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>
37 #include "dynamic-string.h"
41 #include "mac-learning.h"
44 #include "ofp-print.h"
46 #include "ofproto/ofproto.h"
48 #include "poll-loop.h"
49 #include "port-array.h"
50 #include "proc-net-compat.h"
52 #include "socket-util.h"
59 #include "vconn-ssl.h"
60 #include "xenserver.h"
63 #define THIS_MODULE VLM_bridge
71 extern uint64_t mgmt_id;
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? */
88 #define BOND_MASK 0xff
90 int iface_idx; /* Index of assigned iface, or -1 if none. */
91 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
92 tag_type iface_tag; /* Tag associated with iface_idx. */
95 #define MAX_MIRRORS 32
96 typedef uint32_t mirror_mask_t;
97 #define MIRROR_MASK_C(X) UINT32_C(X)
98 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
100 struct bridge *bridge;
104 /* Selection criteria. */
105 struct svec src_ports;
106 struct svec dst_ports;
111 struct port *out_port;
115 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
117 struct bridge *bridge;
119 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
120 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
123 /* An ordinary bridge port has 1 interface.
124 * A bridge port for bonding has at least 2 interfaces. */
125 struct iface **ifaces;
126 size_t n_ifaces, allocated_ifaces;
129 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
130 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
131 tag_type active_iface_tag; /* Tag for bcast flows. */
132 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
133 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
134 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
136 /* Port mirroring info. */
137 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
138 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
139 bool is_mirror_output_port; /* Does port mirroring send frames here? */
141 /* Spanning tree info. */
142 enum stp_state stp_state; /* Always STP_FORWARDING if STP not in use. */
143 tag_type stp_state_tag; /* Tag for STP state change. */
146 #define DP_MAX_PORTS 255
148 struct list node; /* Node in global list of bridges. */
149 char *name; /* User-specified arbitrary name. */
150 struct mac_learning *ml; /* MAC learning table, or null not to learn. */
151 bool sent_config_request; /* Successfully sent config request? */
152 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
154 /* Support for remote controllers. */
155 char *controller; /* NULL if there is no remote controller;
156 * "discover" to do controller discovery;
157 * otherwise a vconn name. */
159 /* OpenFlow switch processing. */
160 struct ofproto *ofproto; /* OpenFlow switch. */
162 /* Kernel datapath information. */
163 struct dpif *dpif; /* Datapath. */
164 struct port_array ifaces; /* Indexed by kernel datapath port number. */
168 size_t n_ports, allocated_ports;
171 bool has_bonded_ports;
172 long long int bond_next_rebalance;
177 /* Flow statistics gathering. */
178 time_t next_stats_request;
180 /* Port mirroring. */
181 struct mirror *mirrors[MAX_MIRRORS];
185 long long int stp_last_tick;
188 /* List of all bridges. */
189 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
191 /* Maximum number of datapaths. */
192 enum { DP_MAX = 256 };
194 static struct bridge *bridge_create(const char *name);
195 static void bridge_destroy(struct bridge *);
196 static struct bridge *bridge_lookup(const char *name);
197 static void bridge_unixctl_dump_flows(struct unixctl_conn *, const char *);
198 static int bridge_run_one(struct bridge *);
199 static void bridge_reconfigure_one(struct bridge *);
200 static void bridge_reconfigure_controller(struct bridge *);
201 static void bridge_get_all_ifaces(const struct bridge *, struct svec *ifaces);
202 static void bridge_fetch_dp_ifaces(struct bridge *);
203 static void bridge_flush(struct bridge *);
204 static void bridge_pick_local_hw_addr(struct bridge *,
205 uint8_t ea[ETH_ADDR_LEN],
206 struct iface **hw_addr_iface);
207 static uint64_t bridge_pick_datapath_id(struct bridge *,
208 const uint8_t bridge_ea[ETH_ADDR_LEN],
209 struct iface *hw_addr_iface);
210 static struct iface *bridge_get_local_iface(struct bridge *);
211 static uint64_t dpid_from_hash(const void *, size_t nbytes);
213 static void bridge_unixctl_fdb_show(struct unixctl_conn *, const char *args);
215 static void bond_init(void);
216 static void bond_run(struct bridge *);
217 static void bond_wait(struct bridge *);
218 static void bond_rebalance_port(struct port *);
219 static void bond_send_learning_packets(struct port *);
221 static void port_create(struct bridge *, const char *name);
222 static void port_reconfigure(struct port *);
223 static void port_destroy(struct port *);
224 static struct port *port_lookup(const struct bridge *, const char *name);
225 static struct iface *port_lookup_iface(const struct port *, const char *name);
226 static struct port *port_from_dp_ifidx(const struct bridge *,
228 static void port_update_bond_compat(struct port *);
229 static void port_update_vlan_compat(struct port *);
230 static void port_update_bonding(struct port *);
232 static void mirror_create(struct bridge *, const char *name);
233 static void mirror_destroy(struct mirror *);
234 static void mirror_reconfigure(struct bridge *);
235 static void mirror_reconfigure_one(struct mirror *);
236 static bool vlan_is_mirrored(const struct mirror *, int vlan);
238 static void brstp_reconfigure(struct bridge *);
239 static void brstp_adjust_timers(struct bridge *);
240 static void brstp_run(struct bridge *);
241 static void brstp_wait(struct bridge *);
243 static void iface_create(struct port *, const char *name);
244 static void iface_destroy(struct iface *);
245 static struct iface *iface_lookup(const struct bridge *, const char *name);
246 static struct iface *iface_from_dp_ifidx(const struct bridge *,
249 /* Hooks into ofproto processing. */
250 static struct ofhooks bridge_ofhooks;
252 /* Public functions. */
254 /* Adds the name of each interface used by a bridge, including local and
255 * internal ports, to 'svec'. */
257 bridge_get_ifaces(struct svec *svec)
259 struct bridge *br, *next;
262 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
263 for (i = 0; i < br->n_ports; i++) {
264 struct port *port = br->ports[i];
266 for (j = 0; j < port->n_ifaces; j++) {
267 struct iface *iface = port->ifaces[j];
268 if (iface->dp_ifidx < 0) {
269 VLOG_ERR("%s interface not in datapath %s, ignoring",
270 iface->name, dpif_name(br->dpif));
272 if (iface->dp_ifidx != ODPP_LOCAL) {
273 svec_add(svec, iface->name);
281 /* The caller must already have called cfg_read(). */
285 struct svec dpif_names;
288 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show);
290 svec_init(&dpif_names);
291 dp_enumerate(&dpif_names);
292 for (i = 0; i < dpif_names.n; i++) {
293 const char *dpif_name = dpif_names.names[i];
297 retval = dpif_open(dpif_name, &dpif);
299 struct svec all_names;
302 svec_init(&all_names);
303 dpif_get_all_names(dpif, &all_names);
304 for (j = 0; j < all_names.n; j++) {
305 if (cfg_has("bridge.%s.port", all_names.names[j])) {
311 svec_destroy(&all_names);
316 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows);
319 bridge_reconfigure();
324 config_string_change(const char *key, char **valuep)
326 const char *value = cfg_get_string(0, "%s", key);
327 if (value && (!*valuep || strcmp(value, *valuep))) {
329 *valuep = xstrdup(value);
337 bridge_configure_ssl(void)
339 /* XXX SSL should be configurable on a per-bridge basis.
340 * XXX should be possible to de-configure SSL. */
341 static char *private_key_file;
342 static char *certificate_file;
343 static char *cacert_file;
346 if (config_string_change("ssl.private-key", &private_key_file)) {
347 vconn_ssl_set_private_key_file(private_key_file);
350 if (config_string_change("ssl.certificate", &certificate_file)) {
351 vconn_ssl_set_certificate_file(certificate_file);
354 /* We assume that even if the filename hasn't changed, if the CA cert
355 * file has been removed, that we want to move back into
356 * boot-strapping mode. This opens a small security hole, because
357 * the old certificate will still be trusted until vSwitch is
358 * restarted. We may want to address this in vconn's SSL library. */
359 if (config_string_change("ssl.ca-cert", &cacert_file)
360 || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) {
361 vconn_ssl_set_ca_cert_file(cacert_file,
362 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
367 /* iterate_and_prune_ifaces() callback function that opens the network device
368 * for 'iface', if it is not already open, and retrieves the interface's MAC
369 * address and carrier status. */
371 init_iface_netdev(struct bridge *br UNUSED, struct iface *iface,
376 } else if (!netdev_open(iface->name, NETDEV_ETH_TYPE_NONE,
378 netdev_get_carrier(iface->netdev, &iface->enabled);
381 /* If the network device can't be opened, then we're not going to try
382 * to do anything with this interface. */
388 check_iface_dp_ifidx(struct bridge *br, struct iface *iface, void *aux UNUSED)
390 if (iface->dp_ifidx >= 0) {
391 VLOG_DBG("%s has interface %s on port %d",
393 iface->name, iface->dp_ifidx);
396 VLOG_ERR("%s interface not in %s, dropping",
397 iface->name, dpif_name(br->dpif));
403 set_iface_policing(struct bridge *br UNUSED, struct iface *iface,
406 int rate = cfg_get_int(0, "port.%s.ingress.policing-rate", iface->name);
407 int burst = cfg_get_int(0, "port.%s.ingress.policing-burst", iface->name);
408 netdev_set_policing(iface->netdev, rate, burst);
412 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
413 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
414 * deletes from 'br' any ports that no longer have any interfaces. */
416 iterate_and_prune_ifaces(struct bridge *br,
417 bool (*cb)(struct bridge *, struct iface *,
423 for (i = 0; i < br->n_ports; ) {
424 struct port *port = br->ports[i];
425 for (j = 0; j < port->n_ifaces; ) {
426 struct iface *iface = port->ifaces[j];
427 if (cb(br, iface, aux)) {
430 iface_destroy(iface);
434 if (port->n_ifaces) {
437 VLOG_ERR("%s port has no interfaces, dropping", port->name);
444 bridge_reconfigure(void)
446 struct svec old_br, new_br;
447 struct bridge *br, *next;
450 COVERAGE_INC(bridge_reconfigure);
452 /* Collect old and new bridges. */
455 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
456 svec_add(&old_br, br->name);
458 cfg_get_subsections(&new_br, "bridge");
460 /* Get rid of deleted bridges and add new bridges. */
463 assert(svec_is_unique(&old_br));
464 assert(svec_is_unique(&new_br));
465 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
466 if (!svec_contains(&new_br, br->name)) {
470 for (i = 0; i < new_br.n; i++) {
471 const char *name = new_br.names[i];
472 if (!svec_contains(&old_br, name)) {
476 svec_destroy(&old_br);
477 svec_destroy(&new_br);
481 bridge_configure_ssl();
484 /* Reconfigure all bridges. */
485 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
486 bridge_reconfigure_one(br);
489 /* Add and delete ports on all datapaths.
491 * The kernel will reject any attempt to add a given port to a datapath if
492 * that port already belongs to a different datapath, so we must do all
493 * port deletions before any port additions. */
494 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
495 struct odp_port *dpif_ports;
497 struct svec want_ifaces;
499 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
500 bridge_get_all_ifaces(br, &want_ifaces);
501 for (i = 0; i < n_dpif_ports; i++) {
502 const struct odp_port *p = &dpif_ports[i];
503 if (!svec_contains(&want_ifaces, p->devname)
504 && strcmp(p->devname, br->name)) {
505 int retval = dpif_port_del(br->dpif, p->port);
507 VLOG_ERR("failed to remove %s interface from %s: %s",
508 p->devname, dpif_name(br->dpif),
513 svec_destroy(&want_ifaces);
516 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
517 struct odp_port *dpif_ports;
519 struct svec cur_ifaces, want_ifaces, add_ifaces;
521 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
522 svec_init(&cur_ifaces);
523 for (i = 0; i < n_dpif_ports; i++) {
524 svec_add(&cur_ifaces, dpif_ports[i].devname);
527 svec_sort_unique(&cur_ifaces);
528 bridge_get_all_ifaces(br, &want_ifaces);
529 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
531 for (i = 0; i < add_ifaces.n; i++) {
532 const char *if_name = add_ifaces.names[i];
536 /* It's an internal interface if it's marked that way, or if
537 * it's a bonded interface for which we're faking up a network
539 internal = cfg_get_bool(0, "iface.%s.internal", if_name);
540 if (cfg_get_bool(0, "bonding.%s.fake-iface", if_name)) {
541 struct port *port = port_lookup(br, if_name);
542 if (port && port->n_ifaces > 1) {
547 /* Add to datapath. */
548 error = dpif_port_add(br->dpif, if_name,
549 internal ? ODP_PORT_INTERNAL : 0, NULL);
550 if (error == EFBIG) {
551 VLOG_ERR("ran out of valid port numbers on %s",
552 dpif_name(br->dpif));
555 VLOG_ERR("failed to add %s interface to %s: %s",
556 if_name, dpif_name(br->dpif), strerror(error));
559 svec_destroy(&cur_ifaces);
560 svec_destroy(&want_ifaces);
561 svec_destroy(&add_ifaces);
563 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
566 struct iface *local_iface;
567 struct iface *hw_addr_iface;
568 uint8_t engine_type, engine_id;
569 bool add_id_to_iface = false;
570 struct svec nf_hosts;
572 bridge_fetch_dp_ifaces(br);
573 iterate_and_prune_ifaces(br, init_iface_netdev, NULL);
575 iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
577 /* Pick local port hardware address, datapath ID. */
578 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
579 local_iface = bridge_get_local_iface(br);
581 int error = netdev_set_etheraddr(local_iface->netdev, ea);
583 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
584 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
585 "Ethernet address: %s",
586 br->name, strerror(error));
590 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
591 ofproto_set_datapath_id(br->ofproto, dpid);
593 /* Set NetFlow configuration on this bridge. */
594 dpif_get_netflow_ids(br->dpif, &engine_type, &engine_id);
595 if (cfg_has("netflow.%s.engine-type", br->name)) {
596 engine_type = cfg_get_int(0, "netflow.%s.engine-type",
599 if (cfg_has("netflow.%s.engine-id", br->name)) {
600 engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name);
602 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
603 add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface",
606 if (add_id_to_iface && engine_id > 0x7f) {
607 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
608 "another vswitch, choose an engine id less than 128",
611 if (add_id_to_iface && br->n_ports > 0x1ff) {
612 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
613 "another port when 512 or more ports are used",
616 svec_init(&nf_hosts);
617 cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name);
618 if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type,
619 engine_id, add_id_to_iface)) {
620 VLOG_ERR("bridge %s: problem setting netflow collectors",
624 /* Update the controller and related settings. It would be more
625 * straightforward to call this from bridge_reconfigure_one(), but we
626 * can't do it there for two reasons. First, and most importantly, at
627 * that point we don't know the dp_ifidx of any interfaces that have
628 * been added to the bridge (because we haven't actually added them to
629 * the datapath). Second, at that point we haven't set the datapath ID
630 * yet; when a controller is configured, resetting the datapath ID will
631 * immediately disconnect from the controller, so it's better to set
632 * the datapath ID before the controller. */
633 bridge_reconfigure_controller(br);
635 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
636 for (i = 0; i < br->n_ports; i++) {
637 struct port *port = br->ports[i];
638 port_update_vlan_compat(port);
639 port_update_bonding(port);
642 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
643 brstp_reconfigure(br);
644 iterate_and_prune_ifaces(br, set_iface_policing, NULL);
649 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
650 struct iface **hw_addr_iface)
652 uint64_t requested_ea;
656 *hw_addr_iface = NULL;
658 /* Did the user request a particular MAC? */
659 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
661 eth_addr_from_uint64(requested_ea, ea);
662 if (eth_addr_is_multicast(ea)) {
663 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
664 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
665 } else if (eth_addr_is_zero(ea)) {
666 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
672 /* Otherwise choose the minimum MAC address among all of the interfaces.
673 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
674 * MAC of the physical interface in such an environment.) */
675 memset(ea, 0xff, sizeof ea);
676 for (i = 0; i < br->n_ports; i++) {
677 struct port *port = br->ports[i];
678 uint8_t iface_ea[ETH_ADDR_LEN];
679 uint64_t iface_ea_u64;
682 /* Mirror output ports don't participate. */
683 if (port->is_mirror_output_port) {
687 /* Choose the MAC address to represent the port. */
688 iface_ea_u64 = cfg_get_mac(0, "port.%s.mac", port->name);
690 /* User specified explicitly. */
691 eth_addr_from_uint64(iface_ea_u64, iface_ea);
693 /* Find the interface with this Ethernet address (if any) so that
694 * we can provide the correct devname to the caller. */
696 for (j = 0; j < port->n_ifaces; j++) {
697 struct iface *candidate = port->ifaces[j];
698 uint8_t candidate_ea[ETH_ADDR_LEN];
699 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
700 && eth_addr_equals(iface_ea, candidate_ea)) {
705 /* Choose the interface whose MAC address will represent the port.
706 * The Linux kernel bonding code always chooses the MAC address of
707 * the first slave added to a bond, and the Fedora networking
708 * scripts always add slaves to a bond in alphabetical order, so
709 * for compatibility we choose the interface with the name that is
710 * first in alphabetical order. */
711 iface = port->ifaces[0];
712 for (j = 1; j < port->n_ifaces; j++) {
713 struct iface *candidate = port->ifaces[j];
714 if (strcmp(candidate->name, iface->name) < 0) {
719 /* The local port doesn't count (since we're trying to choose its
720 * MAC address anyway). Other internal ports don't count because
721 * we really want a physical MAC if we can get it, and internal
722 * ports typically have randomly generated MACs. */
723 if (iface->dp_ifidx == ODPP_LOCAL
724 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
729 error = netdev_get_etheraddr(iface->netdev, iface_ea);
731 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
732 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
733 iface->name, strerror(error));
738 /* Compare against our current choice. */
739 if (!eth_addr_is_multicast(iface_ea) &&
740 !eth_addr_is_reserved(iface_ea) &&
741 !eth_addr_is_zero(iface_ea) &&
742 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
744 memcpy(ea, iface_ea, ETH_ADDR_LEN);
745 *hw_addr_iface = iface;
748 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
749 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
750 *hw_addr_iface = NULL;
751 VLOG_WARN("bridge %s: using default bridge Ethernet "
752 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
754 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
755 br->name, ETH_ADDR_ARGS(ea));
759 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
760 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
761 * an interface on 'br', then that interface must be passed in as
762 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
763 * 'hw_addr_iface' must be passed in as a null pointer. */
765 bridge_pick_datapath_id(struct bridge *br,
766 const uint8_t bridge_ea[ETH_ADDR_LEN],
767 struct iface *hw_addr_iface)
770 * The procedure for choosing a bridge MAC address will, in the most
771 * ordinary case, also choose a unique MAC that we can use as a datapath
772 * ID. In some special cases, though, multiple bridges will end up with
773 * the same MAC address. This is OK for the bridges, but it will confuse
774 * the OpenFlow controller, because each datapath needs a unique datapath
777 * Datapath IDs must be unique. It is also very desirable that they be
778 * stable from one run to the next, so that policy set on a datapath
783 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
790 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
792 * A bridge whose MAC address is taken from a VLAN network device
793 * (that is, a network device created with vconfig(8) or similar
794 * tool) will have the same MAC address as a bridge on the VLAN
795 * device's physical network device.
797 * Handle this case by hashing the physical network device MAC
798 * along with the VLAN identifier.
800 uint8_t buf[ETH_ADDR_LEN + 2];
801 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
802 buf[ETH_ADDR_LEN] = vlan >> 8;
803 buf[ETH_ADDR_LEN + 1] = vlan;
804 return dpid_from_hash(buf, sizeof buf);
807 * Assume that this bridge's MAC address is unique, since it
808 * doesn't fit any of the cases we handle specially.
813 * A purely internal bridge, that is, one that has no non-virtual
814 * network devices on it at all, is more difficult because it has no
815 * natural unique identifier at all.
817 * When the host is a XenServer, we handle this case by hashing the
818 * host's UUID with the name of the bridge. Names of bridges are
819 * persistent across XenServer reboots, although they can be reused if
820 * an internal network is destroyed and then a new one is later
821 * created, so this is fairly effective.
823 * When the host is not a XenServer, we punt by using a random MAC
824 * address on each run.
826 const char *host_uuid = xenserver_get_host_uuid();
828 char *combined = xasprintf("%s,%s", host_uuid, br->name);
829 dpid = dpid_from_hash(combined, strlen(combined));
835 return eth_addr_to_uint64(bridge_ea);
839 dpid_from_hash(const void *data, size_t n)
841 uint8_t hash[SHA1_DIGEST_SIZE];
843 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
844 sha1_bytes(data, n, hash);
845 eth_addr_mark_random(hash);
846 return eth_addr_to_uint64(hash);
852 struct bridge *br, *next;
856 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
857 int error = bridge_run_one(br);
859 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
860 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
861 "forcing reconfiguration", br->name);
875 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
876 ofproto_wait(br->ofproto);
877 if (br->controller) {
882 mac_learning_wait(br->ml);
889 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
890 * configuration changes. */
892 bridge_flush(struct bridge *br)
894 COVERAGE_INC(bridge_flush);
897 mac_learning_flush(br->ml);
901 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
903 static struct iface *
904 bridge_get_local_iface(struct bridge *br)
908 for (i = 0; i < br->n_ports; i++) {
909 struct port *port = br->ports[i];
910 for (j = 0; j < port->n_ifaces; j++) {
911 struct iface *iface = port->ifaces[j];
912 if (iface->dp_ifidx == ODPP_LOCAL) {
921 /* Bridge unixctl user interface functions. */
923 bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args)
925 struct ds ds = DS_EMPTY_INITIALIZER;
926 const struct bridge *br;
928 br = bridge_lookup(args);
930 unixctl_command_reply(conn, 501, "no such bridge");
934 ds_put_cstr(&ds, " port VLAN MAC Age\n");
936 const struct mac_entry *e;
937 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
938 if (e->port < 0 || e->port >= br->n_ports) {
941 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
942 br->ports[e->port]->ifaces[0]->dp_ifidx,
943 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
946 unixctl_command_reply(conn, 200, ds_cstr(&ds));
950 /* Bridge reconfiguration functions. */
952 static struct bridge *
953 bridge_create(const char *name)
958 assert(!bridge_lookup(name));
959 br = xcalloc(1, sizeof *br);
961 error = dpif_create(name, &br->dpif);
962 if (error == EEXIST || error == EBUSY) {
963 error = dpif_open(name, &br->dpif);
965 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
966 name, strerror(error));
970 dpif_flow_flush(br->dpif);
972 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
977 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
979 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
980 dpif_delete(br->dpif);
981 dpif_close(br->dpif);
986 br->name = xstrdup(name);
987 br->ml = mac_learning_create();
988 br->sent_config_request = false;
989 eth_addr_random(br->default_ea);
991 port_array_init(&br->ifaces);
994 br->bond_next_rebalance = time_msec() + 10000;
996 list_push_back(&all_bridges, &br->node);
998 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1004 bridge_destroy(struct bridge *br)
1009 while (br->n_ports > 0) {
1010 port_destroy(br->ports[br->n_ports - 1]);
1012 list_remove(&br->node);
1013 error = dpif_delete(br->dpif);
1014 if (error && error != ENOENT) {
1015 VLOG_ERR("failed to delete %s: %s",
1016 dpif_name(br->dpif), strerror(error));
1018 dpif_close(br->dpif);
1019 ofproto_destroy(br->ofproto);
1020 free(br->controller);
1021 mac_learning_destroy(br->ml);
1022 port_array_destroy(&br->ifaces);
1029 static struct bridge *
1030 bridge_lookup(const char *name)
1034 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
1035 if (!strcmp(br->name, name)) {
1043 bridge_exists(const char *name)
1045 return bridge_lookup(name) ? true : false;
1049 bridge_get_datapathid(const char *name)
1051 struct bridge *br = bridge_lookup(name);
1052 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
1055 /* Handle requests for a listing of all flows known by the OpenFlow
1056 * stack, including those normally hidden. */
1058 bridge_unixctl_dump_flows(struct unixctl_conn *conn, const char *args)
1063 br = bridge_lookup(args);
1065 unixctl_command_reply(conn, 501, "Unknown bridge");
1070 ofproto_get_all_flows(br->ofproto, &results);
1072 unixctl_command_reply(conn, 200, ds_cstr(&results));
1073 ds_destroy(&results);
1077 bridge_run_one(struct bridge *br)
1081 error = ofproto_run1(br->ofproto);
1087 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1092 error = ofproto_run2(br->ofproto, br->flush);
1099 bridge_get_controller(const struct bridge *br)
1101 const char *controller;
1103 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
1105 controller = cfg_get_string(0, "mgmt.controller");
1107 return controller && controller[0] ? controller : NULL;
1111 check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
1113 struct svec *ifaces = ifaces_;
1114 if (!svec_contains(ifaces, iface->name)) {
1115 svec_add(ifaces, iface->name);
1119 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1121 br->name, iface->name, iface->port->name);
1127 bridge_reconfigure_one(struct bridge *br)
1129 struct svec old_ports, new_ports, ifaces;
1130 struct svec listeners, old_listeners;
1131 struct svec snoops, old_snoops;
1134 /* Collect old ports. */
1135 svec_init(&old_ports);
1136 for (i = 0; i < br->n_ports; i++) {
1137 svec_add(&old_ports, br->ports[i]->name);
1139 svec_sort(&old_ports);
1140 assert(svec_is_unique(&old_ports));
1142 /* Collect new ports. */
1143 svec_init(&new_ports);
1144 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
1145 svec_sort(&new_ports);
1146 if (bridge_get_controller(br)) {
1147 char local_name[IF_NAMESIZE];
1150 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1151 local_name, sizeof local_name);
1152 if (!error && !svec_contains(&new_ports, local_name)) {
1153 svec_add(&new_ports, local_name);
1154 svec_sort(&new_ports);
1157 if (!svec_is_unique(&new_ports)) {
1158 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1159 br->name, svec_get_duplicate(&new_ports));
1160 svec_unique(&new_ports);
1163 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
1165 /* Get rid of deleted ports and add new ports. */
1166 for (i = 0; i < br->n_ports; ) {
1167 struct port *port = br->ports[i];
1168 if (!svec_contains(&new_ports, port->name)) {
1174 for (i = 0; i < new_ports.n; i++) {
1175 const char *name = new_ports.names[i];
1176 if (!svec_contains(&old_ports, name)) {
1177 port_create(br, name);
1180 svec_destroy(&old_ports);
1181 svec_destroy(&new_ports);
1183 /* Reconfigure all ports. */
1184 for (i = 0; i < br->n_ports; i++) {
1185 port_reconfigure(br->ports[i]);
1188 /* Check and delete duplicate interfaces. */
1190 iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
1191 svec_destroy(&ifaces);
1193 /* Delete all flows if we're switching from connected to standalone or vice
1194 * versa. (XXX Should we delete all flows if we are switching from one
1195 * controller to another?) */
1197 /* Configure OpenFlow management listeners. */
1198 svec_init(&listeners);
1199 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1201 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1202 ovs_rundir, br->name));
1203 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1204 svec_clear(&listeners);
1206 svec_sort_unique(&listeners);
1208 svec_init(&old_listeners);
1209 ofproto_get_listeners(br->ofproto, &old_listeners);
1210 svec_sort_unique(&old_listeners);
1212 if (!svec_equal(&listeners, &old_listeners)) {
1213 ofproto_set_listeners(br->ofproto, &listeners);
1215 svec_destroy(&listeners);
1216 svec_destroy(&old_listeners);
1218 /* Configure OpenFlow controller connection snooping. */
1220 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1222 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1223 ovs_rundir, br->name));
1224 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1225 svec_clear(&snoops);
1227 svec_sort_unique(&snoops);
1229 svec_init(&old_snoops);
1230 ofproto_get_snoops(br->ofproto, &old_snoops);
1231 svec_sort_unique(&old_snoops);
1233 if (!svec_equal(&snoops, &old_snoops)) {
1234 ofproto_set_snoops(br->ofproto, &snoops);
1236 svec_destroy(&snoops);
1237 svec_destroy(&old_snoops);
1239 mirror_reconfigure(br);
1243 bridge_reconfigure_controller(struct bridge *br)
1245 char *pfx = xasprintf("bridge.%s.controller", br->name);
1246 const char *controller;
1248 controller = bridge_get_controller(br);
1249 if ((br->controller != NULL) != (controller != NULL)) {
1250 ofproto_flush_flows(br->ofproto);
1252 free(br->controller);
1253 br->controller = controller ? xstrdup(controller) : NULL;
1256 const char *fail_mode;
1257 int max_backoff, probe;
1258 int rate_limit, burst_limit;
1260 if (!strcmp(controller, "discover")) {
1261 bool update_resolv_conf = true;
1263 if (cfg_has("%s.update-resolv.conf", pfx)) {
1264 update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf",
1267 ofproto_set_discovery(br->ofproto, true,
1268 cfg_get_string(0, "%s.accept-regex", pfx),
1269 update_resolv_conf);
1271 struct iface *local_iface;
1274 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1276 || cfg_get_bool(0, "%s.in-band", pfx));
1277 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1278 ofproto_set_in_band(br->ofproto, in_band);
1280 local_iface = bridge_get_local_iface(br);
1282 && cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1283 struct netdev *netdev = local_iface->netdev;
1284 struct in_addr ip, mask, gateway;
1285 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1286 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1287 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1289 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1291 mask.s_addr = guess_netmask(ip.s_addr);
1293 if (!netdev_set_in4(netdev, ip, mask)) {
1294 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1296 br->name, IP_ARGS(&ip.s_addr),
1297 IP_ARGS(&mask.s_addr));
1300 if (gateway.s_addr) {
1301 if (!netdev_add_router(netdev, gateway)) {
1302 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1303 br->name, IP_ARGS(&gateway.s_addr));
1309 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1311 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1313 ofproto_set_failure(br->ofproto,
1315 || !strcmp(fail_mode, "standalone")
1316 || !strcmp(fail_mode, "open")));
1318 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1320 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1325 ofproto_set_probe_interval(br->ofproto, probe);
1327 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1329 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1334 ofproto_set_max_backoff(br->ofproto, max_backoff);
1336 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1338 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1340 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1342 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1344 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1346 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1348 if (cfg_has("%s.commands.acl", pfx)) {
1349 struct svec command_acls;
1352 svec_init(&command_acls);
1353 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1354 command_acl = svec_join(&command_acls, ",", "");
1356 ofproto_set_remote_execution(br->ofproto, command_acl,
1357 cfg_get_string(0, "%s.commands.dir",
1360 svec_destroy(&command_acls);
1363 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1366 union ofp_action action;
1369 /* Set up a flow that matches every packet and directs them to
1370 * OFPP_NORMAL (which goes to us). */
1371 memset(&action, 0, sizeof action);
1372 action.type = htons(OFPAT_OUTPUT);
1373 action.output.len = htons(sizeof action);
1374 action.output.port = htons(OFPP_NORMAL);
1375 memset(&flow, 0, sizeof flow);
1376 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1379 ofproto_set_in_band(br->ofproto, false);
1380 ofproto_set_max_backoff(br->ofproto, 1);
1381 ofproto_set_probe_interval(br->ofproto, 5);
1382 ofproto_set_failure(br->ofproto, false);
1383 ofproto_set_stp(br->ofproto, false);
1387 ofproto_set_controller(br->ofproto, br->controller);
1391 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1396 for (i = 0; i < br->n_ports; i++) {
1397 struct port *port = br->ports[i];
1398 for (j = 0; j < port->n_ifaces; j++) {
1399 struct iface *iface = port->ifaces[j];
1400 svec_add(ifaces, iface->name);
1402 if (port->n_ifaces > 1
1403 && cfg_get_bool(0, "bonding.%s.fake-iface", port->name)) {
1404 svec_add(ifaces, port->name);
1407 svec_sort_unique(ifaces);
1410 /* For robustness, in case the administrator moves around datapath ports behind
1411 * our back, we re-check all the datapath port numbers here.
1413 * This function will set the 'dp_ifidx' members of interfaces that have
1414 * disappeared to -1, so only call this function from a context where those
1415 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1416 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1417 * datapath, which doesn't support UINT16_MAX+1 ports. */
1419 bridge_fetch_dp_ifaces(struct bridge *br)
1421 struct odp_port *dpif_ports;
1422 size_t n_dpif_ports;
1425 /* Reset all interface numbers. */
1426 for (i = 0; i < br->n_ports; i++) {
1427 struct port *port = br->ports[i];
1428 for (j = 0; j < port->n_ifaces; j++) {
1429 struct iface *iface = port->ifaces[j];
1430 iface->dp_ifidx = -1;
1433 port_array_clear(&br->ifaces);
1435 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1436 for (i = 0; i < n_dpif_ports; i++) {
1437 struct odp_port *p = &dpif_ports[i];
1438 struct iface *iface = iface_lookup(br, p->devname);
1440 if (iface->dp_ifidx >= 0) {
1441 VLOG_WARN("%s reported interface %s twice",
1442 dpif_name(br->dpif), p->devname);
1443 } else if (iface_from_dp_ifidx(br, p->port)) {
1444 VLOG_WARN("%s reported interface %"PRIu16" twice",
1445 dpif_name(br->dpif), p->port);
1447 port_array_set(&br->ifaces, p->port, iface);
1448 iface->dp_ifidx = p->port;
1455 /* Bridge packet processing functions. */
1458 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1460 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1463 static struct bond_entry *
1464 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1466 return &port->bond_hash[bond_hash(mac)];
1470 bond_choose_iface(const struct port *port)
1473 for (i = 0; i < port->n_ifaces; i++) {
1474 if (port->ifaces[i]->enabled) {
1482 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1483 uint16_t *dp_ifidx, tag_type *tags)
1485 struct iface *iface;
1487 assert(port->n_ifaces);
1488 if (port->n_ifaces == 1) {
1489 iface = port->ifaces[0];
1491 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1492 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1493 || !port->ifaces[e->iface_idx]->enabled) {
1494 /* XXX select interface properly. The current interface selection
1495 * is only good for testing the rebalancing code. */
1496 e->iface_idx = bond_choose_iface(port);
1497 if (e->iface_idx < 0) {
1498 *tags |= port->no_ifaces_tag;
1501 e->iface_tag = tag_create_random();
1502 ((struct port *) port)->bond_compat_is_stale = true;
1504 *tags |= e->iface_tag;
1505 iface = port->ifaces[e->iface_idx];
1507 *dp_ifidx = iface->dp_ifidx;
1508 *tags |= iface->tag; /* Currently only used for bonding. */
1513 bond_link_status_update(struct iface *iface, bool carrier)
1515 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1516 struct port *port = iface->port;
1518 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1519 /* Nothing to do. */
1522 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1523 iface->name, carrier ? "detected" : "dropped");
1524 if (carrier == iface->enabled) {
1525 iface->delay_expires = LLONG_MAX;
1526 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1527 iface->name, carrier ? "disabled" : "enabled");
1528 } else if (carrier && port->updelay && port->active_iface < 0) {
1529 iface->delay_expires = time_msec();
1530 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1531 "other interface is up", iface->name, port->updelay);
1533 int delay = carrier ? port->updelay : port->downdelay;
1534 iface->delay_expires = time_msec() + delay;
1537 "interface %s: will be %s if it stays %s for %d ms",
1539 carrier ? "enabled" : "disabled",
1540 carrier ? "up" : "down",
1547 bond_choose_active_iface(struct port *port)
1549 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1551 port->active_iface = bond_choose_iface(port);
1552 port->active_iface_tag = tag_create_random();
1553 if (port->active_iface >= 0) {
1554 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1555 port->name, port->ifaces[port->active_iface]->name);
1557 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1563 bond_enable_slave(struct iface *iface, bool enable)
1565 struct port *port = iface->port;
1566 struct bridge *br = port->bridge;
1568 iface->delay_expires = LLONG_MAX;
1569 if (enable == iface->enabled) {
1573 iface->enabled = enable;
1574 if (!iface->enabled) {
1575 VLOG_WARN("interface %s: disabled", iface->name);
1576 ofproto_revalidate(br->ofproto, iface->tag);
1577 if (iface->port_ifidx == port->active_iface) {
1578 ofproto_revalidate(br->ofproto,
1579 port->active_iface_tag);
1580 bond_choose_active_iface(port);
1582 bond_send_learning_packets(port);
1584 VLOG_WARN("interface %s: enabled", iface->name);
1585 if (port->active_iface < 0) {
1586 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1587 bond_choose_active_iface(port);
1588 bond_send_learning_packets(port);
1590 iface->tag = tag_create_random();
1595 bond_run(struct bridge *br)
1599 for (i = 0; i < br->n_ports; i++) {
1600 struct port *port = br->ports[i];
1602 if (port->bond_compat_is_stale) {
1603 port->bond_compat_is_stale = false;
1604 port_update_bond_compat(port);
1607 if (port->n_ifaces < 2) {
1610 for (j = 0; j < port->n_ifaces; j++) {
1611 struct iface *iface = port->ifaces[j];
1612 if (time_msec() >= iface->delay_expires) {
1613 bond_enable_slave(iface, !iface->enabled);
1620 bond_wait(struct bridge *br)
1624 for (i = 0; i < br->n_ports; i++) {
1625 struct port *port = br->ports[i];
1626 if (port->n_ifaces < 2) {
1629 for (j = 0; j < port->n_ifaces; j++) {
1630 struct iface *iface = port->ifaces[j];
1631 if (iface->delay_expires != LLONG_MAX) {
1632 poll_timer_wait(iface->delay_expires - time_msec());
1639 set_dst(struct dst *p, const flow_t *flow,
1640 const struct port *in_port, const struct port *out_port,
1645 * XXX This uses too many tags: any broadcast flow will get one tag per
1646 * destination port, and thus a broadcast on a switch of any size is likely
1647 * to have all tag bits set. We should figure out a way to be smarter.
1649 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1650 *tags |= out_port->stp_state_tag;
1651 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1655 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1656 : in_port->vlan >= 0 ? in_port->vlan
1657 : ntohs(flow->dl_vlan));
1658 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1662 swap_dst(struct dst *p, struct dst *q)
1664 struct dst tmp = *p;
1669 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1670 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1671 * that we push to the datapath. We could in fact fully sort the array by
1672 * vlan, but in most cases there are at most two different vlan tags so that's
1673 * possibly overkill.) */
1675 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1677 struct dst *first = dsts;
1678 struct dst *last = dsts + n_dsts;
1680 while (first != last) {
1682 * - All dsts < first have vlan == 'vlan'.
1683 * - All dsts >= last have vlan != 'vlan'.
1684 * - first < last. */
1685 while (first->vlan == vlan) {
1686 if (++first == last) {
1691 /* Same invariants, plus one additional:
1692 * - first->vlan != vlan.
1694 while (last[-1].vlan != vlan) {
1695 if (--last == first) {
1700 /* Same invariants, plus one additional:
1701 * - last[-1].vlan == vlan.*/
1702 swap_dst(first++, --last);
1707 mirror_mask_ffs(mirror_mask_t mask)
1709 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1714 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1715 const struct dst *test)
1718 for (i = 0; i < n_dsts; i++) {
1719 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1727 port_trunks_vlan(const struct port *port, uint16_t vlan)
1729 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1733 port_includes_vlan(const struct port *port, uint16_t vlan)
1735 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1739 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1740 const struct port *in_port, const struct port *out_port,
1741 struct dst dsts[], tag_type *tags)
1743 mirror_mask_t mirrors = in_port->src_mirrors;
1744 struct dst *dst = dsts;
1747 *tags |= in_port->stp_state_tag;
1748 if (out_port == FLOOD_PORT) {
1749 /* XXX use ODP_FLOOD if no vlans or bonding. */
1750 /* XXX even better, define each VLAN as a datapath port group */
1751 for (i = 0; i < br->n_ports; i++) {
1752 struct port *port = br->ports[i];
1753 if (port != in_port && port_includes_vlan(port, vlan)
1754 && !port->is_mirror_output_port
1755 && set_dst(dst, flow, in_port, port, tags)) {
1756 mirrors |= port->dst_mirrors;
1760 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1761 mirrors |= out_port->dst_mirrors;
1766 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1767 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1769 if (set_dst(dst, flow, in_port, m->out_port, tags)
1770 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1774 for (i = 0; i < br->n_ports; i++) {
1775 struct port *port = br->ports[i];
1776 if (port_includes_vlan(port, m->out_vlan)
1777 && set_dst(dst, flow, in_port, port, tags))
1779 if (port->vlan < 0) {
1780 dst->vlan = m->out_vlan;
1782 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
1785 if (dst->dp_ifidx == flow->in_port
1786 && dst->vlan == vlan) {
1787 /* Don't send out input port on same VLAN. */
1795 mirrors &= mirrors - 1;
1798 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1803 print_dsts(const struct dst *dsts, size_t n)
1805 for (; n--; dsts++) {
1806 printf(">p%"PRIu16, dsts->dp_ifidx);
1807 if (dsts->vlan != OFP_VLAN_NONE) {
1808 printf("v%"PRIu16, dsts->vlan);
1814 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1815 const struct port *in_port, const struct port *out_port,
1816 tag_type *tags, struct odp_actions *actions)
1818 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1820 const struct dst *p;
1823 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1825 cur_vlan = ntohs(flow->dl_vlan);
1826 for (p = dsts; p < &dsts[n_dsts]; p++) {
1827 union odp_action *a;
1828 if (p->vlan != cur_vlan) {
1829 if (p->vlan == OFP_VLAN_NONE) {
1830 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1832 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1833 a->vlan_vid.vlan_vid = htons(p->vlan);
1837 a = odp_actions_add(actions, ODPAT_OUTPUT);
1838 a->output.port = p->dp_ifidx;
1843 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1845 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1846 return (flow->dl_type == htons(ETH_TYPE_ARP)
1847 && eth_addr_is_broadcast(flow->dl_dst)
1848 && packet->size >= sizeof(struct arp_eth_header)
1849 && arp->ar_op == ARP_OP_REQUEST);
1852 /* If the composed actions may be applied to any packet in the given 'flow',
1853 * returns true. Otherwise, the actions should only be applied to 'packet', or
1854 * not at all, if 'packet' was NULL. */
1856 process_flow(struct bridge *br, const flow_t *flow,
1857 const struct ofpbuf *packet, struct odp_actions *actions,
1860 struct iface *in_iface;
1861 struct port *in_port;
1862 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1865 /* Find the interface and port structure for the received packet. */
1866 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1868 /* No interface? Something fishy... */
1869 if (packet != NULL) {
1870 /* Odd. A few possible reasons here:
1872 * - We deleted an interface but there are still a few packets
1873 * queued up from it.
1875 * - Someone externally added an interface (e.g. with "ovs-dpctl
1876 * add-if") that we don't know about.
1878 * - Packet arrived on the local port but the local port is not
1879 * one of our bridge ports.
1881 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1883 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1884 "interface %"PRIu16, br->name, flow->in_port);
1887 /* Return without adding any actions, to drop packets on this flow. */
1890 in_port = in_iface->port;
1892 /* Figure out what VLAN this packet belongs to.
1894 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1895 * belongs to VLAN 0, so we should treat both cases identically. (In the
1896 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1897 * presumably to allow a priority to be specified. In the latter case, the
1898 * packet does not have any 802.1Q header.) */
1899 vlan = ntohs(flow->dl_vlan);
1900 if (vlan == OFP_VLAN_NONE) {
1903 if (in_port->vlan >= 0) {
1905 /* XXX support double tagging? */
1906 if (packet != NULL) {
1907 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1908 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1909 "packet received on port %s configured with "
1910 "implicit VLAN %"PRIu16,
1911 br->name, ntohs(flow->dl_vlan),
1912 in_port->name, in_port->vlan);
1916 vlan = in_port->vlan;
1918 if (!port_includes_vlan(in_port, vlan)) {
1919 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1920 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1921 "packet received on port %s not configured for "
1923 br->name, vlan, in_port->name, vlan);
1928 /* Drop frames for ports that STP wants entirely killed (both for
1929 * forwarding and for learning). Later, after we do learning, we'll drop
1930 * the frames that STP wants to do learning but not forwarding on. */
1931 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1935 /* Drop frames for reserved multicast addresses. */
1936 if (eth_addr_is_reserved(flow->dl_dst)) {
1940 /* Drop frames on ports reserved for mirroring. */
1941 if (in_port->is_mirror_output_port) {
1942 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1943 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1944 "which is reserved exclusively for mirroring",
1945 br->name, in_port->name);
1949 /* Multicast (and broadcast) packets on bonds need special attention, to
1950 * avoid receiving duplicates. */
1951 if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) {
1952 *tags |= in_port->active_iface_tag;
1953 if (in_port->active_iface != in_iface->port_ifidx) {
1954 /* Drop all multicast packets on inactive slaves. */
1957 /* Drop all multicast packets for which we have learned a different
1958 * input port, because we probably sent the packet on one slaves
1959 * and got it back on the active slave. Broadcast ARP replies are
1960 * an exception to this rule: the host has moved to another
1962 int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1963 if (src_idx != -1 && src_idx != in_port->port_idx) {
1965 if (!is_bcast_arp_reply(flow, packet)) {
1969 /* No way to know whether it's an ARP reply, because the
1970 * flow entry doesn't include enough information and we
1971 * don't have a packet. Punt. */
1979 out_port = FLOOD_PORT;
1983 /* Learn source MAC (but don't try to learn from revalidation). */
1985 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1986 vlan, in_port->port_idx);
1988 /* The log messages here could actually be useful in debugging,
1989 * so keep the rate limit relatively high. */
1990 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
1992 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1993 "on port %s in VLAN %d",
1994 br->name, ETH_ADDR_ARGS(flow->dl_src),
1995 in_port->name, vlan);
1996 ofproto_revalidate(br->ofproto, rev_tag);
2000 /* Determine output port. */
2001 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
2003 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2004 out_port = br->ports[out_port_idx];
2008 /* Don't send packets out their input ports. Don't forward frames that STP
2009 * wants us to discard. */
2010 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
2015 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
2018 * We send out only a single packet, instead of setting up a flow, if the
2019 * packet is an ARP directed to broadcast that arrived on a bonded
2020 * interface. In such a situation ARP requests and replies must be handled
2021 * differently, but OpenFlow unfortunately can't distinguish them.
2023 return (in_port->n_ifaces < 2
2024 || flow->dl_type != htons(ETH_TYPE_ARP)
2025 || !eth_addr_is_broadcast(flow->dl_dst));
2028 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2031 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2032 const struct ofp_phy_port *opp,
2035 struct bridge *br = br_;
2036 struct iface *iface;
2039 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
2045 if (reason == OFPPR_DELETE) {
2046 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2047 br->name, iface->name);
2048 iface_destroy(iface);
2049 if (!port->n_ifaces) {
2050 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2051 br->name, port->name);
2057 if (port->n_ifaces > 1) {
2058 bool up = !(opp->state & OFPPS_LINK_DOWN);
2059 bond_link_status_update(iface, up);
2060 port_update_bond_compat(port);
2066 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2067 struct odp_actions *actions, tag_type *tags, void *br_)
2069 struct bridge *br = br_;
2072 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
2073 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
2074 brstp_receive(br, flow, payload);
2079 COVERAGE_INC(bridge_process_flow);
2080 return process_flow(br, flow, packet, actions, tags);
2084 bridge_account_flow_ofhook_cb(const flow_t *flow,
2085 const union odp_action *actions,
2086 size_t n_actions, unsigned long long int n_bytes,
2089 struct bridge *br = br_;
2090 const union odp_action *a;
2092 if (!br->has_bonded_ports) {
2096 for (a = actions; a < &actions[n_actions]; a++) {
2097 if (a->type == ODPAT_OUTPUT) {
2098 struct port *port = port_from_dp_ifidx(br, a->output.port);
2099 if (port && port->n_ifaces >= 2) {
2100 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
2101 e->tx_bytes += n_bytes;
2108 bridge_account_checkpoint_ofhook_cb(void *br_)
2110 struct bridge *br = br_;
2113 if (!br->has_bonded_ports) {
2117 /* The current ofproto implementation calls this callback at least once a
2118 * second, so this timer implementation is sufficient. */
2119 if (time_msec() < br->bond_next_rebalance) {
2122 br->bond_next_rebalance = time_msec() + 10000;
2124 for (i = 0; i < br->n_ports; i++) {
2125 struct port *port = br->ports[i];
2126 if (port->n_ifaces > 1) {
2127 bond_rebalance_port(port);
2132 static struct ofhooks bridge_ofhooks = {
2133 bridge_port_changed_ofhook_cb,
2134 bridge_normal_ofhook_cb,
2135 bridge_account_flow_ofhook_cb,
2136 bridge_account_checkpoint_ofhook_cb,
2139 /* Bonding functions. */
2141 /* Statistics for a single interface on a bonded port, used for load-based
2142 * bond rebalancing. */
2143 struct slave_balance {
2144 struct iface *iface; /* The interface. */
2145 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2147 /* All the "bond_entry"s that are assigned to this interface, in order of
2148 * increasing tx_bytes. */
2149 struct bond_entry **hashes;
2153 /* Sorts pointers to pointers to bond_entries in ascending order by the
2154 * interface to which they are assigned, and within a single interface in
2155 * ascending order of bytes transmitted. */
2157 compare_bond_entries(const void *a_, const void *b_)
2159 const struct bond_entry *const *ap = a_;
2160 const struct bond_entry *const *bp = b_;
2161 const struct bond_entry *a = *ap;
2162 const struct bond_entry *b = *bp;
2163 if (a->iface_idx != b->iface_idx) {
2164 return a->iface_idx > b->iface_idx ? 1 : -1;
2165 } else if (a->tx_bytes != b->tx_bytes) {
2166 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2172 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2173 * *descending* order by number of bytes transmitted. */
2175 compare_slave_balance(const void *a_, const void *b_)
2177 const struct slave_balance *a = a_;
2178 const struct slave_balance *b = b_;
2179 if (a->iface->enabled != b->iface->enabled) {
2180 return a->iface->enabled ? -1 : 1;
2181 } else if (a->tx_bytes != b->tx_bytes) {
2182 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2189 swap_bals(struct slave_balance *a, struct slave_balance *b)
2191 struct slave_balance tmp = *a;
2196 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2197 * given that 'p' (and only 'p') might be in the wrong location.
2199 * This function invalidates 'p', since it might now be in a different memory
2202 resort_bals(struct slave_balance *p,
2203 struct slave_balance bals[], size_t n_bals)
2206 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2207 swap_bals(p, p - 1);
2209 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2210 swap_bals(p, p + 1);
2216 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2218 if (VLOG_IS_DBG_ENABLED()) {
2219 struct ds ds = DS_EMPTY_INITIALIZER;
2220 const struct slave_balance *b;
2222 for (b = bals; b < bals + n_bals; b++) {
2226 ds_put_char(&ds, ',');
2228 ds_put_format(&ds, " %s %"PRIu64"kB",
2229 b->iface->name, b->tx_bytes / 1024);
2231 if (!b->iface->enabled) {
2232 ds_put_cstr(&ds, " (disabled)");
2234 if (b->n_hashes > 0) {
2235 ds_put_cstr(&ds, " (");
2236 for (i = 0; i < b->n_hashes; i++) {
2237 const struct bond_entry *e = b->hashes[i];
2239 ds_put_cstr(&ds, " + ");
2241 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2242 e - port->bond_hash, e->tx_bytes / 1024);
2244 ds_put_cstr(&ds, ")");
2247 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2252 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2254 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2255 struct bond_entry *hash)
2257 struct port *port = from->iface->port;
2258 uint64_t delta = hash->tx_bytes;
2260 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2261 "from %s to %s (now carrying %"PRIu64"kB and "
2262 "%"PRIu64"kB load, respectively)",
2263 port->name, delta / 1024, hash - port->bond_hash,
2264 from->iface->name, to->iface->name,
2265 (from->tx_bytes - delta) / 1024,
2266 (to->tx_bytes + delta) / 1024);
2268 /* Delete element from from->hashes.
2270 * We don't bother to add the element to to->hashes because not only would
2271 * it require more work, the only purpose it would be to allow that hash to
2272 * be migrated to another slave in this rebalancing run, and there is no
2273 * point in doing that. */
2274 if (from->hashes[0] == hash) {
2277 int i = hash - from->hashes[0];
2278 memmove(from->hashes + i, from->hashes + i + 1,
2279 (from->n_hashes - (i + 1)) * sizeof *from->hashes);
2283 /* Shift load away from 'from' to 'to'. */
2284 from->tx_bytes -= delta;
2285 to->tx_bytes += delta;
2287 /* Arrange for flows to be revalidated. */
2288 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2289 hash->iface_idx = to->iface->port_ifidx;
2290 hash->iface_tag = tag_create_random();
2294 bond_rebalance_port(struct port *port)
2296 struct slave_balance bals[DP_MAX_PORTS];
2298 struct bond_entry *hashes[BOND_MASK + 1];
2299 struct slave_balance *b, *from, *to;
2300 struct bond_entry *e;
2303 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2304 * descending order of tx_bytes, so that bals[0] represents the most
2305 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2308 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2309 * array for each slave_balance structure, we sort our local array of
2310 * hashes in order by slave, so that all of the hashes for a given slave
2311 * become contiguous in memory, and then we point each 'hashes' members of
2312 * a slave_balance structure to the start of a contiguous group. */
2313 n_bals = port->n_ifaces;
2314 for (b = bals; b < &bals[n_bals]; b++) {
2315 b->iface = port->ifaces[b - bals];
2320 for (i = 0; i <= BOND_MASK; i++) {
2321 hashes[i] = &port->bond_hash[i];
2323 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2324 for (i = 0; i <= BOND_MASK; i++) {
2326 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2327 b = &bals[e->iface_idx];
2328 b->tx_bytes += e->tx_bytes;
2330 b->hashes = &hashes[i];
2335 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2336 log_bals(bals, n_bals, port);
2338 /* Discard slaves that aren't enabled (which were sorted to the back of the
2339 * array earlier). */
2340 while (!bals[n_bals - 1].iface->enabled) {
2347 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2348 to = &bals[n_bals - 1];
2349 for (from = bals; from < to; ) {
2350 uint64_t overload = from->tx_bytes - to->tx_bytes;
2351 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2352 /* The extra load on 'from' (and all less-loaded slaves), compared
2353 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2354 * it is less than ~1Mbps. No point in rebalancing. */
2356 } else if (from->n_hashes == 1) {
2357 /* 'from' only carries a single MAC hash, so we can't shift any
2358 * load away from it, even though we want to. */
2361 /* 'from' is carrying significantly more load than 'to', and that
2362 * load is split across at least two different hashes. Pick a hash
2363 * to migrate to 'to' (the least-loaded slave), given that doing so
2364 * must not cause 'to''s load to exceed 'from''s load.
2366 * The sort order we use means that we prefer to shift away the
2367 * smallest hashes instead of the biggest ones. There is little
2368 * reason behind this decision; we could use the opposite sort
2369 * order to shift away big hashes ahead of small ones. */
2372 for (i = 0; i < from->n_hashes; i++) {
2373 uint64_t delta = from->hashes[i]->tx_bytes;
2374 if (to->tx_bytes + delta < from->tx_bytes - delta) {
2378 if (i < from->n_hashes) {
2379 bond_shift_load(from, to, from->hashes[i]);
2381 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2382 * point to different slave_balance structures. It is only
2383 * valid to do these two operations in a row at all because we
2384 * know that 'from' will not move past 'to' and vice versa. */
2385 resort_bals(from, bals, n_bals);
2386 resort_bals(to, bals, n_bals);
2390 port->bond_compat_is_stale = true;
2394 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2395 * historical data to decay to <1% in 7 rebalancing runs. */
2396 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2402 bond_send_learning_packets(struct port *port)
2404 struct bridge *br = port->bridge;
2405 struct mac_entry *e;
2406 struct ofpbuf packet;
2407 int error, n_packets, n_errors;
2409 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2413 ofpbuf_init(&packet, 128);
2414 error = n_packets = n_errors = 0;
2415 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2416 static const char s[] = "Open vSwitch Bond Failover";
2417 union ofp_action actions[2], *a;
2418 struct eth_header *eth;
2419 struct llc_snap_header *llc_snap;
2425 if (e->port == port->port_idx
2426 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2430 /* Compose packet to send. */
2431 ofpbuf_clear(&packet);
2432 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
2433 llc_snap = ofpbuf_put_zeros(&packet, LLC_SNAP_HEADER_LEN);
2434 ofpbuf_put(&packet, s, sizeof s); /* Includes null byte. */
2435 ofpbuf_put(&packet, e->mac, ETH_ADDR_LEN);
2437 memcpy(eth->eth_dst, eth_addr_broadcast, ETH_ADDR_LEN);
2438 memcpy(eth->eth_src, e->mac, ETH_ADDR_LEN);
2439 eth->eth_type = htons(packet.size - ETH_HEADER_LEN);
2441 llc_snap->llc.llc_dsap = LLC_DSAP_SNAP;
2442 llc_snap->llc.llc_ssap = LLC_SSAP_SNAP;
2443 llc_snap->llc.llc_cntl = LLC_CNTL_SNAP;
2444 memcpy(llc_snap->snap.snap_org, "\x00\x23\x20", 3);
2445 llc_snap->snap.snap_type = htons(0xf177); /* Random number. */
2447 /* Compose actions. */
2448 memset(actions, 0, sizeof actions);
2451 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2452 a->vlan_vid.len = htons(sizeof *a);
2453 a->vlan_vid.vlan_vid = htons(e->vlan);
2456 a->output.type = htons(OFPAT_OUTPUT);
2457 a->output.len = htons(sizeof *a);
2458 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2463 flow_extract(&packet, ODPP_NONE, &flow);
2464 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2471 ofpbuf_uninit(&packet);
2474 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2475 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2476 "packets, last error was: %s",
2477 port->name, n_errors, n_packets, strerror(error));
2479 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2480 port->name, n_packets);
2484 /* Bonding unixctl user interface functions. */
2487 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2489 struct ds ds = DS_EMPTY_INITIALIZER;
2490 const struct bridge *br;
2492 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2494 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2497 for (i = 0; i < br->n_ports; i++) {
2498 const struct port *port = br->ports[i];
2499 if (port->n_ifaces > 1) {
2502 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2503 for (j = 0; j < port->n_ifaces; j++) {
2504 const struct iface *iface = port->ifaces[j];
2506 ds_put_cstr(&ds, ", ");
2508 ds_put_cstr(&ds, iface->name);
2510 ds_put_char(&ds, '\n');
2514 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2518 static struct port *
2519 bond_find(const char *name)
2521 const struct bridge *br;
2523 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2526 for (i = 0; i < br->n_ports; i++) {
2527 struct port *port = br->ports[i];
2528 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2537 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2539 struct ds ds = DS_EMPTY_INITIALIZER;
2540 const struct port *port;
2543 port = bond_find(args);
2545 unixctl_command_reply(conn, 501, "no such bond");
2549 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2550 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2551 ds_put_format(&ds, "next rebalance: %lld ms\n",
2552 port->bridge->bond_next_rebalance - time_msec());
2553 for (j = 0; j < port->n_ifaces; j++) {
2554 const struct iface *iface = port->ifaces[j];
2555 struct bond_entry *be;
2558 ds_put_format(&ds, "slave %s: %s\n",
2559 iface->name, iface->enabled ? "enabled" : "disabled");
2560 if (j == port->active_iface) {
2561 ds_put_cstr(&ds, "\tactive slave\n");
2563 if (iface->delay_expires != LLONG_MAX) {
2564 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2565 iface->enabled ? "downdelay" : "updelay",
2566 iface->delay_expires - time_msec());
2570 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2571 int hash = be - port->bond_hash;
2572 struct mac_entry *me;
2574 if (be->iface_idx != j) {
2578 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2579 hash, be->tx_bytes / 1024);
2582 if (!port->bridge->ml) {
2586 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2587 &port->bridge->ml->lrus) {
2590 if (bond_hash(me->mac) == hash
2591 && me->port != port->port_idx
2592 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2593 && dp_ifidx == iface->dp_ifidx)
2595 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2596 ETH_ADDR_ARGS(me->mac));
2601 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2606 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2608 char *args = (char *) args_;
2609 char *save_ptr = NULL;
2610 char *bond_s, *hash_s, *slave_s;
2611 uint8_t mac[ETH_ADDR_LEN];
2613 struct iface *iface;
2614 struct bond_entry *entry;
2617 bond_s = strtok_r(args, " ", &save_ptr);
2618 hash_s = strtok_r(NULL, " ", &save_ptr);
2619 slave_s = strtok_r(NULL, " ", &save_ptr);
2621 unixctl_command_reply(conn, 501,
2622 "usage: bond/migrate BOND HASH SLAVE");
2626 port = bond_find(bond_s);
2628 unixctl_command_reply(conn, 501, "no such bond");
2632 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2633 == ETH_ADDR_SCAN_COUNT) {
2634 hash = bond_hash(mac);
2635 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2636 hash = atoi(hash_s) & BOND_MASK;
2638 unixctl_command_reply(conn, 501, "bad hash");
2642 iface = port_lookup_iface(port, slave_s);
2644 unixctl_command_reply(conn, 501, "no such slave");
2648 if (!iface->enabled) {
2649 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2653 entry = &port->bond_hash[hash];
2654 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2655 entry->iface_idx = iface->port_ifidx;
2656 entry->iface_tag = tag_create_random();
2657 port->bond_compat_is_stale = true;
2658 unixctl_command_reply(conn, 200, "migrated");
2662 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2664 char *args = (char *) args_;
2665 char *save_ptr = NULL;
2666 char *bond_s, *slave_s;
2668 struct iface *iface;
2670 bond_s = strtok_r(args, " ", &save_ptr);
2671 slave_s = strtok_r(NULL, " ", &save_ptr);
2673 unixctl_command_reply(conn, 501,
2674 "usage: bond/set-active-slave BOND SLAVE");
2678 port = bond_find(bond_s);
2680 unixctl_command_reply(conn, 501, "no such bond");
2684 iface = port_lookup_iface(port, slave_s);
2686 unixctl_command_reply(conn, 501, "no such slave");
2690 if (!iface->enabled) {
2691 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2695 if (port->active_iface != iface->port_ifidx) {
2696 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2697 port->active_iface = iface->port_ifidx;
2698 port->active_iface_tag = tag_create_random();
2699 VLOG_INFO("port %s: active interface is now %s",
2700 port->name, iface->name);
2701 bond_send_learning_packets(port);
2702 unixctl_command_reply(conn, 200, "done");
2704 unixctl_command_reply(conn, 200, "no change");
2709 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2711 char *args = (char *) args_;
2712 char *save_ptr = NULL;
2713 char *bond_s, *slave_s;
2715 struct iface *iface;
2717 bond_s = strtok_r(args, " ", &save_ptr);
2718 slave_s = strtok_r(NULL, " ", &save_ptr);
2720 unixctl_command_reply(conn, 501,
2721 "usage: bond/enable/disable-slave BOND SLAVE");
2725 port = bond_find(bond_s);
2727 unixctl_command_reply(conn, 501, "no such bond");
2731 iface = port_lookup_iface(port, slave_s);
2733 unixctl_command_reply(conn, 501, "no such slave");
2737 bond_enable_slave(iface, enable);
2738 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2742 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2744 enable_slave(conn, args, true);
2748 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2750 enable_slave(conn, args, false);
2756 unixctl_command_register("bond/list", bond_unixctl_list);
2757 unixctl_command_register("bond/show", bond_unixctl_show);
2758 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2759 unixctl_command_register("bond/set-active-slave",
2760 bond_unixctl_set_active_slave);
2761 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2762 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2765 /* Port functions. */
2768 port_create(struct bridge *br, const char *name)
2772 port = xcalloc(1, sizeof *port);
2774 port->port_idx = br->n_ports;
2776 port->trunks = NULL;
2777 port->name = xstrdup(name);
2778 port->active_iface = -1;
2779 port->stp_state = STP_DISABLED;
2780 port->stp_state_tag = 0;
2782 if (br->n_ports >= br->allocated_ports) {
2783 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2786 br->ports[br->n_ports++] = port;
2788 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2793 port_reconfigure(struct port *port)
2795 bool bonded = cfg_has_section("bonding.%s", port->name);
2796 struct svec old_ifaces, new_ifaces;
2797 unsigned long *trunks;
2801 /* Collect old and new interfaces. */
2802 svec_init(&old_ifaces);
2803 svec_init(&new_ifaces);
2804 for (i = 0; i < port->n_ifaces; i++) {
2805 svec_add(&old_ifaces, port->ifaces[i]->name);
2807 svec_sort(&old_ifaces);
2809 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2810 if (!new_ifaces.n) {
2811 VLOG_ERR("port %s: no interfaces specified for bonded port",
2813 } else if (new_ifaces.n == 1) {
2814 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2818 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2819 if (port->updelay < 0) {
2822 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2823 if (port->downdelay < 0) {
2824 port->downdelay = 0;
2827 svec_init(&new_ifaces);
2828 svec_add(&new_ifaces, port->name);
2831 /* Get rid of deleted interfaces and add new interfaces. */
2832 for (i = 0; i < port->n_ifaces; i++) {
2833 struct iface *iface = port->ifaces[i];
2834 if (!svec_contains(&new_ifaces, iface->name)) {
2835 iface_destroy(iface);
2840 for (i = 0; i < new_ifaces.n; i++) {
2841 const char *name = new_ifaces.names[i];
2842 if (!svec_contains(&old_ifaces, name)) {
2843 iface_create(port, name);
2849 if (cfg_has("vlan.%s.tag", port->name)) {
2851 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2852 if (vlan >= 0 && vlan <= 4095) {
2853 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2856 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2857 * they even work as-is. But they have not been tested. */
2858 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2862 if (port->vlan != vlan) {
2864 bridge_flush(port->bridge);
2867 /* Get trunked VLANs. */
2870 size_t n_trunks, n_errors;
2873 trunks = bitmap_allocate(4096);
2874 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2876 for (i = 0; i < n_trunks; i++) {
2877 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2879 bitmap_set1(trunks, trunk);
2885 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2886 port->name, n_trunks);
2888 if (n_errors == n_trunks) {
2890 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2893 bitmap_set_multiple(trunks, 0, 4096, 1);
2896 if (cfg_has("vlan.%s.trunks", port->name)) {
2897 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2898 port->name, port->name);
2902 ? port->trunks != NULL
2903 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2904 bridge_flush(port->bridge);
2906 bitmap_free(port->trunks);
2907 port->trunks = trunks;
2909 svec_destroy(&old_ifaces);
2910 svec_destroy(&new_ifaces);
2914 port_destroy(struct port *port)
2917 struct bridge *br = port->bridge;
2921 proc_net_compat_update_vlan(port->name, NULL, 0);
2922 proc_net_compat_update_bond(port->name, NULL);
2924 for (i = 0; i < MAX_MIRRORS; i++) {
2925 struct mirror *m = br->mirrors[i];
2926 if (m && m->out_port == port) {
2931 while (port->n_ifaces > 0) {
2932 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2935 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2936 del->port_idx = port->port_idx;
2939 bitmap_free(port->trunks);
2946 static struct port *
2947 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2949 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
2950 return iface ? iface->port : NULL;
2953 static struct port *
2954 port_lookup(const struct bridge *br, const char *name)
2958 for (i = 0; i < br->n_ports; i++) {
2959 struct port *port = br->ports[i];
2960 if (!strcmp(port->name, name)) {
2967 static struct iface *
2968 port_lookup_iface(const struct port *port, const char *name)
2972 for (j = 0; j < port->n_ifaces; j++) {
2973 struct iface *iface = port->ifaces[j];
2974 if (!strcmp(iface->name, name)) {
2982 port_update_bonding(struct port *port)
2984 if (port->n_ifaces < 2) {
2985 /* Not a bonded port. */
2986 if (port->bond_hash) {
2987 free(port->bond_hash);
2988 port->bond_hash = NULL;
2989 port->bond_compat_is_stale = true;
2992 if (!port->bond_hash) {
2995 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
2996 for (i = 0; i <= BOND_MASK; i++) {
2997 struct bond_entry *e = &port->bond_hash[i];
3001 port->no_ifaces_tag = tag_create_random();
3002 bond_choose_active_iface(port);
3004 port->bond_compat_is_stale = true;
3009 port_update_bond_compat(struct port *port)
3011 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3012 struct compat_bond bond;
3015 if (port->n_ifaces < 2) {
3016 proc_net_compat_update_bond(port->name, NULL);
3021 bond.updelay = port->updelay;
3022 bond.downdelay = port->downdelay;
3025 bond.hashes = compat_hashes;
3026 if (port->bond_hash) {
3027 const struct bond_entry *e;
3028 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3029 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3030 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3031 cbh->hash = e - port->bond_hash;
3032 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3037 bond.n_slaves = port->n_ifaces;
3038 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3039 for (i = 0; i < port->n_ifaces; i++) {
3040 struct iface *iface = port->ifaces[i];
3041 struct compat_bond_slave *slave = &bond.slaves[i];
3042 slave->name = iface->name;
3043 slave->up = ((iface->enabled && iface->delay_expires == LLONG_MAX) ||
3044 (!iface->enabled && iface->delay_expires != LLONG_MAX));
3048 netdev_get_etheraddr(iface->netdev, slave->mac);
3051 proc_net_compat_update_bond(port->name, &bond);
3056 port_update_vlan_compat(struct port *port)
3058 struct bridge *br = port->bridge;
3059 char *vlandev_name = NULL;
3061 if (port->vlan > 0) {
3062 /* Figure out the name that the VLAN device should actually have, if it
3063 * existed. This takes some work because the VLAN device would not
3064 * have port->name in its name; rather, it would have the trunk port's
3065 * name, and 'port' would be attached to a bridge that also had the
3066 * VLAN device one of its ports. So we need to find a trunk port that
3067 * includes port->vlan.
3069 * There might be more than one candidate. This doesn't happen on
3070 * XenServer, so if it happens we just pick the first choice in
3071 * alphabetical order instead of creating multiple VLAN devices. */
3073 for (i = 0; i < br->n_ports; i++) {
3074 struct port *p = br->ports[i];
3075 if (port_trunks_vlan(p, port->vlan)
3077 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3079 uint8_t ea[ETH_ADDR_LEN];
3080 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3081 if (!eth_addr_is_multicast(ea) &&
3082 !eth_addr_is_reserved(ea) &&
3083 !eth_addr_is_zero(ea)) {
3084 vlandev_name = p->name;
3089 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3092 /* Interface functions. */
3095 iface_create(struct port *port, const char *name)
3097 struct iface *iface;
3099 iface = xcalloc(1, sizeof *iface);
3101 iface->port_ifidx = port->n_ifaces;
3102 iface->name = xstrdup(name);
3103 iface->dp_ifidx = -1;
3104 iface->tag = tag_create_random();
3105 iface->delay_expires = LLONG_MAX;
3106 iface->netdev = NULL;
3108 if (port->n_ifaces >= port->allocated_ifaces) {
3109 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3110 sizeof *port->ifaces);
3112 port->ifaces[port->n_ifaces++] = iface;
3113 if (port->n_ifaces > 1) {
3114 port->bridge->has_bonded_ports = true;
3117 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3119 bridge_flush(port->bridge);
3123 iface_destroy(struct iface *iface)
3126 struct port *port = iface->port;
3127 struct bridge *br = port->bridge;
3128 bool del_active = port->active_iface == iface->port_ifidx;
3131 if (iface->dp_ifidx >= 0) {
3132 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
3135 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3136 del->port_ifidx = iface->port_ifidx;
3138 netdev_close(iface->netdev);
3143 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3144 bond_choose_active_iface(port);
3145 bond_send_learning_packets(port);
3148 bridge_flush(port->bridge);
3152 static struct iface *
3153 iface_lookup(const struct bridge *br, const char *name)
3157 for (i = 0; i < br->n_ports; i++) {
3158 struct port *port = br->ports[i];
3159 for (j = 0; j < port->n_ifaces; j++) {
3160 struct iface *iface = port->ifaces[j];
3161 if (!strcmp(iface->name, name)) {
3169 static struct iface *
3170 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
3172 return port_array_get(&br->ifaces, dp_ifidx);
3175 /* Port mirroring. */
3178 mirror_reconfigure(struct bridge *br)
3180 struct svec old_mirrors, new_mirrors;
3183 /* Collect old and new mirrors. */
3184 svec_init(&old_mirrors);
3185 svec_init(&new_mirrors);
3186 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
3187 for (i = 0; i < MAX_MIRRORS; i++) {
3188 if (br->mirrors[i]) {
3189 svec_add(&old_mirrors, br->mirrors[i]->name);
3193 /* Get rid of deleted mirrors and add new mirrors. */
3194 svec_sort(&old_mirrors);
3195 assert(svec_is_unique(&old_mirrors));
3196 svec_sort(&new_mirrors);
3197 assert(svec_is_unique(&new_mirrors));
3198 for (i = 0; i < MAX_MIRRORS; i++) {
3199 struct mirror *m = br->mirrors[i];
3200 if (m && !svec_contains(&new_mirrors, m->name)) {
3204 for (i = 0; i < new_mirrors.n; i++) {
3205 const char *name = new_mirrors.names[i];
3206 if (!svec_contains(&old_mirrors, name)) {
3207 mirror_create(br, name);
3210 svec_destroy(&old_mirrors);
3211 svec_destroy(&new_mirrors);
3213 /* Reconfigure all mirrors. */
3214 for (i = 0; i < MAX_MIRRORS; i++) {
3215 if (br->mirrors[i]) {
3216 mirror_reconfigure_one(br->mirrors[i]);
3220 /* Update port reserved status. */
3221 for (i = 0; i < br->n_ports; i++) {
3222 br->ports[i]->is_mirror_output_port = false;
3224 for (i = 0; i < MAX_MIRRORS; i++) {
3225 struct mirror *m = br->mirrors[i];
3226 if (m && m->out_port) {
3227 m->out_port->is_mirror_output_port = true;
3233 mirror_create(struct bridge *br, const char *name)
3238 for (i = 0; ; i++) {
3239 if (i >= MAX_MIRRORS) {
3240 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3241 "cannot create %s", br->name, MAX_MIRRORS, name);
3244 if (!br->mirrors[i]) {
3249 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3252 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3255 m->name = xstrdup(name);
3256 svec_init(&m->src_ports);
3257 svec_init(&m->dst_ports);
3265 mirror_destroy(struct mirror *m)
3268 struct bridge *br = m->bridge;
3271 for (i = 0; i < br->n_ports; i++) {
3272 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3273 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3276 svec_destroy(&m->src_ports);
3277 svec_destroy(&m->dst_ports);
3280 m->bridge->mirrors[m->idx] = NULL;
3288 prune_ports(struct mirror *m, struct svec *ports)
3293 svec_sort_unique(ports);
3296 for (i = 0; i < ports->n; i++) {
3297 const char *name = ports->names[i];
3298 if (port_lookup(m->bridge, name)) {
3299 svec_add(&tmp, name);
3301 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3302 m->bridge->name, m->name, name);
3305 svec_swap(ports, &tmp);
3310 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3314 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3315 * order won't give us numeric sort order. But that's good enough for what
3316 * we need right now. */
3317 svec_sort_unique(vlan_strings);
3319 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3321 for (i = 0; i < vlan_strings->n; i++) {
3322 const char *name = vlan_strings->names[i];
3324 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3325 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3326 m->bridge->name, m->name, name);
3328 (*vlans)[n_vlans++] = vlan;
3335 vlan_is_mirrored(const struct mirror *m, int vlan)
3339 for (i = 0; i < m->n_vlans; i++) {
3340 if (m->vlans[i] == vlan) {
3348 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3352 for (i = 0; i < m->n_vlans; i++) {
3353 if (port_trunks_vlan(p, m->vlans[i])) {
3361 mirror_reconfigure_one(struct mirror *m)
3363 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3364 struct svec src_ports, dst_ports, ports;
3365 struct svec vlan_strings;
3366 mirror_mask_t mirror_bit;
3367 const char *out_port_name;
3368 struct port *out_port;
3373 bool mirror_all_ports;
3374 bool any_ports_specified;
3376 /* Get output port. */
3377 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3378 m->bridge->name, m->name);
3379 if (out_port_name) {
3380 out_port = port_lookup(m->bridge, out_port_name);
3382 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3383 "named %s", pfx, m->bridge->name, out_port_name);
3390 if (cfg_has("%s.output.vlan", pfx)) {
3391 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3392 "ignoring %s.output.vlan", pfx, pfx, pfx);
3394 } else if (cfg_has("%s.output.vlan", pfx)) {
3396 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3398 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3399 "but exactly one is required; disabling port mirror %s",
3400 pfx, pfx, pfx, pfx);
3406 /* Get all the ports, and drop duplicates and ports that don't exist. */
3407 svec_init(&src_ports);
3408 svec_init(&dst_ports);
3410 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3411 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3412 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3413 any_ports_specified = src_ports.n || dst_ports.n || ports.n;
3414 svec_append(&src_ports, &ports);
3415 svec_append(&dst_ports, &ports);
3416 svec_destroy(&ports);
3417 prune_ports(m, &src_ports);
3418 prune_ports(m, &dst_ports);
3419 if (any_ports_specified && !src_ports.n && !dst_ports.n) {
3420 VLOG_ERR("%s: none of the specified ports exist; "
3421 "disabling port mirror %s", pfx, pfx);
3426 /* Get all the vlans, and drop duplicate and invalid vlans. */
3427 svec_init(&vlan_strings);
3428 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3429 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3430 svec_destroy(&vlan_strings);
3432 /* Update mirror data. */
3433 if (!svec_equal(&m->src_ports, &src_ports)
3434 || !svec_equal(&m->dst_ports, &dst_ports)
3435 || m->n_vlans != n_vlans
3436 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3437 || m->out_port != out_port
3438 || m->out_vlan != out_vlan) {
3439 bridge_flush(m->bridge);
3441 svec_swap(&m->src_ports, &src_ports);
3442 svec_swap(&m->dst_ports, &dst_ports);
3445 m->n_vlans = n_vlans;
3446 m->out_port = out_port;
3447 m->out_vlan = out_vlan;
3449 /* If no selection criteria have been given, mirror for all ports. */
3450 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3453 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3454 for (i = 0; i < m->bridge->n_ports; i++) {
3455 struct port *port = m->bridge->ports[i];
3457 if (mirror_all_ports
3458 || svec_contains(&m->src_ports, port->name)
3461 ? port_trunks_any_mirrored_vlan(m, port)
3462 : vlan_is_mirrored(m, port->vlan)))) {
3463 port->src_mirrors |= mirror_bit;
3465 port->src_mirrors &= ~mirror_bit;
3468 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3469 port->dst_mirrors |= mirror_bit;
3471 port->dst_mirrors &= ~mirror_bit;
3477 svec_destroy(&src_ports);
3478 svec_destroy(&dst_ports);
3482 /* Spanning tree protocol. */
3484 static void brstp_update_port_state(struct port *);
3487 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3489 struct bridge *br = br_;
3490 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3491 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3493 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3496 struct eth_header *eth = pkt->l2;
3498 netdev_get_etheraddr(iface->netdev, eth->eth_src);
3499 if (eth_addr_is_zero(eth->eth_src)) {
3500 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
3501 "with unknown MAC", br->name, port_no);
3503 union ofp_action action;
3506 memset(&action, 0, sizeof action);
3507 action.type = htons(OFPAT_OUTPUT);
3508 action.output.len = htons(sizeof action);
3509 action.output.port = htons(port_no);
3511 flow_extract(pkt, ODPP_NONE, &flow);
3512 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3519 brstp_reconfigure(struct bridge *br)
3523 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3525 stp_destroy(br->stp);
3531 uint64_t bridge_address, bridge_id;
3532 int bridge_priority;
3534 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3535 if (!bridge_address) {
3537 bridge_address = (stp_get_bridge_id(br->stp)
3538 & ((UINT64_C(1) << 48) - 1));
3540 uint8_t mac[ETH_ADDR_LEN];
3541 eth_addr_random(mac);
3542 bridge_address = eth_addr_to_uint64(mac);
3546 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3548 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3550 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3553 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3555 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3556 br->stp_last_tick = time_msec();
3559 if (bridge_id != stp_get_bridge_id(br->stp)) {
3560 stp_set_bridge_id(br->stp, bridge_id);
3565 for (i = 0; i < br->n_ports; i++) {
3566 struct port *p = br->ports[i];
3568 struct stp_port *sp;
3569 int path_cost, priority;
3575 dp_ifidx = p->ifaces[0]->dp_ifidx;
3576 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3580 sp = stp_get_port(br->stp, dp_ifidx);
3581 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3582 "stp.%s.port.%s.enabled",
3584 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3585 br->name, p->name));
3586 if (p->is_mirror_output_port) {
3589 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3590 bridge_flush(br); /* Might not be necessary. */
3592 stp_port_enable(sp);
3594 stp_port_disable(sp);
3598 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3600 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3602 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3603 "stp.%s.port.%s.priority",
3605 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3607 : STP_DEFAULT_PORT_PRIORITY);
3608 stp_port_set_priority(sp, priority);
3611 brstp_adjust_timers(br);
3613 for (i = 0; i < br->n_ports; i++) {
3614 brstp_update_port_state(br->ports[i]);
3619 brstp_update_port_state(struct port *p)
3621 struct bridge *br = p->bridge;
3622 enum stp_state state;
3624 /* Figure out new state. */
3625 state = STP_DISABLED;
3626 if (br->stp && p->n_ifaces > 0) {
3627 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3628 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3629 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3634 if (p->stp_state != state) {
3635 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3636 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3637 p->name, stp_state_name(p->stp_state),
3638 stp_state_name(state));
3639 if (p->stp_state == STP_DISABLED) {
3642 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3644 p->stp_state = state;
3645 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3646 : tag_create_random());
3651 brstp_adjust_timers(struct bridge *br)
3653 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3654 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3655 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3657 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3658 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3659 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3663 brstp_run(struct bridge *br)
3666 long long int now = time_msec();
3667 long long int elapsed = now - br->stp_last_tick;
3668 struct stp_port *sp;
3671 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3672 br->stp_last_tick = now;
3674 while (stp_get_changed_port(br->stp, &sp)) {
3675 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3677 brstp_update_port_state(p);
3684 brstp_wait(struct bridge *br)
3687 poll_timer_wait(1000);