1 /* Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
23 #include <sys/socket.h>
25 #include <openflow/openflow.h>
30 #include <sys/socket.h>
31 #include <sys/types.h>
36 #include "dynamic-string.h"
42 #include "mac-learning.h"
44 #include "xflow-util.h"
45 #include "ofp-print.h"
47 #include "ofproto/netflow.h"
48 #include "ofproto/ofproto.h"
49 #include "ofproto/wdp.h"
50 #include "ovsdb-data.h"
52 #include "poll-loop.h"
53 #include "proc-net-compat.h"
57 #include "socket-util.h"
58 #include "stream-ssl.h"
60 #include "system-stats.h"
65 #include "vswitchd/vswitch-idl.h"
66 #include "xenserver.h"
69 #include "sflow_api.h"
71 VLOG_DEFINE_THIS_MODULE(bridge)
79 /* These members are always valid. */
80 struct port *port; /* Containing port. */
81 size_t port_ifidx; /* Index within containing port. */
82 char *name; /* Host network device name. */
83 tag_type tag; /* Tag associated with this interface. */
84 long long delay_expires; /* Time after which 'enabled' may change. */
86 /* These members are valid only after bridge_reconfigure() causes them to
88 struct hmap_node xf_ifidx_node; /* In struct bridge's "ifaces" hmap. */
89 int xf_ifidx; /* Index within kernel datapath. */
90 struct netdev *netdev; /* Network device. */
91 bool enabled; /* May be chosen for flows? */
92 const struct ovsrec_interface *cfg;
95 #define BOND_MASK 0xff
97 int iface_idx; /* Index of assigned iface, or -1 if none. */
98 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
99 tag_type iface_tag; /* Tag associated with iface_idx. */
102 #define MAX_MIRRORS 32
103 typedef uint32_t mirror_mask_t;
104 #define MIRROR_MASK_C(X) UINT32_C(X)
105 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
107 struct bridge *bridge;
110 struct uuid uuid; /* UUID of this "mirror" record in database. */
112 /* Selection criteria. */
113 struct shash src_ports; /* Name is port name; data is always NULL. */
114 struct shash dst_ports; /* Name is port name; data is always NULL. */
119 struct port *out_port;
123 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
125 struct bridge *bridge;
127 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
128 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
129 * NULL if all VLANs are trunked. */
130 const struct ovsrec_port *cfg;
133 /* An ordinary bridge port has 1 interface.
134 * A bridge port for bonding has at least 2 interfaces. */
135 struct iface **ifaces;
136 size_t n_ifaces, allocated_ifaces;
139 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
140 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
141 tag_type active_iface_tag; /* Tag for bcast flows. */
142 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
143 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
144 bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
145 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
146 long long int bond_next_fake_iface_update; /* Time of next update. */
147 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
148 long long int bond_next_rebalance; /* Next rebalancing time. */
150 /* Port mirroring info. */
151 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
152 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
153 bool is_mirror_output_port; /* Does port mirroring send frames here? */
156 #define DP_MAX_PORTS 255
158 struct list node; /* Node in global list of bridges. */
159 char *name; /* User-specified arbitrary name. */
160 struct mac_learning *ml; /* MAC learning table. */
161 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
162 const struct ovsrec_bridge *cfg;
164 /* OpenFlow switch processing. */
165 struct ofproto *ofproto; /* OpenFlow switch. */
167 /* Kernel datapath information. */
168 struct wdp *wdp; /* Datapath. */
169 struct hmap ifaces; /* Contains "struct iface"s. */
173 size_t n_ports, allocated_ports;
174 struct shash iface_by_name; /* "struct iface"s indexed by name. */
175 struct shash port_by_name; /* "struct port"s indexed by name. */
178 bool has_bonded_ports;
183 /* Port mirroring. */
184 struct mirror *mirrors[MAX_MIRRORS];
187 /* List of all bridges. */
188 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
190 /* OVSDB IDL used to obtain configuration. */
191 static struct ovsdb_idl *idl;
193 /* Each time this timer expires, the bridge fetches systems and interface
194 * statistics and pushes them into the database. */
195 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
196 static long long int stats_timer = LLONG_MIN;
198 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
199 static void bridge_destroy(struct bridge *);
200 static struct bridge *bridge_lookup(const char *name);
201 static unixctl_cb_func bridge_unixctl_dump_flows;
202 static unixctl_cb_func bridge_unixctl_reconnect;
203 static int bridge_run_one(struct bridge *);
204 static size_t bridge_get_controllers(const struct bridge *br,
205 struct ovsrec_controller ***controllersp);
206 static void bridge_reconfigure_one(struct bridge *);
207 static void bridge_reconfigure_remotes(struct bridge *,
208 const struct sockaddr_in *managers,
210 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
211 static void bridge_fetch_dp_ifaces(struct bridge *);
212 static void bridge_flush(struct bridge *);
213 static void bridge_pick_local_hw_addr(struct bridge *,
214 uint8_t ea[ETH_ADDR_LEN],
215 struct iface **hw_addr_iface);
216 static uint64_t bridge_pick_datapath_id(struct bridge *,
217 const uint8_t bridge_ea[ETH_ADDR_LEN],
218 struct iface *hw_addr_iface);
219 static struct iface *bridge_get_local_iface(struct bridge *);
220 static uint64_t dpid_from_hash(const void *, size_t nbytes);
222 static unixctl_cb_func bridge_unixctl_fdb_show;
224 static void bond_init(void);
225 static void bond_run(struct bridge *);
226 static void bond_wait(struct bridge *);
227 static void bond_rebalance_port(struct port *);
228 static void bond_send_learning_packets(struct port *);
229 static void bond_enable_slave(struct iface *iface, bool enable);
231 static struct port *port_create(struct bridge *, const char *name);
232 static void port_reconfigure(struct port *, const struct ovsrec_port *);
233 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
234 static void port_destroy(struct port *);
235 static struct port *port_lookup(const struct bridge *, const char *name);
236 static struct iface *port_lookup_iface(const struct port *, const char *name);
237 static struct port *port_from_xf_ifidx(const struct bridge *,
239 static void port_update_bond_compat(struct port *);
240 static void port_update_vlan_compat(struct port *);
241 static void port_update_bonding(struct port *);
243 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
244 static void mirror_destroy(struct mirror *);
245 static void mirror_reconfigure(struct bridge *);
246 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
247 static bool vlan_is_mirrored(const struct mirror *, int vlan);
249 static struct iface *iface_create(struct port *port,
250 const struct ovsrec_interface *if_cfg);
251 static void iface_destroy(struct iface *);
252 static struct iface *iface_lookup(const struct bridge *, const char *name);
253 static struct iface *iface_from_xf_ifidx(const struct bridge *,
255 static bool iface_is_internal(const struct bridge *, const char *name);
256 static void iface_set_mac(struct iface *);
257 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
259 /* Hooks into ofproto processing. */
260 static struct ofhooks bridge_ofhooks;
262 /* Public functions. */
264 /* Initializes the bridge module, configuring it to obtain its configuration
265 * from an OVSDB server accessed over 'remote', which should be a string in a
266 * form acceptable to ovsdb_idl_create(). */
268 bridge_init(const char *remote)
270 /* Create connection to database. */
271 idl = ovsdb_idl_create(remote, &ovsrec_idl_class);
273 ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_cur_cfg);
274 ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_statistics);
275 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
277 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
279 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
280 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
282 ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_ofport);
283 ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_statistics);
284 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
286 /* Register unixctl commands. */
287 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
288 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
290 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
295 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
296 * but for which the ovs-vswitchd configuration 'cfg' is required. */
298 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
300 static bool already_configured_once;
301 struct svec bridge_names;
302 struct svec wdp_names, wdp_types;
305 /* Only do this once per ovs-vswitchd run. */
306 if (already_configured_once) {
309 already_configured_once = true;
311 stats_timer = time_msec() + STATS_INTERVAL;
313 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
314 svec_init(&bridge_names);
315 for (i = 0; i < cfg->n_bridges; i++) {
316 svec_add(&bridge_names, cfg->bridges[i]->name);
318 svec_sort(&bridge_names);
320 /* Iterate over all system wdps and delete any of them that do not appear
322 svec_init(&wdp_names);
323 svec_init(&wdp_types);
324 wdp_enumerate_types(&wdp_types);
325 for (i = 0; i < wdp_types.n; i++) {
330 wdp_enumerate_names(wdp_types.names[i], &wdp_names);
332 /* For each wdp... */
333 for (j = 0; j < wdp_names.n; j++) {
334 retval = wdp_open(wdp_names.names[j], wdp_types.names[i], &wdp);
336 struct svec all_names;
339 /* ...check whether any of its names is in 'bridge_names'. */
340 svec_init(&all_names);
341 wdp_get_all_names(wdp, &all_names);
342 for (k = 0; k < all_names.n; k++) {
343 if (svec_contains(&bridge_names, all_names.names[k])) {
348 /* No. Delete the wdp. */
352 svec_destroy(&all_names);
357 svec_destroy(&bridge_names);
358 svec_destroy(&wdp_names);
359 svec_destroy(&wdp_types);
362 /* Attempt to create the network device 'iface_name' through the netdev
365 set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
368 struct shash options;
372 shash_init(&options);
373 for (i = 0; i < iface_cfg->n_options; i++) {
374 shash_add(&options, iface_cfg->key_options[i],
375 xstrdup(iface_cfg->value_options[i]));
378 /* Include 'other_config' keys in hash of netdev options. The
379 * namespace of 'other_config' and 'options' must be disjoint.
380 * Prefer 'options' keys over 'other_config' keys. */
381 for (i = 0; i < iface_cfg->n_other_config; i++) {
382 char *value = xstrdup(iface_cfg->value_other_config[i]);
383 if (!shash_add_once(&options, iface_cfg->key_other_config[i],
385 VLOG_WARN("%s: \"other_config\" key %s conflicts with existing "
386 "\"other_config\" or \"options\" entry...ignoring",
387 iface_cfg->name, iface_cfg->key_other_config[i]);
393 struct netdev_options netdev_options;
395 memset(&netdev_options, 0, sizeof netdev_options);
396 netdev_options.name = iface_cfg->name;
397 if (!strcmp(iface_cfg->type, "internal")) {
398 /* An "internal" config type maps to a netdev "system" type. */
399 netdev_options.type = "system";
401 netdev_options.type = iface_cfg->type;
403 netdev_options.args = &options;
404 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
405 error = netdev_open(&netdev_options, &iface->netdev);
408 netdev_get_carrier(iface->netdev, &iface->enabled);
410 } else if (iface->netdev) {
411 const char *netdev_type = netdev_get_type(iface->netdev);
412 const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
413 ? iface_cfg->type : NULL;
415 /* An "internal" config type maps to a netdev "system" type. */
416 if (iface_type && !strcmp(iface_type, "internal")) {
417 iface_type = "system";
420 if (!iface_type || !strcmp(netdev_type, iface_type)) {
421 error = netdev_reconfigure(iface->netdev, &options);
423 VLOG_WARN("%s: attempting change device type from %s to %s",
424 iface_cfg->name, netdev_type, iface_type);
428 shash_destroy_free_data(&options);
434 reconfigure_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface)
436 return set_up_iface(iface_cfg, iface, false);
440 check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
441 void *aux OVS_UNUSED)
443 if (!iface->netdev) {
444 int error = set_up_iface(iface->cfg, iface, true);
446 VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name,
456 check_iface_xf_ifidx(struct bridge *br, struct iface *iface,
457 void *aux OVS_UNUSED)
459 if (iface->xf_ifidx >= 0) {
460 VLOG_DBG("%s has interface %s on port %d",
462 iface->name, iface->xf_ifidx);
465 VLOG_ERR("%s interface not in %s, dropping",
466 iface->name, wdp_name(br->wdp));
472 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
473 void *aux OVS_UNUSED)
475 /* Set policing attributes. */
476 netdev_set_policing(iface->netdev,
477 iface->cfg->ingress_policing_rate,
478 iface->cfg->ingress_policing_burst);
480 /* Set MAC address of internal interfaces other than the local
482 if (iface->xf_ifidx != XFLOWP_LOCAL
483 && iface_is_internal(br, iface->name)) {
484 iface_set_mac(iface);
490 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
491 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
492 * deletes from 'br' any ports that no longer have any interfaces. */
494 iterate_and_prune_ifaces(struct bridge *br,
495 bool (*cb)(struct bridge *, struct iface *,
501 for (i = 0; i < br->n_ports; ) {
502 struct port *port = br->ports[i];
503 for (j = 0; j < port->n_ifaces; ) {
504 struct iface *iface = port->ifaces[j];
505 if (cb(br, iface, aux)) {
508 iface_destroy(iface);
512 if (port->n_ifaces) {
515 VLOG_ERR("%s port has no interfaces, dropping", port->name);
521 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
522 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
523 * responsible for freeing '*managersp' (with free()).
525 * You may be asking yourself "why does ovs-vswitchd care?", because
526 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
527 * should not be and in fact is not directly involved in that. But
528 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
529 * it has to tell in-band control where the managers are to enable that.
532 collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
533 struct sockaddr_in **managersp, size_t *n_managersp)
535 struct sockaddr_in *managers = NULL;
536 size_t n_managers = 0;
538 if (ovs_cfg->n_managers > 0) {
541 managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
542 for (i = 0; i < ovs_cfg->n_managers; i++) {
543 const char *name = ovs_cfg->managers[i];
544 struct sockaddr_in *sin = &managers[i];
546 if ((!strncmp(name, "tcp:", 4)
547 && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
548 (!strncmp(name, "ssl:", 4)
549 && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
555 *managersp = managers;
556 *n_managersp = n_managers;
560 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
562 struct shash old_br, new_br;
563 struct shash_node *node;
564 struct bridge *br, *next;
565 struct sockaddr_in *managers;
568 int sflow_bridge_number;
570 COVERAGE_INC(bridge_reconfigure);
572 collect_managers(ovs_cfg, &managers, &n_managers);
574 /* Collect old and new bridges. */
577 LIST_FOR_EACH (br, node, &all_bridges) {
578 shash_add(&old_br, br->name, br);
580 for (i = 0; i < ovs_cfg->n_bridges; i++) {
581 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
582 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
583 VLOG_WARN("more than one bridge named %s", br_cfg->name);
587 /* Get rid of deleted bridges and add new bridges. */
588 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
589 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
596 SHASH_FOR_EACH (node, &new_br) {
597 const char *br_name = node->name;
598 const struct ovsrec_bridge *br_cfg = node->data;
599 br = shash_find_data(&old_br, br_name);
601 /* If the bridge datapath type has changed, we need to tear it
602 * down and recreate. */
603 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
605 bridge_create(br_cfg);
608 bridge_create(br_cfg);
611 shash_destroy(&old_br);
612 shash_destroy(&new_br);
614 /* Reconfigure all bridges. */
615 LIST_FOR_EACH (br, node, &all_bridges) {
616 bridge_reconfigure_one(br);
619 /* Add and delete ports on all datapaths.
621 * The kernel will reject any attempt to add a given port to a datapath if
622 * that port already belongs to a different datapath, so we must do all
623 * port deletions before any port additions. */
624 LIST_FOR_EACH (br, node, &all_bridges) {
625 struct wdp_port *wdp_ports;
627 struct shash want_ifaces;
629 wdp_port_list(br->wdp, &wdp_ports, &n_wdp_ports);
630 bridge_get_all_ifaces(br, &want_ifaces);
631 for (i = 0; i < n_wdp_ports; i++) {
632 const struct wdp_port *p = &wdp_ports[i];
633 if (!shash_find(&want_ifaces, p->devname)
634 && strcmp(p->devname, br->name)) {
637 retval = wdp_port_del(br->wdp, p->opp.port_no);
639 VLOG_ERR("failed to remove %s interface from %s: %s",
640 p->devname, wdp_name(br->wdp),
645 shash_destroy(&want_ifaces);
646 wdp_port_array_free(wdp_ports, n_wdp_ports);
648 LIST_FOR_EACH (br, node, &all_bridges) {
649 struct wdp_port *wdp_ports;
651 struct shash cur_ifaces, want_ifaces;
653 /* Get the set of interfaces currently in this datapath. */
654 wdp_port_list(br->wdp, &wdp_ports, &n_wdp_ports);
655 shash_init(&cur_ifaces);
656 for (i = 0; i < n_wdp_ports; i++) {
657 const char *name = wdp_ports[i].devname;
658 shash_add_once(&cur_ifaces, name, NULL);
660 wdp_port_array_free(wdp_ports, n_wdp_ports);
662 /* Get the set of interfaces we want on this datapath. */
663 bridge_get_all_ifaces(br, &want_ifaces);
665 SHASH_FOR_EACH (node, &want_ifaces) {
666 const char *if_name = node->name;
667 struct iface *iface = node->data;
669 if (shash_find(&cur_ifaces, if_name)) {
670 /* Already exists, just reconfigure it. */
672 reconfigure_iface(iface->cfg, iface);
675 /* Need to add to datapath. */
679 /* Add to datapath. */
680 internal = iface_is_internal(br, if_name);
681 error = wdp_port_add(br->wdp, if_name, internal, NULL);
682 if (error == EFBIG) {
683 VLOG_ERR("ran out of valid port numbers on %s",
687 VLOG_ERR("failed to add %s interface to %s: %s",
688 if_name, wdp_name(br->wdp), strerror(error));
692 shash_destroy(&cur_ifaces);
693 shash_destroy(&want_ifaces);
695 sflow_bridge_number = 0;
696 LIST_FOR_EACH (br, node, &all_bridges) {
699 struct iface *local_iface;
700 struct iface *hw_addr_iface;
703 bridge_fetch_dp_ifaces(br);
705 iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
706 iterate_and_prune_ifaces(br, check_iface_xf_ifidx, NULL);
708 /* Pick local port hardware address, datapath ID. */
709 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
710 local_iface = bridge_get_local_iface(br);
712 int error = netdev_set_etheraddr(local_iface->netdev, ea);
714 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
715 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
716 "Ethernet address: %s",
717 br->name, strerror(error));
721 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
722 ofproto_set_datapath_id(br->ofproto, dpid);
724 dpid_string = xasprintf("%016"PRIx64, dpid);
725 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
728 /* Set NetFlow configuration on this bridge. */
729 if (br->cfg->netflow) {
730 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
731 struct netflow_options opts;
733 memset(&opts, 0, sizeof opts);
735 wdp_get_netflow_ids(br->wdp, &opts.engine_type, &opts.engine_id);
736 if (nf_cfg->engine_type) {
737 opts.engine_type = *nf_cfg->engine_type;
739 if (nf_cfg->engine_id) {
740 opts.engine_id = *nf_cfg->engine_id;
743 opts.active_timeout = nf_cfg->active_timeout;
744 if (!opts.active_timeout) {
745 opts.active_timeout = -1;
746 } else if (opts.active_timeout < 0) {
747 VLOG_WARN("bridge %s: active timeout interval set to negative "
748 "value, using default instead (%d seconds)", br->name,
749 NF_ACTIVE_TIMEOUT_DEFAULT);
750 opts.active_timeout = -1;
753 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
754 if (opts.add_id_to_iface) {
755 if (opts.engine_id > 0x7f) {
756 VLOG_WARN("bridge %s: netflow port mangling may conflict "
757 "with another vswitch, choose an engine id less "
758 "than 128", br->name);
760 if (br->n_ports > 508) {
761 VLOG_WARN("bridge %s: netflow port mangling will conflict "
762 "with another port when more than 508 ports are "
767 opts.collectors.n = nf_cfg->n_targets;
768 opts.collectors.names = nf_cfg->targets;
769 if (ofproto_set_netflow(br->ofproto, &opts)) {
770 VLOG_ERR("bridge %s: problem setting netflow collectors",
774 ofproto_set_netflow(br->ofproto, NULL);
777 /* Set sFlow configuration on this bridge. */
778 if (br->cfg->sflow) {
779 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
780 struct ovsrec_controller **controllers;
781 struct ofproto_sflow_options oso;
782 size_t n_controllers;
784 memset(&oso, 0, sizeof oso);
786 oso.targets.n = sflow_cfg->n_targets;
787 oso.targets.names = sflow_cfg->targets;
789 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
790 if (sflow_cfg->sampling) {
791 oso.sampling_rate = *sflow_cfg->sampling;
794 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
795 if (sflow_cfg->polling) {
796 oso.polling_interval = *sflow_cfg->polling;
799 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
800 if (sflow_cfg->header) {
801 oso.header_len = *sflow_cfg->header;
804 oso.sub_id = sflow_bridge_number++;
805 oso.agent_device = sflow_cfg->agent;
807 oso.control_ip = NULL;
808 n_controllers = bridge_get_controllers(br, &controllers);
809 for (i = 0; i < n_controllers; i++) {
810 if (controllers[i]->local_ip) {
811 oso.control_ip = controllers[i]->local_ip;
815 ofproto_set_sflow(br->ofproto, &oso);
817 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
819 ofproto_set_sflow(br->ofproto, NULL);
822 /* Update the controller and related settings. It would be more
823 * straightforward to call this from bridge_reconfigure_one(), but we
824 * can't do it there for two reasons. First, and most importantly, at
825 * that point we don't know the xf_ifidx of any interfaces that have
826 * been added to the bridge (because we haven't actually added them to
827 * the datapath). Second, at that point we haven't set the datapath ID
828 * yet; when a controller is configured, resetting the datapath ID will
829 * immediately disconnect from the controller, so it's better to set
830 * the datapath ID before the controller. */
831 bridge_reconfigure_remotes(br, managers, n_managers);
833 LIST_FOR_EACH (br, node, &all_bridges) {
834 for (i = 0; i < br->n_ports; i++) {
835 struct port *port = br->ports[i];
838 port_update_vlan_compat(port);
839 port_update_bonding(port);
841 for (j = 0; j < port->n_ifaces; j++) {
842 iface_update_qos(port->ifaces[j], port->cfg->qos);
846 LIST_FOR_EACH (br, node, &all_bridges) {
847 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
854 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
855 const struct ovsdb_idl_column *column,
858 const struct ovsdb_datum *datum;
859 union ovsdb_atom atom;
862 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
863 atom.string = (char *) key;
864 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
865 return idx == UINT_MAX ? NULL : datum->values[idx].string;
869 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
871 return get_ovsrec_key_value(&br_cfg->header_,
872 &ovsrec_bridge_col_other_config, key);
876 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
877 struct iface **hw_addr_iface)
883 *hw_addr_iface = NULL;
885 /* Did the user request a particular MAC? */
886 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
887 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
888 if (eth_addr_is_multicast(ea)) {
889 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
890 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
891 } else if (eth_addr_is_zero(ea)) {
892 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
898 /* Otherwise choose the minimum non-local MAC address among all of the
900 memset(ea, 0xff, sizeof ea);
901 for (i = 0; i < br->n_ports; i++) {
902 struct port *port = br->ports[i];
903 uint8_t iface_ea[ETH_ADDR_LEN];
906 /* Mirror output ports don't participate. */
907 if (port->is_mirror_output_port) {
911 /* Choose the MAC address to represent the port. */
912 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
913 /* Find the interface with this Ethernet address (if any) so that
914 * we can provide the correct devname to the caller. */
916 for (j = 0; j < port->n_ifaces; j++) {
917 struct iface *candidate = port->ifaces[j];
918 uint8_t candidate_ea[ETH_ADDR_LEN];
919 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
920 && eth_addr_equals(iface_ea, candidate_ea)) {
925 /* Choose the interface whose MAC address will represent the port.
926 * The Linux kernel bonding code always chooses the MAC address of
927 * the first slave added to a bond, and the Fedora networking
928 * scripts always add slaves to a bond in alphabetical order, so
929 * for compatibility we choose the interface with the name that is
930 * first in alphabetical order. */
931 iface = port->ifaces[0];
932 for (j = 1; j < port->n_ifaces; j++) {
933 struct iface *candidate = port->ifaces[j];
934 if (strcmp(candidate->name, iface->name) < 0) {
939 /* The local port doesn't count (since we're trying to choose its
940 * MAC address anyway). */
941 if (iface->xf_ifidx == XFLOWP_LOCAL) {
946 error = netdev_get_etheraddr(iface->netdev, iface_ea);
948 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
949 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
950 iface->name, strerror(error));
955 /* Compare against our current choice. */
956 if (!eth_addr_is_multicast(iface_ea) &&
957 !eth_addr_is_local(iface_ea) &&
958 !eth_addr_is_reserved(iface_ea) &&
959 !eth_addr_is_zero(iface_ea) &&
960 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
962 memcpy(ea, iface_ea, ETH_ADDR_LEN);
963 *hw_addr_iface = iface;
966 if (eth_addr_is_multicast(ea)) {
967 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
968 *hw_addr_iface = NULL;
969 VLOG_WARN("bridge %s: using default bridge Ethernet "
970 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
972 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
973 br->name, ETH_ADDR_ARGS(ea));
977 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
978 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
979 * an interface on 'br', then that interface must be passed in as
980 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
981 * 'hw_addr_iface' must be passed in as a null pointer. */
983 bridge_pick_datapath_id(struct bridge *br,
984 const uint8_t bridge_ea[ETH_ADDR_LEN],
985 struct iface *hw_addr_iface)
988 * The procedure for choosing a bridge MAC address will, in the most
989 * ordinary case, also choose a unique MAC that we can use as a datapath
990 * ID. In some special cases, though, multiple bridges will end up with
991 * the same MAC address. This is OK for the bridges, but it will confuse
992 * the OpenFlow controller, because each datapath needs a unique datapath
995 * Datapath IDs must be unique. It is also very desirable that they be
996 * stable from one run to the next, so that policy set on a datapath
999 const char *datapath_id;
1002 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1003 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1007 if (hw_addr_iface) {
1009 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1011 * A bridge whose MAC address is taken from a VLAN network device
1012 * (that is, a network device created with vconfig(8) or similar
1013 * tool) will have the same MAC address as a bridge on the VLAN
1014 * device's physical network device.
1016 * Handle this case by hashing the physical network device MAC
1017 * along with the VLAN identifier.
1019 uint8_t buf[ETH_ADDR_LEN + 2];
1020 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1021 buf[ETH_ADDR_LEN] = vlan >> 8;
1022 buf[ETH_ADDR_LEN + 1] = vlan;
1023 return dpid_from_hash(buf, sizeof buf);
1026 * Assume that this bridge's MAC address is unique, since it
1027 * doesn't fit any of the cases we handle specially.
1032 * A purely internal bridge, that is, one that has no non-virtual
1033 * network devices on it at all, is more difficult because it has no
1034 * natural unique identifier at all.
1036 * When the host is a XenServer, we handle this case by hashing the
1037 * host's UUID with the name of the bridge. Names of bridges are
1038 * persistent across XenServer reboots, although they can be reused if
1039 * an internal network is destroyed and then a new one is later
1040 * created, so this is fairly effective.
1042 * When the host is not a XenServer, we punt by using a random MAC
1043 * address on each run.
1045 const char *host_uuid = xenserver_get_host_uuid();
1047 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1048 dpid = dpid_from_hash(combined, strlen(combined));
1054 return eth_addr_to_uint64(bridge_ea);
1058 dpid_from_hash(const void *data, size_t n)
1060 uint8_t hash[SHA1_DIGEST_SIZE];
1062 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1063 sha1_bytes(data, n, hash);
1064 eth_addr_mark_random(hash);
1065 return eth_addr_to_uint64(hash);
1069 iface_refresh_stats(struct iface *iface)
1075 static const struct iface_stat iface_stats[] = {
1076 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1077 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1078 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1079 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1080 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1081 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1082 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1083 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1084 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1085 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1086 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1087 { "collisions", offsetof(struct netdev_stats, collisions) },
1089 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1090 const struct iface_stat *s;
1092 char *keys[N_STATS];
1093 int64_t values[N_STATS];
1096 struct netdev_stats stats;
1098 /* Intentionally ignore return value, since errors will set 'stats' to
1099 * all-1s, and we will deal with that correctly below. */
1100 netdev_get_stats(iface->netdev, &stats);
1103 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1104 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1105 if (value != UINT64_MAX) {
1112 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1116 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1118 struct ovsdb_datum datum;
1122 get_system_stats(&stats);
1124 ovsdb_datum_from_shash(&datum, &stats);
1125 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1132 const struct ovsrec_open_vswitch *cfg;
1134 bool datapath_destroyed;
1135 bool database_changed;
1138 /* Let each bridge do the work that it needs to do. */
1139 datapath_destroyed = false;
1140 LIST_FOR_EACH (br, node, &all_bridges) {
1141 int error = bridge_run_one(br);
1143 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1144 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1145 "forcing reconfiguration", br->name);
1146 datapath_destroyed = true;
1150 /* (Re)configure if necessary. */
1151 database_changed = ovsdb_idl_run(idl);
1152 cfg = ovsrec_open_vswitch_first(idl);
1153 if (database_changed || datapath_destroyed) {
1155 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1157 bridge_configure_once(cfg);
1158 bridge_reconfigure(cfg);
1160 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1161 ovsdb_idl_txn_commit(txn);
1162 ovsdb_idl_txn_destroy(txn); /* XXX */
1164 /* We still need to reconfigure to avoid dangling pointers to
1165 * now-destroyed ovsrec structures inside bridge data. */
1166 static const struct ovsrec_open_vswitch null_cfg;
1168 bridge_reconfigure(&null_cfg);
1173 /* Re-configure SSL. We do this on every trip through the main loop,
1174 * instead of just when the database changes, because the contents of the
1175 * key and certificate files can change without the database changing. */
1176 if (cfg && cfg->ssl) {
1177 const struct ovsrec_ssl *ssl = cfg->ssl;
1179 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1180 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1184 /* Refresh system and interface stats if necessary. */
1185 if (time_msec() >= stats_timer) {
1187 struct ovsdb_idl_txn *txn;
1189 txn = ovsdb_idl_txn_create(idl);
1190 LIST_FOR_EACH (br, node, &all_bridges) {
1193 for (i = 0; i < br->n_ports; i++) {
1194 struct port *port = br->ports[i];
1197 for (j = 0; j < port->n_ifaces; j++) {
1198 struct iface *iface = port->ifaces[j];
1199 iface_refresh_stats(iface);
1203 refresh_system_stats(cfg);
1204 ovsdb_idl_txn_commit(txn);
1205 ovsdb_idl_txn_destroy(txn); /* XXX */
1208 stats_timer = time_msec() + STATS_INTERVAL;
1217 LIST_FOR_EACH (br, node, &all_bridges) {
1218 ofproto_wait(br->ofproto);
1219 if (ofproto_has_primary_controller(br->ofproto)) {
1223 mac_learning_wait(br->ml);
1226 ovsdb_idl_wait(idl);
1227 poll_timer_wait_until(stats_timer);
1230 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1231 * configuration changes. */
1233 bridge_flush(struct bridge *br)
1235 COVERAGE_INC(bridge_flush);
1236 ofproto_revalidate_all(br->ofproto);
1237 mac_learning_flush(br->ml);
1240 /* Returns the 'br' interface for the XFLOWP_LOCAL port, or null if 'br' has no
1241 * such interface. */
1242 static struct iface *
1243 bridge_get_local_iface(struct bridge *br)
1247 for (i = 0; i < br->n_ports; i++) {
1248 struct port *port = br->ports[i];
1249 for (j = 0; j < port->n_ifaces; j++) {
1250 struct iface *iface = port->ifaces[j];
1251 if (iface->xf_ifidx == XFLOWP_LOCAL) {
1260 /* Bridge unixctl user interface functions. */
1262 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1263 const char *args, void *aux OVS_UNUSED)
1265 struct ds ds = DS_EMPTY_INITIALIZER;
1266 const struct bridge *br;
1267 const struct mac_entry *e;
1269 br = bridge_lookup(args);
1271 unixctl_command_reply(conn, 501, "no such bridge");
1275 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1276 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1277 if (e->port < 0 || e->port >= br->n_ports) {
1280 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1281 br->ports[e->port]->ifaces[0]->xf_ifidx,
1282 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1284 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1288 /* Bridge reconfiguration functions. */
1289 static struct bridge *
1290 bridge_create(const struct ovsrec_bridge *br_cfg)
1295 assert(!bridge_lookup(br_cfg->name));
1296 br = xzalloc(sizeof *br);
1298 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1301 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1307 br->wdp = ofproto_get_wdp(br->ofproto);
1308 br->name = xstrdup(br_cfg->name);
1310 br->ml = mac_learning_create();
1311 eth_addr_nicira_random(br->default_ea);
1313 hmap_init(&br->ifaces);
1315 shash_init(&br->port_by_name);
1316 shash_init(&br->iface_by_name);
1318 list_push_back(&all_bridges, &br->node);
1320 VLOG_INFO("created bridge %s on %s", br->name, wdp_name(br->wdp));
1326 bridge_destroy(struct bridge *br)
1331 while (br->n_ports > 0) {
1332 port_destroy(br->ports[br->n_ports - 1]);
1334 list_remove(&br->node);
1335 error = wdp_delete(br->wdp);
1336 if (error && error != ENOENT) {
1337 VLOG_ERR("failed to delete %s: %s",
1338 wdp_name(br->wdp), strerror(error));
1340 ofproto_destroy(br->ofproto);
1341 mac_learning_destroy(br->ml);
1342 hmap_destroy(&br->ifaces);
1343 shash_destroy(&br->port_by_name);
1344 shash_destroy(&br->iface_by_name);
1351 static struct bridge *
1352 bridge_lookup(const char *name)
1356 LIST_FOR_EACH (br, node, &all_bridges) {
1357 if (!strcmp(br->name, name)) {
1364 /* Handle requests for a listing of all flows known by the OpenFlow
1365 * stack, including those normally hidden. */
1367 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1368 const char *args, void *aux OVS_UNUSED)
1373 br = bridge_lookup(args);
1375 unixctl_command_reply(conn, 501, "Unknown bridge");
1380 ofproto_get_all_flows(br->ofproto, &results);
1382 unixctl_command_reply(conn, 200, ds_cstr(&results));
1383 ds_destroy(&results);
1386 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1387 * connections and reconnect. If BRIDGE is not specified, then all bridges
1388 * drop their controller connections and reconnect. */
1390 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1391 const char *args, void *aux OVS_UNUSED)
1394 if (args[0] != '\0') {
1395 br = bridge_lookup(args);
1397 unixctl_command_reply(conn, 501, "Unknown bridge");
1400 ofproto_reconnect_controllers(br->ofproto);
1402 LIST_FOR_EACH (br, node, &all_bridges) {
1403 ofproto_reconnect_controllers(br->ofproto);
1406 unixctl_command_reply(conn, 200, NULL);
1410 bridge_run_one(struct bridge *br)
1412 ofproto_revalidate(br->ofproto, mac_learning_run(br->ml));
1414 return ofproto_run(br->ofproto);
1418 bridge_get_controllers(const struct bridge *br,
1419 struct ovsrec_controller ***controllersp)
1421 struct ovsrec_controller **controllers;
1422 size_t n_controllers;
1424 controllers = br->cfg->controller;
1425 n_controllers = br->cfg->n_controller;
1427 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1433 *controllersp = controllers;
1435 return n_controllers;
1439 bridge_reconfigure_one(struct bridge *br)
1441 struct shash old_ports, new_ports;
1442 struct svec snoops, old_snoops;
1443 struct shash_node *node;
1444 enum ofproto_fail_mode fail_mode;
1447 /* Collect old ports. */
1448 shash_init(&old_ports);
1449 for (i = 0; i < br->n_ports; i++) {
1450 shash_add(&old_ports, br->ports[i]->name, br->ports[i]);
1453 /* Collect new ports. */
1454 shash_init(&new_ports);
1455 for (i = 0; i < br->cfg->n_ports; i++) {
1456 const char *name = br->cfg->ports[i]->name;
1457 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1458 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1463 /* If we have a controller, then we need a local port. Complain if the
1464 * user didn't specify one.
1466 * XXX perhaps we should synthesize a port ourselves in this case. */
1467 if (bridge_get_controllers(br, NULL)) {
1471 error = wdp_port_get_name(br->wdp, OFPP_LOCAL, &local_name);
1472 if (!error && !shash_find(&new_ports, local_name)) {
1473 VLOG_WARN("bridge %s: controller specified but no local port "
1474 "(port named %s) defined",
1475 br->name, local_name);
1480 /* Get rid of deleted ports.
1481 * Get rid of deleted interfaces on ports that still exist. */
1482 SHASH_FOR_EACH (node, &old_ports) {
1483 struct port *port = node->data;
1484 const struct ovsrec_port *port_cfg;
1486 port_cfg = shash_find_data(&new_ports, node->name);
1490 port_del_ifaces(port, port_cfg);
1494 /* Create new ports.
1495 * Add new interfaces to existing ports.
1496 * Reconfigure existing ports. */
1497 SHASH_FOR_EACH (node, &new_ports) {
1498 struct port *port = shash_find_data(&old_ports, node->name);
1500 port = port_create(br, node->name);
1503 port_reconfigure(port, node->data);
1504 if (!port->n_ifaces) {
1505 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1506 br->name, port->name);
1510 shash_destroy(&old_ports);
1511 shash_destroy(&new_ports);
1513 /* Set the fail-mode */
1514 fail_mode = !br->cfg->fail_mode
1515 || !strcmp(br->cfg->fail_mode, "standalone")
1516 ? OFPROTO_FAIL_STANDALONE
1517 : OFPROTO_FAIL_SECURE;
1518 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1519 && !ofproto_has_primary_controller(br->ofproto)) {
1520 ofproto_flush_flows(br->ofproto);
1522 ofproto_set_fail_mode(br->ofproto, fail_mode);
1524 /* Delete all flows if we're switching from connected to standalone or vice
1525 * versa. (XXX Should we delete all flows if we are switching from one
1526 * controller to another?) */
1528 /* Configure OpenFlow controller connection snooping. */
1530 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1531 ovs_rundir, br->name));
1532 svec_init(&old_snoops);
1533 ofproto_get_snoops(br->ofproto, &old_snoops);
1534 if (!svec_equal(&snoops, &old_snoops)) {
1535 ofproto_set_snoops(br->ofproto, &snoops);
1537 svec_destroy(&snoops);
1538 svec_destroy(&old_snoops);
1540 mirror_reconfigure(br);
1543 /* Initializes 'oc' appropriately as a management service controller for
1546 * The caller must free oc->target when it is no longer needed. */
1548 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1549 struct ofproto_controller *oc)
1551 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir, br->name);
1552 oc->max_backoff = 0;
1553 oc->probe_interval = 60;
1554 oc->band = OFPROTO_OUT_OF_BAND;
1555 oc->accept_re = NULL;
1556 oc->update_resolv_conf = false;
1558 oc->burst_limit = 0;
1561 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1563 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1564 struct ofproto_controller *oc)
1566 oc->target = c->target;
1567 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1568 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1569 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1570 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1571 oc->accept_re = c->discover_accept_regex;
1572 oc->update_resolv_conf = c->discover_update_resolv_conf;
1573 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1574 oc->burst_limit = (c->controller_burst_limit
1575 ? *c->controller_burst_limit : 0);
1578 /* Configures the IP stack for 'br''s local interface properly according to the
1579 * configuration in 'c'. */
1581 bridge_configure_local_iface_netdev(struct bridge *br,
1582 struct ovsrec_controller *c)
1584 struct netdev *netdev;
1585 struct in_addr mask, gateway;
1587 struct iface *local_iface;
1590 /* Controller discovery does its own TCP/IP configuration later. */
1591 if (strcmp(c->target, "discover")) {
1595 /* If there's no local interface or no IP address, give up. */
1596 local_iface = bridge_get_local_iface(br);
1597 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1601 /* Bring up the local interface. */
1602 netdev = local_iface->netdev;
1603 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1605 /* Configure the IP address and netmask. */
1606 if (!c->local_netmask
1607 || !inet_aton(c->local_netmask, &mask)
1609 mask.s_addr = guess_netmask(ip.s_addr);
1611 if (!netdev_set_in4(netdev, ip, mask)) {
1612 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
1613 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
1616 /* Configure the default gateway. */
1617 if (c->local_gateway
1618 && inet_aton(c->local_gateway, &gateway)
1619 && gateway.s_addr) {
1620 if (!netdev_add_router(netdev, gateway)) {
1621 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1622 br->name, IP_ARGS(&gateway.s_addr));
1628 bridge_reconfigure_remotes(struct bridge *br,
1629 const struct sockaddr_in *managers,
1632 struct ovsrec_controller **controllers;
1633 size_t n_controllers;
1636 struct ofproto_controller *ocs;
1640 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
1641 had_primary = ofproto_has_primary_controller(br->ofproto);
1643 n_controllers = bridge_get_controllers(br, &controllers);
1645 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
1648 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
1649 for (i = 0; i < n_controllers; i++) {
1650 struct ovsrec_controller *c = controllers[i];
1652 if (!strncmp(c->target, "punix:", 6)
1653 || !strncmp(c->target, "unix:", 5)) {
1654 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1656 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1657 * domain sockets and overwriting arbitrary local files. */
1658 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
1659 "\"%s\" due to possibility for remote exploit",
1660 br->name, c->target);
1664 bridge_configure_local_iface_netdev(br, c);
1665 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs++]);
1668 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
1669 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
1672 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
1673 ofproto_flush_flows(br->ofproto);
1676 /* If there are no controllers and the bridge is in standalone
1677 * mode, set up a flow that matches every packet and directs
1678 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1679 * switch is in secure mode and we won't pass any traffic until
1680 * a controller has been defined and it tells us to do so. */
1682 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
1683 union ofp_action action;
1686 memset(&action, 0, sizeof action);
1687 action.type = htons(OFPAT_OUTPUT);
1688 action.output.len = htons(sizeof action);
1689 action.output.port = htons(OFPP_NORMAL);
1690 memset(&flow, 0, sizeof flow);
1691 flow.wildcards = OVSFW_ALL;
1692 ofproto_add_flow(br->ofproto, &flow, &action, 1, 0);
1697 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
1702 for (i = 0; i < br->n_ports; i++) {
1703 struct port *port = br->ports[i];
1704 for (j = 0; j < port->n_ifaces; j++) {
1705 struct iface *iface = port->ifaces[j];
1706 shash_add_once(ifaces, iface->name, iface);
1708 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
1709 shash_add_once(ifaces, port->name, NULL);
1714 /* For robustness, in case the administrator moves around datapath ports behind
1715 * our back, we re-check all the datapath port numbers here.
1717 * This function will set the 'xf_ifidx' members of interfaces that have
1718 * disappeared to -1, so only call this function from a context where those
1719 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1720 * 'xf_ifidx'es will cause trouble later when we try to send them to the
1721 * datapath, which doesn't support UINT16_MAX+1 ports. */
1723 bridge_fetch_dp_ifaces(struct bridge *br)
1725 struct wdp_port *wdp_ports;
1729 /* Reset all interface numbers. */
1730 for (i = 0; i < br->n_ports; i++) {
1731 struct port *port = br->ports[i];
1732 for (j = 0; j < port->n_ifaces; j++) {
1733 struct iface *iface = port->ifaces[j];
1734 iface->xf_ifidx = -1;
1737 hmap_clear(&br->ifaces);
1739 wdp_port_list(br->wdp, &wdp_ports, &n_wdp_ports);
1740 for (i = 0; i < n_wdp_ports; i++) {
1741 struct wdp_port *p = &wdp_ports[i];
1742 struct iface *iface = iface_lookup(br, p->devname);
1744 if (iface->xf_ifidx >= 0) {
1745 VLOG_WARN("%s reported interface %s twice",
1746 wdp_name(br->wdp), p->devname);
1747 } else if (iface_from_xf_ifidx(
1748 br, ofp_port_to_xflow_port(p->opp.port_no))) {
1749 VLOG_WARN("%s reported interface %"PRIu16" twice",
1750 wdp_name(br->wdp), p->opp.port_no);
1752 iface->xf_ifidx = ofp_port_to_xflow_port(p->opp.port_no);
1753 hmap_insert(&br->ifaces, &iface->xf_ifidx_node,
1754 hash_int(iface->xf_ifidx, 0));
1758 int64_t ofport = iface->xf_ifidx >= 0 ? p->opp.port_no : -1;
1759 ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
1763 wdp_port_array_free(wdp_ports, n_wdp_ports);
1766 /* Bridge packet processing functions. */
1769 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1771 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1774 static struct bond_entry *
1775 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1777 return &port->bond_hash[bond_hash(mac)];
1781 bond_choose_iface(const struct port *port)
1783 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1784 size_t i, best_down_slave = -1;
1785 long long next_delay_expiration = LLONG_MAX;
1787 for (i = 0; i < port->n_ifaces; i++) {
1788 struct iface *iface = port->ifaces[i];
1790 if (iface->enabled) {
1792 } else if (iface->delay_expires < next_delay_expiration) {
1793 best_down_slave = i;
1794 next_delay_expiration = iface->delay_expires;
1798 if (best_down_slave != -1) {
1799 struct iface *iface = port->ifaces[best_down_slave];
1801 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
1802 "since no other interface is up", iface->name,
1803 iface->delay_expires - time_msec());
1804 bond_enable_slave(iface, true);
1807 return best_down_slave;
1811 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1812 uint16_t *xf_ifidx, tag_type *tags)
1814 struct iface *iface;
1816 assert(port->n_ifaces);
1817 if (port->n_ifaces == 1) {
1818 iface = port->ifaces[0];
1820 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1821 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1822 || !port->ifaces[e->iface_idx]->enabled) {
1823 /* XXX select interface properly. The current interface selection
1824 * is only good for testing the rebalancing code. */
1825 e->iface_idx = bond_choose_iface(port);
1826 if (e->iface_idx < 0) {
1827 *tags |= port->no_ifaces_tag;
1830 e->iface_tag = tag_create_random();
1831 ((struct port *) port)->bond_compat_is_stale = true;
1833 *tags |= e->iface_tag;
1834 iface = port->ifaces[e->iface_idx];
1836 *xf_ifidx = iface->xf_ifidx;
1837 *tags |= iface->tag; /* Currently only used for bonding. */
1842 bond_link_status_update(struct iface *iface, bool carrier)
1844 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1845 struct port *port = iface->port;
1847 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1848 /* Nothing to do. */
1851 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1852 iface->name, carrier ? "detected" : "dropped");
1853 if (carrier == iface->enabled) {
1854 iface->delay_expires = LLONG_MAX;
1855 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1856 iface->name, carrier ? "disabled" : "enabled");
1857 } else if (carrier && port->active_iface < 0) {
1858 bond_enable_slave(iface, true);
1859 if (port->updelay) {
1860 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1861 "other interface is up", iface->name, port->updelay);
1864 int delay = carrier ? port->updelay : port->downdelay;
1865 iface->delay_expires = time_msec() + delay;
1868 "interface %s: will be %s if it stays %s for %d ms",
1870 carrier ? "enabled" : "disabled",
1871 carrier ? "up" : "down",
1878 bond_choose_active_iface(struct port *port)
1880 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1882 port->active_iface = bond_choose_iface(port);
1883 port->active_iface_tag = tag_create_random();
1884 if (port->active_iface >= 0) {
1885 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1886 port->name, port->ifaces[port->active_iface]->name);
1888 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1894 bond_enable_slave(struct iface *iface, bool enable)
1896 struct port *port = iface->port;
1897 struct bridge *br = port->bridge;
1899 /* This acts as a recursion check. If the act of disabling a slave
1900 * causes a different slave to be enabled, the flag will allow us to
1901 * skip redundant work when we reenter this function. It must be
1902 * cleared on exit to keep things safe with multiple bonds. */
1903 static bool moving_active_iface = false;
1905 iface->delay_expires = LLONG_MAX;
1906 if (enable == iface->enabled) {
1910 iface->enabled = enable;
1911 if (!iface->enabled) {
1912 VLOG_WARN("interface %s: disabled", iface->name);
1913 ofproto_revalidate(br->ofproto, iface->tag);
1914 if (iface->port_ifidx == port->active_iface) {
1915 ofproto_revalidate(br->ofproto,
1916 port->active_iface_tag);
1918 /* Disabling a slave can lead to another slave being immediately
1919 * enabled if there will be no active slaves but one is waiting
1920 * on an updelay. In this case we do not need to run most of the
1921 * code for the newly enabled slave since there was no period
1922 * without an active slave and it is redundant with the disabling
1924 moving_active_iface = true;
1925 bond_choose_active_iface(port);
1927 bond_send_learning_packets(port);
1929 VLOG_WARN("interface %s: enabled", iface->name);
1930 if (port->active_iface < 0 && !moving_active_iface) {
1931 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1932 bond_choose_active_iface(port);
1933 bond_send_learning_packets(port);
1935 iface->tag = tag_create_random();
1938 moving_active_iface = false;
1939 port->bond_compat_is_stale = true;
1942 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
1943 * bond interface. */
1945 bond_update_fake_iface_stats(struct port *port)
1947 struct netdev_stats bond_stats;
1948 struct netdev *bond_dev;
1951 memset(&bond_stats, 0, sizeof bond_stats);
1953 for (i = 0; i < port->n_ifaces; i++) {
1954 struct netdev_stats slave_stats;
1956 if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
1957 /* XXX: We swap the stats here because they are swapped back when
1958 * reported by the internal device. The reason for this is
1959 * internal devices normally represent packets going into the system
1960 * but when used as fake bond device they represent packets leaving
1961 * the system. We really should do this in the internal device
1962 * itself because changing it here reverses the counts from the
1963 * perspective of the switch. However, the internal device doesn't
1964 * know what type of device it represents so we have to do it here
1966 bond_stats.tx_packets += slave_stats.rx_packets;
1967 bond_stats.tx_bytes += slave_stats.rx_bytes;
1968 bond_stats.rx_packets += slave_stats.tx_packets;
1969 bond_stats.rx_bytes += slave_stats.tx_bytes;
1973 if (!netdev_open_default(port->name, &bond_dev)) {
1974 netdev_set_stats(bond_dev, &bond_stats);
1975 netdev_close(bond_dev);
1980 bond_run(struct bridge *br)
1984 for (i = 0; i < br->n_ports; i++) {
1985 struct port *port = br->ports[i];
1987 if (port->n_ifaces >= 2) {
1988 for (j = 0; j < port->n_ifaces; j++) {
1989 struct iface *iface = port->ifaces[j];
1990 if (time_msec() >= iface->delay_expires) {
1991 bond_enable_slave(iface, !iface->enabled);
1995 if (port->bond_fake_iface
1996 && time_msec() >= port->bond_next_fake_iface_update) {
1997 bond_update_fake_iface_stats(port);
1998 port->bond_next_fake_iface_update = time_msec() + 1000;
2002 if (port->bond_compat_is_stale) {
2003 port->bond_compat_is_stale = false;
2004 port_update_bond_compat(port);
2010 bond_wait(struct bridge *br)
2014 for (i = 0; i < br->n_ports; i++) {
2015 struct port *port = br->ports[i];
2016 if (port->n_ifaces < 2) {
2019 for (j = 0; j < port->n_ifaces; j++) {
2020 struct iface *iface = port->ifaces[j];
2021 if (iface->delay_expires != LLONG_MAX) {
2022 poll_timer_wait_until(iface->delay_expires);
2025 if (port->bond_fake_iface) {
2026 poll_timer_wait_until(port->bond_next_fake_iface_update);
2032 set_dst(struct dst *p, const flow_t *flow,
2033 const struct port *in_port, const struct port *out_port,
2036 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2037 : in_port->vlan >= 0 ? in_port->vlan
2038 : ntohs(flow->dl_vlan));
2039 return choose_output_iface(out_port, flow->dl_src, &p->xf_ifidx, tags);
2043 swap_dst(struct dst *p, struct dst *q)
2045 struct dst tmp = *p;
2050 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2051 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2052 * that we push to the datapath. We could in fact fully sort the array by
2053 * vlan, but in most cases there are at most two different vlan tags so that's
2054 * possibly overkill.) */
2056 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
2058 struct dst *first = dsts;
2059 struct dst *last = dsts + n_dsts;
2061 while (first != last) {
2063 * - All dsts < first have vlan == 'vlan'.
2064 * - All dsts >= last have vlan != 'vlan'.
2065 * - first < last. */
2066 while (first->vlan == vlan) {
2067 if (++first == last) {
2072 /* Same invariants, plus one additional:
2073 * - first->vlan != vlan.
2075 while (last[-1].vlan != vlan) {
2076 if (--last == first) {
2081 /* Same invariants, plus one additional:
2082 * - last[-1].vlan == vlan.*/
2083 swap_dst(first++, --last);
2088 mirror_mask_ffs(mirror_mask_t mask)
2090 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2095 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
2096 const struct dst *test)
2099 for (i = 0; i < n_dsts; i++) {
2100 if (dsts[i].vlan == test->vlan && dsts[i].xf_ifidx == test->xf_ifidx) {
2108 port_trunks_vlan(const struct port *port, uint16_t vlan)
2110 return (port->vlan < 0
2111 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2115 port_includes_vlan(const struct port *port, uint16_t vlan)
2117 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2121 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
2122 const struct port *in_port, const struct port *out_port,
2123 struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface)
2125 mirror_mask_t mirrors = in_port->src_mirrors;
2126 struct dst *dst = dsts;
2129 if (out_port == FLOOD_PORT) {
2130 /* XXX use XFLOW_FLOOD if no vlans or bonding. */
2131 /* XXX even better, define each VLAN as a datapath port group */
2132 for (i = 0; i < br->n_ports; i++) {
2133 struct port *port = br->ports[i];
2134 if (port != in_port && port_includes_vlan(port, vlan)
2135 && !port->is_mirror_output_port
2136 && set_dst(dst, flow, in_port, port, tags)) {
2137 mirrors |= port->dst_mirrors;
2141 *nf_output_iface = NF_OUT_FLOOD;
2142 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
2143 *nf_output_iface = dst->xf_ifidx;
2144 mirrors |= out_port->dst_mirrors;
2149 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2150 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2152 if (set_dst(dst, flow, in_port, m->out_port, tags)
2153 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
2157 for (i = 0; i < br->n_ports; i++) {
2158 struct port *port = br->ports[i];
2159 if (port_includes_vlan(port, m->out_vlan)
2160 && set_dst(dst, flow, in_port, port, tags))
2164 if (port->vlan < 0) {
2165 dst->vlan = m->out_vlan;
2167 if (dst_is_duplicate(dsts, dst - dsts, dst)) {
2171 /* Use the vlan tag on the original flow instead of
2172 * the one passed in the vlan parameter. This ensures
2173 * that we compare the vlan from before any implicit
2174 * tagging tags place. This is necessary because
2175 * dst->vlan is the final vlan, after removing implicit
2177 flow_vlan = ntohs(flow->dl_vlan);
2178 if (flow_vlan == 0) {
2179 flow_vlan = OFP_VLAN_NONE;
2181 if (port == in_port && dst->vlan == flow_vlan) {
2182 /* Don't send out input port on same VLAN. */
2190 mirrors &= mirrors - 1;
2193 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
2197 static void OVS_UNUSED
2198 print_dsts(const struct dst *dsts, size_t n)
2200 for (; n--; dsts++) {
2201 printf(">p%"PRIu16, dsts->xf_ifidx);
2202 if (dsts->vlan != OFP_VLAN_NONE) {
2203 printf("v%"PRIu16, dsts->vlan);
2209 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
2210 const struct port *in_port, const struct port *out_port,
2211 tag_type *tags, struct xflow_actions *actions,
2212 uint16_t *nf_output_iface)
2214 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
2216 const struct dst *p;
2219 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags,
2222 cur_vlan = ntohs(flow->dl_vlan);
2223 for (p = dsts; p < &dsts[n_dsts]; p++) {
2224 union xflow_action *a;
2225 if (p->vlan != cur_vlan) {
2226 if (p->vlan == OFP_VLAN_NONE) {
2227 xflow_actions_add(actions, XFLOWAT_STRIP_VLAN);
2229 a = xflow_actions_add(actions, XFLOWAT_SET_DL_TCI);
2230 a->dl_tci.tci = htons(p->vlan & VLAN_VID_MASK);
2231 a->dl_tci.mask = htons(VLAN_VID_MASK);
2235 a = xflow_actions_add(actions, XFLOWAT_OUTPUT);
2236 a->output.port = p->xf_ifidx;
2240 /* Returns the effective vlan of a packet, taking into account both the
2241 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2242 * the packet is untagged and -1 indicates it has an invalid header and
2243 * should be dropped. */
2244 static int flow_get_vlan(struct bridge *br, const flow_t *flow,
2245 struct port *in_port, bool have_packet)
2247 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2248 * belongs to VLAN 0, so we should treat both cases identically. (In the
2249 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2250 * presumably to allow a priority to be specified. In the latter case, the
2251 * packet does not have any 802.1Q header.) */
2252 int vlan = ntohs(flow->dl_vlan);
2253 if (vlan == OFP_VLAN_NONE) {
2256 if (in_port->vlan >= 0) {
2258 /* XXX support double tagging? */
2260 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2261 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
2262 "packet received on port %s configured with "
2263 "implicit VLAN %"PRIu16,
2264 br->name, ntohs(flow->dl_vlan),
2265 in_port->name, in_port->vlan);
2269 vlan = in_port->vlan;
2271 if (!port_includes_vlan(in_port, vlan)) {
2273 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2274 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2275 "packet received on port %s not configured for "
2277 br->name, vlan, in_port->name, vlan);
2286 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2287 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2288 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2290 is_gratuitous_arp(const flow_t *flow)
2292 return (flow->dl_type == htons(ETH_TYPE_ARP)
2293 && eth_addr_is_broadcast(flow->dl_dst)
2294 && (flow->nw_proto == ARP_OP_REPLY
2295 || (flow->nw_proto == ARP_OP_REQUEST
2296 && flow->nw_src == flow->nw_dst)));
2300 update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
2301 struct port *in_port)
2303 enum grat_arp_lock_type lock_type;
2306 /* We don't want to learn from gratuitous ARP packets that are reflected
2307 * back over bond slaves so we lock the learning table. */
2308 lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
2309 (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
2310 GRAT_ARP_LOCK_CHECK;
2312 rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
2315 /* The log messages here could actually be useful in debugging,
2316 * so keep the rate limit relatively high. */
2317 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
2319 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2320 "on port %s in VLAN %d",
2321 br->name, ETH_ADDR_ARGS(flow->dl_src),
2322 in_port->name, vlan);
2323 ofproto_revalidate(br->ofproto, rev_tag);
2327 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2328 * dropped. Returns true if they may be forwarded, false if they should be
2331 * If 'have_packet' is true, it indicates that the caller is processing a
2332 * received packet. If 'have_packet' is false, then the caller is just
2333 * revalidating an existing flow because configuration has changed. Either
2334 * way, 'have_packet' only affects logging (there is no point in logging errors
2335 * during revalidation).
2337 * Sets '*in_portp' to the input port. This will be a null pointer if
2338 * flow->in_port does not designate a known input port (in which case
2339 * is_admissible() returns false).
2341 * When returning true, sets '*vlanp' to the effective VLAN of the input
2342 * packet, as returned by flow_get_vlan().
2344 * May also add tags to '*tags', although the current implementation only does
2345 * so in one special case.
2348 is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
2349 tag_type *tags, int *vlanp, struct port **in_portp)
2351 struct iface *in_iface;
2352 struct port *in_port;
2355 /* Find the interface and port structure for the received packet. */
2356 in_iface = iface_from_xf_ifidx(br, ofp_port_to_xflow_port(flow->in_port));
2358 /* No interface? Something fishy... */
2360 /* Odd. A few possible reasons here:
2362 * - We deleted an interface but there are still a few packets
2363 * queued up from it.
2365 * - Someone externally added an interface (e.g. with "ovs-dpctl
2366 * add-if") that we don't know about.
2368 * - Packet arrived on the local port but the local port is not
2369 * one of our bridge ports.
2371 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2373 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2374 "interface %"PRIu16, br->name, flow->in_port);
2380 *in_portp = in_port = in_iface->port;
2381 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2386 /* Drop frames for reserved multicast addresses. */
2387 if (eth_addr_is_reserved(flow->dl_dst)) {
2391 /* Drop frames on ports reserved for mirroring. */
2392 if (in_port->is_mirror_output_port) {
2394 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2395 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2396 "%s, which is reserved exclusively for mirroring",
2397 br->name, in_port->name);
2402 /* Packets received on bonds need special attention to avoid duplicates. */
2403 if (in_port->n_ifaces > 1) {
2405 bool is_grat_arp_locked;
2407 if (eth_addr_is_multicast(flow->dl_dst)) {
2408 *tags |= in_port->active_iface_tag;
2409 if (in_port->active_iface != in_iface->port_ifidx) {
2410 /* Drop all multicast packets on inactive slaves. */
2415 /* Drop all packets for which we have learned a different input
2416 * port, because we probably sent the packet on one slave and got
2417 * it back on the other. Gratuitous ARP packets are an exception
2418 * to this rule: the host has moved to another switch. The exception
2419 * to the exception is if we locked the learning table to avoid
2420 * reflections on bond slaves. If this is the case, just drop the
2422 src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
2423 &is_grat_arp_locked);
2424 if (src_idx != -1 && src_idx != in_port->port_idx &&
2425 (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
2433 /* If the composed actions may be applied to any packet in the given 'flow',
2434 * returns true. Otherwise, the actions should only be applied to 'packet', or
2435 * not at all, if 'packet' was NULL. */
2437 process_flow(struct bridge *br, const flow_t *flow,
2438 const struct ofpbuf *packet, struct xflow_actions *actions,
2439 tag_type *tags, uint16_t *nf_output_iface)
2441 struct port *in_port;
2442 struct port *out_port;
2446 /* Check whether we should drop packets in this flow. */
2447 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2452 /* Learn source MAC (but don't try to learn from revalidation). */
2454 update_learning_table(br, flow, vlan, in_port);
2457 /* Determine output port. */
2458 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
2460 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
2461 out_port = br->ports[out_port_idx];
2462 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2463 /* If we are revalidating but don't have a learning entry then
2464 * eject the flow. Installing a flow that floods packets opens
2465 * up a window of time where we could learn from a packet reflected
2466 * on a bond and blackhole packets before the learning table is
2467 * updated to reflect the correct port. */
2470 out_port = FLOOD_PORT;
2473 /* Don't send packets out their input ports. */
2474 if (in_port == out_port) {
2480 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
2487 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2490 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
2491 const struct ofp_phy_port *opp,
2494 struct bridge *br = br_;
2495 struct iface *iface;
2498 iface = iface_from_xf_ifidx(br, ofp_port_to_xflow_port(opp->port_no));
2504 if (reason == OFPPR_DELETE) {
2505 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2506 br->name, iface->name);
2507 iface_destroy(iface);
2508 if (!port->n_ifaces) {
2509 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2510 br->name, port->name);
2516 if (port->n_ifaces > 1) {
2517 bool up = !(opp->state & OFPPS_LINK_DOWN);
2518 bond_link_status_update(iface, up);
2519 port_update_bond_compat(port);
2525 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
2526 struct xflow_actions *actions, tag_type *tags,
2527 uint16_t *nf_output_iface, void *br_)
2529 struct bridge *br = br_;
2531 COVERAGE_INC(bridge_process_flow);
2533 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
2537 bridge_account_flow_ofhook_cb(const flow_t *flow, tag_type tags,
2538 const union xflow_action *actions,
2539 size_t n_actions, unsigned long long int n_bytes,
2542 struct bridge *br = br_;
2543 const union xflow_action *a;
2544 struct port *in_port;
2548 /* Feed information from the active flows back into the learning table to
2549 * ensure that table is always in sync with what is actually flowing
2550 * through the datapath.
2552 * We test that 'tags' is nonzero to ensure that only flows that include an
2553 * OFPP_NORMAL action are used for learning. This works because
2554 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2555 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
2556 update_learning_table(br, flow, vlan, in_port);
2559 /* Account for bond slave utilization. */
2560 if (!br->has_bonded_ports) {
2563 for (a = actions; a < &actions[n_actions]; a++) {
2564 if (a->type == XFLOWAT_OUTPUT) {
2565 struct port *out_port = port_from_xf_ifidx(br, a->output.port);
2566 if (out_port && out_port->n_ifaces >= 2) {
2567 struct bond_entry *e = lookup_bond_entry(out_port,
2569 e->tx_bytes += n_bytes;
2576 bridge_account_checkpoint_ofhook_cb(void *br_)
2578 struct bridge *br = br_;
2582 if (!br->has_bonded_ports) {
2587 for (i = 0; i < br->n_ports; i++) {
2588 struct port *port = br->ports[i];
2589 if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
2590 port->bond_next_rebalance = now + port->bond_rebalance_interval;
2591 bond_rebalance_port(port);
2596 static struct ofhooks bridge_ofhooks = {
2597 bridge_port_changed_ofhook_cb,
2598 bridge_normal_ofhook_cb,
2599 bridge_account_flow_ofhook_cb,
2600 bridge_account_checkpoint_ofhook_cb,
2603 /* Bonding functions. */
2605 /* Statistics for a single interface on a bonded port, used for load-based
2606 * bond rebalancing. */
2607 struct slave_balance {
2608 struct iface *iface; /* The interface. */
2609 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
2611 /* All the "bond_entry"s that are assigned to this interface, in order of
2612 * increasing tx_bytes. */
2613 struct bond_entry **hashes;
2617 /* Sorts pointers to pointers to bond_entries in ascending order by the
2618 * interface to which they are assigned, and within a single interface in
2619 * ascending order of bytes transmitted. */
2621 compare_bond_entries(const void *a_, const void *b_)
2623 const struct bond_entry *const *ap = a_;
2624 const struct bond_entry *const *bp = b_;
2625 const struct bond_entry *a = *ap;
2626 const struct bond_entry *b = *bp;
2627 if (a->iface_idx != b->iface_idx) {
2628 return a->iface_idx > b->iface_idx ? 1 : -1;
2629 } else if (a->tx_bytes != b->tx_bytes) {
2630 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2636 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2637 * *descending* order by number of bytes transmitted. */
2639 compare_slave_balance(const void *a_, const void *b_)
2641 const struct slave_balance *a = a_;
2642 const struct slave_balance *b = b_;
2643 if (a->iface->enabled != b->iface->enabled) {
2644 return a->iface->enabled ? -1 : 1;
2645 } else if (a->tx_bytes != b->tx_bytes) {
2646 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2653 swap_bals(struct slave_balance *a, struct slave_balance *b)
2655 struct slave_balance tmp = *a;
2660 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2661 * given that 'p' (and only 'p') might be in the wrong location.
2663 * This function invalidates 'p', since it might now be in a different memory
2666 resort_bals(struct slave_balance *p,
2667 struct slave_balance bals[], size_t n_bals)
2670 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2671 swap_bals(p, p - 1);
2673 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2674 swap_bals(p, p + 1);
2680 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2682 if (VLOG_IS_DBG_ENABLED()) {
2683 struct ds ds = DS_EMPTY_INITIALIZER;
2684 const struct slave_balance *b;
2686 for (b = bals; b < bals + n_bals; b++) {
2690 ds_put_char(&ds, ',');
2692 ds_put_format(&ds, " %s %"PRIu64"kB",
2693 b->iface->name, b->tx_bytes / 1024);
2695 if (!b->iface->enabled) {
2696 ds_put_cstr(&ds, " (disabled)");
2698 if (b->n_hashes > 0) {
2699 ds_put_cstr(&ds, " (");
2700 for (i = 0; i < b->n_hashes; i++) {
2701 const struct bond_entry *e = b->hashes[i];
2703 ds_put_cstr(&ds, " + ");
2705 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2706 e - port->bond_hash, e->tx_bytes / 1024);
2708 ds_put_cstr(&ds, ")");
2711 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2716 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2718 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2721 struct bond_entry *hash = from->hashes[hash_idx];
2722 struct port *port = from->iface->port;
2723 uint64_t delta = hash->tx_bytes;
2725 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2726 "from %s to %s (now carrying %"PRIu64"kB and "
2727 "%"PRIu64"kB load, respectively)",
2728 port->name, delta / 1024, hash - port->bond_hash,
2729 from->iface->name, to->iface->name,
2730 (from->tx_bytes - delta) / 1024,
2731 (to->tx_bytes + delta) / 1024);
2733 /* Delete element from from->hashes.
2735 * We don't bother to add the element to to->hashes because not only would
2736 * it require more work, the only purpose it would be to allow that hash to
2737 * be migrated to another slave in this rebalancing run, and there is no
2738 * point in doing that. */
2739 if (hash_idx == 0) {
2742 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
2743 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
2747 /* Shift load away from 'from' to 'to'. */
2748 from->tx_bytes -= delta;
2749 to->tx_bytes += delta;
2751 /* Arrange for flows to be revalidated. */
2752 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2753 hash->iface_idx = to->iface->port_ifidx;
2754 hash->iface_tag = tag_create_random();
2758 bond_rebalance_port(struct port *port)
2760 struct slave_balance bals[DP_MAX_PORTS];
2762 struct bond_entry *hashes[BOND_MASK + 1];
2763 struct slave_balance *b, *from, *to;
2764 struct bond_entry *e;
2767 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2768 * descending order of tx_bytes, so that bals[0] represents the most
2769 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2772 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2773 * array for each slave_balance structure, we sort our local array of
2774 * hashes in order by slave, so that all of the hashes for a given slave
2775 * become contiguous in memory, and then we point each 'hashes' members of
2776 * a slave_balance structure to the start of a contiguous group. */
2777 n_bals = port->n_ifaces;
2778 for (b = bals; b < &bals[n_bals]; b++) {
2779 b->iface = port->ifaces[b - bals];
2784 for (i = 0; i <= BOND_MASK; i++) {
2785 hashes[i] = &port->bond_hash[i];
2787 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2788 for (i = 0; i <= BOND_MASK; i++) {
2790 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2791 b = &bals[e->iface_idx];
2792 b->tx_bytes += e->tx_bytes;
2794 b->hashes = &hashes[i];
2799 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2800 log_bals(bals, n_bals, port);
2802 /* Discard slaves that aren't enabled (which were sorted to the back of the
2803 * array earlier). */
2804 while (!bals[n_bals - 1].iface->enabled) {
2811 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2812 to = &bals[n_bals - 1];
2813 for (from = bals; from < to; ) {
2814 uint64_t overload = from->tx_bytes - to->tx_bytes;
2815 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2816 /* The extra load on 'from' (and all less-loaded slaves), compared
2817 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2818 * it is less than ~1Mbps. No point in rebalancing. */
2820 } else if (from->n_hashes == 1) {
2821 /* 'from' only carries a single MAC hash, so we can't shift any
2822 * load away from it, even though we want to. */
2825 /* 'from' is carrying significantly more load than 'to', and that
2826 * load is split across at least two different hashes. Pick a hash
2827 * to migrate to 'to' (the least-loaded slave), given that doing so
2828 * must decrease the ratio of the load on the two slaves by at
2831 * The sort order we use means that we prefer to shift away the
2832 * smallest hashes instead of the biggest ones. There is little
2833 * reason behind this decision; we could use the opposite sort
2834 * order to shift away big hashes ahead of small ones. */
2837 for (i = 0; i < from->n_hashes; i++) {
2838 double old_ratio, new_ratio;
2839 uint64_t delta = from->hashes[i]->tx_bytes;
2841 if (delta == 0 || from->tx_bytes - delta == 0) {
2842 /* Pointless move. */
2846 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
2848 if (to->tx_bytes == 0) {
2849 /* Nothing on the new slave, move it. */
2853 old_ratio = (double)from->tx_bytes / to->tx_bytes;
2854 new_ratio = (double)(from->tx_bytes - delta) /
2855 (to->tx_bytes + delta);
2857 if (new_ratio == 0) {
2858 /* Should already be covered but check to prevent division
2863 if (new_ratio < 1) {
2864 new_ratio = 1 / new_ratio;
2867 if (old_ratio - new_ratio > 0.1) {
2868 /* Would decrease the ratio, move it. */
2872 if (i < from->n_hashes) {
2873 bond_shift_load(from, to, i);
2874 port->bond_compat_is_stale = true;
2876 /* If the result of the migration changed the relative order of
2877 * 'from' and 'to' swap them back to maintain invariants. */
2878 if (order_swapped) {
2879 swap_bals(from, to);
2882 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2883 * point to different slave_balance structures. It is only
2884 * valid to do these two operations in a row at all because we
2885 * know that 'from' will not move past 'to' and vice versa. */
2886 resort_bals(from, bals, n_bals);
2887 resort_bals(to, bals, n_bals);
2894 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2895 * historical data to decay to <1% in 7 rebalancing runs. */
2896 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2902 bond_send_learning_packets(struct port *port)
2904 struct bridge *br = port->bridge;
2905 struct mac_entry *e;
2906 struct ofpbuf packet;
2907 int error, n_packets, n_errors;
2909 if (!port->n_ifaces || port->active_iface < 0) {
2913 ofpbuf_init(&packet, 128);
2914 error = n_packets = n_errors = 0;
2915 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
2916 union ofp_action actions[2], *a;
2922 if (e->port == port->port_idx
2923 || !choose_output_iface(port, e->mac, &xf_ifidx, &tags)) {
2927 /* Compose actions. */
2928 memset(actions, 0, sizeof actions);
2931 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2932 a->vlan_vid.len = htons(sizeof *a);
2933 a->vlan_vid.vlan_vid = htons(e->vlan);
2936 a->output.type = htons(OFPAT_OUTPUT);
2937 a->output.len = htons(sizeof *a);
2938 a->output.port = htons(xflow_port_to_ofp_port(xf_ifidx));
2943 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
2945 flow_extract(&packet, 0, OFPP_NONE, &flow);
2946 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2953 ofpbuf_uninit(&packet);
2956 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2957 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2958 "packets, last error was: %s",
2959 port->name, n_errors, n_packets, strerror(error));
2961 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2962 port->name, n_packets);
2966 /* Bonding unixctl user interface functions. */
2969 bond_unixctl_list(struct unixctl_conn *conn,
2970 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
2972 struct ds ds = DS_EMPTY_INITIALIZER;
2973 const struct bridge *br;
2975 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2977 LIST_FOR_EACH (br, node, &all_bridges) {
2980 for (i = 0; i < br->n_ports; i++) {
2981 const struct port *port = br->ports[i];
2982 if (port->n_ifaces > 1) {
2985 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2986 for (j = 0; j < port->n_ifaces; j++) {
2987 const struct iface *iface = port->ifaces[j];
2989 ds_put_cstr(&ds, ", ");
2991 ds_put_cstr(&ds, iface->name);
2993 ds_put_char(&ds, '\n');
2997 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3001 static struct port *
3002 bond_find(const char *name)
3004 const struct bridge *br;
3006 LIST_FOR_EACH (br, node, &all_bridges) {
3009 for (i = 0; i < br->n_ports; i++) {
3010 struct port *port = br->ports[i];
3011 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3020 bond_unixctl_show(struct unixctl_conn *conn,
3021 const char *args, void *aux OVS_UNUSED)
3023 struct ds ds = DS_EMPTY_INITIALIZER;
3024 const struct port *port;
3027 port = bond_find(args);
3029 unixctl_command_reply(conn, 501, "no such bond");
3033 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3034 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3035 ds_put_format(&ds, "next rebalance: %lld ms\n",
3036 port->bond_next_rebalance - time_msec());
3037 for (j = 0; j < port->n_ifaces; j++) {
3038 const struct iface *iface = port->ifaces[j];
3039 struct bond_entry *be;
3042 ds_put_format(&ds, "slave %s: %s\n",
3043 iface->name, iface->enabled ? "enabled" : "disabled");
3044 if (j == port->active_iface) {
3045 ds_put_cstr(&ds, "\tactive slave\n");
3047 if (iface->delay_expires != LLONG_MAX) {
3048 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3049 iface->enabled ? "downdelay" : "updelay",
3050 iface->delay_expires - time_msec());
3054 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3055 int hash = be - port->bond_hash;
3056 struct mac_entry *me;
3058 if (be->iface_idx != j) {
3062 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3063 hash, be->tx_bytes / 1024);
3066 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3069 if (bond_hash(me->mac) == hash
3070 && me->port != port->port_idx
3071 && choose_output_iface(port, me->mac, &xf_ifidx, &tags)
3072 && xf_ifidx == iface->xf_ifidx)
3074 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3075 ETH_ADDR_ARGS(me->mac));
3080 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3085 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3086 void *aux OVS_UNUSED)
3088 char *args = (char *) args_;
3089 char *save_ptr = NULL;
3090 char *bond_s, *hash_s, *slave_s;
3091 uint8_t mac[ETH_ADDR_LEN];
3093 struct iface *iface;
3094 struct bond_entry *entry;
3097 bond_s = strtok_r(args, " ", &save_ptr);
3098 hash_s = strtok_r(NULL, " ", &save_ptr);
3099 slave_s = strtok_r(NULL, " ", &save_ptr);
3101 unixctl_command_reply(conn, 501,
3102 "usage: bond/migrate BOND HASH SLAVE");
3106 port = bond_find(bond_s);
3108 unixctl_command_reply(conn, 501, "no such bond");
3112 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3113 == ETH_ADDR_SCAN_COUNT) {
3114 hash = bond_hash(mac);
3115 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3116 hash = atoi(hash_s) & BOND_MASK;
3118 unixctl_command_reply(conn, 501, "bad hash");
3122 iface = port_lookup_iface(port, slave_s);
3124 unixctl_command_reply(conn, 501, "no such slave");
3128 if (!iface->enabled) {
3129 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3133 entry = &port->bond_hash[hash];
3134 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
3135 entry->iface_idx = iface->port_ifidx;
3136 entry->iface_tag = tag_create_random();
3137 port->bond_compat_is_stale = true;
3138 unixctl_command_reply(conn, 200, "migrated");
3142 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3143 void *aux OVS_UNUSED)
3145 char *args = (char *) args_;
3146 char *save_ptr = NULL;
3147 char *bond_s, *slave_s;
3149 struct iface *iface;
3151 bond_s = strtok_r(args, " ", &save_ptr);
3152 slave_s = strtok_r(NULL, " ", &save_ptr);
3154 unixctl_command_reply(conn, 501,
3155 "usage: bond/set-active-slave BOND SLAVE");
3159 port = bond_find(bond_s);
3161 unixctl_command_reply(conn, 501, "no such bond");
3165 iface = port_lookup_iface(port, slave_s);
3167 unixctl_command_reply(conn, 501, "no such slave");
3171 if (!iface->enabled) {
3172 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3176 if (port->active_iface != iface->port_ifidx) {
3177 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3178 port->active_iface = iface->port_ifidx;
3179 port->active_iface_tag = tag_create_random();
3180 VLOG_INFO("port %s: active interface is now %s",
3181 port->name, iface->name);
3182 bond_send_learning_packets(port);
3183 unixctl_command_reply(conn, 200, "done");
3185 unixctl_command_reply(conn, 200, "no change");
3190 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3192 char *args = (char *) args_;
3193 char *save_ptr = NULL;
3194 char *bond_s, *slave_s;
3196 struct iface *iface;
3198 bond_s = strtok_r(args, " ", &save_ptr);
3199 slave_s = strtok_r(NULL, " ", &save_ptr);
3201 unixctl_command_reply(conn, 501,
3202 "usage: bond/enable/disable-slave BOND SLAVE");
3206 port = bond_find(bond_s);
3208 unixctl_command_reply(conn, 501, "no such bond");
3212 iface = port_lookup_iface(port, slave_s);
3214 unixctl_command_reply(conn, 501, "no such slave");
3218 bond_enable_slave(iface, enable);
3219 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3223 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3224 void *aux OVS_UNUSED)
3226 enable_slave(conn, args, true);
3230 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3231 void *aux OVS_UNUSED)
3233 enable_slave(conn, args, false);
3237 bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
3238 void *aux OVS_UNUSED)
3240 uint8_t mac[ETH_ADDR_LEN];
3244 if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3245 == ETH_ADDR_SCAN_COUNT) {
3246 hash = bond_hash(mac);
3248 hash_cstr = xasprintf("%u", hash);
3249 unixctl_command_reply(conn, 200, hash_cstr);
3252 unixctl_command_reply(conn, 501, "invalid mac");
3259 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3260 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3261 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3262 unixctl_command_register("bond/set-active-slave",
3263 bond_unixctl_set_active_slave, NULL);
3264 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3266 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3268 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3271 /* Port functions. */
3273 static struct port *
3274 port_create(struct bridge *br, const char *name)
3278 port = xzalloc(sizeof *port);
3280 port->port_idx = br->n_ports;
3282 port->trunks = NULL;
3283 port->name = xstrdup(name);
3284 port->active_iface = -1;
3286 if (br->n_ports >= br->allocated_ports) {
3287 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
3290 br->ports[br->n_ports++] = port;
3291 shash_add_assert(&br->port_by_name, port->name, port);
3293 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3300 get_port_other_config(const struct ovsrec_port *port, const char *key,
3301 const char *default_value)
3305 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3307 return value ? value : default_value;
3311 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3313 struct shash new_ifaces;
3316 /* Collect list of new interfaces. */
3317 shash_init(&new_ifaces);
3318 for (i = 0; i < cfg->n_interfaces; i++) {
3319 const char *name = cfg->interfaces[i]->name;
3320 shash_add_once(&new_ifaces, name, NULL);
3323 /* Get rid of deleted interfaces. */
3324 for (i = 0; i < port->n_ifaces; ) {
3325 if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
3326 iface_destroy(port->ifaces[i]);
3332 shash_destroy(&new_ifaces);
3336 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
3338 struct shash new_ifaces;
3339 long long int next_rebalance;
3340 unsigned long *trunks;
3346 /* Update settings. */
3347 port->updelay = cfg->bond_updelay;
3348 if (port->updelay < 0) {
3351 port->downdelay = cfg->bond_downdelay;
3352 if (port->downdelay < 0) {
3353 port->downdelay = 0;
3355 port->bond_rebalance_interval = atoi(
3356 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
3357 if (port->bond_rebalance_interval < 1000) {
3358 port->bond_rebalance_interval = 1000;
3360 next_rebalance = time_msec() + port->bond_rebalance_interval;
3361 if (port->bond_next_rebalance > next_rebalance) {
3362 port->bond_next_rebalance = next_rebalance;
3365 /* Add new interfaces and update 'cfg' member of existing ones. */
3366 shash_init(&new_ifaces);
3367 for (i = 0; i < cfg->n_interfaces; i++) {
3368 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
3369 struct iface *iface;
3371 if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
3372 VLOG_WARN("port %s: %s specified twice as port interface",
3373 port->name, if_cfg->name);
3377 iface = iface_lookup(port->bridge, if_cfg->name);
3379 if (iface->port != port) {
3380 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3382 port->bridge->name, if_cfg->name, iface->port->name);
3385 iface->cfg = if_cfg;
3387 iface_create(port, if_cfg);
3390 shash_destroy(&new_ifaces);
3395 if (port->n_ifaces < 2) {
3397 if (vlan >= 0 && vlan <= 4095) {
3398 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
3403 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3404 * they even work as-is. But they have not been tested. */
3405 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3409 if (port->vlan != vlan) {
3411 bridge_flush(port->bridge);
3414 /* Get trunked VLANs. */
3416 if (vlan < 0 && cfg->n_trunks) {
3419 trunks = bitmap_allocate(4096);
3421 for (i = 0; i < cfg->n_trunks; i++) {
3422 int trunk = cfg->trunks[i];
3424 bitmap_set1(trunks, trunk);
3430 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3431 port->name, cfg->n_trunks);
3433 if (n_errors == cfg->n_trunks) {
3434 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3436 bitmap_free(trunks);
3439 } else if (vlan >= 0 && cfg->n_trunks) {
3440 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3444 ? port->trunks != NULL
3445 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
3446 bridge_flush(port->bridge);
3448 bitmap_free(port->trunks);
3449 port->trunks = trunks;
3453 port_destroy(struct port *port)
3456 struct bridge *br = port->bridge;
3460 proc_net_compat_update_vlan(port->name, NULL, 0);
3461 proc_net_compat_update_bond(port->name, NULL);
3463 for (i = 0; i < MAX_MIRRORS; i++) {
3464 struct mirror *m = br->mirrors[i];
3465 if (m && m->out_port == port) {
3470 while (port->n_ifaces > 0) {
3471 iface_destroy(port->ifaces[port->n_ifaces - 1]);
3474 shash_find_and_delete_assert(&br->port_by_name, port->name);
3476 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
3477 del->port_idx = port->port_idx;
3480 bitmap_free(port->trunks);
3487 static struct port *
3488 port_from_xf_ifidx(const struct bridge *br, uint16_t xf_ifidx)
3490 struct iface *iface = iface_from_xf_ifidx(br, xf_ifidx);
3491 return iface ? iface->port : NULL;
3494 static struct port *
3495 port_lookup(const struct bridge *br, const char *name)
3497 return shash_find_data(&br->port_by_name, name);
3500 static struct iface *
3501 port_lookup_iface(const struct port *port, const char *name)
3503 struct iface *iface = iface_lookup(port->bridge, name);
3504 return iface && iface->port == port ? iface : NULL;
3508 port_update_bonding(struct port *port)
3510 if (port->n_ifaces < 2) {
3511 /* Not a bonded port. */
3512 if (port->bond_hash) {
3513 free(port->bond_hash);
3514 port->bond_hash = NULL;
3515 port->bond_compat_is_stale = true;
3516 port->bond_fake_iface = false;
3519 if (!port->bond_hash) {
3522 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
3523 for (i = 0; i <= BOND_MASK; i++) {
3524 struct bond_entry *e = &port->bond_hash[i];
3528 port->no_ifaces_tag = tag_create_random();
3529 bond_choose_active_iface(port);
3530 port->bond_next_rebalance
3531 = time_msec() + port->bond_rebalance_interval;
3533 if (port->cfg->bond_fake_iface) {
3534 port->bond_next_fake_iface_update = time_msec();
3537 port->bond_compat_is_stale = true;
3538 port->bond_fake_iface = port->cfg->bond_fake_iface;
3543 port_update_bond_compat(struct port *port)
3545 struct compat_bond_hash compat_hashes[BOND_MASK + 1];
3546 struct compat_bond bond;
3549 if (port->n_ifaces < 2) {
3550 proc_net_compat_update_bond(port->name, NULL);
3555 bond.updelay = port->updelay;
3556 bond.downdelay = port->downdelay;
3559 bond.hashes = compat_hashes;
3560 if (port->bond_hash) {
3561 const struct bond_entry *e;
3562 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
3563 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
3564 struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
3565 cbh->hash = e - port->bond_hash;
3566 cbh->netdev_name = port->ifaces[e->iface_idx]->name;
3571 bond.n_slaves = port->n_ifaces;
3572 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
3573 for (i = 0; i < port->n_ifaces; i++) {
3574 struct iface *iface = port->ifaces[i];
3575 struct compat_bond_slave *slave = &bond.slaves[i];
3576 slave->name = iface->name;
3578 /* We need to make the same determination as the Linux bonding
3579 * code to determine whether a slave should be consider "up".
3580 * The Linux function bond_miimon_inspect() supports four
3581 * BOND_LINK_* states:
3583 * - BOND_LINK_UP: carrier detected, updelay has passed.
3584 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3585 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3586 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3588 * The function bond_info_show_slave() only considers BOND_LINK_UP
3589 * to be "up" and anything else to be "down".
3591 slave->up = iface->enabled && iface->delay_expires == LLONG_MAX;
3595 netdev_get_etheraddr(iface->netdev, slave->mac);
3598 if (port->bond_fake_iface) {
3599 struct netdev *bond_netdev;
3601 if (!netdev_open_default(port->name, &bond_netdev)) {
3603 netdev_turn_flags_on(bond_netdev, NETDEV_UP, true);
3605 netdev_turn_flags_off(bond_netdev, NETDEV_UP, true);
3607 netdev_close(bond_netdev);
3611 proc_net_compat_update_bond(port->name, &bond);
3616 port_update_vlan_compat(struct port *port)
3618 struct bridge *br = port->bridge;
3619 char *vlandev_name = NULL;
3621 if (port->vlan > 0) {
3622 /* Figure out the name that the VLAN device should actually have, if it
3623 * existed. This takes some work because the VLAN device would not
3624 * have port->name in its name; rather, it would have the trunk port's
3625 * name, and 'port' would be attached to a bridge that also had the
3626 * VLAN device one of its ports. So we need to find a trunk port that
3627 * includes port->vlan.
3629 * There might be more than one candidate. This doesn't happen on
3630 * XenServer, so if it happens we just pick the first choice in
3631 * alphabetical order instead of creating multiple VLAN devices. */
3633 for (i = 0; i < br->n_ports; i++) {
3634 struct port *p = br->ports[i];
3635 if (port_trunks_vlan(p, port->vlan)
3637 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
3639 uint8_t ea[ETH_ADDR_LEN];
3640 netdev_get_etheraddr(p->ifaces[0]->netdev, ea);
3641 if (!eth_addr_is_multicast(ea) &&
3642 !eth_addr_is_reserved(ea) &&
3643 !eth_addr_is_zero(ea)) {
3644 vlandev_name = p->name;
3649 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
3652 /* Interface functions. */
3654 static struct iface *
3655 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
3657 struct bridge *br = port->bridge;
3658 struct iface *iface;
3659 char *name = if_cfg->name;
3662 iface = xzalloc(sizeof *iface);
3664 iface->port_ifidx = port->n_ifaces;
3665 iface->name = xstrdup(name);
3666 iface->xf_ifidx = -1;
3667 iface->tag = tag_create_random();
3668 iface->delay_expires = LLONG_MAX;
3669 iface->netdev = NULL;
3670 iface->cfg = if_cfg;
3672 shash_add_assert(&br->iface_by_name, iface->name, iface);
3674 /* Attempt to create the network interface in case it doesn't exist yet. */
3675 if (!iface_is_internal(br, iface->name)) {
3676 error = set_up_iface(if_cfg, iface, true);
3678 VLOG_WARN("could not create iface %s: %s", iface->name,
3681 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3688 if (port->n_ifaces >= port->allocated_ifaces) {
3689 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
3690 sizeof *port->ifaces);
3692 port->ifaces[port->n_ifaces++] = iface;
3693 if (port->n_ifaces > 1) {
3694 br->has_bonded_ports = true;
3697 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
3705 iface_destroy(struct iface *iface)
3708 struct port *port = iface->port;
3709 struct bridge *br = port->bridge;
3710 bool del_active = port->active_iface == iface->port_ifidx;
3713 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
3715 if (iface->xf_ifidx >= 0) {
3716 hmap_remove(&br->ifaces, &iface->xf_ifidx_node);
3719 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
3720 del->port_ifidx = iface->port_ifidx;
3722 netdev_close(iface->netdev);
3725 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
3726 bond_choose_active_iface(port);
3727 bond_send_learning_packets(port);
3733 bridge_flush(port->bridge);
3737 static struct iface *
3738 iface_lookup(const struct bridge *br, const char *name)
3740 return shash_find_data(&br->iface_by_name, name);
3743 static struct iface *
3744 iface_from_xf_ifidx(const struct bridge *br, uint16_t xf_ifidx)
3746 struct iface *iface;
3748 HMAP_FOR_EACH_IN_BUCKET (iface, xf_ifidx_node,
3749 hash_int(xf_ifidx, 0), &br->ifaces) {
3750 if (iface->xf_ifidx == xf_ifidx) {
3757 /* Returns true if 'iface' is the name of an "internal" interface on bridge
3758 * 'br', that is, an interface that is entirely simulated within the datapath.
3759 * The local port (XFLOWP_LOCAL) is always an internal interface. Other local
3760 * interfaces are created by setting "iface.<iface>.internal = true".
3762 * In addition, we have a kluge-y feature that creates an internal port with
3763 * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set.
3764 * This feature needs to go away in the long term. Until then, this is one
3765 * reason why this function takes a name instead of a struct iface: the fake
3766 * interfaces created this way do not have a struct iface. */
3768 iface_is_internal(const struct bridge *br, const char *if_name)
3770 struct iface *iface;
3773 if (!strcmp(if_name, br->name)) {
3777 iface = iface_lookup(br, if_name);
3778 if (iface && !strcmp(iface->cfg->type, "internal")) {
3782 port = port_lookup(br, if_name);
3783 if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
3789 /* Set Ethernet address of 'iface', if one is specified in the configuration
3792 iface_set_mac(struct iface *iface)
3794 uint8_t ea[ETH_ADDR_LEN];
3796 if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
3797 if (eth_addr_is_multicast(ea)) {
3798 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3800 } else if (iface->xf_ifidx == XFLOWP_LOCAL) {
3801 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3802 iface->name, iface->name);
3804 int error = netdev_set_etheraddr(iface->netdev, ea);
3806 VLOG_ERR("interface %s: setting MAC failed (%s)",
3807 iface->name, strerror(error));
3814 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
3815 struct shash *shash)
3820 for (i = 0; i < n; i++) {
3821 shash_add(shash, keys[i], values[i]);
3825 struct iface_delete_queues_cbdata {
3826 struct netdev *netdev;
3827 const struct ovsdb_datum *queues;
3831 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
3833 union ovsdb_atom atom;
3835 atom.integer = target;
3836 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
3840 iface_delete_queues(unsigned int queue_id,
3841 const struct shash *details OVS_UNUSED, void *cbdata_)
3843 struct iface_delete_queues_cbdata *cbdata = cbdata_;
3845 if (!queue_ids_include(cbdata->queues, queue_id)) {
3846 netdev_delete_queue(cbdata->netdev, queue_id);
3851 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
3853 if (!qos || qos->type[0] == '\0') {
3854 netdev_set_qos(iface->netdev, NULL, NULL);
3856 struct iface_delete_queues_cbdata cbdata;
3857 struct shash details;
3860 /* Configure top-level Qos for 'iface'. */
3861 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
3862 qos->n_other_config, &details);
3863 netdev_set_qos(iface->netdev, qos->type, &details);
3864 shash_destroy(&details);
3866 /* Deconfigure queues that were deleted. */
3867 cbdata.netdev = iface->netdev;
3868 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
3870 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
3872 /* Configure queues for 'iface'. */
3873 for (i = 0; i < qos->n_queues; i++) {
3874 const struct ovsrec_queue *queue = qos->value_queues[i];
3875 unsigned int queue_id = qos->key_queues[i];
3877 shash_from_ovs_idl_map(queue->key_other_config,
3878 queue->value_other_config,
3879 queue->n_other_config, &details);
3880 netdev_set_queue(iface->netdev, queue_id, &details);
3881 shash_destroy(&details);
3886 /* Port mirroring. */
3888 static struct mirror *
3889 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
3893 for (i = 0; i < MAX_MIRRORS; i++) {
3894 struct mirror *m = br->mirrors[i];
3895 if (m && uuid_equals(uuid, &m->uuid)) {
3903 mirror_reconfigure(struct bridge *br)
3905 unsigned long *rspan_vlans;
3908 /* Get rid of deleted mirrors. */
3909 for (i = 0; i < MAX_MIRRORS; i++) {
3910 struct mirror *m = br->mirrors[i];
3912 const struct ovsdb_datum *mc;
3913 union ovsdb_atom atom;
3915 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
3916 atom.uuid = br->mirrors[i]->uuid;
3917 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
3923 /* Add new mirrors and reconfigure existing ones. */
3924 for (i = 0; i < br->cfg->n_mirrors; i++) {
3925 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
3926 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
3928 mirror_reconfigure_one(m, cfg);
3930 mirror_create(br, cfg);
3934 /* Update port reserved status. */
3935 for (i = 0; i < br->n_ports; i++) {
3936 br->ports[i]->is_mirror_output_port = false;
3938 for (i = 0; i < MAX_MIRRORS; i++) {
3939 struct mirror *m = br->mirrors[i];
3940 if (m && m->out_port) {
3941 m->out_port->is_mirror_output_port = true;
3945 /* Update flooded vlans (for RSPAN). */
3947 if (br->cfg->n_flood_vlans) {
3948 rspan_vlans = bitmap_allocate(4096);
3950 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
3951 int64_t vlan = br->cfg->flood_vlans[i];
3952 if (vlan >= 0 && vlan < 4096) {
3953 bitmap_set1(rspan_vlans, vlan);
3954 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
3957 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
3962 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
3968 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
3973 for (i = 0; ; i++) {
3974 if (i >= MAX_MIRRORS) {
3975 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3976 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
3979 if (!br->mirrors[i]) {
3984 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
3987 br->mirrors[i] = m = xzalloc(sizeof *m);
3990 m->name = xstrdup(cfg->name);
3991 shash_init(&m->src_ports);
3992 shash_init(&m->dst_ports);
3998 mirror_reconfigure_one(m, cfg);
4002 mirror_destroy(struct mirror *m)
4005 struct bridge *br = m->bridge;
4008 for (i = 0; i < br->n_ports; i++) {
4009 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4010 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4013 shash_destroy(&m->src_ports);
4014 shash_destroy(&m->dst_ports);
4017 m->bridge->mirrors[m->idx] = NULL;
4026 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4027 struct shash *names)
4031 for (i = 0; i < n_ports; i++) {
4032 const char *name = ports[i]->name;
4033 if (port_lookup(m->bridge, name)) {
4034 shash_add_once(names, name, NULL);
4036 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4037 "port %s", m->bridge->name, m->name, name);
4043 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4049 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4051 for (i = 0; i < cfg->n_select_vlan; i++) {
4052 int64_t vlan = cfg->select_vlan[i];
4053 if (vlan < 0 || vlan > 4095) {
4054 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4055 m->bridge->name, m->name, vlan);
4057 (*vlans)[n_vlans++] = vlan;
4064 vlan_is_mirrored(const struct mirror *m, int vlan)
4068 for (i = 0; i < m->n_vlans; i++) {
4069 if (m->vlans[i] == vlan) {
4077 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4081 for (i = 0; i < m->n_vlans; i++) {
4082 if (port_trunks_vlan(p, m->vlans[i])) {
4090 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4092 struct shash src_ports, dst_ports;
4093 mirror_mask_t mirror_bit;
4094 struct port *out_port;
4101 if (strcmp(cfg->name, m->name)) {
4103 m->name = xstrdup(cfg->name);
4106 /* Get output port. */
4107 if (cfg->output_port) {
4108 out_port = port_lookup(m->bridge, cfg->output_port->name);
4110 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4111 m->bridge->name, m->name);
4117 if (cfg->output_vlan) {
4118 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4119 "output vlan; ignoring output vlan",
4120 m->bridge->name, m->name);
4122 } else if (cfg->output_vlan) {
4124 out_vlan = *cfg->output_vlan;
4126 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4127 m->bridge->name, m->name);
4132 shash_init(&src_ports);
4133 shash_init(&dst_ports);
4134 if (cfg->select_all) {
4135 for (i = 0; i < m->bridge->n_ports; i++) {
4136 const char *name = m->bridge->ports[i]->name;
4137 shash_add_once(&src_ports, name, NULL);
4138 shash_add_once(&dst_ports, name, NULL);
4143 /* Get ports, and drop duplicates and ports that don't exist. */
4144 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4146 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4149 /* Get all the vlans, and drop duplicate and invalid vlans. */
4150 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4153 /* Update mirror data. */
4154 if (!shash_equal_keys(&m->src_ports, &src_ports)
4155 || !shash_equal_keys(&m->dst_ports, &dst_ports)
4156 || m->n_vlans != n_vlans
4157 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4158 || m->out_port != out_port
4159 || m->out_vlan != out_vlan) {
4160 bridge_flush(m->bridge);
4162 shash_swap(&m->src_ports, &src_ports);
4163 shash_swap(&m->dst_ports, &dst_ports);
4166 m->n_vlans = n_vlans;
4167 m->out_port = out_port;
4168 m->out_vlan = out_vlan;
4171 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4172 for (i = 0; i < m->bridge->n_ports; i++) {
4173 struct port *port = m->bridge->ports[i];
4175 if (shash_find(&m->src_ports, port->name)
4178 ? port_trunks_any_mirrored_vlan(m, port)
4179 : vlan_is_mirrored(m, port->vlan)))) {
4180 port->src_mirrors |= mirror_bit;
4182 port->src_mirrors &= ~mirror_bit;
4185 if (shash_find(&m->dst_ports, port->name)) {
4186 port->dst_mirrors |= mirror_bit;
4188 port->dst_mirrors &= ~mirror_bit;
4193 shash_destroy(&src_ports);
4194 shash_destroy(&dst_ports);