1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
23 #include <sys/socket.h>
25 #include <openflow/openflow.h>
30 #include <sys/socket.h>
31 #include <sys/types.h>
37 #include "dynamic-string.h"
42 #include "mac-learning.h"
45 #include "ofp-print.h"
47 #include "ofproto/netflow.h"
48 #include "ofproto/ofproto.h"
50 #include "poll-loop.h"
51 #include "port-array.h"
52 #include "proc-net-compat.h"
56 #include "socket-util.h"
57 #include "stream-ssl.h"
63 #include "vswitchd/vswitch-idl.h"
64 #include "xenserver.h"
66 #include "sflow_api.h"
68 #define THIS_MODULE VLM_bridge
77 /* These members are always valid. */
78 struct port *port; /* Containing port. */
79 size_t port_ifidx; /* Index within containing port. */
80 char *name; /* Host network device name. */
81 tag_type tag; /* Tag associated with this interface. */
82 long long delay_expires; /* Time after which 'enabled' may change. */
84 /* These members are valid only after bridge_reconfigure() causes them to
86 int dp_ifidx; /* Index within kernel datapath. */
87 struct netdev *netdev; /* Network device. */
88 bool enabled; /* May be chosen for flows? */
90 /* This member is only valid *during* bridge_reconfigure(). */
91 const struct ovsrec_interface *cfg;
94 #define BOND_MASK 0xff
96 int iface_idx; /* Index of assigned iface, or -1 if none. */
97 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
98 tag_type iface_tag; /* Tag associated with iface_idx. */
101 #define MAX_MIRRORS 32
102 typedef uint32_t mirror_mask_t;
103 #define MIRROR_MASK_C(X) UINT32_C(X)
104 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
106 struct bridge *bridge;
110 /* Selection criteria. */
111 struct shash src_ports; /* Name is port name; data is always NULL. */
112 struct shash dst_ports; /* Name is port name; data is always NULL. */
117 struct port *out_port;
121 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
123 struct bridge *bridge;
125 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
126 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
127 * NULL if all VLANs are trunked. */
130 /* An ordinary bridge port has 1 interface.
131 * A bridge port for bonding has at least 2 interfaces. */
132 struct iface **ifaces;
133 size_t n_ifaces, allocated_ifaces;
136 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
137 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
138 tag_type active_iface_tag; /* Tag for bcast flows. */
139 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
140 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
141 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
142 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
143 long bond_next_fake_iface_update; /* Next update to fake bond stats. */
144 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
145 long long int bond_next_rebalance; /* Next rebalancing time. */
147 /* Port mirroring info. */
148 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
149 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
150 bool is_mirror_output_port; /* Does port mirroring send frames here? */
152 /* This member is only valid *during* bridge_reconfigure(). */
153 const struct ovsrec_port *cfg;
156 #define DP_MAX_PORTS 255
158 struct list node; /* Node in global list of bridges. */
159 char *name; /* User-specified arbitrary name. */
160 struct mac_learning *ml; /* MAC learning table. */
161 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
163 /* OpenFlow switch processing. */
164 struct ofproto *ofproto; /* OpenFlow switch. */
166 /* Kernel datapath information. */
167 struct dpif *dpif; /* Datapath. */
168 struct port_array ifaces; /* Indexed by kernel datapath port number. */
172 size_t n_ports, allocated_ports;
173 struct shash iface_by_name; /* "struct iface"s indexed by name. */
174 struct shash port_by_name; /* "struct port"s indexed by name. */
177 bool has_bonded_ports;
182 /* Port mirroring. */
183 struct mirror *mirrors[MAX_MIRRORS];
185 /* This member is only valid *during* bridge_reconfigure(). */
186 const struct ovsrec_bridge *cfg;
189 /* List of all bridges. */
190 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
192 /* Maximum number of datapaths. */
193 enum { DP_MAX = 256 };
195 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
196 static void bridge_destroy(struct bridge *);
197 static struct bridge *bridge_lookup(const char *name);
198 static unixctl_cb_func bridge_unixctl_dump_flows;
199 static int bridge_run_one(struct bridge *);
200 static size_t bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
201 const struct bridge *br,
202 struct ovsrec_controller ***controllersp);
203 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
205 static void bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *,
207 const struct sockaddr_in *managers,
209 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
210 static void bridge_fetch_dp_ifaces(struct bridge *);
211 static void bridge_flush(struct bridge *);
212 static void bridge_pick_local_hw_addr(struct bridge *,
213 uint8_t ea[ETH_ADDR_LEN],
214 struct iface **hw_addr_iface);
215 static uint64_t bridge_pick_datapath_id(struct bridge *,
216 const uint8_t bridge_ea[ETH_ADDR_LEN],
217 struct iface *hw_addr_iface);
218 static struct iface *bridge_get_local_iface(struct bridge *);
219 static uint64_t dpid_from_hash(const void *, size_t nbytes);
221 static unixctl_cb_func bridge_unixctl_fdb_show;
223 static void bond_init(void);
224 static void bond_run(struct bridge *);
225 static void bond_wait(struct bridge *);
226 static void bond_rebalance_port(struct port *);
227 static void bond_send_learning_packets(struct port *);
228 static void bond_enable_slave(struct iface *iface, bool enable);
230 static struct port *port_create(struct bridge *, const char *name);
231 static void port_reconfigure(struct port *, const struct ovsrec_port *);
232 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
233 static void port_destroy(struct port *);
234 static struct port *port_lookup(const struct bridge *, const char *name);
235 static struct iface *port_lookup_iface(const struct port *, const char *name);
236 static struct port *port_from_dp_ifidx(const struct bridge *,
238 static void port_update_bond_compat(struct port *);
239 static void port_update_vlan_compat(struct port *);
240 static void port_update_bonding(struct port *);
242 static struct mirror *mirror_create(struct bridge *, const char *name);
243 static void mirror_destroy(struct mirror *);
244 static void mirror_reconfigure(struct bridge *);
245 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
246 static bool vlan_is_mirrored(const struct mirror *, int vlan);
248 static struct iface *iface_create(struct port *port,
249 const struct ovsrec_interface *if_cfg);
250 static void iface_destroy(struct iface *);
251 static struct iface *iface_lookup(const struct bridge *, const char *name);
252 static struct iface *iface_from_dp_ifidx(const struct bridge *,
254 static bool iface_is_internal(const struct bridge *, const char *name);
255 static void iface_set_mac(struct iface *);
257 /* Hooks into ofproto processing. */
258 static struct ofhooks bridge_ofhooks;
260 /* Public functions. */
262 /* Adds the name of each interface used by a bridge, including local and
263 * internal ports, to 'svec'. */
265 bridge_get_ifaces(struct svec *svec)
267 struct bridge *br, *next;
270 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
271 for (i = 0; i < br->n_ports; i++) {
272 struct port *port = br->ports[i];
274 for (j = 0; j < port->n_ifaces; j++) {
275 struct iface *iface = port->ifaces[j];
276 if (iface->dp_ifidx < 0) {
277 VLOG_ERR("%s interface not in datapath %s, ignoring",
278 iface->name, dpif_name(br->dpif));
280 if (iface->dp_ifidx != ODPP_LOCAL) {
281 svec_add(svec, iface->name);
290 bridge_init(const struct ovsrec_open_vswitch *cfg)
292 struct svec bridge_names;
293 struct svec dpif_names, dpif_types;
296 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
298 svec_init(&bridge_names);
299 for (i = 0; i < cfg->n_bridges; i++) {
300 svec_add(&bridge_names, cfg->bridges[i]->name);
302 svec_sort(&bridge_names);
304 svec_init(&dpif_names);
305 svec_init(&dpif_types);
306 dp_enumerate_types(&dpif_types);
307 for (i = 0; i < dpif_types.n; i++) {
312 dp_enumerate_names(dpif_types.names[i], &dpif_names);
314 for (j = 0; j < dpif_names.n; j++) {
315 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
317 struct svec all_names;
320 svec_init(&all_names);
321 dpif_get_all_names(dpif, &all_names);
322 for (k = 0; k < all_names.n; k++) {
323 if (svec_contains(&bridge_names, all_names.names[k])) {
329 svec_destroy(&all_names);
334 svec_destroy(&bridge_names);
335 svec_destroy(&dpif_names);
336 svec_destroy(&dpif_types);
338 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
342 bridge_reconfigure(cfg);
347 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
349 /* XXX SSL should be configurable on a per-bridge basis. */
351 stream_ssl_set_private_key_file(ssl->private_key);
352 stream_ssl_set_certificate_file(ssl->certificate);
353 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
358 /* Attempt to create the network device 'iface_name' through the netdev
361 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
364 struct shash_node *node;
365 struct shash options;
369 shash_init(&options);
370 for (i = 0; i < iface_cfg->n_options; i++) {
371 shash_add(&options, iface_cfg->key_options[i],
372 xstrdup(iface_cfg->value_options[i]));
376 struct netdev_options netdev_options;
378 memset(&netdev_options, 0, sizeof netdev_options);
379 netdev_options.name = iface_cfg->name;
380 if (!strcmp(iface_cfg->type, "internal")) {
381 /* An "internal" config type maps to a netdev "system" type. */
382 netdev_options.type = "system";
384 netdev_options.type = iface_cfg->type;
386 netdev_options.args = &options;
387 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
389 error = netdev_open(&netdev_options, &iface->netdev);
392 netdev_get_carrier(iface->netdev, &iface->enabled);
394 } else if (iface->netdev) {
395 const char *netdev_type = netdev_get_type(iface->netdev);
396 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
397 ? iface_cfg->type : NULL;
399 /* An "internal" config type maps to a netdev "system" type. */
400 if (iface_type && !strcmp(iface_type, "internal")) {
401 iface_type = "system";
404 if (!iface_type || !strcmp(netdev_type, iface_type)) {
405 error = netdev_reconfigure(iface->netdev, &options);
407 VLOG_WARN("%s: attempting change device type from %s to %s",
408 iface_cfg->name, netdev_type, iface_type);
413 SHASH_FOR_EACH (node, &options) {
416 shash_destroy(&options);
422 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
424 return set_up_iface(iface_cfg, iface, false);
428 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
429 void *aux OVS_UNUSED)
431 if (!iface->netdev) {
432 int error = set_up_iface(iface->cfg, iface, true);
434 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
444 check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
445 void *aux OVS_UNUSED)
447 if (iface->dp_ifidx >= 0) {
448 VLOG_DBG("%s has interface %s on port %d",
450 iface->name, iface->dp_ifidx);
453 VLOG_ERR("%s interface not in %s, dropping",
454 iface->name, dpif_name(br->dpif));
460 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
461 void *aux OVS_UNUSED)
463 /* Set policing attributes. */
464 netdev_set_policing(iface->netdev,
465 iface->cfg->ingress_policing_rate,
466 iface->cfg->ingress_policing_burst);
468 /* Set MAC address of internal interfaces other than the local
470 if (iface->dp_ifidx != ODPP_LOCAL
471 && iface_is_internal(br, iface->name)) {
472 iface_set_mac(iface);
478 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
479 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
480 * deletes from 'br' any ports that no longer have any interfaces. */
482 iterate_and_prune_ifaces(struct bridge *br,
483 bool (*cb)(struct bridge *, struct iface *,
489 for (i = 0; i < br->n_ports; ) {
490 struct port *port = br->ports[i];
491 for (j = 0; j < port->n_ifaces; ) {
492 struct iface *iface = port->ifaces[j];
493 if (cb(br, iface, aux)) {
496 iface_destroy(iface);
500 if (port->n_ifaces) {
503 VLOG_ERR("%s port has no interfaces, dropping", port->name);
509 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
510 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
511 * responsible for freeing '*managersp' (with free()).
513 * You may be asking yourself "why does ovs-vswitchd care?", because
514 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
515 * should not be and in fact is not directly involved in that. But
516 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
517 * it has to tell in-band control where the managers are to enable that.
520 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
521 struct sockaddr_in **managersp, size_t *n_managersp)
523 struct sockaddr_in *managers = NULL;
524 size_t n_managers = 0;
526 if (ovs_cfg->n_managers > 0) {
529 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
530 for (i = 0; i < ovs_cfg->n_managers; i++) {
531 const char *name = ovs_cfg->managers[i];
532 struct sockaddr_in *sin = &managers[i];
534 if ((!strncmp(name, "tcp:", 4)
535 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
536 (!strncmp(name, "ssl:", 4)
537 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
543 *managersp = managers;
544 *n_managersp = n_managers;
548 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
550 struct ovsdb_idl_txn *txn;
551 struct shash old_br, new_br;
552 struct shash_node *node;
553 struct bridge *br, *next;
554 struct sockaddr_in *managers;
557 int sflow_bridge_number;
559 COVERAGE_INC(bridge_reconfigure);
561 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
563 collect_managers(ovs_cfg, &managers, &n_managers);
565 /* Collect old and new bridges. */
568 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
569 shash_add(&old_br, br->name, br);
571 for (i = 0; i < ovs_cfg->n_bridges; i++) {
572 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
573 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
574 VLOG_WARN("more than one bridge named %s", br_cfg->name);
578 /* Get rid of deleted bridges and add new bridges. */
579 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
580 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
587 SHASH_FOR_EACH (node, &new_br) {
588 const char *br_name = node->name;
589 const struct ovsrec_bridge *br_cfg = node->data;
590 br = shash_find_data(&old_br, br_name);
592 /* If the bridge datapath type has changed, we need to tear it
593 * down and recreate. */
594 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
596 bridge_create(br_cfg);
599 bridge_create(br_cfg);
602 shash_destroy(&old_br);
603 shash_destroy(&new_br);
607 bridge_configure_ssl(ovs_cfg->ssl);
610 /* Reconfigure all bridges. */
611 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
612 bridge_reconfigure_one(ovs_cfg, br);
615 /* Add and delete ports on all datapaths.
617 * The kernel will reject any attempt to add a given port to a datapath if
618 * that port already belongs to a different datapath, so we must do all
619 * port deletions before any port additions. */
620 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
621 struct odp_port *dpif_ports;
623 struct shash want_ifaces;
625 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
626 bridge_get_all_ifaces(br, &want_ifaces);
627 for (i = 0; i < n_dpif_ports; i++) {
628 const struct odp_port *p = &dpif_ports[i];
629 if (!shash_find(&want_ifaces, p->devname)
630 && strcmp(p->devname, br->name)) {
631 int retval = dpif_port_del(br->dpif, p->port);
633 VLOG_ERR("failed to remove %s interface from %s: %s",
634 p->devname, dpif_name(br->dpif),
639 shash_destroy(&want_ifaces);
642 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
643 struct odp_port *dpif_ports;
645 struct shash cur_ifaces, want_ifaces;
646 struct shash_node *node;
648 /* Get the set of interfaces currently in this datapath. */
649 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
650 shash_init(&cur_ifaces);
651 for (i = 0; i < n_dpif_ports; i++) {
652 const char *name = dpif_ports[i].devname;
653 if (!shash_find(&cur_ifaces, name)) {
654 shash_add(&cur_ifaces, name, NULL);
659 /* Get the set of interfaces we want on this datapath. */
660 bridge_get_all_ifaces(br, &want_ifaces);
662 SHASH_FOR_EACH (node, &want_ifaces) {
663 const char *if_name = node->name;
664 struct iface *iface = node->data;
666 if (shash_find(&cur_ifaces, if_name)) {
667 /* Already exists, just reconfigure it. */
669 reconfigure_iface(iface->cfg, iface);
672 /* Need to add to datapath. */
676 /* Add to datapath. */
677 internal = iface_is_internal(br, if_name);
678 error = dpif_port_add(br->dpif, if_name,
679 internal ? ODP_PORT_INTERNAL : 0, NULL);
680 if (error == EFBIG) {
681 VLOG_ERR("ran out of valid port numbers on %s",
682 dpif_name(br->dpif));
685 VLOG_ERR("failed to add %s interface to %s: %s",
686 if_name, dpif_name(br->dpif), strerror(error));
690 shash_destroy(&cur_ifaces);
691 shash_destroy(&want_ifaces);
693 sflow_bridge_number = 0;
694 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
697 struct iface *local_iface;
698 struct iface *hw_addr_iface;
701 bridge_fetch_dp_ifaces(br);
703 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
704 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
706 /* Pick local port hardware address, datapath ID. */
707 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
708 local_iface = bridge_get_local_iface(br);
710 int error = netdev_set_etheraddr(local_iface->netdev, ea);
712 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
713 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
714 "Ethernet address: %s",
715 br->name, strerror(error));
719 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
720 ofproto_set_datapath_id(br->ofproto, dpid);
722 dpid_string = xasprintf("%012"PRIx64, dpid);
723 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
726 /* Set NetFlow configuration on this bridge. */
727 if (br->cfg->netflow) {
728 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
729 struct netflow_options opts;
731 memset(&opts, 0, sizeof opts);
733 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
734 if (nf_cfg->engine_type) {
735 opts.engine_type = *nf_cfg->engine_type;
737 if (nf_cfg->engine_id) {
738 opts.engine_id = *nf_cfg->engine_id;
741 opts.active_timeout = nf_cfg->active_timeout;
742 if (!opts.active_timeout) {
743 opts.active_timeout = -1;
744 } else if (opts.active_timeout < 0) {
745 VLOG_WARN("bridge %s: active timeout interval set to negative "
746 "value, using default instead (%d seconds)", br->name,
747 NF_ACTIVE_TIMEOUT_DEFAULT);
748 opts.active_timeout = -1;
751 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
752 if (opts.add_id_to_iface) {
753 if (opts.engine_id > 0x7f) {
754 VLOG_WARN("bridge %s: netflow port mangling may conflict "
755 "with another vswitch, choose an engine id less "
756 "than 128", br->name);
758 if (br->n_ports > 508) {
759 VLOG_WARN("bridge %s: netflow port mangling will conflict "
760 "with another port when more than 508 ports are "
765 opts.collectors.n = nf_cfg->n_targets;
766 opts.collectors.names = nf_cfg->targets;
767 if (ofproto_set_netflow(br->ofproto, &opts)) {
768 VLOG_ERR("bridge %s: problem setting netflow collectors",
772 ofproto_set_netflow(br->ofproto, NULL);
775 /* Set sFlow configuration on this bridge. */
776 if (br->cfg->sflow) {
777 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
778 struct ovsrec_controller **controllers;
779 struct ofproto_sflow_options oso;
780 size_t n_controllers;
783 memset(&oso, 0, sizeof oso);
785 oso.targets.n = sflow_cfg->n_targets;
786 oso.targets.names = sflow_cfg->targets;
788 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
789 if (sflow_cfg->sampling) {
790 oso.sampling_rate = *sflow_cfg->sampling;
793 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
794 if (sflow_cfg->polling) {
795 oso.polling_interval = *sflow_cfg->polling;
798 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
799 if (sflow_cfg->header) {
800 oso.header_len = *sflow_cfg->header;
803 oso.sub_id = sflow_bridge_number++;
804 oso.agent_device = sflow_cfg->agent;
806 oso.control_ip = NULL;
807 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
808 for (i = 0; i < n_controllers; i++) {
809 if (controllers[i]->local_ip) {
810 oso.control_ip = controllers[i]->local_ip;
814 ofproto_set_sflow(br->ofproto, &oso);
816 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
818 ofproto_set_sflow(br->ofproto, NULL);
821 /* Update the controller and related settings. It would be more
822 * straightforward to call this from bridge_reconfigure_one(), but we
823 * can't do it there for two reasons. First, and most importantly, at
824 * that point we don't know the dp_ifidx of any interfaces that have
825 * been added to the bridge (because we haven't actually added them to
826 * the datapath). Second, at that point we haven't set the datapath ID
827 * yet; when a controller is configured, resetting the datapath ID will
828 * immediately disconnect from the controller, so it's better to set
829 * the datapath ID before the controller. */
830 bridge_reconfigure_remotes(ovs_cfg, br, managers, n_managers);
832 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
833 for (i = 0; i < br->n_ports; i++) {
834 struct port *port = br->ports[i];
836 port_update_vlan_compat(port);
837 port_update_bonding(port);
840 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
841 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
844 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
846 ovsdb_idl_txn_commit(txn);
847 ovsdb_idl_txn_destroy(txn); /* XXX */
853 get_ovsrec_key_value(const char *key, char **keys, char **values, size_t n)
857 for (i = 0; i < n; i++) {
858 if (!strcmp(keys[i], key)) {
866 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
868 return get_ovsrec_key_value(key,
869 br_cfg->key_other_config,
870 br_cfg->value_other_config,
871 br_cfg->n_other_config);
875 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
876 struct iface **hw_addr_iface)
882 *hw_addr_iface = NULL;
884 /* Did the user request a particular MAC? */
885 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
886 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
887 if (eth_addr_is_multicast(ea)) {
888 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
889 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
890 } else if (eth_addr_is_zero(ea)) {
891 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
897 /* Otherwise choose the minimum non-local MAC address among all of the
899 memset(ea, 0xff, sizeof ea);
900 for (i = 0; i < br->n_ports; i++) {
901 struct port *port = br->ports[i];
902 uint8_t iface_ea[ETH_ADDR_LEN];
905 /* Mirror output ports don't participate. */
906 if (port->is_mirror_output_port) {
910 /* Choose the MAC address to represent the port. */
911 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
912 /* Find the interface with this Ethernet address (if any) so that
913 * we can provide the correct devname to the caller. */
915 for (j = 0; j < port->n_ifaces; j++) {
916 struct iface *candidate = port->ifaces[j];
917 uint8_t candidate_ea[ETH_ADDR_LEN];
918 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
919 && eth_addr_equals(iface_ea, candidate_ea)) {
924 /* Choose the interface whose MAC address will represent the port.
925 * The Linux kernel bonding code always chooses the MAC address of
926 * the first slave added to a bond, and the Fedora networking
927 * scripts always add slaves to a bond in alphabetical order, so
928 * for compatibility we choose the interface with the name that is
929 * first in alphabetical order. */
930 iface = port->ifaces[0];
931 for (j = 1; j < port->n_ifaces; j++) {
932 struct iface *candidate = port->ifaces[j];
933 if (strcmp(candidate->name, iface->name) < 0) {
938 /* The local port doesn't count (since we're trying to choose its
939 * MAC address anyway). */
940 if (iface->dp_ifidx == ODPP_LOCAL) {
945 error = netdev_get_etheraddr(iface->netdev, iface_ea);
947 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
948 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
949 iface->name, strerror(error));
954 /* Compare against our current choice. */
955 if (!eth_addr_is_multicast(iface_ea) &&
956 !eth_addr_is_local(iface_ea) &&
957 !eth_addr_is_reserved(iface_ea) &&
958 !eth_addr_is_zero(iface_ea) &&
959 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
961 memcpy(ea, iface_ea, ETH_ADDR_LEN);
962 *hw_addr_iface = iface;
965 if (eth_addr_is_multicast(ea)) {
966 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
967 *hw_addr_iface = NULL;
968 VLOG_WARN("bridge %s: using default bridge Ethernet "
969 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
971 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
972 br->name, ETH_ADDR_ARGS(ea));
976 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
977 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
978 * an interface on 'br', then that interface must be passed in as
979 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
980 * 'hw_addr_iface' must be passed in as a null pointer. */
982 bridge_pick_datapath_id(struct bridge *br,
983 const uint8_t bridge_ea[ETH_ADDR_LEN],
984 struct iface *hw_addr_iface)
987 * The procedure for choosing a bridge MAC address will, in the most
988 * ordinary case, also choose a unique MAC that we can use as a datapath
989 * ID. In some special cases, though, multiple bridges will end up with
990 * the same MAC address. This is OK for the bridges, but it will confuse
991 * the OpenFlow controller, because each datapath needs a unique datapath
994 * Datapath IDs must be unique. It is also very desirable that they be
995 * stable from one run to the next, so that policy set on a datapath
998 const char *datapath_id;
1001 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1002 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1006 if (hw_addr_iface) {
1008 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1010 * A bridge whose MAC address is taken from a VLAN network device
1011 * (that is, a network device created with vconfig(8) or similar
1012 * tool) will have the same MAC address as a bridge on the VLAN
1013 * device's physical network device.
1015 * Handle this case by hashing the physical network device MAC
1016 * along with the VLAN identifier.
1018 uint8_t buf[ETH_ADDR_LEN + 2];
1019 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1020 buf[ETH_ADDR_LEN] = vlan >> 8;
1021 buf[ETH_ADDR_LEN + 1] = vlan;
1022 return dpid_from_hash(buf, sizeof buf);
1025 * Assume that this bridge's MAC address is unique, since it
1026 * doesn't fit any of the cases we handle specially.
1031 * A purely internal bridge, that is, one that has no non-virtual
1032 * network devices on it at all, is more difficult because it has no
1033 * natural unique identifier at all.
1035 * When the host is a XenServer, we handle this case by hashing the
1036 * host's UUID with the name of the bridge. Names of bridges are
1037 * persistent across XenServer reboots, although they can be reused if
1038 * an internal network is destroyed and then a new one is later
1039 * created, so this is fairly effective.
1041 * When the host is not a XenServer, we punt by using a random MAC
1042 * address on each run.
1044 const char *host_uuid = xenserver_get_host_uuid();
1046 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1047 dpid = dpid_from_hash(combined, strlen(combined));
1053 return eth_addr_to_uint64(bridge_ea);
1057 dpid_from_hash(const void *data, size_t n)
1059 uint8_t hash[SHA1_DIGEST_SIZE];
1061 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1062 sha1_bytes(data, n, hash);
1063 eth_addr_mark_random(hash);
1064 return eth_addr_to_uint64(hash);
1070 struct bridge *br, *next;
1074 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
1075 int error = bridge_run_one(br);
1077 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1078 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1079 "forcing reconfiguration", br->name);
1093 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1094 ofproto_wait(br->ofproto);
1095 if (ofproto_has_controller(br->ofproto)) {
1099 mac_learning_wait(br->ml);
1104 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1105 * configuration changes. */
1107 bridge_flush(struct bridge *br)
1109 COVERAGE_INC(bridge_flush);
1111 mac_learning_flush(br->ml);
1114 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1115 * such interface. */
1116 static struct iface *
1117 bridge_get_local_iface(struct bridge *br)
1121 for (i = 0; i < br->n_ports; i++) {
1122 struct port *port = br->ports[i];
1123 for (j = 0; j < port->n_ifaces; j++) {
1124 struct iface *iface = port->ifaces[j];
1125 if (iface->dp_ifidx == ODPP_LOCAL) {
1134 /* Bridge unixctl user interface functions. */
1136 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1137 const char *args, void *aux OVS_UNUSED)
1139 struct ds ds = DS_EMPTY_INITIALIZER;
1140 const struct bridge *br;
1141 const struct mac_entry *e;
1143 br = bridge_lookup(args);
1145 unixctl_command_reply(conn, 501, "no such bridge");
1149 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1150 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1151 if (e->port < 0 || e->port >= br->n_ports) {
1154 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1155 br->ports[e->port]->ifaces[0]->dp_ifidx,
1156 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1158 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1162 /* Bridge reconfiguration functions. */
1163 static struct bridge *
1164 bridge_create(const struct ovsrec_bridge *br_cfg)
1169 assert(!bridge_lookup(br_cfg->name));
1170 br = xzalloc(sizeof *br);
1172 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1178 dpif_flow_flush(br->dpif);
1180 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1183 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1185 dpif_delete(br->dpif);
1186 dpif_close(br->dpif);
1191 br->name = xstrdup(br_cfg->name);
1193 br->ml = mac_learning_create();
1194 eth_addr_nicira_random(br->default_ea);
1196 port_array_init(&br->ifaces);
1198 shash_init(&br->port_by_name);
1199 shash_init(&br->iface_by_name);
1203 list_push_back(&all_bridges, &br->node);
1205 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1211 bridge_destroy(struct bridge *br)
1216 while (br->n_ports > 0) {
1217 port_destroy(br->ports[br->n_ports - 1]);
1219 list_remove(&br->node);
1220 error = dpif_delete(br->dpif);
1221 if (error && error != ENOENT) {
1222 VLOG_ERR("failed to delete %s: %s",
1223 dpif_name(br->dpif), strerror(error));
1225 dpif_close(br->dpif);
1226 ofproto_destroy(br->ofproto);
1227 mac_learning_destroy(br->ml);
1228 port_array_destroy(&br->ifaces);
1229 shash_destroy(&br->port_by_name);
1230 shash_destroy(&br->iface_by_name);
1237 static struct bridge *
1238 bridge_lookup(const char *name)
1242 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1243 if (!strcmp(br->name, name)) {
1251 bridge_exists(const char *name)
1253 return bridge_lookup(name) ? true : false;
1257 bridge_get_datapathid(const char *name)
1259 struct bridge *br = bridge_lookup(name);
1260 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1263 /* Handle requests for a listing of all flows known by the OpenFlow
1264 * stack, including those normally hidden. */
1266 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1267 const char *args, void *aux OVS_UNUSED)
1272 br = bridge_lookup(args);
1274 unixctl_command_reply(conn, 501, "Unknown bridge");
1279 ofproto_get_all_flows(br->ofproto, &results);
1281 unixctl_command_reply(conn, 200, ds_cstr(&results));
1282 ds_destroy(&results);
1286 bridge_run_one(struct bridge *br)
1290 error = ofproto_run1(br->ofproto);
1295 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1298 error = ofproto_run2(br->ofproto, br->flush);
1305 bridge_get_controllers(const struct ovsrec_open_vswitch *ovs_cfg,
1306 const struct bridge *br,
1307 struct ovsrec_controller ***controllersp)
1309 struct ovsrec_controller **controllers;
1310 size_t n_controllers;
1312 if (br->cfg->n_controller) {
1313 controllers = br->cfg->controller;
1314 n_controllers = br->cfg->n_controller;
1316 controllers = ovs_cfg->controller;
1317 n_controllers = ovs_cfg->n_controller;
1320 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1326 *controllersp = controllers;
1328 return n_controllers;
1332 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1335 struct shash old_ports, new_ports;
1336 struct svec listeners, old_listeners;
1337 struct svec snoops, old_snoops;
1338 struct shash_node *node;
1341 /* Collect old ports. */
1342 shash_init(&old_ports);
1343 for (i = 0; i < br->n_ports; i++) {
1344 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1347 /* Collect new ports. */
1348 shash_init(&new_ports);
1349 for (i = 0; i < br->cfg->n_ports; i++) {
1350 const char *name = br->cfg->ports[i]->name;
1351 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1352 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1357 /* If we have a controller, then we need a local port. Complain if the
1358 * user didn't specify one.
1360 * XXX perhaps we should synthesize a port ourselves in this case. */
1361 if (bridge_get_controllers(ovs_cfg, br, NULL)) {
1362 char local_name[IF_NAMESIZE];
1365 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1366 local_name, sizeof local_name);
1367 if (!error && !shash_find(&new_ports, local_name)) {
1368 VLOG_WARN("bridge %s: controller specified but no local port "
1369 "(port named %s) defined",
1370 br->name, local_name);
1374 /* Get rid of deleted ports.
1375 * Get rid of deleted interfaces on ports that still exist. */
1376 SHASH_FOR_EACH (node, &old_ports) {
1377 struct port *port = node->data;
1378 const struct ovsrec_port *port_cfg;
1380 port_cfg = shash_find_data(&new_ports, node->name);
1384 port_del_ifaces(port, port_cfg);
1388 /* Create new ports.
1389 * Add new interfaces to existing ports.
1390 * Reconfigure existing ports. */
1391 SHASH_FOR_EACH (node, &new_ports) {
1392 struct port *port = shash_find_data(&old_ports, node->name);
1394 port = port_create(br, node->name);
1397 port_reconfigure(port, node->data);
1398 if (!port->n_ifaces) {
1399 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1400 br->name, port->name);
1404 shash_destroy(&old_ports);
1405 shash_destroy(&new_ports);
1407 /* Delete all flows if we're switching from connected to standalone or vice
1408 * versa. (XXX Should we delete all flows if we are switching from one
1409 * controller to another?) */
1412 /* Configure OpenFlow management listeners. */
1413 svec_init(&listeners);
1414 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1416 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1417 ovs_rundir, br->name));
1418 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1419 svec_clear(&listeners);
1421 svec_sort_unique(&listeners);
1423 svec_init(&old_listeners);
1424 ofproto_get_listeners(br->ofproto, &old_listeners);
1425 svec_sort_unique(&old_listeners);
1427 if (!svec_equal(&listeners, &old_listeners)) {
1428 ofproto_set_listeners(br->ofproto, &listeners);
1430 svec_destroy(&listeners);
1431 svec_destroy(&old_listeners);
1433 /* Configure OpenFlow controller connection snooping. */
1435 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1437 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1438 ovs_rundir, br->name));
1439 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1440 svec_clear(&snoops);
1442 svec_sort_unique(&snoops);
1444 svec_init(&old_snoops);
1445 ofproto_get_snoops(br->ofproto, &old_snoops);
1446 svec_sort_unique(&old_snoops);
1448 if (!svec_equal(&snoops, &old_snoops)) {
1449 ofproto_set_snoops(br->ofproto, &snoops);
1451 svec_destroy(&snoops);
1452 svec_destroy(&old_snoops);
1454 /* Default listener. */
1455 svec_init(&listeners);
1456 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1457 ovs_rundir, br->name));
1458 svec_init(&old_listeners);
1459 ofproto_get_listeners(br->ofproto, &old_listeners);
1460 if (!svec_equal(&listeners, &old_listeners)) {
1461 ofproto_set_listeners(br->ofproto, &listeners);
1463 svec_destroy(&listeners);
1464 svec_destroy(&old_listeners);
1466 /* Default snoop. */
1468 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1469 ovs_rundir, br->name));
1470 svec_init(&old_snoops);
1471 ofproto_get_snoops(br->ofproto, &old_snoops);
1472 if (!svec_equal(&snoops, &old_snoops)) {
1473 ofproto_set_snoops(br->ofproto, &snoops);
1475 svec_destroy(&snoops);
1476 svec_destroy(&old_snoops);
1479 mirror_reconfigure(br);
1483 bridge_reconfigure_remotes(const struct ovsrec_open_vswitch *ovs_cfg,
1485 const struct sockaddr_in *managers,
1488 struct ovsrec_controller **controllers;
1489 size_t n_controllers;
1491 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1493 n_controllers = bridge_get_controllers(ovs_cfg, br, &controllers);
1494 if (ofproto_has_controller(br->ofproto) != (n_controllers != 0)) {
1495 ofproto_flush_flows(br->ofproto);
1498 if (!n_controllers) {
1499 union ofp_action action;
1502 /* Clear out controllers. */
1503 ofproto_set_controllers(br->ofproto, NULL, 0);
1505 /* Set up a flow that matches every packet and directs them to
1506 * OFPP_NORMAL (which goes to us). */
1507 memset(&action, 0, sizeof action);
1508 action.type = htons(OFPAT_OUTPUT);
1509 action.output.len = htons(sizeof action);
1510 action.output.port = htons(OFPP_NORMAL);
1511 memset(&flow, 0, sizeof flow);
1512 ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
1514 struct ofproto_controller *ocs;
1517 ocs = xmalloc(n_controllers * sizeof *ocs);
1518 for (i = 0; i < n_controllers; i++) {
1519 struct ovsrec_controller *c = controllers[i];
1520 struct ofproto_controller *oc = &ocs[i];
1522 if (strcmp(c->target, "discover")) {
1523 struct iface *local_iface;
1526 local_iface = bridge_get_local_iface(br);
1527 if (local_iface && c->local_ip
1528 && inet_aton(c->local_ip, &ip)) {
1529 struct netdev *netdev = local_iface->netdev;
1530 struct in_addr mask, gateway;
1532 if (!c->local_netmask
1533 || !inet_aton(c->local_netmask, &mask)) {
1536 if (!c->local_gateway
1537 || !inet_aton(c->local_gateway, &gateway)) {
1541 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1543 mask.s_addr = guess_netmask(ip.s_addr);
1545 if (!netdev_set_in4(netdev, ip, mask)) {
1546 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1548 br->name, IP_ARGS(&ip.s_addr),
1549 IP_ARGS(&mask.s_addr));
1552 if (gateway.s_addr) {
1553 if (!netdev_add_router(netdev, gateway)) {
1554 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1555 br->name, IP_ARGS(&gateway.s_addr));
1561 oc->target = c->target;
1562 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1563 oc->probe_interval = (c->inactivity_probe
1564 ? *c->inactivity_probe / 1000 : 5);
1565 oc->fail = (!c->fail_mode
1566 || !strcmp(c->fail_mode, "standalone")
1567 || !strcmp(c->fail_mode, "open")
1568 ? OFPROTO_FAIL_STANDALONE
1569 : OFPROTO_FAIL_SECURE);
1570 oc->band = (!c->connection_mode
1571 || !strcmp(c->connection_mode, "in-band")
1573 : OFPROTO_OUT_OF_BAND);
1574 oc->accept_re = c->discover_accept_regex;
1575 oc->update_resolv_conf = c->discover_update_resolv_conf;
1576 oc->rate_limit = (c->controller_rate_limit
1577 ? *c->controller_rate_limit : 0);
1578 oc->burst_limit = (c->controller_burst_limit
1579 ? *c->controller_burst_limit : 0);
1581 ofproto_set_controllers(br->ofproto, ocs, n_controllers);
1587 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1592 for (i = 0; i < br->n_ports; i++) {
1593 struct port *port = br->ports[i];
1594 for (j = 0; j < port->n_ifaces; j++) {
1595 struct iface *iface = port->ifaces[j];
1596 shash_add_once(ifaces, iface->name, iface);
1598 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1599 shash_add_once(ifaces, port->name, NULL);
1604 /* For robustness, in case the administrator moves around datapath ports behind
1605 * our back, we re-check all the datapath port numbers here.
1607 * This function will set the 'dp_ifidx' members of interfaces that have
1608 * disappeared to -1, so only call this function from a context where those
1609 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1610 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1611 * datapath, which doesn't support UINT16_MAX+1 ports. */
1613 bridge_fetch_dp_ifaces(struct bridge *br)
1615 struct odp_port *dpif_ports;
1616 size_t n_dpif_ports;
1619 /* Reset all interface numbers. */
1620 for (i = 0; i < br->n_ports; i++) {
1621 struct port *port = br->ports[i];
1622 for (j = 0; j < port->n_ifaces; j++) {
1623 struct iface *iface = port->ifaces[j];
1624 iface->dp_ifidx = -1;
1627 port_array_clear(&br->ifaces);
1629 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1630 for (i = 0; i < n_dpif_ports; i++) {
1631 struct odp_port *p = &dpif_ports[i];
1632 struct iface *iface = iface_lookup(br, p->devname);
1634 if (iface->dp_ifidx >= 0) {
1635 VLOG_WARN("%s reported interface %s twice",
1636 dpif_name(br->dpif), p->devname);
1637 } else if (iface_from_dp_ifidx(br, p->port)) {
1638 VLOG_WARN("%s reported interface %"PRIu16" twice",
1639 dpif_name(br->dpif), p->port);
1641 port_array_set(&br->ifaces, p->port, iface);
1642 iface->dp_ifidx = p->port;
1646 int64_t ofport = (iface->dp_ifidx >= 0
1647 ? odp_port_to_ofp_port(iface->dp_ifidx)
1649 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1656 /* Bridge packet processing functions. */
1659 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1661 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1664 static struct bond_entry *
1665 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1667 return &port->bond_hash[bond_hash(mac)];
1671 bond_choose_iface(const struct port *port)
1673 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1674 size_t i, best_down_slave = -1;
1675 long long next_delay_expiration = LLONG_MAX;
1677 for (i = 0; i < port->n_ifaces; i++) {
1678 struct iface *iface = port->ifaces[i];
1680 if (iface->enabled) {
1682 } else if (iface->delay_expires < next_delay_expiration) {
1683 best_down_slave = i;
1684 next_delay_expiration = iface->delay_expires;
1688 if (best_down_slave != -1) {
1689 struct iface *iface = port->ifaces[best_down_slave];
1691 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1692 "since no other interface is up", iface->name,
1693 iface->delay_expires - time_msec());
1694 bond_enable_slave(iface, true);
1697 return best_down_slave;
1701 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1702 uint16_t *dp_ifidx, tag_type *tags)
1704 struct iface *iface;
1706 assert(port->n_ifaces);
1707 if (port->n_ifaces == 1) {
1708 iface = port->ifaces[0];
1710 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1711 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1712 || !port->ifaces[e->iface_idx]->enabled) {
1713 /* XXX select interface properly. The current interface selection
1714 * is only good for testing the rebalancing code. */
1715 e->iface_idx = bond_choose_iface(port);
1716 if (e->iface_idx < 0) {
1717 *tags |= port->no_ifaces_tag;
1720 e->iface_tag = tag_create_random();
1721 ((struct port *) port)->bond_compat_is_stale = true;
1723 *tags |= e->iface_tag;
1724 iface = port->ifaces[e->iface_idx];
1726 *dp_ifidx = iface->dp_ifidx;
1727 *tags |= iface->tag; /* Currently only used for bonding. */
1732 bond_link_status_update(struct iface *iface, bool carrier)
1734 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1735 struct port *port = iface->port;
1737 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1738 /* Nothing to do. */
1741 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1742 iface->name, carrier ? "detected" : "dropped");
1743 if (carrier == iface->enabled) {
1744 iface->delay_expires = LLONG_MAX;
1745 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1746 iface->name, carrier ? "disabled" : "enabled");
1747 } else if (carrier && port->active_iface < 0) {
1748 bond_enable_slave(iface, true);
1749 if (port->updelay) {
1750 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1751 "other interface is up", iface->name, port->updelay);
1754 int delay = carrier ? port->updelay : port->downdelay;
1755 iface->delay_expires = time_msec() + delay;
1758 "interface %s: will be %s if it stays %s for %d ms",
1760 carrier ? "enabled" : "disabled",
1761 carrier ? "up" : "down",
1768 bond_choose_active_iface(struct port *port)
1770 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1772 port->active_iface = bond_choose_iface(port);
1773 port->active_iface_tag = tag_create_random();
1774 if (port->active_iface >= 0) {
1775 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1776 port->name, port->ifaces[port->active_iface]->name);
1778 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1784 bond_enable_slave(struct iface *iface, bool enable)
1786 struct port *port = iface->port;
1787 struct bridge *br = port->bridge;
1789 /* This acts as a recursion check. If the act of disabling a slave
1790 * causes a different slave to be enabled, the flag will allow us to
1791 * skip redundant work when we reenter this function. It must be
1792 * cleared on exit to keep things safe with multiple bonds. */
1793 static bool moving_active_iface = false;
1795 iface->delay_expires = LLONG_MAX;
1796 if (enable == iface->enabled) {
1800 iface->enabled = enable;
1801 if (!iface->enabled) {
1802 VLOG_WARN("interface %s: disabled", iface->name);
1803 ofproto_revalidate(br->ofproto, iface->tag);
1804 if (iface->port_ifidx == port->active_iface) {
1805 ofproto_revalidate(br->ofproto,
1806 port->active_iface_tag);
1808 /* Disabling a slave can lead to another slave being immediately
1809 * enabled if there will be no active slaves but one is waiting
1810 * on an updelay. In this case we do not need to run most of the
1811 * code for the newly enabled slave since there was no period
1812 * without an active slave and it is redundant with the disabling
1814 moving_active_iface = true;
1815 bond_choose_active_iface(port);
1817 bond_send_learning_packets(port);
1819 VLOG_WARN("interface %s: enabled", iface->name);
1820 if (port->active_iface < 0 && !moving_active_iface) {
1821 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1822 bond_choose_active_iface(port);
1823 bond_send_learning_packets(port);
1825 iface->tag = tag_create_random();
1828 moving_active_iface = false;
1829 port->bond_compat_is_stale = true;
1832 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1833 * bond interface. */
1835 bond_update_fake_iface_stats(struct port *port)
1837 struct netdev_stats bond_stats;
1838 struct netdev *bond_dev;
1841 memset(&bond_stats, 0, sizeof bond_stats);
1843 for (i = 0; i < port->n_ifaces; i++) {
1844 struct netdev_stats slave_stats;
1846 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1847 /* XXX: We swap the stats here because they are swapped back when
1848 * reported by the internal device. The reason for this is
1849 * internal devices normally represent packets going into the system
1850 * but when used as fake bond device they represent packets leaving
1851 * the system. We really should do this in the internal device
1852 * itself because changing it here reverses the counts from the
1853 * perspective of the switch. However, the internal device doesn't
1854 * know what type of device it represents so we have to do it here
1856 bond_stats.tx_packets += slave_stats.rx_packets;
1857 bond_stats.tx_bytes += slave_stats.rx_bytes;
1858 bond_stats.rx_packets += slave_stats.tx_packets;
1859 bond_stats.rx_bytes += slave_stats.tx_bytes;
1863 if (!netdev_open_default(port->name, &bond_dev)) {
1864 netdev_set_stats(bond_dev, &bond_stats);
1865 netdev_close(bond_dev);
1870 bond_run(struct bridge *br)
1874 for (i = 0; i < br->n_ports; i++) {
1875 struct port *port = br->ports[i];
1877 if (port->n_ifaces >= 2) {
1878 for (j = 0; j < port->n_ifaces; j++) {
1879 struct iface *iface = port->ifaces[j];
1880 if (time_msec() >= iface->delay_expires) {
1881 bond_enable_slave(iface, !iface->enabled);
1885 if (port->bond_fake_iface
1886 && time_msec() >= port->bond_next_fake_iface_update) {
1887 bond_update_fake_iface_stats(port);
1888 port->bond_next_fake_iface_update = time_msec() + 1000;
1892 if (port->bond_compat_is_stale) {
1893 port->bond_compat_is_stale = false;
1894 port_update_bond_compat(port);
1900 bond_wait(struct bridge *br)
1904 for (i = 0; i < br->n_ports; i++) {
1905 struct port *port = br->ports[i];
1906 if (port->n_ifaces < 2) {
1909 for (j = 0; j < port->n_ifaces; j++) {
1910 struct iface *iface = port->ifaces[j];
1911 if (iface->delay_expires != LLONG_MAX) {
1912 poll_timer_wait_until(iface->delay_expires);
1915 if (port->bond_fake_iface) {
1916 poll_timer_wait_until(port->bond_next_fake_iface_update);
1922 set_dst(struct dst *p, const flow_t *flow,
1923 const struct port *in_port, const struct port *out_port,
1926 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1927 : in_port->vlan >= 0 ? in_port->vlan
1928 : ntohs(flow->dl_vlan));
1929 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1933 swap_dst(struct dst *p, struct dst *q)
1935 struct dst tmp = *p;
1940 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1941 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1942 * that we push to the datapath. We could in fact fully sort the array by
1943 * vlan, but in most cases there are at most two different vlan tags so that's
1944 * possibly overkill.) */
1946 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1948 struct dst *first = dsts;
1949 struct dst *last = dsts + n_dsts;
1951 while (first != last) {
1953 * - All dsts < first have vlan == 'vlan'.
1954 * - All dsts >= last have vlan != 'vlan'.
1955 * - first < last. */
1956 while (first->vlan == vlan) {
1957 if (++first == last) {
1962 /* Same invariants, plus one additional:
1963 * - first->vlan != vlan.
1965 while (last[-1].vlan != vlan) {
1966 if (--last == first) {
1971 /* Same invariants, plus one additional:
1972 * - last[-1].vlan == vlan.*/
1973 swap_dst(first++, --last);
1978 mirror_mask_ffs(mirror_mask_t mask)
1980 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1985 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1986 const struct dst *test)
1989 for (i = 0; i < n_dsts; i++) {
1990 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1998 port_trunks_vlan(const struct port *port, uint16_t vlan)
2000 return (port->vlan < 0
2001 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2005 port_includes_vlan(const struct port *port, uint16_t vlan)
2007 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2011 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2012 const struct port *in_port, const struct port *out_port,
2013 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2015 mirror_mask_t mirrors = in_port->src_mirrors;
2016 struct dst *dst = dsts;
2019 if (out_port == FLOOD_PORT) {
2020 /* XXX use ODP_FLOOD if no vlans or bonding. */
2021 /* XXX even better, define each VLAN as a datapath port group */
2022 for (i = 0; i < br->n_ports; i++) {
2023 struct port *port = br->ports[i];
2024 if (port != in_port && port_includes_vlan(port, vlan)
2025 && !port->is_mirror_output_port
2026 && set_dst(dst, flow, in_port, port, tags)) {
2027 mirrors |= port->dst_mirrors;
2031 *nf_output_iface = NF_OUT_FLOOD;
2032 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2033 *nf_output_iface = dst->dp_ifidx;
2034 mirrors |= out_port->dst_mirrors;
2039 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2040 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2042 if (set_dst(dst, flow, in_port, m->out_port, tags)
2043 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2047 for (i = 0; i < br->n_ports; i++) {
2048 struct port *port = br->ports[i];
2049 if (port_includes_vlan(port, m->out_vlan)
2050 && set_dst(dst, flow, in_port, port, tags))
2054 if (port->vlan < 0) {
2055 dst->vlan = m->out_vlan;
2057 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2061 /* Use the vlan tag on the original flow instead of
2062 * the one passed in the vlan parameter. This ensures
2063 * that we compare the vlan from before any implicit
2064 * tagging tags place. This is necessary because
2065 * dst->vlan is the final vlan, after removing implicit
2067 flow_vlan = ntohs(flow->dl_vlan);
2068 if (flow_vlan == 0) {
2069 flow_vlan = OFP_VLAN_NONE;
2071 if (port == in_port && dst->vlan == flow_vlan) {
2072 /* Don't send out input port on same VLAN. */
2080 mirrors &= mirrors - 1;
2083 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2087 static void OVS_UNUSED
2088 print_dsts(const struct dst *dsts, size_t n)
2090 for (; n--; dsts++) {
2091 printf(">p%"PRIu16, dsts->dp_ifidx);
2092 if (dsts->vlan != OFP_VLAN_NONE) {
2093 printf("v%"PRIu16, dsts->vlan);
2099 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2100 const struct port *in_port, const struct port *out_port,
2101 tag_type *tags, struct odp_actions *actions,
2102 uint16_t *nf_output_iface)
2104 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2106 const struct dst *p;
2109 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2112 cur_vlan = ntohs(flow->dl_vlan);
2113 for (p = dsts; p < &dsts[n_dsts]; p++) {
2114 union odp_action *a;
2115 if (p->vlan != cur_vlan) {
2116 if (p->vlan == OFP_VLAN_NONE) {
2117 odp_actions_add(actions, ODPAT_STRIP_VLAN);
2119 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
2120 a->vlan_vid.vlan_vid = htons(p->vlan);
2124 a = odp_actions_add(actions, ODPAT_OUTPUT);
2125 a->output.port = p->dp_ifidx;
2129 /* Returns the effective vlan of a packet, taking into account both the
2130 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2131 * the packet is untagged and -1 indicates it has an invalid header and
2132 * should be dropped. */
2133 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2134 struct port *in_port, bool have_packet)
2136 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2137 * belongs to VLAN 0, so we should treat both cases identically. (In the
2138 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2139 * presumably to allow a priority to be specified. In the latter case, the
2140 * packet does not have any 802.1Q header.) */
2141 int vlan = ntohs(flow->dl_vlan);
2142 if (vlan == OFP_VLAN_NONE) {
2145 if (in_port->vlan >= 0) {
2147 /* XXX support double tagging? */
2149 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2150 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2151 "packet received on port %s configured with "
2152 "implicit VLAN %"PRIu16,
2153 br->name, ntohs(flow->dl_vlan),
2154 in_port->name, in_port->vlan);
2158 vlan = in_port->vlan;
2160 if (!port_includes_vlan(in_port, vlan)) {
2162 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2163 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2164 "packet received on port %s not configured for "
2166 br->name, vlan, in_port->name, vlan);
2175 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2176 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2177 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2179 is_gratuitous_arp(const flow_t *flow)
2181 return (flow->dl_type == htons(ETH_TYPE_ARP)
2182 && eth_addr_is_broadcast(flow->dl_dst)
2183 && (flow->nw_proto == ARP_OP_REPLY
2184 || (flow->nw_proto == ARP_OP_REQUEST
2185 && flow->nw_src == flow->nw_dst)));
2189 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2190 struct port *in_port)
2192 enum grat_arp_lock_type lock_type;
2195 /* We don't want to learn from gratuitous ARP packets that are reflected
2196 * back over bond slaves so we lock the learning table. */
2197 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2198 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2199 GRAT_ARP_LOCK_CHECK;
2201 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2204 /* The log messages here could actually be useful in debugging,
2205 * so keep the rate limit relatively high. */
2206 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2208 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2209 "on port %s in VLAN %d",
2210 br->name, ETH_ADDR_ARGS(flow->dl_src),
2211 in_port->name, vlan);
2212 ofproto_revalidate(br->ofproto, rev_tag);
2216 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2217 * dropped. Returns true if they may be forwarded, false if they should be
2220 * If 'have_packet' is true, it indicates that the caller is processing a
2221 * received packet. If 'have_packet' is false, then the caller is just
2222 * revalidating an existing flow because configuration has changed. Either
2223 * way, 'have_packet' only affects logging (there is no point in logging errors
2224 * during revalidation).
2226 * Sets '*in_portp' to the input port. This will be a null pointer if
2227 * flow->in_port does not designate a known input port (in which case
2228 * is_admissible() returns false).
2230 * When returning true, sets '*vlanp' to the effective VLAN of the input
2231 * packet, as returned by flow_get_vlan().
2233 * May also add tags to '*tags', although the current implementation only does
2234 * so in one special case.
2237 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2238 tag_type *tags, int *vlanp, struct port **in_portp)
2240 struct iface *in_iface;
2241 struct port *in_port;
2244 /* Find the interface and port structure for the received packet. */
2245 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2247 /* No interface? Something fishy... */
2249 /* Odd. A few possible reasons here:
2251 * - We deleted an interface but there are still a few packets
2252 * queued up from it.
2254 * - Someone externally added an interface (e.g. with "ovs-dpctl
2255 * add-if") that we don't know about.
2257 * - Packet arrived on the local port but the local port is not
2258 * one of our bridge ports.
2260 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2262 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2263 "interface %"PRIu16, br->name, flow->in_port);
2269 *in_portp = in_port = in_iface->port;
2270 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2275 /* Drop frames for reserved multicast addresses. */
2276 if (eth_addr_is_reserved(flow->dl_dst)) {
2280 /* Drop frames on ports reserved for mirroring. */
2281 if (in_port->is_mirror_output_port) {
2283 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2284 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2285 "%s, which is reserved exclusively for mirroring",
2286 br->name, in_port->name);
2291 /* Packets received on bonds need special attention to avoid duplicates. */
2292 if (in_port->n_ifaces > 1) {
2294 bool is_grat_arp_locked;
2296 if (eth_addr_is_multicast(flow->dl_dst)) {
2297 *tags |= in_port->active_iface_tag;
2298 if (in_port->active_iface != in_iface->port_ifidx) {
2299 /* Drop all multicast packets on inactive slaves. */
2304 /* Drop all packets for which we have learned a different input
2305 * port, because we probably sent the packet on one slave and got
2306 * it back on the other. Gratuitous ARP packets are an exception
2307 * to this rule: the host has moved to another switch. The exception
2308 * to the exception is if we locked the learning table to avoid
2309 * reflections on bond slaves. If this is the case, just drop the
2311 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2312 &is_grat_arp_locked);
2313 if (src_idx != -1 && src_idx != in_port->port_idx &&
2314 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2322 /* If the composed actions may be applied to any packet in the given 'flow',
2323 * returns true. Otherwise, the actions should only be applied to 'packet', or
2324 * not at all, if 'packet' was NULL. */
2326 process_flow(struct bridge *br, const flow_t *flow,
2327 const struct ofpbuf *packet, struct odp_actions *actions,
2328 tag_type *tags, uint16_t *nf_output_iface)
2330 struct port *in_port;
2331 struct port *out_port;
2335 /* Check whether we should drop packets in this flow. */
2336 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2341 /* Learn source MAC (but don't try to learn from revalidation). */
2343 update_learning_table(br, flow, vlan, in_port);
2346 /* Determine output port. */
2347 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2349 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2350 out_port = br->ports[out_port_idx];
2351 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2352 /* If we are revalidating but don't have a learning entry then
2353 * eject the flow. Installing a flow that floods packets opens
2354 * up a window of time where we could learn from a packet reflected
2355 * on a bond and blackhole packets before the learning table is
2356 * updated to reflect the correct port. */
2359 out_port = FLOOD_PORT;
2362 /* Don't send packets out their input ports. */
2363 if (in_port == out_port) {
2369 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2376 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2379 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2380 const struct ofp_phy_port *opp,
2383 struct bridge *br = br_;
2384 struct iface *iface;
2387 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2393 if (reason == OFPPR_DELETE) {
2394 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2395 br->name, iface->name);
2396 iface_destroy(iface);
2397 if (!port->n_ifaces) {
2398 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2399 br->name, port->name);
2405 if (port->n_ifaces > 1) {
2406 bool up = !(opp->state & OFPPS_LINK_DOWN);
2407 bond_link_status_update(iface, up);
2408 port_update_bond_compat(port);
2414 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2415 struct odp_actions *actions, tag_type *tags,
2416 uint16_t *nf_output_iface, void *br_)
2418 struct bridge *br = br_;
2420 COVERAGE_INC(bridge_process_flow);
2421 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2425 bridge_account_flow_ofhook_cb(const flow_t *flow,
2426 const union odp_action *actions,
2427 size_t n_actions, unsigned long long int n_bytes,
2430 struct bridge *br = br_;
2431 const union odp_action *a;
2432 struct port *in_port;
2436 /* Feed information from the active flows back into the learning table
2437 * to ensure that table is always in sync with what is actually flowing
2438 * through the datapath. */
2439 if (is_admissible(br, flow, false, &tags, &vlan, &in_port)) {
2440 update_learning_table(br, flow, vlan, in_port);
2443 if (!br->has_bonded_ports) {
2447 for (a = actions; a < &actions[n_actions]; a++) {
2448 if (a->type == ODPAT_OUTPUT) {
2449 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2450 if (out_port && out_port->n_ifaces >= 2) {
2451 struct bond_entry *e = lookup_bond_entry(out_port,
2453 e->tx_bytes += n_bytes;
2460 bridge_account_checkpoint_ofhook_cb(void *br_)
2462 struct bridge *br = br_;
2466 if (!br->has_bonded_ports) {
2471 for (i = 0; i < br->n_ports; i++) {
2472 struct port *port = br->ports[i];
2473 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2474 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2475 bond_rebalance_port(port);
2480 static struct ofhooks bridge_ofhooks = {
2481 bridge_port_changed_ofhook_cb,
2482 bridge_normal_ofhook_cb,
2483 bridge_account_flow_ofhook_cb,
2484 bridge_account_checkpoint_ofhook_cb,
2487 /* Bonding functions. */
2489 /* Statistics for a single interface on a bonded port, used for load-based
2490 * bond rebalancing. */
2491 struct slave_balance {
2492 struct iface *iface; /* The interface. */
2493 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2495 /* All the "bond_entry"s that are assigned to this interface, in order of
2496 * increasing tx_bytes. */
2497 struct bond_entry **hashes;
2501 /* Sorts pointers to pointers to bond_entries in ascending order by the
2502 * interface to which they are assigned, and within a single interface in
2503 * ascending order of bytes transmitted. */
2505 compare_bond_entries(const void *a_, const void *b_)
2507 const struct bond_entry *const *ap = a_;
2508 const struct bond_entry *const *bp = b_;
2509 const struct bond_entry *a = *ap;
2510 const struct bond_entry *b = *bp;
2511 if (a->iface_idx != b->iface_idx) {
2512 return a->iface_idx > b->iface_idx ? 1 : -1;
2513 } else if (a->tx_bytes != b->tx_bytes) {
2514 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2520 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2521 * *descending* order by number of bytes transmitted. */
2523 compare_slave_balance(const void *a_, const void *b_)
2525 const struct slave_balance *a = a_;
2526 const struct slave_balance *b = b_;
2527 if (a->iface->enabled != b->iface->enabled) {
2528 return a->iface->enabled ? -1 : 1;
2529 } else if (a->tx_bytes != b->tx_bytes) {
2530 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2537 swap_bals(struct slave_balance *a, struct slave_balance *b)
2539 struct slave_balance tmp = *a;
2544 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2545 * given that 'p' (and only 'p') might be in the wrong location.
2547 * This function invalidates 'p', since it might now be in a different memory
2550 resort_bals(struct slave_balance *p,
2551 struct slave_balance bals[], size_t n_bals)
2554 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2555 swap_bals(p, p - 1);
2557 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2558 swap_bals(p, p + 1);
2564 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2566 if (VLOG_IS_DBG_ENABLED()) {
2567 struct ds ds = DS_EMPTY_INITIALIZER;
2568 const struct slave_balance *b;
2570 for (b = bals; b < bals + n_bals; b++) {
2574 ds_put_char(&ds, ',');
2576 ds_put_format(&ds, " %s %"PRIu64"kB",
2577 b->iface->name, b->tx_bytes / 1024);
2579 if (!b->iface->enabled) {
2580 ds_put_cstr(&ds, " (disabled)");
2582 if (b->n_hashes > 0) {
2583 ds_put_cstr(&ds, " (");
2584 for (i = 0; i < b->n_hashes; i++) {
2585 const struct bond_entry *e = b->hashes[i];
2587 ds_put_cstr(&ds, " + ");
2589 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2590 e - port->bond_hash, e->tx_bytes / 1024);
2592 ds_put_cstr(&ds, ")");
2595 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2600 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2602 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2605 struct bond_entry *hash = from->hashes[hash_idx];
2606 struct port *port = from->iface->port;
2607 uint64_t delta = hash->tx_bytes;
2609 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2610 "from %s to %s (now carrying %"PRIu64"kB and "
2611 "%"PRIu64"kB load, respectively)",
2612 port->name, delta / 1024, hash - port->bond_hash,
2613 from->iface->name, to->iface->name,
2614 (from->tx_bytes - delta) / 1024,
2615 (to->tx_bytes + delta) / 1024);
2617 /* Delete element from from->hashes.
2619 * We don't bother to add the element to to->hashes because not only would
2620 * it require more work, the only purpose it would be to allow that hash to
2621 * be migrated to another slave in this rebalancing run, and there is no
2622 * point in doing that. */
2623 if (hash_idx == 0) {
2626 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2627 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2631 /* Shift load away from 'from' to 'to'. */
2632 from->tx_bytes -= delta;
2633 to->tx_bytes += delta;
2635 /* Arrange for flows to be revalidated. */
2636 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2637 hash->iface_idx = to->iface->port_ifidx;
2638 hash->iface_tag = tag_create_random();
2642 bond_rebalance_port(struct port *port)
2644 struct slave_balance bals[DP_MAX_PORTS];
2646 struct bond_entry *hashes[BOND_MASK + 1];
2647 struct slave_balance *b, *from, *to;
2648 struct bond_entry *e;
2651 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2652 * descending order of tx_bytes, so that bals[0] represents the most
2653 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2656 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2657 * array for each slave_balance structure, we sort our local array of
2658 * hashes in order by slave, so that all of the hashes for a given slave
2659 * become contiguous in memory, and then we point each 'hashes' members of
2660 * a slave_balance structure to the start of a contiguous group. */
2661 n_bals = port->n_ifaces;
2662 for (b = bals; b < &bals[n_bals]; b++) {
2663 b->iface = port->ifaces[b - bals];
2668 for (i = 0; i <= BOND_MASK; i++) {
2669 hashes[i] = &port->bond_hash[i];
2671 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2672 for (i = 0; i <= BOND_MASK; i++) {
2674 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2675 b = &bals[e->iface_idx];
2676 b->tx_bytes += e->tx_bytes;
2678 b->hashes = &hashes[i];
2683 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2684 log_bals(bals, n_bals, port);
2686 /* Discard slaves that aren't enabled (which were sorted to the back of the
2687 * array earlier). */
2688 while (!bals[n_bals - 1].iface->enabled) {
2695 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2696 to = &bals[n_bals - 1];
2697 for (from = bals; from < to; ) {
2698 uint64_t overload = from->tx_bytes - to->tx_bytes;
2699 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2700 /* The extra load on 'from' (and all less-loaded slaves), compared
2701 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2702 * it is less than ~1Mbps. No point in rebalancing. */
2704 } else if (from->n_hashes == 1) {
2705 /* 'from' only carries a single MAC hash, so we can't shift any
2706 * load away from it, even though we want to. */
2709 /* 'from' is carrying significantly more load than 'to', and that
2710 * load is split across at least two different hashes. Pick a hash
2711 * to migrate to 'to' (the least-loaded slave), given that doing so
2712 * must decrease the ratio of the load on the two slaves by at
2715 * The sort order we use means that we prefer to shift away the
2716 * smallest hashes instead of the biggest ones. There is little
2717 * reason behind this decision; we could use the opposite sort
2718 * order to shift away big hashes ahead of small ones. */
2722 for (i = 0; i < from->n_hashes; i++) {
2723 double old_ratio, new_ratio;
2724 uint64_t delta = from->hashes[i]->tx_bytes;
2726 if (delta == 0 || from->tx_bytes - delta == 0) {
2727 /* Pointless move. */
2731 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2733 if (to->tx_bytes == 0) {
2734 /* Nothing on the new slave, move it. */
2738 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2739 new_ratio = (double)(from->tx_bytes - delta) /
2740 (to->tx_bytes + delta);
2742 if (new_ratio == 0) {
2743 /* Should already be covered but check to prevent division
2748 if (new_ratio < 1) {
2749 new_ratio = 1 / new_ratio;
2752 if (old_ratio - new_ratio > 0.1) {
2753 /* Would decrease the ratio, move it. */
2757 if (i < from->n_hashes) {
2758 bond_shift_load(from, to, i);
2759 port->bond_compat_is_stale = true;
2761 /* If the result of the migration changed the relative order of
2762 * 'from' and 'to' swap them back to maintain invariants. */
2763 if (order_swapped) {
2764 swap_bals(from, to);
2767 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2768 * point to different slave_balance structures. It is only
2769 * valid to do these two operations in a row at all because we
2770 * know that 'from' will not move past 'to' and vice versa. */
2771 resort_bals(from, bals, n_bals);
2772 resort_bals(to, bals, n_bals);
2779 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2780 * historical data to decay to <1% in 7 rebalancing runs. */
2781 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2787 bond_send_learning_packets(struct port *port)
2789 struct bridge *br = port->bridge;
2790 struct mac_entry *e;
2791 struct ofpbuf packet;
2792 int error, n_packets, n_errors;
2794 if (!port->n_ifaces || port->active_iface < 0) {
2798 ofpbuf_init(&packet, 128);
2799 error = n_packets = n_errors = 0;
2800 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2801 union ofp_action actions[2], *a;
2807 if (e->port == port->port_idx
2808 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2812 /* Compose actions. */
2813 memset(actions, 0, sizeof actions);
2816 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2817 a->vlan_vid.len = htons(sizeof *a);
2818 a->vlan_vid.vlan_vid = htons(e->vlan);
2821 a->output.type = htons(OFPAT_OUTPUT);
2822 a->output.len = htons(sizeof *a);
2823 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2828 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2830 flow_extract(&packet, 0, ODPP_NONE, &flow);
2831 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2838 ofpbuf_uninit(&packet);
2841 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2842 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2843 "packets, last error was: %s",
2844 port->name, n_errors, n_packets, strerror(error));
2846 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2847 port->name, n_packets);
2851 /* Bonding unixctl user interface functions. */
2854 bond_unixctl_list(struct unixctl_conn *conn,
2855 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2857 struct ds ds = DS_EMPTY_INITIALIZER;
2858 const struct bridge *br;
2860 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2862 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2865 for (i = 0; i < br->n_ports; i++) {
2866 const struct port *port = br->ports[i];
2867 if (port->n_ifaces > 1) {
2870 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2871 for (j = 0; j < port->n_ifaces; j++) {
2872 const struct iface *iface = port->ifaces[j];
2874 ds_put_cstr(&ds, ", ");
2876 ds_put_cstr(&ds, iface->name);
2878 ds_put_char(&ds, '\n');
2882 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2886 static struct port *
2887 bond_find(const char *name)
2889 const struct bridge *br;
2891 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2894 for (i = 0; i < br->n_ports; i++) {
2895 struct port *port = br->ports[i];
2896 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2905 bond_unixctl_show(struct unixctl_conn *conn,
2906 const char *args, void *aux OVS_UNUSED)
2908 struct ds ds = DS_EMPTY_INITIALIZER;
2909 const struct port *port;
2912 port = bond_find(args);
2914 unixctl_command_reply(conn, 501, "no such bond");
2918 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2919 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2920 ds_put_format(&ds, "next rebalance: %lld ms\n",
2921 port->bond_next_rebalance - time_msec());
2922 for (j = 0; j < port->n_ifaces; j++) {
2923 const struct iface *iface = port->ifaces[j];
2924 struct bond_entry *be;
2927 ds_put_format(&ds, "slave %s: %s\n",
2928 iface->name, iface->enabled ? "enabled" : "disabled");
2929 if (j == port->active_iface) {
2930 ds_put_cstr(&ds, "\tactive slave\n");
2932 if (iface->delay_expires != LLONG_MAX) {
2933 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2934 iface->enabled ? "downdelay" : "updelay",
2935 iface->delay_expires - time_msec());
2939 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2940 int hash = be - port->bond_hash;
2941 struct mac_entry *me;
2943 if (be->iface_idx != j) {
2947 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2948 hash, be->tx_bytes / 1024);
2951 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2952 &port->bridge->ml->lrus) {
2955 if (bond_hash(me->mac) == hash
2956 && me->port != port->port_idx
2957 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2958 && dp_ifidx == iface->dp_ifidx)
2960 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2961 ETH_ADDR_ARGS(me->mac));
2966 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2971 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2972 void *aux OVS_UNUSED)
2974 char *args = (char *) args_;
2975 char *save_ptr = NULL;
2976 char *bond_s, *hash_s, *slave_s;
2977 uint8_t mac[ETH_ADDR_LEN];
2979 struct iface *iface;
2980 struct bond_entry *entry;
2983 bond_s = strtok_r(args, " ", &save_ptr);
2984 hash_s = strtok_r(NULL, " ", &save_ptr);
2985 slave_s = strtok_r(NULL, " ", &save_ptr);
2987 unixctl_command_reply(conn, 501,
2988 "usage: bond/migrate BOND HASH SLAVE");
2992 port = bond_find(bond_s);
2994 unixctl_command_reply(conn, 501, "no such bond");
2998 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2999 == ETH_ADDR_SCAN_COUNT) {
3000 hash = bond_hash(mac);
3001 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3002 hash = atoi(hash_s) & BOND_MASK;
3004 unixctl_command_reply(conn, 501, "bad hash");
3008 iface = port_lookup_iface(port, slave_s);
3010 unixctl_command_reply(conn, 501, "no such slave");
3014 if (!iface->enabled) {
3015 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3019 entry = &port->bond_hash[hash];
3020 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3021 entry->iface_idx = iface->port_ifidx;
3022 entry->iface_tag = tag_create_random();
3023 port->bond_compat_is_stale = true;
3024 unixctl_command_reply(conn, 200, "migrated");
3028 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3029 void *aux OVS_UNUSED)
3031 char *args = (char *) args_;
3032 char *save_ptr = NULL;
3033 char *bond_s, *slave_s;
3035 struct iface *iface;
3037 bond_s = strtok_r(args, " ", &save_ptr);
3038 slave_s = strtok_r(NULL, " ", &save_ptr);
3040 unixctl_command_reply(conn, 501,
3041 "usage: bond/set-active-slave BOND SLAVE");
3045 port = bond_find(bond_s);
3047 unixctl_command_reply(conn, 501, "no such bond");
3051 iface = port_lookup_iface(port, slave_s);
3053 unixctl_command_reply(conn, 501, "no such slave");
3057 if (!iface->enabled) {
3058 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3062 if (port->active_iface != iface->port_ifidx) {
3063 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3064 port->active_iface = iface->port_ifidx;
3065 port->active_iface_tag = tag_create_random();
3066 VLOG_INFO("port %s: active interface is now %s",
3067 port->name, iface->name);
3068 bond_send_learning_packets(port);
3069 unixctl_command_reply(conn, 200, "done");
3071 unixctl_command_reply(conn, 200, "no change");
3076 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3078 char *args = (char *) args_;
3079 char *save_ptr = NULL;
3080 char *bond_s, *slave_s;
3082 struct iface *iface;
3084 bond_s = strtok_r(args, " ", &save_ptr);
3085 slave_s = strtok_r(NULL, " ", &save_ptr);
3087 unixctl_command_reply(conn, 501,
3088 "usage: bond/enable/disable-slave BOND SLAVE");
3092 port = bond_find(bond_s);
3094 unixctl_command_reply(conn, 501, "no such bond");
3098 iface = port_lookup_iface(port, slave_s);
3100 unixctl_command_reply(conn, 501, "no such slave");
3104 bond_enable_slave(iface, enable);
3105 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3109 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3110 void *aux OVS_UNUSED)
3112 enable_slave(conn, args, true);
3116 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3117 void *aux OVS_UNUSED)
3119 enable_slave(conn, args, false);
3123 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3124 void *aux OVS_UNUSED)
3126 uint8_t mac[ETH_ADDR_LEN];
3130 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3131 == ETH_ADDR_SCAN_COUNT) {
3132 hash = bond_hash(mac);
3134 hash_cstr = xasprintf("%u", hash);
3135 unixctl_command_reply(conn, 200, hash_cstr);
3138 unixctl_command_reply(conn, 501, "invalid mac");
3145 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3146 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3147 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3148 unixctl_command_register("bond/set-active-slave",
3149 bond_unixctl_set_active_slave, NULL);
3150 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3152 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3154 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3157 /* Port functions. */
3159 static struct port *
3160 port_create(struct bridge *br, const char *name)
3164 port = xzalloc(sizeof *port);
3166 port->port_idx = br->n_ports;
3168 port->trunks = NULL;
3169 port->name = xstrdup(name);
3170 port->active_iface = -1;
3172 if (br->n_ports >= br->allocated_ports) {
3173 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3176 br->ports[br->n_ports++] = port;
3177 shash_add_assert(&br->port_by_name, port->name, port);
3179 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3186 get_port_other_config(const struct ovsrec_port *port, const char *key,
3187 const char *default_value)
3189 const char *value = get_ovsrec_key_value(key,
3190 port->key_other_config,
3191 port->value_other_config,
3192 port->n_other_config);
3193 return value ? value : default_value;
3197 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3199 struct shash new_ifaces;
3202 /* Collect list of new interfaces. */
3203 shash_init(&new_ifaces);
3204 for (i = 0; i < cfg->n_interfaces; i++) {
3205 const char *name = cfg->interfaces[i]->name;
3206 shash_add_once(&new_ifaces, name, NULL);
3209 /* Get rid of deleted interfaces. */
3210 for (i = 0; i < port->n_ifaces; ) {
3211 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3212 iface_destroy(port->ifaces[i]);
3218 shash_destroy(&new_ifaces);
3222 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3224 struct shash new_ifaces;
3225 long long int next_rebalance;
3226 unsigned long *trunks;
3232 /* Update settings. */
3233 port->updelay = cfg->bond_updelay;
3234 if (port->updelay < 0) {
3237 port->updelay = cfg->bond_downdelay;
3238 if (port->downdelay < 0) {
3239 port->downdelay = 0;
3241 port->bond_rebalance_interval = atoi(
3242 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3243 if (port->bond_rebalance_interval < 1000) {
3244 port->bond_rebalance_interval = 1000;
3246 next_rebalance = time_msec() + port->bond_rebalance_interval;
3247 if (port->bond_next_rebalance > next_rebalance) {
3248 port->bond_next_rebalance = next_rebalance;
3251 /* Add new interfaces and update 'cfg' member of existing ones. */
3252 shash_init(&new_ifaces);
3253 for (i = 0; i < cfg->n_interfaces; i++) {
3254 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3255 struct iface *iface;
3257 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3258 VLOG_WARN("port %s: %s specified twice as port interface",
3259 port->name, if_cfg->name);
3263 iface = iface_lookup(port->bridge, if_cfg->name);
3265 if (iface->port != port) {
3266 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3268 port->bridge->name, if_cfg->name, iface->port->name);
3271 iface->cfg = if_cfg;
3273 iface_create(port, if_cfg);
3276 shash_destroy(&new_ifaces);
3281 if (port->n_ifaces < 2) {
3283 if (vlan >= 0 && vlan <= 4095) {
3284 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3289 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3290 * they even work as-is. But they have not been tested. */
3291 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3295 if (port->vlan != vlan) {
3297 bridge_flush(port->bridge);
3300 /* Get trunked VLANs. */
3302 if (vlan < 0 && cfg->n_trunks) {
3306 trunks = bitmap_allocate(4096);
3308 for (i = 0; i < cfg->n_trunks; i++) {
3309 int trunk = cfg->trunks[i];
3311 bitmap_set1(trunks, trunk);
3317 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3318 port->name, cfg->n_trunks);
3320 if (n_errors == cfg->n_trunks) {
3321 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3323 bitmap_free(trunks);
3326 } else if (vlan >= 0 && cfg->n_trunks) {
3327 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3331 ? port->trunks != NULL
3332 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3333 bridge_flush(port->bridge);
3335 bitmap_free(port->trunks);
3336 port->trunks = trunks;
3340 port_destroy(struct port *port)
3343 struct bridge *br = port->bridge;
3347 proc_net_compat_update_vlan(port->name, NULL, 0);
3348 proc_net_compat_update_bond(port->name, NULL);
3350 for (i = 0; i < MAX_MIRRORS; i++) {
3351 struct mirror *m = br->mirrors[i];
3352 if (m && m->out_port == port) {
3357 while (port->n_ifaces > 0) {
3358 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3361 shash_find_and_delete_assert(&br->port_by_name, port->name);
3363 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3364 del->port_idx = port->port_idx;
3367 bitmap_free(port->trunks);
3374 static struct port *
3375 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3377 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3378 return iface ? iface->port : NULL;
3381 static struct port *
3382 port_lookup(const struct bridge *br, const char *name)
3384 return shash_find_data(&br->port_by_name, name);
3387 static struct iface *
3388 port_lookup_iface(const struct port *port, const char *name)
3390 struct iface *iface = iface_lookup(port->bridge, name);
3391 return iface && iface->port == port ? iface : NULL;
3395 port_update_bonding(struct port *port)
3397 if (port->n_ifaces < 2) {
3398 /* Not a bonded port. */
3399 if (port->bond_hash) {
3400 free(port->bond_hash);
3401 port->bond_hash = NULL;
3402 port->bond_compat_is_stale = true;
3403 port->bond_fake_iface = false;
3406 if (!port->bond_hash) {
3409 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3410 for (i = 0; i <= BOND_MASK; i++) {
3411 struct bond_entry *e = &port->bond_hash[i];
3415 port->no_ifaces_tag = tag_create_random();
3416 bond_choose_active_iface(port);
3417 port->bond_next_rebalance
3418 = time_msec() + port->bond_rebalance_interval;
3420 if (port->cfg->bond_fake_iface) {
3421 port->bond_next_fake_iface_update = time_msec();
3424 port->bond_compat_is_stale = true;
3425 port->bond_fake_iface = port->cfg->bond_fake_iface;
3430 port_update_bond_compat(struct port *port)
3432 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3433 struct compat_bond bond;
3436 if (port->n_ifaces < 2) {
3437 proc_net_compat_update_bond(port->name, NULL);
3442 bond.updelay = port->updelay;
3443 bond.downdelay = port->downdelay;
3446 bond.hashes = compat_hashes;
3447 if (port->bond_hash) {
3448 const struct bond_entry *e;
3449 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3450 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3451 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3452 cbh->hash = e - port->bond_hash;
3453 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3458 bond.n_slaves = port->n_ifaces;
3459 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3460 for (i = 0; i < port->n_ifaces; i++) {
3461 struct iface *iface = port->ifaces[i];
3462 struct compat_bond_slave *slave = &bond.slaves[i];
3463 slave->name = iface->name;
3465 /* We need to make the same determination as the Linux bonding
3466 * code to determine whether a slave should be consider "up".
3467 * The Linux function bond_miimon_inspect() supports four
3468 * BOND_LINK_* states:
3470 * - BOND_LINK_UP: carrier detected, updelay has passed.
3471 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3472 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3473 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3475 * The function bond_info_show_slave() only considers BOND_LINK_UP
3476 * to be "up" and anything else to be "down".
3478 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3482 netdev_get_etheraddr(iface->netdev, slave->mac);
3485 if (port->bond_fake_iface) {
3486 struct netdev *bond_netdev;
3488 if (!netdev_open_default(port->name, &bond_netdev)) {
3490 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3492 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3494 netdev_close(bond_netdev);
3498 proc_net_compat_update_bond(port->name, &bond);
3503 port_update_vlan_compat(struct port *port)
3505 struct bridge *br = port->bridge;
3506 char *vlandev_name = NULL;
3508 if (port->vlan > 0) {
3509 /* Figure out the name that the VLAN device should actually have, if it
3510 * existed. This takes some work because the VLAN device would not
3511 * have port->name in its name; rather, it would have the trunk port's
3512 * name, and 'port' would be attached to a bridge that also had the
3513 * VLAN device one of its ports. So we need to find a trunk port that
3514 * includes port->vlan.
3516 * There might be more than one candidate. This doesn't happen on
3517 * XenServer, so if it happens we just pick the first choice in
3518 * alphabetical order instead of creating multiple VLAN devices. */
3520 for (i = 0; i < br->n_ports; i++) {
3521 struct port *p = br->ports[i];
3522 if (port_trunks_vlan(p, port->vlan)
3524 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3526 uint8_t ea[ETH_ADDR_LEN];
3527 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3528 if (!eth_addr_is_multicast(ea) &&
3529 !eth_addr_is_reserved(ea) &&
3530 !eth_addr_is_zero(ea)) {
3531 vlandev_name = p->name;
3536 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3539 /* Interface functions. */
3541 static struct iface *
3542 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3544 struct bridge *br = port->bridge;
3545 struct iface *iface;
3546 char *name = if_cfg->name;
3549 iface = xzalloc(sizeof *iface);
3551 iface->port_ifidx = port->n_ifaces;
3552 iface->name = xstrdup(name);
3553 iface->dp_ifidx = -1;
3554 iface->tag = tag_create_random();
3555 iface->delay_expires = LLONG_MAX;
3556 iface->netdev = NULL;
3557 iface->cfg = if_cfg;
3559 shash_add_assert(&br->iface_by_name, iface->name, iface);
3561 /* Attempt to create the network interface in case it doesn't exist yet. */
3562 if (!iface_is_internal(br, iface->name)) {
3563 error = set_up_iface(if_cfg, iface, true);
3565 VLOG_WARN("could not create iface %s: %s", iface->name,
3568 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3575 if (port->n_ifaces >= port->allocated_ifaces) {
3576 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3577 sizeof *port->ifaces);
3579 port->ifaces[port->n_ifaces++] = iface;
3580 if (port->n_ifaces > 1) {
3581 br->has_bonded_ports = true;
3584 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3592 iface_destroy(struct iface *iface)
3595 struct port *port = iface->port;
3596 struct bridge *br = port->bridge;
3597 bool del_active = port->active_iface == iface->port_ifidx;
3600 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3602 if (iface->dp_ifidx >= 0) {
3603 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3606 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3607 del->port_ifidx = iface->port_ifidx;
3609 netdev_close(iface->netdev);
3612 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3613 bond_choose_active_iface(port);
3614 bond_send_learning_packets(port);
3620 bridge_flush(port->bridge);
3624 static struct iface *
3625 iface_lookup(const struct bridge *br, const char *name)
3627 return shash_find_data(&br->iface_by_name, name);
3630 static struct iface *
3631 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3633 return port_array_get(&br->ifaces, dp_ifidx);
3636 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3637 * 'br', that is, an interface that is entirely simulated within the datapath.
3638 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3639 * interfaces are created by setting "iface.<iface>.internal = true".
3641 * In addition, we have a kluge-y feature that creates an internal port with
3642 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3643 * This feature needs to go away in the long term. Until then, this is one
3644 * reason why this function takes a name instead of a struct iface: the fake
3645 * interfaces created this way do not have a struct iface. */
3647 iface_is_internal(const struct bridge *br, const char *if_name)
3649 struct iface *iface;
3652 if (!strcmp(if_name, br->name)) {
3656 iface = iface_lookup(br, if_name);
3657 if (iface && !strcmp(iface->cfg->type, "internal")) {
3661 port = port_lookup(br, if_name);
3662 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3668 /* Set Ethernet address of 'iface', if one is specified in the configuration
3671 iface_set_mac(struct iface *iface)
3673 uint8_t ea[ETH_ADDR_LEN];
3675 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3676 if (eth_addr_is_multicast(ea)) {
3677 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3679 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3680 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3681 iface->name, iface->name);
3683 int error = netdev_set_etheraddr(iface->netdev, ea);
3685 VLOG_ERR("interface %s: setting MAC failed (%s)",
3686 iface->name, strerror(error));
3692 /* Port mirroring. */
3695 mirror_reconfigure(struct bridge *br)
3697 struct shash old_mirrors, new_mirrors;
3698 struct shash_node *node;
3699 unsigned long *rspan_vlans;
3702 /* Collect old mirrors. */
3703 shash_init(&old_mirrors);
3704 for (i = 0; i < MAX_MIRRORS; i++) {
3705 if (br->mirrors[i]) {
3706 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3710 /* Collect new mirrors. */
3711 shash_init(&new_mirrors);
3712 for (i = 0; i < br->cfg->n_mirrors; i++) {
3713 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3714 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3715 VLOG_WARN("bridge %s: %s specified twice as mirror",
3716 br->name, cfg->name);
3720 /* Get rid of deleted mirrors and add new mirrors. */
3721 SHASH_FOR_EACH (node, &old_mirrors) {
3722 if (!shash_find(&new_mirrors, node->name)) {
3723 mirror_destroy(node->data);
3726 SHASH_FOR_EACH (node, &new_mirrors) {
3727 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3729 mirror = mirror_create(br, node->name);
3734 mirror_reconfigure_one(mirror, node->data);
3736 shash_destroy(&old_mirrors);
3737 shash_destroy(&new_mirrors);
3739 /* Update port reserved status. */
3740 for (i = 0; i < br->n_ports; i++) {
3741 br->ports[i]->is_mirror_output_port = false;
3743 for (i = 0; i < MAX_MIRRORS; i++) {
3744 struct mirror *m = br->mirrors[i];
3745 if (m && m->out_port) {
3746 m->out_port->is_mirror_output_port = true;
3750 /* Update flooded vlans (for RSPAN). */
3752 if (br->cfg->n_flood_vlans) {
3753 rspan_vlans = bitmap_allocate(4096);
3755 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3756 int64_t vlan = br->cfg->flood_vlans[i];
3757 if (vlan >= 0 && vlan < 4096) {
3758 bitmap_set1(rspan_vlans, vlan);
3759 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3762 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3767 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3772 static struct mirror *
3773 mirror_create(struct bridge *br, const char *name)
3778 for (i = 0; ; i++) {
3779 if (i >= MAX_MIRRORS) {
3780 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3781 "cannot create %s", br->name, MAX_MIRRORS, name);
3784 if (!br->mirrors[i]) {
3789 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3792 br->mirrors[i] = m = xzalloc(sizeof *m);
3795 m->name = xstrdup(name);
3796 shash_init(&m->src_ports);
3797 shash_init(&m->dst_ports);
3807 mirror_destroy(struct mirror *m)
3810 struct bridge *br = m->bridge;
3813 for (i = 0; i < br->n_ports; i++) {
3814 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3815 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3818 shash_destroy(&m->src_ports);
3819 shash_destroy(&m->dst_ports);
3822 m->bridge->mirrors[m->idx] = NULL;
3830 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3831 struct shash *names)
3835 for (i = 0; i < n_ports; i++) {
3836 const char *name = ports[i]->name;
3837 if (port_lookup(m->bridge, name)) {
3838 shash_add_once(names, name, NULL);
3840 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3841 "port %s", m->bridge->name, m->name, name);
3847 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3853 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
3855 for (i = 0; i < cfg->n_select_vlan; i++) {
3856 int64_t vlan = cfg->select_vlan[i];
3857 if (vlan < 0 || vlan > 4095) {
3858 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3859 m->bridge->name, m->name, vlan);
3861 (*vlans)[n_vlans++] = vlan;
3868 vlan_is_mirrored(const struct mirror *m, int vlan)
3872 for (i = 0; i < m->n_vlans; i++) {
3873 if (m->vlans[i] == vlan) {
3881 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3885 for (i = 0; i < m->n_vlans; i++) {
3886 if (port_trunks_vlan(p, m->vlans[i])) {
3894 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3896 struct shash src_ports, dst_ports;
3897 mirror_mask_t mirror_bit;
3898 struct port *out_port;
3904 /* Get output port. */
3905 if (cfg->output_port) {
3906 out_port = port_lookup(m->bridge, cfg->output_port->name);
3908 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3909 m->bridge->name, m->name);
3915 if (cfg->output_vlan) {
3916 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3917 "output vlan; ignoring output vlan",
3918 m->bridge->name, m->name);
3920 } else if (cfg->output_vlan) {
3922 out_vlan = *cfg->output_vlan;
3924 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3925 m->bridge->name, m->name);
3930 shash_init(&src_ports);
3931 shash_init(&dst_ports);
3932 if (cfg->select_all) {
3933 for (i = 0; i < m->bridge->n_ports; i++) {
3934 const char *name = m->bridge->ports[i]->name;
3935 shash_add_once(&src_ports, name, NULL);
3936 shash_add_once(&dst_ports, name, NULL);
3941 /* Get ports, and drop duplicates and ports that don't exist. */
3942 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
3944 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
3947 /* Get all the vlans, and drop duplicate and invalid vlans. */
3948 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
3951 /* Update mirror data. */
3952 if (!shash_equal_keys(&m->src_ports, &src_ports)
3953 || !shash_equal_keys(&m->dst_ports, &dst_ports)
3954 || m->n_vlans != n_vlans
3955 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3956 || m->out_port != out_port
3957 || m->out_vlan != out_vlan) {
3958 bridge_flush(m->bridge);
3960 shash_swap(&m->src_ports, &src_ports);
3961 shash_swap(&m->dst_ports, &dst_ports);
3964 m->n_vlans = n_vlans;
3965 m->out_port = out_port;
3966 m->out_vlan = out_vlan;
3969 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3970 for (i = 0; i < m->bridge->n_ports; i++) {
3971 struct port *port = m->bridge->ports[i];
3973 if (shash_find(&m->src_ports, port->name)
3976 ? port_trunks_any_mirrored_vlan(m, port)
3977 : vlan_is_mirrored(m, port->vlan)))) {
3978 port->src_mirrors |= mirror_bit;
3980 port->src_mirrors &= ~mirror_bit;
3983 if (shash_find(&m->dst_ports, port->name)) {
3984 port->dst_mirrors |= mirror_bit;
3986 port->dst_mirrors &= ~mirror_bit;
3991 shash_destroy(&src_ports);
3992 shash_destroy(&dst_ports);