1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
18 #include "byte-order.h"
21 #include <arpa/inet.h>
24 #include <sys/socket.h>
26 #include <openflow/openflow.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
36 #include "classifier.h"
40 #include "dynamic-string.h"
46 #include "mac-learning.h"
50 #include "ofp-print.h"
52 #include "ofproto/netflow.h"
53 #include "ofproto/ofproto.h"
54 #include "ovsdb-data.h"
56 #include "poll-loop.h"
57 #include "proc-net-compat.h"
61 #include "socket-util.h"
62 #include "stream-ssl.h"
64 #include "system-stats.h"
69 #include "vswitchd/vswitch-idl.h"
70 #include "xenserver.h"
72 #include "sflow_api.h"
74 VLOG_DEFINE_THIS_MODULE(bridge);
76 COVERAGE_DEFINE(bridge_flush);
77 COVERAGE_DEFINE(bridge_process_flow);
78 COVERAGE_DEFINE(bridge_reconfigure);
86 /* These members are always valid. */
87 struct port *port; /* Containing port. */
88 size_t port_ifidx; /* Index within containing port. */
89 char *name; /* Host network device name. */
90 tag_type tag; /* Tag associated with this interface. */
91 long long delay_expires; /* Time after which 'enabled' may change. */
93 /* These members are valid only after bridge_reconfigure() causes them to
95 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
96 int dp_ifidx; /* Index within kernel datapath. */
97 struct netdev *netdev; /* Network device. */
98 bool enabled; /* May be chosen for flows? */
99 const char *type; /* Usually same as cfg->type. */
100 struct cfm *cfm; /* Connectivity Fault Management */
101 const struct ovsrec_interface *cfg;
104 #define BOND_MASK 0xff
106 int iface_idx; /* Index of assigned iface, or -1 if none. */
107 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
108 tag_type iface_tag; /* Tag associated with iface_idx. */
111 #define MAX_MIRRORS 32
112 typedef uint32_t mirror_mask_t;
113 #define MIRROR_MASK_C(X) UINT32_C(X)
114 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
116 struct bridge *bridge;
119 struct uuid uuid; /* UUID of this "mirror" record in database. */
121 /* Selection criteria. */
122 struct shash src_ports; /* Name is port name; data is always NULL. */
123 struct shash dst_ports; /* Name is port name; data is always NULL. */
128 struct port *out_port;
132 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
134 struct bridge *bridge;
136 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
137 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
138 * NULL if all VLANs are trunked. */
139 const struct ovsrec_port *cfg;
142 /* An ordinary bridge port has 1 interface.
143 * A bridge port for bonding has at least 2 interfaces. */
144 struct iface **ifaces;
145 size_t n_ifaces, allocated_ifaces;
148 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
149 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
150 tag_type active_iface_tag; /* Tag for bcast flows. */
151 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
152 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
153 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
154 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
155 long long int bond_next_fake_iface_update; /* Time of next update. */
156 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
157 long long int bond_next_rebalance; /* Next rebalancing time. */
158 struct netdev_monitor *monitor; /* Tracks carrier up/down status. */
160 /* Port mirroring info. */
161 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
162 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
163 bool is_mirror_output_port; /* Does port mirroring send frames here? */
166 #define DP_MAX_PORTS 255
168 struct list node; /* Node in global list of bridges. */
169 char *name; /* User-specified arbitrary name. */
170 struct mac_learning *ml; /* MAC learning table. */
171 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
172 const struct ovsrec_bridge *cfg;
174 /* OpenFlow switch processing. */
175 struct ofproto *ofproto; /* OpenFlow switch. */
177 /* Kernel datapath information. */
178 struct dpif *dpif; /* Datapath. */
179 struct hmap ifaces; /* Contains "struct iface"s. */
183 size_t n_ports, allocated_ports;
184 struct shash iface_by_name; /* "struct iface"s indexed by name. */
185 struct shash port_by_name; /* "struct port"s indexed by name. */
188 bool has_bonded_ports;
193 /* Port mirroring. */
194 struct mirror *mirrors[MAX_MIRRORS];
197 /* List of all bridges. */
198 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
200 /* OVSDB IDL used to obtain configuration. */
201 static struct ovsdb_idl *idl;
203 /* Each time this timer expires, the bridge fetches systems and interface
204 * statistics and pushes them into the database. */
205 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
206 static long long int stats_timer = LLONG_MIN;
208 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
209 static void bridge_destroy(struct bridge *);
210 static struct bridge *bridge_lookup(const char *name);
211 static unixctl_cb_func bridge_unixctl_dump_flows;
212 static unixctl_cb_func bridge_unixctl_reconnect;
213 static int bridge_run_one(struct bridge *);
214 static size_t bridge_get_controllers(const struct bridge *br,
215 struct ovsrec_controller ***controllersp);
216 static void bridge_reconfigure_one(struct bridge *);
217 static void bridge_reconfigure_remotes(struct bridge *,
218 const struct sockaddr_in *managers,
220 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
221 static void bridge_fetch_dp_ifaces(struct bridge *);
222 static void bridge_flush(struct bridge *);
223 static void bridge_pick_local_hw_addr(struct bridge *,
224 uint8_t ea[ETH_ADDR_LEN],
225 struct iface **hw_addr_iface);
226 static uint64_t bridge_pick_datapath_id(struct bridge *,
227 const uint8_t bridge_ea[ETH_ADDR_LEN],
228 struct iface *hw_addr_iface);
229 static struct iface *bridge_get_local_iface(struct bridge *);
230 static uint64_t dpid_from_hash(const void *, size_t nbytes);
232 static unixctl_cb_func bridge_unixctl_fdb_show;
234 static void bond_init(void);
235 static void bond_run(struct bridge *);
236 static void bond_wait(struct bridge *);
237 static void bond_rebalance_port(struct port *);
238 static void bond_send_learning_packets(struct port *);
239 static void bond_enable_slave(struct iface *iface, bool enable);
241 static struct port *port_create(struct bridge *, const char *name);
242 static void port_reconfigure(struct port *, const struct ovsrec_port *);
243 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
244 static void port_destroy(struct port *);
245 static struct port *port_lookup(const struct bridge *, const char *name);
246 static struct iface *port_lookup_iface(const struct port *, const char *name);
247 static struct port *port_from_dp_ifidx(const struct bridge *,
249 static void port_update_bond_compat(struct port *);
250 static void port_update_vlan_compat(struct port *);
251 static void port_update_bonding(struct port *);
253 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
254 static void mirror_destroy(struct mirror *);
255 static void mirror_reconfigure(struct bridge *);
256 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
257 static bool vlan_is_mirrored(const struct mirror *, int vlan);
259 static struct iface *iface_create(struct port *port,
260 const struct ovsrec_interface *if_cfg);
261 static void iface_destroy(struct iface *);
262 static struct iface *iface_lookup(const struct bridge *, const char *name);
263 static struct iface *iface_from_dp_ifidx(const struct bridge *,
265 static void iface_set_mac(struct iface *);
266 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
267 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
268 static void iface_update_cfm(struct iface *);
269 static void iface_refresh_cfm_stats(struct iface *iface);
270 static void iface_send_packet(struct iface *, struct ofpbuf *packet);
272 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
275 /* Hooks into ofproto processing. */
276 static struct ofhooks bridge_ofhooks;
278 /* Public functions. */
280 /* Initializes the bridge module, configuring it to obtain its configuration
281 * from an OVSDB server accessed over 'remote', which should be a string in a
282 * form acceptable to ovsdb_idl_create(). */
284 bridge_init(const char *remote)
286 /* Create connection to database. */
287 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
289 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
290 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
291 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
293 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
295 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
296 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
298 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
299 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
300 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
302 /* Register unixctl commands. */
303 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
304 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
306 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
311 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
312 * but for which the ovs-vswitchd configuration 'cfg' is required. */
314 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
316 static bool already_configured_once;
317 struct svec bridge_names;
318 struct svec dpif_names, dpif_types;
321 /* Only do this once per ovs-vswitchd run. */
322 if (already_configured_once) {
325 already_configured_once = true;
327 stats_timer = time_msec() + STATS_INTERVAL;
329 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
330 svec_init(&bridge_names);
331 for (i = 0; i < cfg->n_bridges; i++) {
332 svec_add(&bridge_names, cfg->bridges[i]->name);
334 svec_sort(&bridge_names);
336 /* Iterate over all system dpifs and delete any of them that do not appear
338 svec_init(&dpif_names);
339 svec_init(&dpif_types);
340 dp_enumerate_types(&dpif_types);
341 for (i = 0; i < dpif_types.n; i++) {
346 dp_enumerate_names(dpif_types.names[i], &dpif_names);
348 /* For each dpif... */
349 for (j = 0; j < dpif_names.n; j++) {
350 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
352 struct svec all_names;
355 /* ...check whether any of its names is in 'bridge_names'. */
356 svec_init(&all_names);
357 dpif_get_all_names(dpif, &all_names);
358 for (k = 0; k < all_names.n; k++) {
359 if (svec_contains(&bridge_names, all_names.names[k])) {
364 /* No. Delete the dpif. */
368 svec_destroy(&all_names);
373 svec_destroy(&bridge_names);
374 svec_destroy(&dpif_names);
375 svec_destroy(&dpif_types);
378 /* Initializes 'options' and fills it with the options for 'if_cfg'. Merges
379 * keys from "options" and "other_config", preferring "options" keys over
380 * "other_config" keys. */
382 iface_get_options(const struct ovsrec_interface *if_cfg, struct shash *options)
386 shash_from_ovs_idl_map(if_cfg->key_options, if_cfg->value_options,
387 if_cfg->n_options, options);
389 for (i = 0; i < if_cfg->n_other_config; i++) {
390 char *key = if_cfg->key_other_config[i];
391 char *value = if_cfg->value_other_config[i];
393 if (!shash_find_data(options, key)) {
394 shash_add(options, key, value);
396 VLOG_WARN("%s: ignoring \"other_config\" key %s that conflicts "
397 "with \"options\" key %s", if_cfg->name, key, key);
402 /* Callback for iterate_and_prune_ifaces(). */
404 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
406 if (!iface->netdev) {
407 /* We already reported a related error, don't bother duplicating it. */
411 if (iface->dp_ifidx < 0) {
412 VLOG_ERR("%s interface not in %s, dropping",
413 iface->name, dpif_name(br->dpif));
417 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
418 iface->name, iface->dp_ifidx);
422 /* Callback for iterate_and_prune_ifaces(). */
424 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
425 void *aux OVS_UNUSED)
427 /* Set policing attributes. */
428 netdev_set_policing(iface->netdev,
429 iface->cfg->ingress_policing_rate,
430 iface->cfg->ingress_policing_burst);
432 /* Set MAC address of internal interfaces other than the local
434 if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) {
435 iface_set_mac(iface);
441 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
442 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
443 * deletes from 'br' any ports that no longer have any interfaces. */
445 iterate_and_prune_ifaces(struct bridge *br,
446 bool (*cb)(struct bridge *, struct iface *,
452 for (i = 0; i < br->n_ports; ) {
453 struct port *port = br->ports[i];
454 for (j = 0; j < port->n_ifaces; ) {
455 struct iface *iface = port->ifaces[j];
456 if (cb(br, iface, aux)) {
459 iface_set_ofport(iface->cfg, -1);
460 iface_destroy(iface);
464 if (port->n_ifaces) {
467 VLOG_ERR("%s port has no interfaces, dropping", port->name);
473 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
474 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
475 * responsible for freeing '*managersp' (with free()).
477 * You may be asking yourself "why does ovs-vswitchd care?", because
478 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
479 * should not be and in fact is not directly involved in that. But
480 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
481 * it has to tell in-band control where the managers are to enable that.
482 * (Thus, only managers connected in-band are collected.)
485 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
486 struct sockaddr_in **managersp, size_t *n_managersp)
488 struct sockaddr_in *managers = NULL;
489 size_t n_managers = 0;
490 struct shash targets;
493 /* Collect all of the potential targets, as the union of the "managers"
494 * column and the "targets" columns of the rows pointed to by
495 * "manager_options", excluding any that are out-of-band. */
496 shash_init(&targets);
497 for (i = 0; i < ovs_cfg->n_managers; i++) {
498 shash_add_once(&targets, ovs_cfg->managers[i], NULL);
500 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
501 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
503 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
504 shash_find_and_delete(&targets, m->target);
506 shash_add_once(&targets, m->target, NULL);
510 /* Now extract the targets' IP addresses. */
511 if (!shash_is_empty(&targets)) {
512 struct shash_node *node;
514 managers = xmalloc(shash_count(&targets) * sizeof *managers);
515 SHASH_FOR_EACH (node, &targets) {
516 const char *target = node->name;
517 struct sockaddr_in *sin = &managers[n_managers];
519 if ((!strncmp(target, "tcp:", 4)
520 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
521 (!strncmp(target, "ssl:", 4)
522 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
527 shash_destroy(&targets);
529 *managersp = managers;
530 *n_managersp = n_managers;
534 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
536 struct shash old_br, new_br;
537 struct shash_node *node;
538 struct bridge *br, *next;
539 struct sockaddr_in *managers;
542 int sflow_bridge_number;
544 COVERAGE_INC(bridge_reconfigure);
546 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
548 /* Collect old and new bridges. */
551 LIST_FOR_EACH (br, node, &all_bridges) {
552 shash_add(&old_br, br->name, br);
554 for (i = 0; i < ovs_cfg->n_bridges; i++) {
555 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
556 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
557 VLOG_WARN("more than one bridge named %s", br_cfg->name);
561 /* Get rid of deleted bridges and add new bridges. */
562 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
563 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
570 SHASH_FOR_EACH (node, &new_br) {
571 const char *br_name = node->name;
572 const struct ovsrec_bridge *br_cfg = node->data;
573 br = shash_find_data(&old_br, br_name);
575 /* If the bridge datapath type has changed, we need to tear it
576 * down and recreate. */
577 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
579 bridge_create(br_cfg);
582 bridge_create(br_cfg);
585 shash_destroy(&old_br);
586 shash_destroy(&new_br);
588 /* Reconfigure all bridges. */
589 LIST_FOR_EACH (br, node, &all_bridges) {
590 bridge_reconfigure_one(br);
593 /* Add and delete ports on all datapaths.
595 * The kernel will reject any attempt to add a given port to a datapath if
596 * that port already belongs to a different datapath, so we must do all
597 * port deletions before any port additions. */
598 LIST_FOR_EACH (br, node, &all_bridges) {
599 struct odp_port *dpif_ports;
601 struct shash want_ifaces;
603 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
604 bridge_get_all_ifaces(br, &want_ifaces);
605 for (i = 0; i < n_dpif_ports; i++) {
606 const struct odp_port *p = &dpif_ports[i];
607 if (!shash_find(&want_ifaces, p->devname)
608 && strcmp(p->devname, br->name)) {
609 int retval = dpif_port_del(br->dpif, p->port);
611 VLOG_ERR("failed to remove %s interface from %s: %s",
612 p->devname, dpif_name(br->dpif),
617 shash_destroy(&want_ifaces);
620 LIST_FOR_EACH (br, node, &all_bridges) {
621 struct odp_port *dpif_ports;
623 struct shash cur_ifaces, want_ifaces;
625 /* Get the set of interfaces currently in this datapath. */
626 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
627 shash_init(&cur_ifaces);
628 for (i = 0; i < n_dpif_ports; i++) {
629 const char *name = dpif_ports[i].devname;
630 shash_add_once(&cur_ifaces, name, &dpif_ports[i]);
633 /* Get the set of interfaces we want on this datapath. */
634 bridge_get_all_ifaces(br, &want_ifaces);
636 hmap_clear(&br->ifaces);
637 SHASH_FOR_EACH (node, &want_ifaces) {
638 const char *if_name = node->name;
639 struct iface *iface = node->data;
640 struct odp_port *dpif_port = shash_find_data(&cur_ifaces, if_name);
641 const char *type = iface ? iface->type : "internal";
644 /* If we have a port or a netdev already, and it's not the type we
645 * want, then delete the port (if any) and close the netdev (if
647 if ((dpif_port && strcmp(dpif_port->type, type))
648 || (iface && iface->netdev
649 && strcmp(type, netdev_get_type(iface->netdev)))) {
651 error = ofproto_port_del(br->ofproto, dpif_port->port);
658 netdev_close(iface->netdev);
659 iface->netdev = NULL;
663 /* If the port doesn't exist or we don't have the netdev open,
664 * we need to do more work. */
665 if (!dpif_port || (iface && !iface->netdev)) {
666 struct netdev_options options;
667 struct netdev *netdev;
670 /* First open the network device. */
671 options.name = if_name;
673 options.args = &args;
674 options.ethertype = NETDEV_ETH_TYPE_NONE;
678 iface_get_options(iface->cfg, &args);
680 error = netdev_open(&options, &netdev);
681 shash_destroy(&args);
684 VLOG_WARN("could not open network device %s (%s)",
685 if_name, strerror(error));
689 /* Then add the port if we haven't already. */
691 error = dpif_port_add(br->dpif, netdev, NULL);
693 netdev_close(netdev);
694 if (error == EFBIG) {
695 VLOG_ERR("ran out of valid port numbers on %s",
696 dpif_name(br->dpif));
699 VLOG_ERR("failed to add %s interface to %s: %s",
700 if_name, dpif_name(br->dpif),
707 /* Update 'iface'. */
709 iface->netdev = netdev;
710 iface->enabled = netdev_get_carrier(iface->netdev);
712 } else if (iface && iface->netdev) {
716 iface_get_options(iface->cfg, &args);
717 netdev_reconfigure(iface->netdev, &args);
718 shash_destroy(&args);
722 shash_destroy(&cur_ifaces);
723 shash_destroy(&want_ifaces);
725 sflow_bridge_number = 0;
726 LIST_FOR_EACH (br, node, &all_bridges) {
729 struct iface *local_iface;
730 struct iface *hw_addr_iface;
733 bridge_fetch_dp_ifaces(br);
735 iterate_and_prune_ifaces(br, check_iface, NULL);
737 /* Pick local port hardware address, datapath ID. */
738 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
739 local_iface = bridge_get_local_iface(br);
741 int error = netdev_set_etheraddr(local_iface->netdev, ea);
743 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
744 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
745 "Ethernet address: %s",
746 br->name, strerror(error));
750 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
751 ofproto_set_datapath_id(br->ofproto, dpid);
753 dpid_string = xasprintf("%016"PRIx64, dpid);
754 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
757 /* Set NetFlow configuration on this bridge. */
758 if (br->cfg->netflow) {
759 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
760 struct netflow_options opts;
762 memset(&opts, 0, sizeof opts);
764 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
765 if (nf_cfg->engine_type) {
766 opts.engine_type = *nf_cfg->engine_type;
768 if (nf_cfg->engine_id) {
769 opts.engine_id = *nf_cfg->engine_id;
772 opts.active_timeout = nf_cfg->active_timeout;
773 if (!opts.active_timeout) {
774 opts.active_timeout = -1;
775 } else if (opts.active_timeout < 0) {
776 VLOG_WARN("bridge %s: active timeout interval set to negative "
777 "value, using default instead (%d seconds)", br->name,
778 NF_ACTIVE_TIMEOUT_DEFAULT);
779 opts.active_timeout = -1;
782 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
783 if (opts.add_id_to_iface) {
784 if (opts.engine_id > 0x7f) {
785 VLOG_WARN("bridge %s: netflow port mangling may conflict "
786 "with another vswitch, choose an engine id less "
787 "than 128", br->name);
789 if (br->n_ports > 508) {
790 VLOG_WARN("bridge %s: netflow port mangling will conflict "
791 "with another port when more than 508 ports are "
796 opts.collectors.n = nf_cfg->n_targets;
797 opts.collectors.names = nf_cfg->targets;
798 if (ofproto_set_netflow(br->ofproto, &opts)) {
799 VLOG_ERR("bridge %s: problem setting netflow collectors",
803 ofproto_set_netflow(br->ofproto, NULL);
806 /* Set sFlow configuration on this bridge. */
807 if (br->cfg->sflow) {
808 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
809 struct ovsrec_controller **controllers;
810 struct ofproto_sflow_options oso;
811 size_t n_controllers;
813 memset(&oso, 0, sizeof oso);
815 oso.targets.n = sflow_cfg->n_targets;
816 oso.targets.names = sflow_cfg->targets;
818 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
819 if (sflow_cfg->sampling) {
820 oso.sampling_rate = *sflow_cfg->sampling;
823 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
824 if (sflow_cfg->polling) {
825 oso.polling_interval = *sflow_cfg->polling;
828 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
829 if (sflow_cfg->header) {
830 oso.header_len = *sflow_cfg->header;
833 oso.sub_id = sflow_bridge_number++;
834 oso.agent_device = sflow_cfg->agent;
836 oso.control_ip = NULL;
837 n_controllers = bridge_get_controllers(br, &controllers);
838 for (i = 0; i < n_controllers; i++) {
839 if (controllers[i]->local_ip) {
840 oso.control_ip = controllers[i]->local_ip;
844 ofproto_set_sflow(br->ofproto, &oso);
846 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
848 ofproto_set_sflow(br->ofproto, NULL);
851 /* Update the controller and related settings. It would be more
852 * straightforward to call this from bridge_reconfigure_one(), but we
853 * can't do it there for two reasons. First, and most importantly, at
854 * that point we don't know the dp_ifidx of any interfaces that have
855 * been added to the bridge (because we haven't actually added them to
856 * the datapath). Second, at that point we haven't set the datapath ID
857 * yet; when a controller is configured, resetting the datapath ID will
858 * immediately disconnect from the controller, so it's better to set
859 * the datapath ID before the controller. */
860 bridge_reconfigure_remotes(br, managers, n_managers);
862 LIST_FOR_EACH (br, node, &all_bridges) {
863 for (i = 0; i < br->n_ports; i++) {
864 struct port *port = br->ports[i];
867 port_update_vlan_compat(port);
868 port_update_bonding(port);
870 for (j = 0; j < port->n_ifaces; j++) {
871 iface_update_qos(port->ifaces[j], port->cfg->qos);
875 LIST_FOR_EACH (br, node, &all_bridges) {
876 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
879 LIST_FOR_EACH (br, node, &all_bridges) {
881 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
882 iface_update_cfm(iface);
890 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
891 const struct ovsdb_idl_column *column,
894 const struct ovsdb_datum *datum;
895 union ovsdb_atom atom;
898 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
899 atom.string = (char *) key;
900 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
901 return idx == UINT_MAX ? NULL : datum->values[idx].string;
905 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
907 return get_ovsrec_key_value(&br_cfg->header_,
908 &ovsrec_bridge_col_other_config, key);
912 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
913 struct iface **hw_addr_iface)
919 *hw_addr_iface = NULL;
921 /* Did the user request a particular MAC? */
922 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
923 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
924 if (eth_addr_is_multicast(ea)) {
925 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
926 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
927 } else if (eth_addr_is_zero(ea)) {
928 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
934 /* Otherwise choose the minimum non-local MAC address among all of the
936 memset(ea, 0xff, sizeof ea);
937 for (i = 0; i < br->n_ports; i++) {
938 struct port *port = br->ports[i];
939 uint8_t iface_ea[ETH_ADDR_LEN];
942 /* Mirror output ports don't participate. */
943 if (port->is_mirror_output_port) {
947 /* Choose the MAC address to represent the port. */
948 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
949 /* Find the interface with this Ethernet address (if any) so that
950 * we can provide the correct devname to the caller. */
952 for (j = 0; j < port->n_ifaces; j++) {
953 struct iface *candidate = port->ifaces[j];
954 uint8_t candidate_ea[ETH_ADDR_LEN];
955 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
956 && eth_addr_equals(iface_ea, candidate_ea)) {
961 /* Choose the interface whose MAC address will represent the port.
962 * The Linux kernel bonding code always chooses the MAC address of
963 * the first slave added to a bond, and the Fedora networking
964 * scripts always add slaves to a bond in alphabetical order, so
965 * for compatibility we choose the interface with the name that is
966 * first in alphabetical order. */
967 iface = port->ifaces[0];
968 for (j = 1; j < port->n_ifaces; j++) {
969 struct iface *candidate = port->ifaces[j];
970 if (strcmp(candidate->name, iface->name) < 0) {
975 /* The local port doesn't count (since we're trying to choose its
976 * MAC address anyway). */
977 if (iface->dp_ifidx == ODPP_LOCAL) {
982 error = netdev_get_etheraddr(iface->netdev, iface_ea);
984 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
985 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
986 iface->name, strerror(error));
991 /* Compare against our current choice. */
992 if (!eth_addr_is_multicast(iface_ea) &&
993 !eth_addr_is_local(iface_ea) &&
994 !eth_addr_is_reserved(iface_ea) &&
995 !eth_addr_is_zero(iface_ea) &&
996 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
998 memcpy(ea, iface_ea, ETH_ADDR_LEN);
999 *hw_addr_iface = iface;
1002 if (eth_addr_is_multicast(ea)) {
1003 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1004 *hw_addr_iface = NULL;
1005 VLOG_WARN("bridge %s: using default bridge Ethernet "
1006 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1008 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1009 br->name, ETH_ADDR_ARGS(ea));
1013 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1014 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1015 * an interface on 'br', then that interface must be passed in as
1016 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1017 * 'hw_addr_iface' must be passed in as a null pointer. */
1019 bridge_pick_datapath_id(struct bridge *br,
1020 const uint8_t bridge_ea[ETH_ADDR_LEN],
1021 struct iface *hw_addr_iface)
1024 * The procedure for choosing a bridge MAC address will, in the most
1025 * ordinary case, also choose a unique MAC that we can use as a datapath
1026 * ID. In some special cases, though, multiple bridges will end up with
1027 * the same MAC address. This is OK for the bridges, but it will confuse
1028 * the OpenFlow controller, because each datapath needs a unique datapath
1031 * Datapath IDs must be unique. It is also very desirable that they be
1032 * stable from one run to the next, so that policy set on a datapath
1035 const char *datapath_id;
1038 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1039 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1043 if (hw_addr_iface) {
1045 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1047 * A bridge whose MAC address is taken from a VLAN network device
1048 * (that is, a network device created with vconfig(8) or similar
1049 * tool) will have the same MAC address as a bridge on the VLAN
1050 * device's physical network device.
1052 * Handle this case by hashing the physical network device MAC
1053 * along with the VLAN identifier.
1055 uint8_t buf[ETH_ADDR_LEN + 2];
1056 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1057 buf[ETH_ADDR_LEN] = vlan >> 8;
1058 buf[ETH_ADDR_LEN + 1] = vlan;
1059 return dpid_from_hash(buf, sizeof buf);
1062 * Assume that this bridge's MAC address is unique, since it
1063 * doesn't fit any of the cases we handle specially.
1068 * A purely internal bridge, that is, one that has no non-virtual
1069 * network devices on it at all, is more difficult because it has no
1070 * natural unique identifier at all.
1072 * When the host is a XenServer, we handle this case by hashing the
1073 * host's UUID with the name of the bridge. Names of bridges are
1074 * persistent across XenServer reboots, although they can be reused if
1075 * an internal network is destroyed and then a new one is later
1076 * created, so this is fairly effective.
1078 * When the host is not a XenServer, we punt by using a random MAC
1079 * address on each run.
1081 const char *host_uuid = xenserver_get_host_uuid();
1083 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1084 dpid = dpid_from_hash(combined, strlen(combined));
1090 return eth_addr_to_uint64(bridge_ea);
1094 dpid_from_hash(const void *data, size_t n)
1096 uint8_t hash[SHA1_DIGEST_SIZE];
1098 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1099 sha1_bytes(data, n, hash);
1100 eth_addr_mark_random(hash);
1101 return eth_addr_to_uint64(hash);
1105 iface_refresh_cfm_stats(struct iface *iface)
1109 const struct ovsrec_monitor *mon;
1111 mon = iface->cfg->monitor;
1118 for (i = 0; i < mon->n_remote_mps; i++) {
1119 const struct ovsrec_maintenance_point *mp;
1120 const struct remote_mp *rmp;
1122 mp = mon->remote_mps[i];
1123 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1125 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1128 if (hmap_is_empty(&cfm->x_remote_mps)) {
1129 ovsrec_monitor_set_unexpected_remote_mpids(mon, NULL, 0);
1132 struct remote_mp *rmp;
1133 int64_t *x_remote_mps;
1135 length = hmap_count(&cfm->x_remote_mps);
1136 x_remote_mps = xzalloc(length * sizeof *x_remote_mps);
1139 HMAP_FOR_EACH (rmp, node, &cfm->x_remote_mps) {
1140 x_remote_mps[i++] = rmp->mpid;
1143 ovsrec_monitor_set_unexpected_remote_mpids(mon, x_remote_mps, length);
1147 if (hmap_is_empty(&cfm->x_remote_maids)) {
1148 ovsrec_monitor_set_unexpected_remote_maids(mon, NULL, 0);
1151 char **x_remote_maids;
1152 struct remote_maid *rmaid;
1154 length = hmap_count(&cfm->x_remote_maids);
1155 x_remote_maids = xzalloc(length * sizeof *x_remote_maids);
1158 HMAP_FOR_EACH (rmaid, node, &cfm->x_remote_maids) {
1161 x_remote_maids[i] = xzalloc(CCM_MAID_LEN * 2 + 1);
1163 for (j = 0; j < CCM_MAID_LEN; j++) {
1164 snprintf(&x_remote_maids[i][j * 2], 3, "%02hhx",
1169 ovsrec_monitor_set_unexpected_remote_maids(mon, x_remote_maids, length);
1171 for (i = 0; i < length; i++) {
1172 free(x_remote_maids[i]);
1174 free(x_remote_maids);
1177 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1181 iface_refresh_stats(struct iface *iface)
1187 static const struct iface_stat iface_stats[] = {
1188 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1189 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1190 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1191 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1192 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1193 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1194 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1195 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1196 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1197 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1198 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1199 { "collisions", offsetof(struct netdev_stats, collisions) },
1201 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1202 const struct iface_stat *s;
1204 char *keys[N_STATS];
1205 int64_t values[N_STATS];
1208 struct netdev_stats stats;
1210 /* Intentionally ignore return value, since errors will set 'stats' to
1211 * all-1s, and we will deal with that correctly below. */
1212 netdev_get_stats(iface->netdev, &stats);
1215 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1216 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1217 if (value != UINT64_MAX) {
1224 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1228 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1230 struct ovsdb_datum datum;
1234 get_system_stats(&stats);
1236 ovsdb_datum_from_shash(&datum, &stats);
1237 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1244 const struct ovsrec_open_vswitch *cfg;
1246 bool datapath_destroyed;
1247 bool database_changed;
1250 /* Let each bridge do the work that it needs to do. */
1251 datapath_destroyed = false;
1252 LIST_FOR_EACH (br, node, &all_bridges) {
1253 int error = bridge_run_one(br);
1255 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1256 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1257 "forcing reconfiguration", br->name);
1258 datapath_destroyed = true;
1262 /* (Re)configure if necessary. */
1263 database_changed = ovsdb_idl_run(idl);
1264 cfg = ovsrec_open_vswitch_first(idl);
1265 if (database_changed || datapath_destroyed) {
1267 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1269 bridge_configure_once(cfg);
1270 bridge_reconfigure(cfg);
1272 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1273 ovsdb_idl_txn_commit(txn);
1274 ovsdb_idl_txn_destroy(txn); /* XXX */
1276 /* We still need to reconfigure to avoid dangling pointers to
1277 * now-destroyed ovsrec structures inside bridge data. */
1278 static const struct ovsrec_open_vswitch null_cfg;
1280 bridge_reconfigure(&null_cfg);
1285 /* Re-configure SSL. We do this on every trip through the main loop,
1286 * instead of just when the database changes, because the contents of the
1287 * key and certificate files can change without the database changing. */
1288 if (cfg && cfg->ssl) {
1289 const struct ovsrec_ssl *ssl = cfg->ssl;
1291 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1292 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1296 /* Refresh system and interface stats if necessary. */
1297 if (time_msec() >= stats_timer) {
1299 struct ovsdb_idl_txn *txn;
1301 txn = ovsdb_idl_txn_create(idl);
1302 LIST_FOR_EACH (br, node, &all_bridges) {
1305 for (i = 0; i < br->n_ports; i++) {
1306 struct port *port = br->ports[i];
1309 for (j = 0; j < port->n_ifaces; j++) {
1310 struct iface *iface = port->ifaces[j];
1311 iface_refresh_stats(iface);
1312 iface_refresh_cfm_stats(iface);
1316 refresh_system_stats(cfg);
1317 ovsdb_idl_txn_commit(txn);
1318 ovsdb_idl_txn_destroy(txn); /* XXX */
1321 stats_timer = time_msec() + STATS_INTERVAL;
1329 struct iface *iface;
1331 LIST_FOR_EACH (br, node, &all_bridges) {
1332 ofproto_wait(br->ofproto);
1333 if (ofproto_has_primary_controller(br->ofproto)) {
1337 mac_learning_wait(br->ml);
1340 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1342 cfm_wait(iface->cfm);
1346 ovsdb_idl_wait(idl);
1347 poll_timer_wait_until(stats_timer);
1350 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1351 * configuration changes. */
1353 bridge_flush(struct bridge *br)
1355 COVERAGE_INC(bridge_flush);
1357 mac_learning_flush(br->ml);
1360 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1361 * such interface. */
1362 static struct iface *
1363 bridge_get_local_iface(struct bridge *br)
1367 for (i = 0; i < br->n_ports; i++) {
1368 struct port *port = br->ports[i];
1369 for (j = 0; j < port->n_ifaces; j++) {
1370 struct iface *iface = port->ifaces[j];
1371 if (iface->dp_ifidx == ODPP_LOCAL) {
1380 /* Bridge unixctl user interface functions. */
1382 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1383 const char *args, void *aux OVS_UNUSED)
1385 struct ds ds = DS_EMPTY_INITIALIZER;
1386 const struct bridge *br;
1387 const struct mac_entry *e;
1389 br = bridge_lookup(args);
1391 unixctl_command_reply(conn, 501, "no such bridge");
1395 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1396 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1397 if (e->port < 0 || e->port >= br->n_ports) {
1400 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1401 br->ports[e->port]->ifaces[0]->dp_ifidx,
1402 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1404 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1408 /* Bridge reconfiguration functions. */
1409 static struct bridge *
1410 bridge_create(const struct ovsrec_bridge *br_cfg)
1415 assert(!bridge_lookup(br_cfg->name));
1416 br = xzalloc(sizeof *br);
1418 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1424 dpif_flow_flush(br->dpif);
1426 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1429 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1431 dpif_delete(br->dpif);
1432 dpif_close(br->dpif);
1437 br->name = xstrdup(br_cfg->name);
1439 br->ml = mac_learning_create();
1440 eth_addr_nicira_random(br->default_ea);
1442 hmap_init(&br->ifaces);
1444 shash_init(&br->port_by_name);
1445 shash_init(&br->iface_by_name);
1449 list_push_back(&all_bridges, &br->node);
1451 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1457 bridge_destroy(struct bridge *br)
1462 while (br->n_ports > 0) {
1463 port_destroy(br->ports[br->n_ports - 1]);
1465 list_remove(&br->node);
1466 error = dpif_delete(br->dpif);
1467 if (error && error != ENOENT) {
1468 VLOG_ERR("failed to delete %s: %s",
1469 dpif_name(br->dpif), strerror(error));
1471 dpif_close(br->dpif);
1472 ofproto_destroy(br->ofproto);
1473 mac_learning_destroy(br->ml);
1474 hmap_destroy(&br->ifaces);
1475 shash_destroy(&br->port_by_name);
1476 shash_destroy(&br->iface_by_name);
1483 static struct bridge *
1484 bridge_lookup(const char *name)
1488 LIST_FOR_EACH (br, node, &all_bridges) {
1489 if (!strcmp(br->name, name)) {
1496 /* Handle requests for a listing of all flows known by the OpenFlow
1497 * stack, including those normally hidden. */
1499 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1500 const char *args, void *aux OVS_UNUSED)
1505 br = bridge_lookup(args);
1507 unixctl_command_reply(conn, 501, "Unknown bridge");
1512 ofproto_get_all_flows(br->ofproto, &results);
1514 unixctl_command_reply(conn, 200, ds_cstr(&results));
1515 ds_destroy(&results);
1518 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1519 * connections and reconnect. If BRIDGE is not specified, then all bridges
1520 * drop their controller connections and reconnect. */
1522 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1523 const char *args, void *aux OVS_UNUSED)
1526 if (args[0] != '\0') {
1527 br = bridge_lookup(args);
1529 unixctl_command_reply(conn, 501, "Unknown bridge");
1532 ofproto_reconnect_controllers(br->ofproto);
1534 LIST_FOR_EACH (br, node, &all_bridges) {
1535 ofproto_reconnect_controllers(br->ofproto);
1538 unixctl_command_reply(conn, 200, NULL);
1542 bridge_run_one(struct bridge *br)
1545 struct iface *iface;
1547 error = ofproto_run1(br->ofproto);
1552 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1555 error = ofproto_run2(br->ofproto, br->flush);
1558 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
1559 struct ofpbuf *packet;
1565 packet = cfm_run(iface->cfm);
1567 iface_send_packet(iface, packet);
1568 ofpbuf_uninit(packet);
1577 bridge_get_controllers(const struct bridge *br,
1578 struct ovsrec_controller ***controllersp)
1580 struct ovsrec_controller **controllers;
1581 size_t n_controllers;
1583 controllers = br->cfg->controller;
1584 n_controllers = br->cfg->n_controller;
1586 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1592 *controllersp = controllers;
1594 return n_controllers;
1598 bridge_reconfigure_one(struct bridge *br)
1600 struct shash old_ports, new_ports;
1601 struct svec snoops, old_snoops;
1602 struct shash_node *node;
1603 enum ofproto_fail_mode fail_mode;
1606 /* Collect old ports. */
1607 shash_init(&old_ports);
1608 for (i = 0; i < br->n_ports; i++) {
1609 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1612 /* Collect new ports. */
1613 shash_init(&new_ports);
1614 for (i = 0; i < br->cfg->n_ports; i++) {
1615 const char *name = br->cfg->ports[i]->name;
1616 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1617 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1622 /* If we have a controller, then we need a local port. Complain if the
1623 * user didn't specify one.
1625 * XXX perhaps we should synthesize a port ourselves in this case. */
1626 if (bridge_get_controllers(br, NULL)) {
1627 char local_name[IF_NAMESIZE];
1630 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1631 local_name, sizeof local_name);
1632 if (!error && !shash_find(&new_ports, local_name)) {
1633 VLOG_WARN("bridge %s: controller specified but no local port "
1634 "(port named %s) defined",
1635 br->name, local_name);
1639 /* Get rid of deleted ports.
1640 * Get rid of deleted interfaces on ports that still exist. */
1641 SHASH_FOR_EACH (node, &old_ports) {
1642 struct port *port = node->data;
1643 const struct ovsrec_port *port_cfg;
1645 port_cfg = shash_find_data(&new_ports, node->name);
1649 port_del_ifaces(port, port_cfg);
1653 /* Create new ports.
1654 * Add new interfaces to existing ports.
1655 * Reconfigure existing ports. */
1656 SHASH_FOR_EACH (node, &new_ports) {
1657 struct port *port = shash_find_data(&old_ports, node->name);
1659 port = port_create(br, node->name);
1662 port_reconfigure(port, node->data);
1663 if (!port->n_ifaces) {
1664 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1665 br->name, port->name);
1669 shash_destroy(&old_ports);
1670 shash_destroy(&new_ports);
1672 /* Set the fail-mode */
1673 fail_mode = !br->cfg->fail_mode
1674 || !strcmp(br->cfg->fail_mode, "standalone")
1675 ? OFPROTO_FAIL_STANDALONE
1676 : OFPROTO_FAIL_SECURE;
1677 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1678 && !ofproto_has_primary_controller(br->ofproto)) {
1679 ofproto_flush_flows(br->ofproto);
1681 ofproto_set_fail_mode(br->ofproto, fail_mode);
1683 /* Delete all flows if we're switching from connected to standalone or vice
1684 * versa. (XXX Should we delete all flows if we are switching from one
1685 * controller to another?) */
1687 /* Configure OpenFlow controller connection snooping. */
1689 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1690 ovs_rundir(), br->name));
1691 svec_init(&old_snoops);
1692 ofproto_get_snoops(br->ofproto, &old_snoops);
1693 if (!svec_equal(&snoops, &old_snoops)) {
1694 ofproto_set_snoops(br->ofproto, &snoops);
1696 svec_destroy(&snoops);
1697 svec_destroy(&old_snoops);
1699 mirror_reconfigure(br);
1702 /* Initializes 'oc' appropriately as a management service controller for
1705 * The caller must free oc->target when it is no longer needed. */
1707 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1708 struct ofproto_controller *oc)
1710 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1711 oc->max_backoff = 0;
1712 oc->probe_interval = 60;
1713 oc->band = OFPROTO_OUT_OF_BAND;
1714 oc->accept_re = NULL;
1715 oc->update_resolv_conf = false;
1717 oc->burst_limit = 0;
1720 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1722 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1723 struct ofproto_controller *oc)
1725 oc->target = c->target;
1726 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1727 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1728 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1729 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1730 oc->accept_re = c->discover_accept_regex;
1731 oc->update_resolv_conf = c->discover_update_resolv_conf;
1732 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1733 oc->burst_limit = (c->controller_burst_limit
1734 ? *c->controller_burst_limit : 0);
1737 /* Configures the IP stack for 'br''s local interface properly according to the
1738 * configuration in 'c'. */
1740 bridge_configure_local_iface_netdev(struct bridge *br,
1741 struct ovsrec_controller *c)
1743 struct netdev *netdev;
1744 struct in_addr mask, gateway;
1746 struct iface *local_iface;
1749 /* Controller discovery does its own TCP/IP configuration later. */
1750 if (strcmp(c->target, "discover")) {
1754 /* If there's no local interface or no IP address, give up. */
1755 local_iface = bridge_get_local_iface(br);
1756 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1760 /* Bring up the local interface. */
1761 netdev = local_iface->netdev;
1762 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1764 /* Configure the IP address and netmask. */
1765 if (!c->local_netmask
1766 || !inet_aton(c->local_netmask, &mask)
1768 mask.s_addr = guess_netmask(ip.s_addr);
1770 if (!netdev_set_in4(netdev, ip, mask)) {
1771 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1772 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1775 /* Configure the default gateway. */
1776 if (c->local_gateway
1777 && inet_aton(c->local_gateway, &gateway)
1778 && gateway.s_addr) {
1779 if (!netdev_add_router(netdev, gateway)) {
1780 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1781 br->name, IP_ARGS(&gateway.s_addr));
1787 bridge_reconfigure_remotes(struct bridge *br,
1788 const struct sockaddr_in *managers,
1791 const char *disable_ib_str, *queue_id_str;
1792 bool disable_in_band = false;
1795 struct ovsrec_controller **controllers;
1796 size_t n_controllers;
1799 struct ofproto_controller *ocs;
1803 /* Check if we should disable in-band control on this bridge. */
1804 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
1805 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
1806 disable_in_band = true;
1809 /* Set OpenFlow queue ID for in-band control. */
1810 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
1811 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
1812 ofproto_set_in_band_queue(br->ofproto, queue_id);
1814 if (disable_in_band) {
1815 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
1817 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1819 had_primary = ofproto_has_primary_controller(br->ofproto);
1821 n_controllers = bridge_get_controllers(br, &controllers);
1823 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1826 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1827 for (i = 0; i < n_controllers; i++) {
1828 struct ovsrec_controller *c = controllers[i];
1830 if (!strncmp(c->target, "punix:", 6)
1831 || !strncmp(c->target, "unix:", 5)) {
1832 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1834 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1835 * domain sockets and overwriting arbitrary local files. */
1836 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1837 "\"%s\" due to possibility for remote exploit",
1838 dpif_name(br->dpif), c->target);
1842 bridge_configure_local_iface_netdev(br, c);
1843 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
1844 if (disable_in_band) {
1845 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
1850 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1851 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1854 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1855 ofproto_flush_flows(br->ofproto);
1858 /* If there are no controllers and the bridge is in standalone
1859 * mode, set up a flow that matches every packet and directs
1860 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1861 * switch is in secure mode and we won't pass any traffic until
1862 * a controller has been defined and it tells us to do so. */
1864 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1865 union ofp_action action;
1866 struct cls_rule rule;
1868 memset(&action, 0, sizeof action);
1869 action.type = htons(OFPAT_OUTPUT);
1870 action.output.len = htons(sizeof action);
1871 action.output.port = htons(OFPP_NORMAL);
1872 cls_rule_init_catchall(&rule, 0);
1873 ofproto_add_flow(br->ofproto, &rule, &action, 1);
1878 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1883 for (i = 0; i < br->n_ports; i++) {
1884 struct port *port = br->ports[i];
1885 for (j = 0; j < port->n_ifaces; j++) {
1886 struct iface *iface = port->ifaces[j];
1887 shash_add_once(ifaces, iface->name, iface);
1889 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1890 shash_add_once(ifaces, port->name, NULL);
1895 /* For robustness, in case the administrator moves around datapath ports behind
1896 * our back, we re-check all the datapath port numbers here.
1898 * This function will set the 'dp_ifidx' members of interfaces that have
1899 * disappeared to -1, so only call this function from a context where those
1900 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1901 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1902 * datapath, which doesn't support UINT16_MAX+1 ports. */
1904 bridge_fetch_dp_ifaces(struct bridge *br)
1906 struct odp_port *dpif_ports;
1907 size_t n_dpif_ports;
1910 /* Reset all interface numbers. */
1911 for (i = 0; i < br->n_ports; i++) {
1912 struct port *port = br->ports[i];
1913 for (j = 0; j < port->n_ifaces; j++) {
1914 struct iface *iface = port->ifaces[j];
1915 iface->dp_ifidx = -1;
1918 hmap_clear(&br->ifaces);
1920 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1921 for (i = 0; i < n_dpif_ports; i++) {
1922 struct odp_port *p = &dpif_ports[i];
1923 struct iface *iface = iface_lookup(br, p->devname);
1925 if (iface->dp_ifidx >= 0) {
1926 VLOG_WARN("%s reported interface %s twice",
1927 dpif_name(br->dpif), p->devname);
1928 } else if (iface_from_dp_ifidx(br, p->port)) {
1929 VLOG_WARN("%s reported interface %"PRIu16" twice",
1930 dpif_name(br->dpif), p->port);
1932 iface->dp_ifidx = p->port;
1933 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
1934 hash_int(iface->dp_ifidx, 0));
1937 iface_set_ofport(iface->cfg,
1938 (iface->dp_ifidx >= 0
1939 ? odp_port_to_ofp_port(iface->dp_ifidx)
1946 /* Bridge packet processing functions. */
1949 bond_hash(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
1951 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
1954 static struct bond_entry *
1955 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN],
1958 return &port->bond_hash[bond_hash(mac, vlan)];
1962 bond_choose_iface(const struct port *port)
1964 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1965 size_t i, best_down_slave = -1;
1966 long long next_delay_expiration = LLONG_MAX;
1968 for (i = 0; i < port->n_ifaces; i++) {
1969 struct iface *iface = port->ifaces[i];
1971 if (iface->enabled) {
1973 } else if (iface->delay_expires < next_delay_expiration) {
1974 best_down_slave = i;
1975 next_delay_expiration = iface->delay_expires;
1979 if (best_down_slave != -1) {
1980 struct iface *iface = port->ifaces[best_down_slave];
1982 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1983 "since no other interface is up", iface->name,
1984 iface->delay_expires - time_msec());
1985 bond_enable_slave(iface, true);
1988 return best_down_slave;
1992 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1993 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
1995 struct iface *iface;
1997 assert(port->n_ifaces);
1998 if (port->n_ifaces == 1) {
1999 iface = port->ifaces[0];
2001 struct bond_entry *e = lookup_bond_entry(port, dl_src, vlan);
2002 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
2003 || !port->ifaces[e->iface_idx]->enabled) {
2004 /* XXX select interface properly. The current interface selection
2005 * is only good for testing the rebalancing code. */
2006 e->iface_idx = bond_choose_iface(port);
2007 if (e->iface_idx < 0) {
2008 *tags |= port->no_ifaces_tag;
2011 e->iface_tag = tag_create_random();
2012 ((struct port *) port)->bond_compat_is_stale = true;
2014 *tags |= e->iface_tag;
2015 iface = port->ifaces[e->iface_idx];
2017 *dp_ifidx = iface->dp_ifidx;
2018 *tags |= iface->tag; /* Currently only used for bonding. */
2023 bond_link_status_update(struct iface *iface, bool carrier)
2025 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2026 struct port *port = iface->port;
2028 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2029 /* Nothing to do. */
2032 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
2033 iface->name, carrier ? "detected" : "dropped");
2034 if (carrier == iface->enabled) {
2035 iface->delay_expires = LLONG_MAX;
2036 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2037 iface->name, carrier ? "disabled" : "enabled");
2038 } else if (carrier && port->active_iface < 0) {
2039 bond_enable_slave(iface, true);
2040 if (port->updelay) {
2041 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2042 "other interface is up", iface->name, port->updelay);
2045 int delay = carrier ? port->updelay : port->downdelay;
2046 iface->delay_expires = time_msec() + delay;
2049 "interface %s: will be %s if it stays %s for %d ms",
2051 carrier ? "enabled" : "disabled",
2052 carrier ? "up" : "down",
2059 bond_choose_active_iface(struct port *port)
2061 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2063 port->active_iface = bond_choose_iface(port);
2064 port->active_iface_tag = tag_create_random();
2065 if (port->active_iface >= 0) {
2066 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2067 port->name, port->ifaces[port->active_iface]->name);
2069 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2075 bond_enable_slave(struct iface *iface, bool enable)
2077 struct port *port = iface->port;
2078 struct bridge *br = port->bridge;
2080 /* This acts as a recursion check. If the act of disabling a slave
2081 * causes a different slave to be enabled, the flag will allow us to
2082 * skip redundant work when we reenter this function. It must be
2083 * cleared on exit to keep things safe with multiple bonds. */
2084 static bool moving_active_iface = false;
2086 iface->delay_expires = LLONG_MAX;
2087 if (enable == iface->enabled) {
2091 iface->enabled = enable;
2092 if (!iface->enabled) {
2093 VLOG_WARN("interface %s: disabled", iface->name);
2094 ofproto_revalidate(br->ofproto, iface->tag);
2095 if (iface->port_ifidx == port->active_iface) {
2096 ofproto_revalidate(br->ofproto,
2097 port->active_iface_tag);
2099 /* Disabling a slave can lead to another slave being immediately
2100 * enabled if there will be no active slaves but one is waiting
2101 * on an updelay. In this case we do not need to run most of the
2102 * code for the newly enabled slave since there was no period
2103 * without an active slave and it is redundant with the disabling
2105 moving_active_iface = true;
2106 bond_choose_active_iface(port);
2108 bond_send_learning_packets(port);
2110 VLOG_WARN("interface %s: enabled", iface->name);
2111 if (port->active_iface < 0 && !moving_active_iface) {
2112 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2113 bond_choose_active_iface(port);
2114 bond_send_learning_packets(port);
2116 iface->tag = tag_create_random();
2119 moving_active_iface = false;
2120 port->bond_compat_is_stale = true;
2123 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2124 * bond interface. */
2126 bond_update_fake_iface_stats(struct port *port)
2128 struct netdev_stats bond_stats;
2129 struct netdev *bond_dev;
2132 memset(&bond_stats, 0, sizeof bond_stats);
2134 for (i = 0; i < port->n_ifaces; i++) {
2135 struct netdev_stats slave_stats;
2137 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
2138 /* XXX: We swap the stats here because they are swapped back when
2139 * reported by the internal device. The reason for this is
2140 * internal devices normally represent packets going into the system
2141 * but when used as fake bond device they represent packets leaving
2142 * the system. We really should do this in the internal device
2143 * itself because changing it here reverses the counts from the
2144 * perspective of the switch. However, the internal device doesn't
2145 * know what type of device it represents so we have to do it here
2147 bond_stats.tx_packets += slave_stats.rx_packets;
2148 bond_stats.tx_bytes += slave_stats.rx_bytes;
2149 bond_stats.rx_packets += slave_stats.tx_packets;
2150 bond_stats.rx_bytes += slave_stats.tx_bytes;
2154 if (!netdev_open_default(port->name, &bond_dev)) {
2155 netdev_set_stats(bond_dev, &bond_stats);
2156 netdev_close(bond_dev);
2161 bond_run(struct bridge *br)
2165 for (i = 0; i < br->n_ports; i++) {
2166 struct port *port = br->ports[i];
2168 if (port->n_ifaces >= 2) {
2171 /* Track carrier going up and down on interfaces. */
2172 while (!netdev_monitor_poll(port->monitor, &devname)) {
2173 struct iface *iface;
2175 iface = port_lookup_iface(port, devname);
2177 bool carrier = netdev_get_carrier(iface->netdev);
2179 bond_link_status_update(iface, carrier);
2180 port_update_bond_compat(port);
2185 for (j = 0; j < port->n_ifaces; j++) {
2186 struct iface *iface = port->ifaces[j];
2187 if (time_msec() >= iface->delay_expires) {
2188 bond_enable_slave(iface, !iface->enabled);
2192 if (port->bond_fake_iface
2193 && time_msec() >= port->bond_next_fake_iface_update) {
2194 bond_update_fake_iface_stats(port);
2195 port->bond_next_fake_iface_update = time_msec() + 1000;
2199 if (port->bond_compat_is_stale) {
2200 port->bond_compat_is_stale = false;
2201 port_update_bond_compat(port);
2207 bond_wait(struct bridge *br)
2211 for (i = 0; i < br->n_ports; i++) {
2212 struct port *port = br->ports[i];
2213 if (port->n_ifaces < 2) {
2216 netdev_monitor_poll_wait(port->monitor);
2217 for (j = 0; j < port->n_ifaces; j++) {
2218 struct iface *iface = port->ifaces[j];
2219 if (iface->delay_expires != LLONG_MAX) {
2220 poll_timer_wait_until(iface->delay_expires);
2223 if (port->bond_fake_iface) {
2224 poll_timer_wait_until(port->bond_next_fake_iface_update);
2230 set_dst(struct dst *p, const struct flow *flow,
2231 const struct port *in_port, const struct port *out_port,
2234 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2235 : in_port->vlan >= 0 ? in_port->vlan
2236 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2237 : vlan_tci_to_vid(flow->vlan_tci));
2238 return choose_output_iface(out_port, flow->dl_src, p->vlan,
2239 &p->dp_ifidx, tags);
2243 swap_dst(struct dst *p, struct dst *q)
2245 struct dst tmp = *p;
2250 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2251 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2252 * that we push to the datapath. We could in fact fully sort the array by
2253 * vlan, but in most cases there are at most two different vlan tags so that's
2254 * possibly overkill.) */
2256 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2258 struct dst *first = dsts;
2259 struct dst *last = dsts + n_dsts;
2261 while (first != last) {
2263 * - All dsts < first have vlan == 'vlan'.
2264 * - All dsts >= last have vlan != 'vlan'.
2265 * - first < last. */
2266 while (first->vlan == vlan) {
2267 if (++first == last) {
2272 /* Same invariants, plus one additional:
2273 * - first->vlan != vlan.
2275 while (last[-1].vlan != vlan) {
2276 if (--last == first) {
2281 /* Same invariants, plus one additional:
2282 * - last[-1].vlan == vlan.*/
2283 swap_dst(first++, --last);
2288 mirror_mask_ffs(mirror_mask_t mask)
2290 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2295 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2296 const struct dst *test)
2299 for (i = 0; i < n_dsts; i++) {
2300 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
2308 port_trunks_vlan(const struct port *port, uint16_t vlan)
2310 return (port->vlan < 0
2311 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2315 port_includes_vlan(const struct port *port, uint16_t vlan)
2317 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2321 port_is_floodable(const struct port *port)
2325 for (i = 0; i < port->n_ifaces; i++) {
2326 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2327 port->ifaces[i]->dp_ifidx)) {
2335 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2336 const struct port *in_port, const struct port *out_port,
2337 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2339 mirror_mask_t mirrors = in_port->src_mirrors;
2341 struct dst *dst = dsts;
2344 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2345 if (flow_vlan == 0) {
2346 flow_vlan = OFP_VLAN_NONE;
2349 if (out_port == FLOOD_PORT) {
2350 /* XXX use ODP_FLOOD if no vlans or bonding. */
2351 /* XXX even better, define each VLAN as a datapath port group */
2352 for (i = 0; i < br->n_ports; i++) {
2353 struct port *port = br->ports[i];
2355 && port_is_floodable(port)
2356 && port_includes_vlan(port, vlan)
2357 && !port->is_mirror_output_port
2358 && set_dst(dst, flow, in_port, port, tags)) {
2359 mirrors |= port->dst_mirrors;
2363 *nf_output_iface = NF_OUT_FLOOD;
2364 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2365 *nf_output_iface = dst->dp_ifidx;
2366 mirrors |= out_port->dst_mirrors;
2371 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2372 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2374 if (set_dst(dst, flow, in_port, m->out_port, tags)
2375 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2379 for (i = 0; i < br->n_ports; i++) {
2380 struct port *port = br->ports[i];
2381 if (port_includes_vlan(port, m->out_vlan)
2382 && set_dst(dst, flow, in_port, port, tags))
2385 if (port->vlan < 0) {
2386 dst->vlan = m->out_vlan;
2388 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2392 /* Use the vlan tag on the original flow instead of
2393 * the one passed in the vlan parameter. This ensures
2394 * that we compare the vlan from before any implicit
2395 * tagging tags place. This is necessary because
2396 * dst->vlan is the final vlan, after removing implicit
2398 if (port == in_port && dst->vlan == flow_vlan) {
2399 /* Don't send out input port on same VLAN. */
2407 mirrors &= mirrors - 1;
2410 partition_dsts(dsts, dst - dsts, flow_vlan);
2414 static void OVS_UNUSED
2415 print_dsts(const struct dst *dsts, size_t n)
2417 for (; n--; dsts++) {
2418 printf(">p%"PRIu16, dsts->dp_ifidx);
2419 if (dsts->vlan != OFP_VLAN_NONE) {
2420 printf("v%"PRIu16, dsts->vlan);
2426 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2427 const struct port *in_port, const struct port *out_port,
2428 tag_type *tags, struct ofpbuf *actions,
2429 uint16_t *nf_output_iface)
2431 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2433 const struct dst *p;
2436 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2439 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2440 if (cur_vlan == 0) {
2441 cur_vlan = OFP_VLAN_NONE;
2443 for (p = dsts; p < &dsts[n_dsts]; p++) {
2444 if (p->vlan != cur_vlan) {
2445 if (p->vlan == OFP_VLAN_NONE) {
2446 nl_msg_put_flag(actions, ODPAT_STRIP_VLAN);
2449 tci = htons(p->vlan & VLAN_VID_MASK);
2450 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2451 nl_msg_put_be16(actions, ODPAT_SET_DL_TCI, tci);
2455 nl_msg_put_u32(actions, ODPAT_OUTPUT, p->dp_ifidx);
2459 /* Returns the effective vlan of a packet, taking into account both the
2460 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2461 * the packet is untagged and -1 indicates it has an invalid header and
2462 * should be dropped. */
2463 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2464 struct port *in_port, bool have_packet)
2466 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2467 if (in_port->vlan >= 0) {
2469 /* XXX support double tagging? */
2471 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2472 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2473 "packet received on port %s configured with "
2474 "implicit VLAN %"PRIu16,
2475 br->name, vlan, in_port->name, in_port->vlan);
2479 vlan = in_port->vlan;
2481 if (!port_includes_vlan(in_port, vlan)) {
2483 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2484 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2485 "packet received on port %s not configured for "
2487 br->name, vlan, in_port->name, vlan);
2496 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2497 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2498 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2500 is_gratuitous_arp(const struct flow *flow)
2502 return (flow->dl_type == htons(ETH_TYPE_ARP)
2503 && eth_addr_is_broadcast(flow->dl_dst)
2504 && (flow->nw_proto == ARP_OP_REPLY
2505 || (flow->nw_proto == ARP_OP_REQUEST
2506 && flow->nw_src == flow->nw_dst)));
2510 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2511 struct port *in_port)
2513 enum grat_arp_lock_type lock_type;
2516 /* We don't want to learn from gratuitous ARP packets that are reflected
2517 * back over bond slaves so we lock the learning table. */
2518 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2519 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2520 GRAT_ARP_LOCK_CHECK;
2522 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2525 /* The log messages here could actually be useful in debugging,
2526 * so keep the rate limit relatively high. */
2527 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2529 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2530 "on port %s in VLAN %d",
2531 br->name, ETH_ADDR_ARGS(flow->dl_src),
2532 in_port->name, vlan);
2533 ofproto_revalidate(br->ofproto, rev_tag);
2537 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2538 * dropped. Returns true if they may be forwarded, false if they should be
2541 * If 'have_packet' is true, it indicates that the caller is processing a
2542 * received packet. If 'have_packet' is false, then the caller is just
2543 * revalidating an existing flow because configuration has changed. Either
2544 * way, 'have_packet' only affects logging (there is no point in logging errors
2545 * during revalidation).
2547 * Sets '*in_portp' to the input port. This will be a null pointer if
2548 * flow->in_port does not designate a known input port (in which case
2549 * is_admissible() returns false).
2551 * When returning true, sets '*vlanp' to the effective VLAN of the input
2552 * packet, as returned by flow_get_vlan().
2554 * May also add tags to '*tags', although the current implementation only does
2555 * so in one special case.
2558 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2559 tag_type *tags, int *vlanp, struct port **in_portp)
2561 struct iface *in_iface;
2562 struct port *in_port;
2565 /* Find the interface and port structure for the received packet. */
2566 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2568 /* No interface? Something fishy... */
2570 /* Odd. A few possible reasons here:
2572 * - We deleted an interface but there are still a few packets
2573 * queued up from it.
2575 * - Someone externally added an interface (e.g. with "ovs-dpctl
2576 * add-if") that we don't know about.
2578 * - Packet arrived on the local port but the local port is not
2579 * one of our bridge ports.
2581 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2583 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2584 "interface %"PRIu16, br->name, flow->in_port);
2590 *in_portp = in_port = in_iface->port;
2591 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2596 /* Drop frames for reserved multicast addresses. */
2597 if (eth_addr_is_reserved(flow->dl_dst)) {
2601 /* Drop frames on ports reserved for mirroring. */
2602 if (in_port->is_mirror_output_port) {
2604 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2605 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2606 "%s, which is reserved exclusively for mirroring",
2607 br->name, in_port->name);
2612 /* Packets received on bonds need special attention to avoid duplicates. */
2613 if (in_port->n_ifaces > 1) {
2615 bool is_grat_arp_locked;
2617 if (eth_addr_is_multicast(flow->dl_dst)) {
2618 *tags |= in_port->active_iface_tag;
2619 if (in_port->active_iface != in_iface->port_ifidx) {
2620 /* Drop all multicast packets on inactive slaves. */
2625 /* Drop all packets for which we have learned a different input
2626 * port, because we probably sent the packet on one slave and got
2627 * it back on the other. Gratuitous ARP packets are an exception
2628 * to this rule: the host has moved to another switch. The exception
2629 * to the exception is if we locked the learning table to avoid
2630 * reflections on bond slaves. If this is the case, just drop the
2632 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2633 &is_grat_arp_locked);
2634 if (src_idx != -1 && src_idx != in_port->port_idx &&
2635 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2643 /* If the composed actions may be applied to any packet in the given 'flow',
2644 * returns true. Otherwise, the actions should only be applied to 'packet', or
2645 * not at all, if 'packet' was NULL. */
2647 process_flow(struct bridge *br, const struct flow *flow,
2648 const struct ofpbuf *packet, struct ofpbuf *actions,
2649 tag_type *tags, uint16_t *nf_output_iface)
2651 struct port *in_port;
2652 struct port *out_port;
2656 /* Check whether we should drop packets in this flow. */
2657 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2662 /* Learn source MAC (but don't try to learn from revalidation). */
2664 update_learning_table(br, flow, vlan, in_port);
2667 /* Determine output port. */
2668 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2670 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2671 out_port = br->ports[out_port_idx];
2672 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2673 /* If we are revalidating but don't have a learning entry then
2674 * eject the flow. Installing a flow that floods packets opens
2675 * up a window of time where we could learn from a packet reflected
2676 * on a bond and blackhole packets before the learning table is
2677 * updated to reflect the correct port. */
2680 out_port = FLOOD_PORT;
2683 /* Don't send packets out their input ports. */
2684 if (in_port == out_port) {
2690 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2698 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
2699 struct ofpbuf *actions, tag_type *tags,
2700 uint16_t *nf_output_iface, void *br_)
2702 struct iface *iface;
2703 struct bridge *br = br_;
2705 COVERAGE_INC(bridge_process_flow);
2707 iface = iface_from_dp_ifidx(br, flow->in_port);
2709 if (cfm_should_process_flow(flow)) {
2710 if (packet && iface->cfm) {
2711 cfm_process_heartbeat(iface->cfm, packet);
2716 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2720 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
2721 const struct nlattr *actions,
2722 unsigned int actions_len,
2723 unsigned long long int n_bytes, void *br_)
2725 struct bridge *br = br_;
2726 const struct nlattr *a;
2727 struct port *in_port;
2732 /* Feed information from the active flows back into the learning table to
2733 * ensure that table is always in sync with what is actually flowing
2734 * through the datapath.
2736 * We test that 'tags' is nonzero to ensure that only flows that include an
2737 * OFPP_NORMAL action are used for learning. This works because
2738 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2739 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2740 update_learning_table(br, flow, vlan, in_port);
2743 /* Account for bond slave utilization. */
2744 if (!br->has_bonded_ports) {
2747 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
2748 if (nl_attr_type(a) == ODPAT_OUTPUT) {
2749 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
2750 if (out_port && out_port->n_ifaces >= 2) {
2751 uint16_t vlan = (flow->vlan_tci
2752 ? vlan_tci_to_vid(flow->vlan_tci)
2754 struct bond_entry *e = lookup_bond_entry(out_port,
2755 flow->dl_src, vlan);
2756 e->tx_bytes += n_bytes;
2763 bridge_account_checkpoint_ofhook_cb(void *br_)
2765 struct bridge *br = br_;
2769 if (!br->has_bonded_ports) {
2774 for (i = 0; i < br->n_ports; i++) {
2775 struct port *port = br->ports[i];
2776 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2777 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2778 bond_rebalance_port(port);
2783 static struct ofhooks bridge_ofhooks = {
2784 bridge_normal_ofhook_cb,
2785 bridge_account_flow_ofhook_cb,
2786 bridge_account_checkpoint_ofhook_cb,
2789 /* Bonding functions. */
2791 /* Statistics for a single interface on a bonded port, used for load-based
2792 * bond rebalancing. */
2793 struct slave_balance {
2794 struct iface *iface; /* The interface. */
2795 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2797 /* All the "bond_entry"s that are assigned to this interface, in order of
2798 * increasing tx_bytes. */
2799 struct bond_entry **hashes;
2803 /* Sorts pointers to pointers to bond_entries in ascending order by the
2804 * interface to which they are assigned, and within a single interface in
2805 * ascending order of bytes transmitted. */
2807 compare_bond_entries(const void *a_, const void *b_)
2809 const struct bond_entry *const *ap = a_;
2810 const struct bond_entry *const *bp = b_;
2811 const struct bond_entry *a = *ap;
2812 const struct bond_entry *b = *bp;
2813 if (a->iface_idx != b->iface_idx) {
2814 return a->iface_idx > b->iface_idx ? 1 : -1;
2815 } else if (a->tx_bytes != b->tx_bytes) {
2816 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2822 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2823 * *descending* order by number of bytes transmitted. */
2825 compare_slave_balance(const void *a_, const void *b_)
2827 const struct slave_balance *a = a_;
2828 const struct slave_balance *b = b_;
2829 if (a->iface->enabled != b->iface->enabled) {
2830 return a->iface->enabled ? -1 : 1;
2831 } else if (a->tx_bytes != b->tx_bytes) {
2832 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2839 swap_bals(struct slave_balance *a, struct slave_balance *b)
2841 struct slave_balance tmp = *a;
2846 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2847 * given that 'p' (and only 'p') might be in the wrong location.
2849 * This function invalidates 'p', since it might now be in a different memory
2852 resort_bals(struct slave_balance *p,
2853 struct slave_balance bals[], size_t n_bals)
2856 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2857 swap_bals(p, p - 1);
2859 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2860 swap_bals(p, p + 1);
2866 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2868 if (VLOG_IS_DBG_ENABLED()) {
2869 struct ds ds = DS_EMPTY_INITIALIZER;
2870 const struct slave_balance *b;
2872 for (b = bals; b < bals + n_bals; b++) {
2876 ds_put_char(&ds, ',');
2878 ds_put_format(&ds, " %s %"PRIu64"kB",
2879 b->iface->name, b->tx_bytes / 1024);
2881 if (!b->iface->enabled) {
2882 ds_put_cstr(&ds, " (disabled)");
2884 if (b->n_hashes > 0) {
2885 ds_put_cstr(&ds, " (");
2886 for (i = 0; i < b->n_hashes; i++) {
2887 const struct bond_entry *e = b->hashes[i];
2889 ds_put_cstr(&ds, " + ");
2891 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2892 e - port->bond_hash, e->tx_bytes / 1024);
2894 ds_put_cstr(&ds, ")");
2897 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2902 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2904 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2907 struct bond_entry *hash = from->hashes[hash_idx];
2908 struct port *port = from->iface->port;
2909 uint64_t delta = hash->tx_bytes;
2911 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2912 "from %s to %s (now carrying %"PRIu64"kB and "
2913 "%"PRIu64"kB load, respectively)",
2914 port->name, delta / 1024, hash - port->bond_hash,
2915 from->iface->name, to->iface->name,
2916 (from->tx_bytes - delta) / 1024,
2917 (to->tx_bytes + delta) / 1024);
2919 /* Delete element from from->hashes.
2921 * We don't bother to add the element to to->hashes because not only would
2922 * it require more work, the only purpose it would be to allow that hash to
2923 * be migrated to another slave in this rebalancing run, and there is no
2924 * point in doing that. */
2925 if (hash_idx == 0) {
2928 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2929 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2933 /* Shift load away from 'from' to 'to'. */
2934 from->tx_bytes -= delta;
2935 to->tx_bytes += delta;
2937 /* Arrange for flows to be revalidated. */
2938 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2939 hash->iface_idx = to->iface->port_ifidx;
2940 hash->iface_tag = tag_create_random();
2944 bond_rebalance_port(struct port *port)
2946 struct slave_balance bals[DP_MAX_PORTS];
2948 struct bond_entry *hashes[BOND_MASK + 1];
2949 struct slave_balance *b, *from, *to;
2950 struct bond_entry *e;
2953 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2954 * descending order of tx_bytes, so that bals[0] represents the most
2955 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2958 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2959 * array for each slave_balance structure, we sort our local array of
2960 * hashes in order by slave, so that all of the hashes for a given slave
2961 * become contiguous in memory, and then we point each 'hashes' members of
2962 * a slave_balance structure to the start of a contiguous group. */
2963 n_bals = port->n_ifaces;
2964 for (b = bals; b < &bals[n_bals]; b++) {
2965 b->iface = port->ifaces[b - bals];
2970 for (i = 0; i <= BOND_MASK; i++) {
2971 hashes[i] = &port->bond_hash[i];
2973 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2974 for (i = 0; i <= BOND_MASK; i++) {
2976 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2977 b = &bals[e->iface_idx];
2978 b->tx_bytes += e->tx_bytes;
2980 b->hashes = &hashes[i];
2985 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2986 log_bals(bals, n_bals, port);
2988 /* Discard slaves that aren't enabled (which were sorted to the back of the
2989 * array earlier). */
2990 while (!bals[n_bals - 1].iface->enabled) {
2997 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2998 to = &bals[n_bals - 1];
2999 for (from = bals; from < to; ) {
3000 uint64_t overload = from->tx_bytes - to->tx_bytes;
3001 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3002 /* The extra load on 'from' (and all less-loaded slaves), compared
3003 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3004 * it is less than ~1Mbps. No point in rebalancing. */
3006 } else if (from->n_hashes == 1) {
3007 /* 'from' only carries a single MAC hash, so we can't shift any
3008 * load away from it, even though we want to. */
3011 /* 'from' is carrying significantly more load than 'to', and that
3012 * load is split across at least two different hashes. Pick a hash
3013 * to migrate to 'to' (the least-loaded slave), given that doing so
3014 * must decrease the ratio of the load on the two slaves by at
3017 * The sort order we use means that we prefer to shift away the
3018 * smallest hashes instead of the biggest ones. There is little
3019 * reason behind this decision; we could use the opposite sort
3020 * order to shift away big hashes ahead of small ones. */
3023 for (i = 0; i < from->n_hashes; i++) {
3024 double old_ratio, new_ratio;
3025 uint64_t delta = from->hashes[i]->tx_bytes;
3027 if (delta == 0 || from->tx_bytes - delta == 0) {
3028 /* Pointless move. */
3032 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3034 if (to->tx_bytes == 0) {
3035 /* Nothing on the new slave, move it. */
3039 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3040 new_ratio = (double)(from->tx_bytes - delta) /
3041 (to->tx_bytes + delta);
3043 if (new_ratio == 0) {
3044 /* Should already be covered but check to prevent division
3049 if (new_ratio < 1) {
3050 new_ratio = 1 / new_ratio;
3053 if (old_ratio - new_ratio > 0.1) {
3054 /* Would decrease the ratio, move it. */
3058 if (i < from->n_hashes) {
3059 bond_shift_load(from, to, i);
3060 port->bond_compat_is_stale = true;
3062 /* If the result of the migration changed the relative order of
3063 * 'from' and 'to' swap them back to maintain invariants. */
3064 if (order_swapped) {
3065 swap_bals(from, to);
3068 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3069 * point to different slave_balance structures. It is only
3070 * valid to do these two operations in a row at all because we
3071 * know that 'from' will not move past 'to' and vice versa. */
3072 resort_bals(from, bals, n_bals);
3073 resort_bals(to, bals, n_bals);
3080 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3081 * historical data to decay to <1% in 7 rebalancing runs. */
3082 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3088 bond_send_learning_packets(struct port *port)
3090 struct bridge *br = port->bridge;
3091 struct mac_entry *e;
3092 struct ofpbuf packet;
3093 int error, n_packets, n_errors;
3095 if (!port->n_ifaces || port->active_iface < 0) {
3099 ofpbuf_init(&packet, 128);
3100 error = n_packets = n_errors = 0;
3101 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3102 union ofp_action actions[2], *a;
3108 if (e->port == port->port_idx
3109 || !choose_output_iface(port, e->mac, e->vlan, &dp_ifidx, &tags)) {
3113 /* Compose actions. */
3114 memset(actions, 0, sizeof actions);
3117 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
3118 a->vlan_vid.len = htons(sizeof *a);
3119 a->vlan_vid.vlan_vid = htons(e->vlan);
3122 a->output.type = htons(OFPAT_OUTPUT);
3123 a->output.len = htons(sizeof *a);
3124 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
3129 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3131 flow_extract(&packet, 0, ODPP_NONE, &flow);
3132 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
3139 ofpbuf_uninit(&packet);
3142 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3143 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3144 "packets, last error was: %s",
3145 port->name, n_errors, n_packets, strerror(error));
3147 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3148 port->name, n_packets);
3152 /* Bonding unixctl user interface functions. */
3155 bond_unixctl_list(struct unixctl_conn *conn,
3156 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3158 struct ds ds = DS_EMPTY_INITIALIZER;
3159 const struct bridge *br;
3161 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
3163 LIST_FOR_EACH (br, node, &all_bridges) {
3166 for (i = 0; i < br->n_ports; i++) {
3167 const struct port *port = br->ports[i];
3168 if (port->n_ifaces > 1) {
3171 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
3172 for (j = 0; j < port->n_ifaces; j++) {
3173 const struct iface *iface = port->ifaces[j];
3175 ds_put_cstr(&ds, ", ");
3177 ds_put_cstr(&ds, iface->name);
3179 ds_put_char(&ds, '\n');
3183 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3187 static struct port *
3188 bond_find(const char *name)
3190 const struct bridge *br;
3192 LIST_FOR_EACH (br, node, &all_bridges) {
3195 for (i = 0; i < br->n_ports; i++) {
3196 struct port *port = br->ports[i];
3197 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3206 bond_unixctl_show(struct unixctl_conn *conn,
3207 const char *args, void *aux OVS_UNUSED)
3209 struct ds ds = DS_EMPTY_INITIALIZER;
3210 const struct port *port;
3213 port = bond_find(args);
3215 unixctl_command_reply(conn, 501, "no such bond");
3219 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3220 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3221 ds_put_format(&ds, "next rebalance: %lld ms\n",
3222 port->bond_next_rebalance - time_msec());
3223 for (j = 0; j < port->n_ifaces; j++) {
3224 const struct iface *iface = port->ifaces[j];
3225 struct bond_entry *be;
3228 ds_put_format(&ds, "slave %s: %s\n",
3229 iface->name, iface->enabled ? "enabled" : "disabled");
3230 if (j == port->active_iface) {
3231 ds_put_cstr(&ds, "\tactive slave\n");
3233 if (iface->delay_expires != LLONG_MAX) {
3234 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3235 iface->enabled ? "downdelay" : "updelay",
3236 iface->delay_expires - time_msec());
3240 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3241 int hash = be - port->bond_hash;
3242 struct mac_entry *me;
3244 if (be->iface_idx != j) {
3248 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3249 hash, be->tx_bytes / 1024);
3252 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3255 if (bond_hash(me->mac, me->vlan) == hash
3256 && me->port != port->port_idx
3257 && choose_output_iface(port, me->mac, me->vlan,
3259 && dp_ifidx == iface->dp_ifidx)
3261 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3262 ETH_ADDR_ARGS(me->mac));
3267 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3272 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3273 void *aux OVS_UNUSED)
3275 char *args = (char *) args_;
3276 char *save_ptr = NULL;
3277 char *bond_s, *hash_s, *slave_s;
3279 struct iface *iface;
3280 struct bond_entry *entry;
3283 bond_s = strtok_r(args, " ", &save_ptr);
3284 hash_s = strtok_r(NULL, " ", &save_ptr);
3285 slave_s = strtok_r(NULL, " ", &save_ptr);
3287 unixctl_command_reply(conn, 501,
3288 "usage: bond/migrate BOND HASH SLAVE");
3292 port = bond_find(bond_s);
3294 unixctl_command_reply(conn, 501, "no such bond");
3298 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3299 hash = atoi(hash_s) & BOND_MASK;
3301 unixctl_command_reply(conn, 501, "bad hash");
3305 iface = port_lookup_iface(port, slave_s);
3307 unixctl_command_reply(conn, 501, "no such slave");
3311 if (!iface->enabled) {
3312 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3316 entry = &port->bond_hash[hash];
3317 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3318 entry->iface_idx = iface->port_ifidx;
3319 entry->iface_tag = tag_create_random();
3320 port->bond_compat_is_stale = true;
3321 unixctl_command_reply(conn, 200, "migrated");
3325 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3326 void *aux OVS_UNUSED)
3328 char *args = (char *) args_;
3329 char *save_ptr = NULL;
3330 char *bond_s, *slave_s;
3332 struct iface *iface;
3334 bond_s = strtok_r(args, " ", &save_ptr);
3335 slave_s = strtok_r(NULL, " ", &save_ptr);
3337 unixctl_command_reply(conn, 501,
3338 "usage: bond/set-active-slave BOND SLAVE");
3342 port = bond_find(bond_s);
3344 unixctl_command_reply(conn, 501, "no such bond");
3348 iface = port_lookup_iface(port, slave_s);
3350 unixctl_command_reply(conn, 501, "no such slave");
3354 if (!iface->enabled) {
3355 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3359 if (port->active_iface != iface->port_ifidx) {
3360 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3361 port->active_iface = iface->port_ifidx;
3362 port->active_iface_tag = tag_create_random();
3363 VLOG_INFO("port %s: active interface is now %s",
3364 port->name, iface->name);
3365 bond_send_learning_packets(port);
3366 unixctl_command_reply(conn, 200, "done");
3368 unixctl_command_reply(conn, 200, "no change");
3373 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3375 char *args = (char *) args_;
3376 char *save_ptr = NULL;
3377 char *bond_s, *slave_s;
3379 struct iface *iface;
3381 bond_s = strtok_r(args, " ", &save_ptr);
3382 slave_s = strtok_r(NULL, " ", &save_ptr);
3384 unixctl_command_reply(conn, 501,
3385 "usage: bond/enable/disable-slave BOND SLAVE");
3389 port = bond_find(bond_s);
3391 unixctl_command_reply(conn, 501, "no such bond");
3395 iface = port_lookup_iface(port, slave_s);
3397 unixctl_command_reply(conn, 501, "no such slave");
3401 bond_enable_slave(iface, enable);
3402 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3406 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3407 void *aux OVS_UNUSED)
3409 enable_slave(conn, args, true);
3413 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3414 void *aux OVS_UNUSED)
3416 enable_slave(conn, args, false);
3420 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3421 void *aux OVS_UNUSED)
3423 char *args = (char *) args_;
3424 uint8_t mac[ETH_ADDR_LEN];
3428 char *mac_s, *vlan_s;
3429 char *save_ptr = NULL;
3431 mac_s = strtok_r(args, " ", &save_ptr);
3432 vlan_s = strtok_r(NULL, " ", &save_ptr);
3435 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3436 unixctl_command_reply(conn, 501, "invalid vlan");
3440 vlan = OFP_VLAN_NONE;
3443 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3444 == ETH_ADDR_SCAN_COUNT) {
3445 hash = bond_hash(mac, vlan);
3447 hash_cstr = xasprintf("%u", hash);
3448 unixctl_command_reply(conn, 200, hash_cstr);
3451 unixctl_command_reply(conn, 501, "invalid mac");
3458 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3459 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3460 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3461 unixctl_command_register("bond/set-active-slave",
3462 bond_unixctl_set_active_slave, NULL);
3463 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3465 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3467 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3470 /* Port functions. */
3472 static struct port *
3473 port_create(struct bridge *br, const char *name)
3477 port = xzalloc(sizeof *port);
3479 port->port_idx = br->n_ports;
3481 port->trunks = NULL;
3482 port->name = xstrdup(name);
3483 port->active_iface = -1;
3485 if (br->n_ports >= br->allocated_ports) {
3486 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3489 br->ports[br->n_ports++] = port;
3490 shash_add_assert(&br->port_by_name, port->name, port);
3492 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3499 get_port_other_config(const struct ovsrec_port *port, const char *key,
3500 const char *default_value)
3504 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3506 return value ? value : default_value;
3510 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3512 struct shash new_ifaces;
3515 /* Collect list of new interfaces. */
3516 shash_init(&new_ifaces);
3517 for (i = 0; i < cfg->n_interfaces; i++) {
3518 const char *name = cfg->interfaces[i]->name;
3519 shash_add_once(&new_ifaces, name, NULL);
3522 /* Get rid of deleted interfaces. */
3523 for (i = 0; i < port->n_ifaces; ) {
3524 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3525 iface_destroy(port->ifaces[i]);
3531 shash_destroy(&new_ifaces);
3535 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3537 struct shash new_ifaces;
3538 long long int next_rebalance;
3539 unsigned long *trunks;
3545 /* Update settings. */
3546 port->updelay = cfg->bond_updelay;
3547 if (port->updelay < 0) {
3550 port->downdelay = cfg->bond_downdelay;
3551 if (port->downdelay < 0) {
3552 port->downdelay = 0;
3554 port->bond_rebalance_interval = atoi(
3555 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3556 if (port->bond_rebalance_interval < 1000) {
3557 port->bond_rebalance_interval = 1000;
3559 next_rebalance = time_msec() + port->bond_rebalance_interval;
3560 if (port->bond_next_rebalance > next_rebalance) {
3561 port->bond_next_rebalance = next_rebalance;
3564 /* Add new interfaces and update 'cfg' member of existing ones. */
3565 shash_init(&new_ifaces);
3566 for (i = 0; i < cfg->n_interfaces; i++) {
3567 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3568 struct iface *iface;
3570 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3571 VLOG_WARN("port %s: %s specified twice as port interface",
3572 port->name, if_cfg->name);
3573 iface_set_ofport(if_cfg, -1);
3577 iface = iface_lookup(port->bridge, if_cfg->name);
3579 if (iface->port != port) {
3580 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3582 port->bridge->name, if_cfg->name, iface->port->name);
3585 iface->cfg = if_cfg;
3587 iface = iface_create(port, if_cfg);
3590 /* Determine interface type. The local port always has type
3591 * "internal". Other ports take their type from the database and
3592 * default to "system" if none is specified. */
3593 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
3594 : if_cfg->type[0] ? if_cfg->type
3597 shash_destroy(&new_ifaces);
3602 if (port->n_ifaces < 2) {
3604 if (vlan >= 0 && vlan <= 4095) {
3605 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3610 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3611 * they even work as-is. But they have not been tested. */
3612 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3616 if (port->vlan != vlan) {
3618 bridge_flush(port->bridge);
3621 /* Get trunked VLANs. */
3623 if (vlan < 0 && cfg->n_trunks) {
3626 trunks = bitmap_allocate(4096);
3628 for (i = 0; i < cfg->n_trunks; i++) {
3629 int trunk = cfg->trunks[i];
3631 bitmap_set1(trunks, trunk);
3637 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3638 port->name, cfg->n_trunks);
3640 if (n_errors == cfg->n_trunks) {
3641 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3643 bitmap_free(trunks);
3646 } else if (vlan >= 0 && cfg->n_trunks) {
3647 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3651 ? port->trunks != NULL
3652 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3653 bridge_flush(port->bridge);
3655 bitmap_free(port->trunks);
3656 port->trunks = trunks;
3660 port_destroy(struct port *port)
3663 struct bridge *br = port->bridge;
3667 proc_net_compat_update_vlan(port->name, NULL, 0);
3668 proc_net_compat_update_bond(port->name, NULL);
3670 for (i = 0; i < MAX_MIRRORS; i++) {
3671 struct mirror *m = br->mirrors[i];
3672 if (m && m->out_port == port) {
3677 while (port->n_ifaces > 0) {
3678 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3681 shash_find_and_delete_assert(&br->port_by_name, port->name);
3683 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3684 del->port_idx = port->port_idx;
3686 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
3688 netdev_monitor_destroy(port->monitor);
3690 bitmap_free(port->trunks);
3697 static struct port *
3698 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3700 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3701 return iface ? iface->port : NULL;
3704 static struct port *
3705 port_lookup(const struct bridge *br, const char *name)
3707 return shash_find_data(&br->port_by_name, name);
3710 static struct iface *
3711 port_lookup_iface(const struct port *port, const char *name)
3713 struct iface *iface = iface_lookup(port->bridge, name);
3714 return iface && iface->port == port ? iface : NULL;
3718 port_update_bonding(struct port *port)
3720 if (port->monitor) {
3721 netdev_monitor_destroy(port->monitor);
3722 port->monitor = NULL;
3724 if (port->n_ifaces < 2) {
3725 /* Not a bonded port. */
3726 if (port->bond_hash) {
3727 free(port->bond_hash);
3728 port->bond_hash = NULL;
3729 port->bond_compat_is_stale = true;
3730 port->bond_fake_iface = false;
3735 if (!port->bond_hash) {
3736 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3737 for (i = 0; i <= BOND_MASK; i++) {
3738 struct bond_entry *e = &port->bond_hash[i];
3742 port->no_ifaces_tag = tag_create_random();
3743 bond_choose_active_iface(port);
3744 port->bond_next_rebalance
3745 = time_msec() + port->bond_rebalance_interval;
3747 if (port->cfg->bond_fake_iface) {
3748 port->bond_next_fake_iface_update = time_msec();
3751 port->bond_compat_is_stale = true;
3752 port->bond_fake_iface = port->cfg->bond_fake_iface;
3754 port->monitor = netdev_monitor_create();
3755 for (i = 0; i < port->n_ifaces; i++) {
3756 netdev_monitor_add(port->monitor, port->ifaces[i]->netdev);
3762 port_update_bond_compat(struct port *port)
3764 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3765 struct compat_bond bond;
3768 if (port->n_ifaces < 2) {
3769 proc_net_compat_update_bond(port->name, NULL);
3774 bond.updelay = port->updelay;
3775 bond.downdelay = port->downdelay;
3778 bond.hashes = compat_hashes;
3779 if (port->bond_hash) {
3780 const struct bond_entry *e;
3781 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3782 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3783 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3784 cbh->hash = e - port->bond_hash;
3785 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3790 bond.n_slaves = port->n_ifaces;
3791 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3792 for (i = 0; i < port->n_ifaces; i++) {
3793 struct iface *iface = port->ifaces[i];
3794 struct compat_bond_slave *slave = &bond.slaves[i];
3795 slave->name = iface->name;
3797 /* We need to make the same determination as the Linux bonding
3798 * code to determine whether a slave should be consider "up".
3799 * The Linux function bond_miimon_inspect() supports four
3800 * BOND_LINK_* states:
3802 * - BOND_LINK_UP: carrier detected, updelay has passed.
3803 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3804 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3805 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3807 * The function bond_info_show_slave() only considers BOND_LINK_UP
3808 * to be "up" and anything else to be "down".
3810 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3814 netdev_get_etheraddr(iface->netdev, slave->mac);
3817 if (port->bond_fake_iface) {
3818 struct netdev *bond_netdev;
3820 if (!netdev_open_default(port->name, &bond_netdev)) {
3822 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3824 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3826 netdev_close(bond_netdev);
3830 proc_net_compat_update_bond(port->name, &bond);
3835 port_update_vlan_compat(struct port *port)
3837 struct bridge *br = port->bridge;
3838 char *vlandev_name = NULL;
3840 if (port->vlan > 0) {
3841 /* Figure out the name that the VLAN device should actually have, if it
3842 * existed. This takes some work because the VLAN device would not
3843 * have port->name in its name; rather, it would have the trunk port's
3844 * name, and 'port' would be attached to a bridge that also had the
3845 * VLAN device one of its ports. So we need to find a trunk port that
3846 * includes port->vlan.
3848 * There might be more than one candidate. This doesn't happen on
3849 * XenServer, so if it happens we just pick the first choice in
3850 * alphabetical order instead of creating multiple VLAN devices. */
3852 for (i = 0; i < br->n_ports; i++) {
3853 struct port *p = br->ports[i];
3854 if (port_trunks_vlan(p, port->vlan)
3856 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3858 uint8_t ea[ETH_ADDR_LEN];
3859 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3860 if (!eth_addr_is_multicast(ea) &&
3861 !eth_addr_is_reserved(ea) &&
3862 !eth_addr_is_zero(ea)) {
3863 vlandev_name = p->name;
3868 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3871 /* Interface functions. */
3874 iface_send_packet(struct iface *iface, struct ofpbuf *packet)
3877 union ofp_action action;
3879 memset(&action, 0, sizeof action);
3880 action.output.type = htons(OFPAT_OUTPUT);
3881 action.output.len = htons(sizeof action);
3882 action.output.port = htons(odp_port_to_ofp_port(iface->dp_ifidx));
3884 flow_extract(packet, 0, ODPP_NONE, &flow);
3886 if (ofproto_send_packet(iface->port->bridge->ofproto, &flow, &action, 1,
3888 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3889 VLOG_WARN_RL(&rl, "interface %s: Failed to send packet.", iface->name);
3893 static struct iface *
3894 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3896 struct bridge *br = port->bridge;
3897 struct iface *iface;
3898 char *name = if_cfg->name;
3900 iface = xzalloc(sizeof *iface);
3902 iface->port_ifidx = port->n_ifaces;
3903 iface->name = xstrdup(name);
3904 iface->dp_ifidx = -1;
3905 iface->tag = tag_create_random();
3906 iface->delay_expires = LLONG_MAX;
3907 iface->netdev = NULL;
3908 iface->cfg = if_cfg;
3910 shash_add_assert(&br->iface_by_name, iface->name, iface);
3912 if (port->n_ifaces >= port->allocated_ifaces) {
3913 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3914 sizeof *port->ifaces);
3916 port->ifaces[port->n_ifaces++] = iface;
3917 if (port->n_ifaces > 1) {
3918 br->has_bonded_ports = true;
3921 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3929 iface_destroy(struct iface *iface)
3932 struct port *port = iface->port;
3933 struct bridge *br = port->bridge;
3934 bool del_active = port->active_iface == iface->port_ifidx;
3937 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3939 if (iface->dp_ifidx >= 0) {
3940 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
3943 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3944 del->port_ifidx = iface->port_ifidx;
3946 netdev_close(iface->netdev);
3949 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3950 bond_choose_active_iface(port);
3951 bond_send_learning_packets(port);
3954 cfm_destroy(iface->cfm);
3959 bridge_flush(port->bridge);
3963 static struct iface *
3964 iface_lookup(const struct bridge *br, const char *name)
3966 return shash_find_data(&br->iface_by_name, name);
3969 static struct iface *
3970 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3972 struct iface *iface;
3974 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
3975 hash_int(dp_ifidx, 0), &br->ifaces) {
3976 if (iface->dp_ifidx == dp_ifidx) {
3983 /* Set Ethernet address of 'iface', if one is specified in the configuration
3986 iface_set_mac(struct iface *iface)
3988 uint8_t ea[ETH_ADDR_LEN];
3990 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3991 if (eth_addr_is_multicast(ea)) {
3992 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3994 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3995 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3996 iface->name, iface->name);
3998 int error = netdev_set_etheraddr(iface->netdev, ea);
4000 VLOG_ERR("interface %s: setting MAC failed (%s)",
4001 iface->name, strerror(error));
4007 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4009 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4012 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4016 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4018 * The value strings in '*shash' are taken directly from values[], not copied,
4019 * so the caller should not modify or free them. */
4021 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4022 struct shash *shash)
4027 for (i = 0; i < n; i++) {
4028 shash_add(shash, keys[i], values[i]);
4032 struct iface_delete_queues_cbdata {
4033 struct netdev *netdev;
4034 const struct ovsdb_datum *queues;
4038 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4040 union ovsdb_atom atom;
4042 atom.integer = target;
4043 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4047 iface_delete_queues(unsigned int queue_id,
4048 const struct shash *details OVS_UNUSED, void *cbdata_)
4050 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4052 if (!queue_ids_include(cbdata->queues, queue_id)) {
4053 netdev_delete_queue(cbdata->netdev, queue_id);
4058 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4060 if (!qos || qos->type[0] == '\0') {
4061 netdev_set_qos(iface->netdev, NULL, NULL);
4063 struct iface_delete_queues_cbdata cbdata;
4064 struct shash details;
4067 /* Configure top-level Qos for 'iface'. */
4068 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4069 qos->n_other_config, &details);
4070 netdev_set_qos(iface->netdev, qos->type, &details);
4071 shash_destroy(&details);
4073 /* Deconfigure queues that were deleted. */
4074 cbdata.netdev = iface->netdev;
4075 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4077 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4079 /* Configure queues for 'iface'. */
4080 for (i = 0; i < qos->n_queues; i++) {
4081 const struct ovsrec_queue *queue = qos->value_queues[i];
4082 unsigned int queue_id = qos->key_queues[i];
4084 shash_from_ovs_idl_map(queue->key_other_config,
4085 queue->value_other_config,
4086 queue->n_other_config, &details);
4087 netdev_set_queue(iface->netdev, queue_id, &details);
4088 shash_destroy(&details);
4094 iface_update_cfm(struct iface *iface)
4098 uint16_t *remote_mps;
4099 struct ovsrec_monitor *mon;
4100 uint8_t ea[ETH_ADDR_LEN], maid[CCM_MAID_LEN];
4102 mon = iface->cfg->monitor;
4108 if (netdev_get_etheraddr(iface->netdev, ea)) {
4109 VLOG_WARN("interface %s: Failed to get ethernet address. "
4110 "Skipping Monitor.", iface->name);
4114 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4115 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4120 iface->cfm = cfm_create();
4124 cfm->mpid = mon->mpid;
4125 cfm->interval = mon->interval ? *mon->interval : 1000;
4127 memcpy(cfm->eth_src, ea, sizeof cfm->eth_src);
4128 memcpy(cfm->maid, maid, sizeof cfm->maid);
4130 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4131 for(i = 0; i < mon->n_remote_mps; i++) {
4132 remote_mps[i] = mon->remote_mps[i]->mpid;
4134 cfm_update_remote_mps(cfm, remote_mps, mon->n_remote_mps);
4137 if (!cfm_configure(iface->cfm)) {
4138 cfm_destroy(iface->cfm);
4143 /* Port mirroring. */
4145 static struct mirror *
4146 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4150 for (i = 0; i < MAX_MIRRORS; i++) {
4151 struct mirror *m = br->mirrors[i];
4152 if (m && uuid_equals(uuid, &m->uuid)) {
4160 mirror_reconfigure(struct bridge *br)
4162 unsigned long *rspan_vlans;
4165 /* Get rid of deleted mirrors. */
4166 for (i = 0; i < MAX_MIRRORS; i++) {
4167 struct mirror *m = br->mirrors[i];
4169 const struct ovsdb_datum *mc;
4170 union ovsdb_atom atom;
4172 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4173 atom.uuid = br->mirrors[i]->uuid;
4174 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4180 /* Add new mirrors and reconfigure existing ones. */
4181 for (i = 0; i < br->cfg->n_mirrors; i++) {
4182 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4183 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4185 mirror_reconfigure_one(m, cfg);
4187 mirror_create(br, cfg);
4191 /* Update port reserved status. */
4192 for (i = 0; i < br->n_ports; i++) {
4193 br->ports[i]->is_mirror_output_port = false;
4195 for (i = 0; i < MAX_MIRRORS; i++) {
4196 struct mirror *m = br->mirrors[i];
4197 if (m && m->out_port) {
4198 m->out_port->is_mirror_output_port = true;
4202 /* Update flooded vlans (for RSPAN). */
4204 if (br->cfg->n_flood_vlans) {
4205 rspan_vlans = bitmap_allocate(4096);
4207 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4208 int64_t vlan = br->cfg->flood_vlans[i];
4209 if (vlan >= 0 && vlan < 4096) {
4210 bitmap_set1(rspan_vlans, vlan);
4211 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4214 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4219 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4225 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4230 for (i = 0; ; i++) {
4231 if (i >= MAX_MIRRORS) {
4232 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4233 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4236 if (!br->mirrors[i]) {
4241 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4244 br->mirrors[i] = m = xzalloc(sizeof *m);
4247 m->name = xstrdup(cfg->name);
4248 shash_init(&m->src_ports);
4249 shash_init(&m->dst_ports);
4255 mirror_reconfigure_one(m, cfg);
4259 mirror_destroy(struct mirror *m)
4262 struct bridge *br = m->bridge;
4265 for (i = 0; i < br->n_ports; i++) {
4266 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4267 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4270 shash_destroy(&m->src_ports);
4271 shash_destroy(&m->dst_ports);
4274 m->bridge->mirrors[m->idx] = NULL;
4283 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4284 struct shash *names)
4288 for (i = 0; i < n_ports; i++) {
4289 const char *name = ports[i]->name;
4290 if (port_lookup(m->bridge, name)) {
4291 shash_add_once(names, name, NULL);
4293 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4294 "port %s", m->bridge->name, m->name, name);
4300 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4306 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4308 for (i = 0; i < cfg->n_select_vlan; i++) {
4309 int64_t vlan = cfg->select_vlan[i];
4310 if (vlan < 0 || vlan > 4095) {
4311 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4312 m->bridge->name, m->name, vlan);
4314 (*vlans)[n_vlans++] = vlan;
4321 vlan_is_mirrored(const struct mirror *m, int vlan)
4325 for (i = 0; i < m->n_vlans; i++) {
4326 if (m->vlans[i] == vlan) {
4334 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4338 for (i = 0; i < m->n_vlans; i++) {
4339 if (port_trunks_vlan(p, m->vlans[i])) {
4347 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4349 struct shash src_ports, dst_ports;
4350 mirror_mask_t mirror_bit;
4351 struct port *out_port;
4358 if (strcmp(cfg->name, m->name)) {
4360 m->name = xstrdup(cfg->name);
4363 /* Get output port. */
4364 if (cfg->output_port) {
4365 out_port = port_lookup(m->bridge, cfg->output_port->name);
4367 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4368 m->bridge->name, m->name);
4374 if (cfg->output_vlan) {
4375 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4376 "output vlan; ignoring output vlan",
4377 m->bridge->name, m->name);
4379 } else if (cfg->output_vlan) {
4381 out_vlan = *cfg->output_vlan;
4383 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4384 m->bridge->name, m->name);
4389 shash_init(&src_ports);
4390 shash_init(&dst_ports);
4391 if (cfg->select_all) {
4392 for (i = 0; i < m->bridge->n_ports; i++) {
4393 const char *name = m->bridge->ports[i]->name;
4394 shash_add_once(&src_ports, name, NULL);
4395 shash_add_once(&dst_ports, name, NULL);
4400 /* Get ports, and drop duplicates and ports that don't exist. */
4401 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4403 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4406 /* Get all the vlans, and drop duplicate and invalid vlans. */
4407 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4410 /* Update mirror data. */
4411 if (!shash_equal_keys(&m->src_ports, &src_ports)
4412 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4413 || m->n_vlans != n_vlans
4414 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4415 || m->out_port != out_port
4416 || m->out_vlan != out_vlan) {
4417 bridge_flush(m->bridge);
4419 shash_swap(&m->src_ports, &src_ports);
4420 shash_swap(&m->dst_ports, &dst_ports);
4423 m->n_vlans = n_vlans;
4424 m->out_port = out_port;
4425 m->out_vlan = out_vlan;
4428 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4429 for (i = 0; i < m->bridge->n_ports; i++) {
4430 struct port *port = m->bridge->ports[i];
4432 if (shash_find(&m->src_ports, port->name)
4435 ? port_trunks_any_mirrored_vlan(m, port)
4436 : vlan_is_mirrored(m, port->vlan)))) {
4437 port->src_mirrors |= mirror_bit;
4439 port->src_mirrors &= ~mirror_bit;
4442 if (shash_find(&m->dst_ports, port->name)) {
4443 port->dst_mirrors |= mirror_bit;
4445 port->dst_mirrors &= ~mirror_bit;
4450 shash_destroy(&src_ports);
4451 shash_destroy(&dst_ports);