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>
24 #include <openflow/openflow.h>
29 #include <sys/socket.h>
30 #include <sys/types.h>
36 #include "dynamic-string.h"
40 #include "mac-learning.h"
43 #include "ofp-print.h"
45 #include "ofproto/netflow.h"
46 #include "ofproto/ofproto.h"
48 #include "poll-loop.h"
49 #include "port-array.h"
50 #include "proc-net-compat.h"
54 #include "socket-util.h"
55 #include "stream-ssl.h"
61 #include "vswitchd/vswitch-idl.h"
62 #include "xenserver.h"
65 #define THIS_MODULE VLM_bridge
74 /* These members are always valid. */
75 struct port *port; /* Containing port. */
76 size_t port_ifidx; /* Index within containing port. */
77 char *name; /* Host network device name. */
78 tag_type tag; /* Tag associated with this interface. */
79 long long delay_expires; /* Time after which 'enabled' may change. */
81 /* These members are valid only after bridge_reconfigure() causes them to
83 int dp_ifidx; /* Index within kernel datapath. */
84 struct netdev *netdev; /* Network device. */
85 bool enabled; /* May be chosen for flows? */
87 /* This member is only valid *during* bridge_reconfigure(). */
88 const struct ovsrec_interface *cfg;
91 #define BOND_MASK 0xff
93 int iface_idx; /* Index of assigned iface, or -1 if none. */
94 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
95 tag_type iface_tag; /* Tag associated with iface_idx. */
98 #define MAX_MIRRORS 32
99 typedef uint32_t mirror_mask_t;
100 #define MIRROR_MASK_C(X) UINT32_C(X)
101 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
103 struct bridge *bridge;
107 /* Selection criteria. */
108 struct shash src_ports; /* Name is port name; data is always NULL. */
109 struct shash dst_ports; /* Name is port name; data is always NULL. */
114 struct port *out_port;
118 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
120 struct bridge *bridge;
122 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
123 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
126 /* An ordinary bridge port has 1 interface.
127 * A bridge port for bonding has at least 2 interfaces. */
128 struct iface **ifaces;
129 size_t n_ifaces, allocated_ifaces;
132 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
133 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
134 tag_type active_iface_tag; /* Tag for bcast flows. */
135 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
136 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
137 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
138 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
140 /* Port mirroring info. */
141 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
142 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
143 bool is_mirror_output_port; /* Does port mirroring send frames here? */
145 /* This member is only valid *during* bridge_reconfigure(). */
146 const struct ovsrec_port *cfg;
149 #define DP_MAX_PORTS 255
151 struct list node; /* Node in global list of bridges. */
152 char *name; /* User-specified arbitrary name. */
153 struct mac_learning *ml; /* MAC learning table. */
154 bool sent_config_request; /* Successfully sent config request? */
155 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
157 /* Support for remote controllers. */
158 char *controller; /* NULL if there is no remote controller;
159 * "discover" to do controller discovery;
160 * otherwise a vconn name. */
162 /* OpenFlow switch processing. */
163 struct ofproto *ofproto; /* OpenFlow switch. */
165 /* Kernel datapath information. */
166 struct dpif *dpif; /* Datapath. */
167 struct port_array ifaces; /* Indexed by kernel datapath port number. */
171 size_t n_ports, allocated_ports;
174 bool has_bonded_ports;
175 long long int bond_next_rebalance;
180 /* Flow statistics gathering. */
181 time_t next_stats_request;
183 /* Port mirroring. */
184 struct mirror *mirrors[MAX_MIRRORS];
186 /* This member is only valid *during* bridge_reconfigure(). */
187 const struct ovsrec_bridge *cfg;
190 /* List of all bridges. */
191 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
193 /* Maximum number of datapaths. */
194 enum { DP_MAX = 256 };
196 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
197 static void bridge_destroy(struct bridge *);
198 static struct bridge *bridge_lookup(const char *name);
199 static unixctl_cb_func bridge_unixctl_dump_flows;
200 static int bridge_run_one(struct bridge *);
201 static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
203 static void bridge_reconfigure_controller(const struct ovsrec_open_vswitch *,
205 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
206 static void bridge_fetch_dp_ifaces(struct bridge *);
207 static void bridge_flush(struct bridge *);
208 static void bridge_pick_local_hw_addr(struct bridge *,
209 uint8_t ea[ETH_ADDR_LEN],
210 struct iface **hw_addr_iface);
211 static uint64_t bridge_pick_datapath_id(struct bridge *,
212 const uint8_t bridge_ea[ETH_ADDR_LEN],
213 struct iface *hw_addr_iface);
214 static struct iface *bridge_get_local_iface(struct bridge *);
215 static uint64_t dpid_from_hash(const void *, size_t nbytes);
217 static unixctl_cb_func bridge_unixctl_fdb_show;
219 static void bond_init(void);
220 static void bond_run(struct bridge *);
221 static void bond_wait(struct bridge *);
222 static void bond_rebalance_port(struct port *);
223 static void bond_send_learning_packets(struct port *);
224 static void bond_enable_slave(struct iface *iface, bool enable);
226 static struct port *port_create(struct bridge *, const char *name);
227 static void port_reconfigure(struct port *, const struct ovsrec_port *);
228 static void port_destroy(struct port *);
229 static struct port *port_lookup(const struct bridge *, const char *name);
230 static struct iface *port_lookup_iface(const struct port *, const char *name);
231 static struct port *port_from_dp_ifidx(const struct bridge *,
233 static void port_update_bond_compat(struct port *);
234 static void port_update_vlan_compat(struct port *);
235 static void port_update_bonding(struct port *);
237 static struct mirror *mirror_create(struct bridge *, const char *name);
238 static void mirror_destroy(struct mirror *);
239 static void mirror_reconfigure(struct bridge *);
240 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
241 static bool vlan_is_mirrored(const struct mirror *, int vlan);
243 static struct iface *iface_create(struct port *port,
244 const struct ovsrec_interface *if_cfg);
245 static void iface_destroy(struct iface *);
246 static struct iface *iface_lookup(const struct bridge *, const char *name);
247 static struct iface *iface_from_dp_ifidx(const struct bridge *,
249 static bool iface_is_internal(const struct bridge *, const char *name);
250 static void iface_set_mac(struct iface *);
252 /* Hooks into ofproto processing. */
253 static struct ofhooks bridge_ofhooks;
255 /* Public functions. */
257 /* Adds the name of each interface used by a bridge, including local and
258 * internal ports, to 'svec'. */
260 bridge_get_ifaces(struct svec *svec)
262 struct bridge *br, *next;
265 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
266 for (i = 0; i < br->n_ports; i++) {
267 struct port *port = br->ports[i];
269 for (j = 0; j < port->n_ifaces; j++) {
270 struct iface *iface = port->ifaces[j];
271 if (iface->dp_ifidx < 0) {
272 VLOG_ERR("%s interface not in datapath %s, ignoring",
273 iface->name, dpif_name(br->dpif));
275 if (iface->dp_ifidx != ODPP_LOCAL) {
276 svec_add(svec, iface->name);
285 bridge_init(const struct ovsrec_open_vswitch *cfg)
287 struct svec bridge_names;
288 struct svec dpif_names, dpif_types;
291 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
293 svec_init(&bridge_names);
294 for (i = 0; i < cfg->n_bridges; i++) {
295 svec_add(&bridge_names, cfg->bridges[i]->name);
297 svec_sort(&bridge_names);
299 svec_init(&dpif_names);
300 svec_init(&dpif_types);
301 dp_enumerate_types(&dpif_types);
302 for (i = 0; i < dpif_types.n; i++) {
307 dp_enumerate_names(dpif_types.names[i], &dpif_names);
309 for (j = 0; j < dpif_names.n; j++) {
310 retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
312 struct svec all_names;
315 svec_init(&all_names);
316 dpif_get_all_names(dpif, &all_names);
317 for (k = 0; k < all_names.n; k++) {
318 if (svec_contains(&bridge_names, all_names.names[k])) {
324 svec_destroy(&all_names);
329 svec_destroy(&dpif_names);
330 svec_destroy(&dpif_types);
332 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
336 bridge_reconfigure(cfg);
341 config_string_change(const char *value, char **valuep)
343 if (value && (!*valuep || strcmp(value, *valuep))) {
345 *valuep = xstrdup(value);
353 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
355 /* XXX SSL should be configurable on a per-bridge basis.
356 * XXX should be possible to de-configure SSL. */
357 static char *private_key_file;
358 static char *certificate_file;
359 static char *cacert_file;
363 /* XXX We can't un-set SSL settings. */
367 if (config_string_change(ssl->private_key, &private_key_file)) {
368 stream_ssl_set_private_key_file(private_key_file);
371 if (config_string_change(ssl->certificate, &certificate_file)) {
372 stream_ssl_set_certificate_file(certificate_file);
375 /* We assume that even if the filename hasn't changed, if the CA cert
376 * file has been removed, that we want to move back into
377 * boot-strapping mode. This opens a small security hole, because
378 * the old certificate will still be trusted until vSwitch is
379 * restarted. We may want to address this in vconn's SSL library. */
380 if (config_string_change(ssl->ca_cert, &cacert_file)
381 || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) {
382 stream_ssl_set_ca_cert_file(cacert_file, ssl->bootstrap_ca_cert);
387 /* Attempt to create the network device 'iface_name' through the netdev
390 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
393 struct shash_node *node;
394 struct shash options;
398 shash_init(&options);
399 for (i = 0; i < iface_cfg->n_options; i++) {
400 shash_add(&options, iface_cfg->key_options[i],
401 xstrdup(iface_cfg->value_options[i]));
405 struct netdev_options netdev_options;
407 memset(&netdev_options, 0, sizeof netdev_options);
408 netdev_options.name = iface_cfg->name;
409 netdev_options.type = iface_cfg->type;
410 netdev_options.args = &options;
411 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
412 netdev_options.may_create = true;
413 if (iface_is_internal(iface->port->bridge, iface_cfg->name)) {
414 netdev_options.may_open = true;
417 error = netdev_open(&netdev_options, &iface->netdev);
420 netdev_get_carrier(iface->netdev, &iface->enabled);
422 } else if (iface->netdev) {
423 const char *netdev_type = netdev_get_type(iface->netdev);
424 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
425 ? iface_cfg->type : NULL;
427 if (!iface_type || !strcmp(netdev_type, iface_type)) {
428 error = netdev_reconfigure(iface->netdev, &options);
430 VLOG_WARN("%s: attempting change device type from %s to %s",
431 iface_cfg->name, netdev_type, iface_type);
436 SHASH_FOR_EACH (node, &options) {
439 shash_destroy(&options);
445 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
447 return set_up_iface(iface_cfg, iface, false);
451 check_iface_netdev(struct bridge *br UNUSED, struct iface *iface,
454 if (!iface->netdev) {
455 int error = set_up_iface(iface->cfg, iface, true);
457 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
467 check_iface_dp_ifidx(struct bridge *br, struct iface *iface, void *aux UNUSED)
469 if (iface->dp_ifidx >= 0) {
470 VLOG_DBG("%s has interface %s on port %d",
472 iface->name, iface->dp_ifidx);
475 VLOG_ERR("%s interface not in %s, dropping",
476 iface->name, dpif_name(br->dpif));
482 set_iface_properties(struct bridge *br UNUSED, struct iface *iface,
485 /* Set policing attributes. */
486 netdev_set_policing(iface->netdev,
487 iface->cfg->ingress_policing_rate,
488 iface->cfg->ingress_policing_burst);
490 /* Set MAC address of internal interfaces other than the local
492 if (iface->dp_ifidx != ODPP_LOCAL
493 && iface_is_internal(br, iface->name)) {
494 iface_set_mac(iface);
500 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
501 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
502 * deletes from 'br' any ports that no longer have any interfaces. */
504 iterate_and_prune_ifaces(struct bridge *br,
505 bool (*cb)(struct bridge *, struct iface *,
511 for (i = 0; i < br->n_ports; ) {
512 struct port *port = br->ports[i];
513 for (j = 0; j < port->n_ifaces; ) {
514 struct iface *iface = port->ifaces[j];
515 if (cb(br, iface, aux)) {
518 iface_destroy(iface);
522 if (port->n_ifaces) {
525 VLOG_ERR("%s port has no interfaces, dropping", port->name);
532 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
534 struct ovsdb_idl_txn *txn;
535 struct shash old_br, new_br;
536 struct shash_node *node;
537 struct bridge *br, *next;
540 COVERAGE_INC(bridge_reconfigure);
542 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
544 /* Collect old and new bridges. */
547 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
548 shash_add(&old_br, br->name, br);
550 for (i = 0; i < ovs_cfg->n_bridges; i++) {
551 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
552 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
553 VLOG_WARN("more than one bridge named %s", br_cfg->name);
557 /* Get rid of deleted bridges and add new bridges. */
558 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
559 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
566 SHASH_FOR_EACH (node, &new_br) {
567 const char *br_name = node->name;
568 const struct ovsrec_bridge *br_cfg = node->data;
569 br = shash_find_data(&old_br, br_name);
571 /* If the bridge datapath type has changed, we need to tear it
572 * down and recreate. */
573 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
575 bridge_create(br_cfg);
578 bridge_create(br_cfg);
581 shash_destroy(&old_br);
582 shash_destroy(&new_br);
586 bridge_configure_ssl(ovs_cfg->ssl);
589 /* Reconfigure all bridges. */
590 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
591 bridge_reconfigure_one(ovs_cfg, br);
594 /* Add and delete ports on all datapaths.
596 * The kernel will reject any attempt to add a given port to a datapath if
597 * that port already belongs to a different datapath, so we must do all
598 * port deletions before any port additions. */
599 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
600 struct odp_port *dpif_ports;
602 struct shash want_ifaces;
604 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
605 bridge_get_all_ifaces(br, &want_ifaces);
606 for (i = 0; i < n_dpif_ports; i++) {
607 const struct odp_port *p = &dpif_ports[i];
608 if (!shash_find(&want_ifaces, p->devname)
609 && strcmp(p->devname, br->name)) {
610 int retval = dpif_port_del(br->dpif, p->port);
612 VLOG_ERR("failed to remove %s interface from %s: %s",
613 p->devname, dpif_name(br->dpif),
618 shash_destroy(&want_ifaces);
621 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
622 struct odp_port *dpif_ports;
624 struct shash cur_ifaces, want_ifaces;
625 struct shash_node *node;
627 /* Get the set of interfaces currently in this datapath. */
628 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
629 shash_init(&cur_ifaces);
630 for (i = 0; i < n_dpif_ports; i++) {
631 const char *name = dpif_ports[i].devname;
632 if (!shash_find(&cur_ifaces, name)) {
633 shash_add(&cur_ifaces, name, NULL);
638 /* Get the set of interfaces we want on this datapath. */
639 bridge_get_all_ifaces(br, &want_ifaces);
641 SHASH_FOR_EACH (node, &want_ifaces) {
642 const char *if_name = node->name;
643 struct iface *iface = node->data;
645 if (shash_find(&cur_ifaces, if_name)) {
646 /* Already exists, just reconfigure it. */
648 reconfigure_iface(iface->cfg, iface);
651 /* Need to add to datapath. */
655 /* Add to datapath. */
656 internal = iface_is_internal(br, if_name);
657 error = dpif_port_add(br->dpif, if_name,
658 internal ? ODP_PORT_INTERNAL : 0, NULL);
659 if (error == EFBIG) {
660 VLOG_ERR("ran out of valid port numbers on %s",
661 dpif_name(br->dpif));
664 VLOG_ERR("failed to add %s interface to %s: %s",
665 if_name, dpif_name(br->dpif), strerror(error));
669 shash_destroy(&cur_ifaces);
670 shash_destroy(&want_ifaces);
672 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
675 struct iface *local_iface;
676 struct iface *hw_addr_iface;
679 bridge_fetch_dp_ifaces(br);
681 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
682 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
684 /* Pick local port hardware address, datapath ID. */
685 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
686 local_iface = bridge_get_local_iface(br);
688 int error = netdev_set_etheraddr(local_iface->netdev, ea);
690 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
691 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
692 "Ethernet address: %s",
693 br->name, strerror(error));
697 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
698 ofproto_set_datapath_id(br->ofproto, dpid);
700 dpid_string = xasprintf("%012"PRIx64, dpid);
701 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
704 /* Set NetFlow configuration on this bridge. */
705 if (br->cfg->netflow) {
706 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
707 struct netflow_options opts;
709 memset(&opts, 0, sizeof opts);
711 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
712 if (nf_cfg->engine_type) {
713 opts.engine_type = *nf_cfg->engine_type;
715 if (nf_cfg->engine_id) {
716 opts.engine_id = *nf_cfg->engine_id;
719 opts.active_timeout = nf_cfg->active_timeout;
720 if (!opts.active_timeout) {
721 opts.active_timeout = -1;
722 } else if (opts.active_timeout < 0) {
723 VLOG_WARN("bridge %s: active timeout interval set to negative "
724 "value, using default instead (%d seconds)", br->name,
725 NF_ACTIVE_TIMEOUT_DEFAULT);
726 opts.active_timeout = -1;
729 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
730 if (opts.add_id_to_iface) {
731 if (opts.engine_id > 0x7f) {
732 VLOG_WARN("bridge %s: netflow port mangling may conflict "
733 "with another vswitch, choose an engine id less "
734 "than 128", br->name);
736 if (br->n_ports > 508) {
737 VLOG_WARN("bridge %s: netflow port mangling will conflict "
738 "with another port when more than 508 ports are "
743 opts.collectors.n = nf_cfg->n_targets;
744 opts.collectors.names = nf_cfg->targets;
745 if (ofproto_set_netflow(br->ofproto, &opts)) {
746 VLOG_ERR("bridge %s: problem setting netflow collectors",
750 ofproto_set_netflow(br->ofproto, NULL);
753 /* Update the controller and related settings. It would be more
754 * straightforward to call this from bridge_reconfigure_one(), but we
755 * can't do it there for two reasons. First, and most importantly, at
756 * that point we don't know the dp_ifidx of any interfaces that have
757 * been added to the bridge (because we haven't actually added them to
758 * the datapath). Second, at that point we haven't set the datapath ID
759 * yet; when a controller is configured, resetting the datapath ID will
760 * immediately disconnect from the controller, so it's better to set
761 * the datapath ID before the controller. */
762 bridge_reconfigure_controller(ovs_cfg, br);
764 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
765 for (i = 0; i < br->n_ports; i++) {
766 struct port *port = br->ports[i];
768 port_update_vlan_compat(port);
769 port_update_bonding(port);
772 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
773 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
776 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
778 ovsdb_idl_txn_commit(txn);
779 ovsdb_idl_txn_destroy(txn); /* XXX */
783 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
787 for (i = 0; i < br_cfg->n_other_config; i++) {
788 if (!strcmp(br_cfg->key_other_config[i], key)) {
789 return br_cfg->value_other_config[i];
796 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
797 struct iface **hw_addr_iface)
803 *hw_addr_iface = NULL;
805 /* Did the user request a particular MAC? */
806 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
807 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
808 if (eth_addr_is_multicast(ea)) {
809 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
810 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
811 } else if (eth_addr_is_zero(ea)) {
812 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
818 /* Otherwise choose the minimum non-local MAC address among all of the
820 memset(ea, 0xff, sizeof ea);
821 for (i = 0; i < br->n_ports; i++) {
822 struct port *port = br->ports[i];
823 uint8_t iface_ea[ETH_ADDR_LEN];
826 /* Mirror output ports don't participate. */
827 if (port->is_mirror_output_port) {
831 /* Choose the MAC address to represent the port. */
832 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
833 /* Find the interface with this Ethernet address (if any) so that
834 * we can provide the correct devname to the caller. */
836 for (j = 0; j < port->n_ifaces; j++) {
837 struct iface *candidate = port->ifaces[j];
838 uint8_t candidate_ea[ETH_ADDR_LEN];
839 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
840 && eth_addr_equals(iface_ea, candidate_ea)) {
845 /* Choose the interface whose MAC address will represent the port.
846 * The Linux kernel bonding code always chooses the MAC address of
847 * the first slave added to a bond, and the Fedora networking
848 * scripts always add slaves to a bond in alphabetical order, so
849 * for compatibility we choose the interface with the name that is
850 * first in alphabetical order. */
851 iface = port->ifaces[0];
852 for (j = 1; j < port->n_ifaces; j++) {
853 struct iface *candidate = port->ifaces[j];
854 if (strcmp(candidate->name, iface->name) < 0) {
859 /* The local port doesn't count (since we're trying to choose its
860 * MAC address anyway). */
861 if (iface->dp_ifidx == ODPP_LOCAL) {
866 error = netdev_get_etheraddr(iface->netdev, iface_ea);
868 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
869 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
870 iface->name, strerror(error));
875 /* Compare against our current choice. */
876 if (!eth_addr_is_multicast(iface_ea) &&
877 !eth_addr_is_local(iface_ea) &&
878 !eth_addr_is_reserved(iface_ea) &&
879 !eth_addr_is_zero(iface_ea) &&
880 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
882 memcpy(ea, iface_ea, ETH_ADDR_LEN);
883 *hw_addr_iface = iface;
886 if (eth_addr_is_multicast(ea)) {
887 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
888 *hw_addr_iface = NULL;
889 VLOG_WARN("bridge %s: using default bridge Ethernet "
890 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
892 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
893 br->name, ETH_ADDR_ARGS(ea));
897 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
898 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
899 * an interface on 'br', then that interface must be passed in as
900 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
901 * 'hw_addr_iface' must be passed in as a null pointer. */
903 bridge_pick_datapath_id(struct bridge *br,
904 const uint8_t bridge_ea[ETH_ADDR_LEN],
905 struct iface *hw_addr_iface)
908 * The procedure for choosing a bridge MAC address will, in the most
909 * ordinary case, also choose a unique MAC that we can use as a datapath
910 * ID. In some special cases, though, multiple bridges will end up with
911 * the same MAC address. This is OK for the bridges, but it will confuse
912 * the OpenFlow controller, because each datapath needs a unique datapath
915 * Datapath IDs must be unique. It is also very desirable that they be
916 * stable from one run to the next, so that policy set on a datapath
919 const char *datapath_id;
922 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
923 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
929 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
931 * A bridge whose MAC address is taken from a VLAN network device
932 * (that is, a network device created with vconfig(8) or similar
933 * tool) will have the same MAC address as a bridge on the VLAN
934 * device's physical network device.
936 * Handle this case by hashing the physical network device MAC
937 * along with the VLAN identifier.
939 uint8_t buf[ETH_ADDR_LEN + 2];
940 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
941 buf[ETH_ADDR_LEN] = vlan >> 8;
942 buf[ETH_ADDR_LEN + 1] = vlan;
943 return dpid_from_hash(buf, sizeof buf);
946 * Assume that this bridge's MAC address is unique, since it
947 * doesn't fit any of the cases we handle specially.
952 * A purely internal bridge, that is, one that has no non-virtual
953 * network devices on it at all, is more difficult because it has no
954 * natural unique identifier at all.
956 * When the host is a XenServer, we handle this case by hashing the
957 * host's UUID with the name of the bridge. Names of bridges are
958 * persistent across XenServer reboots, although they can be reused if
959 * an internal network is destroyed and then a new one is later
960 * created, so this is fairly effective.
962 * When the host is not a XenServer, we punt by using a random MAC
963 * address on each run.
965 const char *host_uuid = xenserver_get_host_uuid();
967 char *combined = xasprintf("%s,%s", host_uuid, br->name);
968 dpid = dpid_from_hash(combined, strlen(combined));
974 return eth_addr_to_uint64(bridge_ea);
978 dpid_from_hash(const void *data, size_t n)
980 uint8_t hash[SHA1_DIGEST_SIZE];
982 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
983 sha1_bytes(data, n, hash);
984 eth_addr_mark_random(hash);
985 return eth_addr_to_uint64(hash);
991 struct bridge *br, *next;
995 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
996 int error = bridge_run_one(br);
998 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
999 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1000 "forcing reconfiguration", br->name);
1014 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1015 ofproto_wait(br->ofproto);
1016 if (br->controller) {
1020 mac_learning_wait(br->ml);
1025 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1026 * configuration changes. */
1028 bridge_flush(struct bridge *br)
1030 COVERAGE_INC(bridge_flush);
1032 mac_learning_flush(br->ml);
1035 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1036 * such interface. */
1037 static struct iface *
1038 bridge_get_local_iface(struct bridge *br)
1042 for (i = 0; i < br->n_ports; i++) {
1043 struct port *port = br->ports[i];
1044 for (j = 0; j < port->n_ifaces; j++) {
1045 struct iface *iface = port->ifaces[j];
1046 if (iface->dp_ifidx == ODPP_LOCAL) {
1055 /* Bridge unixctl user interface functions. */
1057 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1058 const char *args, void *aux UNUSED)
1060 struct ds ds = DS_EMPTY_INITIALIZER;
1061 const struct bridge *br;
1062 const struct mac_entry *e;
1064 br = bridge_lookup(args);
1066 unixctl_command_reply(conn, 501, "no such bridge");
1070 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1071 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1072 if (e->port < 0 || e->port >= br->n_ports) {
1075 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1076 br->ports[e->port]->ifaces[0]->dp_ifidx,
1077 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1079 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1083 /* Bridge reconfiguration functions. */
1084 static struct bridge *
1085 bridge_create(const struct ovsrec_bridge *br_cfg)
1090 assert(!bridge_lookup(br_cfg->name));
1091 br = xzalloc(sizeof *br);
1093 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1099 dpif_flow_flush(br->dpif);
1101 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1104 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1106 dpif_delete(br->dpif);
1107 dpif_close(br->dpif);
1112 br->name = xstrdup(br_cfg->name);
1114 br->ml = mac_learning_create();
1115 br->sent_config_request = false;
1116 eth_addr_random(br->default_ea);
1118 port_array_init(&br->ifaces);
1121 br->bond_next_rebalance = time_msec() + 10000;
1123 list_push_back(&all_bridges, &br->node);
1125 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1131 bridge_destroy(struct bridge *br)
1136 while (br->n_ports > 0) {
1137 port_destroy(br->ports[br->n_ports - 1]);
1139 list_remove(&br->node);
1140 error = dpif_delete(br->dpif);
1141 if (error && error != ENOENT) {
1142 VLOG_ERR("failed to delete %s: %s",
1143 dpif_name(br->dpif), strerror(error));
1145 dpif_close(br->dpif);
1146 ofproto_destroy(br->ofproto);
1147 free(br->controller);
1148 mac_learning_destroy(br->ml);
1149 port_array_destroy(&br->ifaces);
1156 static struct bridge *
1157 bridge_lookup(const char *name)
1161 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1162 if (!strcmp(br->name, name)) {
1170 bridge_exists(const char *name)
1172 return bridge_lookup(name) ? true : false;
1176 bridge_get_datapathid(const char *name)
1178 struct bridge *br = bridge_lookup(name);
1179 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1182 /* Handle requests for a listing of all flows known by the OpenFlow
1183 * stack, including those normally hidden. */
1185 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1186 const char *args, void *aux UNUSED)
1191 br = bridge_lookup(args);
1193 unixctl_command_reply(conn, 501, "Unknown bridge");
1198 ofproto_get_all_flows(br->ofproto, &results);
1200 unixctl_command_reply(conn, 200, ds_cstr(&results));
1201 ds_destroy(&results);
1205 bridge_run_one(struct bridge *br)
1209 error = ofproto_run1(br->ofproto);
1214 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1217 error = ofproto_run2(br->ofproto, br->flush);
1223 static const struct ovsrec_controller *
1224 bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1225 const struct bridge *br)
1227 const struct ovsrec_controller *controller;
1229 controller = (br->cfg->controller ? br->cfg->controller
1230 : ovs_cfg->controller ? ovs_cfg->controller
1233 if (controller && !strcmp(controller->target, "none")) {
1241 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1243 struct svec *ifaces = ifaces_;
1244 if (!svec_contains(ifaces, iface->name)) {
1245 svec_add(ifaces, iface->name);
1249 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1251 br->name, iface->name, iface->port->name);
1257 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1260 struct shash old_ports, new_ports;
1262 struct svec listeners, old_listeners;
1263 struct svec snoops, old_snoops;
1264 struct shash_node *node;
1268 /* Collect old ports. */
1269 shash_init(&old_ports);
1270 for (i = 0; i < br->n_ports; i++) {
1271 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1274 /* Collect new ports. */
1275 shash_init(&new_ports);
1276 for (i = 0; i < br->cfg->n_ports; i++) {
1277 const char *name = br->cfg->ports[i]->name;
1278 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1279 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1284 /* If we have a controller, then we need a local port. Complain if the
1285 * user didn't specify one.
1287 * XXX perhaps we should synthesize a port ourselves in this case. */
1288 if (bridge_get_controller(ovs_cfg, br)) {
1289 char local_name[IF_NAMESIZE];
1292 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1293 local_name, sizeof local_name);
1294 if (!error && !shash_find(&new_ports, local_name)) {
1295 VLOG_WARN("bridge %s: controller specified but no local port "
1296 "(port named %s) defined",
1297 br->name, local_name);
1301 dpid_from_string(ovs_cfg->management_id, &mgmt_id);
1302 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
1304 /* Get rid of deleted ports and add new ports. */
1305 SHASH_FOR_EACH (node, &old_ports) {
1306 if (!shash_find(&new_ports, node->name)) {
1307 port_destroy(node->data);
1310 SHASH_FOR_EACH (node, &new_ports) {
1311 struct port *port = shash_find_data(&old_ports, node->name);
1313 port = port_create(br, node->name);
1315 port_reconfigure(port, node->data);
1317 shash_destroy(&old_ports);
1318 shash_destroy(&new_ports);
1320 /* Check and delete duplicate interfaces. */
1322 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1323 svec_destroy(&ifaces);
1325 /* Delete all flows if we're switching from connected to standalone or vice
1326 * versa. (XXX Should we delete all flows if we are switching from one
1327 * controller to another?) */
1330 /* Configure OpenFlow management listeners. */
1331 svec_init(&listeners);
1332 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1334 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1335 ovs_rundir, br->name));
1336 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1337 svec_clear(&listeners);
1339 svec_sort_unique(&listeners);
1341 svec_init(&old_listeners);
1342 ofproto_get_listeners(br->ofproto, &old_listeners);
1343 svec_sort_unique(&old_listeners);
1345 if (!svec_equal(&listeners, &old_listeners)) {
1346 ofproto_set_listeners(br->ofproto, &listeners);
1348 svec_destroy(&listeners);
1349 svec_destroy(&old_listeners);
1351 /* Configure OpenFlow controller connection snooping. */
1353 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1355 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1356 ovs_rundir, br->name));
1357 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1358 svec_clear(&snoops);
1360 svec_sort_unique(&snoops);
1362 svec_init(&old_snoops);
1363 ofproto_get_snoops(br->ofproto, &old_snoops);
1364 svec_sort_unique(&old_snoops);
1366 if (!svec_equal(&snoops, &old_snoops)) {
1367 ofproto_set_snoops(br->ofproto, &snoops);
1369 svec_destroy(&snoops);
1370 svec_destroy(&old_snoops);
1372 /* Default listener. */
1373 svec_init(&listeners);
1374 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1375 ovs_rundir, br->name));
1376 svec_init(&old_listeners);
1377 ofproto_get_listeners(br->ofproto, &old_listeners);
1378 if (!svec_equal(&listeners, &old_listeners)) {
1379 ofproto_set_listeners(br->ofproto, &listeners);
1381 svec_destroy(&listeners);
1382 svec_destroy(&old_listeners);
1384 /* Default snoop. */
1386 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1387 ovs_rundir, br->name));
1388 svec_init(&old_snoops);
1389 ofproto_get_snoops(br->ofproto, &old_snoops);
1390 if (!svec_equal(&snoops, &old_snoops)) {
1391 ofproto_set_snoops(br->ofproto, &snoops);
1393 svec_destroy(&snoops);
1394 svec_destroy(&old_snoops);
1397 mirror_reconfigure(br);
1401 bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1404 char *pfx = xasprintf("bridge.%s.controller", br->name);
1405 const struct ovsrec_controller *c;
1407 c = bridge_get_controller(ovs_cfg, br);
1408 if ((br->controller != NULL) != (c != NULL)) {
1409 ofproto_flush_flows(br->ofproto);
1411 free(br->controller);
1412 br->controller = c ? xstrdup(c->target) : NULL;
1415 int max_backoff, probe;
1416 int rate_limit, burst_limit;
1418 if (!strcmp(c->target, "discover")) {
1419 ofproto_set_discovery(br->ofproto, true,
1420 c->discover_accept_regex,
1421 c->discover_update_resolv_conf);
1423 struct iface *local_iface;
1427 in_band = (!c->connection_mode
1428 || !strcmp(c->connection_mode, "out-of-band"));
1429 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1430 ofproto_set_in_band(br->ofproto, in_band);
1432 local_iface = bridge_get_local_iface(br);
1433 if (local_iface && c->local_ip && inet_aton(c->local_ip, &ip)) {
1434 struct netdev *netdev = local_iface->netdev;
1435 struct in_addr ip, mask, gateway;
1437 if (!c->local_netmask || !inet_aton(c->local_netmask, &mask)) {
1440 if (!c->local_gateway
1441 || !inet_aton(c->local_gateway, &gateway)) {
1445 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1447 mask.s_addr = guess_netmask(ip.s_addr);
1449 if (!netdev_set_in4(netdev, ip, mask)) {
1450 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1452 br->name, IP_ARGS(&ip.s_addr),
1453 IP_ARGS(&mask.s_addr));
1456 if (gateway.s_addr) {
1457 if (!netdev_add_router(netdev, gateway)) {
1458 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1459 br->name, IP_ARGS(&gateway.s_addr));
1465 ofproto_set_failure(br->ofproto,
1467 || !strcmp(c->fail_mode, "standalone")
1468 || !strcmp(c->fail_mode, "open")));
1470 probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1471 ofproto_set_probe_interval(br->ofproto, probe);
1473 max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1474 ofproto_set_max_backoff(br->ofproto, max_backoff);
1476 rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1477 burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0;
1478 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1480 union ofp_action action;
1483 /* Set up a flow that matches every packet and directs them to
1484 * OFPP_NORMAL (which goes to us). */
1485 memset(&action, 0, sizeof action);
1486 action.type = htons(OFPAT_OUTPUT);
1487 action.output.len = htons(sizeof action);
1488 action.output.port = htons(OFPP_NORMAL);
1489 memset(&flow, 0, sizeof flow);
1490 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1493 ofproto_set_in_band(br->ofproto, false);
1494 ofproto_set_max_backoff(br->ofproto, 1);
1495 ofproto_set_probe_interval(br->ofproto, 5);
1496 ofproto_set_failure(br->ofproto, false);
1500 ofproto_set_controller(br->ofproto, br->controller);
1504 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1509 for (i = 0; i < br->n_ports; i++) {
1510 struct port *port = br->ports[i];
1511 for (j = 0; j < port->n_ifaces; j++) {
1512 struct iface *iface = port->ifaces[j];
1513 shash_add_once(ifaces, iface->name, iface);
1515 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1516 shash_add_once(ifaces, port->name, NULL);
1521 /* For robustness, in case the administrator moves around datapath ports behind
1522 * our back, we re-check all the datapath port numbers here.
1524 * This function will set the 'dp_ifidx' members of interfaces that have
1525 * disappeared to -1, so only call this function from a context where those
1526 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1527 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1528 * datapath, which doesn't support UINT16_MAX+1 ports. */
1530 bridge_fetch_dp_ifaces(struct bridge *br)
1532 struct odp_port *dpif_ports;
1533 size_t n_dpif_ports;
1536 /* Reset all interface numbers. */
1537 for (i = 0; i < br->n_ports; i++) {
1538 struct port *port = br->ports[i];
1539 for (j = 0; j < port->n_ifaces; j++) {
1540 struct iface *iface = port->ifaces[j];
1541 iface->dp_ifidx = -1;
1544 port_array_clear(&br->ifaces);
1546 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1547 for (i = 0; i < n_dpif_ports; i++) {
1548 struct odp_port *p = &dpif_ports[i];
1549 struct iface *iface = iface_lookup(br, p->devname);
1551 if (iface->dp_ifidx >= 0) {
1552 VLOG_WARN("%s reported interface %s twice",
1553 dpif_name(br->dpif), p->devname);
1554 } else if (iface_from_dp_ifidx(br, p->port)) {
1555 VLOG_WARN("%s reported interface %"PRIu16" twice",
1556 dpif_name(br->dpif), p->port);
1558 port_array_set(&br->ifaces, p->port, iface);
1559 iface->dp_ifidx = p->port;
1563 int64_t ofport = (iface->dp_ifidx >= 0
1564 ? odp_port_to_ofp_port(iface->dp_ifidx)
1566 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1573 /* Bridge packet processing functions. */
1576 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1578 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1581 static struct bond_entry *
1582 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1584 return &port->bond_hash[bond_hash(mac)];
1588 bond_choose_iface(const struct port *port)
1590 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1591 size_t i, best_down_slave = -1;
1592 long long next_delay_expiration = LLONG_MAX;
1594 for (i = 0; i < port->n_ifaces; i++) {
1595 struct iface *iface = port->ifaces[i];
1597 if (iface->enabled) {
1599 } else if (iface->delay_expires < next_delay_expiration) {
1600 best_down_slave = i;
1601 next_delay_expiration = iface->delay_expires;
1605 if (best_down_slave != -1) {
1606 struct iface *iface = port->ifaces[best_down_slave];
1608 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1609 "since no other interface is up", iface->name,
1610 iface->delay_expires - time_msec());
1611 bond_enable_slave(iface, true);
1614 return best_down_slave;
1618 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1619 uint16_t *dp_ifidx, tag_type *tags)
1621 struct iface *iface;
1623 assert(port->n_ifaces);
1624 if (port->n_ifaces == 1) {
1625 iface = port->ifaces[0];
1627 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1628 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1629 || !port->ifaces[e->iface_idx]->enabled) {
1630 /* XXX select interface properly. The current interface selection
1631 * is only good for testing the rebalancing code. */
1632 e->iface_idx = bond_choose_iface(port);
1633 if (e->iface_idx < 0) {
1634 *tags |= port->no_ifaces_tag;
1637 e->iface_tag = tag_create_random();
1638 ((struct port *) port)->bond_compat_is_stale = true;
1640 *tags |= e->iface_tag;
1641 iface = port->ifaces[e->iface_idx];
1643 *dp_ifidx = iface->dp_ifidx;
1644 *tags |= iface->tag; /* Currently only used for bonding. */
1649 bond_link_status_update(struct iface *iface, bool carrier)
1651 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1652 struct port *port = iface->port;
1654 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1655 /* Nothing to do. */
1658 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1659 iface->name, carrier ? "detected" : "dropped");
1660 if (carrier == iface->enabled) {
1661 iface->delay_expires = LLONG_MAX;
1662 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1663 iface->name, carrier ? "disabled" : "enabled");
1664 } else if (carrier && port->active_iface < 0) {
1665 bond_enable_slave(iface, true);
1666 if (port->updelay) {
1667 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1668 "other interface is up", iface->name, port->updelay);
1671 int delay = carrier ? port->updelay : port->downdelay;
1672 iface->delay_expires = time_msec() + delay;
1675 "interface %s: will be %s if it stays %s for %d ms",
1677 carrier ? "enabled" : "disabled",
1678 carrier ? "up" : "down",
1685 bond_choose_active_iface(struct port *port)
1687 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1689 port->active_iface = bond_choose_iface(port);
1690 port->active_iface_tag = tag_create_random();
1691 if (port->active_iface >= 0) {
1692 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1693 port->name, port->ifaces[port->active_iface]->name);
1695 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1701 bond_enable_slave(struct iface *iface, bool enable)
1703 struct port *port = iface->port;
1704 struct bridge *br = port->bridge;
1706 /* This acts as a recursion check. If the act of disabling a slave
1707 * causes a different slave to be enabled, the flag will allow us to
1708 * skip redundant work when we reenter this function. It must be
1709 * cleared on exit to keep things safe with multiple bonds. */
1710 static bool moving_active_iface = false;
1712 iface->delay_expires = LLONG_MAX;
1713 if (enable == iface->enabled) {
1717 iface->enabled = enable;
1718 if (!iface->enabled) {
1719 VLOG_WARN("interface %s: disabled", iface->name);
1720 ofproto_revalidate(br->ofproto, iface->tag);
1721 if (iface->port_ifidx == port->active_iface) {
1722 ofproto_revalidate(br->ofproto,
1723 port->active_iface_tag);
1725 /* Disabling a slave can lead to another slave being immediately
1726 * enabled if there will be no active slaves but one is waiting
1727 * on an updelay. In this case we do not need to run most of the
1728 * code for the newly enabled slave since there was no period
1729 * without an active slave and it is redundant with the disabling
1731 moving_active_iface = true;
1732 bond_choose_active_iface(port);
1734 bond_send_learning_packets(port);
1736 VLOG_WARN("interface %s: enabled", iface->name);
1737 if (port->active_iface < 0 && !moving_active_iface) {
1738 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1739 bond_choose_active_iface(port);
1740 bond_send_learning_packets(port);
1742 iface->tag = tag_create_random();
1745 moving_active_iface = false;
1746 port->bond_compat_is_stale = true;
1750 bond_run(struct bridge *br)
1754 for (i = 0; i < br->n_ports; i++) {
1755 struct port *port = br->ports[i];
1757 if (port->n_ifaces >= 2) {
1758 for (j = 0; j < port->n_ifaces; j++) {
1759 struct iface *iface = port->ifaces[j];
1760 if (time_msec() >= iface->delay_expires) {
1761 bond_enable_slave(iface, !iface->enabled);
1766 if (port->bond_compat_is_stale) {
1767 port->bond_compat_is_stale = false;
1768 port_update_bond_compat(port);
1774 bond_wait(struct bridge *br)
1778 for (i = 0; i < br->n_ports; i++) {
1779 struct port *port = br->ports[i];
1780 if (port->n_ifaces < 2) {
1783 for (j = 0; j < port->n_ifaces; j++) {
1784 struct iface *iface = port->ifaces[j];
1785 if (iface->delay_expires != LLONG_MAX) {
1786 poll_timer_wait(iface->delay_expires - time_msec());
1793 set_dst(struct dst *p, const flow_t *flow,
1794 const struct port *in_port, const struct port *out_port,
1797 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1798 : in_port->vlan >= 0 ? in_port->vlan
1799 : ntohs(flow->dl_vlan));
1800 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1804 swap_dst(struct dst *p, struct dst *q)
1806 struct dst tmp = *p;
1811 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1812 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1813 * that we push to the datapath. We could in fact fully sort the array by
1814 * vlan, but in most cases there are at most two different vlan tags so that's
1815 * possibly overkill.) */
1817 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1819 struct dst *first = dsts;
1820 struct dst *last = dsts + n_dsts;
1822 while (first != last) {
1824 * - All dsts < first have vlan == 'vlan'.
1825 * - All dsts >= last have vlan != 'vlan'.
1826 * - first < last. */
1827 while (first->vlan == vlan) {
1828 if (++first == last) {
1833 /* Same invariants, plus one additional:
1834 * - first->vlan != vlan.
1836 while (last[-1].vlan != vlan) {
1837 if (--last == first) {
1842 /* Same invariants, plus one additional:
1843 * - last[-1].vlan == vlan.*/
1844 swap_dst(first++, --last);
1849 mirror_mask_ffs(mirror_mask_t mask)
1851 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1856 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1857 const struct dst *test)
1860 for (i = 0; i < n_dsts; i++) {
1861 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1869 port_trunks_vlan(const struct port *port, uint16_t vlan)
1871 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1875 port_includes_vlan(const struct port *port, uint16_t vlan)
1877 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1881 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1882 const struct port *in_port, const struct port *out_port,
1883 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
1885 mirror_mask_t mirrors = in_port->src_mirrors;
1886 struct dst *dst = dsts;
1889 if (out_port == FLOOD_PORT) {
1890 /* XXX use ODP_FLOOD if no vlans or bonding. */
1891 /* XXX even better, define each VLAN as a datapath port group */
1892 for (i = 0; i < br->n_ports; i++) {
1893 struct port *port = br->ports[i];
1894 if (port != in_port && port_includes_vlan(port, vlan)
1895 && !port->is_mirror_output_port
1896 && set_dst(dst, flow, in_port, port, tags)) {
1897 mirrors |= port->dst_mirrors;
1901 *nf_output_iface = NF_OUT_FLOOD;
1902 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1903 *nf_output_iface = dst->dp_ifidx;
1904 mirrors |= out_port->dst_mirrors;
1909 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1910 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1912 if (set_dst(dst, flow, in_port, m->out_port, tags)
1913 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1917 for (i = 0; i < br->n_ports; i++) {
1918 struct port *port = br->ports[i];
1919 if (port_includes_vlan(port, m->out_vlan)
1920 && set_dst(dst, flow, in_port, port, tags))
1924 if (port->vlan < 0) {
1925 dst->vlan = m->out_vlan;
1927 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
1931 /* Use the vlan tag on the original flow instead of
1932 * the one passed in the vlan parameter. This ensures
1933 * that we compare the vlan from before any implicit
1934 * tagging tags place. This is necessary because
1935 * dst->vlan is the final vlan, after removing implicit
1937 flow_vlan = ntohs(flow->dl_vlan);
1938 if (flow_vlan == 0) {
1939 flow_vlan = OFP_VLAN_NONE;
1941 if (port == in_port && dst->vlan == flow_vlan) {
1942 /* Don't send out input port on same VLAN. */
1950 mirrors &= mirrors - 1;
1953 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1958 print_dsts(const struct dst *dsts, size_t n)
1960 for (; n--; dsts++) {
1961 printf(">p%"PRIu16, dsts->dp_ifidx);
1962 if (dsts->vlan != OFP_VLAN_NONE) {
1963 printf("v%"PRIu16, dsts->vlan);
1969 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1970 const struct port *in_port, const struct port *out_port,
1971 tag_type *tags, struct odp_actions *actions,
1972 uint16_t *nf_output_iface)
1974 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1976 const struct dst *p;
1979 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
1982 cur_vlan = ntohs(flow->dl_vlan);
1983 for (p = dsts; p < &dsts[n_dsts]; p++) {
1984 union odp_action *a;
1985 if (p->vlan != cur_vlan) {
1986 if (p->vlan == OFP_VLAN_NONE) {
1987 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1989 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1990 a->vlan_vid.vlan_vid = htons(p->vlan);
1994 a = odp_actions_add(actions, ODPAT_OUTPUT);
1995 a->output.port = p->dp_ifidx;
1999 /* Returns the effective vlan of a packet, taking into account both the
2000 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2001 * the packet is untagged and -1 indicates it has an invalid header and
2002 * should be dropped. */
2003 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2004 struct port *in_port, bool have_packet)
2006 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2007 * belongs to VLAN 0, so we should treat both cases identically. (In the
2008 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2009 * presumably to allow a priority to be specified. In the latter case, the
2010 * packet does not have any 802.1Q header.) */
2011 int vlan = ntohs(flow->dl_vlan);
2012 if (vlan == OFP_VLAN_NONE) {
2015 if (in_port->vlan >= 0) {
2017 /* XXX support double tagging? */
2019 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2020 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2021 "packet received on port %s configured with "
2022 "implicit VLAN %"PRIu16,
2023 br->name, ntohs(flow->dl_vlan),
2024 in_port->name, in_port->vlan);
2028 vlan = in_port->vlan;
2030 if (!port_includes_vlan(in_port, vlan)) {
2032 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2033 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2034 "packet received on port %s not configured for "
2036 br->name, vlan, in_port->name, vlan);
2046 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2047 struct port *in_port)
2049 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
2050 vlan, in_port->port_idx);
2052 /* The log messages here could actually be useful in debugging,
2053 * so keep the rate limit relatively high. */
2054 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2056 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2057 "on port %s in VLAN %d",
2058 br->name, ETH_ADDR_ARGS(flow->dl_src),
2059 in_port->name, vlan);
2060 ofproto_revalidate(br->ofproto, rev_tag);
2065 is_bcast_arp_reply(const flow_t *flow)
2067 return (flow->dl_type == htons(ETH_TYPE_ARP)
2068 && flow->nw_proto == ARP_OP_REPLY
2069 && eth_addr_is_broadcast(flow->dl_dst));
2072 /* If the composed actions may be applied to any packet in the given 'flow',
2073 * returns true. Otherwise, the actions should only be applied to 'packet', or
2074 * not at all, if 'packet' was NULL. */
2076 process_flow(struct bridge *br, const flow_t *flow,
2077 const struct ofpbuf *packet, struct odp_actions *actions,
2078 tag_type *tags, uint16_t *nf_output_iface)
2080 struct iface *in_iface;
2081 struct port *in_port;
2082 struct port *out_port = NULL; /* By default, drop the packet/flow. */
2086 /* Find the interface and port structure for the received packet. */
2087 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2089 /* No interface? Something fishy... */
2090 if (packet != NULL) {
2091 /* Odd. A few possible reasons here:
2093 * - We deleted an interface but there are still a few packets
2094 * queued up from it.
2096 * - Someone externally added an interface (e.g. with "ovs-dpctl
2097 * add-if") that we don't know about.
2099 * - Packet arrived on the local port but the local port is not
2100 * one of our bridge ports.
2102 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2104 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2105 "interface %"PRIu16, br->name, flow->in_port);
2108 /* Return without adding any actions, to drop packets on this flow. */
2111 in_port = in_iface->port;
2112 vlan = flow_get_vlan(br, flow, in_port, !!packet);
2117 /* Drop frames for reserved multicast addresses. */
2118 if (eth_addr_is_reserved(flow->dl_dst)) {
2122 /* Drop frames on ports reserved for mirroring. */
2123 if (in_port->is_mirror_output_port) {
2124 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2125 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
2126 "which is reserved exclusively for mirroring",
2127 br->name, in_port->name);
2131 /* Packets received on bonds need special attention to avoid duplicates. */
2132 if (in_port->n_ifaces > 1) {
2135 if (eth_addr_is_multicast(flow->dl_dst)) {
2136 *tags |= in_port->active_iface_tag;
2137 if (in_port->active_iface != in_iface->port_ifidx) {
2138 /* Drop all multicast packets on inactive slaves. */
2143 /* Drop all packets for which we have learned a different input
2144 * port, because we probably sent the packet on one slave and got
2145 * it back on the other. Broadcast ARP replies are an exception
2146 * to this rule: the host has moved to another switch. */
2147 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
2148 if (src_idx != -1 && src_idx != in_port->port_idx &&
2149 !is_bcast_arp_reply(flow)) {
2155 out_port = FLOOD_PORT;
2156 /* Learn source MAC (but don't try to learn from revalidation). */
2158 update_learning_table(br, flow, vlan, in_port);
2161 /* Determine output port. */
2162 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
2164 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2165 out_port = br->ports[out_port_idx];
2166 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2167 /* If we are revalidating but don't have a learning entry then
2168 * eject the flow. Installing a flow that floods packets opens
2169 * up a window of time where we could learn from a packet reflected
2170 * on a bond and blackhole packets before the learning table is
2171 * updated to reflect the correct port. */
2175 /* Don't send packets out their input ports. */
2176 if (in_port == out_port) {
2181 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2187 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2190 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2191 const struct ofp_phy_port *opp,
2194 struct bridge *br = br_;
2195 struct iface *iface;
2198 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2204 if (reason == OFPPR_DELETE) {
2205 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2206 br->name, iface->name);
2207 iface_destroy(iface);
2208 if (!port->n_ifaces) {
2209 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2210 br->name, port->name);
2216 if (port->n_ifaces > 1) {
2217 bool up = !(opp->state & OFPPS_LINK_DOWN);
2218 bond_link_status_update(iface, up);
2219 port_update_bond_compat(port);
2225 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2226 struct odp_actions *actions, tag_type *tags,
2227 uint16_t *nf_output_iface, void *br_)
2229 struct bridge *br = br_;
2231 COVERAGE_INC(bridge_process_flow);
2232 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2236 bridge_account_flow_ofhook_cb(const flow_t *flow,
2237 const union odp_action *actions,
2238 size_t n_actions, unsigned long long int n_bytes,
2241 struct bridge *br = br_;
2242 struct port *in_port;
2243 const union odp_action *a;
2245 /* Feed information from the active flows back into the learning table
2246 * to ensure that table is always in sync with what is actually flowing
2247 * through the datapath. */
2248 in_port = port_from_dp_ifidx(br, flow->in_port);
2250 int vlan = flow_get_vlan(br, flow, in_port, false);
2252 update_learning_table(br, flow, vlan, in_port);
2256 if (!br->has_bonded_ports) {
2260 for (a = actions; a < &actions[n_actions]; a++) {
2261 if (a->type == ODPAT_OUTPUT) {
2262 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2263 if (out_port && out_port->n_ifaces >= 2) {
2264 struct bond_entry *e = lookup_bond_entry(out_port,
2266 e->tx_bytes += n_bytes;
2273 bridge_account_checkpoint_ofhook_cb(void *br_)
2275 struct bridge *br = br_;
2278 if (!br->has_bonded_ports) {
2282 /* The current ofproto implementation calls this callback at least once a
2283 * second, so this timer implementation is sufficient. */
2284 if (time_msec() < br->bond_next_rebalance) {
2287 br->bond_next_rebalance = time_msec() + 10000;
2289 for (i = 0; i < br->n_ports; i++) {
2290 struct port *port = br->ports[i];
2291 if (port->n_ifaces > 1) {
2292 bond_rebalance_port(port);
2297 static struct ofhooks bridge_ofhooks = {
2298 bridge_port_changed_ofhook_cb,
2299 bridge_normal_ofhook_cb,
2300 bridge_account_flow_ofhook_cb,
2301 bridge_account_checkpoint_ofhook_cb,
2304 /* Bonding functions. */
2306 /* Statistics for a single interface on a bonded port, used for load-based
2307 * bond rebalancing. */
2308 struct slave_balance {
2309 struct iface *iface; /* The interface. */
2310 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2312 /* All the "bond_entry"s that are assigned to this interface, in order of
2313 * increasing tx_bytes. */
2314 struct bond_entry **hashes;
2318 /* Sorts pointers to pointers to bond_entries in ascending order by the
2319 * interface to which they are assigned, and within a single interface in
2320 * ascending order of bytes transmitted. */
2322 compare_bond_entries(const void *a_, const void *b_)
2324 const struct bond_entry *const *ap = a_;
2325 const struct bond_entry *const *bp = b_;
2326 const struct bond_entry *a = *ap;
2327 const struct bond_entry *b = *bp;
2328 if (a->iface_idx != b->iface_idx) {
2329 return a->iface_idx > b->iface_idx ? 1 : -1;
2330 } else if (a->tx_bytes != b->tx_bytes) {
2331 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2337 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2338 * *descending* order by number of bytes transmitted. */
2340 compare_slave_balance(const void *a_, const void *b_)
2342 const struct slave_balance *a = a_;
2343 const struct slave_balance *b = b_;
2344 if (a->iface->enabled != b->iface->enabled) {
2345 return a->iface->enabled ? -1 : 1;
2346 } else if (a->tx_bytes != b->tx_bytes) {
2347 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2354 swap_bals(struct slave_balance *a, struct slave_balance *b)
2356 struct slave_balance tmp = *a;
2361 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2362 * given that 'p' (and only 'p') might be in the wrong location.
2364 * This function invalidates 'p', since it might now be in a different memory
2367 resort_bals(struct slave_balance *p,
2368 struct slave_balance bals[], size_t n_bals)
2371 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2372 swap_bals(p, p - 1);
2374 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2375 swap_bals(p, p + 1);
2381 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2383 if (VLOG_IS_DBG_ENABLED()) {
2384 struct ds ds = DS_EMPTY_INITIALIZER;
2385 const struct slave_balance *b;
2387 for (b = bals; b < bals + n_bals; b++) {
2391 ds_put_char(&ds, ',');
2393 ds_put_format(&ds, " %s %"PRIu64"kB",
2394 b->iface->name, b->tx_bytes / 1024);
2396 if (!b->iface->enabled) {
2397 ds_put_cstr(&ds, " (disabled)");
2399 if (b->n_hashes > 0) {
2400 ds_put_cstr(&ds, " (");
2401 for (i = 0; i < b->n_hashes; i++) {
2402 const struct bond_entry *e = b->hashes[i];
2404 ds_put_cstr(&ds, " + ");
2406 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2407 e - port->bond_hash, e->tx_bytes / 1024);
2409 ds_put_cstr(&ds, ")");
2412 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2417 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2419 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2422 struct bond_entry *hash = from->hashes[hash_idx];
2423 struct port *port = from->iface->port;
2424 uint64_t delta = hash->tx_bytes;
2426 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2427 "from %s to %s (now carrying %"PRIu64"kB and "
2428 "%"PRIu64"kB load, respectively)",
2429 port->name, delta / 1024, hash - port->bond_hash,
2430 from->iface->name, to->iface->name,
2431 (from->tx_bytes - delta) / 1024,
2432 (to->tx_bytes + delta) / 1024);
2434 /* Delete element from from->hashes.
2436 * We don't bother to add the element to to->hashes because not only would
2437 * it require more work, the only purpose it would be to allow that hash to
2438 * be migrated to another slave in this rebalancing run, and there is no
2439 * point in doing that. */
2440 if (hash_idx == 0) {
2443 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2444 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2448 /* Shift load away from 'from' to 'to'. */
2449 from->tx_bytes -= delta;
2450 to->tx_bytes += delta;
2452 /* Arrange for flows to be revalidated. */
2453 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2454 hash->iface_idx = to->iface->port_ifidx;
2455 hash->iface_tag = tag_create_random();
2459 bond_rebalance_port(struct port *port)
2461 struct slave_balance bals[DP_MAX_PORTS];
2463 struct bond_entry *hashes[BOND_MASK + 1];
2464 struct slave_balance *b, *from, *to;
2465 struct bond_entry *e;
2468 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2469 * descending order of tx_bytes, so that bals[0] represents the most
2470 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2473 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2474 * array for each slave_balance structure, we sort our local array of
2475 * hashes in order by slave, so that all of the hashes for a given slave
2476 * become contiguous in memory, and then we point each 'hashes' members of
2477 * a slave_balance structure to the start of a contiguous group. */
2478 n_bals = port->n_ifaces;
2479 for (b = bals; b < &bals[n_bals]; b++) {
2480 b->iface = port->ifaces[b - bals];
2485 for (i = 0; i <= BOND_MASK; i++) {
2486 hashes[i] = &port->bond_hash[i];
2488 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2489 for (i = 0; i <= BOND_MASK; i++) {
2491 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2492 b = &bals[e->iface_idx];
2493 b->tx_bytes += e->tx_bytes;
2495 b->hashes = &hashes[i];
2500 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2501 log_bals(bals, n_bals, port);
2503 /* Discard slaves that aren't enabled (which were sorted to the back of the
2504 * array earlier). */
2505 while (!bals[n_bals - 1].iface->enabled) {
2512 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2513 to = &bals[n_bals - 1];
2514 for (from = bals; from < to; ) {
2515 uint64_t overload = from->tx_bytes - to->tx_bytes;
2516 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2517 /* The extra load on 'from' (and all less-loaded slaves), compared
2518 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2519 * it is less than ~1Mbps. No point in rebalancing. */
2521 } else if (from->n_hashes == 1) {
2522 /* 'from' only carries a single MAC hash, so we can't shift any
2523 * load away from it, even though we want to. */
2526 /* 'from' is carrying significantly more load than 'to', and that
2527 * load is split across at least two different hashes. Pick a hash
2528 * to migrate to 'to' (the least-loaded slave), given that doing so
2529 * must decrease the ratio of the load on the two slaves by at
2532 * The sort order we use means that we prefer to shift away the
2533 * smallest hashes instead of the biggest ones. There is little
2534 * reason behind this decision; we could use the opposite sort
2535 * order to shift away big hashes ahead of small ones. */
2539 for (i = 0; i < from->n_hashes; i++) {
2540 double old_ratio, new_ratio;
2541 uint64_t delta = from->hashes[i]->tx_bytes;
2543 if (delta == 0 || from->tx_bytes - delta == 0) {
2544 /* Pointless move. */
2548 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2550 if (to->tx_bytes == 0) {
2551 /* Nothing on the new slave, move it. */
2555 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2556 new_ratio = (double)(from->tx_bytes - delta) /
2557 (to->tx_bytes + delta);
2559 if (new_ratio == 0) {
2560 /* Should already be covered but check to prevent division
2565 if (new_ratio < 1) {
2566 new_ratio = 1 / new_ratio;
2569 if (old_ratio - new_ratio > 0.1) {
2570 /* Would decrease the ratio, move it. */
2574 if (i < from->n_hashes) {
2575 bond_shift_load(from, to, i);
2576 port->bond_compat_is_stale = true;
2578 /* If the result of the migration changed the relative order of
2579 * 'from' and 'to' swap them back to maintain invariants. */
2580 if (order_swapped) {
2581 swap_bals(from, to);
2584 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2585 * point to different slave_balance structures. It is only
2586 * valid to do these two operations in a row at all because we
2587 * know that 'from' will not move past 'to' and vice versa. */
2588 resort_bals(from, bals, n_bals);
2589 resort_bals(to, bals, n_bals);
2596 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2597 * historical data to decay to <1% in 7 rebalancing runs. */
2598 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2604 bond_send_learning_packets(struct port *port)
2606 struct bridge *br = port->bridge;
2607 struct mac_entry *e;
2608 struct ofpbuf packet;
2609 int error, n_packets, n_errors;
2611 if (!port->n_ifaces || port->active_iface < 0) {
2615 ofpbuf_init(&packet, 128);
2616 error = n_packets = n_errors = 0;
2617 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2618 union ofp_action actions[2], *a;
2624 if (e->port == port->port_idx
2625 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2629 /* Compose actions. */
2630 memset(actions, 0, sizeof actions);
2633 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2634 a->vlan_vid.len = htons(sizeof *a);
2635 a->vlan_vid.vlan_vid = htons(e->vlan);
2638 a->output.type = htons(OFPAT_OUTPUT);
2639 a->output.len = htons(sizeof *a);
2640 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2645 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2647 flow_extract(&packet, ODPP_NONE, &flow);
2648 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2655 ofpbuf_uninit(&packet);
2658 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2659 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2660 "packets, last error was: %s",
2661 port->name, n_errors, n_packets, strerror(error));
2663 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2664 port->name, n_packets);
2668 /* Bonding unixctl user interface functions. */
2671 bond_unixctl_list(struct unixctl_conn *conn,
2672 const char *args UNUSED, void *aux UNUSED)
2674 struct ds ds = DS_EMPTY_INITIALIZER;
2675 const struct bridge *br;
2677 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2679 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2682 for (i = 0; i < br->n_ports; i++) {
2683 const struct port *port = br->ports[i];
2684 if (port->n_ifaces > 1) {
2687 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2688 for (j = 0; j < port->n_ifaces; j++) {
2689 const struct iface *iface = port->ifaces[j];
2691 ds_put_cstr(&ds, ", ");
2693 ds_put_cstr(&ds, iface->name);
2695 ds_put_char(&ds, '\n');
2699 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2703 static struct port *
2704 bond_find(const char *name)
2706 const struct bridge *br;
2708 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2711 for (i = 0; i < br->n_ports; i++) {
2712 struct port *port = br->ports[i];
2713 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2722 bond_unixctl_show(struct unixctl_conn *conn,
2723 const char *args, void *aux UNUSED)
2725 struct ds ds = DS_EMPTY_INITIALIZER;
2726 const struct port *port;
2729 port = bond_find(args);
2731 unixctl_command_reply(conn, 501, "no such bond");
2735 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2736 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2737 ds_put_format(&ds, "next rebalance: %lld ms\n",
2738 port->bridge->bond_next_rebalance - time_msec());
2739 for (j = 0; j < port->n_ifaces; j++) {
2740 const struct iface *iface = port->ifaces[j];
2741 struct bond_entry *be;
2744 ds_put_format(&ds, "slave %s: %s\n",
2745 iface->name, iface->enabled ? "enabled" : "disabled");
2746 if (j == port->active_iface) {
2747 ds_put_cstr(&ds, "\tactive slave\n");
2749 if (iface->delay_expires != LLONG_MAX) {
2750 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2751 iface->enabled ? "downdelay" : "updelay",
2752 iface->delay_expires - time_msec());
2756 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2757 int hash = be - port->bond_hash;
2758 struct mac_entry *me;
2760 if (be->iface_idx != j) {
2764 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2765 hash, be->tx_bytes / 1024);
2768 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2769 &port->bridge->ml->lrus) {
2772 if (bond_hash(me->mac) == hash
2773 && me->port != port->port_idx
2774 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2775 && dp_ifidx == iface->dp_ifidx)
2777 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2778 ETH_ADDR_ARGS(me->mac));
2783 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2788 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2791 char *args = (char *) args_;
2792 char *save_ptr = NULL;
2793 char *bond_s, *hash_s, *slave_s;
2794 uint8_t mac[ETH_ADDR_LEN];
2796 struct iface *iface;
2797 struct bond_entry *entry;
2800 bond_s = strtok_r(args, " ", &save_ptr);
2801 hash_s = strtok_r(NULL, " ", &save_ptr);
2802 slave_s = strtok_r(NULL, " ", &save_ptr);
2804 unixctl_command_reply(conn, 501,
2805 "usage: bond/migrate BOND HASH SLAVE");
2809 port = bond_find(bond_s);
2811 unixctl_command_reply(conn, 501, "no such bond");
2815 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2816 == ETH_ADDR_SCAN_COUNT) {
2817 hash = bond_hash(mac);
2818 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2819 hash = atoi(hash_s) & BOND_MASK;
2821 unixctl_command_reply(conn, 501, "bad hash");
2825 iface = port_lookup_iface(port, slave_s);
2827 unixctl_command_reply(conn, 501, "no such slave");
2831 if (!iface->enabled) {
2832 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2836 entry = &port->bond_hash[hash];
2837 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2838 entry->iface_idx = iface->port_ifidx;
2839 entry->iface_tag = tag_create_random();
2840 port->bond_compat_is_stale = true;
2841 unixctl_command_reply(conn, 200, "migrated");
2845 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
2848 char *args = (char *) args_;
2849 char *save_ptr = NULL;
2850 char *bond_s, *slave_s;
2852 struct iface *iface;
2854 bond_s = strtok_r(args, " ", &save_ptr);
2855 slave_s = strtok_r(NULL, " ", &save_ptr);
2857 unixctl_command_reply(conn, 501,
2858 "usage: bond/set-active-slave BOND SLAVE");
2862 port = bond_find(bond_s);
2864 unixctl_command_reply(conn, 501, "no such bond");
2868 iface = port_lookup_iface(port, slave_s);
2870 unixctl_command_reply(conn, 501, "no such slave");
2874 if (!iface->enabled) {
2875 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2879 if (port->active_iface != iface->port_ifidx) {
2880 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2881 port->active_iface = iface->port_ifidx;
2882 port->active_iface_tag = tag_create_random();
2883 VLOG_INFO("port %s: active interface is now %s",
2884 port->name, iface->name);
2885 bond_send_learning_packets(port);
2886 unixctl_command_reply(conn, 200, "done");
2888 unixctl_command_reply(conn, 200, "no change");
2893 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2895 char *args = (char *) args_;
2896 char *save_ptr = NULL;
2897 char *bond_s, *slave_s;
2899 struct iface *iface;
2901 bond_s = strtok_r(args, " ", &save_ptr);
2902 slave_s = strtok_r(NULL, " ", &save_ptr);
2904 unixctl_command_reply(conn, 501,
2905 "usage: bond/enable/disable-slave BOND SLAVE");
2909 port = bond_find(bond_s);
2911 unixctl_command_reply(conn, 501, "no such bond");
2915 iface = port_lookup_iface(port, slave_s);
2917 unixctl_command_reply(conn, 501, "no such slave");
2921 bond_enable_slave(iface, enable);
2922 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2926 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
2929 enable_slave(conn, args, true);
2933 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
2936 enable_slave(conn, args, false);
2940 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
2943 uint8_t mac[ETH_ADDR_LEN];
2947 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2948 == ETH_ADDR_SCAN_COUNT) {
2949 hash = bond_hash(mac);
2951 hash_cstr = xasprintf("%u", hash);
2952 unixctl_command_reply(conn, 200, hash_cstr);
2955 unixctl_command_reply(conn, 501, "invalid mac");
2962 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
2963 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
2964 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
2965 unixctl_command_register("bond/set-active-slave",
2966 bond_unixctl_set_active_slave, NULL);
2967 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
2969 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
2971 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
2974 /* Port functions. */
2976 static struct port *
2977 port_create(struct bridge *br, const char *name)
2981 port = xzalloc(sizeof *port);
2983 port->port_idx = br->n_ports;
2985 port->trunks = NULL;
2986 port->name = xstrdup(name);
2987 port->active_iface = -1;
2989 if (br->n_ports >= br->allocated_ports) {
2990 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2993 br->ports[br->n_ports++] = port;
2995 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3002 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3004 struct shash old_ifaces, new_ifaces;
3005 struct shash_node *node;
3006 unsigned long *trunks;
3012 /* Collect old and new interfaces. */
3013 shash_init(&old_ifaces);
3014 shash_init(&new_ifaces);
3015 for (i = 0; i < port->n_ifaces; i++) {
3016 shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]);
3018 for (i = 0; i < cfg->n_interfaces; i++) {
3019 const char *name = cfg->interfaces[i]->name;
3020 if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) {
3021 VLOG_WARN("port %s: %s specified twice as port interface",
3025 port->updelay = cfg->bond_updelay;
3026 if (port->updelay < 0) {
3029 port->updelay = cfg->bond_downdelay;
3030 if (port->downdelay < 0) {
3031 port->downdelay = 0;
3034 /* Get rid of deleted interfaces and add new interfaces. */
3035 SHASH_FOR_EACH (node, &old_ifaces) {
3036 if (!shash_find(&new_ifaces, node->name)) {
3037 iface_destroy(node->data);
3040 SHASH_FOR_EACH (node, &new_ifaces) {
3041 const struct ovsrec_interface *if_cfg = node->data;
3042 struct iface *iface;
3044 iface = shash_find_data(&old_ifaces, if_cfg->name);
3046 iface = iface_create(port, if_cfg);
3048 iface->cfg = if_cfg;
3055 if (port->n_ifaces < 2) {
3057 if (vlan >= 0 && vlan <= 4095) {
3058 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3063 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3064 * they even work as-is. But they have not been tested. */
3065 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3069 if (port->vlan != vlan) {
3071 bridge_flush(port->bridge);
3074 /* Get trunked VLANs. */
3080 trunks = bitmap_allocate(4096);
3082 for (i = 0; i < cfg->n_trunks; i++) {
3083 int trunk = cfg->trunks[i];
3085 bitmap_set1(trunks, trunk);
3091 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3092 port->name, cfg->n_trunks);
3094 if (n_errors == cfg->n_trunks) {
3096 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3099 bitmap_set_multiple(trunks, 0, 4096, 1);
3102 if (cfg->n_trunks) {
3103 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3108 ? port->trunks != NULL
3109 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3110 bridge_flush(port->bridge);
3112 bitmap_free(port->trunks);
3113 port->trunks = trunks;
3115 shash_destroy(&old_ifaces);
3116 shash_destroy(&new_ifaces);
3120 port_destroy(struct port *port)
3123 struct bridge *br = port->bridge;
3127 proc_net_compat_update_vlan(port->name, NULL, 0);
3128 proc_net_compat_update_bond(port->name, NULL);
3130 for (i = 0; i < MAX_MIRRORS; i++) {
3131 struct mirror *m = br->mirrors[i];
3132 if (m && m->out_port == port) {
3137 while (port->n_ifaces > 0) {
3138 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3141 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3142 del->port_idx = port->port_idx;
3145 bitmap_free(port->trunks);
3152 static struct port *
3153 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3155 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3156 return iface ? iface->port : NULL;
3159 static struct port *
3160 port_lookup(const struct bridge *br, const char *name)
3164 for (i = 0; i < br->n_ports; i++) {
3165 struct port *port = br->ports[i];
3166 if (!strcmp(port->name, name)) {
3173 static struct iface *
3174 port_lookup_iface(const struct port *port, const char *name)
3178 for (j = 0; j < port->n_ifaces; j++) {
3179 struct iface *iface = port->ifaces[j];
3180 if (!strcmp(iface->name, name)) {
3188 port_update_bonding(struct port *port)
3190 if (port->n_ifaces < 2) {
3191 /* Not a bonded port. */
3192 if (port->bond_hash) {
3193 free(port->bond_hash);
3194 port->bond_hash = NULL;
3195 port->bond_compat_is_stale = true;
3196 port->bond_fake_iface = false;
3199 if (!port->bond_hash) {
3202 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3203 for (i = 0; i <= BOND_MASK; i++) {
3204 struct bond_entry *e = &port->bond_hash[i];
3208 port->no_ifaces_tag = tag_create_random();
3209 bond_choose_active_iface(port);
3211 port->bond_compat_is_stale = true;
3212 port->bond_fake_iface = port->cfg->bond_fake_iface;
3217 port_update_bond_compat(struct port *port)
3219 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3220 struct compat_bond bond;
3223 if (port->n_ifaces < 2) {
3224 proc_net_compat_update_bond(port->name, NULL);
3229 bond.updelay = port->updelay;
3230 bond.downdelay = port->downdelay;
3233 bond.hashes = compat_hashes;
3234 if (port->bond_hash) {
3235 const struct bond_entry *e;
3236 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3237 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3238 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3239 cbh->hash = e - port->bond_hash;
3240 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3245 bond.n_slaves = port->n_ifaces;
3246 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3247 for (i = 0; i < port->n_ifaces; i++) {
3248 struct iface *iface = port->ifaces[i];
3249 struct compat_bond_slave *slave = &bond.slaves[i];
3250 slave->name = iface->name;
3252 /* We need to make the same determination as the Linux bonding
3253 * code to determine whether a slave should be consider "up".
3254 * The Linux function bond_miimon_inspect() supports four
3255 * BOND_LINK_* states:
3257 * - BOND_LINK_UP: carrier detected, updelay has passed.
3258 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3259 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3260 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3262 * The function bond_info_show_slave() only considers BOND_LINK_UP
3263 * to be "up" and anything else to be "down".
3265 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3269 netdev_get_etheraddr(iface->netdev, slave->mac);
3272 if (port->bond_fake_iface) {
3273 struct netdev *bond_netdev;
3275 if (!netdev_open_default(port->name, &bond_netdev)) {
3277 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3279 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3281 netdev_close(bond_netdev);
3285 proc_net_compat_update_bond(port->name, &bond);
3290 port_update_vlan_compat(struct port *port)
3292 struct bridge *br = port->bridge;
3293 char *vlandev_name = NULL;
3295 if (port->vlan > 0) {
3296 /* Figure out the name that the VLAN device should actually have, if it
3297 * existed. This takes some work because the VLAN device would not
3298 * have port->name in its name; rather, it would have the trunk port's
3299 * name, and 'port' would be attached to a bridge that also had the
3300 * VLAN device one of its ports. So we need to find a trunk port that
3301 * includes port->vlan.
3303 * There might be more than one candidate. This doesn't happen on
3304 * XenServer, so if it happens we just pick the first choice in
3305 * alphabetical order instead of creating multiple VLAN devices. */
3307 for (i = 0; i < br->n_ports; i++) {
3308 struct port *p = br->ports[i];
3309 if (port_trunks_vlan(p, port->vlan)
3311 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3313 uint8_t ea[ETH_ADDR_LEN];
3314 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3315 if (!eth_addr_is_multicast(ea) &&
3316 !eth_addr_is_reserved(ea) &&
3317 !eth_addr_is_zero(ea)) {
3318 vlandev_name = p->name;
3323 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3326 /* Interface functions. */
3328 static struct iface *
3329 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3331 struct iface *iface;
3332 char *name = if_cfg->name;
3335 iface = xzalloc(sizeof *iface);
3337 iface->port_ifidx = port->n_ifaces;
3338 iface->name = xstrdup(name);
3339 iface->dp_ifidx = -1;
3340 iface->tag = tag_create_random();
3341 iface->delay_expires = LLONG_MAX;
3342 iface->netdev = NULL;
3343 iface->cfg = if_cfg;
3345 if (port->n_ifaces >= port->allocated_ifaces) {
3346 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3347 sizeof *port->ifaces);
3349 port->ifaces[port->n_ifaces++] = iface;
3350 if (port->n_ifaces > 1) {
3351 port->bridge->has_bonded_ports = true;
3354 /* Attempt to create the network interface in case it
3355 * doesn't exist yet. */
3356 if (!iface_is_internal(port->bridge, iface->name)) {
3357 error = set_up_iface(if_cfg, iface, true);
3359 VLOG_WARN("could not create iface %s: %s", iface->name,
3364 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3366 bridge_flush(port->bridge);
3372 iface_destroy(struct iface *iface)
3375 struct port *port = iface->port;
3376 struct bridge *br = port->bridge;
3377 bool del_active = port->active_iface == iface->port_ifidx;
3380 if (iface->dp_ifidx >= 0) {
3381 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3384 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3385 del->port_ifidx = iface->port_ifidx;
3387 netdev_close(iface->netdev);
3390 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3391 bond_choose_active_iface(port);
3392 bond_send_learning_packets(port);
3398 bridge_flush(port->bridge);
3402 static struct iface *
3403 iface_lookup(const struct bridge *br, const char *name)
3407 for (i = 0; i < br->n_ports; i++) {
3408 struct port *port = br->ports[i];
3409 for (j = 0; j < port->n_ifaces; j++) {
3410 struct iface *iface = port->ifaces[j];
3411 if (!strcmp(iface->name, name)) {
3419 static struct iface *
3420 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3422 return port_array_get(&br->ifaces, dp_ifidx);
3425 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3426 * 'br', that is, an interface that is entirely simulated within the datapath.
3427 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3428 * interfaces are created by setting "iface.<iface>.internal = true".
3430 * In addition, we have a kluge-y feature that creates an internal port with
3431 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3432 * This feature needs to go away in the long term. Until then, this is one
3433 * reason why this function takes a name instead of a struct iface: the fake
3434 * interfaces created this way do not have a struct iface. */
3436 iface_is_internal(const struct bridge *br, const char *if_name)
3438 /* XXX wastes time */
3439 struct iface *iface;
3442 if (!strcmp(if_name, br->name)) {
3446 iface = iface_lookup(br, if_name);
3447 if (iface && !strcmp(iface->cfg->type, "internal")) {
3451 port = port_lookup(br, if_name);
3452 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3458 /* Set Ethernet address of 'iface', if one is specified in the configuration
3461 iface_set_mac(struct iface *iface)
3463 uint8_t ea[ETH_ADDR_LEN];
3465 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3466 if (eth_addr_is_multicast(ea)) {
3467 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3469 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3470 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3471 iface->name, iface->name);
3473 int error = netdev_set_etheraddr(iface->netdev, ea);
3475 VLOG_ERR("interface %s: setting MAC failed (%s)",
3476 iface->name, strerror(error));
3482 /* Port mirroring. */
3485 mirror_reconfigure(struct bridge *br)
3487 struct shash old_mirrors, new_mirrors;
3488 struct shash_node *node;
3489 unsigned long *rspan_vlans;
3492 /* Collect old mirrors. */
3493 shash_init(&old_mirrors);
3494 for (i = 0; i < MAX_MIRRORS; i++) {
3495 if (br->mirrors[i]) {
3496 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3500 /* Collect new mirrors. */
3501 shash_init(&new_mirrors);
3502 for (i = 0; i < br->cfg->n_mirrors; i++) {
3503 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3504 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3505 VLOG_WARN("bridge %s: %s specified twice as mirror",
3506 br->name, cfg->name);
3510 /* Get rid of deleted mirrors and add new mirrors. */
3511 SHASH_FOR_EACH (node, &old_mirrors) {
3512 if (!shash_find(&new_mirrors, node->name)) {
3513 mirror_destroy(node->data);
3516 SHASH_FOR_EACH (node, &new_mirrors) {
3517 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3519 mirror = mirror_create(br, node->name);
3524 mirror_reconfigure_one(mirror, node->data);
3526 shash_destroy(&old_mirrors);
3527 shash_destroy(&new_mirrors);
3529 /* Update port reserved status. */
3530 for (i = 0; i < br->n_ports; i++) {
3531 br->ports[i]->is_mirror_output_port = false;
3533 for (i = 0; i < MAX_MIRRORS; i++) {
3534 struct mirror *m = br->mirrors[i];
3535 if (m && m->out_port) {
3536 m->out_port->is_mirror_output_port = true;
3540 /* Update flooded vlans (for RSPAN). */
3542 if (br->cfg->n_flood_vlans) {
3543 rspan_vlans = bitmap_allocate(4096);
3545 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3546 int64_t vlan = br->cfg->flood_vlans[i];
3547 if (vlan >= 0 && vlan < 4096) {
3548 bitmap_set1(rspan_vlans, vlan);
3549 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3552 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3557 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3562 static struct mirror *
3563 mirror_create(struct bridge *br, const char *name)
3568 for (i = 0; ; i++) {
3569 if (i >= MAX_MIRRORS) {
3570 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3571 "cannot create %s", br->name, MAX_MIRRORS, name);
3574 if (!br->mirrors[i]) {
3579 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3582 br->mirrors[i] = m = xzalloc(sizeof *m);
3585 m->name = xstrdup(name);
3586 shash_init(&m->src_ports);
3587 shash_init(&m->dst_ports);
3597 mirror_destroy(struct mirror *m)
3600 struct bridge *br = m->bridge;
3603 for (i = 0; i < br->n_ports; i++) {
3604 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3605 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3608 shash_destroy(&m->src_ports);
3609 shash_destroy(&m->dst_ports);
3612 m->bridge->mirrors[m->idx] = NULL;
3620 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3621 struct shash *names)
3625 for (i = 0; i < n_ports; i++) {
3626 const char *name = ports[i]->name;
3627 if (port_lookup(m->bridge, name)) {
3628 shash_add_once(names, name, NULL);
3630 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3631 "port %s", m->bridge->name, m->name, name);
3637 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3643 *vlans = xmalloc(sizeof *vlans * cfg->n_select_vlan);
3645 for (i = 0; i < cfg->n_select_vlan; i++) {
3646 int64_t vlan = cfg->select_vlan[i];
3647 if (vlan < 0 || vlan > 4095) {
3648 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3649 m->bridge->name, m->name, vlan);
3651 (*vlans)[n_vlans++] = vlan;
3658 vlan_is_mirrored(const struct mirror *m, int vlan)
3662 for (i = 0; i < m->n_vlans; i++) {
3663 if (m->vlans[i] == vlan) {
3671 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3675 for (i = 0; i < m->n_vlans; i++) {
3676 if (port_trunks_vlan(p, m->vlans[i])) {
3684 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3686 struct shash src_ports, dst_ports;
3687 mirror_mask_t mirror_bit;
3688 struct port *out_port;
3693 bool mirror_all_ports;
3694 bool any_ports_specified;
3695 bool any_vlans_specified;
3697 /* Get output port. */
3698 if (cfg->output_port) {
3699 out_port = port_lookup(m->bridge, cfg->output_port->name);
3701 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3702 m->bridge->name, m->name);
3708 if (cfg->output_vlan) {
3709 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3710 "output vlan; ignoring output vlan",
3711 m->bridge->name, m->name);
3713 } else if (cfg->output_vlan) {
3715 out_vlan = *cfg->output_vlan;
3717 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3718 m->bridge->name, m->name);
3723 /* Get all the ports, and drop duplicates and ports that don't exist. */
3724 shash_init(&src_ports);
3725 shash_init(&dst_ports);
3726 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
3728 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
3730 any_ports_specified = cfg->n_select_dst_port || cfg->n_select_dst_port;
3731 if (any_ports_specified
3732 && shash_is_empty(&src_ports) && shash_is_empty(&dst_ports)) {
3733 VLOG_ERR("bridge %s: disabling mirror %s since none of the specified "
3734 "selection ports exists", m->bridge->name, m->name);
3739 /* Get all the vlans, and drop duplicate and invalid vlans. */
3740 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
3741 any_vlans_specified = cfg->n_select_vlan > 0;
3742 if (any_vlans_specified && !n_vlans) {
3743 VLOG_ERR("bridge %s: disabling mirror %s since none of the specified "
3744 "VLANs exists", m->bridge->name, m->name);
3749 /* Update mirror data. */
3750 if (!shash_equal_keys(&m->src_ports, &src_ports)
3751 || !shash_equal_keys(&m->dst_ports, &dst_ports)
3752 || m->n_vlans != n_vlans
3753 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3754 || m->out_port != out_port
3755 || m->out_vlan != out_vlan) {
3756 bridge_flush(m->bridge);
3758 shash_swap(&m->src_ports, &src_ports);
3759 shash_swap(&m->dst_ports, &dst_ports);
3762 m->n_vlans = n_vlans;
3763 m->out_port = out_port;
3764 m->out_vlan = out_vlan;
3766 /* If no selection criteria have been given, mirror for all ports. */
3767 mirror_all_ports = !any_ports_specified && !any_vlans_specified;
3770 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3771 for (i = 0; i < m->bridge->n_ports; i++) {
3772 struct port *port = m->bridge->ports[i];
3774 if (mirror_all_ports
3775 || shash_find(&m->src_ports, port->name)
3778 ? port_trunks_any_mirrored_vlan(m, port)
3779 : vlan_is_mirrored(m, port->vlan)))) {
3780 port->src_mirrors |= mirror_bit;
3782 port->src_mirrors &= ~mirror_bit;
3785 if (mirror_all_ports || shash_find(&m->dst_ports, port->name)) {
3786 port->dst_mirrors |= mirror_bit;
3788 port->dst_mirrors &= ~mirror_bit;
3794 shash_destroy(&src_ports);
3795 shash_destroy(&dst_ports);