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 char *name);
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;
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 dp_enumerate(&dpif_names);
301 for (i = 0; i < dpif_names.n; i++) {
302 const char *dpif_name = dpif_names.names[i];
306 retval = dpif_open(dpif_name, &dpif);
308 struct svec all_names;
311 svec_init(&all_names);
312 dpif_get_all_names(dpif, &all_names);
313 for (j = 0; j < all_names.n; j++) {
314 if (svec_contains(&bridge_names, all_names.names[j])) {
320 svec_destroy(&all_names);
324 svec_destroy(&dpif_names);
326 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
330 bridge_reconfigure(cfg);
335 config_string_change(const char *value, char **valuep)
337 if (value && (!*valuep || strcmp(value, *valuep))) {
339 *valuep = xstrdup(value);
347 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
349 /* XXX SSL should be configurable on a per-bridge basis.
350 * XXX should be possible to de-configure SSL. */
351 static char *private_key_file;
352 static char *certificate_file;
353 static char *cacert_file;
357 /* XXX We can't un-set SSL settings. */
361 if (config_string_change(ssl->private_key, &private_key_file)) {
362 stream_ssl_set_private_key_file(private_key_file);
365 if (config_string_change(ssl->certificate, &certificate_file)) {
366 stream_ssl_set_certificate_file(certificate_file);
369 /* We assume that even if the filename hasn't changed, if the CA cert
370 * file has been removed, that we want to move back into
371 * boot-strapping mode. This opens a small security hole, because
372 * the old certificate will still be trusted until vSwitch is
373 * restarted. We may want to address this in vconn's SSL library. */
374 if (config_string_change(ssl->ca_cert, &cacert_file)
375 || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) {
376 stream_ssl_set_ca_cert_file(cacert_file, ssl->bootstrap_ca_cert);
381 /* Attempt to create the network device 'iface_name' through the netdev
384 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
387 struct shash_node *node;
388 struct shash options;
392 shash_init(&options);
393 for (i = 0; i < iface_cfg->n_options; i++) {
394 shash_add(&options, iface_cfg->key_options[i],
395 xstrdup(iface_cfg->value_options[i]));
399 struct netdev_options netdev_options;
401 memset(&netdev_options, 0, sizeof netdev_options);
402 netdev_options.name = iface_cfg->name;
403 netdev_options.type = iface_cfg->type;
404 netdev_options.args = &options;
405 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
406 netdev_options.may_create = true;
407 if (iface_is_internal(iface->port->bridge, iface_cfg->name)) {
408 netdev_options.may_open = true;
411 error = netdev_open(&netdev_options, &iface->netdev);
414 netdev_get_carrier(iface->netdev, &iface->enabled);
416 } else if (iface->netdev) {
417 const char *netdev_type = netdev_get_type(iface->netdev);
418 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
419 ? iface_cfg->type : NULL;
421 if (!iface_type || !strcmp(netdev_type, iface_type)) {
422 error = netdev_reconfigure(iface->netdev, &options);
424 VLOG_WARN("%s: attempting change device type from %s to %s",
425 iface_cfg->name, netdev_type, iface_type);
430 SHASH_FOR_EACH (node, &options) {
433 shash_destroy(&options);
439 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
441 return set_up_iface(iface_cfg, iface, false);
445 check_iface_netdev(struct bridge *br UNUSED, struct iface *iface,
448 if (!iface->netdev) {
449 int error = set_up_iface(iface->cfg, iface, true);
451 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
461 check_iface_dp_ifidx(struct bridge *br, struct iface *iface, void *aux UNUSED)
463 if (iface->dp_ifidx >= 0) {
464 VLOG_DBG("%s has interface %s on port %d",
466 iface->name, iface->dp_ifidx);
469 VLOG_ERR("%s interface not in %s, dropping",
470 iface->name, dpif_name(br->dpif));
476 set_iface_properties(struct bridge *br UNUSED, struct iface *iface,
479 /* Set policing attributes. */
480 netdev_set_policing(iface->netdev,
481 iface->cfg->ingress_policing_rate,
482 iface->cfg->ingress_policing_burst);
484 /* Set MAC address of internal interfaces other than the local
486 if (iface->dp_ifidx != ODPP_LOCAL
487 && iface_is_internal(br, iface->name)) {
488 iface_set_mac(iface);
494 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
495 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
496 * deletes from 'br' any ports that no longer have any interfaces. */
498 iterate_and_prune_ifaces(struct bridge *br,
499 bool (*cb)(struct bridge *, struct iface *,
505 for (i = 0; i < br->n_ports; ) {
506 struct port *port = br->ports[i];
507 for (j = 0; j < port->n_ifaces; ) {
508 struct iface *iface = port->ifaces[j];
509 if (cb(br, iface, aux)) {
512 iface_destroy(iface);
516 if (port->n_ifaces) {
519 VLOG_ERR("%s port has no interfaces, dropping", port->name);
526 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
528 struct ovsdb_idl_txn *txn;
529 struct shash old_br, new_br;
530 struct shash_node *node;
531 struct bridge *br, *next;
534 COVERAGE_INC(bridge_reconfigure);
536 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
538 /* Collect old and new bridges. */
541 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
542 shash_add(&old_br, br->name, br);
544 for (i = 0; i < ovs_cfg->n_bridges; i++) {
545 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
546 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
547 VLOG_WARN("more than one bridge named %s", br_cfg->name);
551 /* Get rid of deleted bridges and add new bridges. */
552 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
553 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
560 SHASH_FOR_EACH (node, &new_br) {
561 const char *br_name = node->name;
562 const struct ovsrec_bridge *br_cfg = node->data;
563 if (!shash_find_data(&old_br, br_name)) {
564 br = bridge_create(br_name);
570 shash_destroy(&old_br);
571 shash_destroy(&new_br);
575 bridge_configure_ssl(ovs_cfg->ssl);
578 /* Reconfigure all bridges. */
579 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
580 bridge_reconfigure_one(ovs_cfg, br);
583 /* Add and delete ports on all datapaths.
585 * The kernel will reject any attempt to add a given port to a datapath if
586 * that port already belongs to a different datapath, so we must do all
587 * port deletions before any port additions. */
588 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
589 struct odp_port *dpif_ports;
591 struct shash want_ifaces;
593 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
594 bridge_get_all_ifaces(br, &want_ifaces);
595 for (i = 0; i < n_dpif_ports; i++) {
596 const struct odp_port *p = &dpif_ports[i];
597 if (!shash_find(&want_ifaces, p->devname)
598 && strcmp(p->devname, br->name)) {
599 int retval = dpif_port_del(br->dpif, p->port);
601 VLOG_ERR("failed to remove %s interface from %s: %s",
602 p->devname, dpif_name(br->dpif),
607 shash_destroy(&want_ifaces);
610 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
611 struct odp_port *dpif_ports;
613 struct shash cur_ifaces, want_ifaces;
614 struct shash_node *node;
616 /* Get the set of interfaces currently in this datapath. */
617 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
618 shash_init(&cur_ifaces);
619 for (i = 0; i < n_dpif_ports; i++) {
620 const char *name = dpif_ports[i].devname;
621 if (!shash_find(&cur_ifaces, name)) {
622 shash_add(&cur_ifaces, name, NULL);
627 /* Get the set of interfaces we want on this datapath. */
628 bridge_get_all_ifaces(br, &want_ifaces);
630 SHASH_FOR_EACH (node, &want_ifaces) {
631 const char *if_name = node->name;
632 struct iface *iface = node->data;
634 if (shash_find(&cur_ifaces, if_name)) {
635 /* Already exists, just reconfigure it. */
637 reconfigure_iface(iface->cfg, iface);
640 /* Need to add to datapath. */
644 /* Add to datapath. */
645 internal = iface_is_internal(br, if_name);
646 error = dpif_port_add(br->dpif, if_name,
647 internal ? ODP_PORT_INTERNAL : 0, NULL);
648 if (error == EFBIG) {
649 VLOG_ERR("ran out of valid port numbers on %s",
650 dpif_name(br->dpif));
653 VLOG_ERR("failed to add %s interface to %s: %s",
654 if_name, dpif_name(br->dpif), strerror(error));
658 shash_destroy(&cur_ifaces);
659 shash_destroy(&want_ifaces);
661 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
664 struct iface *local_iface;
665 struct iface *hw_addr_iface;
668 bridge_fetch_dp_ifaces(br);
670 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
671 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
673 /* Pick local port hardware address, datapath ID. */
674 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
675 local_iface = bridge_get_local_iface(br);
677 int error = netdev_set_etheraddr(local_iface->netdev, ea);
679 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
680 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
681 "Ethernet address: %s",
682 br->name, strerror(error));
686 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
687 ofproto_set_datapath_id(br->ofproto, dpid);
689 dpid_string = xasprintf("%012"PRIx64, dpid);
690 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
693 /* Set NetFlow configuration on this bridge. */
694 if (br->cfg->netflow) {
695 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
696 struct netflow_options opts;
698 memset(&opts, 0, sizeof opts);
700 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
701 if (nf_cfg->engine_type) {
702 opts.engine_type = nf_cfg->engine_type;
704 if (nf_cfg->engine_id) {
705 opts.engine_id = nf_cfg->engine_id;
708 opts.active_timeout = nf_cfg->active_timeout;
709 if (!opts.active_timeout) {
710 opts.active_timeout = -1;
711 } else if (opts.active_timeout < 0) {
712 VLOG_WARN("bridge %s: active timeout interval set to negative "
713 "value, using default instead (%d seconds)", br->name,
714 NF_ACTIVE_TIMEOUT_DEFAULT);
715 opts.active_timeout = -1;
718 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
719 if (opts.add_id_to_iface) {
720 if (opts.engine_id > 0x7f) {
721 VLOG_WARN("bridge %s: netflow port mangling may conflict "
722 "with another vswitch, choose an engine id less "
723 "than 128", br->name);
725 if (br->n_ports > 508) {
726 VLOG_WARN("bridge %s: netflow port mangling will conflict "
727 "with another port when more than 508 ports are "
732 opts.collectors.n = nf_cfg->n_targets;
733 opts.collectors.names = nf_cfg->targets;
734 if (ofproto_set_netflow(br->ofproto, &opts)) {
735 VLOG_ERR("bridge %s: problem setting netflow collectors",
739 ofproto_set_netflow(br->ofproto, NULL);
742 /* Update the controller and related settings. It would be more
743 * straightforward to call this from bridge_reconfigure_one(), but we
744 * can't do it there for two reasons. First, and most importantly, at
745 * that point we don't know the dp_ifidx of any interfaces that have
746 * been added to the bridge (because we haven't actually added them to
747 * the datapath). Second, at that point we haven't set the datapath ID
748 * yet; when a controller is configured, resetting the datapath ID will
749 * immediately disconnect from the controller, so it's better to set
750 * the datapath ID before the controller. */
751 bridge_reconfigure_controller(ovs_cfg, br);
753 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
754 for (i = 0; i < br->n_ports; i++) {
755 struct port *port = br->ports[i];
757 port_update_vlan_compat(port);
758 port_update_bonding(port);
761 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
762 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
765 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
767 ovsdb_idl_txn_commit(txn);
768 ovsdb_idl_txn_destroy(txn); /* XXX */
772 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
776 for (i = 0; i < br_cfg->n_other_config; i++) {
777 if (!strcmp(br_cfg->key_other_config[i], key)) {
778 return br_cfg->value_other_config[i];
785 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
786 struct iface **hw_addr_iface)
792 *hw_addr_iface = NULL;
794 /* Did the user request a particular MAC? */
795 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
796 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
797 if (eth_addr_is_multicast(ea)) {
798 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
799 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
800 } else if (eth_addr_is_zero(ea)) {
801 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
807 /* Otherwise choose the minimum non-local MAC address among all of the
809 memset(ea, 0xff, sizeof ea);
810 for (i = 0; i < br->n_ports; i++) {
811 struct port *port = br->ports[i];
812 uint8_t iface_ea[ETH_ADDR_LEN];
815 /* Mirror output ports don't participate. */
816 if (port->is_mirror_output_port) {
820 /* Choose the MAC address to represent the port. */
821 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
822 /* Find the interface with this Ethernet address (if any) so that
823 * we can provide the correct devname to the caller. */
825 for (j = 0; j < port->n_ifaces; j++) {
826 struct iface *candidate = port->ifaces[j];
827 uint8_t candidate_ea[ETH_ADDR_LEN];
828 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
829 && eth_addr_equals(iface_ea, candidate_ea)) {
834 /* Choose the interface whose MAC address will represent the port.
835 * The Linux kernel bonding code always chooses the MAC address of
836 * the first slave added to a bond, and the Fedora networking
837 * scripts always add slaves to a bond in alphabetical order, so
838 * for compatibility we choose the interface with the name that is
839 * first in alphabetical order. */
840 iface = port->ifaces[0];
841 for (j = 1; j < port->n_ifaces; j++) {
842 struct iface *candidate = port->ifaces[j];
843 if (strcmp(candidate->name, iface->name) < 0) {
848 /* The local port doesn't count (since we're trying to choose its
849 * MAC address anyway). */
850 if (iface->dp_ifidx == ODPP_LOCAL) {
855 error = netdev_get_etheraddr(iface->netdev, iface_ea);
857 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
858 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
859 iface->name, strerror(error));
864 /* Compare against our current choice. */
865 if (!eth_addr_is_multicast(iface_ea) &&
866 !eth_addr_is_local(iface_ea) &&
867 !eth_addr_is_reserved(iface_ea) &&
868 !eth_addr_is_zero(iface_ea) &&
869 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
871 memcpy(ea, iface_ea, ETH_ADDR_LEN);
872 *hw_addr_iface = iface;
875 if (eth_addr_is_multicast(ea)) {
876 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
877 *hw_addr_iface = NULL;
878 VLOG_WARN("bridge %s: using default bridge Ethernet "
879 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
881 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
882 br->name, ETH_ADDR_ARGS(ea));
886 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
887 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
888 * an interface on 'br', then that interface must be passed in as
889 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
890 * 'hw_addr_iface' must be passed in as a null pointer. */
892 bridge_pick_datapath_id(struct bridge *br,
893 const uint8_t bridge_ea[ETH_ADDR_LEN],
894 struct iface *hw_addr_iface)
897 * The procedure for choosing a bridge MAC address will, in the most
898 * ordinary case, also choose a unique MAC that we can use as a datapath
899 * ID. In some special cases, though, multiple bridges will end up with
900 * the same MAC address. This is OK for the bridges, but it will confuse
901 * the OpenFlow controller, because each datapath needs a unique datapath
904 * Datapath IDs must be unique. It is also very desirable that they be
905 * stable from one run to the next, so that policy set on a datapath
908 const char *datapath_id;
911 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
912 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
918 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
920 * A bridge whose MAC address is taken from a VLAN network device
921 * (that is, a network device created with vconfig(8) or similar
922 * tool) will have the same MAC address as a bridge on the VLAN
923 * device's physical network device.
925 * Handle this case by hashing the physical network device MAC
926 * along with the VLAN identifier.
928 uint8_t buf[ETH_ADDR_LEN + 2];
929 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
930 buf[ETH_ADDR_LEN] = vlan >> 8;
931 buf[ETH_ADDR_LEN + 1] = vlan;
932 return dpid_from_hash(buf, sizeof buf);
935 * Assume that this bridge's MAC address is unique, since it
936 * doesn't fit any of the cases we handle specially.
941 * A purely internal bridge, that is, one that has no non-virtual
942 * network devices on it at all, is more difficult because it has no
943 * natural unique identifier at all.
945 * When the host is a XenServer, we handle this case by hashing the
946 * host's UUID with the name of the bridge. Names of bridges are
947 * persistent across XenServer reboots, although they can be reused if
948 * an internal network is destroyed and then a new one is later
949 * created, so this is fairly effective.
951 * When the host is not a XenServer, we punt by using a random MAC
952 * address on each run.
954 const char *host_uuid = xenserver_get_host_uuid();
956 char *combined = xasprintf("%s,%s", host_uuid, br->name);
957 dpid = dpid_from_hash(combined, strlen(combined));
963 return eth_addr_to_uint64(bridge_ea);
967 dpid_from_hash(const void *data, size_t n)
969 uint8_t hash[SHA1_DIGEST_SIZE];
971 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
972 sha1_bytes(data, n, hash);
973 eth_addr_mark_random(hash);
974 return eth_addr_to_uint64(hash);
980 struct bridge *br, *next;
984 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
985 int error = bridge_run_one(br);
987 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
988 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
989 "forcing reconfiguration", br->name);
1003 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1004 ofproto_wait(br->ofproto);
1005 if (br->controller) {
1009 mac_learning_wait(br->ml);
1014 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1015 * configuration changes. */
1017 bridge_flush(struct bridge *br)
1019 COVERAGE_INC(bridge_flush);
1021 mac_learning_flush(br->ml);
1024 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1025 * such interface. */
1026 static struct iface *
1027 bridge_get_local_iface(struct bridge *br)
1031 for (i = 0; i < br->n_ports; i++) {
1032 struct port *port = br->ports[i];
1033 for (j = 0; j < port->n_ifaces; j++) {
1034 struct iface *iface = port->ifaces[j];
1035 if (iface->dp_ifidx == ODPP_LOCAL) {
1044 /* Bridge unixctl user interface functions. */
1046 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1047 const char *args, void *aux UNUSED)
1049 struct ds ds = DS_EMPTY_INITIALIZER;
1050 const struct bridge *br;
1051 const struct mac_entry *e;
1053 br = bridge_lookup(args);
1055 unixctl_command_reply(conn, 501, "no such bridge");
1059 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1060 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1061 if (e->port < 0 || e->port >= br->n_ports) {
1064 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1065 br->ports[e->port]->ifaces[0]->dp_ifidx,
1066 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1068 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1072 /* Bridge reconfiguration functions. */
1073 static struct bridge *
1074 bridge_create(const char *name)
1079 assert(!bridge_lookup(name));
1080 br = xzalloc(sizeof *br);
1082 error = dpif_create_and_open(name, &br->dpif);
1087 dpif_flow_flush(br->dpif);
1089 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
1091 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
1092 dpif_delete(br->dpif);
1093 dpif_close(br->dpif);
1098 br->name = xstrdup(name);
1099 br->ml = mac_learning_create();
1100 br->sent_config_request = false;
1101 eth_addr_random(br->default_ea);
1103 port_array_init(&br->ifaces);
1106 br->bond_next_rebalance = time_msec() + 10000;
1108 list_push_back(&all_bridges, &br->node);
1110 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1116 bridge_destroy(struct bridge *br)
1121 while (br->n_ports > 0) {
1122 port_destroy(br->ports[br->n_ports - 1]);
1124 list_remove(&br->node);
1125 error = dpif_delete(br->dpif);
1126 if (error && error != ENOENT) {
1127 VLOG_ERR("failed to delete %s: %s",
1128 dpif_name(br->dpif), strerror(error));
1130 dpif_close(br->dpif);
1131 ofproto_destroy(br->ofproto);
1132 free(br->controller);
1133 mac_learning_destroy(br->ml);
1134 port_array_destroy(&br->ifaces);
1141 static struct bridge *
1142 bridge_lookup(const char *name)
1146 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1147 if (!strcmp(br->name, name)) {
1155 bridge_exists(const char *name)
1157 return bridge_lookup(name) ? true : false;
1161 bridge_get_datapathid(const char *name)
1163 struct bridge *br = bridge_lookup(name);
1164 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1167 /* Handle requests for a listing of all flows known by the OpenFlow
1168 * stack, including those normally hidden. */
1170 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1171 const char *args, void *aux UNUSED)
1176 br = bridge_lookup(args);
1178 unixctl_command_reply(conn, 501, "Unknown bridge");
1183 ofproto_get_all_flows(br->ofproto, &results);
1185 unixctl_command_reply(conn, 200, ds_cstr(&results));
1186 ds_destroy(&results);
1190 bridge_run_one(struct bridge *br)
1194 error = ofproto_run1(br->ofproto);
1199 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1202 error = ofproto_run2(br->ofproto, br->flush);
1208 static const struct ovsrec_controller *
1209 bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1210 const struct bridge *br)
1212 const struct ovsrec_controller *controller;
1214 controller = (br->cfg->controller ? br->cfg->controller
1215 : ovs_cfg->controller ? ovs_cfg->controller
1218 if (controller && !strcmp(controller->target, "none")) {
1226 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1228 struct svec *ifaces = ifaces_;
1229 if (!svec_contains(ifaces, iface->name)) {
1230 svec_add(ifaces, iface->name);
1234 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1236 br->name, iface->name, iface->port->name);
1242 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1245 struct shash old_ports, new_ports;
1247 struct svec listeners, old_listeners;
1248 struct svec snoops, old_snoops;
1249 struct shash_node *node;
1253 /* Collect old ports. */
1254 shash_init(&old_ports);
1255 for (i = 0; i < br->n_ports; i++) {
1256 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1259 /* Collect new ports. */
1260 shash_init(&new_ports);
1261 for (i = 0; i < br->cfg->n_ports; i++) {
1262 const char *name = br->cfg->ports[i]->name;
1263 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1264 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1269 /* If we have a controller, then we need a local port. Complain if the
1270 * user didn't specify one.
1272 * XXX perhaps we should synthesize a port ourselves in this case. */
1273 if (bridge_get_controller(ovs_cfg, br)) {
1274 char local_name[IF_NAMESIZE];
1277 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1278 local_name, sizeof local_name);
1279 if (!error && !shash_find(&new_ports, local_name)) {
1280 VLOG_WARN("bridge %s: controller specified but no local port "
1281 "(port named %s) defined",
1282 br->name, local_name);
1286 dpid_from_string(ovs_cfg->management_id, &mgmt_id);
1287 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
1289 /* Get rid of deleted ports and add new ports. */
1290 SHASH_FOR_EACH (node, &old_ports) {
1291 if (!shash_find(&new_ports, node->name)) {
1292 port_destroy(node->data);
1295 SHASH_FOR_EACH (node, &new_ports) {
1296 struct port *port = shash_find_data(&old_ports, node->name);
1298 port = port_create(br, node->name);
1300 port_reconfigure(port, node->data);
1302 shash_destroy(&old_ports);
1303 shash_destroy(&new_ports);
1305 /* Check and delete duplicate interfaces. */
1307 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1308 svec_destroy(&ifaces);
1310 /* Delete all flows if we're switching from connected to standalone or vice
1311 * versa. (XXX Should we delete all flows if we are switching from one
1312 * controller to another?) */
1315 /* Configure OpenFlow management listeners. */
1316 svec_init(&listeners);
1317 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1319 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1320 ovs_rundir, br->name));
1321 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1322 svec_clear(&listeners);
1324 svec_sort_unique(&listeners);
1326 svec_init(&old_listeners);
1327 ofproto_get_listeners(br->ofproto, &old_listeners);
1328 svec_sort_unique(&old_listeners);
1330 if (!svec_equal(&listeners, &old_listeners)) {
1331 ofproto_set_listeners(br->ofproto, &listeners);
1333 svec_destroy(&listeners);
1334 svec_destroy(&old_listeners);
1336 /* Configure OpenFlow controller connection snooping. */
1338 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1340 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1341 ovs_rundir, br->name));
1342 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1343 svec_clear(&snoops);
1345 svec_sort_unique(&snoops);
1347 svec_init(&old_snoops);
1348 ofproto_get_snoops(br->ofproto, &old_snoops);
1349 svec_sort_unique(&old_snoops);
1351 if (!svec_equal(&snoops, &old_snoops)) {
1352 ofproto_set_snoops(br->ofproto, &snoops);
1354 svec_destroy(&snoops);
1355 svec_destroy(&old_snoops);
1357 /* Default listener. */
1358 svec_init(&listeners);
1359 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1360 ovs_rundir, br->name));
1361 svec_init(&old_listeners);
1362 ofproto_get_listeners(br->ofproto, &old_listeners);
1363 if (!svec_equal(&listeners, &old_listeners)) {
1364 ofproto_set_listeners(br->ofproto, &listeners);
1366 svec_destroy(&listeners);
1367 svec_destroy(&old_listeners);
1369 /* Default snoop. */
1371 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1372 ovs_rundir, br->name));
1373 svec_init(&old_snoops);
1374 ofproto_get_snoops(br->ofproto, &old_snoops);
1375 if (!svec_equal(&snoops, &old_snoops)) {
1376 ofproto_set_snoops(br->ofproto, &snoops);
1378 svec_destroy(&snoops);
1379 svec_destroy(&old_snoops);
1382 mirror_reconfigure(br);
1386 bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1389 char *pfx = xasprintf("bridge.%s.controller", br->name);
1390 const struct ovsrec_controller *c;
1392 c = bridge_get_controller(ovs_cfg, br);
1393 if ((br->controller != NULL) != (c != NULL)) {
1394 ofproto_flush_flows(br->ofproto);
1396 free(br->controller);
1397 br->controller = c ? xstrdup(c->target) : NULL;
1400 int max_backoff, probe;
1401 int rate_limit, burst_limit;
1403 if (!strcmp(c->target, "discover")) {
1404 ofproto_set_discovery(br->ofproto, true,
1405 c->discover_accept_regex,
1406 c->discover_update_resolv_conf);
1408 struct iface *local_iface;
1412 in_band = (!c->connection_mode
1413 || !strcmp(c->connection_mode, "out-of-band"));
1414 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1415 ofproto_set_in_band(br->ofproto, in_band);
1417 local_iface = bridge_get_local_iface(br);
1418 if (local_iface && c->local_ip && inet_aton(c->local_ip, &ip)) {
1419 struct netdev *netdev = local_iface->netdev;
1420 struct in_addr ip, mask, gateway;
1422 if (!c->local_netmask || !inet_aton(c->local_netmask, &mask)) {
1425 if (!c->local_gateway
1426 || !inet_aton(c->local_gateway, &gateway)) {
1430 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1432 mask.s_addr = guess_netmask(ip.s_addr);
1434 if (!netdev_set_in4(netdev, ip, mask)) {
1435 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1437 br->name, IP_ARGS(&ip.s_addr),
1438 IP_ARGS(&mask.s_addr));
1441 if (gateway.s_addr) {
1442 if (!netdev_add_router(netdev, gateway)) {
1443 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1444 br->name, IP_ARGS(&gateway.s_addr));
1450 ofproto_set_failure(br->ofproto,
1452 || !strcmp(c->fail_mode, "standalone")
1453 || !strcmp(c->fail_mode, "open")));
1455 probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1456 ofproto_set_probe_interval(br->ofproto, probe);
1458 max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1459 ofproto_set_max_backoff(br->ofproto, max_backoff);
1461 rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1462 burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0;
1463 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1465 union ofp_action action;
1468 /* Set up a flow that matches every packet and directs them to
1469 * OFPP_NORMAL (which goes to us). */
1470 memset(&action, 0, sizeof action);
1471 action.type = htons(OFPAT_OUTPUT);
1472 action.output.len = htons(sizeof action);
1473 action.output.port = htons(OFPP_NORMAL);
1474 memset(&flow, 0, sizeof flow);
1475 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1478 ofproto_set_in_band(br->ofproto, false);
1479 ofproto_set_max_backoff(br->ofproto, 1);
1480 ofproto_set_probe_interval(br->ofproto, 5);
1481 ofproto_set_failure(br->ofproto, false);
1485 ofproto_set_controller(br->ofproto, br->controller);
1489 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1494 for (i = 0; i < br->n_ports; i++) {
1495 struct port *port = br->ports[i];
1496 for (j = 0; j < port->n_ifaces; j++) {
1497 struct iface *iface = port->ifaces[j];
1498 shash_add_once(ifaces, iface->name, iface);
1500 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1501 shash_add_once(ifaces, port->name, NULL);
1506 /* For robustness, in case the administrator moves around datapath ports behind
1507 * our back, we re-check all the datapath port numbers here.
1509 * This function will set the 'dp_ifidx' members of interfaces that have
1510 * disappeared to -1, so only call this function from a context where those
1511 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1512 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1513 * datapath, which doesn't support UINT16_MAX+1 ports. */
1515 bridge_fetch_dp_ifaces(struct bridge *br)
1517 struct odp_port *dpif_ports;
1518 size_t n_dpif_ports;
1521 /* Reset all interface numbers. */
1522 for (i = 0; i < br->n_ports; i++) {
1523 struct port *port = br->ports[i];
1524 for (j = 0; j < port->n_ifaces; j++) {
1525 struct iface *iface = port->ifaces[j];
1526 iface->dp_ifidx = -1;
1529 port_array_clear(&br->ifaces);
1531 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1532 for (i = 0; i < n_dpif_ports; i++) {
1533 struct odp_port *p = &dpif_ports[i];
1534 struct iface *iface = iface_lookup(br, p->devname);
1536 if (iface->dp_ifidx >= 0) {
1537 VLOG_WARN("%s reported interface %s twice",
1538 dpif_name(br->dpif), p->devname);
1539 } else if (iface_from_dp_ifidx(br, p->port)) {
1540 VLOG_WARN("%s reported interface %"PRIu16" twice",
1541 dpif_name(br->dpif), p->port);
1543 port_array_set(&br->ifaces, p->port, iface);
1544 iface->dp_ifidx = p->port;
1548 int64_t ofport = (iface->dp_ifidx >= 0
1549 ? odp_port_to_ofp_port(iface->dp_ifidx)
1551 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1558 /* Bridge packet processing functions. */
1561 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1563 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1566 static struct bond_entry *
1567 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1569 return &port->bond_hash[bond_hash(mac)];
1573 bond_choose_iface(const struct port *port)
1575 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1576 size_t i, best_down_slave = -1;
1577 long long next_delay_expiration = LLONG_MAX;
1579 for (i = 0; i < port->n_ifaces; i++) {
1580 struct iface *iface = port->ifaces[i];
1582 if (iface->enabled) {
1584 } else if (iface->delay_expires < next_delay_expiration) {
1585 best_down_slave = i;
1586 next_delay_expiration = iface->delay_expires;
1590 if (best_down_slave != -1) {
1591 struct iface *iface = port->ifaces[best_down_slave];
1593 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1594 "since no other interface is up", iface->name,
1595 iface->delay_expires - time_msec());
1596 bond_enable_slave(iface, true);
1599 return best_down_slave;
1603 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1604 uint16_t *dp_ifidx, tag_type *tags)
1606 struct iface *iface;
1608 assert(port->n_ifaces);
1609 if (port->n_ifaces == 1) {
1610 iface = port->ifaces[0];
1612 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1613 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1614 || !port->ifaces[e->iface_idx]->enabled) {
1615 /* XXX select interface properly. The current interface selection
1616 * is only good for testing the rebalancing code. */
1617 e->iface_idx = bond_choose_iface(port);
1618 if (e->iface_idx < 0) {
1619 *tags |= port->no_ifaces_tag;
1622 e->iface_tag = tag_create_random();
1623 ((struct port *) port)->bond_compat_is_stale = true;
1625 *tags |= e->iface_tag;
1626 iface = port->ifaces[e->iface_idx];
1628 *dp_ifidx = iface->dp_ifidx;
1629 *tags |= iface->tag; /* Currently only used for bonding. */
1634 bond_link_status_update(struct iface *iface, bool carrier)
1636 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1637 struct port *port = iface->port;
1639 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1640 /* Nothing to do. */
1643 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1644 iface->name, carrier ? "detected" : "dropped");
1645 if (carrier == iface->enabled) {
1646 iface->delay_expires = LLONG_MAX;
1647 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1648 iface->name, carrier ? "disabled" : "enabled");
1649 } else if (carrier && port->active_iface < 0) {
1650 bond_enable_slave(iface, true);
1651 if (port->updelay) {
1652 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1653 "other interface is up", iface->name, port->updelay);
1656 int delay = carrier ? port->updelay : port->downdelay;
1657 iface->delay_expires = time_msec() + delay;
1660 "interface %s: will be %s if it stays %s for %d ms",
1662 carrier ? "enabled" : "disabled",
1663 carrier ? "up" : "down",
1670 bond_choose_active_iface(struct port *port)
1672 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1674 port->active_iface = bond_choose_iface(port);
1675 port->active_iface_tag = tag_create_random();
1676 if (port->active_iface >= 0) {
1677 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1678 port->name, port->ifaces[port->active_iface]->name);
1680 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1686 bond_enable_slave(struct iface *iface, bool enable)
1688 struct port *port = iface->port;
1689 struct bridge *br = port->bridge;
1691 /* This acts as a recursion check. If the act of disabling a slave
1692 * causes a different slave to be enabled, the flag will allow us to
1693 * skip redundant work when we reenter this function. It must be
1694 * cleared on exit to keep things safe with multiple bonds. */
1695 static bool moving_active_iface = false;
1697 iface->delay_expires = LLONG_MAX;
1698 if (enable == iface->enabled) {
1702 iface->enabled = enable;
1703 if (!iface->enabled) {
1704 VLOG_WARN("interface %s: disabled", iface->name);
1705 ofproto_revalidate(br->ofproto, iface->tag);
1706 if (iface->port_ifidx == port->active_iface) {
1707 ofproto_revalidate(br->ofproto,
1708 port->active_iface_tag);
1710 /* Disabling a slave can lead to another slave being immediately
1711 * enabled if there will be no active slaves but one is waiting
1712 * on an updelay. In this case we do not need to run most of the
1713 * code for the newly enabled slave since there was no period
1714 * without an active slave and it is redundant with the disabling
1716 moving_active_iface = true;
1717 bond_choose_active_iface(port);
1719 bond_send_learning_packets(port);
1721 VLOG_WARN("interface %s: enabled", iface->name);
1722 if (port->active_iface < 0 && !moving_active_iface) {
1723 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1724 bond_choose_active_iface(port);
1725 bond_send_learning_packets(port);
1727 iface->tag = tag_create_random();
1730 moving_active_iface = false;
1731 port->bond_compat_is_stale = true;
1735 bond_run(struct bridge *br)
1739 for (i = 0; i < br->n_ports; i++) {
1740 struct port *port = br->ports[i];
1742 if (port->n_ifaces >= 2) {
1743 for (j = 0; j < port->n_ifaces; j++) {
1744 struct iface *iface = port->ifaces[j];
1745 if (time_msec() >= iface->delay_expires) {
1746 bond_enable_slave(iface, !iface->enabled);
1751 if (port->bond_compat_is_stale) {
1752 port->bond_compat_is_stale = false;
1753 port_update_bond_compat(port);
1759 bond_wait(struct bridge *br)
1763 for (i = 0; i < br->n_ports; i++) {
1764 struct port *port = br->ports[i];
1765 if (port->n_ifaces < 2) {
1768 for (j = 0; j < port->n_ifaces; j++) {
1769 struct iface *iface = port->ifaces[j];
1770 if (iface->delay_expires != LLONG_MAX) {
1771 poll_timer_wait(iface->delay_expires - time_msec());
1778 set_dst(struct dst *p, const flow_t *flow,
1779 const struct port *in_port, const struct port *out_port,
1782 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1783 : in_port->vlan >= 0 ? in_port->vlan
1784 : ntohs(flow->dl_vlan));
1785 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1789 swap_dst(struct dst *p, struct dst *q)
1791 struct dst tmp = *p;
1796 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1797 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1798 * that we push to the datapath. We could in fact fully sort the array by
1799 * vlan, but in most cases there are at most two different vlan tags so that's
1800 * possibly overkill.) */
1802 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1804 struct dst *first = dsts;
1805 struct dst *last = dsts + n_dsts;
1807 while (first != last) {
1809 * - All dsts < first have vlan == 'vlan'.
1810 * - All dsts >= last have vlan != 'vlan'.
1811 * - first < last. */
1812 while (first->vlan == vlan) {
1813 if (++first == last) {
1818 /* Same invariants, plus one additional:
1819 * - first->vlan != vlan.
1821 while (last[-1].vlan != vlan) {
1822 if (--last == first) {
1827 /* Same invariants, plus one additional:
1828 * - last[-1].vlan == vlan.*/
1829 swap_dst(first++, --last);
1834 mirror_mask_ffs(mirror_mask_t mask)
1836 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1841 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1842 const struct dst *test)
1845 for (i = 0; i < n_dsts; i++) {
1846 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1854 port_trunks_vlan(const struct port *port, uint16_t vlan)
1856 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1860 port_includes_vlan(const struct port *port, uint16_t vlan)
1862 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1866 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1867 const struct port *in_port, const struct port *out_port,
1868 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
1870 mirror_mask_t mirrors = in_port->src_mirrors;
1871 struct dst *dst = dsts;
1874 if (out_port == FLOOD_PORT) {
1875 /* XXX use ODP_FLOOD if no vlans or bonding. */
1876 /* XXX even better, define each VLAN as a datapath port group */
1877 for (i = 0; i < br->n_ports; i++) {
1878 struct port *port = br->ports[i];
1879 if (port != in_port && port_includes_vlan(port, vlan)
1880 && !port->is_mirror_output_port
1881 && set_dst(dst, flow, in_port, port, tags)) {
1882 mirrors |= port->dst_mirrors;
1886 *nf_output_iface = NF_OUT_FLOOD;
1887 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1888 *nf_output_iface = dst->dp_ifidx;
1889 mirrors |= out_port->dst_mirrors;
1894 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1895 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1897 if (set_dst(dst, flow, in_port, m->out_port, tags)
1898 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1902 for (i = 0; i < br->n_ports; i++) {
1903 struct port *port = br->ports[i];
1904 if (port_includes_vlan(port, m->out_vlan)
1905 && set_dst(dst, flow, in_port, port, tags))
1909 if (port->vlan < 0) {
1910 dst->vlan = m->out_vlan;
1912 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
1916 /* Use the vlan tag on the original flow instead of
1917 * the one passed in the vlan parameter. This ensures
1918 * that we compare the vlan from before any implicit
1919 * tagging tags place. This is necessary because
1920 * dst->vlan is the final vlan, after removing implicit
1922 flow_vlan = ntohs(flow->dl_vlan);
1923 if (flow_vlan == 0) {
1924 flow_vlan = OFP_VLAN_NONE;
1926 if (port == in_port && dst->vlan == flow_vlan) {
1927 /* Don't send out input port on same VLAN. */
1935 mirrors &= mirrors - 1;
1938 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1943 print_dsts(const struct dst *dsts, size_t n)
1945 for (; n--; dsts++) {
1946 printf(">p%"PRIu16, dsts->dp_ifidx);
1947 if (dsts->vlan != OFP_VLAN_NONE) {
1948 printf("v%"PRIu16, dsts->vlan);
1954 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1955 const struct port *in_port, const struct port *out_port,
1956 tag_type *tags, struct odp_actions *actions,
1957 uint16_t *nf_output_iface)
1959 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1961 const struct dst *p;
1964 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
1967 cur_vlan = ntohs(flow->dl_vlan);
1968 for (p = dsts; p < &dsts[n_dsts]; p++) {
1969 union odp_action *a;
1970 if (p->vlan != cur_vlan) {
1971 if (p->vlan == OFP_VLAN_NONE) {
1972 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1974 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1975 a->vlan_vid.vlan_vid = htons(p->vlan);
1979 a = odp_actions_add(actions, ODPAT_OUTPUT);
1980 a->output.port = p->dp_ifidx;
1984 /* Returns the effective vlan of a packet, taking into account both the
1985 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
1986 * the packet is untagged and -1 indicates it has an invalid header and
1987 * should be dropped. */
1988 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
1989 struct port *in_port, bool have_packet)
1991 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1992 * belongs to VLAN 0, so we should treat both cases identically. (In the
1993 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1994 * presumably to allow a priority to be specified. In the latter case, the
1995 * packet does not have any 802.1Q header.) */
1996 int vlan = ntohs(flow->dl_vlan);
1997 if (vlan == OFP_VLAN_NONE) {
2000 if (in_port->vlan >= 0) {
2002 /* XXX support double tagging? */
2004 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2005 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2006 "packet received on port %s configured with "
2007 "implicit VLAN %"PRIu16,
2008 br->name, ntohs(flow->dl_vlan),
2009 in_port->name, in_port->vlan);
2013 vlan = in_port->vlan;
2015 if (!port_includes_vlan(in_port, vlan)) {
2017 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2018 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2019 "packet received on port %s not configured for "
2021 br->name, vlan, in_port->name, vlan);
2031 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2032 struct port *in_port)
2034 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
2035 vlan, in_port->port_idx);
2037 /* The log messages here could actually be useful in debugging,
2038 * so keep the rate limit relatively high. */
2039 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2041 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2042 "on port %s in VLAN %d",
2043 br->name, ETH_ADDR_ARGS(flow->dl_src),
2044 in_port->name, vlan);
2045 ofproto_revalidate(br->ofproto, rev_tag);
2050 is_bcast_arp_reply(const flow_t *flow)
2052 return (flow->dl_type == htons(ETH_TYPE_ARP)
2053 && flow->nw_proto == ARP_OP_REPLY
2054 && eth_addr_is_broadcast(flow->dl_dst));
2057 /* If the composed actions may be applied to any packet in the given 'flow',
2058 * returns true. Otherwise, the actions should only be applied to 'packet', or
2059 * not at all, if 'packet' was NULL. */
2061 process_flow(struct bridge *br, const flow_t *flow,
2062 const struct ofpbuf *packet, struct odp_actions *actions,
2063 tag_type *tags, uint16_t *nf_output_iface)
2065 struct iface *in_iface;
2066 struct port *in_port;
2067 struct port *out_port = NULL; /* By default, drop the packet/flow. */
2071 /* Find the interface and port structure for the received packet. */
2072 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2074 /* No interface? Something fishy... */
2075 if (packet != NULL) {
2076 /* Odd. A few possible reasons here:
2078 * - We deleted an interface but there are still a few packets
2079 * queued up from it.
2081 * - Someone externally added an interface (e.g. with "ovs-dpctl
2082 * add-if") that we don't know about.
2084 * - Packet arrived on the local port but the local port is not
2085 * one of our bridge ports.
2087 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2089 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2090 "interface %"PRIu16, br->name, flow->in_port);
2093 /* Return without adding any actions, to drop packets on this flow. */
2096 in_port = in_iface->port;
2097 vlan = flow_get_vlan(br, flow, in_port, !!packet);
2102 /* Drop frames for reserved multicast addresses. */
2103 if (eth_addr_is_reserved(flow->dl_dst)) {
2107 /* Drop frames on ports reserved for mirroring. */
2108 if (in_port->is_mirror_output_port) {
2109 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2110 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
2111 "which is reserved exclusively for mirroring",
2112 br->name, in_port->name);
2116 /* Packets received on bonds need special attention to avoid duplicates. */
2117 if (in_port->n_ifaces > 1) {
2120 if (eth_addr_is_multicast(flow->dl_dst)) {
2121 *tags |= in_port->active_iface_tag;
2122 if (in_port->active_iface != in_iface->port_ifidx) {
2123 /* Drop all multicast packets on inactive slaves. */
2128 /* Drop all packets for which we have learned a different input
2129 * port, because we probably sent the packet on one slave and got
2130 * it back on the other. Broadcast ARP replies are an exception
2131 * to this rule: the host has moved to another switch. */
2132 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
2133 if (src_idx != -1 && src_idx != in_port->port_idx &&
2134 !is_bcast_arp_reply(flow)) {
2140 out_port = FLOOD_PORT;
2141 /* Learn source MAC (but don't try to learn from revalidation). */
2143 update_learning_table(br, flow, vlan, in_port);
2146 /* Determine output port. */
2147 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
2149 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2150 out_port = br->ports[out_port_idx];
2151 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2152 /* If we are revalidating but don't have a learning entry then
2153 * eject the flow. Installing a flow that floods packets opens
2154 * up a window of time where we could learn from a packet reflected
2155 * on a bond and blackhole packets before the learning table is
2156 * updated to reflect the correct port. */
2160 /* Don't send packets out their input ports. */
2161 if (in_port == out_port) {
2166 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2172 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2175 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2176 const struct ofp_phy_port *opp,
2179 struct bridge *br = br_;
2180 struct iface *iface;
2183 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2189 if (reason == OFPPR_DELETE) {
2190 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2191 br->name, iface->name);
2192 iface_destroy(iface);
2193 if (!port->n_ifaces) {
2194 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2195 br->name, port->name);
2201 if (port->n_ifaces > 1) {
2202 bool up = !(opp->state & OFPPS_LINK_DOWN);
2203 bond_link_status_update(iface, up);
2204 port_update_bond_compat(port);
2210 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2211 struct odp_actions *actions, tag_type *tags,
2212 uint16_t *nf_output_iface, void *br_)
2214 struct bridge *br = br_;
2216 COVERAGE_INC(bridge_process_flow);
2217 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2221 bridge_account_flow_ofhook_cb(const flow_t *flow,
2222 const union odp_action *actions,
2223 size_t n_actions, unsigned long long int n_bytes,
2226 struct bridge *br = br_;
2227 struct port *in_port;
2228 const union odp_action *a;
2230 /* Feed information from the active flows back into the learning table
2231 * to ensure that table is always in sync with what is actually flowing
2232 * through the datapath. */
2233 in_port = port_from_dp_ifidx(br, flow->in_port);
2235 int vlan = flow_get_vlan(br, flow, in_port, false);
2237 update_learning_table(br, flow, vlan, in_port);
2241 if (!br->has_bonded_ports) {
2245 for (a = actions; a < &actions[n_actions]; a++) {
2246 if (a->type == ODPAT_OUTPUT) {
2247 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2248 if (out_port && out_port->n_ifaces >= 2) {
2249 struct bond_entry *e = lookup_bond_entry(out_port,
2251 e->tx_bytes += n_bytes;
2258 bridge_account_checkpoint_ofhook_cb(void *br_)
2260 struct bridge *br = br_;
2263 if (!br->has_bonded_ports) {
2267 /* The current ofproto implementation calls this callback at least once a
2268 * second, so this timer implementation is sufficient. */
2269 if (time_msec() < br->bond_next_rebalance) {
2272 br->bond_next_rebalance = time_msec() + 10000;
2274 for (i = 0; i < br->n_ports; i++) {
2275 struct port *port = br->ports[i];
2276 if (port->n_ifaces > 1) {
2277 bond_rebalance_port(port);
2282 static struct ofhooks bridge_ofhooks = {
2283 bridge_port_changed_ofhook_cb,
2284 bridge_normal_ofhook_cb,
2285 bridge_account_flow_ofhook_cb,
2286 bridge_account_checkpoint_ofhook_cb,
2289 /* Bonding functions. */
2291 /* Statistics for a single interface on a bonded port, used for load-based
2292 * bond rebalancing. */
2293 struct slave_balance {
2294 struct iface *iface; /* The interface. */
2295 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2297 /* All the "bond_entry"s that are assigned to this interface, in order of
2298 * increasing tx_bytes. */
2299 struct bond_entry **hashes;
2303 /* Sorts pointers to pointers to bond_entries in ascending order by the
2304 * interface to which they are assigned, and within a single interface in
2305 * ascending order of bytes transmitted. */
2307 compare_bond_entries(const void *a_, const void *b_)
2309 const struct bond_entry *const *ap = a_;
2310 const struct bond_entry *const *bp = b_;
2311 const struct bond_entry *a = *ap;
2312 const struct bond_entry *b = *bp;
2313 if (a->iface_idx != b->iface_idx) {
2314 return a->iface_idx > b->iface_idx ? 1 : -1;
2315 } else if (a->tx_bytes != b->tx_bytes) {
2316 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2322 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2323 * *descending* order by number of bytes transmitted. */
2325 compare_slave_balance(const void *a_, const void *b_)
2327 const struct slave_balance *a = a_;
2328 const struct slave_balance *b = b_;
2329 if (a->iface->enabled != b->iface->enabled) {
2330 return a->iface->enabled ? -1 : 1;
2331 } else if (a->tx_bytes != b->tx_bytes) {
2332 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2339 swap_bals(struct slave_balance *a, struct slave_balance *b)
2341 struct slave_balance tmp = *a;
2346 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2347 * given that 'p' (and only 'p') might be in the wrong location.
2349 * This function invalidates 'p', since it might now be in a different memory
2352 resort_bals(struct slave_balance *p,
2353 struct slave_balance bals[], size_t n_bals)
2356 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2357 swap_bals(p, p - 1);
2359 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2360 swap_bals(p, p + 1);
2366 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2368 if (VLOG_IS_DBG_ENABLED()) {
2369 struct ds ds = DS_EMPTY_INITIALIZER;
2370 const struct slave_balance *b;
2372 for (b = bals; b < bals + n_bals; b++) {
2376 ds_put_char(&ds, ',');
2378 ds_put_format(&ds, " %s %"PRIu64"kB",
2379 b->iface->name, b->tx_bytes / 1024);
2381 if (!b->iface->enabled) {
2382 ds_put_cstr(&ds, " (disabled)");
2384 if (b->n_hashes > 0) {
2385 ds_put_cstr(&ds, " (");
2386 for (i = 0; i < b->n_hashes; i++) {
2387 const struct bond_entry *e = b->hashes[i];
2389 ds_put_cstr(&ds, " + ");
2391 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2392 e - port->bond_hash, e->tx_bytes / 1024);
2394 ds_put_cstr(&ds, ")");
2397 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2402 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2404 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2407 struct bond_entry *hash = from->hashes[hash_idx];
2408 struct port *port = from->iface->port;
2409 uint64_t delta = hash->tx_bytes;
2411 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2412 "from %s to %s (now carrying %"PRIu64"kB and "
2413 "%"PRIu64"kB load, respectively)",
2414 port->name, delta / 1024, hash - port->bond_hash,
2415 from->iface->name, to->iface->name,
2416 (from->tx_bytes - delta) / 1024,
2417 (to->tx_bytes + delta) / 1024);
2419 /* Delete element from from->hashes.
2421 * We don't bother to add the element to to->hashes because not only would
2422 * it require more work, the only purpose it would be to allow that hash to
2423 * be migrated to another slave in this rebalancing run, and there is no
2424 * point in doing that. */
2425 if (hash_idx == 0) {
2428 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2429 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2433 /* Shift load away from 'from' to 'to'. */
2434 from->tx_bytes -= delta;
2435 to->tx_bytes += delta;
2437 /* Arrange for flows to be revalidated. */
2438 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2439 hash->iface_idx = to->iface->port_ifidx;
2440 hash->iface_tag = tag_create_random();
2444 bond_rebalance_port(struct port *port)
2446 struct slave_balance bals[DP_MAX_PORTS];
2448 struct bond_entry *hashes[BOND_MASK + 1];
2449 struct slave_balance *b, *from, *to;
2450 struct bond_entry *e;
2453 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2454 * descending order of tx_bytes, so that bals[0] represents the most
2455 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2458 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2459 * array for each slave_balance structure, we sort our local array of
2460 * hashes in order by slave, so that all of the hashes for a given slave
2461 * become contiguous in memory, and then we point each 'hashes' members of
2462 * a slave_balance structure to the start of a contiguous group. */
2463 n_bals = port->n_ifaces;
2464 for (b = bals; b < &bals[n_bals]; b++) {
2465 b->iface = port->ifaces[b - bals];
2470 for (i = 0; i <= BOND_MASK; i++) {
2471 hashes[i] = &port->bond_hash[i];
2473 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2474 for (i = 0; i <= BOND_MASK; i++) {
2476 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2477 b = &bals[e->iface_idx];
2478 b->tx_bytes += e->tx_bytes;
2480 b->hashes = &hashes[i];
2485 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2486 log_bals(bals, n_bals, port);
2488 /* Discard slaves that aren't enabled (which were sorted to the back of the
2489 * array earlier). */
2490 while (!bals[n_bals - 1].iface->enabled) {
2497 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2498 to = &bals[n_bals - 1];
2499 for (from = bals; from < to; ) {
2500 uint64_t overload = from->tx_bytes - to->tx_bytes;
2501 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2502 /* The extra load on 'from' (and all less-loaded slaves), compared
2503 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2504 * it is less than ~1Mbps. No point in rebalancing. */
2506 } else if (from->n_hashes == 1) {
2507 /* 'from' only carries a single MAC hash, so we can't shift any
2508 * load away from it, even though we want to. */
2511 /* 'from' is carrying significantly more load than 'to', and that
2512 * load is split across at least two different hashes. Pick a hash
2513 * to migrate to 'to' (the least-loaded slave), given that doing so
2514 * must decrease the ratio of the load on the two slaves by at
2517 * The sort order we use means that we prefer to shift away the
2518 * smallest hashes instead of the biggest ones. There is little
2519 * reason behind this decision; we could use the opposite sort
2520 * order to shift away big hashes ahead of small ones. */
2524 for (i = 0; i < from->n_hashes; i++) {
2525 double old_ratio, new_ratio;
2526 uint64_t delta = from->hashes[i]->tx_bytes;
2528 if (delta == 0 || from->tx_bytes - delta == 0) {
2529 /* Pointless move. */
2533 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2535 if (to->tx_bytes == 0) {
2536 /* Nothing on the new slave, move it. */
2540 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2541 new_ratio = (double)(from->tx_bytes - delta) /
2542 (to->tx_bytes + delta);
2544 if (new_ratio == 0) {
2545 /* Should already be covered but check to prevent division
2550 if (new_ratio < 1) {
2551 new_ratio = 1 / new_ratio;
2554 if (old_ratio - new_ratio > 0.1) {
2555 /* Would decrease the ratio, move it. */
2559 if (i < from->n_hashes) {
2560 bond_shift_load(from, to, i);
2561 port->bond_compat_is_stale = true;
2563 /* If the result of the migration changed the relative order of
2564 * 'from' and 'to' swap them back to maintain invariants. */
2565 if (order_swapped) {
2566 swap_bals(from, to);
2569 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2570 * point to different slave_balance structures. It is only
2571 * valid to do these two operations in a row at all because we
2572 * know that 'from' will not move past 'to' and vice versa. */
2573 resort_bals(from, bals, n_bals);
2574 resort_bals(to, bals, n_bals);
2581 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2582 * historical data to decay to <1% in 7 rebalancing runs. */
2583 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2589 bond_send_learning_packets(struct port *port)
2591 struct bridge *br = port->bridge;
2592 struct mac_entry *e;
2593 struct ofpbuf packet;
2594 int error, n_packets, n_errors;
2596 if (!port->n_ifaces || port->active_iface < 0) {
2600 ofpbuf_init(&packet, 128);
2601 error = n_packets = n_errors = 0;
2602 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2603 union ofp_action actions[2], *a;
2609 if (e->port == port->port_idx
2610 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2614 /* Compose actions. */
2615 memset(actions, 0, sizeof actions);
2618 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2619 a->vlan_vid.len = htons(sizeof *a);
2620 a->vlan_vid.vlan_vid = htons(e->vlan);
2623 a->output.type = htons(OFPAT_OUTPUT);
2624 a->output.len = htons(sizeof *a);
2625 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2630 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2632 flow_extract(&packet, ODPP_NONE, &flow);
2633 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2640 ofpbuf_uninit(&packet);
2643 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2644 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2645 "packets, last error was: %s",
2646 port->name, n_errors, n_packets, strerror(error));
2648 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2649 port->name, n_packets);
2653 /* Bonding unixctl user interface functions. */
2656 bond_unixctl_list(struct unixctl_conn *conn,
2657 const char *args UNUSED, void *aux UNUSED)
2659 struct ds ds = DS_EMPTY_INITIALIZER;
2660 const struct bridge *br;
2662 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2664 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2667 for (i = 0; i < br->n_ports; i++) {
2668 const struct port *port = br->ports[i];
2669 if (port->n_ifaces > 1) {
2672 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2673 for (j = 0; j < port->n_ifaces; j++) {
2674 const struct iface *iface = port->ifaces[j];
2676 ds_put_cstr(&ds, ", ");
2678 ds_put_cstr(&ds, iface->name);
2680 ds_put_char(&ds, '\n');
2684 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2688 static struct port *
2689 bond_find(const char *name)
2691 const struct bridge *br;
2693 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2696 for (i = 0; i < br->n_ports; i++) {
2697 struct port *port = br->ports[i];
2698 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2707 bond_unixctl_show(struct unixctl_conn *conn,
2708 const char *args, void *aux UNUSED)
2710 struct ds ds = DS_EMPTY_INITIALIZER;
2711 const struct port *port;
2714 port = bond_find(args);
2716 unixctl_command_reply(conn, 501, "no such bond");
2720 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2721 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2722 ds_put_format(&ds, "next rebalance: %lld ms\n",
2723 port->bridge->bond_next_rebalance - time_msec());
2724 for (j = 0; j < port->n_ifaces; j++) {
2725 const struct iface *iface = port->ifaces[j];
2726 struct bond_entry *be;
2729 ds_put_format(&ds, "slave %s: %s\n",
2730 iface->name, iface->enabled ? "enabled" : "disabled");
2731 if (j == port->active_iface) {
2732 ds_put_cstr(&ds, "\tactive slave\n");
2734 if (iface->delay_expires != LLONG_MAX) {
2735 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2736 iface->enabled ? "downdelay" : "updelay",
2737 iface->delay_expires - time_msec());
2741 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2742 int hash = be - port->bond_hash;
2743 struct mac_entry *me;
2745 if (be->iface_idx != j) {
2749 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2750 hash, be->tx_bytes / 1024);
2753 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2754 &port->bridge->ml->lrus) {
2757 if (bond_hash(me->mac) == hash
2758 && me->port != port->port_idx
2759 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2760 && dp_ifidx == iface->dp_ifidx)
2762 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2763 ETH_ADDR_ARGS(me->mac));
2768 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2773 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2776 char *args = (char *) args_;
2777 char *save_ptr = NULL;
2778 char *bond_s, *hash_s, *slave_s;
2779 uint8_t mac[ETH_ADDR_LEN];
2781 struct iface *iface;
2782 struct bond_entry *entry;
2785 bond_s = strtok_r(args, " ", &save_ptr);
2786 hash_s = strtok_r(NULL, " ", &save_ptr);
2787 slave_s = strtok_r(NULL, " ", &save_ptr);
2789 unixctl_command_reply(conn, 501,
2790 "usage: bond/migrate BOND HASH SLAVE");
2794 port = bond_find(bond_s);
2796 unixctl_command_reply(conn, 501, "no such bond");
2800 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2801 == ETH_ADDR_SCAN_COUNT) {
2802 hash = bond_hash(mac);
2803 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2804 hash = atoi(hash_s) & BOND_MASK;
2806 unixctl_command_reply(conn, 501, "bad hash");
2810 iface = port_lookup_iface(port, slave_s);
2812 unixctl_command_reply(conn, 501, "no such slave");
2816 if (!iface->enabled) {
2817 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2821 entry = &port->bond_hash[hash];
2822 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2823 entry->iface_idx = iface->port_ifidx;
2824 entry->iface_tag = tag_create_random();
2825 port->bond_compat_is_stale = true;
2826 unixctl_command_reply(conn, 200, "migrated");
2830 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
2833 char *args = (char *) args_;
2834 char *save_ptr = NULL;
2835 char *bond_s, *slave_s;
2837 struct iface *iface;
2839 bond_s = strtok_r(args, " ", &save_ptr);
2840 slave_s = strtok_r(NULL, " ", &save_ptr);
2842 unixctl_command_reply(conn, 501,
2843 "usage: bond/set-active-slave BOND SLAVE");
2847 port = bond_find(bond_s);
2849 unixctl_command_reply(conn, 501, "no such bond");
2853 iface = port_lookup_iface(port, slave_s);
2855 unixctl_command_reply(conn, 501, "no such slave");
2859 if (!iface->enabled) {
2860 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2864 if (port->active_iface != iface->port_ifidx) {
2865 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2866 port->active_iface = iface->port_ifidx;
2867 port->active_iface_tag = tag_create_random();
2868 VLOG_INFO("port %s: active interface is now %s",
2869 port->name, iface->name);
2870 bond_send_learning_packets(port);
2871 unixctl_command_reply(conn, 200, "done");
2873 unixctl_command_reply(conn, 200, "no change");
2878 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2880 char *args = (char *) args_;
2881 char *save_ptr = NULL;
2882 char *bond_s, *slave_s;
2884 struct iface *iface;
2886 bond_s = strtok_r(args, " ", &save_ptr);
2887 slave_s = strtok_r(NULL, " ", &save_ptr);
2889 unixctl_command_reply(conn, 501,
2890 "usage: bond/enable/disable-slave BOND SLAVE");
2894 port = bond_find(bond_s);
2896 unixctl_command_reply(conn, 501, "no such bond");
2900 iface = port_lookup_iface(port, slave_s);
2902 unixctl_command_reply(conn, 501, "no such slave");
2906 bond_enable_slave(iface, enable);
2907 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2911 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
2914 enable_slave(conn, args, true);
2918 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
2921 enable_slave(conn, args, false);
2925 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
2928 uint8_t mac[ETH_ADDR_LEN];
2932 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2933 == ETH_ADDR_SCAN_COUNT) {
2934 hash = bond_hash(mac);
2936 hash_cstr = xasprintf("%u", hash);
2937 unixctl_command_reply(conn, 200, hash_cstr);
2940 unixctl_command_reply(conn, 501, "invalid mac");
2947 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
2948 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
2949 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
2950 unixctl_command_register("bond/set-active-slave",
2951 bond_unixctl_set_active_slave, NULL);
2952 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
2954 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
2956 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
2959 /* Port functions. */
2961 static struct port *
2962 port_create(struct bridge *br, const char *name)
2966 port = xzalloc(sizeof *port);
2968 port->port_idx = br->n_ports;
2970 port->trunks = NULL;
2971 port->name = xstrdup(name);
2972 port->active_iface = -1;
2974 if (br->n_ports >= br->allocated_ports) {
2975 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2978 br->ports[br->n_ports++] = port;
2980 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2987 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
2989 struct shash old_ifaces, new_ifaces;
2990 struct shash_node *node;
2991 unsigned long *trunks;
2997 /* Collect old and new interfaces. */
2998 shash_init(&old_ifaces);
2999 shash_init(&new_ifaces);
3000 for (i = 0; i < port->n_ifaces; i++) {
3001 shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]);
3003 for (i = 0; i < cfg->n_interfaces; i++) {
3004 const char *name = cfg->interfaces[i]->name;
3005 if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) {
3006 VLOG_WARN("port %s: %s specified twice as port interface",
3010 port->updelay = cfg->bond_updelay;
3011 if (port->updelay < 0) {
3014 port->updelay = cfg->bond_downdelay;
3015 if (port->downdelay < 0) {
3016 port->downdelay = 0;
3019 /* Get rid of deleted interfaces and add new interfaces. */
3020 SHASH_FOR_EACH (node, &old_ifaces) {
3021 if (!shash_find(&new_ifaces, node->name)) {
3022 iface_destroy(node->data);
3025 SHASH_FOR_EACH (node, &new_ifaces) {
3026 const struct ovsrec_interface *if_cfg = node->data;
3027 struct iface *iface;
3029 iface = shash_find_data(&old_ifaces, if_cfg->name);
3031 iface = iface_create(port, if_cfg);
3033 iface->cfg = if_cfg;
3040 if (port->n_ifaces < 2) {
3042 if (vlan >= 0 && vlan <= 4095) {
3043 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3048 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3049 * they even work as-is. But they have not been tested. */
3050 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3054 if (port->vlan != vlan) {
3056 bridge_flush(port->bridge);
3059 /* Get trunked VLANs. */
3065 trunks = bitmap_allocate(4096);
3067 for (i = 0; i < cfg->n_trunks; i++) {
3068 int trunk = cfg->trunks[i];
3070 bitmap_set1(trunks, trunk);
3076 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3077 port->name, cfg->n_trunks);
3079 if (n_errors == cfg->n_trunks) {
3081 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3084 bitmap_set_multiple(trunks, 0, 4096, 1);
3087 if (cfg->n_trunks) {
3088 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3093 ? port->trunks != NULL
3094 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3095 bridge_flush(port->bridge);
3097 bitmap_free(port->trunks);
3098 port->trunks = trunks;
3100 shash_destroy(&old_ifaces);
3101 shash_destroy(&new_ifaces);
3105 port_destroy(struct port *port)
3108 struct bridge *br = port->bridge;
3112 proc_net_compat_update_vlan(port->name, NULL, 0);
3113 proc_net_compat_update_bond(port->name, NULL);
3115 for (i = 0; i < MAX_MIRRORS; i++) {
3116 struct mirror *m = br->mirrors[i];
3117 if (m && m->out_port == port) {
3122 while (port->n_ifaces > 0) {
3123 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3126 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3127 del->port_idx = port->port_idx;
3130 bitmap_free(port->trunks);
3137 static struct port *
3138 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3140 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3141 return iface ? iface->port : NULL;
3144 static struct port *
3145 port_lookup(const struct bridge *br, const char *name)
3149 for (i = 0; i < br->n_ports; i++) {
3150 struct port *port = br->ports[i];
3151 if (!strcmp(port->name, name)) {
3158 static struct iface *
3159 port_lookup_iface(const struct port *port, const char *name)
3163 for (j = 0; j < port->n_ifaces; j++) {
3164 struct iface *iface = port->ifaces[j];
3165 if (!strcmp(iface->name, name)) {
3173 port_update_bonding(struct port *port)
3175 if (port->n_ifaces < 2) {
3176 /* Not a bonded port. */
3177 if (port->bond_hash) {
3178 free(port->bond_hash);
3179 port->bond_hash = NULL;
3180 port->bond_compat_is_stale = true;
3181 port->bond_fake_iface = false;
3184 if (!port->bond_hash) {
3187 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3188 for (i = 0; i <= BOND_MASK; i++) {
3189 struct bond_entry *e = &port->bond_hash[i];
3193 port->no_ifaces_tag = tag_create_random();
3194 bond_choose_active_iface(port);
3196 port->bond_compat_is_stale = true;
3197 port->bond_fake_iface = port->cfg->bond_fake_iface;
3202 port_update_bond_compat(struct port *port)
3204 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3205 struct compat_bond bond;
3208 if (port->n_ifaces < 2) {
3209 proc_net_compat_update_bond(port->name, NULL);
3214 bond.updelay = port->updelay;
3215 bond.downdelay = port->downdelay;
3218 bond.hashes = compat_hashes;
3219 if (port->bond_hash) {
3220 const struct bond_entry *e;
3221 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3222 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3223 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3224 cbh->hash = e - port->bond_hash;
3225 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3230 bond.n_slaves = port->n_ifaces;
3231 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3232 for (i = 0; i < port->n_ifaces; i++) {
3233 struct iface *iface = port->ifaces[i];
3234 struct compat_bond_slave *slave = &bond.slaves[i];
3235 slave->name = iface->name;
3237 /* We need to make the same determination as the Linux bonding
3238 * code to determine whether a slave should be consider "up".
3239 * The Linux function bond_miimon_inspect() supports four
3240 * BOND_LINK_* states:
3242 * - BOND_LINK_UP: carrier detected, updelay has passed.
3243 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3244 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3245 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3247 * The function bond_info_show_slave() only considers BOND_LINK_UP
3248 * to be "up" and anything else to be "down".
3250 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3254 netdev_get_etheraddr(iface->netdev, slave->mac);
3257 if (port->bond_fake_iface) {
3258 struct netdev *bond_netdev;
3260 if (!netdev_open_default(port->name, &bond_netdev)) {
3262 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3264 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3266 netdev_close(bond_netdev);
3270 proc_net_compat_update_bond(port->name, &bond);
3275 port_update_vlan_compat(struct port *port)
3277 struct bridge *br = port->bridge;
3278 char *vlandev_name = NULL;
3280 if (port->vlan > 0) {
3281 /* Figure out the name that the VLAN device should actually have, if it
3282 * existed. This takes some work because the VLAN device would not
3283 * have port->name in its name; rather, it would have the trunk port's
3284 * name, and 'port' would be attached to a bridge that also had the
3285 * VLAN device one of its ports. So we need to find a trunk port that
3286 * includes port->vlan.
3288 * There might be more than one candidate. This doesn't happen on
3289 * XenServer, so if it happens we just pick the first choice in
3290 * alphabetical order instead of creating multiple VLAN devices. */
3292 for (i = 0; i < br->n_ports; i++) {
3293 struct port *p = br->ports[i];
3294 if (port_trunks_vlan(p, port->vlan)
3296 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3298 uint8_t ea[ETH_ADDR_LEN];
3299 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3300 if (!eth_addr_is_multicast(ea) &&
3301 !eth_addr_is_reserved(ea) &&
3302 !eth_addr_is_zero(ea)) {
3303 vlandev_name = p->name;
3308 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3311 /* Interface functions. */
3313 static struct iface *
3314 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3316 struct iface *iface;
3317 char *name = if_cfg->name;
3320 iface = xzalloc(sizeof *iface);
3322 iface->port_ifidx = port->n_ifaces;
3323 iface->name = xstrdup(name);
3324 iface->dp_ifidx = -1;
3325 iface->tag = tag_create_random();
3326 iface->delay_expires = LLONG_MAX;
3327 iface->netdev = NULL;
3328 iface->cfg = if_cfg;
3330 if (port->n_ifaces >= port->allocated_ifaces) {
3331 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3332 sizeof *port->ifaces);
3334 port->ifaces[port->n_ifaces++] = iface;
3335 if (port->n_ifaces > 1) {
3336 port->bridge->has_bonded_ports = true;
3339 /* Attempt to create the network interface in case it
3340 * doesn't exist yet. */
3341 if (!iface_is_internal(port->bridge, iface->name)) {
3342 error = set_up_iface(if_cfg, iface, true);
3344 VLOG_WARN("could not create iface %s: %s", iface->name,
3349 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3351 bridge_flush(port->bridge);
3357 iface_destroy(struct iface *iface)
3360 struct port *port = iface->port;
3361 struct bridge *br = port->bridge;
3362 bool del_active = port->active_iface == iface->port_ifidx;
3365 if (iface->dp_ifidx >= 0) {
3366 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3369 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3370 del->port_ifidx = iface->port_ifidx;
3372 netdev_close(iface->netdev);
3375 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3376 bond_choose_active_iface(port);
3377 bond_send_learning_packets(port);
3383 bridge_flush(port->bridge);
3387 static struct iface *
3388 iface_lookup(const struct bridge *br, const char *name)
3392 for (i = 0; i < br->n_ports; i++) {
3393 struct port *port = br->ports[i];
3394 for (j = 0; j < port->n_ifaces; j++) {
3395 struct iface *iface = port->ifaces[j];
3396 if (!strcmp(iface->name, name)) {
3404 static struct iface *
3405 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3407 return port_array_get(&br->ifaces, dp_ifidx);
3410 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3411 * 'br', that is, an interface that is entirely simulated within the datapath.
3412 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3413 * interfaces are created by setting "iface.<iface>.internal = true".
3415 * In addition, we have a kluge-y feature that creates an internal port with
3416 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3417 * This feature needs to go away in the long term. Until then, this is one
3418 * reason why this function takes a name instead of a struct iface: the fake
3419 * interfaces created this way do not have a struct iface. */
3421 iface_is_internal(const struct bridge *br, const char *if_name)
3423 /* XXX wastes time */
3424 struct iface *iface;
3427 if (!strcmp(if_name, br->name)) {
3431 iface = iface_lookup(br, if_name);
3432 if (iface && !strcmp(iface->cfg->type, "internal")) {
3436 port = port_lookup(br, if_name);
3437 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3443 /* Set Ethernet address of 'iface', if one is specified in the configuration
3446 iface_set_mac(struct iface *iface)
3448 uint8_t ea[ETH_ADDR_LEN];
3450 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3451 if (eth_addr_is_multicast(ea)) {
3452 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3454 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3455 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3456 iface->name, iface->name);
3458 int error = netdev_set_etheraddr(iface->netdev, ea);
3460 VLOG_ERR("interface %s: setting MAC failed (%s)",
3461 iface->name, strerror(error));
3467 /* Port mirroring. */
3470 mirror_reconfigure(struct bridge *br)
3472 struct shash old_mirrors, new_mirrors;
3473 struct shash_node *node;
3474 unsigned long *rspan_vlans;
3477 /* Collect old mirrors. */
3478 shash_init(&old_mirrors);
3479 for (i = 0; i < MAX_MIRRORS; i++) {
3480 if (br->mirrors[i]) {
3481 shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
3485 /* Collect new mirrors. */
3486 shash_init(&new_mirrors);
3487 for (i = 0; i < br->cfg->n_mirrors; i++) {
3488 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3489 if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
3490 VLOG_WARN("bridge %s: %s specified twice as mirror",
3491 br->name, cfg->name);
3495 /* Get rid of deleted mirrors and add new mirrors. */
3496 SHASH_FOR_EACH (node, &old_mirrors) {
3497 if (!shash_find(&new_mirrors, node->name)) {
3498 mirror_destroy(node->data);
3501 SHASH_FOR_EACH (node, &new_mirrors) {
3502 struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
3504 mirror = mirror_create(br, node->name);
3509 mirror_reconfigure_one(mirror, node->data);
3511 shash_destroy(&old_mirrors);
3512 shash_destroy(&new_mirrors);
3514 /* Update port reserved status. */
3515 for (i = 0; i < br->n_ports; i++) {
3516 br->ports[i]->is_mirror_output_port = false;
3518 for (i = 0; i < MAX_MIRRORS; i++) {
3519 struct mirror *m = br->mirrors[i];
3520 if (m && m->out_port) {
3521 m->out_port->is_mirror_output_port = true;
3525 /* Update learning disabled vlans (for RSPAN). */
3527 if (br->cfg->n_flood_vlans) {
3528 rspan_vlans = bitmap_allocate(4096);
3530 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3531 int64_t vlan = br->cfg->flood_vlans[i];
3532 if (vlan >= 0 && vlan < 4096) {
3533 bitmap_set1(rspan_vlans, vlan);
3534 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3537 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3542 if (mac_learning_set_disabled_vlans(br->ml, rspan_vlans)) {
3547 static struct mirror *
3548 mirror_create(struct bridge *br, const char *name)
3553 for (i = 0; ; i++) {
3554 if (i >= MAX_MIRRORS) {
3555 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3556 "cannot create %s", br->name, MAX_MIRRORS, name);
3559 if (!br->mirrors[i]) {
3564 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3567 br->mirrors[i] = m = xzalloc(sizeof *m);
3570 m->name = xstrdup(name);
3571 shash_init(&m->src_ports);
3572 shash_init(&m->dst_ports);
3582 mirror_destroy(struct mirror *m)
3585 struct bridge *br = m->bridge;
3588 for (i = 0; i < br->n_ports; i++) {
3589 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3590 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3593 shash_destroy(&m->src_ports);
3594 shash_destroy(&m->dst_ports);
3597 m->bridge->mirrors[m->idx] = NULL;
3605 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
3606 struct shash *names)
3610 for (i = 0; i < n_ports; i++) {
3611 const char *name = ports[i]->name;
3612 if (port_lookup(m->bridge, name)) {
3613 shash_add_once(names, name, NULL);
3615 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
3616 "port %s", m->bridge->name, m->name, name);
3622 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
3628 *vlans = xmalloc(sizeof *vlans * cfg->n_select_vlan);
3630 for (i = 0; i < cfg->n_select_vlan; i++) {
3631 int64_t vlan = cfg->select_vlan[i];
3632 if (vlan < 0 || vlan > 4095) {
3633 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
3634 m->bridge->name, m->name, vlan);
3636 (*vlans)[n_vlans++] = vlan;
3643 vlan_is_mirrored(const struct mirror *m, int vlan)
3647 for (i = 0; i < m->n_vlans; i++) {
3648 if (m->vlans[i] == vlan) {
3656 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3660 for (i = 0; i < m->n_vlans; i++) {
3661 if (port_trunks_vlan(p, m->vlans[i])) {
3669 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
3671 struct shash src_ports, dst_ports;
3672 mirror_mask_t mirror_bit;
3673 struct port *out_port;
3678 bool mirror_all_ports;
3679 bool any_ports_specified;
3680 bool any_vlans_specified;
3682 /* Get output port. */
3683 if (cfg->output_port) {
3684 out_port = port_lookup(m->bridge, cfg->output_port->name);
3686 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
3687 m->bridge->name, m->name);
3693 if (cfg->output_vlan) {
3694 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
3695 "output vlan; ignoring output vlan",
3696 m->bridge->name, m->name);
3698 } else if (cfg->output_vlan) {
3700 out_vlan = *cfg->output_vlan;
3702 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
3703 m->bridge->name, m->name);
3708 /* Get all the ports, and drop duplicates and ports that don't exist. */
3709 shash_init(&src_ports);
3710 shash_init(&dst_ports);
3711 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
3713 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
3715 any_ports_specified = cfg->n_select_dst_port || cfg->n_select_dst_port;
3716 if (any_ports_specified
3717 && shash_is_empty(&src_ports) && shash_is_empty(&dst_ports)) {
3718 VLOG_ERR("bridge %s: disabling mirror %s since none of the specified "
3719 "selection ports exists", m->bridge->name, m->name);
3724 /* Get all the vlans, and drop duplicate and invalid vlans. */
3725 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
3726 any_vlans_specified = cfg->n_select_vlan > 0;
3727 if (any_vlans_specified && !n_vlans) {
3728 VLOG_ERR("bridge %s: disabling mirror %s since none of the specified "
3729 "VLANs exists", m->bridge->name, m->name);
3734 /* Update mirror data. */
3735 if (!shash_equal_keys(&m->src_ports, &src_ports)
3736 || !shash_equal_keys(&m->dst_ports, &dst_ports)
3737 || m->n_vlans != n_vlans
3738 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3739 || m->out_port != out_port
3740 || m->out_vlan != out_vlan) {
3741 bridge_flush(m->bridge);
3743 shash_swap(&m->src_ports, &src_ports);
3744 shash_swap(&m->dst_ports, &dst_ports);
3747 m->n_vlans = n_vlans;
3748 m->out_port = out_port;
3749 m->out_vlan = out_vlan;
3751 /* If no selection criteria have been given, mirror for all ports. */
3752 mirror_all_ports = !any_ports_specified && !any_vlans_specified;
3755 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3756 for (i = 0; i < m->bridge->n_ports; i++) {
3757 struct port *port = m->bridge->ports[i];
3759 if (mirror_all_ports
3760 || shash_find(&m->src_ports, port->name)
3763 ? port_trunks_any_mirrored_vlan(m, port)
3764 : vlan_is_mirrored(m, port->vlan)))) {
3765 port->src_mirrors |= mirror_bit;
3767 port->src_mirrors &= ~mirror_bit;
3770 if (mirror_all_ports || shash_find(&m->dst_ports, port->name)) {
3771 port->dst_mirrors |= mirror_bit;
3773 port->dst_mirrors &= ~mirror_bit;
3779 shash_destroy(&src_ports);
3780 shash_destroy(&dst_ports);