1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
18 #include "byte-order.h"
21 #include <arpa/inet.h>
24 #include <sys/socket.h>
26 #include <openflow/openflow.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
35 #include "classifier.h"
39 #include "dynamic-string.h"
45 #include "mac-learning.h"
48 #include "ofp-print.h"
50 #include "ofproto/netflow.h"
51 #include "ofproto/ofproto.h"
52 #include "ovsdb-data.h"
54 #include "poll-loop.h"
55 #include "proc-net-compat.h"
59 #include "socket-util.h"
60 #include "stream-ssl.h"
62 #include "system-stats.h"
67 #include "vswitchd/vswitch-idl.h"
68 #include "xenserver.h"
70 #include "sflow_api.h"
72 VLOG_DEFINE_THIS_MODULE(bridge);
80 /* These members are always valid. */
81 struct port *port; /* Containing port. */
82 size_t port_ifidx; /* Index within containing port. */
83 char *name; /* Host network device name. */
84 tag_type tag; /* Tag associated with this interface. */
85 long long delay_expires; /* Time after which 'enabled' may change. */
87 /* These members are valid only after bridge_reconfigure() causes them to
89 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
90 int dp_ifidx; /* Index within kernel datapath. */
91 struct netdev *netdev; /* Network device. */
92 bool enabled; /* May be chosen for flows? */
93 const char *type; /* Usually same as cfg->type. */
94 const struct ovsrec_interface *cfg;
97 #define BOND_MASK 0xff
99 int iface_idx; /* Index of assigned iface, or -1 if none. */
100 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
101 tag_type iface_tag; /* Tag associated with iface_idx. */
104 #define MAX_MIRRORS 32
105 typedef uint32_t mirror_mask_t;
106 #define MIRROR_MASK_C(X) UINT32_C(X)
107 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
109 struct bridge *bridge;
112 struct uuid uuid; /* UUID of this "mirror" record in database. */
114 /* Selection criteria. */
115 struct shash src_ports; /* Name is port name; data is always NULL. */
116 struct shash dst_ports; /* Name is port name; data is always NULL. */
121 struct port *out_port;
125 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
127 struct bridge *bridge;
129 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
130 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
131 * NULL if all VLANs are trunked. */
132 const struct ovsrec_port *cfg;
135 /* An ordinary bridge port has 1 interface.
136 * A bridge port for bonding has at least 2 interfaces. */
137 struct iface **ifaces;
138 size_t n_ifaces, allocated_ifaces;
141 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
142 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
143 tag_type active_iface_tag; /* Tag for bcast flows. */
144 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
145 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
146 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
147 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
148 long long int bond_next_fake_iface_update; /* Time of next update. */
149 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
150 long long int bond_next_rebalance; /* Next rebalancing time. */
151 struct netdev_monitor *monitor; /* Tracks carrier up/down status. */
153 /* Port mirroring info. */
154 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
155 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
156 bool is_mirror_output_port; /* Does port mirroring send frames here? */
159 #define DP_MAX_PORTS 255
161 struct list node; /* Node in global list of bridges. */
162 char *name; /* User-specified arbitrary name. */
163 struct mac_learning *ml; /* MAC learning table. */
164 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
165 const struct ovsrec_bridge *cfg;
167 /* OpenFlow switch processing. */
168 struct ofproto *ofproto; /* OpenFlow switch. */
170 /* Kernel datapath information. */
171 struct dpif *dpif; /* Datapath. */
172 struct hmap ifaces; /* Contains "struct iface"s. */
176 size_t n_ports, allocated_ports;
177 struct shash iface_by_name; /* "struct iface"s indexed by name. */
178 struct shash port_by_name; /* "struct port"s indexed by name. */
181 bool has_bonded_ports;
186 /* Port mirroring. */
187 struct mirror *mirrors[MAX_MIRRORS];
190 /* List of all bridges. */
191 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
193 /* OVSDB IDL used to obtain configuration. */
194 static struct ovsdb_idl *idl;
196 /* Each time this timer expires, the bridge fetches systems and interface
197 * statistics and pushes them into the database. */
198 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
199 static long long int stats_timer = LLONG_MIN;
201 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
202 static void bridge_destroy(struct bridge *);
203 static struct bridge *bridge_lookup(const char *name);
204 static unixctl_cb_func bridge_unixctl_dump_flows;
205 static unixctl_cb_func bridge_unixctl_reconnect;
206 static int bridge_run_one(struct bridge *);
207 static size_t bridge_get_controllers(const struct bridge *br,
208 struct ovsrec_controller ***controllersp);
209 static void bridge_reconfigure_one(struct bridge *);
210 static void bridge_reconfigure_remotes(struct bridge *,
211 const struct sockaddr_in *managers,
213 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
214 static void bridge_fetch_dp_ifaces(struct bridge *);
215 static void bridge_flush(struct bridge *);
216 static void bridge_pick_local_hw_addr(struct bridge *,
217 uint8_t ea[ETH_ADDR_LEN],
218 struct iface **hw_addr_iface);
219 static uint64_t bridge_pick_datapath_id(struct bridge *,
220 const uint8_t bridge_ea[ETH_ADDR_LEN],
221 struct iface *hw_addr_iface);
222 static struct iface *bridge_get_local_iface(struct bridge *);
223 static uint64_t dpid_from_hash(const void *, size_t nbytes);
225 static unixctl_cb_func bridge_unixctl_fdb_show;
227 static void bond_init(void);
228 static void bond_run(struct bridge *);
229 static void bond_wait(struct bridge *);
230 static void bond_rebalance_port(struct port *);
231 static void bond_send_learning_packets(struct port *);
232 static void bond_enable_slave(struct iface *iface, bool enable);
234 static struct port *port_create(struct bridge *, const char *name);
235 static void port_reconfigure(struct port *, const struct ovsrec_port *);
236 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
237 static void port_destroy(struct port *);
238 static struct port *port_lookup(const struct bridge *, const char *name);
239 static struct iface *port_lookup_iface(const struct port *, const char *name);
240 static struct port *port_from_dp_ifidx(const struct bridge *,
242 static void port_update_bond_compat(struct port *);
243 static void port_update_vlan_compat(struct port *);
244 static void port_update_bonding(struct port *);
246 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
247 static void mirror_destroy(struct mirror *);
248 static void mirror_reconfigure(struct bridge *);
249 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
250 static bool vlan_is_mirrored(const struct mirror *, int vlan);
252 static struct iface *iface_create(struct port *port,
253 const struct ovsrec_interface *if_cfg);
254 static void iface_destroy(struct iface *);
255 static struct iface *iface_lookup(const struct bridge *, const char *name);
256 static struct iface *iface_from_dp_ifidx(const struct bridge *,
258 static void iface_set_mac(struct iface *);
259 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
260 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
262 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
265 /* Hooks into ofproto processing. */
266 static struct ofhooks bridge_ofhooks;
268 /* Public functions. */
270 /* Initializes the bridge module, configuring it to obtain its configuration
271 * from an OVSDB server accessed over 'remote', which should be a string in a
272 * form acceptable to ovsdb_idl_create(). */
274 bridge_init(const char *remote)
276 /* Create connection to database. */
277 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
279 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
280 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
281 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
283 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
285 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
286 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
288 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
289 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
290 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
292 /* Register unixctl commands. */
293 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
294 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
296 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
301 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
302 * but for which the ovs-vswitchd configuration 'cfg' is required. */
304 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
306 static bool already_configured_once;
307 struct svec bridge_names;
308 struct svec dpif_names, dpif_types;
311 /* Only do this once per ovs-vswitchd run. */
312 if (already_configured_once) {
315 already_configured_once = true;
317 stats_timer = time_msec() + STATS_INTERVAL;
319 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
320 svec_init(&bridge_names);
321 for (i = 0; i < cfg->n_bridges; i++) {
322 svec_add(&bridge_names, cfg->bridges[i]->name);
324 svec_sort(&bridge_names);
326 /* Iterate over all system dpifs and delete any of them that do not appear
328 svec_init(&dpif_names);
329 svec_init(&dpif_types);
330 dp_enumerate_types(&dpif_types);
331 for (i = 0; i < dpif_types.n; i++) {
336 dp_enumerate_names(dpif_types.names[i], &dpif_names);
338 /* For each dpif... */
339 for (j = 0; j < dpif_names.n; j++) {
340 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
342 struct svec all_names;
345 /* ...check whether any of its names is in 'bridge_names'. */
346 svec_init(&all_names);
347 dpif_get_all_names(dpif, &all_names);
348 for (k = 0; k < all_names.n; k++) {
349 if (svec_contains(&bridge_names, all_names.names[k])) {
354 /* No. Delete the dpif. */
358 svec_destroy(&all_names);
363 svec_destroy(&bridge_names);
364 svec_destroy(&dpif_names);
365 svec_destroy(&dpif_types);
368 /* Initializes 'options' and fills it with the options for 'if_cfg'. Merges
369 * keys from "options" and "other_config", preferring "options" keys over
370 * "other_config" keys.
372 * The value strings in '*options' are taken directly from if_cfg, not copied,
373 * so the caller should not modify or free them. */
375 iface_get_options(const struct ovsrec_interface *if_cfg, struct shash *options)
379 shash_from_ovs_idl_map(if_cfg->key_options, if_cfg->value_options,
380 if_cfg->n_options, options);
382 for (i = 0; i < if_cfg->n_other_config; i++) {
383 char *key = if_cfg->key_other_config[i];
384 char *value = if_cfg->value_other_config[i];
386 if (!shash_find_data(options, key)) {
387 shash_add(options, key, value);
389 VLOG_WARN("%s: ignoring \"other_config\" key %s that conflicts "
390 "with \"options\" key %s", if_cfg->name, key, key);
395 /* Returns the type of network device that 'iface' should have. (This is
396 * ordinarily the same type as the interface, but the network devices for
397 * "internal" ports have type "system".) */
399 iface_get_netdev_type(const struct iface *iface)
401 return !strcmp(iface->type, "internal") ? "system" : iface->type;
404 /* Attempt to create the network device for 'iface' through the netdev
407 create_iface_netdev(struct iface *iface)
409 struct netdev_options netdev_options;
410 struct shash options;
413 memset(&netdev_options, 0, sizeof netdev_options);
414 netdev_options.name = iface->cfg->name;
415 netdev_options.type = iface_get_netdev_type(iface);
416 netdev_options.args = &options;
417 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
419 iface_get_options(iface->cfg, &options);
421 error = netdev_open(&netdev_options, &iface->netdev);
424 iface->enabled = netdev_get_carrier(iface->netdev);
427 shash_destroy(&options);
433 reconfigure_iface_netdev(struct iface *iface)
435 const char *netdev_type, *iface_type;
436 struct shash options;
439 /* Skip reconfiguration if the device has the wrong type. This shouldn't
441 iface_type = iface_get_netdev_type(iface);
442 netdev_type = netdev_get_type(iface->netdev);
443 if (iface_type && strcmp(netdev_type, iface_type)) {
444 VLOG_WARN("%s: attempting change device type from %s to %s",
445 iface->cfg->name, netdev_type, iface_type);
449 /* Reconfigure device. */
450 iface_get_options(iface->cfg, &options);
451 error = netdev_reconfigure(iface->netdev, &options);
452 shash_destroy(&options);
457 /* Callback for iterate_and_prune_ifaces(). */
459 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
461 if (!iface->netdev) {
462 /* We already reported a related error, don't bother duplicating it. */
466 if (iface->dp_ifidx < 0) {
467 VLOG_ERR("%s interface not in %s, dropping",
468 iface->name, dpif_name(br->dpif));
472 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
473 iface->name, iface->dp_ifidx);
477 /* Callback for iterate_and_prune_ifaces(). */
479 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
480 void *aux OVS_UNUSED)
482 /* Set policing attributes. */
483 netdev_set_policing(iface->netdev,
484 iface->cfg->ingress_policing_rate,
485 iface->cfg->ingress_policing_burst);
487 /* Set MAC address of internal interfaces other than the local
489 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
490 iface_set_mac(iface);
496 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
497 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
498 * deletes from 'br' any ports that no longer have any interfaces. */
500 iterate_and_prune_ifaces(struct bridge *br,
501 bool (*cb)(struct bridge *, struct iface *,
507 for (i = 0; i < br->n_ports; ) {
508 struct port *port = br->ports[i];
509 for (j = 0; j < port->n_ifaces; ) {
510 struct iface *iface = port->ifaces[j];
511 if (cb(br, iface, aux)) {
514 iface_set_ofport(iface->cfg, -1);
515 iface_destroy(iface);
519 if (port->n_ifaces) {
522 VLOG_ERR("%s port has no interfaces, dropping", port->name);
528 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
529 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
530 * responsible for freeing '*managersp' (with free()).
532 * You may be asking yourself "why does ovs-vswitchd care?", because
533 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
534 * should not be and in fact is not directly involved in that. But
535 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
536 * it has to tell in-band control where the managers are to enable that.
537 * (Thus, only managers connected in-band are collected.)
540 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
541 struct sockaddr_in **managersp, size_t *n_managersp)
543 struct sockaddr_in *managers = NULL;
544 size_t n_managers = 0;
545 struct shash targets;
548 /* Collect all of the potential targets, as the union of the "managers"
549 * column and the "targets" columns of the rows pointed to by
550 * "manager_options", excluding any that are out-of-band. */
551 shash_init(&targets);
552 for (i = 0; i < ovs_cfg->n_managers; i++) {
553 shash_add_once(&targets, ovs_cfg->managers[i], NULL);
555 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
556 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
558 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
559 shash_find_and_delete(&targets, m->target);
561 shash_add_once(&targets, m->target, NULL);
565 /* Now extract the targets' IP addresses. */
566 if (!shash_is_empty(&targets)) {
567 struct shash_node *node;
569 managers = xmalloc(shash_count(&targets) * sizeof *managers);
570 SHASH_FOR_EACH (node, &targets) {
571 const char *target = node->name;
572 struct sockaddr_in *sin = &managers[n_managers];
574 if ((!strncmp(target, "tcp:", 4)
575 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
576 (!strncmp(target, "ssl:", 4)
577 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
582 shash_destroy(&targets);
584 *managersp = managers;
585 *n_managersp = n_managers;
589 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
591 struct shash old_br, new_br;
592 struct shash_node *node;
593 struct bridge *br, *next;
594 struct sockaddr_in *managers;
597 int sflow_bridge_number;
599 COVERAGE_INC(bridge_reconfigure);
601 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
603 /* Collect old and new bridges. */
606 LIST_FOR_EACH (br, node, &all_bridges) {
607 shash_add(&old_br, br->name, br);
609 for (i = 0; i < ovs_cfg->n_bridges; i++) {
610 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
611 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
612 VLOG_WARN("more than one bridge named %s", br_cfg->name);
616 /* Get rid of deleted bridges and add new bridges. */
617 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
618 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
625 SHASH_FOR_EACH (node, &new_br) {
626 const char *br_name = node->name;
627 const struct ovsrec_bridge *br_cfg = node->data;
628 br = shash_find_data(&old_br, br_name);
630 /* If the bridge datapath type has changed, we need to tear it
631 * down and recreate. */
632 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
634 bridge_create(br_cfg);
637 bridge_create(br_cfg);
640 shash_destroy(&old_br);
641 shash_destroy(&new_br);
643 /* Reconfigure all bridges. */
644 LIST_FOR_EACH (br, node, &all_bridges) {
645 bridge_reconfigure_one(br);
648 /* Add and delete ports on all datapaths.
650 * The kernel will reject any attempt to add a given port to a datapath if
651 * that port already belongs to a different datapath, so we must do all
652 * port deletions before any port additions. */
653 LIST_FOR_EACH (br, node, &all_bridges) {
654 struct odp_port *dpif_ports;
656 struct shash want_ifaces;
658 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
659 bridge_get_all_ifaces(br, &want_ifaces);
660 for (i = 0; i < n_dpif_ports; i++) {
661 const struct odp_port *p = &dpif_ports[i];
662 if (!shash_find(&want_ifaces, p->devname)
663 && strcmp(p->devname, br->name)) {
664 int retval = dpif_port_del(br->dpif, p->port);
666 VLOG_ERR("failed to remove %s interface from %s: %s",
667 p->devname, dpif_name(br->dpif),
672 shash_destroy(&want_ifaces);
675 LIST_FOR_EACH (br, node, &all_bridges) {
676 struct odp_port *dpif_ports;
678 struct shash cur_ifaces, want_ifaces;
680 /* Get the set of interfaces currently in this datapath. */
681 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
682 shash_init(&cur_ifaces);
683 for (i = 0; i < n_dpif_ports; i++) {
684 const char *name = dpif_ports[i].devname;
685 shash_add_once(&cur_ifaces, name, &dpif_ports[i]);
688 /* Get the set of interfaces we want on this datapath. */
689 bridge_get_all_ifaces(br, &want_ifaces);
691 hmap_clear(&br->ifaces);
692 SHASH_FOR_EACH (node, &want_ifaces) {
693 const char *if_name = node->name;
694 struct iface *iface = node->data;
695 bool internal = !iface || !strcmp(iface->type, "internal");
696 struct odp_port *dpif_port = shash_find_data(&cur_ifaces, if_name);
699 /* If we have a port or a netdev already, and it's not the type we
700 * want, then delete the port (if any) and close the netdev (if
703 ? dpif_port && !(dpif_port->flags & ODP_PORT_INTERNAL)
705 && strcmp(iface->type, netdev_get_type(iface->netdev))))
708 error = ofproto_port_del(br->ofproto, dpif_port->port);
715 netdev_close(iface->netdev);
716 iface->netdev = NULL;
720 /* If it's not an internal port, open (possibly create) the
723 if (!iface->netdev) {
724 error = create_iface_netdev(iface);
726 VLOG_WARN("could not create iface %s: %s", iface->name,
731 reconfigure_iface_netdev(iface);
735 /* If it's not part of the datapath, add it. */
737 error = dpif_port_add(br->dpif, if_name,
738 internal ? ODP_PORT_INTERNAL : 0, NULL);
739 if (error == EFBIG) {
740 VLOG_ERR("ran out of valid port numbers on %s",
741 dpif_name(br->dpif));
744 VLOG_ERR("failed to add %s interface to %s: %s",
745 if_name, dpif_name(br->dpif), strerror(error));
750 /* If it's an internal port, open the netdev. */
752 if (iface && !iface->netdev) {
753 error = create_iface_netdev(iface);
755 VLOG_WARN("could not create iface %s: %s", iface->name,
761 assert(iface->netdev != NULL);
765 shash_destroy(&cur_ifaces);
766 shash_destroy(&want_ifaces);
768 sflow_bridge_number = 0;
769 LIST_FOR_EACH (br, node, &all_bridges) {
772 struct iface *local_iface;
773 struct iface *hw_addr_iface;
776 bridge_fetch_dp_ifaces(br);
778 iterate_and_prune_ifaces(br, check_iface, NULL);
780 /* Pick local port hardware address, datapath ID. */
781 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
782 local_iface = bridge_get_local_iface(br);
784 int error = netdev_set_etheraddr(local_iface->netdev, ea);
786 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
787 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
788 "Ethernet address: %s",
789 br->name, strerror(error));
793 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
794 ofproto_set_datapath_id(br->ofproto, dpid);
796 dpid_string = xasprintf("%016"PRIx64, dpid);
797 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
800 /* Set NetFlow configuration on this bridge. */
801 if (br->cfg->netflow) {
802 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
803 struct netflow_options opts;
805 memset(&opts, 0, sizeof opts);
807 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
808 if (nf_cfg->engine_type) {
809 opts.engine_type = *nf_cfg->engine_type;
811 if (nf_cfg->engine_id) {
812 opts.engine_id = *nf_cfg->engine_id;
815 opts.active_timeout = nf_cfg->active_timeout;
816 if (!opts.active_timeout) {
817 opts.active_timeout = -1;
818 } else if (opts.active_timeout < 0) {
819 VLOG_WARN("bridge %s: active timeout interval set to negative "
820 "value, using default instead (%d seconds)", br->name,
821 NF_ACTIVE_TIMEOUT_DEFAULT);
822 opts.active_timeout = -1;
825 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
826 if (opts.add_id_to_iface) {
827 if (opts.engine_id > 0x7f) {
828 VLOG_WARN("bridge %s: netflow port mangling may conflict "
829 "with another vswitch, choose an engine id less "
830 "than 128", br->name);
832 if (br->n_ports > 508) {
833 VLOG_WARN("bridge %s: netflow port mangling will conflict "
834 "with another port when more than 508 ports are "
839 opts.collectors.n = nf_cfg->n_targets;
840 opts.collectors.names = nf_cfg->targets;
841 if (ofproto_set_netflow(br->ofproto, &opts)) {
842 VLOG_ERR("bridge %s: problem setting netflow collectors",
846 ofproto_set_netflow(br->ofproto, NULL);
849 /* Set sFlow configuration on this bridge. */
850 if (br->cfg->sflow) {
851 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
852 struct ovsrec_controller **controllers;
853 struct ofproto_sflow_options oso;
854 size_t n_controllers;
856 memset(&oso, 0, sizeof oso);
858 oso.targets.n = sflow_cfg->n_targets;
859 oso.targets.names = sflow_cfg->targets;
861 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
862 if (sflow_cfg->sampling) {
863 oso.sampling_rate = *sflow_cfg->sampling;
866 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
867 if (sflow_cfg->polling) {
868 oso.polling_interval = *sflow_cfg->polling;
871 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
872 if (sflow_cfg->header) {
873 oso.header_len = *sflow_cfg->header;
876 oso.sub_id = sflow_bridge_number++;
877 oso.agent_device = sflow_cfg->agent;
879 oso.control_ip = NULL;
880 n_controllers = bridge_get_controllers(br, &controllers);
881 for (i = 0; i < n_controllers; i++) {
882 if (controllers[i]->local_ip) {
883 oso.control_ip = controllers[i]->local_ip;
887 ofproto_set_sflow(br->ofproto, &oso);
889 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
891 ofproto_set_sflow(br->ofproto, NULL);
894 /* Update the controller and related settings. It would be more
895 * straightforward to call this from bridge_reconfigure_one(), but we
896 * can't do it there for two reasons. First, and most importantly, at
897 * that point we don't know the dp_ifidx of any interfaces that have
898 * been added to the bridge (because we haven't actually added them to
899 * the datapath). Second, at that point we haven't set the datapath ID
900 * yet; when a controller is configured, resetting the datapath ID will
901 * immediately disconnect from the controller, so it's better to set
902 * the datapath ID before the controller. */
903 bridge_reconfigure_remotes(br, managers, n_managers);
905 LIST_FOR_EACH (br, node, &all_bridges) {
906 for (i = 0; i < br->n_ports; i++) {
907 struct port *port = br->ports[i];
910 port_update_vlan_compat(port);
911 port_update_bonding(port);
913 for (j = 0; j < port->n_ifaces; j++) {
914 iface_update_qos(port->ifaces[j], port->cfg->qos);
918 LIST_FOR_EACH (br, node, &all_bridges) {
919 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
926 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
927 const struct ovsdb_idl_column *column,
930 const struct ovsdb_datum *datum;
931 union ovsdb_atom atom;
934 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
935 atom.string = (char *) key;
936 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
937 return idx == UINT_MAX ? NULL : datum->values[idx].string;
941 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
943 return get_ovsrec_key_value(&br_cfg->header_,
944 &ovsrec_bridge_col_other_config, key);
948 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
949 struct iface **hw_addr_iface)
955 *hw_addr_iface = NULL;
957 /* Did the user request a particular MAC? */
958 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
959 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
960 if (eth_addr_is_multicast(ea)) {
961 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
962 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
963 } else if (eth_addr_is_zero(ea)) {
964 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
970 /* Otherwise choose the minimum non-local MAC address among all of the
972 memset(ea, 0xff, sizeof ea);
973 for (i = 0; i < br->n_ports; i++) {
974 struct port *port = br->ports[i];
975 uint8_t iface_ea[ETH_ADDR_LEN];
978 /* Mirror output ports don't participate. */
979 if (port->is_mirror_output_port) {
983 /* Choose the MAC address to represent the port. */
984 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
985 /* Find the interface with this Ethernet address (if any) so that
986 * we can provide the correct devname to the caller. */
988 for (j = 0; j < port->n_ifaces; j++) {
989 struct iface *candidate = port->ifaces[j];
990 uint8_t candidate_ea[ETH_ADDR_LEN];
991 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
992 && eth_addr_equals(iface_ea, candidate_ea)) {
997 /* Choose the interface whose MAC address will represent the port.
998 * The Linux kernel bonding code always chooses the MAC address of
999 * the first slave added to a bond, and the Fedora networking
1000 * scripts always add slaves to a bond in alphabetical order, so
1001 * for compatibility we choose the interface with the name that is
1002 * first in alphabetical order. */
1003 iface = port->ifaces[0];
1004 for (j = 1; j < port->n_ifaces; j++) {
1005 struct iface *candidate = port->ifaces[j];
1006 if (strcmp(candidate->name, iface->name) < 0) {
1011 /* The local port doesn't count (since we're trying to choose its
1012 * MAC address anyway). */
1013 if (iface->dp_ifidx == ODPP_LOCAL) {
1018 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1020 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1021 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1022 iface->name, strerror(error));
1027 /* Compare against our current choice. */
1028 if (!eth_addr_is_multicast(iface_ea) &&
1029 !eth_addr_is_local(iface_ea) &&
1030 !eth_addr_is_reserved(iface_ea) &&
1031 !eth_addr_is_zero(iface_ea) &&
1032 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
1034 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1035 *hw_addr_iface = iface;
1038 if (eth_addr_is_multicast(ea)) {
1039 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1040 *hw_addr_iface = NULL;
1041 VLOG_WARN("bridge %s: using default bridge Ethernet "
1042 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1044 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1045 br->name, ETH_ADDR_ARGS(ea));
1049 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1050 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1051 * an interface on 'br', then that interface must be passed in as
1052 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1053 * 'hw_addr_iface' must be passed in as a null pointer. */
1055 bridge_pick_datapath_id(struct bridge *br,
1056 const uint8_t bridge_ea[ETH_ADDR_LEN],
1057 struct iface *hw_addr_iface)
1060 * The procedure for choosing a bridge MAC address will, in the most
1061 * ordinary case, also choose a unique MAC that we can use as a datapath
1062 * ID. In some special cases, though, multiple bridges will end up with
1063 * the same MAC address. This is OK for the bridges, but it will confuse
1064 * the OpenFlow controller, because each datapath needs a unique datapath
1067 * Datapath IDs must be unique. It is also very desirable that they be
1068 * stable from one run to the next, so that policy set on a datapath
1071 const char *datapath_id;
1074 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1075 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1079 if (hw_addr_iface) {
1081 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1083 * A bridge whose MAC address is taken from a VLAN network device
1084 * (that is, a network device created with vconfig(8) or similar
1085 * tool) will have the same MAC address as a bridge on the VLAN
1086 * device's physical network device.
1088 * Handle this case by hashing the physical network device MAC
1089 * along with the VLAN identifier.
1091 uint8_t buf[ETH_ADDR_LEN + 2];
1092 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1093 buf[ETH_ADDR_LEN] = vlan >> 8;
1094 buf[ETH_ADDR_LEN + 1] = vlan;
1095 return dpid_from_hash(buf, sizeof buf);
1098 * Assume that this bridge's MAC address is unique, since it
1099 * doesn't fit any of the cases we handle specially.
1104 * A purely internal bridge, that is, one that has no non-virtual
1105 * network devices on it at all, is more difficult because it has no
1106 * natural unique identifier at all.
1108 * When the host is a XenServer, we handle this case by hashing the
1109 * host's UUID with the name of the bridge. Names of bridges are
1110 * persistent across XenServer reboots, although they can be reused if
1111 * an internal network is destroyed and then a new one is later
1112 * created, so this is fairly effective.
1114 * When the host is not a XenServer, we punt by using a random MAC
1115 * address on each run.
1117 const char *host_uuid = xenserver_get_host_uuid();
1119 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1120 dpid = dpid_from_hash(combined, strlen(combined));
1126 return eth_addr_to_uint64(bridge_ea);
1130 dpid_from_hash(const void *data, size_t n)
1132 uint8_t hash[SHA1_DIGEST_SIZE];
1134 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1135 sha1_bytes(data, n, hash);
1136 eth_addr_mark_random(hash);
1137 return eth_addr_to_uint64(hash);
1141 iface_refresh_stats(struct iface *iface)
1147 static const struct iface_stat iface_stats[] = {
1148 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1149 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1150 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1151 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1152 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1153 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1154 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1155 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1156 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1157 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1158 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1159 { "collisions", offsetof(struct netdev_stats, collisions) },
1161 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1162 const struct iface_stat *s;
1164 char *keys[N_STATS];
1165 int64_t values[N_STATS];
1168 struct netdev_stats stats;
1170 /* Intentionally ignore return value, since errors will set 'stats' to
1171 * all-1s, and we will deal with that correctly below. */
1172 netdev_get_stats(iface->netdev, &stats);
1175 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1176 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1177 if (value != UINT64_MAX) {
1184 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1188 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1190 struct ovsdb_datum datum;
1194 get_system_stats(&stats);
1196 ovsdb_datum_from_shash(&datum, &stats);
1197 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1204 const struct ovsrec_open_vswitch *cfg;
1206 bool datapath_destroyed;
1207 bool database_changed;
1210 /* Let each bridge do the work that it needs to do. */
1211 datapath_destroyed = false;
1212 LIST_FOR_EACH (br, node, &all_bridges) {
1213 int error = bridge_run_one(br);
1215 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1216 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1217 "forcing reconfiguration", br->name);
1218 datapath_destroyed = true;
1222 /* (Re)configure if necessary. */
1223 database_changed = ovsdb_idl_run(idl);
1224 cfg = ovsrec_open_vswitch_first(idl);
1225 if (database_changed || datapath_destroyed) {
1227 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1229 bridge_configure_once(cfg);
1230 bridge_reconfigure(cfg);
1232 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1233 ovsdb_idl_txn_commit(txn);
1234 ovsdb_idl_txn_destroy(txn); /* XXX */
1236 /* We still need to reconfigure to avoid dangling pointers to
1237 * now-destroyed ovsrec structures inside bridge data. */
1238 static const struct ovsrec_open_vswitch null_cfg;
1240 bridge_reconfigure(&null_cfg);
1245 /* Re-configure SSL. We do this on every trip through the main loop,
1246 * instead of just when the database changes, because the contents of the
1247 * key and certificate files can change without the database changing. */
1248 if (cfg && cfg->ssl) {
1249 const struct ovsrec_ssl *ssl = cfg->ssl;
1251 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1252 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1256 /* Refresh system and interface stats if necessary. */
1257 if (time_msec() >= stats_timer) {
1259 struct ovsdb_idl_txn *txn;
1261 txn = ovsdb_idl_txn_create(idl);
1262 LIST_FOR_EACH (br, node, &all_bridges) {
1265 for (i = 0; i < br->n_ports; i++) {
1266 struct port *port = br->ports[i];
1269 for (j = 0; j < port->n_ifaces; j++) {
1270 struct iface *iface = port->ifaces[j];
1271 iface_refresh_stats(iface);
1275 refresh_system_stats(cfg);
1276 ovsdb_idl_txn_commit(txn);
1277 ovsdb_idl_txn_destroy(txn); /* XXX */
1280 stats_timer = time_msec() + STATS_INTERVAL;
1289 LIST_FOR_EACH (br, node, &all_bridges) {
1290 ofproto_wait(br->ofproto);
1291 if (ofproto_has_primary_controller(br->ofproto)) {
1295 mac_learning_wait(br->ml);
1298 ovsdb_idl_wait(idl);
1299 poll_timer_wait_until(stats_timer);
1302 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1303 * configuration changes. */
1305 bridge_flush(struct bridge *br)
1307 COVERAGE_INC(bridge_flush);
1309 mac_learning_flush(br->ml);
1312 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1313 * such interface. */
1314 static struct iface *
1315 bridge_get_local_iface(struct bridge *br)
1319 for (i = 0; i < br->n_ports; i++) {
1320 struct port *port = br->ports[i];
1321 for (j = 0; j < port->n_ifaces; j++) {
1322 struct iface *iface = port->ifaces[j];
1323 if (iface->dp_ifidx == ODPP_LOCAL) {
1332 /* Bridge unixctl user interface functions. */
1334 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1335 const char *args, void *aux OVS_UNUSED)
1337 struct ds ds = DS_EMPTY_INITIALIZER;
1338 const struct bridge *br;
1339 const struct mac_entry *e;
1341 br = bridge_lookup(args);
1343 unixctl_command_reply(conn, 501, "no such bridge");
1347 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1348 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1349 if (e->port < 0 || e->port >= br->n_ports) {
1352 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1353 br->ports[e->port]->ifaces[0]->dp_ifidx,
1354 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1356 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1360 /* Bridge reconfiguration functions. */
1361 static struct bridge *
1362 bridge_create(const struct ovsrec_bridge *br_cfg)
1367 assert(!bridge_lookup(br_cfg->name));
1368 br = xzalloc(sizeof *br);
1370 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1376 dpif_flow_flush(br->dpif);
1378 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1381 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1383 dpif_delete(br->dpif);
1384 dpif_close(br->dpif);
1389 br->name = xstrdup(br_cfg->name);
1391 br->ml = mac_learning_create();
1392 eth_addr_nicira_random(br->default_ea);
1394 hmap_init(&br->ifaces);
1396 shash_init(&br->port_by_name);
1397 shash_init(&br->iface_by_name);
1401 list_push_back(&all_bridges, &br->node);
1403 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1409 bridge_destroy(struct bridge *br)
1414 while (br->n_ports > 0) {
1415 port_destroy(br->ports[br->n_ports - 1]);
1417 list_remove(&br->node);
1418 error = dpif_delete(br->dpif);
1419 if (error && error != ENOENT) {
1420 VLOG_ERR("failed to delete %s: %s",
1421 dpif_name(br->dpif), strerror(error));
1423 dpif_close(br->dpif);
1424 ofproto_destroy(br->ofproto);
1425 mac_learning_destroy(br->ml);
1426 hmap_destroy(&br->ifaces);
1427 shash_destroy(&br->port_by_name);
1428 shash_destroy(&br->iface_by_name);
1435 static struct bridge *
1436 bridge_lookup(const char *name)
1440 LIST_FOR_EACH (br, node, &all_bridges) {
1441 if (!strcmp(br->name, name)) {
1448 /* Handle requests for a listing of all flows known by the OpenFlow
1449 * stack, including those normally hidden. */
1451 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1452 const char *args, void *aux OVS_UNUSED)
1457 br = bridge_lookup(args);
1459 unixctl_command_reply(conn, 501, "Unknown bridge");
1464 ofproto_get_all_flows(br->ofproto, &results);
1466 unixctl_command_reply(conn, 200, ds_cstr(&results));
1467 ds_destroy(&results);
1470 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1471 * connections and reconnect. If BRIDGE is not specified, then all bridges
1472 * drop their controller connections and reconnect. */
1474 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1475 const char *args, void *aux OVS_UNUSED)
1478 if (args[0] != '\0') {
1479 br = bridge_lookup(args);
1481 unixctl_command_reply(conn, 501, "Unknown bridge");
1484 ofproto_reconnect_controllers(br->ofproto);
1486 LIST_FOR_EACH (br, node, &all_bridges) {
1487 ofproto_reconnect_controllers(br->ofproto);
1490 unixctl_command_reply(conn, 200, NULL);
1494 bridge_run_one(struct bridge *br)
1498 error = ofproto_run1(br->ofproto);
1503 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1506 error = ofproto_run2(br->ofproto, br->flush);
1513 bridge_get_controllers(const struct bridge *br,
1514 struct ovsrec_controller ***controllersp)
1516 struct ovsrec_controller **controllers;
1517 size_t n_controllers;
1519 controllers = br->cfg->controller;
1520 n_controllers = br->cfg->n_controller;
1522 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1528 *controllersp = controllers;
1530 return n_controllers;
1534 bridge_reconfigure_one(struct bridge *br)
1536 struct shash old_ports, new_ports;
1537 struct svec snoops, old_snoops;
1538 struct shash_node *node;
1539 enum ofproto_fail_mode fail_mode;
1542 /* Collect old ports. */
1543 shash_init(&old_ports);
1544 for (i = 0; i < br->n_ports; i++) {
1545 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1548 /* Collect new ports. */
1549 shash_init(&new_ports);
1550 for (i = 0; i < br->cfg->n_ports; i++) {
1551 const char *name = br->cfg->ports[i]->name;
1552 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1553 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1558 /* If we have a controller, then we need a local port. Complain if the
1559 * user didn't specify one.
1561 * XXX perhaps we should synthesize a port ourselves in this case. */
1562 if (bridge_get_controllers(br, NULL)) {
1563 char local_name[IF_NAMESIZE];
1566 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1567 local_name, sizeof local_name);
1568 if (!error && !shash_find(&new_ports, local_name)) {
1569 VLOG_WARN("bridge %s: controller specified but no local port "
1570 "(port named %s) defined",
1571 br->name, local_name);
1575 /* Get rid of deleted ports.
1576 * Get rid of deleted interfaces on ports that still exist. */
1577 SHASH_FOR_EACH (node, &old_ports) {
1578 struct port *port = node->data;
1579 const struct ovsrec_port *port_cfg;
1581 port_cfg = shash_find_data(&new_ports, node->name);
1585 port_del_ifaces(port, port_cfg);
1589 /* Create new ports.
1590 * Add new interfaces to existing ports.
1591 * Reconfigure existing ports. */
1592 SHASH_FOR_EACH (node, &new_ports) {
1593 struct port *port = shash_find_data(&old_ports, node->name);
1595 port = port_create(br, node->name);
1598 port_reconfigure(port, node->data);
1599 if (!port->n_ifaces) {
1600 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1601 br->name, port->name);
1605 shash_destroy(&old_ports);
1606 shash_destroy(&new_ports);
1608 /* Set the fail-mode */
1609 fail_mode = !br->cfg->fail_mode
1610 || !strcmp(br->cfg->fail_mode, "standalone")
1611 ? OFPROTO_FAIL_STANDALONE
1612 : OFPROTO_FAIL_SECURE;
1613 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1614 && !ofproto_has_primary_controller(br->ofproto)) {
1615 ofproto_flush_flows(br->ofproto);
1617 ofproto_set_fail_mode(br->ofproto, fail_mode);
1619 /* Delete all flows if we're switching from connected to standalone or vice
1620 * versa. (XXX Should we delete all flows if we are switching from one
1621 * controller to another?) */
1623 /* Configure OpenFlow controller connection snooping. */
1625 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1626 ovs_rundir, br->name));
1627 svec_init(&old_snoops);
1628 ofproto_get_snoops(br->ofproto, &old_snoops);
1629 if (!svec_equal(&snoops, &old_snoops)) {
1630 ofproto_set_snoops(br->ofproto, &snoops);
1632 svec_destroy(&snoops);
1633 svec_destroy(&old_snoops);
1635 mirror_reconfigure(br);
1638 /* Initializes 'oc' appropriately as a management service controller for
1641 * The caller must free oc->target when it is no longer needed. */
1643 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1644 struct ofproto_controller *oc)
1646 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir, br->name);
1647 oc->max_backoff = 0;
1648 oc->probe_interval = 60;
1649 oc->band = OFPROTO_OUT_OF_BAND;
1650 oc->accept_re = NULL;
1651 oc->update_resolv_conf = false;
1653 oc->burst_limit = 0;
1656 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1658 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1659 struct ofproto_controller *oc)
1661 oc->target = c->target;
1662 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1663 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1664 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1665 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1666 oc->accept_re = c->discover_accept_regex;
1667 oc->update_resolv_conf = c->discover_update_resolv_conf;
1668 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1669 oc->burst_limit = (c->controller_burst_limit
1670 ? *c->controller_burst_limit : 0);
1673 /* Configures the IP stack for 'br''s local interface properly according to the
1674 * configuration in 'c'. */
1676 bridge_configure_local_iface_netdev(struct bridge *br,
1677 struct ovsrec_controller *c)
1679 struct netdev *netdev;
1680 struct in_addr mask, gateway;
1682 struct iface *local_iface;
1685 /* Controller discovery does its own TCP/IP configuration later. */
1686 if (strcmp(c->target, "discover")) {
1690 /* If there's no local interface or no IP address, give up. */
1691 local_iface = bridge_get_local_iface(br);
1692 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1696 /* Bring up the local interface. */
1697 netdev = local_iface->netdev;
1698 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1700 /* Configure the IP address and netmask. */
1701 if (!c->local_netmask
1702 || !inet_aton(c->local_netmask, &mask)
1704 mask.s_addr = guess_netmask(ip.s_addr);
1706 if (!netdev_set_in4(netdev, ip, mask)) {
1707 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1708 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1711 /* Configure the default gateway. */
1712 if (c->local_gateway
1713 && inet_aton(c->local_gateway, &gateway)
1714 && gateway.s_addr) {
1715 if (!netdev_add_router(netdev, gateway)) {
1716 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1717 br->name, IP_ARGS(&gateway.s_addr));
1723 bridge_reconfigure_remotes(struct bridge *br,
1724 const struct sockaddr_in *managers,
1727 const char *disable_ib_str, *queue_id_str;
1728 bool disable_in_band = false;
1731 struct ovsrec_controller **controllers;
1732 size_t n_controllers;
1735 struct ofproto_controller *ocs;
1739 /* Check if we should disable in-band control on this bridge. */
1740 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1741 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1742 disable_in_band = true;
1745 /* Set OpenFlow queue ID for in-band control. */
1746 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
1747 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
1748 ofproto_set_in_band_queue(br->ofproto, queue_id);
1750 if (disable_in_band) {
1751 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1753 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1755 had_primary = ofproto_has_primary_controller(br->ofproto);
1757 n_controllers = bridge_get_controllers(br, &controllers);
1759 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1762 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1763 for (i = 0; i < n_controllers; i++) {
1764 struct ovsrec_controller *c = controllers[i];
1766 if (!strncmp(c->target, "punix:", 6)
1767 || !strncmp(c->target, "unix:", 5)) {
1768 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1770 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1771 * domain sockets and overwriting arbitrary local files. */
1772 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1773 "\"%s\" due to possibility for remote exploit",
1774 dpif_name(br->dpif), c->target);
1778 bridge_configure_local_iface_netdev(br, c);
1779 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
1780 if (disable_in_band) {
1781 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
1786 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1787 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1790 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1791 ofproto_flush_flows(br->ofproto);
1794 /* If there are no controllers and the bridge is in standalone
1795 * mode, set up a flow that matches every packet and directs
1796 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1797 * switch is in secure mode and we won't pass any traffic until
1798 * a controller has been defined and it tells us to do so. */
1800 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1801 union ofp_action action;
1802 struct cls_rule rule;
1804 memset(&action, 0, sizeof action);
1805 action.type = htons(OFPAT_OUTPUT);
1806 action.output.len = htons(sizeof action);
1807 action.output.port = htons(OFPP_NORMAL);
1808 cls_rule_init_catchall(&rule, 0);
1809 ofproto_add_flow(br->ofproto, &rule, &action, 1);
1814 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1819 for (i = 0; i < br->n_ports; i++) {
1820 struct port *port = br->ports[i];
1821 for (j = 0; j < port->n_ifaces; j++) {
1822 struct iface *iface = port->ifaces[j];
1823 shash_add_once(ifaces, iface->name, iface);
1825 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1826 shash_add_once(ifaces, port->name, NULL);
1831 /* For robustness, in case the administrator moves around datapath ports behind
1832 * our back, we re-check all the datapath port numbers here.
1834 * This function will set the 'dp_ifidx' members of interfaces that have
1835 * disappeared to -1, so only call this function from a context where those
1836 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1837 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1838 * datapath, which doesn't support UINT16_MAX+1 ports. */
1840 bridge_fetch_dp_ifaces(struct bridge *br)
1842 struct odp_port *dpif_ports;
1843 size_t n_dpif_ports;
1846 /* Reset all interface numbers. */
1847 for (i = 0; i < br->n_ports; i++) {
1848 struct port *port = br->ports[i];
1849 for (j = 0; j < port->n_ifaces; j++) {
1850 struct iface *iface = port->ifaces[j];
1851 iface->dp_ifidx = -1;
1854 hmap_clear(&br->ifaces);
1856 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1857 for (i = 0; i < n_dpif_ports; i++) {
1858 struct odp_port *p = &dpif_ports[i];
1859 struct iface *iface = iface_lookup(br, p->devname);
1861 if (iface->dp_ifidx >= 0) {
1862 VLOG_WARN("%s reported interface %s twice",
1863 dpif_name(br->dpif), p->devname);
1864 } else if (iface_from_dp_ifidx(br, p->port)) {
1865 VLOG_WARN("%s reported interface %"PRIu16" twice",
1866 dpif_name(br->dpif), p->port);
1868 iface->dp_ifidx = p->port;
1869 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
1870 hash_int(iface->dp_ifidx, 0));
1873 iface_set_ofport(iface->cfg,
1874 (iface->dp_ifidx >= 0
1875 ? odp_port_to_ofp_port(iface->dp_ifidx)
1882 /* Bridge packet processing functions. */
1885 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1887 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1890 static struct bond_entry *
1891 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1893 return &port->bond_hash[bond_hash(mac)];
1897 bond_choose_iface(const struct port *port)
1899 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1900 size_t i, best_down_slave = -1;
1901 long long next_delay_expiration = LLONG_MAX;
1903 for (i = 0; i < port->n_ifaces; i++) {
1904 struct iface *iface = port->ifaces[i];
1906 if (iface->enabled) {
1908 } else if (iface->delay_expires < next_delay_expiration) {
1909 best_down_slave = i;
1910 next_delay_expiration = iface->delay_expires;
1914 if (best_down_slave != -1) {
1915 struct iface *iface = port->ifaces[best_down_slave];
1917 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1918 "since no other interface is up", iface->name,
1919 iface->delay_expires - time_msec());
1920 bond_enable_slave(iface, true);
1923 return best_down_slave;
1927 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1928 uint16_t *dp_ifidx, tag_type *tags)
1930 struct iface *iface;
1932 assert(port->n_ifaces);
1933 if (port->n_ifaces == 1) {
1934 iface = port->ifaces[0];
1936 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1937 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1938 || !port->ifaces[e->iface_idx]->enabled) {
1939 /* XXX select interface properly. The current interface selection
1940 * is only good for testing the rebalancing code. */
1941 e->iface_idx = bond_choose_iface(port);
1942 if (e->iface_idx < 0) {
1943 *tags |= port->no_ifaces_tag;
1946 e->iface_tag = tag_create_random();
1947 ((struct port *) port)->bond_compat_is_stale = true;
1949 *tags |= e->iface_tag;
1950 iface = port->ifaces[e->iface_idx];
1952 *dp_ifidx = iface->dp_ifidx;
1953 *tags |= iface->tag; /* Currently only used for bonding. */
1958 bond_link_status_update(struct iface *iface, bool carrier)
1960 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1961 struct port *port = iface->port;
1963 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1964 /* Nothing to do. */
1967 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1968 iface->name, carrier ? "detected" : "dropped");
1969 if (carrier == iface->enabled) {
1970 iface->delay_expires = LLONG_MAX;
1971 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1972 iface->name, carrier ? "disabled" : "enabled");
1973 } else if (carrier && port->active_iface < 0) {
1974 bond_enable_slave(iface, true);
1975 if (port->updelay) {
1976 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1977 "other interface is up", iface->name, port->updelay);
1980 int delay = carrier ? port->updelay : port->downdelay;
1981 iface->delay_expires = time_msec() + delay;
1984 "interface %s: will be %s if it stays %s for %d ms",
1986 carrier ? "enabled" : "disabled",
1987 carrier ? "up" : "down",
1994 bond_choose_active_iface(struct port *port)
1996 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1998 port->active_iface = bond_choose_iface(port);
1999 port->active_iface_tag = tag_create_random();
2000 if (port->active_iface >= 0) {
2001 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2002 port->name, port->ifaces[port->active_iface]->name);
2004 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2010 bond_enable_slave(struct iface *iface, bool enable)
2012 struct port *port = iface->port;
2013 struct bridge *br = port->bridge;
2015 /* This acts as a recursion check. If the act of disabling a slave
2016 * causes a different slave to be enabled, the flag will allow us to
2017 * skip redundant work when we reenter this function. It must be
2018 * cleared on exit to keep things safe with multiple bonds. */
2019 static bool moving_active_iface = false;
2021 iface->delay_expires = LLONG_MAX;
2022 if (enable == iface->enabled) {
2026 iface->enabled = enable;
2027 if (!iface->enabled) {
2028 VLOG_WARN("interface %s: disabled", iface->name);
2029 ofproto_revalidate(br->ofproto, iface->tag);
2030 if (iface->port_ifidx == port->active_iface) {
2031 ofproto_revalidate(br->ofproto,
2032 port->active_iface_tag);
2034 /* Disabling a slave can lead to another slave being immediately
2035 * enabled if there will be no active slaves but one is waiting
2036 * on an updelay. In this case we do not need to run most of the
2037 * code for the newly enabled slave since there was no period
2038 * without an active slave and it is redundant with the disabling
2040 moving_active_iface = true;
2041 bond_choose_active_iface(port);
2043 bond_send_learning_packets(port);
2045 VLOG_WARN("interface %s: enabled", iface->name);
2046 if (port->active_iface < 0 && !moving_active_iface) {
2047 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2048 bond_choose_active_iface(port);
2049 bond_send_learning_packets(port);
2051 iface->tag = tag_create_random();
2054 moving_active_iface = false;
2055 port->bond_compat_is_stale = true;
2058 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2059 * bond interface. */
2061 bond_update_fake_iface_stats(struct port *port)
2063 struct netdev_stats bond_stats;
2064 struct netdev *bond_dev;
2067 memset(&bond_stats, 0, sizeof bond_stats);
2069 for (i = 0; i < port->n_ifaces; i++) {
2070 struct netdev_stats slave_stats;
2072 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2073 /* XXX: We swap the stats here because they are swapped back when
2074 * reported by the internal device. The reason for this is
2075 * internal devices normally represent packets going into the system
2076 * but when used as fake bond device they represent packets leaving
2077 * the system. We really should do this in the internal device
2078 * itself because changing it here reverses the counts from the
2079 * perspective of the switch. However, the internal device doesn't
2080 * know what type of device it represents so we have to do it here
2082 bond_stats.tx_packets += slave_stats.rx_packets;
2083 bond_stats.tx_bytes += slave_stats.rx_bytes;
2084 bond_stats.rx_packets += slave_stats.tx_packets;
2085 bond_stats.rx_bytes += slave_stats.tx_bytes;
2089 if (!netdev_open_default(port->name, &bond_dev)) {
2090 netdev_set_stats(bond_dev, &bond_stats);
2091 netdev_close(bond_dev);
2096 bond_run(struct bridge *br)
2100 for (i = 0; i < br->n_ports; i++) {
2101 struct port *port = br->ports[i];
2103 if (port->n_ifaces >= 2) {
2106 /* Track carrier going up and down on interfaces. */
2107 while (!netdev_monitor_poll(port->monitor, &devname)) {
2108 struct iface *iface;
2110 iface = port_lookup_iface(port, devname);
2112 bool carrier = netdev_get_carrier(iface->netdev);
2114 bond_link_status_update(iface, carrier);
2115 port_update_bond_compat(port);
2120 for (j = 0; j < port->n_ifaces; j++) {
2121 struct iface *iface = port->ifaces[j];
2122 if (time_msec() >= iface->delay_expires) {
2123 bond_enable_slave(iface, !iface->enabled);
2127 if (port->bond_fake_iface
2128 && time_msec() >= port->bond_next_fake_iface_update) {
2129 bond_update_fake_iface_stats(port);
2130 port->bond_next_fake_iface_update = time_msec() + 1000;
2134 if (port->bond_compat_is_stale) {
2135 port->bond_compat_is_stale = false;
2136 port_update_bond_compat(port);
2142 bond_wait(struct bridge *br)
2146 for (i = 0; i < br->n_ports; i++) {
2147 struct port *port = br->ports[i];
2148 if (port->n_ifaces < 2) {
2151 netdev_monitor_poll_wait(port->monitor);
2152 for (j = 0; j < port->n_ifaces; j++) {
2153 struct iface *iface = port->ifaces[j];
2154 if (iface->delay_expires != LLONG_MAX) {
2155 poll_timer_wait_until(iface->delay_expires);
2158 if (port->bond_fake_iface) {
2159 poll_timer_wait_until(port->bond_next_fake_iface_update);
2165 set_dst(struct dst *p, const struct flow *flow,
2166 const struct port *in_port, const struct port *out_port,
2169 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2170 : in_port->vlan >= 0 ? in_port->vlan
2171 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2172 : vlan_tci_to_vid(flow->vlan_tci));
2173 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
2177 swap_dst(struct dst *p, struct dst *q)
2179 struct dst tmp = *p;
2184 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2185 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2186 * that we push to the datapath. We could in fact fully sort the array by
2187 * vlan, but in most cases there are at most two different vlan tags so that's
2188 * possibly overkill.) */
2190 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2192 struct dst *first = dsts;
2193 struct dst *last = dsts + n_dsts;
2195 while (first != last) {
2197 * - All dsts < first have vlan == 'vlan'.
2198 * - All dsts >= last have vlan != 'vlan'.
2199 * - first < last. */
2200 while (first->vlan == vlan) {
2201 if (++first == last) {
2206 /* Same invariants, plus one additional:
2207 * - first->vlan != vlan.
2209 while (last[-1].vlan != vlan) {
2210 if (--last == first) {
2215 /* Same invariants, plus one additional:
2216 * - last[-1].vlan == vlan.*/
2217 swap_dst(first++, --last);
2222 mirror_mask_ffs(mirror_mask_t mask)
2224 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2229 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2230 const struct dst *test)
2233 for (i = 0; i < n_dsts; i++) {
2234 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2242 port_trunks_vlan(const struct port *port, uint16_t vlan)
2244 return (port->vlan < 0
2245 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2249 port_includes_vlan(const struct port *port, uint16_t vlan)
2251 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2255 port_is_floodable(const struct port *port)
2259 for (i = 0; i < port->n_ifaces; i++) {
2260 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2261 port->ifaces[i]->dp_ifidx)) {
2269 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2270 const struct port *in_port, const struct port *out_port,
2271 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2273 mirror_mask_t mirrors = in_port->src_mirrors;
2275 struct dst *dst = dsts;
2278 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2279 if (flow_vlan == 0) {
2280 flow_vlan = OFP_VLAN_NONE;
2283 if (out_port == FLOOD_PORT) {
2284 /* XXX use ODP_FLOOD if no vlans or bonding. */
2285 /* XXX even better, define each VLAN as a datapath port group */
2286 for (i = 0; i < br->n_ports; i++) {
2287 struct port *port = br->ports[i];
2289 && port_is_floodable(port)
2290 && port_includes_vlan(port, vlan)
2291 && !port->is_mirror_output_port
2292 && set_dst(dst, flow, in_port, port, tags)) {
2293 mirrors |= port->dst_mirrors;
2297 *nf_output_iface = NF_OUT_FLOOD;
2298 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2299 *nf_output_iface = dst->dp_ifidx;
2300 mirrors |= out_port->dst_mirrors;
2305 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2306 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2308 if (set_dst(dst, flow, in_port, m->out_port, tags)
2309 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2313 for (i = 0; i < br->n_ports; i++) {
2314 struct port *port = br->ports[i];
2315 if (port_includes_vlan(port, m->out_vlan)
2316 && set_dst(dst, flow, in_port, port, tags))
2319 if (port->vlan < 0) {
2320 dst->vlan = m->out_vlan;
2322 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2326 /* Use the vlan tag on the original flow instead of
2327 * the one passed in the vlan parameter. This ensures
2328 * that we compare the vlan from before any implicit
2329 * tagging tags place. This is necessary because
2330 * dst->vlan is the final vlan, after removing implicit
2332 if (port == in_port && dst->vlan == flow_vlan) {
2333 /* Don't send out input port on same VLAN. */
2341 mirrors &= mirrors - 1;
2344 partition_dsts(dsts, dst - dsts, flow_vlan);
2348 static void OVS_UNUSED
2349 print_dsts(const struct dst *dsts, size_t n)
2351 for (; n--; dsts++) {
2352 printf(">p%"PRIu16, dsts->dp_ifidx);
2353 if (dsts->vlan != OFP_VLAN_NONE) {
2354 printf("v%"PRIu16, dsts->vlan);
2360 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2361 const struct port *in_port, const struct port *out_port,
2362 tag_type *tags, struct odp_actions *actions,
2363 uint16_t *nf_output_iface)
2365 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2367 const struct dst *p;
2370 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2373 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2374 if (cur_vlan == 0) {
2375 cur_vlan = OFP_VLAN_NONE;
2377 for (p = dsts; p < &dsts[n_dsts]; p++) {
2378 union odp_action *a;
2379 if (p->vlan != cur_vlan) {
2380 if (p->vlan == OFP_VLAN_NONE) {
2381 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2383 a = odp_actions_add(actions, ODPAT_SET_DL_TCI);
2384 a->dl_tci.tci = htons(p->vlan & VLAN_VID_MASK);
2385 a->dl_tci.tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2389 a = odp_actions_add(actions, ODPAT_OUTPUT);
2390 a->output.port = p->dp_ifidx;
2394 /* Returns the effective vlan of a packet, taking into account both the
2395 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2396 * the packet is untagged and -1 indicates it has an invalid header and
2397 * should be dropped. */
2398 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2399 struct port *in_port, bool have_packet)
2401 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2402 if (in_port->vlan >= 0) {
2404 /* XXX support double tagging? */
2406 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2407 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2408 "packet received on port %s configured with "
2409 "implicit VLAN %"PRIu16,
2410 br->name, vlan, in_port->name, in_port->vlan);
2414 vlan = in_port->vlan;
2416 if (!port_includes_vlan(in_port, vlan)) {
2418 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2419 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2420 "packet received on port %s not configured for "
2422 br->name, vlan, in_port->name, vlan);
2431 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2432 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2433 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2435 is_gratuitous_arp(const struct flow *flow)
2437 return (flow->dl_type == htons(ETH_TYPE_ARP)
2438 && eth_addr_is_broadcast(flow->dl_dst)
2439 && (flow->nw_proto == ARP_OP_REPLY
2440 || (flow->nw_proto == ARP_OP_REQUEST
2441 && flow->nw_src == flow->nw_dst)));
2445 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2446 struct port *in_port)
2448 enum grat_arp_lock_type lock_type;
2451 /* We don't want to learn from gratuitous ARP packets that are reflected
2452 * back over bond slaves so we lock the learning table. */
2453 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2454 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2455 GRAT_ARP_LOCK_CHECK;
2457 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2460 /* The log messages here could actually be useful in debugging,
2461 * so keep the rate limit relatively high. */
2462 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2464 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2465 "on port %s in VLAN %d",
2466 br->name, ETH_ADDR_ARGS(flow->dl_src),
2467 in_port->name, vlan);
2468 ofproto_revalidate(br->ofproto, rev_tag);
2472 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2473 * dropped. Returns true if they may be forwarded, false if they should be
2476 * If 'have_packet' is true, it indicates that the caller is processing a
2477 * received packet. If 'have_packet' is false, then the caller is just
2478 * revalidating an existing flow because configuration has changed. Either
2479 * way, 'have_packet' only affects logging (there is no point in logging errors
2480 * during revalidation).
2482 * Sets '*in_portp' to the input port. This will be a null pointer if
2483 * flow->in_port does not designate a known input port (in which case
2484 * is_admissible() returns false).
2486 * When returning true, sets '*vlanp' to the effective VLAN of the input
2487 * packet, as returned by flow_get_vlan().
2489 * May also add tags to '*tags', although the current implementation only does
2490 * so in one special case.
2493 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2494 tag_type *tags, int *vlanp, struct port **in_portp)
2496 struct iface *in_iface;
2497 struct port *in_port;
2500 /* Find the interface and port structure for the received packet. */
2501 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2503 /* No interface? Something fishy... */
2505 /* Odd. A few possible reasons here:
2507 * - We deleted an interface but there are still a few packets
2508 * queued up from it.
2510 * - Someone externally added an interface (e.g. with "ovs-dpctl
2511 * add-if") that we don't know about.
2513 * - Packet arrived on the local port but the local port is not
2514 * one of our bridge ports.
2516 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2518 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2519 "interface %"PRIu16, br->name, flow->in_port);
2525 *in_portp = in_port = in_iface->port;
2526 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2531 /* Drop frames for reserved multicast addresses. */
2532 if (eth_addr_is_reserved(flow->dl_dst)) {
2536 /* Drop frames on ports reserved for mirroring. */
2537 if (in_port->is_mirror_output_port) {
2539 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2540 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2541 "%s, which is reserved exclusively for mirroring",
2542 br->name, in_port->name);
2547 /* Packets received on bonds need special attention to avoid duplicates. */
2548 if (in_port->n_ifaces > 1) {
2550 bool is_grat_arp_locked;
2552 if (eth_addr_is_multicast(flow->dl_dst)) {
2553 *tags |= in_port->active_iface_tag;
2554 if (in_port->active_iface != in_iface->port_ifidx) {
2555 /* Drop all multicast packets on inactive slaves. */
2560 /* Drop all packets for which we have learned a different input
2561 * port, because we probably sent the packet on one slave and got
2562 * it back on the other. Gratuitous ARP packets are an exception
2563 * to this rule: the host has moved to another switch. The exception
2564 * to the exception is if we locked the learning table to avoid
2565 * reflections on bond slaves. If this is the case, just drop the
2567 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2568 &is_grat_arp_locked);
2569 if (src_idx != -1 && src_idx != in_port->port_idx &&
2570 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2578 /* If the composed actions may be applied to any packet in the given 'flow',
2579 * returns true. Otherwise, the actions should only be applied to 'packet', or
2580 * not at all, if 'packet' was NULL. */
2582 process_flow(struct bridge *br, const struct flow *flow,
2583 const struct ofpbuf *packet, struct odp_actions *actions,
2584 tag_type *tags, uint16_t *nf_output_iface)
2586 struct port *in_port;
2587 struct port *out_port;
2591 /* Check whether we should drop packets in this flow. */
2592 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2597 /* Learn source MAC (but don't try to learn from revalidation). */
2599 update_learning_table(br, flow, vlan, in_port);
2602 /* Determine output port. */
2603 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2605 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2606 out_port = br->ports[out_port_idx];
2607 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2608 /* If we are revalidating but don't have a learning entry then
2609 * eject the flow. Installing a flow that floods packets opens
2610 * up a window of time where we could learn from a packet reflected
2611 * on a bond and blackhole packets before the learning table is
2612 * updated to reflect the correct port. */
2615 out_port = FLOOD_PORT;
2618 /* Don't send packets out their input ports. */
2619 if (in_port == out_port) {
2625 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2633 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2634 struct odp_actions *actions, tag_type *tags,
2635 uint16_t *nf_output_iface, void *br_)
2637 struct bridge *br = br_;
2639 COVERAGE_INC(bridge_process_flow);
2641 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2645 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
2646 const union odp_action *actions,
2647 size_t n_actions, unsigned long long int n_bytes,
2650 struct bridge *br = br_;
2651 const union odp_action *a;
2652 struct port *in_port;
2656 /* Feed information from the active flows back into the learning table to
2657 * ensure that table is always in sync with what is actually flowing
2658 * through the datapath.
2660 * We test that 'tags' is nonzero to ensure that only flows that include an
2661 * OFPP_NORMAL action are used for learning. This works because
2662 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2663 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2664 update_learning_table(br, flow, vlan, in_port);
2667 /* Account for bond slave utilization. */
2668 if (!br->has_bonded_ports) {
2671 for (a = actions; a < &actions[n_actions]; a++) {
2672 if (a->type == ODPAT_OUTPUT) {
2673 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2674 if (out_port && out_port->n_ifaces >= 2) {
2675 struct bond_entry *e = lookup_bond_entry(out_port,
2677 e->tx_bytes += n_bytes;
2684 bridge_account_checkpoint_ofhook_cb(void *br_)
2686 struct bridge *br = br_;
2690 if (!br->has_bonded_ports) {
2695 for (i = 0; i < br->n_ports; i++) {
2696 struct port *port = br->ports[i];
2697 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2698 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2699 bond_rebalance_port(port);
2704 static struct ofhooks bridge_ofhooks = {
2705 bridge_normal_ofhook_cb,
2706 bridge_account_flow_ofhook_cb,
2707 bridge_account_checkpoint_ofhook_cb,
2710 /* Bonding functions. */
2712 /* Statistics for a single interface on a bonded port, used for load-based
2713 * bond rebalancing. */
2714 struct slave_balance {
2715 struct iface *iface; /* The interface. */
2716 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2718 /* All the "bond_entry"s that are assigned to this interface, in order of
2719 * increasing tx_bytes. */
2720 struct bond_entry **hashes;
2724 /* Sorts pointers to pointers to bond_entries in ascending order by the
2725 * interface to which they are assigned, and within a single interface in
2726 * ascending order of bytes transmitted. */
2728 compare_bond_entries(const void *a_, const void *b_)
2730 const struct bond_entry *const *ap = a_;
2731 const struct bond_entry *const *bp = b_;
2732 const struct bond_entry *a = *ap;
2733 const struct bond_entry *b = *bp;
2734 if (a->iface_idx != b->iface_idx) {
2735 return a->iface_idx > b->iface_idx ? 1 : -1;
2736 } else if (a->tx_bytes != b->tx_bytes) {
2737 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2743 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2744 * *descending* order by number of bytes transmitted. */
2746 compare_slave_balance(const void *a_, const void *b_)
2748 const struct slave_balance *a = a_;
2749 const struct slave_balance *b = b_;
2750 if (a->iface->enabled != b->iface->enabled) {
2751 return a->iface->enabled ? -1 : 1;
2752 } else if (a->tx_bytes != b->tx_bytes) {
2753 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2760 swap_bals(struct slave_balance *a, struct slave_balance *b)
2762 struct slave_balance tmp = *a;
2767 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2768 * given that 'p' (and only 'p') might be in the wrong location.
2770 * This function invalidates 'p', since it might now be in a different memory
2773 resort_bals(struct slave_balance *p,
2774 struct slave_balance bals[], size_t n_bals)
2777 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2778 swap_bals(p, p - 1);
2780 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2781 swap_bals(p, p + 1);
2787 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2789 if (VLOG_IS_DBG_ENABLED()) {
2790 struct ds ds = DS_EMPTY_INITIALIZER;
2791 const struct slave_balance *b;
2793 for (b = bals; b < bals + n_bals; b++) {
2797 ds_put_char(&ds, ',');
2799 ds_put_format(&ds, " %s %"PRIu64"kB",
2800 b->iface->name, b->tx_bytes / 1024);
2802 if (!b->iface->enabled) {
2803 ds_put_cstr(&ds, " (disabled)");
2805 if (b->n_hashes > 0) {
2806 ds_put_cstr(&ds, " (");
2807 for (i = 0; i < b->n_hashes; i++) {
2808 const struct bond_entry *e = b->hashes[i];
2810 ds_put_cstr(&ds, " + ");
2812 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2813 e - port->bond_hash, e->tx_bytes / 1024);
2815 ds_put_cstr(&ds, ")");
2818 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2823 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2825 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2828 struct bond_entry *hash = from->hashes[hash_idx];
2829 struct port *port = from->iface->port;
2830 uint64_t delta = hash->tx_bytes;
2832 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2833 "from %s to %s (now carrying %"PRIu64"kB and "
2834 "%"PRIu64"kB load, respectively)",
2835 port->name, delta / 1024, hash - port->bond_hash,
2836 from->iface->name, to->iface->name,
2837 (from->tx_bytes - delta) / 1024,
2838 (to->tx_bytes + delta) / 1024);
2840 /* Delete element from from->hashes.
2842 * We don't bother to add the element to to->hashes because not only would
2843 * it require more work, the only purpose it would be to allow that hash to
2844 * be migrated to another slave in this rebalancing run, and there is no
2845 * point in doing that. */
2846 if (hash_idx == 0) {
2849 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2850 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2854 /* Shift load away from 'from' to 'to'. */
2855 from->tx_bytes -= delta;
2856 to->tx_bytes += delta;
2858 /* Arrange for flows to be revalidated. */
2859 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2860 hash->iface_idx = to->iface->port_ifidx;
2861 hash->iface_tag = tag_create_random();
2865 bond_rebalance_port(struct port *port)
2867 struct slave_balance bals[DP_MAX_PORTS];
2869 struct bond_entry *hashes[BOND_MASK + 1];
2870 struct slave_balance *b, *from, *to;
2871 struct bond_entry *e;
2874 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2875 * descending order of tx_bytes, so that bals[0] represents the most
2876 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2879 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2880 * array for each slave_balance structure, we sort our local array of
2881 * hashes in order by slave, so that all of the hashes for a given slave
2882 * become contiguous in memory, and then we point each 'hashes' members of
2883 * a slave_balance structure to the start of a contiguous group. */
2884 n_bals = port->n_ifaces;
2885 for (b = bals; b < &bals[n_bals]; b++) {
2886 b->iface = port->ifaces[b - bals];
2891 for (i = 0; i <= BOND_MASK; i++) {
2892 hashes[i] = &port->bond_hash[i];
2894 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2895 for (i = 0; i <= BOND_MASK; i++) {
2897 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2898 b = &bals[e->iface_idx];
2899 b->tx_bytes += e->tx_bytes;
2901 b->hashes = &hashes[i];
2906 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2907 log_bals(bals, n_bals, port);
2909 /* Discard slaves that aren't enabled (which were sorted to the back of the
2910 * array earlier). */
2911 while (!bals[n_bals - 1].iface->enabled) {
2918 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2919 to = &bals[n_bals - 1];
2920 for (from = bals; from < to; ) {
2921 uint64_t overload = from->tx_bytes - to->tx_bytes;
2922 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2923 /* The extra load on 'from' (and all less-loaded slaves), compared
2924 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2925 * it is less than ~1Mbps. No point in rebalancing. */
2927 } else if (from->n_hashes == 1) {
2928 /* 'from' only carries a single MAC hash, so we can't shift any
2929 * load away from it, even though we want to. */
2932 /* 'from' is carrying significantly more load than 'to', and that
2933 * load is split across at least two different hashes. Pick a hash
2934 * to migrate to 'to' (the least-loaded slave), given that doing so
2935 * must decrease the ratio of the load on the two slaves by at
2938 * The sort order we use means that we prefer to shift away the
2939 * smallest hashes instead of the biggest ones. There is little
2940 * reason behind this decision; we could use the opposite sort
2941 * order to shift away big hashes ahead of small ones. */
2944 for (i = 0; i < from->n_hashes; i++) {
2945 double old_ratio, new_ratio;
2946 uint64_t delta = from->hashes[i]->tx_bytes;
2948 if (delta == 0 || from->tx_bytes - delta == 0) {
2949 /* Pointless move. */
2953 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2955 if (to->tx_bytes == 0) {
2956 /* Nothing on the new slave, move it. */
2960 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2961 new_ratio = (double)(from->tx_bytes - delta) /
2962 (to->tx_bytes + delta);
2964 if (new_ratio == 0) {
2965 /* Should already be covered but check to prevent division
2970 if (new_ratio < 1) {
2971 new_ratio = 1 / new_ratio;
2974 if (old_ratio - new_ratio > 0.1) {
2975 /* Would decrease the ratio, move it. */
2979 if (i < from->n_hashes) {
2980 bond_shift_load(from, to, i);
2981 port->bond_compat_is_stale = true;
2983 /* If the result of the migration changed the relative order of
2984 * 'from' and 'to' swap them back to maintain invariants. */
2985 if (order_swapped) {
2986 swap_bals(from, to);
2989 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2990 * point to different slave_balance structures. It is only
2991 * valid to do these two operations in a row at all because we
2992 * know that 'from' will not move past 'to' and vice versa. */
2993 resort_bals(from, bals, n_bals);
2994 resort_bals(to, bals, n_bals);
3001 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3002 * historical data to decay to <1% in 7 rebalancing runs. */
3003 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3009 bond_send_learning_packets(struct port *port)
3011 struct bridge *br = port->bridge;
3012 struct mac_entry *e;
3013 struct ofpbuf packet;
3014 int error, n_packets, n_errors;
3016 if (!port->n_ifaces || port->active_iface < 0) {
3020 ofpbuf_init(&packet, 128);
3021 error = n_packets = n_errors = 0;
3022 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3023 union ofp_action actions[2], *a;
3029 if (e->port == port->port_idx
3030 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
3034 /* Compose actions. */
3035 memset(actions, 0, sizeof actions);
3038 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3039 a->vlan_vid.len = htons(sizeof *a);
3040 a->vlan_vid.vlan_vid = htons(e->vlan);
3043 a->output.type = htons(OFPAT_OUTPUT);
3044 a->output.len = htons(sizeof *a);
3045 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3050 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3052 flow_extract(&packet, 0, ODPP_NONE, &flow);
3053 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3060 ofpbuf_uninit(&packet);
3063 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3064 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3065 "packets, last error was: %s",
3066 port->name, n_errors, n_packets, strerror(error));
3068 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3069 port->name, n_packets);
3073 /* Bonding unixctl user interface functions. */
3076 bond_unixctl_list(struct unixctl_conn *conn,
3077 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3079 struct ds ds = DS_EMPTY_INITIALIZER;
3080 const struct bridge *br;
3082 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
3084 LIST_FOR_EACH (br, node, &all_bridges) {
3087 for (i = 0; i < br->n_ports; i++) {
3088 const struct port *port = br->ports[i];
3089 if (port->n_ifaces > 1) {
3092 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
3093 for (j = 0; j < port->n_ifaces; j++) {
3094 const struct iface *iface = port->ifaces[j];
3096 ds_put_cstr(&ds, ", ");
3098 ds_put_cstr(&ds, iface->name);
3100 ds_put_char(&ds, '\n');
3104 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3108 static struct port *
3109 bond_find(const char *name)
3111 const struct bridge *br;
3113 LIST_FOR_EACH (br, node, &all_bridges) {
3116 for (i = 0; i < br->n_ports; i++) {
3117 struct port *port = br->ports[i];
3118 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3127 bond_unixctl_show(struct unixctl_conn *conn,
3128 const char *args, void *aux OVS_UNUSED)
3130 struct ds ds = DS_EMPTY_INITIALIZER;
3131 const struct port *port;
3134 port = bond_find(args);
3136 unixctl_command_reply(conn, 501, "no such bond");
3140 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3141 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3142 ds_put_format(&ds, "next rebalance: %lld ms\n",
3143 port->bond_next_rebalance - time_msec());
3144 for (j = 0; j < port->n_ifaces; j++) {
3145 const struct iface *iface = port->ifaces[j];
3146 struct bond_entry *be;
3149 ds_put_format(&ds, "slave %s: %s\n",
3150 iface->name, iface->enabled ? "enabled" : "disabled");
3151 if (j == port->active_iface) {
3152 ds_put_cstr(&ds, "\tactive slave\n");
3154 if (iface->delay_expires != LLONG_MAX) {
3155 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3156 iface->enabled ? "downdelay" : "updelay",
3157 iface->delay_expires - time_msec());
3161 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3162 int hash = be - port->bond_hash;
3163 struct mac_entry *me;
3165 if (be->iface_idx != j) {
3169 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3170 hash, be->tx_bytes / 1024);
3173 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3176 if (bond_hash(me->mac) == hash
3177 && me->port != port->port_idx
3178 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
3179 && dp_ifidx == iface->dp_ifidx)
3181 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3182 ETH_ADDR_ARGS(me->mac));
3187 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3192 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3193 void *aux OVS_UNUSED)
3195 char *args = (char *) args_;
3196 char *save_ptr = NULL;
3197 char *bond_s, *hash_s, *slave_s;
3198 uint8_t mac[ETH_ADDR_LEN];
3200 struct iface *iface;
3201 struct bond_entry *entry;
3204 bond_s = strtok_r(args, " ", &save_ptr);
3205 hash_s = strtok_r(NULL, " ", &save_ptr);
3206 slave_s = strtok_r(NULL, " ", &save_ptr);
3208 unixctl_command_reply(conn, 501,
3209 "usage: bond/migrate BOND HASH SLAVE");
3213 port = bond_find(bond_s);
3215 unixctl_command_reply(conn, 501, "no such bond");
3219 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3220 == ETH_ADDR_SCAN_COUNT) {
3221 hash = bond_hash(mac);
3222 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3223 hash = atoi(hash_s) & BOND_MASK;
3225 unixctl_command_reply(conn, 501, "bad hash");
3229 iface = port_lookup_iface(port, slave_s);
3231 unixctl_command_reply(conn, 501, "no such slave");
3235 if (!iface->enabled) {
3236 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3240 entry = &port->bond_hash[hash];
3241 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3242 entry->iface_idx = iface->port_ifidx;
3243 entry->iface_tag = tag_create_random();
3244 port->bond_compat_is_stale = true;
3245 unixctl_command_reply(conn, 200, "migrated");
3249 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3250 void *aux OVS_UNUSED)
3252 char *args = (char *) args_;
3253 char *save_ptr = NULL;
3254 char *bond_s, *slave_s;
3256 struct iface *iface;
3258 bond_s = strtok_r(args, " ", &save_ptr);
3259 slave_s = strtok_r(NULL, " ", &save_ptr);
3261 unixctl_command_reply(conn, 501,
3262 "usage: bond/set-active-slave BOND SLAVE");
3266 port = bond_find(bond_s);
3268 unixctl_command_reply(conn, 501, "no such bond");
3272 iface = port_lookup_iface(port, slave_s);
3274 unixctl_command_reply(conn, 501, "no such slave");
3278 if (!iface->enabled) {
3279 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3283 if (port->active_iface != iface->port_ifidx) {
3284 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3285 port->active_iface = iface->port_ifidx;
3286 port->active_iface_tag = tag_create_random();
3287 VLOG_INFO("port %s: active interface is now %s",
3288 port->name, iface->name);
3289 bond_send_learning_packets(port);
3290 unixctl_command_reply(conn, 200, "done");
3292 unixctl_command_reply(conn, 200, "no change");
3297 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3299 char *args = (char *) args_;
3300 char *save_ptr = NULL;
3301 char *bond_s, *slave_s;
3303 struct iface *iface;
3305 bond_s = strtok_r(args, " ", &save_ptr);
3306 slave_s = strtok_r(NULL, " ", &save_ptr);
3308 unixctl_command_reply(conn, 501,
3309 "usage: bond/enable/disable-slave BOND SLAVE");
3313 port = bond_find(bond_s);
3315 unixctl_command_reply(conn, 501, "no such bond");
3319 iface = port_lookup_iface(port, slave_s);
3321 unixctl_command_reply(conn, 501, "no such slave");
3325 bond_enable_slave(iface, enable);
3326 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3330 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3331 void *aux OVS_UNUSED)
3333 enable_slave(conn, args, true);
3337 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3338 void *aux OVS_UNUSED)
3340 enable_slave(conn, args, false);
3344 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3345 void *aux OVS_UNUSED)
3347 uint8_t mac[ETH_ADDR_LEN];
3351 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3352 == ETH_ADDR_SCAN_COUNT) {
3353 hash = bond_hash(mac);
3355 hash_cstr = xasprintf("%u", hash);
3356 unixctl_command_reply(conn, 200, hash_cstr);
3359 unixctl_command_reply(conn, 501, "invalid mac");
3366 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3367 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3368 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3369 unixctl_command_register("bond/set-active-slave",
3370 bond_unixctl_set_active_slave, NULL);
3371 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3373 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3375 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3378 /* Port functions. */
3380 static struct port *
3381 port_create(struct bridge *br, const char *name)
3385 port = xzalloc(sizeof *port);
3387 port->port_idx = br->n_ports;
3389 port->trunks = NULL;
3390 port->name = xstrdup(name);
3391 port->active_iface = -1;
3393 if (br->n_ports >= br->allocated_ports) {
3394 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3397 br->ports[br->n_ports++] = port;
3398 shash_add_assert(&br->port_by_name, port->name, port);
3400 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3407 get_port_other_config(const struct ovsrec_port *port, const char *key,
3408 const char *default_value)
3412 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3414 return value ? value : default_value;
3418 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3420 struct shash new_ifaces;
3423 /* Collect list of new interfaces. */
3424 shash_init(&new_ifaces);
3425 for (i = 0; i < cfg->n_interfaces; i++) {
3426 const char *name = cfg->interfaces[i]->name;
3427 shash_add_once(&new_ifaces, name, NULL);
3430 /* Get rid of deleted interfaces. */
3431 for (i = 0; i < port->n_ifaces; ) {
3432 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3433 iface_destroy(port->ifaces[i]);
3439 shash_destroy(&new_ifaces);
3443 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3445 struct shash new_ifaces;
3446 long long int next_rebalance;
3447 unsigned long *trunks;
3453 /* Update settings. */
3454 port->updelay = cfg->bond_updelay;
3455 if (port->updelay < 0) {
3458 port->downdelay = cfg->bond_downdelay;
3459 if (port->downdelay < 0) {
3460 port->downdelay = 0;
3462 port->bond_rebalance_interval = atoi(
3463 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3464 if (port->bond_rebalance_interval < 1000) {
3465 port->bond_rebalance_interval = 1000;
3467 next_rebalance = time_msec() + port->bond_rebalance_interval;
3468 if (port->bond_next_rebalance > next_rebalance) {
3469 port->bond_next_rebalance = next_rebalance;
3472 /* Add new interfaces and update 'cfg' member of existing ones. */
3473 shash_init(&new_ifaces);
3474 for (i = 0; i < cfg->n_interfaces; i++) {
3475 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3476 struct iface *iface;
3478 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3479 VLOG_WARN("port %s: %s specified twice as port interface",
3480 port->name, if_cfg->name);
3481 iface_set_ofport(if_cfg, -1);
3485 iface = iface_lookup(port->bridge, if_cfg->name);
3487 if (iface->port != port) {
3488 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3490 port->bridge->name, if_cfg->name, iface->port->name);
3493 iface->cfg = if_cfg;
3495 iface = iface_create(port, if_cfg);
3498 /* Determine interface type. The local port always has type
3499 * "internal". Other ports take their type from the database and
3500 * default to "system" if none is specified. */
3501 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
3502 : if_cfg->type[0] ? if_cfg->type
3505 shash_destroy(&new_ifaces);
3510 if (port->n_ifaces < 2) {
3512 if (vlan >= 0 && vlan <= 4095) {
3513 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3518 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3519 * they even work as-is. But they have not been tested. */
3520 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3524 if (port->vlan != vlan) {
3526 bridge_flush(port->bridge);
3529 /* Get trunked VLANs. */
3531 if (vlan < 0 && cfg->n_trunks) {
3534 trunks = bitmap_allocate(4096);
3536 for (i = 0; i < cfg->n_trunks; i++) {
3537 int trunk = cfg->trunks[i];
3539 bitmap_set1(trunks, trunk);
3545 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3546 port->name, cfg->n_trunks);
3548 if (n_errors == cfg->n_trunks) {
3549 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3551 bitmap_free(trunks);
3554 } else if (vlan >= 0 && cfg->n_trunks) {
3555 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3559 ? port->trunks != NULL
3560 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3561 bridge_flush(port->bridge);
3563 bitmap_free(port->trunks);
3564 port->trunks = trunks;
3568 port_destroy(struct port *port)
3571 struct bridge *br = port->bridge;
3575 proc_net_compat_update_vlan(port->name, NULL, 0);
3576 proc_net_compat_update_bond(port->name, NULL);
3578 for (i = 0; i < MAX_MIRRORS; i++) {
3579 struct mirror *m = br->mirrors[i];
3580 if (m && m->out_port == port) {
3585 while (port->n_ifaces > 0) {
3586 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3589 shash_find_and_delete_assert(&br->port_by_name, port->name);
3591 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3592 del->port_idx = port->port_idx;
3594 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
3596 netdev_monitor_destroy(port->monitor);
3598 bitmap_free(port->trunks);
3605 static struct port *
3606 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3608 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3609 return iface ? iface->port : NULL;
3612 static struct port *
3613 port_lookup(const struct bridge *br, const char *name)
3615 return shash_find_data(&br->port_by_name, name);
3618 static struct iface *
3619 port_lookup_iface(const struct port *port, const char *name)
3621 struct iface *iface = iface_lookup(port->bridge, name);
3622 return iface && iface->port == port ? iface : NULL;
3626 port_update_bonding(struct port *port)
3628 if (port->monitor) {
3629 netdev_monitor_destroy(port->monitor);
3630 port->monitor = NULL;
3632 if (port->n_ifaces < 2) {
3633 /* Not a bonded port. */
3634 if (port->bond_hash) {
3635 free(port->bond_hash);
3636 port->bond_hash = NULL;
3637 port->bond_compat_is_stale = true;
3638 port->bond_fake_iface = false;
3643 if (!port->bond_hash) {
3644 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3645 for (i = 0; i <= BOND_MASK; i++) {
3646 struct bond_entry *e = &port->bond_hash[i];
3650 port->no_ifaces_tag = tag_create_random();
3651 bond_choose_active_iface(port);
3652 port->bond_next_rebalance
3653 = time_msec() + port->bond_rebalance_interval;
3655 if (port->cfg->bond_fake_iface) {
3656 port->bond_next_fake_iface_update = time_msec();
3659 port->bond_compat_is_stale = true;
3660 port->bond_fake_iface = port->cfg->bond_fake_iface;
3662 port->monitor = netdev_monitor_create();
3663 for (i = 0; i < port->n_ifaces; i++) {
3664 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
3670 port_update_bond_compat(struct port *port)
3672 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3673 struct compat_bond bond;
3676 if (port->n_ifaces < 2) {
3677 proc_net_compat_update_bond(port->name, NULL);
3682 bond.updelay = port->updelay;
3683 bond.downdelay = port->downdelay;
3686 bond.hashes = compat_hashes;
3687 if (port->bond_hash) {
3688 const struct bond_entry *e;
3689 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3690 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3691 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3692 cbh->hash = e - port->bond_hash;
3693 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3698 bond.n_slaves = port->n_ifaces;
3699 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3700 for (i = 0; i < port->n_ifaces; i++) {
3701 struct iface *iface = port->ifaces[i];
3702 struct compat_bond_slave *slave = &bond.slaves[i];
3703 slave->name = iface->name;
3705 /* We need to make the same determination as the Linux bonding
3706 * code to determine whether a slave should be consider "up".
3707 * The Linux function bond_miimon_inspect() supports four
3708 * BOND_LINK_* states:
3710 * - BOND_LINK_UP: carrier detected, updelay has passed.
3711 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3712 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3713 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3715 * The function bond_info_show_slave() only considers BOND_LINK_UP
3716 * to be "up" and anything else to be "down".
3718 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3722 netdev_get_etheraddr(iface->netdev, slave->mac);
3725 if (port->bond_fake_iface) {
3726 struct netdev *bond_netdev;
3728 if (!netdev_open_default(port->name, &bond_netdev)) {
3730 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3732 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3734 netdev_close(bond_netdev);
3738 proc_net_compat_update_bond(port->name, &bond);
3743 port_update_vlan_compat(struct port *port)
3745 struct bridge *br = port->bridge;
3746 char *vlandev_name = NULL;
3748 if (port->vlan > 0) {
3749 /* Figure out the name that the VLAN device should actually have, if it
3750 * existed. This takes some work because the VLAN device would not
3751 * have port->name in its name; rather, it would have the trunk port's
3752 * name, and 'port' would be attached to a bridge that also had the
3753 * VLAN device one of its ports. So we need to find a trunk port that
3754 * includes port->vlan.
3756 * There might be more than one candidate. This doesn't happen on
3757 * XenServer, so if it happens we just pick the first choice in
3758 * alphabetical order instead of creating multiple VLAN devices. */
3760 for (i = 0; i < br->n_ports; i++) {
3761 struct port *p = br->ports[i];
3762 if (port_trunks_vlan(p, port->vlan)
3764 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3766 uint8_t ea[ETH_ADDR_LEN];
3767 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3768 if (!eth_addr_is_multicast(ea) &&
3769 !eth_addr_is_reserved(ea) &&
3770 !eth_addr_is_zero(ea)) {
3771 vlandev_name = p->name;
3776 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3779 /* Interface functions. */
3781 static struct iface *
3782 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3784 struct bridge *br = port->bridge;
3785 struct iface *iface;
3786 char *name = if_cfg->name;
3788 iface = xzalloc(sizeof *iface);
3790 iface->port_ifidx = port->n_ifaces;
3791 iface->name = xstrdup(name);
3792 iface->dp_ifidx = -1;
3793 iface->tag = tag_create_random();
3794 iface->delay_expires = LLONG_MAX;
3795 iface->netdev = NULL;
3796 iface->cfg = if_cfg;
3798 shash_add_assert(&br->iface_by_name, iface->name, iface);
3800 if (port->n_ifaces >= port->allocated_ifaces) {
3801 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3802 sizeof *port->ifaces);
3804 port->ifaces[port->n_ifaces++] = iface;
3805 if (port->n_ifaces > 1) {
3806 br->has_bonded_ports = true;
3809 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3817 iface_destroy(struct iface *iface)
3820 struct port *port = iface->port;
3821 struct bridge *br = port->bridge;
3822 bool del_active = port->active_iface == iface->port_ifidx;
3825 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3827 if (iface->dp_ifidx >= 0) {
3828 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
3831 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3832 del->port_ifidx = iface->port_ifidx;
3834 netdev_close(iface->netdev);
3837 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3838 bond_choose_active_iface(port);
3839 bond_send_learning_packets(port);
3845 bridge_flush(port->bridge);
3849 static struct iface *
3850 iface_lookup(const struct bridge *br, const char *name)
3852 return shash_find_data(&br->iface_by_name, name);
3855 static struct iface *
3856 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3858 struct iface *iface;
3860 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
3861 hash_int(dp_ifidx, 0), &br->ifaces) {
3862 if (iface->dp_ifidx == dp_ifidx) {
3869 /* Set Ethernet address of 'iface', if one is specified in the configuration
3872 iface_set_mac(struct iface *iface)
3874 uint8_t ea[ETH_ADDR_LEN];
3876 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3877 if (eth_addr_is_multicast(ea)) {
3878 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3880 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3881 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3882 iface->name, iface->name);
3884 int error = netdev_set_etheraddr(iface->netdev, ea);
3886 VLOG_ERR("interface %s: setting MAC failed (%s)",
3887 iface->name, strerror(error));
3893 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
3895 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
3898 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
3902 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
3904 * The value strings in '*shash' are taken directly from values[], not copied,
3905 * so the caller should not modify or free them. */
3907 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3908 struct shash *shash)
3913 for (i = 0; i < n; i++) {
3914 shash_add(shash, keys[i], values[i]);
3918 struct iface_delete_queues_cbdata {
3919 struct netdev *netdev;
3920 const struct ovsdb_datum *queues;
3924 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
3926 union ovsdb_atom atom;
3928 atom.integer = target;
3929 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
3933 iface_delete_queues(unsigned int queue_id,
3934 const struct shash *details OVS_UNUSED, void *cbdata_)
3936 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3938 if (!queue_ids_include(cbdata->queues, queue_id)) {
3939 netdev_delete_queue(cbdata->netdev, queue_id);
3944 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3946 if (!qos || qos->type[0] == '\0') {
3947 netdev_set_qos(iface->netdev, NULL, NULL);
3949 struct iface_delete_queues_cbdata cbdata;
3950 struct shash details;
3953 /* Configure top-level Qos for 'iface'. */
3954 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3955 qos->n_other_config, &details);
3956 netdev_set_qos(iface->netdev, qos->type, &details);
3957 shash_destroy(&details);
3959 /* Deconfigure queues that were deleted. */
3960 cbdata.netdev = iface->netdev;
3961 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
3963 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3965 /* Configure queues for 'iface'. */
3966 for (i = 0; i < qos->n_queues; i++) {
3967 const struct ovsrec_queue *queue = qos->value_queues[i];
3968 unsigned int queue_id = qos->key_queues[i];
3970 shash_from_ovs_idl_map(queue->key_other_config,
3971 queue->value_other_config,
3972 queue->n_other_config, &details);
3973 netdev_set_queue(iface->netdev, queue_id, &details);
3974 shash_destroy(&details);
3979 /* Port mirroring. */
3981 static struct mirror *
3982 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
3986 for (i = 0; i < MAX_MIRRORS; i++) {
3987 struct mirror *m = br->mirrors[i];
3988 if (m && uuid_equals(uuid, &m->uuid)) {
3996 mirror_reconfigure(struct bridge *br)
3998 unsigned long *rspan_vlans;
4001 /* Get rid of deleted mirrors. */
4002 for (i = 0; i < MAX_MIRRORS; i++) {
4003 struct mirror *m = br->mirrors[i];
4005 const struct ovsdb_datum *mc;
4006 union ovsdb_atom atom;
4008 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4009 atom.uuid = br->mirrors[i]->uuid;
4010 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4016 /* Add new mirrors and reconfigure existing ones. */
4017 for (i = 0; i < br->cfg->n_mirrors; i++) {
4018 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4019 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4021 mirror_reconfigure_one(m, cfg);
4023 mirror_create(br, cfg);
4027 /* Update port reserved status. */
4028 for (i = 0; i < br->n_ports; i++) {
4029 br->ports[i]->is_mirror_output_port = false;
4031 for (i = 0; i < MAX_MIRRORS; i++) {
4032 struct mirror *m = br->mirrors[i];
4033 if (m && m->out_port) {
4034 m->out_port->is_mirror_output_port = true;
4038 /* Update flooded vlans (for RSPAN). */
4040 if (br->cfg->n_flood_vlans) {
4041 rspan_vlans = bitmap_allocate(4096);
4043 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4044 int64_t vlan = br->cfg->flood_vlans[i];
4045 if (vlan >= 0 && vlan < 4096) {
4046 bitmap_set1(rspan_vlans, vlan);
4047 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4050 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4055 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4061 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4066 for (i = 0; ; i++) {
4067 if (i >= MAX_MIRRORS) {
4068 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4069 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4072 if (!br->mirrors[i]) {
4077 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4080 br->mirrors[i] = m = xzalloc(sizeof *m);
4083 m->name = xstrdup(cfg->name);
4084 shash_init(&m->src_ports);
4085 shash_init(&m->dst_ports);
4091 mirror_reconfigure_one(m, cfg);
4095 mirror_destroy(struct mirror *m)
4098 struct bridge *br = m->bridge;
4101 for (i = 0; i < br->n_ports; i++) {
4102 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4103 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4106 shash_destroy(&m->src_ports);
4107 shash_destroy(&m->dst_ports);
4110 m->bridge->mirrors[m->idx] = NULL;
4119 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4120 struct shash *names)
4124 for (i = 0; i < n_ports; i++) {
4125 const char *name = ports[i]->name;
4126 if (port_lookup(m->bridge, name)) {
4127 shash_add_once(names, name, NULL);
4129 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4130 "port %s", m->bridge->name, m->name, name);
4136 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4142 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4144 for (i = 0; i < cfg->n_select_vlan; i++) {
4145 int64_t vlan = cfg->select_vlan[i];
4146 if (vlan < 0 || vlan > 4095) {
4147 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4148 m->bridge->name, m->name, vlan);
4150 (*vlans)[n_vlans++] = vlan;
4157 vlan_is_mirrored(const struct mirror *m, int vlan)
4161 for (i = 0; i < m->n_vlans; i++) {
4162 if (m->vlans[i] == vlan) {
4170 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4174 for (i = 0; i < m->n_vlans; i++) {
4175 if (port_trunks_vlan(p, m->vlans[i])) {
4183 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4185 struct shash src_ports, dst_ports;
4186 mirror_mask_t mirror_bit;
4187 struct port *out_port;
4194 if (strcmp(cfg->name, m->name)) {
4196 m->name = xstrdup(cfg->name);
4199 /* Get output port. */
4200 if (cfg->output_port) {
4201 out_port = port_lookup(m->bridge, cfg->output_port->name);
4203 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4204 m->bridge->name, m->name);
4210 if (cfg->output_vlan) {
4211 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4212 "output vlan; ignoring output vlan",
4213 m->bridge->name, m->name);
4215 } else if (cfg->output_vlan) {
4217 out_vlan = *cfg->output_vlan;
4219 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4220 m->bridge->name, m->name);
4225 shash_init(&src_ports);
4226 shash_init(&dst_ports);
4227 if (cfg->select_all) {
4228 for (i = 0; i < m->bridge->n_ports; i++) {
4229 const char *name = m->bridge->ports[i]->name;
4230 shash_add_once(&src_ports, name, NULL);
4231 shash_add_once(&dst_ports, name, NULL);
4236 /* Get ports, and drop duplicates and ports that don't exist. */
4237 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4239 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4242 /* Get all the vlans, and drop duplicate and invalid vlans. */
4243 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4246 /* Update mirror data. */
4247 if (!shash_equal_keys(&m->src_ports, &src_ports)
4248 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4249 || m->n_vlans != n_vlans
4250 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4251 || m->out_port != out_port
4252 || m->out_vlan != out_vlan) {
4253 bridge_flush(m->bridge);
4255 shash_swap(&m->src_ports, &src_ports);
4256 shash_swap(&m->dst_ports, &dst_ports);
4259 m->n_vlans = n_vlans;
4260 m->out_port = out_port;
4261 m->out_vlan = out_vlan;
4264 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4265 for (i = 0; i < m->bridge->n_ports; i++) {
4266 struct port *port = m->bridge->ports[i];
4268 if (shash_find(&m->src_ports, port->name)
4271 ? port_trunks_any_mirrored_vlan(m, port)
4272 : vlan_is_mirrored(m, port->vlan)))) {
4273 port->src_mirrors |= mirror_bit;
4275 port->src_mirrors &= ~mirror_bit;
4278 if (shash_find(&m->dst_ports, port->name)) {
4279 port->dst_mirrors |= mirror_bit;
4281 port->dst_mirrors &= ~mirror_bit;
4286 shash_destroy(&src_ports);
4287 shash_destroy(&dst_ports);