vlan
2003 : ntohs(flow->dl_vlan));
2004 return choose_output_iface(out_port, flow->dl_src, &p->xf_ifidx, tags);
2008 swap_dst(struct dst *p, struct dst *q)
2010 struct dst tmp = *p;
2015 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2016 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2017 * that we push to the datapath. We could in fact fully sort the array by
2018 * vlan, but in most cases there are at most two different vlan tags so that's
2019 * possibly overkill.) */
2021 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2023 struct dst *first = dsts;
2024 struct dst *last = dsts + n_dsts;
2026 while (first != last) {
2028 * - All dsts < first have vlan == 'vlan'.
2029 * - All dsts >= last have vlan != 'vlan'.
2030 * - first < last. */
2031 while (first->vlan == vlan) {
2032 if (++first == last) {
2037 /* Same invariants, plus one additional:
2038 * - first->vlan != vlan.
2040 while (last[-1].vlan != vlan) {
2041 if (--last == first) {
2046 /* Same invariants, plus one additional:
2047 * - last[-1].vlan == vlan.*/
2048 swap_dst(first++, --last);
2053 mirror_mask_ffs(mirror_mask_t mask)
2055 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2060 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2061 const struct dst *test)
2064 for (i = 0; i < n_dsts; i++) {
2065 if (dsts[i].vlan == test->vlan && dsts[i].xf_ifidx == test->xf_ifidx) {
2073 port_trunks_vlan(const struct port *port, uint16_t vlan)
2075 return (port->vlan < 0
2076 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2080 port_includes_vlan(const struct port *port, uint16_t vlan)
2082 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2086 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2087 const struct port *in_port, const struct port *out_port,
2088 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2090 mirror_mask_t mirrors = in_port->src_mirrors;
2091 struct dst *dst = dsts;
2094 if (out_port == FLOOD_PORT) {
2095 /* XXX use XFLOW_FLOOD if no vlans or bonding. */
2096 /* XXX even better, define each VLAN as a datapath port group */
2097 for (i = 0; i < br->n_ports; i++) {
2098 struct port *port = br->ports[i];
2099 if (port != in_port && port_includes_vlan(port, vlan)
2100 && !port->is_mirror_output_port
2101 && set_dst(dst, flow, in_port, port, tags)) {
2102 mirrors |= port->dst_mirrors;
2106 *nf_output_iface = NF_OUT_FLOOD;
2107 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2108 *nf_output_iface = dst->xf_ifidx;
2109 mirrors |= out_port->dst_mirrors;
2114 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2115 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2117 if (set_dst(dst, flow, in_port, m->out_port, tags)
2118 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2122 for (i = 0; i < br->n_ports; i++) {
2123 struct port *port = br->ports[i];
2124 if (port_includes_vlan(port, m->out_vlan)
2125 && set_dst(dst, flow, in_port, port, tags))
2129 if (port->vlan < 0) {
2130 dst->vlan = m->out_vlan;
2132 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2136 /* Use the vlan tag on the original flow instead of
2137 * the one passed in the vlan parameter. This ensures
2138 * that we compare the vlan from before any implicit
2139 * tagging tags place. This is necessary because
2140 * dst->vlan is the final vlan, after removing implicit
2142 flow_vlan = ntohs(flow->dl_vlan);
2143 if (flow_vlan == 0) {
2144 flow_vlan = OFP_VLAN_NONE;
2146 if (port == in_port && dst->vlan == flow_vlan) {
2147 /* Don't send out input port on same VLAN. */
2155 mirrors &= mirrors - 1;
2158 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2162 static void OVS_UNUSED
2163 print_dsts(const struct dst *dsts, size_t n)
2165 for (; n--; dsts++) {
2166 printf(">p%"PRIu16, dsts->xf_ifidx);
2167 if (dsts->vlan != OFP_VLAN_NONE) {
2168 printf("v%"PRIu16, dsts->vlan);
2174 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2175 const struct port *in_port, const struct port *out_port,
2176 tag_type *tags, struct xflow_actions *actions,
2177 uint16_t *nf_output_iface)
2179 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2181 const struct dst *p;
2184 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2187 cur_vlan = ntohs(flow->dl_vlan);
2188 for (p = dsts; p < &dsts[n_dsts]; p++) {
2189 union xflow_action *a;
2190 if (p->vlan != cur_vlan) {
2191 if (p->vlan == OFP_VLAN_NONE) {
2192 xflow_actions_add(actions, XFLOWAT_STRIP_VLAN);
2194 a = xflow_actions_add(actions, XFLOWAT_SET_DL_TCI);
2195 a->dl_tci.tci = htons(p->vlan & VLAN_VID_MASK);
2196 a->dl_tci.mask = htons(VLAN_VID_MASK);
2200 a = xflow_actions_add(actions, XFLOWAT_OUTPUT);
2201 a->output.port = p->xf_ifidx;
2205 /* Returns the effective vlan of a packet, taking into account both the
2206 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2207 * the packet is untagged and -1 indicates it has an invalid header and
2208 * should be dropped. */
2209 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2210 struct port *in_port, bool have_packet)
2212 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2213 * belongs to VLAN 0, so we should treat both cases identically. (In the
2214 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2215 * presumably to allow a priority to be specified. In the latter case, the
2216 * packet does not have any 802.1Q header.) */
2217 int vlan = ntohs(flow->dl_vlan);
2218 if (vlan == OFP_VLAN_NONE) {
2221 if (in_port->vlan >= 0) {
2223 /* XXX support double tagging? */
2225 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2226 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2227 "packet received on port %s configured with "
2228 "implicit VLAN %"PRIu16,
2229 br->name, ntohs(flow->dl_vlan),
2230 in_port->name, in_port->vlan);
2234 vlan = in_port->vlan;
2236 if (!port_includes_vlan(in_port, vlan)) {
2238 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2239 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2240 "packet received on port %s not configured for "
2242 br->name, vlan, in_port->name, vlan);
2251 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2252 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2253 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2255 is_gratuitous_arp(const flow_t *flow)
2257 return (flow->dl_type == htons(ETH_TYPE_ARP)
2258 && eth_addr_is_broadcast(flow->dl_dst)
2259 && (flow->nw_proto == ARP_OP_REPLY
2260 || (flow->nw_proto == ARP_OP_REQUEST
2261 && flow->nw_src == flow->nw_dst)));
2265 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2266 struct port *in_port)
2268 enum grat_arp_lock_type lock_type;
2271 /* We don't want to learn from gratuitous ARP packets that are reflected
2272 * back over bond slaves so we lock the learning table. */
2273 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2274 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2275 GRAT_ARP_LOCK_CHECK;
2277 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2280 /* The log messages here could actually be useful in debugging,
2281 * so keep the rate limit relatively high. */
2282 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2284 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2285 "on port %s in VLAN %d",
2286 br->name, ETH_ADDR_ARGS(flow->dl_src),
2287 in_port->name, vlan);
2288 ofproto_revalidate(br->ofproto, rev_tag);
2292 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2293 * dropped. Returns true if they may be forwarded, false if they should be
2296 * If 'have_packet' is true, it indicates that the caller is processing a
2297 * received packet. If 'have_packet' is false, then the caller is just
2298 * revalidating an existing flow because configuration has changed. Either
2299 * way, 'have_packet' only affects logging (there is no point in logging errors
2300 * during revalidation).
2302 * Sets '*in_portp' to the input port. This will be a null pointer if
2303 * flow->in_port does not designate a known input port (in which case
2304 * is_admissible() returns false).
2306 * When returning true, sets '*vlanp' to the effective VLAN of the input
2307 * packet, as returned by flow_get_vlan().
2309 * May also add tags to '*tags', although the current implementation only does
2310 * so in one special case.
2313 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2314 tag_type *tags, int *vlanp, struct port **in_portp)
2316 struct iface *in_iface;
2317 struct port *in_port;
2320 /* Find the interface and port structure for the received packet. */
2321 in_iface = iface_from_xf_ifidx(br, ofp_port_to_xflow_port(flow->in_port));
2323 /* No interface? Something fishy... */
2325 /* Odd. A few possible reasons here:
2327 * - We deleted an interface but there are still a few packets
2328 * queued up from it.
2330 * - Someone externally added an interface (e.g. with "ovs-dpctl
2331 * add-if") that we don't know about.
2333 * - Packet arrived on the local port but the local port is not
2334 * one of our bridge ports.
2336 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2338 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2339 "interface %"PRIu16, br->name, flow->in_port);
2345 *in_portp = in_port = in_iface->port;
2346 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2351 /* Drop frames for reserved multicast addresses. */
2352 if (eth_addr_is_reserved(flow->dl_dst)) {
2356 /* Drop frames on ports reserved for mirroring. */
2357 if (in_port->is_mirror_output_port) {
2359 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2360 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2361 "%s, which is reserved exclusively for mirroring",
2362 br->name, in_port->name);
2367 /* Packets received on bonds need special attention to avoid duplicates. */
2368 if (in_port->n_ifaces > 1) {
2370 bool is_grat_arp_locked;
2372 if (eth_addr_is_multicast(flow->dl_dst)) {
2373 *tags |= in_port->active_iface_tag;
2374 if (in_port->active_iface != in_iface->port_ifidx) {
2375 /* Drop all multicast packets on inactive slaves. */
2380 /* Drop all packets for which we have learned a different input
2381 * port, because we probably sent the packet on one slave and got
2382 * it back on the other. Gratuitous ARP packets are an exception
2383 * to this rule: the host has moved to another switch. The exception
2384 * to the exception is if we locked the learning table to avoid
2385 * reflections on bond slaves. If this is the case, just drop the
2387 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2388 &is_grat_arp_locked);
2389 if (src_idx != -1 && src_idx != in_port->port_idx &&
2390 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2398 /* If the composed actions may be applied to any packet in the given 'flow',
2399 * returns true. Otherwise, the actions should only be applied to 'packet', or
2400 * not at all, if 'packet' was NULL. */
2402 process_flow(struct bridge *br, const flow_t *flow,
2403 const struct ofpbuf *packet, struct xflow_actions *actions,
2404 tag_type *tags, uint16_t *nf_output_iface)
2406 struct port *in_port;
2407 struct port *out_port;
2411 /* Check whether we should drop packets in this flow. */
2412 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2417 /* Learn source MAC (but don't try to learn from revalidation). */
2419 update_learning_table(br, flow, vlan, in_port);
2422 /* Determine output port. */
2423 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2425 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2426 out_port = br->ports[out_port_idx];
2427 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2428 /* If we are revalidating but don't have a learning entry then
2429 * eject the flow. Installing a flow that floods packets opens
2430 * up a window of time where we could learn from a packet reflected
2431 * on a bond and blackhole packets before the learning table is
2432 * updated to reflect the correct port. */
2435 out_port = FLOOD_PORT;
2438 /* Don't send packets out their input ports. */
2439 if (in_port == out_port) {
2445 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2452 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2455 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2456 const struct ofp_phy_port *opp,
2459 struct bridge *br = br_;
2460 struct iface *iface;
2463 iface = iface_from_xf_ifidx(br, ofp_port_to_xflow_port(opp->port_no));
2469 if (reason == OFPPR_DELETE) {
2470 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2471 br->name, iface->name);
2472 iface_destroy(iface);
2473 if (!port->n_ifaces) {
2474 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2475 br->name, port->name);
2481 if (port->n_ifaces > 1) {
2482 bool up = !(opp->state & OFPPS_LINK_DOWN);
2483 bond_link_status_update(iface, up);
2484 port_update_bond_compat(port);
2490 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2491 struct xflow_actions *actions, tag_type *tags,
2492 uint16_t *nf_output_iface, void *br_)
2494 struct bridge *br = br_;
2496 COVERAGE_INC(bridge_process_flow);
2498 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2502 bridge_account_flow_ofhook_cb(const flow_t *flow, tag_type tags,
2503 const union xflow_action *actions,
2504 size_t n_actions, unsigned long long int n_bytes,
2507 struct bridge *br = br_;
2508 const union xflow_action *a;
2509 struct port *in_port;
2513 /* Feed information from the active flows back into the learning table to
2514 * ensure that table is always in sync with what is actually flowing
2515 * through the datapath.
2517 * We test that 'tags' is nonzero to ensure that only flows that include an
2518 * OFPP_NORMAL action are used for learning. This works because
2519 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2520 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2521 update_learning_table(br, flow, vlan, in_port);
2524 /* Account for bond slave utilization. */
2525 if (!br->has_bonded_ports) {
2528 for (a = actions; a < &actions[n_actions]; a++) {
2529 if (a->type == XFLOWAT_OUTPUT) {
2530 struct port *out_port = port_from_xf_ifidx(br, a->output.port);
2531 if (out_port && out_port->n_ifaces >= 2) {
2532 struct bond_entry *e = lookup_bond_entry(out_port,
2534 e->tx_bytes += n_bytes;
2541 bridge_account_checkpoint_ofhook_cb(void *br_)
2543 struct bridge *br = br_;
2547 if (!br->has_bonded_ports) {
2552 for (i = 0; i < br->n_ports; i++) {
2553 struct port *port = br->ports[i];
2554 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2555 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2556 bond_rebalance_port(port);
2561 static struct ofhooks bridge_ofhooks = {
2562 bridge_port_changed_ofhook_cb,
2563 bridge_normal_ofhook_cb,
2564 bridge_account_flow_ofhook_cb,
2565 bridge_account_checkpoint_ofhook_cb,
2568 /* Bonding functions. */
2570 /* Statistics for a single interface on a bonded port, used for load-based
2571 * bond rebalancing. */
2572 struct slave_balance {
2573 struct iface *iface; /* The interface. */
2574 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2576 /* All the "bond_entry"s that are assigned to this interface, in order of
2577 * increasing tx_bytes. */
2578 struct bond_entry **hashes;
2582 /* Sorts pointers to pointers to bond_entries in ascending order by the
2583 * interface to which they are assigned, and within a single interface in
2584 * ascending order of bytes transmitted. */
2586 compare_bond_entries(const void *a_, const void *b_)
2588 const struct bond_entry *const *ap = a_;
2589 const struct bond_entry *const *bp = b_;
2590 const struct bond_entry *a = *ap;
2591 const struct bond_entry *b = *bp;
2592 if (a->iface_idx != b->iface_idx) {
2593 return a->iface_idx > b->iface_idx ? 1 : -1;
2594 } else if (a->tx_bytes != b->tx_bytes) {
2595 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2601 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2602 * *descending* order by number of bytes transmitted. */
2604 compare_slave_balance(const void *a_, const void *b_)
2606 const struct slave_balance *a = a_;
2607 const struct slave_balance *b = b_;
2608 if (a->iface->enabled != b->iface->enabled) {
2609 return a->iface->enabled ? -1 : 1;
2610 } else if (a->tx_bytes != b->tx_bytes) {
2611 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2618 swap_bals(struct slave_balance *a, struct slave_balance *b)
2620 struct slave_balance tmp = *a;
2625 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2626 * given that 'p' (and only 'p') might be in the wrong location.
2628 * This function invalidates 'p', since it might now be in a different memory
2631 resort_bals(struct slave_balance *p,
2632 struct slave_balance bals[], size_t n_bals)
2635 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2636 swap_bals(p, p - 1);
2638 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2639 swap_bals(p, p + 1);
2645 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2647 if (VLOG_IS_DBG_ENABLED()) {
2648 struct ds ds = DS_EMPTY_INITIALIZER;
2649 const struct slave_balance *b;
2651 for (b = bals; b < bals + n_bals; b++) {
2655 ds_put_char(&ds, ',');
2657 ds_put_format(&ds, " %s %"PRIu64"kB",
2658 b->iface->name, b->tx_bytes / 1024);
2660 if (!b->iface->enabled) {
2661 ds_put_cstr(&ds, " (disabled)");
2663 if (b->n_hashes > 0) {
2664 ds_put_cstr(&ds, " (");
2665 for (i = 0; i < b->n_hashes; i++) {
2666 const struct bond_entry *e = b->hashes[i];
2668 ds_put_cstr(&ds, " + ");
2670 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2671 e - port->bond_hash, e->tx_bytes / 1024);
2673 ds_put_cstr(&ds, ")");
2676 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2681 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2683 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2686 struct bond_entry *hash = from->hashes[hash_idx];
2687 struct port *port = from->iface->port;
2688 uint64_t delta = hash->tx_bytes;
2690 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2691 "from %s to %s (now carrying %"PRIu64"kB and "
2692 "%"PRIu64"kB load, respectively)",
2693 port->name, delta / 1024, hash - port->bond_hash,
2694 from->iface->name, to->iface->name,
2695 (from->tx_bytes - delta) / 1024,
2696 (to->tx_bytes + delta) / 1024);
2698 /* Delete element from from->hashes.
2700 * We don't bother to add the element to to->hashes because not only would
2701 * it require more work, the only purpose it would be to allow that hash to
2702 * be migrated to another slave in this rebalancing run, and there is no
2703 * point in doing that. */
2704 if (hash_idx == 0) {
2707 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2708 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2712 /* Shift load away from 'from' to 'to'. */
2713 from->tx_bytes -= delta;
2714 to->tx_bytes += delta;
2716 /* Arrange for flows to be revalidated. */
2717 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2718 hash->iface_idx = to->iface->port_ifidx;
2719 hash->iface_tag = tag_create_random();
2723 bond_rebalance_port(struct port *port)
2725 struct slave_balance bals[DP_MAX_PORTS];
2727 struct bond_entry *hashes[BOND_MASK + 1];
2728 struct slave_balance *b, *from, *to;
2729 struct bond_entry *e;
2732 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2733 * descending order of tx_bytes, so that bals[0] represents the most
2734 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2737 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2738 * array for each slave_balance structure, we sort our local array of
2739 * hashes in order by slave, so that all of the hashes for a given slave
2740 * become contiguous in memory, and then we point each 'hashes' members of
2741 * a slave_balance structure to the start of a contiguous group. */
2742 n_bals = port->n_ifaces;
2743 for (b = bals; b < &bals[n_bals]; b++) {
2744 b->iface = port->ifaces[b - bals];
2749 for (i = 0; i <= BOND_MASK; i++) {
2750 hashes[i] = &port->bond_hash[i];
2752 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2753 for (i = 0; i <= BOND_MASK; i++) {
2755 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2756 b = &bals[e->iface_idx];
2757 b->tx_bytes += e->tx_bytes;
2759 b->hashes = &hashes[i];
2764 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2765 log_bals(bals, n_bals, port);
2767 /* Discard slaves that aren't enabled (which were sorted to the back of the
2768 * array earlier). */
2769 while (!bals[n_bals - 1].iface->enabled) {
2776 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2777 to = &bals[n_bals - 1];
2778 for (from = bals; from < to; ) {
2779 uint64_t overload = from->tx_bytes - to->tx_bytes;
2780 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2781 /* The extra load on 'from' (and all less-loaded slaves), compared
2782 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2783 * it is less than ~1Mbps. No point in rebalancing. */
2785 } else if (from->n_hashes == 1) {
2786 /* 'from' only carries a single MAC hash, so we can't shift any
2787 * load away from it, even though we want to. */
2790 /* 'from' is carrying significantly more load than 'to', and that
2791 * load is split across at least two different hashes. Pick a hash
2792 * to migrate to 'to' (the least-loaded slave), given that doing so
2793 * must decrease the ratio of the load on the two slaves by at
2796 * The sort order we use means that we prefer to shift away the
2797 * smallest hashes instead of the biggest ones. There is little
2798 * reason behind this decision; we could use the opposite sort
2799 * order to shift away big hashes ahead of small ones. */
2803 for (i = 0; i < from->n_hashes; i++) {
2804 double old_ratio, new_ratio;
2805 uint64_t delta = from->hashes[i]->tx_bytes;
2807 if (delta == 0 || from->tx_bytes - delta == 0) {
2808 /* Pointless move. */
2812 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2814 if (to->tx_bytes == 0) {
2815 /* Nothing on the new slave, move it. */
2819 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2820 new_ratio = (double)(from->tx_bytes - delta) /
2821 (to->tx_bytes + delta);
2823 if (new_ratio == 0) {
2824 /* Should already be covered but check to prevent division
2829 if (new_ratio < 1) {
2830 new_ratio = 1 / new_ratio;
2833 if (old_ratio - new_ratio > 0.1) {
2834 /* Would decrease the ratio, move it. */
2838 if (i < from->n_hashes) {
2839 bond_shift_load(from, to, i);
2840 port->bond_compat_is_stale = true;
2842 /* If the result of the migration changed the relative order of
2843 * 'from' and 'to' swap them back to maintain invariants. */
2844 if (order_swapped) {
2845 swap_bals(from, to);
2848 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2849 * point to different slave_balance structures. It is only
2850 * valid to do these two operations in a row at all because we
2851 * know that 'from' will not move past 'to' and vice versa. */
2852 resort_bals(from, bals, n_bals);
2853 resort_bals(to, bals, n_bals);
2860 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2861 * historical data to decay to <1% in 7 rebalancing runs. */
2862 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2868 bond_send_learning_packets(struct port *port)
2870 struct bridge *br = port->bridge;
2871 struct mac_entry *e;
2872 struct ofpbuf packet;
2873 int error, n_packets, n_errors;
2875 if (!port->n_ifaces || port->active_iface < 0) {
2879 ofpbuf_init(&packet, 128);
2880 error = n_packets = n_errors = 0;
2881 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2882 union ofp_action actions[2], *a;
2888 if (e->port == port->port_idx
2889 || !choose_output_iface(port, e->mac, &xf_ifidx, &tags)) {
2893 /* Compose actions. */
2894 memset(actions, 0, sizeof actions);
2897 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2898 a->vlan_vid.len = htons(sizeof *a);
2899 a->vlan_vid.vlan_vid = htons(e->vlan);
2902 a->output.type = htons(OFPAT_OUTPUT);
2903 a->output.len = htons(sizeof *a);
2904 a->output.port = htons(xflow_port_to_ofp_port(xf_ifidx));
2909 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2911 flow_extract(&packet, 0, OFPP_NONE, &flow);
2912 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2919 ofpbuf_uninit(&packet);
2922 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2923 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2924 "packets, last error was: %s",
2925 port->name, n_errors, n_packets, strerror(error));
2927 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2928 port->name, n_packets);
2932 /* Bonding unixctl user interface functions. */
2935 bond_unixctl_list(struct unixctl_conn *conn,
2936 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2938 struct ds ds = DS_EMPTY_INITIALIZER;
2939 const struct bridge *br;
2941 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2943 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2946 for (i = 0; i < br->n_ports; i++) {
2947 const struct port *port = br->ports[i];
2948 if (port->n_ifaces > 1) {
2951 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2952 for (j = 0; j < port->n_ifaces; j++) {
2953 const struct iface *iface = port->ifaces[j];
2955 ds_put_cstr(&ds, ", ");
2957 ds_put_cstr(&ds, iface->name);
2959 ds_put_char(&ds, '\n');
2963 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2967 static struct port *
2968 bond_find(const char *name)
2970 const struct bridge *br;
2972 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2975 for (i = 0; i < br->n_ports; i++) {
2976 struct port *port = br->ports[i];
2977 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2986 bond_unixctl_show(struct unixctl_conn *conn,
2987 const char *args, void *aux OVS_UNUSED)
2989 struct ds ds = DS_EMPTY_INITIALIZER;
2990 const struct port *port;
2993 port = bond_find(args);
2995 unixctl_command_reply(conn, 501, "no such bond");
2999 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3000 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3001 ds_put_format(&ds, "next rebalance: %lld ms\n",
3002 port->bond_next_rebalance - time_msec());
3003 for (j = 0; j < port->n_ifaces; j++) {
3004 const struct iface *iface = port->ifaces[j];
3005 struct bond_entry *be;
3008 ds_put_format(&ds, "slave %s: %s\n",
3009 iface->name, iface->enabled ? "enabled" : "disabled");
3010 if (j == port->active_iface) {
3011 ds_put_cstr(&ds, "\tactive slave\n");
3013 if (iface->delay_expires != LLONG_MAX) {
3014 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3015 iface->enabled ? "downdelay" : "updelay",
3016 iface->delay_expires - time_msec());
3020 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3021 int hash = be - port->bond_hash;
3022 struct mac_entry *me;
3024 if (be->iface_idx != j) {
3028 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3029 hash, be->tx_bytes / 1024);
3032 LIST_FOR_EACH (me, struct mac_entry, lru_node,
3033 &port->bridge->ml->lrus) {
3036 if (bond_hash(me->mac) == hash
3037 && me->port != port->port_idx
3038 && choose_output_iface(port, me->mac, &xf_ifidx, &tags)
3039 && xf_ifidx == iface->xf_ifidx)
3041 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3042 ETH_ADDR_ARGS(me->mac));
3047 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3052 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3053 void *aux OVS_UNUSED)
3055 char *args = (char *) args_;
3056 char *save_ptr = NULL;
3057 char *bond_s, *hash_s, *slave_s;
3058 uint8_t mac[ETH_ADDR_LEN];
3060 struct iface *iface;
3061 struct bond_entry *entry;
3064 bond_s = strtok_r(args, " ", &save_ptr);
3065 hash_s = strtok_r(NULL, " ", &save_ptr);
3066 slave_s = strtok_r(NULL, " ", &save_ptr);
3068 unixctl_command_reply(conn, 501,
3069 "usage: bond/migrate BOND HASH SLAVE");
3073 port = bond_find(bond_s);
3075 unixctl_command_reply(conn, 501, "no such bond");
3079 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3080 == ETH_ADDR_SCAN_COUNT) {
3081 hash = bond_hash(mac);
3082 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3083 hash = atoi(hash_s) & BOND_MASK;
3085 unixctl_command_reply(conn, 501, "bad hash");
3089 iface = port_lookup_iface(port, slave_s);
3091 unixctl_command_reply(conn, 501, "no such slave");
3095 if (!iface->enabled) {
3096 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3100 entry = &port->bond_hash[hash];
3101 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3102 entry->iface_idx = iface->port_ifidx;
3103 entry->iface_tag = tag_create_random();
3104 port->bond_compat_is_stale = true;
3105 unixctl_command_reply(conn, 200, "migrated");
3109 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3110 void *aux OVS_UNUSED)
3112 char *args = (char *) args_;
3113 char *save_ptr = NULL;
3114 char *bond_s, *slave_s;
3116 struct iface *iface;
3118 bond_s = strtok_r(args, " ", &save_ptr);
3119 slave_s = strtok_r(NULL, " ", &save_ptr);
3121 unixctl_command_reply(conn, 501,
3122 "usage: bond/set-active-slave BOND SLAVE");
3126 port = bond_find(bond_s);
3128 unixctl_command_reply(conn, 501, "no such bond");
3132 iface = port_lookup_iface(port, slave_s);
3134 unixctl_command_reply(conn, 501, "no such slave");
3138 if (!iface->enabled) {
3139 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3143 if (port->active_iface != iface->port_ifidx) {
3144 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3145 port->active_iface = iface->port_ifidx;
3146 port->active_iface_tag = tag_create_random();
3147 VLOG_INFO("port %s: active interface is now %s",
3148 port->name, iface->name);
3149 bond_send_learning_packets(port);
3150 unixctl_command_reply(conn, 200, "done");
3152 unixctl_command_reply(conn, 200, "no change");
3157 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3159 char *args = (char *) args_;
3160 char *save_ptr = NULL;
3161 char *bond_s, *slave_s;
3163 struct iface *iface;
3165 bond_s = strtok_r(args, " ", &save_ptr);
3166 slave_s = strtok_r(NULL, " ", &save_ptr);
3168 unixctl_command_reply(conn, 501,
3169 "usage: bond/enable/disable-slave BOND SLAVE");
3173 port = bond_find(bond_s);
3175 unixctl_command_reply(conn, 501, "no such bond");
3179 iface = port_lookup_iface(port, slave_s);
3181 unixctl_command_reply(conn, 501, "no such slave");
3185 bond_enable_slave(iface, enable);
3186 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3190 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3191 void *aux OVS_UNUSED)
3193 enable_slave(conn, args, true);
3197 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3198 void *aux OVS_UNUSED)
3200 enable_slave(conn, args, false);
3204 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3205 void *aux OVS_UNUSED)
3207 uint8_t mac[ETH_ADDR_LEN];
3211 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3212 == ETH_ADDR_SCAN_COUNT) {
3213 hash = bond_hash(mac);
3215 hash_cstr = xasprintf("%u", hash);
3216 unixctl_command_reply(conn, 200, hash_cstr);
3219 unixctl_command_reply(conn, 501, "invalid mac");
3226 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3227 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3228 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3229 unixctl_command_register("bond/set-active-slave",
3230 bond_unixctl_set_active_slave, NULL);
3231 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3233 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3235 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3238 /* Port functions. */
3240 static struct port *
3241 port_create(struct bridge *br, const char *name)
3245 port = xzalloc(sizeof *port);
3247 port->port_idx = br->n_ports;
3249 port->trunks = NULL;
3250 port->name = xstrdup(name);
3251 port->active_iface = -1;
3253 if (br->n_ports >= br->allocated_ports) {
3254 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3257 br->ports[br->n_ports++] = port;
3258 shash_add_assert(&br->port_by_name, port->name, port);
3260 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3267 get_port_other_config(const struct ovsrec_port *port, const char *key,
3268 const char *default_value)
3272 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3274 return value ? value : default_value;
3278 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3280 struct shash new_ifaces;
3283 /* Collect list of new interfaces. */
3284 shash_init(&new_ifaces);
3285 for (i = 0; i < cfg->n_interfaces; i++) {
3286 const char *name = cfg->interfaces[i]->name;
3287 shash_add_once(&new_ifaces, name, NULL);
3290 /* Get rid of deleted interfaces. */
3291 for (i = 0; i < port->n_ifaces; ) {
3292 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3293 iface_destroy(port->ifaces[i]);
3299 shash_destroy(&new_ifaces);
3303 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3305 struct shash new_ifaces;
3306 long long int next_rebalance;
3307 unsigned long *trunks;
3313 /* Update settings. */
3314 port->updelay = cfg->bond_updelay;
3315 if (port->updelay < 0) {
3318 port->updelay = cfg->bond_downdelay;
3319 if (port->downdelay < 0) {
3320 port->downdelay = 0;
3322 port->bond_rebalance_interval = atoi(
3323 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3324 if (port->bond_rebalance_interval < 1000) {
3325 port->bond_rebalance_interval = 1000;
3327 next_rebalance = time_msec() + port->bond_rebalance_interval;
3328 if (port->bond_next_rebalance > next_rebalance) {
3329 port->bond_next_rebalance = next_rebalance;
3332 /* Add new interfaces and update 'cfg' member of existing ones. */
3333 shash_init(&new_ifaces);
3334 for (i = 0; i < cfg->n_interfaces; i++) {
3335 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3336 struct iface *iface;
3338 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3339 VLOG_WARN("port %s: %s specified twice as port interface",
3340 port->name, if_cfg->name);
3344 iface = iface_lookup(port->bridge, if_cfg->name);
3346 if (iface->port != port) {
3347 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3349 port->bridge->name, if_cfg->name, iface->port->name);
3352 iface->cfg = if_cfg;
3354 iface_create(port, if_cfg);
3357 shash_destroy(&new_ifaces);
3362 if (port->n_ifaces < 2) {
3364 if (vlan >= 0 && vlan <= 4095) {
3365 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3370 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3371 * they even work as-is. But they have not been tested. */
3372 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3376 if (port->vlan != vlan) {
3378 bridge_flush(port->bridge);
3381 /* Get trunked VLANs. */
3383 if (vlan < 0 && cfg->n_trunks) {
3387 trunks = bitmap_allocate(4096);
3389 for (i = 0; i < cfg->n_trunks; i++) {
3390 int trunk = cfg->trunks[i];
3392 bitmap_set1(trunks, trunk);
3398 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3399 port->name, cfg->n_trunks);
3401 if (n_errors == cfg->n_trunks) {
3402 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3404 bitmap_free(trunks);
3407 } else if (vlan >= 0 && cfg->n_trunks) {
3408 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3412 ? port->trunks != NULL
3413 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3414 bridge_flush(port->bridge);
3416 bitmap_free(port->trunks);
3417 port->trunks = trunks;
3421 port_destroy(struct port *port)
3424 struct bridge *br = port->bridge;
3428 proc_net_compat_update_vlan(port->name, NULL, 0);
3429 proc_net_compat_update_bond(port->name, NULL);
3431 for (i = 0; i < MAX_MIRRORS; i++) {
3432 struct mirror *m = br->mirrors[i];
3433 if (m && m->out_port == port) {
3438 while (port->n_ifaces > 0) {
3439 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3442 shash_find_and_delete_assert(&br->port_by_name, port->name);
3444 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3445 del->port_idx = port->port_idx;
3448 bitmap_free(port->trunks);
3455 static struct port *
3456 port_from_xf_ifidx(const struct bridge *br, uint16_t xf_ifidx)
3458 struct iface *iface = iface_from_xf_ifidx(br, xf_ifidx);
3459 return iface ? iface->port : NULL;
3462 static struct port *
3463 port_lookup(const struct bridge *br, const char *name)
3465 return shash_find_data(&br->port_by_name, name);
3468 static struct iface *
3469 port_lookup_iface(const struct port *port, const char *name)
3471 struct iface *iface = iface_lookup(port->bridge, name);
3472 return iface && iface->port == port ? iface : NULL;
3476 port_update_bonding(struct port *port)
3478 if (port->n_ifaces < 2) {
3479 /* Not a bonded port. */
3480 if (port->bond_hash) {
3481 free(port->bond_hash);
3482 port->bond_hash = NULL;
3483 port->bond_compat_is_stale = true;
3484 port->bond_fake_iface = false;
3487 if (!port->bond_hash) {
3490 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3491 for (i = 0; i <= BOND_MASK; i++) {
3492 struct bond_entry *e = &port->bond_hash[i];
3496 port->no_ifaces_tag = tag_create_random();
3497 bond_choose_active_iface(port);
3498 port->bond_next_rebalance
3499 = time_msec() + port->bond_rebalance_interval;
3501 if (port->cfg->bond_fake_iface) {
3502 port->bond_next_fake_iface_update = time_msec();
3505 port->bond_compat_is_stale = true;
3506 port->bond_fake_iface = port->cfg->bond_fake_iface;
3511 port_update_bond_compat(struct port *port)
3513 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3514 struct compat_bond bond;
3517 if (port->n_ifaces < 2) {
3518 proc_net_compat_update_bond(port->name, NULL);
3523 bond.updelay = port->updelay;
3524 bond.downdelay = port->downdelay;
3527 bond.hashes = compat_hashes;
3528 if (port->bond_hash) {
3529 const struct bond_entry *e;
3530 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3531 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3532 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3533 cbh->hash = e - port->bond_hash;
3534 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3539 bond.n_slaves = port->n_ifaces;
3540 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3541 for (i = 0; i < port->n_ifaces; i++) {
3542 struct iface *iface = port->ifaces[i];
3543 struct compat_bond_slave *slave = &bond.slaves[i];
3544 slave->name = iface->name;
3546 /* We need to make the same determination as the Linux bonding
3547 * code to determine whether a slave should be consider "up".
3548 * The Linux function bond_miimon_inspect() supports four
3549 * BOND_LINK_* states:
3551 * - BOND_LINK_UP: carrier detected, updelay has passed.
3552 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3553 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3554 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3556 * The function bond_info_show_slave() only considers BOND_LINK_UP
3557 * to be "up" and anything else to be "down".
3559 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3563 netdev_get_etheraddr(iface->netdev, slave->mac);
3566 if (port->bond_fake_iface) {
3567 struct netdev *bond_netdev;
3569 if (!netdev_open_default(port->name, &bond_netdev)) {
3571 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3573 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3575 netdev_close(bond_netdev);
3579 proc_net_compat_update_bond(port->name, &bond);
3584 port_update_vlan_compat(struct port *port)
3586 struct bridge *br = port->bridge;
3587 char *vlandev_name = NULL;
3589 if (port->vlan > 0) {
3590 /* Figure out the name that the VLAN device should actually have, if it
3591 * existed. This takes some work because the VLAN device would not
3592 * have port->name in its name; rather, it would have the trunk port's
3593 * name, and 'port' would be attached to a bridge that also had the
3594 * VLAN device one of its ports. So we need to find a trunk port that
3595 * includes port->vlan.
3597 * There might be more than one candidate. This doesn't happen on
3598 * XenServer, so if it happens we just pick the first choice in
3599 * alphabetical order instead of creating multiple VLAN devices. */
3601 for (i = 0; i < br->n_ports; i++) {
3602 struct port *p = br->ports[i];
3603 if (port_trunks_vlan(p, port->vlan)
3605 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3607 uint8_t ea[ETH_ADDR_LEN];
3608 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3609 if (!eth_addr_is_multicast(ea) &&
3610 !eth_addr_is_reserved(ea) &&
3611 !eth_addr_is_zero(ea)) {
3612 vlandev_name = p->name;
3617 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3620 /* Interface functions. */
3622 static struct iface *
3623 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3625 struct bridge *br = port->bridge;
3626 struct iface *iface;
3627 char *name = if_cfg->name;
3630 iface = xzalloc(sizeof *iface);
3632 iface->port_ifidx = port->n_ifaces;
3633 iface->name = xstrdup(name);
3634 iface->xf_ifidx = -1;
3635 iface->tag = tag_create_random();
3636 iface->delay_expires = LLONG_MAX;
3637 iface->netdev = NULL;
3638 iface->cfg = if_cfg;
3640 shash_add_assert(&br->iface_by_name, iface->name, iface);
3642 /* Attempt to create the network interface in case it doesn't exist yet. */
3643 if (!iface_is_internal(br, iface->name)) {
3644 error = set_up_iface(if_cfg, iface, true);
3646 VLOG_WARN("could not create iface %s: %s", iface->name,
3649 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3656 if (port->n_ifaces >= port->allocated_ifaces) {
3657 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3658 sizeof *port->ifaces);
3660 port->ifaces[port->n_ifaces++] = iface;
3661 if (port->n_ifaces > 1) {
3662 br->has_bonded_ports = true;
3665 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3673 iface_destroy(struct iface *iface)
3676 struct port *port = iface->port;
3677 struct bridge *br = port->bridge;
3678 bool del_active = port->active_iface == iface->port_ifidx;
3681 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3683 if (iface->xf_ifidx >= 0) {
3684 port_array_set(&br->ifaces, iface->xf_ifidx, NULL);
3687 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3688 del->port_ifidx = iface->port_ifidx;
3690 netdev_close(iface->netdev);
3693 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3694 bond_choose_active_iface(port);
3695 bond_send_learning_packets(port);
3701 bridge_flush(port->bridge);
3705 static struct iface *
3706 iface_lookup(const struct bridge *br, const char *name)
3708 return shash_find_data(&br->iface_by_name, name);
3711 static struct iface *
3712 iface_from_xf_ifidx(const struct bridge *br, uint16_t xf_ifidx)
3714 return port_array_get(&br->ifaces, xf_ifidx);
3717 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3718 * 'br', that is, an interface that is entirely simulated within the datapath.
3719 * The local port (XFLOWP_LOCAL) is always an internal interface. Other local
3720 * interfaces are created by setting "iface.<iface>.internal = true".
3722 * In addition, we have a kluge-y feature that creates an internal port with
3723 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3724 * This feature needs to go away in the long term. Until then, this is one
3725 * reason why this function takes a name instead of a struct iface: the fake
3726 * interfaces created this way do not have a struct iface. */
3728 iface_is_internal(const struct bridge *br, const char *if_name)
3730 struct iface *iface;
3733 if (!strcmp(if_name, br->name)) {
3737 iface = iface_lookup(br, if_name);
3738 if (iface && !strcmp(iface->cfg->type, "internal")) {
3742 port = port_lookup(br, if_name);
3743 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3749 /* Set Ethernet address of 'iface', if one is specified in the configuration
3752 iface_set_mac(struct iface *iface)
3754 uint8_t ea[ETH_ADDR_LEN];
3756 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3757 if (eth_addr_is_multicast(ea)) {
3758 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3760 } else if (iface->xf_ifidx == XFLOWP_LOCAL) {
3761 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3762 iface->name, iface->name);
3764 int error = netdev_set_etheraddr(iface->netdev, ea);
3766 VLOG_ERR("interface %s: setting MAC failed (%s)",
3767 iface->name, strerror(error));
3774 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3775 struct shash *shash)
3780 for (i = 0; i < n; i++) {
3781 shash_add(shash, keys[i], values[i]);
3785 struct iface_delete_queues_cbdata {
3786 struct netdev *netdev;
3787 const struct ovsdb_datum *queues;
3791 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
3793 union ovsdb_atom atom;
3795 atom.integer = target;
3796 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
3800 iface_delete_queues(unsigned int queue_id,
3801 const struct shash *details OVS_UNUSED, void *cbdata_)
3803 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3805 if (!queue_ids_include(cbdata->queues, queue_id)) {
3806 netdev_delete_queue(cbdata->netdev, queue_id);
3811 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3813 if (!qos || qos->type[0] == '\0') {
3814 netdev_set_qos(iface->netdev, NULL, NULL);
3816 struct iface_delete_queues_cbdata cbdata;
3817 struct shash details;
3820 /* Configure top-level Qos for 'iface'. */
3821 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3822 qos->n_other_config, &details);
3823 netdev_set_qos(iface->netdev, qos->type, &details);
3824 shash_destroy(&details);
3826 /* Deconfigure queues that were deleted. */
3827 cbdata.netdev = iface->netdev;
3828 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
3830 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3832 /* Configure queues for 'iface'. */
3833 for (i = 0; i < qos->n_queues; i++) {
3834 const struct ovsrec_queue *queue = qos->value_queues[i];
3835 unsigned int queue_id = qos->key_queues[i];
3837 shash_from_ovs_idl_map(queue->key_other_config,
3838 queue->value_other_config,
3839 queue->n_other_config, &details);
3840 netdev_set_queue(iface->netdev, queue_id, &details);
3841 shash_destroy(&details);
3846 /* Port mirroring. */
3848 static struct mirror *
3849 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
3853 for (i = 0; i < MAX_MIRRORS; i++) {
3854 struct mirror *m = br->mirrors[i];
3855 if (m && uuid_equals(uuid, &m->uuid)) {
3863 mirror_reconfigure(struct bridge *br)
3865 unsigned long *rspan_vlans;
3868 /* Get rid of deleted mirrors. */
3869 for (i = 0; i < MAX_MIRRORS; i++) {
3870 struct mirror *m = br->mirrors[i];
3872 const struct ovsdb_datum *mc;
3873 union ovsdb_atom atom;
3875 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
3876 atom.uuid = br->mirrors[i]->uuid;
3877 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
3883 /* Add new mirrors and reconfigure existing ones. */
3884 for (i = 0; i < br->cfg->n_mirrors; i++) {
3885 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3886 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
3888 mirror_reconfigure_one(m, cfg);
3890 mirror_create(br, cfg);
3894 /* Update port reserved status. */
3895 for (i = 0; i < br->n_ports; i++) {
3896 br->ports[i]->is_mirror_output_port = false;
3898 for (i = 0; i < MAX_MIRRORS; i++) {
3899 struct mirror *m = br->mirrors[i];
3900 if (m && m->out_port) {
3901 m->out_port->is_mirror_output_port = true;
3905 /* Update flooded vlans (for RSPAN). */
3907 if (br->cfg->n_flood_vlans) {
3908 rspan_vlans = bitmap_allocate(4096);
3910 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3911 int64_t vlan = br->cfg->flood_vlans[i];
3912 if (vlan >= 0 && vlan < 4096) {
3913 bitmap_set1(rspan_vlans, vlan);
3914 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3917 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3922 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3928 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
3933 for (i = 0; ; i++) {
3934 if (i >= MAX_MIRRORS) {
3935 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3936 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
3939 if (!br->mirrors[i]) {
3944 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
3947 br->mirrors[i] = m = xzalloc(sizeof *m);
3950 m->name = xstrdup(cfg->name);
3951 shash_init(&m->src_ports);
3952 shash_init(&m->dst_ports);
3958 mirror_reconfigure_one(m, cfg);
3962 mirror_destroy(struct mirror *m)
3965 struct bridge *br = m->bridge;
3968 for (i = 0; i < br->n_ports; i++) {
3969 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3970 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3973 shash_destroy(&m->src_ports);
3974 shash_destroy(&m->dst_ports);
3977 m->bridge->mirrors[m->idx] = NULL;
3986 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3987 struct shash *names)
3991 for (i = 0; i < n_ports; i++) {
3992 const char *name = ports[i]->name;
3993 if (port_lookup(m->bridge, name)) {
3994 shash_add_once(names, name, NULL);
3996 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3997 "port %s", m->bridge->name, m->name, name);
4003 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4009 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4011 for (i = 0; i < cfg->n_select_vlan; i++) {
4012 int64_t vlan = cfg->select_vlan[i];
4013 if (vlan < 0 || vlan > 4095) {
4014 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4015 m->bridge->name, m->name, vlan);
4017 (*vlans)[n_vlans++] = vlan;
4024 vlan_is_mirrored(const struct mirror *m, int vlan)
4028 for (i = 0; i < m->n_vlans; i++) {
4029 if (m->vlans[i] == vlan) {
4037 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4041 for (i = 0; i < m->n_vlans; i++) {
4042 if (port_trunks_vlan(p, m->vlans[i])) {
4050 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4052 struct shash src_ports, dst_ports;
4053 mirror_mask_t mirror_bit;
4054 struct port *out_port;
4061 if (strcmp(cfg->name, m->name)) {
4063 m->name = xstrdup(cfg->name);
4066 /* Get output port. */
4067 if (cfg->output_port) {
4068 out_port = port_lookup(m->bridge, cfg->output_port->name);
4070 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4071 m->bridge->name, m->name);
4077 if (cfg->output_vlan) {
4078 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4079 "output vlan; ignoring output vlan",
4080 m->bridge->name, m->name);
4082 } else if (cfg->output_vlan) {
4084 out_vlan = *cfg->output_vlan;
4086 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4087 m->bridge->name, m->name);
4092 shash_init(&src_ports);
4093 shash_init(&dst_ports);
4094 if (cfg->select_all) {
4095 for (i = 0; i < m->bridge->n_ports; i++) {
4096 const char *name = m->bridge->ports[i]->name;
4097 shash_add_once(&src_ports, name, NULL);
4098 shash_add_once(&dst_ports, name, NULL);
4103 /* Get ports, and drop duplicates and ports that don't exist. */
4104 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4106 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4109 /* Get all the vlans, and drop duplicate and invalid vlans. */
4110 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4113 /* Update mirror data. */
4114 if (!shash_equal_keys(&m->src_ports, &src_ports)
4115 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4116 || m->n_vlans != n_vlans
4117 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4118 || m->out_port != out_port
4119 || m->out_vlan != out_vlan) {
4120 bridge_flush(m->bridge);
4122 shash_swap(&m->src_ports, &src_ports);
4123 shash_swap(&m->dst_ports, &dst_ports);
4126 m->n_vlans = n_vlans;
4127 m->out_port = out_port;
4128 m->out_vlan = out_vlan;
4131 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4132 for (i = 0; i < m->bridge->n_ports; i++) {
4133 struct port *port = m->bridge->ports[i];
4135 if (shash_find(&m->src_ports, port->name)
4138 ? port_trunks_any_mirrored_vlan(m, port)
4139 : vlan_is_mirrored(m, port->vlan)))) {
4140 port->src_mirrors |= mirror_bit;
4142 port->src_mirrors &= ~mirror_bit;
4145 if (shash_find(&m->dst_ports, port->name)) {
4146 port->dst_mirrors |= mirror_bit;
4148 port->dst_mirrors &= ~mirror_bit;
4153 shash_destroy(&src_ports);
4154 shash_destroy(&dst_ports);