1 /* Copyright (c) 2008, 2009 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"
60 #include "vconn-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 svec src_ports;
109 struct svec dst_ports;
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 *);
238 static void mirror_create(struct bridge *, const char *name);
239 static void mirror_destroy(struct mirror *);
240 static void mirror_reconfigure(struct bridge *);
241 static void mirror_reconfigure_one(struct mirror *);
242 static bool vlan_is_mirrored(const struct mirror *, int vlan);
244 static bool vlan_is_mirrored(const struct mirror *m UNUSED, int vlan UNUSED)
250 static struct iface *iface_create(struct port *port,
251 const struct ovsrec_interface *if_cfg);
252 static void iface_destroy(struct iface *);
253 static struct iface *iface_lookup(const struct bridge *, const char *name);
254 static struct iface *iface_from_dp_ifidx(const struct bridge *,
256 static bool iface_is_internal(const struct bridge *, const char *name);
257 static void iface_set_mac(struct iface *);
259 /* Hooks into ofproto processing. */
260 static struct ofhooks bridge_ofhooks;
262 /* Public functions. */
264 /* Adds the name of each interface used by a bridge, including local and
265 * internal ports, to 'svec'. */
267 bridge_get_ifaces(struct svec *svec)
269 struct bridge *br, *next;
272 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
273 for (i = 0; i < br->n_ports; i++) {
274 struct port *port = br->ports[i];
276 for (j = 0; j < port->n_ifaces; j++) {
277 struct iface *iface = port->ifaces[j];
278 if (iface->dp_ifidx < 0) {
279 VLOG_ERR("%s interface not in datapath %s, ignoring",
280 iface->name, dpif_name(br->dpif));
282 if (iface->dp_ifidx != ODPP_LOCAL) {
283 svec_add(svec, iface->name);
292 bridge_init(const struct ovsrec_open_vswitch *cfg)
294 struct svec bridge_names;
295 struct svec dpif_names;
298 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
300 svec_init(&bridge_names);
301 for (i = 0; i < cfg->n_bridges; i++) {
302 svec_add(&bridge_names, cfg->bridges[i]->name);
304 svec_sort(&bridge_names);
306 svec_init(&dpif_names);
307 dp_enumerate(&dpif_names);
308 for (i = 0; i < dpif_names.n; i++) {
309 const char *dpif_name = dpif_names.names[i];
313 retval = dpif_open(dpif_name, &dpif);
315 struct svec all_names;
318 svec_init(&all_names);
319 dpif_get_all_names(dpif, &all_names);
320 for (j = 0; j < all_names.n; j++) {
321 if (svec_contains(&bridge_names, all_names.names[j])) {
327 svec_destroy(&all_names);
331 svec_destroy(&dpif_names);
333 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
337 bridge_reconfigure(cfg);
342 config_string_change(const char *value, char **valuep)
344 if (value && (!*valuep || strcmp(value, *valuep))) {
346 *valuep = xstrdup(value);
354 bridge_configure_ssl(const struct ovsrec_ssl *ssl)
356 /* XXX SSL should be configurable on a per-bridge basis.
357 * XXX should be possible to de-configure SSL. */
358 static char *private_key_file;
359 static char *certificate_file;
360 static char *cacert_file;
364 /* XXX We can't un-set SSL settings. */
368 if (config_string_change(ssl->private_key, &private_key_file)) {
369 vconn_ssl_set_private_key_file(private_key_file);
372 if (config_string_change(ssl->certificate, &certificate_file)) {
373 vconn_ssl_set_certificate_file(certificate_file);
376 /* We assume that even if the filename hasn't changed, if the CA cert
377 * file has been removed, that we want to move back into
378 * boot-strapping mode. This opens a small security hole, because
379 * the old certificate will still be trusted until vSwitch is
380 * restarted. We may want to address this in vconn's SSL library. */
381 if (config_string_change(ssl->ca_cert, &cacert_file)
382 || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) {
383 vconn_ssl_set_ca_cert_file(cacert_file, ssl->bootstrap_ca_cert);
388 /* Attempt to create the network device 'iface_name' through the netdev
391 set_up_iface(const struct ovsrec_interface *iface_cfg, bool create)
393 struct shash_node *node;
394 struct shash options;
398 /* If a type is not explicitly declared, then assume it's an existing
399 * "system" device. */
400 if (iface_cfg->type[0] == '\0' || !strcmp(iface_cfg->type, "system")) {
404 shash_init(&options);
405 for (i = 0; i < iface_cfg->n_options; i++) {
406 shash_add(&options, iface_cfg->key_options[i],
407 xstrdup(iface_cfg->value_options[i]));
411 error = netdev_create(iface_cfg->name, iface_cfg->type, &options);
413 /* xxx Check to make sure that the type hasn't changed. */
414 error = netdev_reconfigure(iface_cfg->name, &options);
417 SHASH_FOR_EACH (node, &options) {
420 shash_destroy(&options);
426 reconfigure_iface(const struct ovsrec_interface *iface_cfg)
428 return set_up_iface(iface_cfg, false);
432 /* iterate_and_prune_ifaces() callback function that opens the network device
433 * for 'iface', if it is not already open, and retrieves the interface's MAC
434 * address and carrier status. */
436 init_iface_netdev(struct bridge *br UNUSED, struct iface *iface,
441 } else if (!netdev_open(iface->name, NETDEV_ETH_TYPE_NONE,
443 netdev_get_carrier(iface->netdev, &iface->enabled);
446 /* If the network device can't be opened, then we're not going to try
447 * to do anything with this interface. */
453 check_iface_dp_ifidx(struct bridge *br, struct iface *iface, void *aux UNUSED)
455 if (iface->dp_ifidx >= 0) {
456 VLOG_DBG("%s has interface %s on port %d",
458 iface->name, iface->dp_ifidx);
461 VLOG_ERR("%s interface not in %s, dropping",
462 iface->name, dpif_name(br->dpif));
468 set_iface_properties(struct bridge *br UNUSED, struct iface *iface,
471 /* Set policing attributes. */
472 netdev_set_policing(iface->netdev,
473 iface->cfg->ingress_policing_rate,
474 iface->cfg->ingress_policing_burst);
476 /* Set MAC address of internal interfaces other than the local
478 if (iface->dp_ifidx != ODPP_LOCAL
479 && iface_is_internal(br, iface->name)) {
480 iface_set_mac(iface);
486 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
487 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
488 * deletes from 'br' any ports that no longer have any interfaces. */
490 iterate_and_prune_ifaces(struct bridge *br,
491 bool (*cb)(struct bridge *, struct iface *,
497 for (i = 0; i < br->n_ports; ) {
498 struct port *port = br->ports[i];
499 for (j = 0; j < port->n_ifaces; ) {
500 struct iface *iface = port->ifaces[j];
501 if (cb(br, iface, aux)) {
504 iface_destroy(iface);
508 if (port->n_ifaces) {
511 VLOG_ERR("%s port has no interfaces, dropping", port->name);
518 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
520 struct ovsdb_idl_txn *txn;
521 struct shash old_br, new_br;
522 struct shash_node *node;
523 struct bridge *br, *next;
526 COVERAGE_INC(bridge_reconfigure);
528 txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
530 /* Collect old and new bridges. */
533 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
534 shash_add(&old_br, br->name, br);
536 for (i = 0; i < ovs_cfg->n_bridges; i++) {
537 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
538 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
539 VLOG_WARN("more than one bridge named %s", br_cfg->name);
543 /* Get rid of deleted bridges and add new bridges. */
544 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
545 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
552 SHASH_FOR_EACH (node, &new_br) {
553 const char *br_name = node->name;
554 const struct ovsrec_bridge *br_cfg = node->data;
555 if (!shash_find_data(&old_br, br_name)) {
556 br = bridge_create(br_name);
562 shash_destroy(&old_br);
563 shash_destroy(&new_br);
567 bridge_configure_ssl(ovs_cfg->ssl);
570 /* Reconfigure all bridges. */
571 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
572 bridge_reconfigure_one(ovs_cfg, br);
575 /* Add and delete ports on all datapaths.
577 * The kernel will reject any attempt to add a given port to a datapath if
578 * that port already belongs to a different datapath, so we must do all
579 * port deletions before any port additions. */
580 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
581 struct odp_port *dpif_ports;
583 struct shash want_ifaces;
585 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
586 bridge_get_all_ifaces(br, &want_ifaces);
587 for (i = 0; i < n_dpif_ports; i++) {
588 const struct odp_port *p = &dpif_ports[i];
589 if (!shash_find(&want_ifaces, p->devname)
590 && strcmp(p->devname, br->name)) {
591 int retval = dpif_port_del(br->dpif, p->port);
593 VLOG_ERR("failed to remove %s interface from %s: %s",
594 p->devname, dpif_name(br->dpif),
599 shash_destroy(&want_ifaces);
602 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
603 struct odp_port *dpif_ports;
605 struct shash cur_ifaces, want_ifaces;
606 struct shash_node *node;
608 /* Get the set of interfaces currently in this datapath. */
609 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
610 shash_init(&cur_ifaces);
611 for (i = 0; i < n_dpif_ports; i++) {
612 const char *name = dpif_ports[i].devname;
613 if (!shash_find(&cur_ifaces, name)) {
614 shash_add(&cur_ifaces, name, NULL);
619 /* Get the set of interfaces we want on this datapath. */
620 bridge_get_all_ifaces(br, &want_ifaces);
622 SHASH_FOR_EACH (node, &want_ifaces) {
623 const char *if_name = node->name;
624 struct iface *iface = node->data;
626 if (shash_find(&cur_ifaces, if_name)) {
627 /* Already exists, just reconfigure it. */
629 reconfigure_iface(iface->cfg);
632 /* Need to add to datapath. */
636 /* Add to datapath. */
637 internal = iface_is_internal(br, if_name);
638 error = dpif_port_add(br->dpif, if_name,
639 internal ? ODP_PORT_INTERNAL : 0, NULL);
640 if (error == EFBIG) {
641 VLOG_ERR("ran out of valid port numbers on %s",
642 dpif_name(br->dpif));
645 VLOG_ERR("failed to add %s interface to %s: %s",
646 if_name, dpif_name(br->dpif), strerror(error));
650 shash_destroy(&cur_ifaces);
651 shash_destroy(&want_ifaces);
653 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
656 struct iface *local_iface;
657 struct iface *hw_addr_iface;
660 bridge_fetch_dp_ifaces(br);
661 iterate_and_prune_ifaces(br, init_iface_netdev, NULL);
663 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
665 /* Pick local port hardware address, datapath ID. */
666 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
667 local_iface = bridge_get_local_iface(br);
669 int error = netdev_set_etheraddr(local_iface->netdev, ea);
671 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
672 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
673 "Ethernet address: %s",
674 br->name, strerror(error));
678 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
679 ofproto_set_datapath_id(br->ofproto, dpid);
681 dpid_string = xasprintf("%012"PRIx64, dpid);
682 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
685 /* Set NetFlow configuration on this bridge. */
686 if (br->cfg->netflow) {
687 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
688 struct netflow_options opts;
690 memset(&opts, 0, sizeof opts);
692 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
693 if (nf_cfg->engine_type) {
694 opts.engine_type = nf_cfg->engine_type;
696 if (nf_cfg->engine_id) {
697 opts.engine_id = nf_cfg->engine_id;
700 opts.active_timeout = nf_cfg->active_timeout;
701 if (!opts.active_timeout) {
702 opts.active_timeout = -1;
703 } else if (opts.active_timeout < 0) {
704 VLOG_WARN("bridge %s: active timeout interval set to negative "
705 "value, using default instead (%d seconds)", br->name,
706 NF_ACTIVE_TIMEOUT_DEFAULT);
707 opts.active_timeout = -1;
710 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
711 if (opts.add_id_to_iface) {
712 if (opts.engine_id > 0x7f) {
713 VLOG_WARN("bridge %s: netflow port mangling may conflict "
714 "with another vswitch, choose an engine id less "
715 "than 128", br->name);
717 if (br->n_ports > 508) {
718 VLOG_WARN("bridge %s: netflow port mangling will conflict "
719 "with another port when more than 508 ports are "
724 opts.collectors.n = nf_cfg->n_targets;
725 opts.collectors.names = nf_cfg->targets;
726 if (ofproto_set_netflow(br->ofproto, &opts)) {
727 VLOG_ERR("bridge %s: problem setting netflow collectors",
731 ofproto_set_netflow(br->ofproto, NULL);
734 /* Update the controller and related settings. It would be more
735 * straightforward to call this from bridge_reconfigure_one(), but we
736 * can't do it there for two reasons. First, and most importantly, at
737 * that point we don't know the dp_ifidx of any interfaces that have
738 * been added to the bridge (because we haven't actually added them to
739 * the datapath). Second, at that point we haven't set the datapath ID
740 * yet; when a controller is configured, resetting the datapath ID will
741 * immediately disconnect from the controller, so it's better to set
742 * the datapath ID before the controller. */
743 bridge_reconfigure_controller(ovs_cfg, br);
745 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
746 for (i = 0; i < br->n_ports; i++) {
747 struct port *port = br->ports[i];
749 port_update_vlan_compat(port);
750 port_update_bonding(port);
753 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
754 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
757 ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
759 ovsdb_idl_txn_commit(txn);
760 ovsdb_idl_txn_destroy(txn); /* XXX */
764 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
768 for (i = 0; i < br_cfg->n_other_config; i++) {
769 if (!strcmp(br_cfg->key_other_config[i], key)) {
770 return br_cfg->value_other_config[i];
777 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
778 struct iface **hw_addr_iface)
784 *hw_addr_iface = NULL;
786 /* Did the user request a particular MAC? */
787 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
788 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
789 if (eth_addr_is_multicast(ea)) {
790 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
791 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
792 } else if (eth_addr_is_zero(ea)) {
793 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
799 /* Otherwise choose the minimum non-local MAC address among all of the
801 memset(ea, 0xff, sizeof ea);
802 for (i = 0; i < br->n_ports; i++) {
803 struct port *port = br->ports[i];
804 uint8_t iface_ea[ETH_ADDR_LEN];
807 /* Mirror output ports don't participate. */
808 if (port->is_mirror_output_port) {
812 /* Choose the MAC address to represent the port. */
813 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
814 /* Find the interface with this Ethernet address (if any) so that
815 * we can provide the correct devname to the caller. */
817 for (j = 0; j < port->n_ifaces; j++) {
818 struct iface *candidate = port->ifaces[j];
819 uint8_t candidate_ea[ETH_ADDR_LEN];
820 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
821 && eth_addr_equals(iface_ea, candidate_ea)) {
826 /* Choose the interface whose MAC address will represent the port.
827 * The Linux kernel bonding code always chooses the MAC address of
828 * the first slave added to a bond, and the Fedora networking
829 * scripts always add slaves to a bond in alphabetical order, so
830 * for compatibility we choose the interface with the name that is
831 * first in alphabetical order. */
832 iface = port->ifaces[0];
833 for (j = 1; j < port->n_ifaces; j++) {
834 struct iface *candidate = port->ifaces[j];
835 if (strcmp(candidate->name, iface->name) < 0) {
840 /* The local port doesn't count (since we're trying to choose its
841 * MAC address anyway). */
842 if (iface->dp_ifidx == ODPP_LOCAL) {
847 error = netdev_get_etheraddr(iface->netdev, iface_ea);
849 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
850 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
851 iface->name, strerror(error));
856 /* Compare against our current choice. */
857 if (!eth_addr_is_multicast(iface_ea) &&
858 !eth_addr_is_local(iface_ea) &&
859 !eth_addr_is_reserved(iface_ea) &&
860 !eth_addr_is_zero(iface_ea) &&
861 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
863 memcpy(ea, iface_ea, ETH_ADDR_LEN);
864 *hw_addr_iface = iface;
867 if (eth_addr_is_multicast(ea)) {
868 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
869 *hw_addr_iface = NULL;
870 VLOG_WARN("bridge %s: using default bridge Ethernet "
871 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
873 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
874 br->name, ETH_ADDR_ARGS(ea));
878 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
879 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
880 * an interface on 'br', then that interface must be passed in as
881 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
882 * 'hw_addr_iface' must be passed in as a null pointer. */
884 bridge_pick_datapath_id(struct bridge *br,
885 const uint8_t bridge_ea[ETH_ADDR_LEN],
886 struct iface *hw_addr_iface)
889 * The procedure for choosing a bridge MAC address will, in the most
890 * ordinary case, also choose a unique MAC that we can use as a datapath
891 * ID. In some special cases, though, multiple bridges will end up with
892 * the same MAC address. This is OK for the bridges, but it will confuse
893 * the OpenFlow controller, because each datapath needs a unique datapath
896 * Datapath IDs must be unique. It is also very desirable that they be
897 * stable from one run to the next, so that policy set on a datapath
900 const char *datapath_id;
903 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
904 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
910 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
912 * A bridge whose MAC address is taken from a VLAN network device
913 * (that is, a network device created with vconfig(8) or similar
914 * tool) will have the same MAC address as a bridge on the VLAN
915 * device's physical network device.
917 * Handle this case by hashing the physical network device MAC
918 * along with the VLAN identifier.
920 uint8_t buf[ETH_ADDR_LEN + 2];
921 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
922 buf[ETH_ADDR_LEN] = vlan >> 8;
923 buf[ETH_ADDR_LEN + 1] = vlan;
924 return dpid_from_hash(buf, sizeof buf);
927 * Assume that this bridge's MAC address is unique, since it
928 * doesn't fit any of the cases we handle specially.
933 * A purely internal bridge, that is, one that has no non-virtual
934 * network devices on it at all, is more difficult because it has no
935 * natural unique identifier at all.
937 * When the host is a XenServer, we handle this case by hashing the
938 * host's UUID with the name of the bridge. Names of bridges are
939 * persistent across XenServer reboots, although they can be reused if
940 * an internal network is destroyed and then a new one is later
941 * created, so this is fairly effective.
943 * When the host is not a XenServer, we punt by using a random MAC
944 * address on each run.
946 const char *host_uuid = xenserver_get_host_uuid();
948 char *combined = xasprintf("%s,%s", host_uuid, br->name);
949 dpid = dpid_from_hash(combined, strlen(combined));
955 return eth_addr_to_uint64(bridge_ea);
959 dpid_from_hash(const void *data, size_t n)
961 uint8_t hash[SHA1_DIGEST_SIZE];
963 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
964 sha1_bytes(data, n, hash);
965 eth_addr_mark_random(hash);
966 return eth_addr_to_uint64(hash);
972 struct bridge *br, *next;
976 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
977 int error = bridge_run_one(br);
979 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
980 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
981 "forcing reconfiguration", br->name);
995 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
996 ofproto_wait(br->ofproto);
997 if (br->controller) {
1001 mac_learning_wait(br->ml);
1006 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1007 * configuration changes. */
1009 bridge_flush(struct bridge *br)
1011 COVERAGE_INC(bridge_flush);
1013 mac_learning_flush(br->ml);
1016 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1017 * such interface. */
1018 static struct iface *
1019 bridge_get_local_iface(struct bridge *br)
1023 for (i = 0; i < br->n_ports; i++) {
1024 struct port *port = br->ports[i];
1025 for (j = 0; j < port->n_ifaces; j++) {
1026 struct iface *iface = port->ifaces[j];
1027 if (iface->dp_ifidx == ODPP_LOCAL) {
1036 /* Bridge unixctl user interface functions. */
1038 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1039 const char *args, void *aux UNUSED)
1041 struct ds ds = DS_EMPTY_INITIALIZER;
1042 const struct bridge *br;
1043 const struct mac_entry *e;
1045 br = bridge_lookup(args);
1047 unixctl_command_reply(conn, 501, "no such bridge");
1051 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1052 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
1053 if (e->port < 0 || e->port >= br->n_ports) {
1056 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1057 br->ports[e->port]->ifaces[0]->dp_ifidx,
1058 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1060 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1064 /* Bridge reconfiguration functions. */
1066 static struct bridge *
1067 bridge_create(const char *name)
1072 assert(!bridge_lookup(name));
1073 br = xzalloc(sizeof *br);
1075 error = dpif_create_and_open(name, &br->dpif);
1080 dpif_flow_flush(br->dpif);
1082 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
1084 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
1085 dpif_delete(br->dpif);
1086 dpif_close(br->dpif);
1091 br->name = xstrdup(name);
1092 br->ml = mac_learning_create();
1093 br->sent_config_request = false;
1094 eth_addr_random(br->default_ea);
1096 port_array_init(&br->ifaces);
1099 br->bond_next_rebalance = time_msec() + 10000;
1101 list_push_back(&all_bridges, &br->node);
1103 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1109 bridge_destroy(struct bridge *br)
1114 while (br->n_ports > 0) {
1115 port_destroy(br->ports[br->n_ports - 1]);
1117 list_remove(&br->node);
1118 error = dpif_delete(br->dpif);
1119 if (error && error != ENOENT) {
1120 VLOG_ERR("failed to delete %s: %s",
1121 dpif_name(br->dpif), strerror(error));
1123 dpif_close(br->dpif);
1124 ofproto_destroy(br->ofproto);
1125 free(br->controller);
1126 mac_learning_destroy(br->ml);
1127 port_array_destroy(&br->ifaces);
1134 static struct bridge *
1135 bridge_lookup(const char *name)
1139 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1140 if (!strcmp(br->name, name)) {
1148 bridge_exists(const char *name)
1150 return bridge_lookup(name) ? true : false;
1154 bridge_get_datapathid(const char *name)
1156 struct bridge *br = bridge_lookup(name);
1157 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1160 /* Handle requests for a listing of all flows known by the OpenFlow
1161 * stack, including those normally hidden. */
1163 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1164 const char *args, void *aux UNUSED)
1169 br = bridge_lookup(args);
1171 unixctl_command_reply(conn, 501, "Unknown bridge");
1176 ofproto_get_all_flows(br->ofproto, &results);
1178 unixctl_command_reply(conn, 200, ds_cstr(&results));
1179 ds_destroy(&results);
1183 bridge_run_one(struct bridge *br)
1187 error = ofproto_run1(br->ofproto);
1192 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1195 error = ofproto_run2(br->ofproto, br->flush);
1201 static const struct ovsrec_controller *
1202 bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1203 const struct bridge *br)
1205 const struct ovsrec_controller *controller;
1207 controller = (br->cfg->controller ? br->cfg->controller
1208 : ovs_cfg->controller ? ovs_cfg->controller
1211 if (controller && !strcmp(controller->target, "none")) {
1219 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1221 struct svec *ifaces = ifaces_;
1222 if (!svec_contains(ifaces, iface->name)) {
1223 svec_add(ifaces, iface->name);
1227 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1229 br->name, iface->name, iface->port->name);
1235 bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
1238 struct shash old_ports, new_ports;
1240 struct svec listeners, old_listeners;
1241 struct svec snoops, old_snoops;
1242 struct shash_node *node;
1246 /* Collect old ports. */
1247 shash_init(&old_ports);
1248 for (i = 0; i < br->n_ports; i++) {
1249 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1252 /* Collect new ports. */
1253 shash_init(&new_ports);
1254 for (i = 0; i < br->cfg->n_ports; i++) {
1255 const char *name = br->cfg->ports[i]->name;
1256 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1257 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1262 /* If we have a controller, then we need a local port. Complain if the
1263 * user didn't specify one.
1265 * XXX perhaps we should synthesize a port ourselves in this case. */
1266 if (bridge_get_controller(ovs_cfg, br)) {
1267 char local_name[IF_NAMESIZE];
1270 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1271 local_name, sizeof local_name);
1272 if (!error && !shash_find(&new_ports, local_name)) {
1273 VLOG_WARN("bridge %s: controller specified but no local port "
1274 "(port named %s) defined",
1275 br->name, local_name);
1279 dpid_from_string(ovs_cfg->management_id, &mgmt_id);
1280 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
1282 /* Get rid of deleted ports and add new ports. */
1283 SHASH_FOR_EACH (node, &old_ports) {
1284 if (!shash_find(&new_ports, node->name)) {
1285 port_destroy(node->data);
1288 SHASH_FOR_EACH (node, &new_ports) {
1289 struct port *port = shash_find_data(&old_ports, node->name);
1291 port = port_create(br, node->name);
1293 port_reconfigure(port, node->data);
1295 shash_destroy(&old_ports);
1296 shash_destroy(&new_ports);
1298 /* Check and delete duplicate interfaces. */
1300 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1301 svec_destroy(&ifaces);
1303 /* Delete all flows if we're switching from connected to standalone or vice
1304 * versa. (XXX Should we delete all flows if we are switching from one
1305 * controller to another?) */
1308 /* Configure OpenFlow management listeners. */
1309 svec_init(&listeners);
1310 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1312 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1313 ovs_rundir, br->name));
1314 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1315 svec_clear(&listeners);
1317 svec_sort_unique(&listeners);
1319 svec_init(&old_listeners);
1320 ofproto_get_listeners(br->ofproto, &old_listeners);
1321 svec_sort_unique(&old_listeners);
1323 if (!svec_equal(&listeners, &old_listeners)) {
1324 ofproto_set_listeners(br->ofproto, &listeners);
1326 svec_destroy(&listeners);
1327 svec_destroy(&old_listeners);
1329 /* Configure OpenFlow controller connection snooping. */
1331 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1333 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1334 ovs_rundir, br->name));
1335 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1336 svec_clear(&snoops);
1338 svec_sort_unique(&snoops);
1340 svec_init(&old_snoops);
1341 ofproto_get_snoops(br->ofproto, &old_snoops);
1342 svec_sort_unique(&old_snoops);
1344 if (!svec_equal(&snoops, &old_snoops)) {
1345 ofproto_set_snoops(br->ofproto, &snoops);
1347 svec_destroy(&snoops);
1348 svec_destroy(&old_snoops);
1350 /* Default listener. */
1351 svec_init(&listeners);
1352 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1353 ovs_rundir, br->name));
1354 svec_init(&old_listeners);
1355 ofproto_get_listeners(br->ofproto, &old_listeners);
1356 if (!svec_equal(&listeners, &old_listeners)) {
1357 ofproto_set_listeners(br->ofproto, &listeners);
1359 svec_destroy(&listeners);
1360 svec_destroy(&old_listeners);
1362 /* Default snoop. */
1364 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1365 ovs_rundir, br->name));
1366 svec_init(&old_snoops);
1367 ofproto_get_snoops(br->ofproto, &old_snoops);
1368 if (!svec_equal(&snoops, &old_snoops)) {
1369 ofproto_set_snoops(br->ofproto, &snoops);
1371 svec_destroy(&snoops);
1372 svec_destroy(&old_snoops);
1376 mirror_reconfigure(br);
1381 bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg,
1384 char *pfx = xasprintf("bridge.%s.controller", br->name);
1385 const struct ovsrec_controller *c;
1387 c = bridge_get_controller(ovs_cfg, br);
1388 if ((br->controller != NULL) != (c != NULL)) {
1389 ofproto_flush_flows(br->ofproto);
1391 free(br->controller);
1392 br->controller = c ? xstrdup(c->target) : NULL;
1395 int max_backoff, probe;
1396 int rate_limit, burst_limit;
1398 if (!strcmp(c->target, "discover")) {
1399 ofproto_set_discovery(br->ofproto, true,
1400 c->discover_accept_regex,
1401 c->discover_update_resolv_conf);
1403 struct iface *local_iface;
1407 in_band = (!c->connection_mode
1408 || !strcmp(c->connection_mode, "out-of-band"));
1409 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1410 ofproto_set_in_band(br->ofproto, in_band);
1412 local_iface = bridge_get_local_iface(br);
1413 if (local_iface && c->local_ip && inet_aton(c->local_ip, &ip)) {
1414 struct netdev *netdev = local_iface->netdev;
1415 struct in_addr ip, mask, gateway;
1417 if (!c->local_netmask || !inet_aton(c->local_netmask, &mask)) {
1420 if (!c->local_gateway
1421 || !inet_aton(c->local_gateway, &gateway)) {
1425 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1427 mask.s_addr = guess_netmask(ip.s_addr);
1429 if (!netdev_set_in4(netdev, ip, mask)) {
1430 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1432 br->name, IP_ARGS(&ip.s_addr),
1433 IP_ARGS(&mask.s_addr));
1436 if (gateway.s_addr) {
1437 if (!netdev_add_router(netdev, gateway)) {
1438 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1439 br->name, IP_ARGS(&gateway.s_addr));
1445 ofproto_set_failure(br->ofproto,
1447 || !strcmp(c->fail_mode, "standalone")
1448 || !strcmp(c->fail_mode, "open")));
1450 probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1451 ofproto_set_probe_interval(br->ofproto, probe);
1453 max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1454 ofproto_set_max_backoff(br->ofproto, max_backoff);
1456 rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1457 burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0;
1458 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1460 union ofp_action action;
1463 /* Set up a flow that matches every packet and directs them to
1464 * OFPP_NORMAL (which goes to us). */
1465 memset(&action, 0, sizeof action);
1466 action.type = htons(OFPAT_OUTPUT);
1467 action.output.len = htons(sizeof action);
1468 action.output.port = htons(OFPP_NORMAL);
1469 memset(&flow, 0, sizeof flow);
1470 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1473 ofproto_set_in_band(br->ofproto, false);
1474 ofproto_set_max_backoff(br->ofproto, 1);
1475 ofproto_set_probe_interval(br->ofproto, 5);
1476 ofproto_set_failure(br->ofproto, false);
1480 ofproto_set_controller(br->ofproto, br->controller);
1484 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1489 for (i = 0; i < br->n_ports; i++) {
1490 struct port *port = br->ports[i];
1491 for (j = 0; j < port->n_ifaces; j++) {
1492 struct iface *iface = port->ifaces[j];
1493 shash_add_once(ifaces, iface->name, iface);
1495 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1496 shash_add_once(ifaces, port->name, NULL);
1501 /* For robustness, in case the administrator moves around datapath ports behind
1502 * our back, we re-check all the datapath port numbers here.
1504 * This function will set the 'dp_ifidx' members of interfaces that have
1505 * disappeared to -1, so only call this function from a context where those
1506 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1507 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1508 * datapath, which doesn't support UINT16_MAX+1 ports. */
1510 bridge_fetch_dp_ifaces(struct bridge *br)
1512 struct odp_port *dpif_ports;
1513 size_t n_dpif_ports;
1516 /* Reset all interface numbers. */
1517 for (i = 0; i < br->n_ports; i++) {
1518 struct port *port = br->ports[i];
1519 for (j = 0; j < port->n_ifaces; j++) {
1520 struct iface *iface = port->ifaces[j];
1521 iface->dp_ifidx = -1;
1524 port_array_clear(&br->ifaces);
1526 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1527 for (i = 0; i < n_dpif_ports; i++) {
1528 struct odp_port *p = &dpif_ports[i];
1529 struct iface *iface = iface_lookup(br, p->devname);
1531 if (iface->dp_ifidx >= 0) {
1532 VLOG_WARN("%s reported interface %s twice",
1533 dpif_name(br->dpif), p->devname);
1534 } else if (iface_from_dp_ifidx(br, p->port)) {
1535 VLOG_WARN("%s reported interface %"PRIu16" twice",
1536 dpif_name(br->dpif), p->port);
1538 port_array_set(&br->ifaces, p->port, iface);
1539 iface->dp_ifidx = p->port;
1543 int64_t ofport = (iface->dp_ifidx >= 0
1544 ? odp_port_to_ofp_port(iface->dp_ifidx)
1546 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1553 /* Bridge packet processing functions. */
1556 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1558 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1561 static struct bond_entry *
1562 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1564 return &port->bond_hash[bond_hash(mac)];
1568 bond_choose_iface(const struct port *port)
1570 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1571 size_t i, best_down_slave = -1;
1572 long long next_delay_expiration = LLONG_MAX;
1574 for (i = 0; i < port->n_ifaces; i++) {
1575 struct iface *iface = port->ifaces[i];
1577 if (iface->enabled) {
1579 } else if (iface->delay_expires < next_delay_expiration) {
1580 best_down_slave = i;
1581 next_delay_expiration = iface->delay_expires;
1585 if (best_down_slave != -1) {
1586 struct iface *iface = port->ifaces[best_down_slave];
1588 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1589 "since no other interface is up", iface->name,
1590 iface->delay_expires - time_msec());
1591 bond_enable_slave(iface, true);
1594 return best_down_slave;
1598 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1599 uint16_t *dp_ifidx, tag_type *tags)
1601 struct iface *iface;
1603 assert(port->n_ifaces);
1604 if (port->n_ifaces == 1) {
1605 iface = port->ifaces[0];
1607 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1608 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1609 || !port->ifaces[e->iface_idx]->enabled) {
1610 /* XXX select interface properly. The current interface selection
1611 * is only good for testing the rebalancing code. */
1612 e->iface_idx = bond_choose_iface(port);
1613 if (e->iface_idx < 0) {
1614 *tags |= port->no_ifaces_tag;
1617 e->iface_tag = tag_create_random();
1618 ((struct port *) port)->bond_compat_is_stale = true;
1620 *tags |= e->iface_tag;
1621 iface = port->ifaces[e->iface_idx];
1623 *dp_ifidx = iface->dp_ifidx;
1624 *tags |= iface->tag; /* Currently only used for bonding. */
1629 bond_link_status_update(struct iface *iface, bool carrier)
1631 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1632 struct port *port = iface->port;
1634 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1635 /* Nothing to do. */
1638 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1639 iface->name, carrier ? "detected" : "dropped");
1640 if (carrier == iface->enabled) {
1641 iface->delay_expires = LLONG_MAX;
1642 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1643 iface->name, carrier ? "disabled" : "enabled");
1644 } else if (carrier && port->active_iface < 0) {
1645 bond_enable_slave(iface, true);
1646 if (port->updelay) {
1647 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1648 "other interface is up", iface->name, port->updelay);
1651 int delay = carrier ? port->updelay : port->downdelay;
1652 iface->delay_expires = time_msec() + delay;
1655 "interface %s: will be %s if it stays %s for %d ms",
1657 carrier ? "enabled" : "disabled",
1658 carrier ? "up" : "down",
1665 bond_choose_active_iface(struct port *port)
1667 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1669 port->active_iface = bond_choose_iface(port);
1670 port->active_iface_tag = tag_create_random();
1671 if (port->active_iface >= 0) {
1672 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1673 port->name, port->ifaces[port->active_iface]->name);
1675 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1681 bond_enable_slave(struct iface *iface, bool enable)
1683 struct port *port = iface->port;
1684 struct bridge *br = port->bridge;
1686 /* This acts as a recursion check. If the act of disabling a slave
1687 * causes a different slave to be enabled, the flag will allow us to
1688 * skip redundant work when we reenter this function. It must be
1689 * cleared on exit to keep things safe with multiple bonds. */
1690 static bool moving_active_iface = false;
1692 iface->delay_expires = LLONG_MAX;
1693 if (enable == iface->enabled) {
1697 iface->enabled = enable;
1698 if (!iface->enabled) {
1699 VLOG_WARN("interface %s: disabled", iface->name);
1700 ofproto_revalidate(br->ofproto, iface->tag);
1701 if (iface->port_ifidx == port->active_iface) {
1702 ofproto_revalidate(br->ofproto,
1703 port->active_iface_tag);
1705 /* Disabling a slave can lead to another slave being immediately
1706 * enabled if there will be no active slaves but one is waiting
1707 * on an updelay. In this case we do not need to run most of the
1708 * code for the newly enabled slave since there was no period
1709 * without an active slave and it is redundant with the disabling
1711 moving_active_iface = true;
1712 bond_choose_active_iface(port);
1714 bond_send_learning_packets(port);
1716 VLOG_WARN("interface %s: enabled", iface->name);
1717 if (port->active_iface < 0 && !moving_active_iface) {
1718 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1719 bond_choose_active_iface(port);
1720 bond_send_learning_packets(port);
1722 iface->tag = tag_create_random();
1725 moving_active_iface = false;
1726 port->bond_compat_is_stale = true;
1730 bond_run(struct bridge *br)
1734 for (i = 0; i < br->n_ports; i++) {
1735 struct port *port = br->ports[i];
1737 if (port->n_ifaces >= 2) {
1738 for (j = 0; j < port->n_ifaces; j++) {
1739 struct iface *iface = port->ifaces[j];
1740 if (time_msec() >= iface->delay_expires) {
1741 bond_enable_slave(iface, !iface->enabled);
1746 if (port->bond_compat_is_stale) {
1747 port->bond_compat_is_stale = false;
1748 port_update_bond_compat(port);
1754 bond_wait(struct bridge *br)
1758 for (i = 0; i < br->n_ports; i++) {
1759 struct port *port = br->ports[i];
1760 if (port->n_ifaces < 2) {
1763 for (j = 0; j < port->n_ifaces; j++) {
1764 struct iface *iface = port->ifaces[j];
1765 if (iface->delay_expires != LLONG_MAX) {
1766 poll_timer_wait(iface->delay_expires - time_msec());
1773 set_dst(struct dst *p, const flow_t *flow,
1774 const struct port *in_port, const struct port *out_port,
1777 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1778 : in_port->vlan >= 0 ? in_port->vlan
1779 : ntohs(flow->dl_vlan));
1780 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1784 swap_dst(struct dst *p, struct dst *q)
1786 struct dst tmp = *p;
1791 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1792 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1793 * that we push to the datapath. We could in fact fully sort the array by
1794 * vlan, but in most cases there are at most two different vlan tags so that's
1795 * possibly overkill.) */
1797 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1799 struct dst *first = dsts;
1800 struct dst *last = dsts + n_dsts;
1802 while (first != last) {
1804 * - All dsts < first have vlan == 'vlan'.
1805 * - All dsts >= last have vlan != 'vlan'.
1806 * - first < last. */
1807 while (first->vlan == vlan) {
1808 if (++first == last) {
1813 /* Same invariants, plus one additional:
1814 * - first->vlan != vlan.
1816 while (last[-1].vlan != vlan) {
1817 if (--last == first) {
1822 /* Same invariants, plus one additional:
1823 * - last[-1].vlan == vlan.*/
1824 swap_dst(first++, --last);
1829 mirror_mask_ffs(mirror_mask_t mask)
1831 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1836 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1837 const struct dst *test)
1840 for (i = 0; i < n_dsts; i++) {
1841 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1849 port_trunks_vlan(const struct port *port, uint16_t vlan)
1851 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1855 port_includes_vlan(const struct port *port, uint16_t vlan)
1857 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1861 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1862 const struct port *in_port, const struct port *out_port,
1863 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
1865 mirror_mask_t mirrors = in_port->src_mirrors;
1866 struct dst *dst = dsts;
1869 if (out_port == FLOOD_PORT) {
1870 /* XXX use ODP_FLOOD if no vlans or bonding. */
1871 /* XXX even better, define each VLAN as a datapath port group */
1872 for (i = 0; i < br->n_ports; i++) {
1873 struct port *port = br->ports[i];
1874 if (port != in_port && port_includes_vlan(port, vlan)
1875 && !port->is_mirror_output_port
1876 && set_dst(dst, flow, in_port, port, tags)) {
1877 mirrors |= port->dst_mirrors;
1881 *nf_output_iface = NF_OUT_FLOOD;
1882 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1883 *nf_output_iface = dst->dp_ifidx;
1884 mirrors |= out_port->dst_mirrors;
1889 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1890 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1892 if (set_dst(dst, flow, in_port, m->out_port, tags)
1893 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1897 for (i = 0; i < br->n_ports; i++) {
1898 struct port *port = br->ports[i];
1899 if (port_includes_vlan(port, m->out_vlan)
1900 && set_dst(dst, flow, in_port, port, tags))
1904 if (port->vlan < 0) {
1905 dst->vlan = m->out_vlan;
1907 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
1911 /* Use the vlan tag on the original flow instead of
1912 * the one passed in the vlan parameter. This ensures
1913 * that we compare the vlan from before any implicit
1914 * tagging tags place. This is necessary because
1915 * dst->vlan is the final vlan, after removing implicit
1917 flow_vlan = ntohs(flow->dl_vlan);
1918 if (flow_vlan == 0) {
1919 flow_vlan = OFP_VLAN_NONE;
1921 if (port == in_port && dst->vlan == flow_vlan) {
1922 /* Don't send out input port on same VLAN. */
1930 mirrors &= mirrors - 1;
1933 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1938 print_dsts(const struct dst *dsts, size_t n)
1940 for (; n--; dsts++) {
1941 printf(">p%"PRIu16, dsts->dp_ifidx);
1942 if (dsts->vlan != OFP_VLAN_NONE) {
1943 printf("v%"PRIu16, dsts->vlan);
1949 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1950 const struct port *in_port, const struct port *out_port,
1951 tag_type *tags, struct odp_actions *actions,
1952 uint16_t *nf_output_iface)
1954 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1956 const struct dst *p;
1959 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
1962 cur_vlan = ntohs(flow->dl_vlan);
1963 for (p = dsts; p < &dsts[n_dsts]; p++) {
1964 union odp_action *a;
1965 if (p->vlan != cur_vlan) {
1966 if (p->vlan == OFP_VLAN_NONE) {
1967 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1969 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1970 a->vlan_vid.vlan_vid = htons(p->vlan);
1974 a = odp_actions_add(actions, ODPAT_OUTPUT);
1975 a->output.port = p->dp_ifidx;
1979 /* Returns the effective vlan of a packet, taking into account both the
1980 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
1981 * the packet is untagged and -1 indicates it has an invalid header and
1982 * should be dropped. */
1983 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
1984 struct port *in_port, bool have_packet)
1986 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1987 * belongs to VLAN 0, so we should treat both cases identically. (In the
1988 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1989 * presumably to allow a priority to be specified. In the latter case, the
1990 * packet does not have any 802.1Q header.) */
1991 int vlan = ntohs(flow->dl_vlan);
1992 if (vlan == OFP_VLAN_NONE) {
1995 if (in_port->vlan >= 0) {
1997 /* XXX support double tagging? */
1999 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2000 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2001 "packet received on port %s configured with "
2002 "implicit VLAN %"PRIu16,
2003 br->name, ntohs(flow->dl_vlan),
2004 in_port->name, in_port->vlan);
2008 vlan = in_port->vlan;
2010 if (!port_includes_vlan(in_port, vlan)) {
2012 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2013 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2014 "packet received on port %s not configured for "
2016 br->name, vlan, in_port->name, vlan);
2026 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2027 struct port *in_port)
2029 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
2030 vlan, in_port->port_idx);
2032 /* The log messages here could actually be useful in debugging,
2033 * so keep the rate limit relatively high. */
2034 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2036 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2037 "on port %s in VLAN %d",
2038 br->name, ETH_ADDR_ARGS(flow->dl_src),
2039 in_port->name, vlan);
2040 ofproto_revalidate(br->ofproto, rev_tag);
2045 is_bcast_arp_reply(const flow_t *flow)
2047 return (flow->dl_type == htons(ETH_TYPE_ARP)
2048 && flow->nw_proto == ARP_OP_REPLY
2049 && eth_addr_is_broadcast(flow->dl_dst));
2052 /* If the composed actions may be applied to any packet in the given 'flow',
2053 * returns true. Otherwise, the actions should only be applied to 'packet', or
2054 * not at all, if 'packet' was NULL. */
2056 process_flow(struct bridge *br, const flow_t *flow,
2057 const struct ofpbuf *packet, struct odp_actions *actions,
2058 tag_type *tags, uint16_t *nf_output_iface)
2060 struct iface *in_iface;
2061 struct port *in_port;
2062 struct port *out_port = NULL; /* By default, drop the packet/flow. */
2066 /* Find the interface and port structure for the received packet. */
2067 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2069 /* No interface? Something fishy... */
2070 if (packet != NULL) {
2071 /* Odd. A few possible reasons here:
2073 * - We deleted an interface but there are still a few packets
2074 * queued up from it.
2076 * - Someone externally added an interface (e.g. with "ovs-dpctl
2077 * add-if") that we don't know about.
2079 * - Packet arrived on the local port but the local port is not
2080 * one of our bridge ports.
2082 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2084 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2085 "interface %"PRIu16, br->name, flow->in_port);
2088 /* Return without adding any actions, to drop packets on this flow. */
2091 in_port = in_iface->port;
2092 vlan = flow_get_vlan(br, flow, in_port, !!packet);
2097 /* Drop frames for reserved multicast addresses. */
2098 if (eth_addr_is_reserved(flow->dl_dst)) {
2102 /* Drop frames on ports reserved for mirroring. */
2103 if (in_port->is_mirror_output_port) {
2104 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2105 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
2106 "which is reserved exclusively for mirroring",
2107 br->name, in_port->name);
2111 /* Packets received on bonds need special attention to avoid duplicates. */
2112 if (in_port->n_ifaces > 1) {
2115 if (eth_addr_is_multicast(flow->dl_dst)) {
2116 *tags |= in_port->active_iface_tag;
2117 if (in_port->active_iface != in_iface->port_ifidx) {
2118 /* Drop all multicast packets on inactive slaves. */
2123 /* Drop all packets for which we have learned a different input
2124 * port, because we probably sent the packet on one slave and got
2125 * it back on the other. Broadcast ARP replies are an exception
2126 * to this rule: the host has moved to another switch. */
2127 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
2128 if (src_idx != -1 && src_idx != in_port->port_idx &&
2129 !is_bcast_arp_reply(flow)) {
2135 out_port = FLOOD_PORT;
2136 /* Learn source MAC (but don't try to learn from revalidation). */
2138 update_learning_table(br, flow, vlan, in_port);
2141 /* Determine output port. */
2142 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
2144 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2145 out_port = br->ports[out_port_idx];
2146 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2147 /* If we are revalidating but don't have a learning entry then
2148 * eject the flow. Installing a flow that floods packets opens
2149 * up a window of time where we could learn from a packet reflected
2150 * on a bond and blackhole packets before the learning table is
2151 * updated to reflect the correct port. */
2155 /* Don't send packets out their input ports. */
2156 if (in_port == out_port) {
2161 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2167 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2170 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2171 const struct ofp_phy_port *opp,
2174 struct bridge *br = br_;
2175 struct iface *iface;
2178 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2184 if (reason == OFPPR_DELETE) {
2185 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2186 br->name, iface->name);
2187 iface_destroy(iface);
2188 if (!port->n_ifaces) {
2189 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2190 br->name, port->name);
2196 if (port->n_ifaces > 1) {
2197 bool up = !(opp->state & OFPPS_LINK_DOWN);
2198 bond_link_status_update(iface, up);
2199 port_update_bond_compat(port);
2205 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2206 struct odp_actions *actions, tag_type *tags,
2207 uint16_t *nf_output_iface, void *br_)
2209 struct bridge *br = br_;
2211 COVERAGE_INC(bridge_process_flow);
2212 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2216 bridge_account_flow_ofhook_cb(const flow_t *flow,
2217 const union odp_action *actions,
2218 size_t n_actions, unsigned long long int n_bytes,
2221 struct bridge *br = br_;
2222 struct port *in_port;
2223 const union odp_action *a;
2225 /* Feed information from the active flows back into the learning table
2226 * to ensure that table is always in sync with what is actually flowing
2227 * through the datapath. */
2228 in_port = port_from_dp_ifidx(br, flow->in_port);
2230 int vlan = flow_get_vlan(br, flow, in_port, false);
2232 update_learning_table(br, flow, vlan, in_port);
2236 if (!br->has_bonded_ports) {
2240 for (a = actions; a < &actions[n_actions]; a++) {
2241 if (a->type == ODPAT_OUTPUT) {
2242 struct port *out_port = port_from_dp_ifidx(br, a->output.port);
2243 if (out_port && out_port->n_ifaces >= 2) {
2244 struct bond_entry *e = lookup_bond_entry(out_port,
2246 e->tx_bytes += n_bytes;
2253 bridge_account_checkpoint_ofhook_cb(void *br_)
2255 struct bridge *br = br_;
2258 if (!br->has_bonded_ports) {
2262 /* The current ofproto implementation calls this callback at least once a
2263 * second, so this timer implementation is sufficient. */
2264 if (time_msec() < br->bond_next_rebalance) {
2267 br->bond_next_rebalance = time_msec() + 10000;
2269 for (i = 0; i < br->n_ports; i++) {
2270 struct port *port = br->ports[i];
2271 if (port->n_ifaces > 1) {
2272 bond_rebalance_port(port);
2277 static struct ofhooks bridge_ofhooks = {
2278 bridge_port_changed_ofhook_cb,
2279 bridge_normal_ofhook_cb,
2280 bridge_account_flow_ofhook_cb,
2281 bridge_account_checkpoint_ofhook_cb,
2284 /* Bonding functions. */
2286 /* Statistics for a single interface on a bonded port, used for load-based
2287 * bond rebalancing. */
2288 struct slave_balance {
2289 struct iface *iface; /* The interface. */
2290 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2292 /* All the "bond_entry"s that are assigned to this interface, in order of
2293 * increasing tx_bytes. */
2294 struct bond_entry **hashes;
2298 /* Sorts pointers to pointers to bond_entries in ascending order by the
2299 * interface to which they are assigned, and within a single interface in
2300 * ascending order of bytes transmitted. */
2302 compare_bond_entries(const void *a_, const void *b_)
2304 const struct bond_entry *const *ap = a_;
2305 const struct bond_entry *const *bp = b_;
2306 const struct bond_entry *a = *ap;
2307 const struct bond_entry *b = *bp;
2308 if (a->iface_idx != b->iface_idx) {
2309 return a->iface_idx > b->iface_idx ? 1 : -1;
2310 } else if (a->tx_bytes != b->tx_bytes) {
2311 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2317 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2318 * *descending* order by number of bytes transmitted. */
2320 compare_slave_balance(const void *a_, const void *b_)
2322 const struct slave_balance *a = a_;
2323 const struct slave_balance *b = b_;
2324 if (a->iface->enabled != b->iface->enabled) {
2325 return a->iface->enabled ? -1 : 1;
2326 } else if (a->tx_bytes != b->tx_bytes) {
2327 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2334 swap_bals(struct slave_balance *a, struct slave_balance *b)
2336 struct slave_balance tmp = *a;
2341 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2342 * given that 'p' (and only 'p') might be in the wrong location.
2344 * This function invalidates 'p', since it might now be in a different memory
2347 resort_bals(struct slave_balance *p,
2348 struct slave_balance bals[], size_t n_bals)
2351 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2352 swap_bals(p, p - 1);
2354 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2355 swap_bals(p, p + 1);
2361 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2363 if (VLOG_IS_DBG_ENABLED()) {
2364 struct ds ds = DS_EMPTY_INITIALIZER;
2365 const struct slave_balance *b;
2367 for (b = bals; b < bals + n_bals; b++) {
2371 ds_put_char(&ds, ',');
2373 ds_put_format(&ds, " %s %"PRIu64"kB",
2374 b->iface->name, b->tx_bytes / 1024);
2376 if (!b->iface->enabled) {
2377 ds_put_cstr(&ds, " (disabled)");
2379 if (b->n_hashes > 0) {
2380 ds_put_cstr(&ds, " (");
2381 for (i = 0; i < b->n_hashes; i++) {
2382 const struct bond_entry *e = b->hashes[i];
2384 ds_put_cstr(&ds, " + ");
2386 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2387 e - port->bond_hash, e->tx_bytes / 1024);
2389 ds_put_cstr(&ds, ")");
2392 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2397 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2399 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2402 struct bond_entry *hash = from->hashes[hash_idx];
2403 struct port *port = from->iface->port;
2404 uint64_t delta = hash->tx_bytes;
2406 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2407 "from %s to %s (now carrying %"PRIu64"kB and "
2408 "%"PRIu64"kB load, respectively)",
2409 port->name, delta / 1024, hash - port->bond_hash,
2410 from->iface->name, to->iface->name,
2411 (from->tx_bytes - delta) / 1024,
2412 (to->tx_bytes + delta) / 1024);
2414 /* Delete element from from->hashes.
2416 * We don't bother to add the element to to->hashes because not only would
2417 * it require more work, the only purpose it would be to allow that hash to
2418 * be migrated to another slave in this rebalancing run, and there is no
2419 * point in doing that. */
2420 if (hash_idx == 0) {
2423 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2424 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2428 /* Shift load away from 'from' to 'to'. */
2429 from->tx_bytes -= delta;
2430 to->tx_bytes += delta;
2432 /* Arrange for flows to be revalidated. */
2433 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2434 hash->iface_idx = to->iface->port_ifidx;
2435 hash->iface_tag = tag_create_random();
2439 bond_rebalance_port(struct port *port)
2441 struct slave_balance bals[DP_MAX_PORTS];
2443 struct bond_entry *hashes[BOND_MASK + 1];
2444 struct slave_balance *b, *from, *to;
2445 struct bond_entry *e;
2448 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2449 * descending order of tx_bytes, so that bals[0] represents the most
2450 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2453 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2454 * array for each slave_balance structure, we sort our local array of
2455 * hashes in order by slave, so that all of the hashes for a given slave
2456 * become contiguous in memory, and then we point each 'hashes' members of
2457 * a slave_balance structure to the start of a contiguous group. */
2458 n_bals = port->n_ifaces;
2459 for (b = bals; b < &bals[n_bals]; b++) {
2460 b->iface = port->ifaces[b - bals];
2465 for (i = 0; i <= BOND_MASK; i++) {
2466 hashes[i] = &port->bond_hash[i];
2468 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2469 for (i = 0; i <= BOND_MASK; i++) {
2471 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2472 b = &bals[e->iface_idx];
2473 b->tx_bytes += e->tx_bytes;
2475 b->hashes = &hashes[i];
2480 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2481 log_bals(bals, n_bals, port);
2483 /* Discard slaves that aren't enabled (which were sorted to the back of the
2484 * array earlier). */
2485 while (!bals[n_bals - 1].iface->enabled) {
2492 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2493 to = &bals[n_bals - 1];
2494 for (from = bals; from < to; ) {
2495 uint64_t overload = from->tx_bytes - to->tx_bytes;
2496 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2497 /* The extra load on 'from' (and all less-loaded slaves), compared
2498 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2499 * it is less than ~1Mbps. No point in rebalancing. */
2501 } else if (from->n_hashes == 1) {
2502 /* 'from' only carries a single MAC hash, so we can't shift any
2503 * load away from it, even though we want to. */
2506 /* 'from' is carrying significantly more load than 'to', and that
2507 * load is split across at least two different hashes. Pick a hash
2508 * to migrate to 'to' (the least-loaded slave), given that doing so
2509 * must decrease the ratio of the load on the two slaves by at
2512 * The sort order we use means that we prefer to shift away the
2513 * smallest hashes instead of the biggest ones. There is little
2514 * reason behind this decision; we could use the opposite sort
2515 * order to shift away big hashes ahead of small ones. */
2519 for (i = 0; i < from->n_hashes; i++) {
2520 double old_ratio, new_ratio;
2521 uint64_t delta = from->hashes[i]->tx_bytes;
2523 if (delta == 0 || from->tx_bytes - delta == 0) {
2524 /* Pointless move. */
2528 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2530 if (to->tx_bytes == 0) {
2531 /* Nothing on the new slave, move it. */
2535 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2536 new_ratio = (double)(from->tx_bytes - delta) /
2537 (to->tx_bytes + delta);
2539 if (new_ratio == 0) {
2540 /* Should already be covered but check to prevent division
2545 if (new_ratio < 1) {
2546 new_ratio = 1 / new_ratio;
2549 if (old_ratio - new_ratio > 0.1) {
2550 /* Would decrease the ratio, move it. */
2554 if (i < from->n_hashes) {
2555 bond_shift_load(from, to, i);
2556 port->bond_compat_is_stale = true;
2558 /* If the result of the migration changed the relative order of
2559 * 'from' and 'to' swap them back to maintain invariants. */
2560 if (order_swapped) {
2561 swap_bals(from, to);
2564 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2565 * point to different slave_balance structures. It is only
2566 * valid to do these two operations in a row at all because we
2567 * know that 'from' will not move past 'to' and vice versa. */
2568 resort_bals(from, bals, n_bals);
2569 resort_bals(to, bals, n_bals);
2576 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2577 * historical data to decay to <1% in 7 rebalancing runs. */
2578 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2584 bond_send_learning_packets(struct port *port)
2586 struct bridge *br = port->bridge;
2587 struct mac_entry *e;
2588 struct ofpbuf packet;
2589 int error, n_packets, n_errors;
2591 if (!port->n_ifaces || port->active_iface < 0) {
2595 ofpbuf_init(&packet, 128);
2596 error = n_packets = n_errors = 0;
2597 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2598 union ofp_action actions[2], *a;
2604 if (e->port == port->port_idx
2605 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2609 /* Compose actions. */
2610 memset(actions, 0, sizeof actions);
2613 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2614 a->vlan_vid.len = htons(sizeof *a);
2615 a->vlan_vid.vlan_vid = htons(e->vlan);
2618 a->output.type = htons(OFPAT_OUTPUT);
2619 a->output.len = htons(sizeof *a);
2620 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2625 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2627 flow_extract(&packet, ODPP_NONE, &flow);
2628 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2635 ofpbuf_uninit(&packet);
2638 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2639 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2640 "packets, last error was: %s",
2641 port->name, n_errors, n_packets, strerror(error));
2643 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2644 port->name, n_packets);
2648 /* Bonding unixctl user interface functions. */
2651 bond_unixctl_list(struct unixctl_conn *conn,
2652 const char *args UNUSED, void *aux UNUSED)
2654 struct ds ds = DS_EMPTY_INITIALIZER;
2655 const struct bridge *br;
2657 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2659 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2662 for (i = 0; i < br->n_ports; i++) {
2663 const struct port *port = br->ports[i];
2664 if (port->n_ifaces > 1) {
2667 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2668 for (j = 0; j < port->n_ifaces; j++) {
2669 const struct iface *iface = port->ifaces[j];
2671 ds_put_cstr(&ds, ", ");
2673 ds_put_cstr(&ds, iface->name);
2675 ds_put_char(&ds, '\n');
2679 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2683 static struct port *
2684 bond_find(const char *name)
2686 const struct bridge *br;
2688 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2691 for (i = 0; i < br->n_ports; i++) {
2692 struct port *port = br->ports[i];
2693 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2702 bond_unixctl_show(struct unixctl_conn *conn,
2703 const char *args, void *aux UNUSED)
2705 struct ds ds = DS_EMPTY_INITIALIZER;
2706 const struct port *port;
2709 port = bond_find(args);
2711 unixctl_command_reply(conn, 501, "no such bond");
2715 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2716 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2717 ds_put_format(&ds, "next rebalance: %lld ms\n",
2718 port->bridge->bond_next_rebalance - time_msec());
2719 for (j = 0; j < port->n_ifaces; j++) {
2720 const struct iface *iface = port->ifaces[j];
2721 struct bond_entry *be;
2724 ds_put_format(&ds, "slave %s: %s\n",
2725 iface->name, iface->enabled ? "enabled" : "disabled");
2726 if (j == port->active_iface) {
2727 ds_put_cstr(&ds, "\tactive slave\n");
2729 if (iface->delay_expires != LLONG_MAX) {
2730 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2731 iface->enabled ? "downdelay" : "updelay",
2732 iface->delay_expires - time_msec());
2736 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2737 int hash = be - port->bond_hash;
2738 struct mac_entry *me;
2740 if (be->iface_idx != j) {
2744 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
2745 hash, be->tx_bytes / 1024);
2748 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2749 &port->bridge->ml->lrus) {
2752 if (bond_hash(me->mac) == hash
2753 && me->port != port->port_idx
2754 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2755 && dp_ifidx == iface->dp_ifidx)
2757 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2758 ETH_ADDR_ARGS(me->mac));
2763 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2768 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
2771 char *args = (char *) args_;
2772 char *save_ptr = NULL;
2773 char *bond_s, *hash_s, *slave_s;
2774 uint8_t mac[ETH_ADDR_LEN];
2776 struct iface *iface;
2777 struct bond_entry *entry;
2780 bond_s = strtok_r(args, " ", &save_ptr);
2781 hash_s = strtok_r(NULL, " ", &save_ptr);
2782 slave_s = strtok_r(NULL, " ", &save_ptr);
2784 unixctl_command_reply(conn, 501,
2785 "usage: bond/migrate BOND HASH SLAVE");
2789 port = bond_find(bond_s);
2791 unixctl_command_reply(conn, 501, "no such bond");
2795 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2796 == ETH_ADDR_SCAN_COUNT) {
2797 hash = bond_hash(mac);
2798 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2799 hash = atoi(hash_s) & BOND_MASK;
2801 unixctl_command_reply(conn, 501, "bad hash");
2805 iface = port_lookup_iface(port, slave_s);
2807 unixctl_command_reply(conn, 501, "no such slave");
2811 if (!iface->enabled) {
2812 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2816 entry = &port->bond_hash[hash];
2817 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2818 entry->iface_idx = iface->port_ifidx;
2819 entry->iface_tag = tag_create_random();
2820 port->bond_compat_is_stale = true;
2821 unixctl_command_reply(conn, 200, "migrated");
2825 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
2828 char *args = (char *) args_;
2829 char *save_ptr = NULL;
2830 char *bond_s, *slave_s;
2832 struct iface *iface;
2834 bond_s = strtok_r(args, " ", &save_ptr);
2835 slave_s = strtok_r(NULL, " ", &save_ptr);
2837 unixctl_command_reply(conn, 501,
2838 "usage: bond/set-active-slave BOND SLAVE");
2842 port = bond_find(bond_s);
2844 unixctl_command_reply(conn, 501, "no such bond");
2848 iface = port_lookup_iface(port, slave_s);
2850 unixctl_command_reply(conn, 501, "no such slave");
2854 if (!iface->enabled) {
2855 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2859 if (port->active_iface != iface->port_ifidx) {
2860 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2861 port->active_iface = iface->port_ifidx;
2862 port->active_iface_tag = tag_create_random();
2863 VLOG_INFO("port %s: active interface is now %s",
2864 port->name, iface->name);
2865 bond_send_learning_packets(port);
2866 unixctl_command_reply(conn, 200, "done");
2868 unixctl_command_reply(conn, 200, "no change");
2873 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2875 char *args = (char *) args_;
2876 char *save_ptr = NULL;
2877 char *bond_s, *slave_s;
2879 struct iface *iface;
2881 bond_s = strtok_r(args, " ", &save_ptr);
2882 slave_s = strtok_r(NULL, " ", &save_ptr);
2884 unixctl_command_reply(conn, 501,
2885 "usage: bond/enable/disable-slave BOND SLAVE");
2889 port = bond_find(bond_s);
2891 unixctl_command_reply(conn, 501, "no such bond");
2895 iface = port_lookup_iface(port, slave_s);
2897 unixctl_command_reply(conn, 501, "no such slave");
2901 bond_enable_slave(iface, enable);
2902 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2906 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
2909 enable_slave(conn, args, true);
2913 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
2916 enable_slave(conn, args, false);
2920 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
2923 uint8_t mac[ETH_ADDR_LEN];
2927 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2928 == ETH_ADDR_SCAN_COUNT) {
2929 hash = bond_hash(mac);
2931 hash_cstr = xasprintf("%u", hash);
2932 unixctl_command_reply(conn, 200, hash_cstr);
2935 unixctl_command_reply(conn, 501, "invalid mac");
2942 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
2943 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
2944 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
2945 unixctl_command_register("bond/set-active-slave",
2946 bond_unixctl_set_active_slave, NULL);
2947 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
2949 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
2951 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
2954 /* Port functions. */
2956 static struct port *
2957 port_create(struct bridge *br, const char *name)
2961 port = xzalloc(sizeof *port);
2963 port->port_idx = br->n_ports;
2965 port->trunks = NULL;
2966 port->name = xstrdup(name);
2967 port->active_iface = -1;
2969 if (br->n_ports >= br->allocated_ports) {
2970 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2973 br->ports[br->n_ports++] = port;
2975 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2982 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
2984 struct shash old_ifaces, new_ifaces;
2985 struct shash_node *node;
2986 unsigned long *trunks;
2992 /* Collect old and new interfaces. */
2993 shash_init(&old_ifaces);
2994 shash_init(&new_ifaces);
2995 for (i = 0; i < port->n_ifaces; i++) {
2996 shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]);
2998 for (i = 0; i < cfg->n_interfaces; i++) {
2999 const char *name = cfg->interfaces[i]->name;
3000 if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) {
3001 VLOG_WARN("port %s: %s specified twice as port interface",
3005 port->updelay = cfg->bond_updelay;
3006 if (port->updelay < 0) {
3009 port->updelay = cfg->bond_downdelay;
3010 if (port->downdelay < 0) {
3011 port->downdelay = 0;
3014 /* Get rid of deleted interfaces and add new interfaces. */
3015 SHASH_FOR_EACH (node, &old_ifaces) {
3016 if (!shash_find(&new_ifaces, node->name)) {
3017 iface_destroy(node->data);
3020 SHASH_FOR_EACH (node, &new_ifaces) {
3021 const struct ovsrec_interface *if_cfg = node->data;
3022 struct iface *iface;
3024 iface = shash_find_data(&old_ifaces, if_cfg->name);
3026 iface = iface_create(port, if_cfg);
3028 iface->cfg = if_cfg;
3034 if (port->n_ifaces < 2) {
3036 if (vlan >= 0 && vlan <= 4095) {
3037 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3042 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3043 * they even work as-is. But they have not been tested. */
3044 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3048 if (port->vlan != vlan) {
3050 bridge_flush(port->bridge);
3053 /* Get trunked VLANs. */
3059 trunks = bitmap_allocate(4096);
3061 for (i = 0; i < cfg->n_trunks; i++) {
3062 int trunk = cfg->trunks[i];
3064 bitmap_set1(trunks, trunk);
3070 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3071 port->name, cfg->n_trunks);
3073 if (n_errors == cfg->n_trunks) {
3075 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3078 bitmap_set_multiple(trunks, 0, 4096, 1);
3081 if (cfg->n_trunks) {
3082 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3087 ? port->trunks != NULL
3088 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3089 bridge_flush(port->bridge);
3091 bitmap_free(port->trunks);
3092 port->trunks = trunks;
3094 shash_destroy(&old_ifaces);
3095 shash_destroy(&new_ifaces);
3099 port_destroy(struct port *port)
3102 struct bridge *br = port->bridge;
3105 proc_net_compat_update_vlan(port->name, NULL, 0);
3106 proc_net_compat_update_bond(port->name, NULL);
3109 for (i = 0; i < MAX_MIRRORS; i++) {
3110 struct mirror *m = br->mirrors[i];
3111 if (m && m->out_port == port) {
3117 while (port->n_ifaces > 0) {
3118 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3121 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3122 del->port_idx = port->port_idx;
3125 bitmap_free(port->trunks);
3132 static struct port *
3133 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3135 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
3136 return iface ? iface->port : NULL;
3139 static struct port *
3140 port_lookup(const struct bridge *br, const char *name)
3144 for (i = 0; i < br->n_ports; i++) {
3145 struct port *port = br->ports[i];
3146 if (!strcmp(port->name, name)) {
3153 static struct iface *
3154 port_lookup_iface(const struct port *port, const char *name)
3158 for (j = 0; j < port->n_ifaces; j++) {
3159 struct iface *iface = port->ifaces[j];
3160 if (!strcmp(iface->name, name)) {
3168 port_update_bonding(struct port *port)
3170 if (port->n_ifaces < 2) {
3171 /* Not a bonded port. */
3172 if (port->bond_hash) {
3173 free(port->bond_hash);
3174 port->bond_hash = NULL;
3175 port->bond_compat_is_stale = true;
3176 port->bond_fake_iface = false;
3179 if (!port->bond_hash) {
3182 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3183 for (i = 0; i <= BOND_MASK; i++) {
3184 struct bond_entry *e = &port->bond_hash[i];
3188 port->no_ifaces_tag = tag_create_random();
3189 bond_choose_active_iface(port);
3191 port->bond_compat_is_stale = true;
3192 port->bond_fake_iface = port->cfg->bond_fake_iface;
3197 port_update_bond_compat(struct port *port)
3199 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3200 struct compat_bond bond;
3203 if (port->n_ifaces < 2) {
3204 proc_net_compat_update_bond(port->name, NULL);
3209 bond.updelay = port->updelay;
3210 bond.downdelay = port->downdelay;
3213 bond.hashes = compat_hashes;
3214 if (port->bond_hash) {
3215 const struct bond_entry *e;
3216 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3217 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3218 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3219 cbh->hash = e - port->bond_hash;
3220 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3225 bond.n_slaves = port->n_ifaces;
3226 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3227 for (i = 0; i < port->n_ifaces; i++) {
3228 struct iface *iface = port->ifaces[i];
3229 struct compat_bond_slave *slave = &bond.slaves[i];
3230 slave->name = iface->name;
3232 /* We need to make the same determination as the Linux bonding
3233 * code to determine whether a slave should be consider "up".
3234 * The Linux function bond_miimon_inspect() supports four
3235 * BOND_LINK_* states:
3237 * - BOND_LINK_UP: carrier detected, updelay has passed.
3238 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3239 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3240 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3242 * The function bond_info_show_slave() only considers BOND_LINK_UP
3243 * to be "up" and anything else to be "down".
3245 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3249 netdev_get_etheraddr(iface->netdev, slave->mac);
3252 if (port->bond_fake_iface) {
3253 struct netdev *bond_netdev;
3255 if (!netdev_open(port->name, NETDEV_ETH_TYPE_NONE, &bond_netdev)) {
3257 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3259 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3261 netdev_close(bond_netdev);
3265 proc_net_compat_update_bond(port->name, &bond);
3270 port_update_vlan_compat(struct port *port)
3272 struct bridge *br = port->bridge;
3273 char *vlandev_name = NULL;
3275 if (port->vlan > 0) {
3276 /* Figure out the name that the VLAN device should actually have, if it
3277 * existed. This takes some work because the VLAN device would not
3278 * have port->name in its name; rather, it would have the trunk port's
3279 * name, and 'port' would be attached to a bridge that also had the
3280 * VLAN device one of its ports. So we need to find a trunk port that
3281 * includes port->vlan.
3283 * There might be more than one candidate. This doesn't happen on
3284 * XenServer, so if it happens we just pick the first choice in
3285 * alphabetical order instead of creating multiple VLAN devices. */
3287 for (i = 0; i < br->n_ports; i++) {
3288 struct port *p = br->ports[i];
3289 if (port_trunks_vlan(p, port->vlan)
3291 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3293 uint8_t ea[ETH_ADDR_LEN];
3294 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3295 if (!eth_addr_is_multicast(ea) &&
3296 !eth_addr_is_reserved(ea) &&
3297 !eth_addr_is_zero(ea)) {
3298 vlandev_name = p->name;
3303 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3306 /* Interface functions. */
3308 static struct iface *
3309 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3311 struct iface *iface;
3312 char *name = if_cfg->name;
3315 iface = xzalloc(sizeof *iface);
3317 iface->port_ifidx = port->n_ifaces;
3318 iface->name = xstrdup(name);
3319 iface->dp_ifidx = -1;
3320 iface->tag = tag_create_random();
3321 iface->delay_expires = LLONG_MAX;
3322 iface->netdev = NULL;
3324 if (port->n_ifaces >= port->allocated_ifaces) {
3325 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3326 sizeof *port->ifaces);
3328 port->ifaces[port->n_ifaces++] = iface;
3329 if (port->n_ifaces > 1) {
3330 port->bridge->has_bonded_ports = true;
3333 /* Attempt to create the network interface in case it
3334 * doesn't exist yet. */
3335 error = set_up_iface(if_cfg, true);
3337 VLOG_WARN("could not create iface %s: %s\n", iface->name,
3341 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3343 bridge_flush(port->bridge);
3349 iface_destroy(struct iface *iface)
3352 struct port *port = iface->port;
3353 struct bridge *br = port->bridge;
3354 bool del_active = port->active_iface == iface->port_ifidx;
3357 if (iface->dp_ifidx >= 0) {
3358 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3361 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3362 del->port_ifidx = iface->port_ifidx;
3364 netdev_close(iface->netdev);
3367 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3368 bond_choose_active_iface(port);
3369 bond_send_learning_packets(port);
3372 netdev_destroy(iface->name);
3376 bridge_flush(port->bridge);
3380 static struct iface *
3381 iface_lookup(const struct bridge *br, const char *name)
3385 for (i = 0; i < br->n_ports; i++) {
3386 struct port *port = br->ports[i];
3387 for (j = 0; j < port->n_ifaces; j++) {
3388 struct iface *iface = port->ifaces[j];
3389 if (!strcmp(iface->name, name)) {
3397 static struct iface *
3398 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3400 return port_array_get(&br->ifaces, dp_ifidx);
3403 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3404 * 'br', that is, an interface that is entirely simulated within the datapath.
3405 * The local port (ODPP_LOCAL) is always an internal interface. Other local
3406 * interfaces are created by setting "iface.<iface>.internal = true".
3408 * In addition, we have a kluge-y feature that creates an internal port with
3409 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3410 * This feature needs to go away in the long term. Until then, this is one
3411 * reason why this function takes a name instead of a struct iface: the fake
3412 * interfaces created this way do not have a struct iface. */
3414 iface_is_internal(const struct bridge *br, const char *if_name)
3416 /* XXX wastes time */
3417 struct iface *iface;
3420 if (!strcmp(if_name, br->name)) {
3424 iface = iface_lookup(br, if_name);
3425 if (iface && !strcmp(iface->cfg->type, "internal")) {
3429 port = port_lookup(br, if_name);
3430 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3436 /* Set Ethernet address of 'iface', if one is specified in the configuration
3439 iface_set_mac(struct iface *iface)
3441 uint8_t ea[ETH_ADDR_LEN];
3443 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3444 if (eth_addr_is_multicast(ea)) {
3445 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3447 } else if (iface->dp_ifidx == ODPP_LOCAL) {
3448 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3449 iface->name, iface->name);
3451 int error = netdev_set_etheraddr(iface->netdev, ea);
3453 VLOG_ERR("interface %s: setting MAC failed (%s)",
3454 iface->name, strerror(error));
3460 /* Port mirroring. */
3464 mirror_reconfigure(struct bridge *br UNUSED)
3466 struct svec old_mirrors, new_mirrors;
3467 size_t i, n_rspan_vlans;
3468 unsigned long *rspan_vlans;
3470 /* Collect old and new mirrors. */
3471 svec_init(&old_mirrors);
3472 svec_init(&new_mirrors);
3473 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
3474 for (i = 0; i < MAX_MIRRORS; i++) {
3475 if (br->mirrors[i]) {
3476 svec_add(&old_mirrors, br->mirrors[i]->name);
3480 /* Get rid of deleted mirrors and add new mirrors. */
3481 svec_sort(&old_mirrors);
3482 assert(svec_is_unique(&old_mirrors));
3483 svec_sort(&new_mirrors);
3484 assert(svec_is_unique(&new_mirrors));
3485 for (i = 0; i < MAX_MIRRORS; i++) {
3486 struct mirror *m = br->mirrors[i];
3487 if (m && !svec_contains(&new_mirrors, m->name)) {
3491 for (i = 0; i < new_mirrors.n; i++) {
3492 const char *name = new_mirrors.names[i];
3493 if (!svec_contains(&old_mirrors, name)) {
3494 mirror_create(br, name);
3497 svec_destroy(&old_mirrors);
3498 svec_destroy(&new_mirrors);
3500 /* Reconfigure all mirrors. */
3501 for (i = 0; i < MAX_MIRRORS; i++) {
3502 if (br->mirrors[i]) {
3503 mirror_reconfigure_one(br->mirrors[i]);
3507 /* Update port reserved status. */
3508 for (i = 0; i < br->n_ports; i++) {
3509 br->ports[i]->is_mirror_output_port = false;
3511 for (i = 0; i < MAX_MIRRORS; i++) {
3512 struct mirror *m = br->mirrors[i];
3513 if (m && m->out_port) {
3514 m->out_port->is_mirror_output_port = true;
3518 /* Update learning disabled vlans (for RSPAN). */
3520 n_rspan_vlans = cfg_count("vlan.%s.disable-learning", br->name);
3521 if (n_rspan_vlans) {
3522 rspan_vlans = bitmap_allocate(4096);
3524 for (i = 0; i < n_rspan_vlans; i++) {
3525 int vlan = cfg_get_vlan(i, "vlan.%s.disable-learning", br->name);
3527 bitmap_set1(rspan_vlans, vlan);
3528 VLOG_INFO("bridge %s: disabling learning on vlan %d\n",
3531 VLOG_ERR("bridge %s: invalid value '%s' for learning disabled "
3533 cfg_get_string(i, "vlan.%s.disable-learning", br->name));
3537 if (mac_learning_set_disabled_vlans(br->ml, rspan_vlans)) {
3543 mirror_create(struct bridge *br, const char *name)
3548 for (i = 0; ; i++) {
3549 if (i >= MAX_MIRRORS) {
3550 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3551 "cannot create %s", br->name, MAX_MIRRORS, name);
3554 if (!br->mirrors[i]) {
3559 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3562 br->mirrors[i] = m = xzalloc(sizeof *m);
3565 m->name = xstrdup(name);
3566 svec_init(&m->src_ports);
3567 svec_init(&m->dst_ports);
3575 mirror_destroy(struct mirror *m)
3578 struct bridge *br = m->bridge;
3581 for (i = 0; i < br->n_ports; i++) {
3582 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3583 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3586 svec_destroy(&m->src_ports);
3587 svec_destroy(&m->dst_ports);
3590 m->bridge->mirrors[m->idx] = NULL;
3598 prune_ports(struct mirror *m, struct svec *ports)
3603 svec_sort_unique(ports);
3606 for (i = 0; i < ports->n; i++) {
3607 const char *name = ports->names[i];
3608 if (port_lookup(m->bridge, name)) {
3609 svec_add(&tmp, name);
3611 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3612 m->bridge->name, m->name, name);
3615 svec_swap(ports, &tmp);
3620 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3624 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3625 * order won't give us numeric sort order. But that's good enough for what
3626 * we need right now. */
3627 svec_sort_unique(vlan_strings);
3629 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3631 for (i = 0; i < vlan_strings->n; i++) {
3632 const char *name = vlan_strings->names[i];
3634 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3635 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3636 m->bridge->name, m->name, name);
3638 (*vlans)[n_vlans++] = vlan;
3645 vlan_is_mirrored(const struct mirror *m, int vlan)
3649 for (i = 0; i < m->n_vlans; i++) {
3650 if (m->vlans[i] == vlan) {
3658 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3662 for (i = 0; i < m->n_vlans; i++) {
3663 if (port_trunks_vlan(p, m->vlans[i])) {
3671 mirror_reconfigure_one(struct mirror *m UNUSED)
3673 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3674 struct svec src_ports, dst_ports, ports;
3675 struct svec vlan_strings;
3676 mirror_mask_t mirror_bit;
3677 const char *out_port_name;
3678 struct port *out_port;
3683 bool mirror_all_ports;
3684 bool any_ports_specified;
3686 /* Get output port. */
3687 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3688 m->bridge->name, m->name);
3689 if (out_port_name) {
3690 out_port = port_lookup(m->bridge, out_port_name);
3692 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3693 "named %s", pfx, m->bridge->name, out_port_name);
3700 if (cfg_has("%s.output.vlan", pfx)) {
3701 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3702 "ignoring %s.output.vlan", pfx, pfx, pfx);
3704 } else if (cfg_has("%s.output.vlan", pfx)) {
3706 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3708 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3709 "but exactly one is required; disabling port mirror %s",
3710 pfx, pfx, pfx, pfx);
3716 /* Get all the ports, and drop duplicates and ports that don't exist. */
3717 svec_init(&src_ports);
3718 svec_init(&dst_ports);
3720 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3721 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3722 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3723 any_ports_specified = src_ports.n || dst_ports.n || ports.n;
3724 svec_append(&src_ports, &ports);
3725 svec_append(&dst_ports, &ports);
3726 svec_destroy(&ports);
3727 prune_ports(m, &src_ports);
3728 prune_ports(m, &dst_ports);
3729 if (any_ports_specified && !src_ports.n && !dst_ports.n) {
3730 VLOG_ERR("%s: none of the specified ports exist; "
3731 "disabling port mirror %s", pfx, pfx);
3736 /* Get all the vlans, and drop duplicate and invalid vlans. */
3737 svec_init(&vlan_strings);
3738 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3739 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3740 svec_destroy(&vlan_strings);
3742 /* Update mirror data. */
3743 if (!svec_equal(&m->src_ports, &src_ports)
3744 || !svec_equal(&m->dst_ports, &dst_ports)
3745 || m->n_vlans != n_vlans
3746 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3747 || m->out_port != out_port
3748 || m->out_vlan != out_vlan) {
3749 bridge_flush(m->bridge);
3751 svec_swap(&m->src_ports, &src_ports);
3752 svec_swap(&m->dst_ports, &dst_ports);
3755 m->n_vlans = n_vlans;
3756 m->out_port = out_port;
3757 m->out_vlan = out_vlan;
3759 /* If no selection criteria have been given, mirror for all ports. */
3760 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3763 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3764 for (i = 0; i < m->bridge->n_ports; i++) {
3765 struct port *port = m->bridge->ports[i];
3767 if (mirror_all_ports
3768 || svec_contains(&m->src_ports, port->name)
3771 ? port_trunks_any_mirrored_vlan(m, port)
3772 : vlan_is_mirrored(m, port->vlan)))) {
3773 port->src_mirrors |= mirror_bit;
3775 port->src_mirrors &= ~mirror_bit;
3778 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3779 port->dst_mirrors |= mirror_bit;
3781 port->dst_mirrors &= ~mirror_bit;
3787 svec_destroy(&src_ports);
3788 svec_destroy(&dst_ports);