2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
26 #include "byte-order.h"
27 #include "classifier.h"
30 #include "dynamic-string.h"
33 #include "meta-flow.h"
36 #include "ofp-actions.h"
37 #include "ofp-errors.h"
39 #include "ofp-print.h"
42 #include "ofproto-provider.h"
43 #include "openflow/nicira-ext.h"
44 #include "openflow/openflow.h"
48 #include "poll-loop.h"
54 #include "unaligned.h"
58 VLOG_DEFINE_THIS_MODULE(ofproto
);
60 COVERAGE_DEFINE(ofproto_error
);
61 COVERAGE_DEFINE(ofproto_flush
);
62 COVERAGE_DEFINE(ofproto_no_packet_in
);
63 COVERAGE_DEFINE(ofproto_packet_out
);
64 COVERAGE_DEFINE(ofproto_queue_req
);
65 COVERAGE_DEFINE(ofproto_recv_openflow
);
66 COVERAGE_DEFINE(ofproto_reinit_ports
);
67 COVERAGE_DEFINE(ofproto_uninstallable
);
68 COVERAGE_DEFINE(ofproto_update_port
);
71 S_OPENFLOW
, /* Processing OpenFlow commands. */
72 S_EVICT
, /* Evicting flows from over-limit tables. */
73 S_FLUSH
, /* Deleting all flow table rules. */
76 enum ofoperation_type
{
83 /* A single OpenFlow request can execute any number of operations. The
84 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
85 * ofconn to which an error reply should be sent if necessary.
87 * ofproto initiates some operations internally. These operations are still
88 * assigned to groups but will not have an associated ofconn. */
90 struct ofproto
*ofproto
; /* Owning ofproto. */
91 struct list ofproto_node
; /* In ofproto's "pending" list. */
92 struct list ops
; /* List of "struct ofoperation"s. */
93 int n_running
; /* Number of ops still pending. */
95 /* Data needed to send OpenFlow reply on failure or to send a buffered
98 * If list_is_empty(ofconn_node) then this ofopgroup never had an
99 * associated ofconn or its ofconn's connection dropped after it initiated
100 * the operation. In the latter case 'ofconn' is a wild pointer that
101 * refers to freed memory, so the 'ofconn' member must be used only if
102 * !list_is_empty(ofconn_node).
104 struct list ofconn_node
; /* In ofconn's list of pending opgroups. */
105 struct ofconn
*ofconn
; /* ofconn for reply (but see note above). */
106 struct ofp_header
*request
; /* Original request (truncated at 64 bytes). */
107 uint32_t buffer_id
; /* Buffer id from original request. */
110 static struct ofopgroup
*ofopgroup_create_unattached(struct ofproto
*);
111 static struct ofopgroup
*ofopgroup_create(struct ofproto
*, struct ofconn
*,
112 const struct ofp_header
*,
114 static void ofopgroup_submit(struct ofopgroup
*);
115 static void ofopgroup_complete(struct ofopgroup
*);
117 /* A single flow table operation. */
119 struct ofopgroup
*group
; /* Owning group. */
120 struct list group_node
; /* In ofopgroup's "ops" list. */
121 struct hmap_node hmap_node
; /* In ofproto's "deletions" hmap. */
122 struct rule
*rule
; /* Rule being operated upon. */
123 enum ofoperation_type type
; /* Type of operation. */
125 /* OFOPERATION_MODIFY, OFOPERATION_REPLACE: The old actions, if the actions
127 struct ofpact
*ofpacts
;
131 /* OFOPERATION_DELETE. */
132 enum ofp_flow_removed_reason reason
; /* Reason flow was removed. */
134 ovs_be64 flow_cookie
; /* Rule's old flow cookie. */
135 uint16_t idle_timeout
; /* Rule's old idle timeout. */
136 uint16_t hard_timeout
; /* Rule's old hard timeout. */
137 bool send_flow_removed
; /* Rule's old 'send_flow_removed'. */
138 enum ofperr error
; /* 0 if no error. */
141 static struct ofoperation
*ofoperation_create(struct ofopgroup
*,
143 enum ofoperation_type
,
144 enum ofp_flow_removed_reason
);
145 static void ofoperation_destroy(struct ofoperation
*);
148 static void oftable_init(struct oftable
*);
149 static void oftable_destroy(struct oftable
*);
151 static void oftable_set_name(struct oftable
*, const char *name
);
153 static void oftable_disable_eviction(struct oftable
*);
154 static void oftable_enable_eviction(struct oftable
*,
155 const struct mf_subfield
*fields
,
158 static void oftable_remove_rule(struct rule
*rule
) OVS_RELEASES(rule
->evict
);
159 static void oftable_remove_rule__(struct ofproto
*ofproto
,
160 struct classifier
*cls
, struct rule
*rule
)
161 OVS_REQ_WRLOCK(cls
->rwlock
) OVS_RELEASES(rule
->evict
);
162 static void oftable_insert_rule(struct rule
*);
164 /* A set of rules within a single OpenFlow table (oftable) that have the same
165 * values for the oftable's eviction_fields. A rule to be evicted, when one is
166 * needed, is taken from the eviction group that contains the greatest number
169 * An oftable owns any number of eviction groups, each of which contains any
172 * Membership in an eviction group is imprecise, based on the hash of the
173 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
174 * That is, if two rules have different eviction_fields, but those
175 * eviction_fields hash to the same value, then they will belong to the same
176 * eviction_group anyway.
178 * (When eviction is not enabled on an oftable, we don't track any eviction
179 * groups, to save time and space.) */
180 struct eviction_group
{
181 struct hmap_node id_node
; /* In oftable's "eviction_groups_by_id". */
182 struct heap_node size_node
; /* In oftable's "eviction_groups_by_size". */
183 struct heap rules
; /* Contains "struct rule"s. */
186 static bool choose_rule_to_evict(struct oftable
*table
, struct rule
**rulep
)
187 OVS_TRY_WRLOCK(true, (*rulep
)->evict
);
188 static void ofproto_evict(struct ofproto
*);
189 static uint32_t rule_eviction_priority(struct rule
*);
190 static void eviction_group_add_rule(struct rule
*);
191 static void eviction_group_remove_rule(struct rule
*);
194 static void ofport_destroy__(struct ofport
*);
195 static void ofport_destroy(struct ofport
*);
197 static void update_port(struct ofproto
*, const char *devname
);
198 static int init_ports(struct ofproto
*);
199 static void reinit_ports(struct ofproto
*);
202 static void ofproto_rule_destroy__(struct rule
*);
203 static void ofproto_rule_send_removed(struct rule
*, uint8_t reason
);
204 static bool rule_is_modifiable(const struct rule
*);
207 static enum ofperr
add_flow(struct ofproto
*, struct ofconn
*,
208 struct ofputil_flow_mod
*,
209 const struct ofp_header
*);
210 static enum ofperr
modify_flows__(struct ofproto
*, struct ofconn
*,
211 struct ofputil_flow_mod
*,
212 const struct ofp_header
*, struct list
*);
213 static void delete_flow__(struct rule
*rule
, struct ofopgroup
*,
214 enum ofp_flow_removed_reason
)
215 OVS_RELEASES(rule
->evict
);
216 static enum ofperr
add_group(struct ofproto
*, struct ofputil_group_mod
*);
217 static bool handle_openflow(struct ofconn
*, const struct ofpbuf
*);
218 static enum ofperr
handle_flow_mod__(struct ofproto
*, struct ofconn
*,
219 struct ofputil_flow_mod
*,
220 const struct ofp_header
*);
221 static void calc_duration(long long int start
, long long int now
,
222 uint32_t *sec
, uint32_t *nsec
);
225 static uint64_t pick_datapath_id(const struct ofproto
*);
226 static uint64_t pick_fallback_dpid(void);
227 static void ofproto_destroy__(struct ofproto
*);
228 static void update_mtu(struct ofproto
*, struct ofport
*);
229 static void meter_delete(struct ofproto
*, uint32_t first
, uint32_t last
);
232 static void ofproto_unixctl_init(void);
234 /* All registered ofproto classes, in probe order. */
235 static const struct ofproto_class
**ofproto_classes
;
236 static size_t n_ofproto_classes
;
237 static size_t allocated_ofproto_classes
;
239 unsigned flow_eviction_threshold
= OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT
;
240 unsigned n_handler_threads
;
241 enum ofproto_flow_miss_model flow_miss_model
= OFPROTO_HANDLE_MISS_AUTO
;
243 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
244 static struct hmap all_ofprotos
= HMAP_INITIALIZER(&all_ofprotos
);
246 /* Initial mappings of port to OpenFlow number mappings. */
247 static struct shash init_ofp_ports
= SHASH_INITIALIZER(&init_ofp_ports
);
249 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
251 /* The default value of true waits for flow restore. */
252 static bool flow_restore_wait
= true;
254 /* Must be called to initialize the ofproto library.
256 * The caller may pass in 'iface_hints', which contains an shash of
257 * "iface_hint" elements indexed by the interface's name. The provider
258 * may use these hints to describe the startup configuration in order to
259 * reinitialize its state. The caller owns the provided data, so a
260 * provider will make copies of anything required. An ofproto provider
261 * will remove any existing state that is not described by the hint, and
262 * may choose to remove it all. */
264 ofproto_init(const struct shash
*iface_hints
)
266 struct shash_node
*node
;
269 ofproto_class_register(&ofproto_dpif_class
);
271 /* Make a local copy, since we don't own 'iface_hints' elements. */
272 SHASH_FOR_EACH(node
, iface_hints
) {
273 const struct iface_hint
*orig_hint
= node
->data
;
274 struct iface_hint
*new_hint
= xmalloc(sizeof *new_hint
);
275 const char *br_type
= ofproto_normalize_type(orig_hint
->br_type
);
277 new_hint
->br_name
= xstrdup(orig_hint
->br_name
);
278 new_hint
->br_type
= xstrdup(br_type
);
279 new_hint
->ofp_port
= orig_hint
->ofp_port
;
281 shash_add(&init_ofp_ports
, node
->name
, new_hint
);
284 for (i
= 0; i
< n_ofproto_classes
; i
++) {
285 ofproto_classes
[i
]->init(&init_ofp_ports
);
289 /* 'type' should be a normalized datapath type, as returned by
290 * ofproto_normalize_type(). Returns the corresponding ofproto_class
291 * structure, or a null pointer if there is none registered for 'type'. */
292 static const struct ofproto_class
*
293 ofproto_class_find__(const char *type
)
297 for (i
= 0; i
< n_ofproto_classes
; i
++) {
298 const struct ofproto_class
*class = ofproto_classes
[i
];
303 class->enumerate_types(&types
);
304 found
= sset_contains(&types
, type
);
305 sset_destroy(&types
);
311 VLOG_WARN("unknown datapath type %s", type
);
315 /* Registers a new ofproto class. After successful registration, new ofprotos
316 * of that type can be created using ofproto_create(). */
318 ofproto_class_register(const struct ofproto_class
*new_class
)
322 for (i
= 0; i
< n_ofproto_classes
; i
++) {
323 if (ofproto_classes
[i
] == new_class
) {
328 if (n_ofproto_classes
>= allocated_ofproto_classes
) {
329 ofproto_classes
= x2nrealloc(ofproto_classes
,
330 &allocated_ofproto_classes
,
331 sizeof *ofproto_classes
);
333 ofproto_classes
[n_ofproto_classes
++] = new_class
;
337 /* Unregisters a datapath provider. 'type' must have been previously
338 * registered and not currently be in use by any ofprotos. After
339 * unregistration new datapaths of that type cannot be opened using
340 * ofproto_create(). */
342 ofproto_class_unregister(const struct ofproto_class
*class)
346 for (i
= 0; i
< n_ofproto_classes
; i
++) {
347 if (ofproto_classes
[i
] == class) {
348 for (i
++; i
< n_ofproto_classes
; i
++) {
349 ofproto_classes
[i
- 1] = ofproto_classes
[i
];
355 VLOG_WARN("attempted to unregister an ofproto class that is not "
360 /* Clears 'types' and enumerates all registered ofproto types into it. The
361 * caller must first initialize the sset. */
363 ofproto_enumerate_types(struct sset
*types
)
367 for (i
= 0; i
< n_ofproto_classes
; i
++) {
368 ofproto_classes
[i
]->enumerate_types(types
);
372 /* Returns the fully spelled out name for the given ofproto 'type'.
374 * Normalized type string can be compared with strcmp(). Unnormalized type
375 * string might be the same even if they have different spellings. */
377 ofproto_normalize_type(const char *type
)
379 return type
&& type
[0] ? type
: "system";
382 /* Clears 'names' and enumerates the names of all known created ofprotos with
383 * the given 'type'. The caller must first initialize the sset. Returns 0 if
384 * successful, otherwise a positive errno value.
386 * Some kinds of datapaths might not be practically enumerable. This is not
387 * considered an error. */
389 ofproto_enumerate_names(const char *type
, struct sset
*names
)
391 const struct ofproto_class
*class = ofproto_class_find__(type
);
392 return class ? class->enumerate_names(type
, names
) : EAFNOSUPPORT
;
396 ofproto_create(const char *datapath_name
, const char *datapath_type
,
397 struct ofproto
**ofprotop
)
399 const struct ofproto_class
*class;
400 struct ofproto
*ofproto
;
406 ofproto_unixctl_init();
408 datapath_type
= ofproto_normalize_type(datapath_type
);
409 class = ofproto_class_find__(datapath_type
);
411 VLOG_WARN("could not create datapath %s of unknown type %s",
412 datapath_name
, datapath_type
);
416 ofproto
= class->alloc();
418 VLOG_ERR("failed to allocate datapath %s of type %s",
419 datapath_name
, datapath_type
);
424 memset(ofproto
, 0, sizeof *ofproto
);
425 ofproto
->ofproto_class
= class;
426 ofproto
->name
= xstrdup(datapath_name
);
427 ofproto
->type
= xstrdup(datapath_type
);
428 hmap_insert(&all_ofprotos
, &ofproto
->hmap_node
,
429 hash_string(ofproto
->name
, 0));
430 ofproto
->datapath_id
= 0;
431 ofproto
->forward_bpdu
= false;
432 ofproto
->fallback_dpid
= pick_fallback_dpid();
433 ofproto
->mfr_desc
= NULL
;
434 ofproto
->hw_desc
= NULL
;
435 ofproto
->sw_desc
= NULL
;
436 ofproto
->serial_desc
= NULL
;
437 ofproto
->dp_desc
= NULL
;
438 ofproto
->frag_handling
= OFPC_FRAG_NORMAL
;
439 hmap_init(&ofproto
->ports
);
440 shash_init(&ofproto
->port_by_name
);
441 simap_init(&ofproto
->ofp_requests
);
442 ofproto
->max_ports
= ofp_to_u16(OFPP_MAX
);
443 ofproto
->tables
= NULL
;
444 ofproto
->n_tables
= 0;
445 hindex_init(&ofproto
->cookies
);
446 list_init(&ofproto
->expirable
);
447 ovs_mutex_init_recursive(&ofproto
->expirable_mutex
);
448 ofproto
->connmgr
= connmgr_create(ofproto
, datapath_name
, datapath_name
);
449 ofproto
->state
= S_OPENFLOW
;
450 list_init(&ofproto
->pending
);
451 ofproto
->n_pending
= 0;
452 hmap_init(&ofproto
->deletions
);
453 ofproto
->n_add
= ofproto
->n_delete
= ofproto
->n_modify
= 0;
454 ofproto
->first_op
= ofproto
->last_op
= LLONG_MIN
;
455 ofproto
->next_op_report
= LLONG_MAX
;
456 ofproto
->op_backoff
= LLONG_MIN
;
457 ofproto
->vlan_bitmap
= NULL
;
458 ofproto
->vlans_changed
= false;
459 ofproto
->min_mtu
= INT_MAX
;
460 ovs_rwlock_init(&ofproto
->groups_rwlock
);
461 hmap_init(&ofproto
->groups
);
463 error
= ofproto
->ofproto_class
->construct(ofproto
);
465 VLOG_ERR("failed to open datapath %s: %s",
466 datapath_name
, ovs_strerror(error
));
467 ofproto_destroy__(ofproto
);
471 /* The "max_ports" member should have been set by ->construct(ofproto).
472 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
473 ofproto
->ofp_port_ids
= bitmap_allocate(ofproto
->max_ports
);
474 bitmap_set1(ofproto
->ofp_port_ids
, 0);
476 /* Check that hidden tables, if any, are at the end. */
477 ovs_assert(ofproto
->n_tables
);
478 for (i
= 0; i
+ 1 < ofproto
->n_tables
; i
++) {
479 enum oftable_flags flags
= ofproto
->tables
[i
].flags
;
480 enum oftable_flags next_flags
= ofproto
->tables
[i
+ 1].flags
;
482 ovs_assert(!(flags
& OFTABLE_HIDDEN
) || next_flags
& OFTABLE_HIDDEN
);
485 ofproto
->datapath_id
= pick_datapath_id(ofproto
);
488 /* Initialize meters table. */
489 if (ofproto
->ofproto_class
->meter_get_features
) {
490 ofproto
->ofproto_class
->meter_get_features(ofproto
,
491 &ofproto
->meter_features
);
493 memset(&ofproto
->meter_features
, 0, sizeof ofproto
->meter_features
);
495 ofproto
->meters
= xzalloc((ofproto
->meter_features
.max_meters
+ 1)
496 * sizeof(struct meter
*));
502 /* Must be called (only) by an ofproto implementation in its constructor
503 * function. See the large comment on 'construct' in struct ofproto_class for
506 ofproto_init_tables(struct ofproto
*ofproto
, int n_tables
)
508 struct oftable
*table
;
510 ovs_assert(!ofproto
->n_tables
);
511 ovs_assert(n_tables
>= 1 && n_tables
<= 255);
513 ofproto
->n_tables
= n_tables
;
514 ofproto
->tables
= xmalloc(n_tables
* sizeof *ofproto
->tables
);
515 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
520 /* To be optionally called (only) by an ofproto implementation in its
521 * constructor function. See the large comment on 'construct' in struct
522 * ofproto_class for details.
524 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
525 * will then ensure that actions passed into the ofproto implementation will
526 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
527 * function is not called, there will be no such restriction.
529 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
530 * the 'max_ports' restriction. */
532 ofproto_init_max_ports(struct ofproto
*ofproto
, uint16_t max_ports
)
534 ovs_assert(max_ports
<= ofp_to_u16(OFPP_MAX
));
535 ofproto
->max_ports
= max_ports
;
539 ofproto_get_datapath_id(const struct ofproto
*ofproto
)
541 return ofproto
->datapath_id
;
545 ofproto_set_datapath_id(struct ofproto
*p
, uint64_t datapath_id
)
547 uint64_t old_dpid
= p
->datapath_id
;
548 p
->datapath_id
= datapath_id
? datapath_id
: pick_datapath_id(p
);
549 if (p
->datapath_id
!= old_dpid
) {
550 /* Force all active connections to reconnect, since there is no way to
551 * notify a controller that the datapath ID has changed. */
552 ofproto_reconnect_controllers(p
);
557 ofproto_set_controllers(struct ofproto
*p
,
558 const struct ofproto_controller
*controllers
,
559 size_t n_controllers
, uint32_t allowed_versions
)
561 connmgr_set_controllers(p
->connmgr
, controllers
, n_controllers
,
566 ofproto_set_fail_mode(struct ofproto
*p
, enum ofproto_fail_mode fail_mode
)
568 connmgr_set_fail_mode(p
->connmgr
, fail_mode
);
571 /* Drops the connections between 'ofproto' and all of its controllers, forcing
572 * them to reconnect. */
574 ofproto_reconnect_controllers(struct ofproto
*ofproto
)
576 connmgr_reconnect(ofproto
->connmgr
);
579 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
580 * in-band control should guarantee access, in the same way that in-band
581 * control guarantees access to OpenFlow controllers. */
583 ofproto_set_extra_in_band_remotes(struct ofproto
*ofproto
,
584 const struct sockaddr_in
*extras
, size_t n
)
586 connmgr_set_extra_in_band_remotes(ofproto
->connmgr
, extras
, n
);
589 /* Sets the OpenFlow queue used by flows set up by in-band control on
590 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
591 * flows will use the default queue. */
593 ofproto_set_in_band_queue(struct ofproto
*ofproto
, int queue_id
)
595 connmgr_set_in_band_queue(ofproto
->connmgr
, queue_id
);
598 /* Sets the number of flows at which eviction from the kernel flow table
601 ofproto_set_flow_eviction_threshold(unsigned threshold
)
603 flow_eviction_threshold
= MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN
,
607 /* Sets the path for handling flow misses. */
609 ofproto_set_flow_miss_model(unsigned model
)
611 flow_miss_model
= model
;
614 /* If forward_bpdu is true, the NORMAL action will forward frames with
615 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
616 * the NORMAL action will drop these frames. */
618 ofproto_set_forward_bpdu(struct ofproto
*ofproto
, bool forward_bpdu
)
620 bool old_val
= ofproto
->forward_bpdu
;
621 ofproto
->forward_bpdu
= forward_bpdu
;
622 if (old_val
!= ofproto
->forward_bpdu
) {
623 if (ofproto
->ofproto_class
->forward_bpdu_changed
) {
624 ofproto
->ofproto_class
->forward_bpdu_changed(ofproto
);
629 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
630 * 'idle_time', in seconds, and the maximum number of MAC table entries to
633 ofproto_set_mac_table_config(struct ofproto
*ofproto
, unsigned idle_time
,
636 if (ofproto
->ofproto_class
->set_mac_table_config
) {
637 ofproto
->ofproto_class
->set_mac_table_config(ofproto
, idle_time
,
642 /* Sets number of upcall handler threads. The default is
643 * (number of online cores - 1). */
645 ofproto_set_n_handler_threads(unsigned limit
)
648 n_handler_threads
= limit
;
650 n_handler_threads
= MAX(1, sysconf(_SC_NPROCESSORS_ONLN
) - 1);
655 ofproto_set_dp_desc(struct ofproto
*p
, const char *dp_desc
)
658 p
->dp_desc
= dp_desc
? xstrdup(dp_desc
) : NULL
;
662 ofproto_set_snoops(struct ofproto
*ofproto
, const struct sset
*snoops
)
664 return connmgr_set_snoops(ofproto
->connmgr
, snoops
);
668 ofproto_set_netflow(struct ofproto
*ofproto
,
669 const struct netflow_options
*nf_options
)
671 if (nf_options
&& sset_is_empty(&nf_options
->collectors
)) {
675 if (ofproto
->ofproto_class
->set_netflow
) {
676 return ofproto
->ofproto_class
->set_netflow(ofproto
, nf_options
);
678 return nf_options
? EOPNOTSUPP
: 0;
683 ofproto_set_sflow(struct ofproto
*ofproto
,
684 const struct ofproto_sflow_options
*oso
)
686 if (oso
&& sset_is_empty(&oso
->targets
)) {
690 if (ofproto
->ofproto_class
->set_sflow
) {
691 return ofproto
->ofproto_class
->set_sflow(ofproto
, oso
);
693 return oso
? EOPNOTSUPP
: 0;
698 ofproto_set_ipfix(struct ofproto
*ofproto
,
699 const struct ofproto_ipfix_bridge_exporter_options
*bo
,
700 const struct ofproto_ipfix_flow_exporter_options
*fo
,
703 if (ofproto
->ofproto_class
->set_ipfix
) {
704 return ofproto
->ofproto_class
->set_ipfix(ofproto
, bo
, fo
, n_fo
);
706 return (bo
|| fo
) ? EOPNOTSUPP
: 0;
711 ofproto_set_flow_restore_wait(bool flow_restore_wait_db
)
713 flow_restore_wait
= flow_restore_wait_db
;
717 ofproto_get_flow_restore_wait(void)
719 return flow_restore_wait
;
723 /* Spanning Tree Protocol (STP) configuration. */
725 /* Configures STP on 'ofproto' using the settings defined in 's'. If
726 * 's' is NULL, disables STP.
728 * Returns 0 if successful, otherwise a positive errno value. */
730 ofproto_set_stp(struct ofproto
*ofproto
,
731 const struct ofproto_stp_settings
*s
)
733 return (ofproto
->ofproto_class
->set_stp
734 ? ofproto
->ofproto_class
->set_stp(ofproto
, s
)
738 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
739 * 'enabled' member of 's' is false, then the other members are not
742 * Returns 0 if successful, otherwise a positive errno value. */
744 ofproto_get_stp_status(struct ofproto
*ofproto
,
745 struct ofproto_stp_status
*s
)
747 return (ofproto
->ofproto_class
->get_stp_status
748 ? ofproto
->ofproto_class
->get_stp_status(ofproto
, s
)
752 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
753 * in 's'. The caller is responsible for assigning STP port numbers
754 * (using the 'port_num' member in the range of 1 through 255, inclusive)
755 * and ensuring there are no duplicates. If the 's' is NULL, then STP
756 * is disabled on the port.
758 * Returns 0 if successful, otherwise a positive errno value.*/
760 ofproto_port_set_stp(struct ofproto
*ofproto
, ofp_port_t ofp_port
,
761 const struct ofproto_port_stp_settings
*s
)
763 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
765 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16
,
766 ofproto
->name
, ofp_port
);
770 return (ofproto
->ofproto_class
->set_stp_port
771 ? ofproto
->ofproto_class
->set_stp_port(ofport
, s
)
775 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
776 * 's'. If the 'enabled' member in 's' is false, then the other members
777 * are not meaningful.
779 * Returns 0 if successful, otherwise a positive errno value.*/
781 ofproto_port_get_stp_status(struct ofproto
*ofproto
, ofp_port_t ofp_port
,
782 struct ofproto_port_stp_status
*s
)
784 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
786 VLOG_WARN_RL(&rl
, "%s: cannot get STP status on nonexistent "
787 "port %"PRIu16
, ofproto
->name
, ofp_port
);
791 return (ofproto
->ofproto_class
->get_stp_port_status
792 ? ofproto
->ofproto_class
->get_stp_port_status(ofport
, s
)
796 /* Queue DSCP configuration. */
798 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
799 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
800 * to implement QoS. Instead, it is used to implement features which require
801 * knowledge of what queues exist on a port, and some basic information about
804 * Returns 0 if successful, otherwise a positive errno value. */
806 ofproto_port_set_queues(struct ofproto
*ofproto
, ofp_port_t ofp_port
,
807 const struct ofproto_port_queue
*queues
,
810 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
813 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16
,
814 ofproto
->name
, ofp_port
);
818 return (ofproto
->ofproto_class
->set_queues
819 ? ofproto
->ofproto_class
->set_queues(ofport
, queues
, n_queues
)
823 /* Connectivity Fault Management configuration. */
825 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
827 ofproto_port_clear_cfm(struct ofproto
*ofproto
, ofp_port_t ofp_port
)
829 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
830 if (ofport
&& ofproto
->ofproto_class
->set_cfm
) {
831 ofproto
->ofproto_class
->set_cfm(ofport
, NULL
);
835 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
836 * basic configuration from the configuration members in 'cfm', and the remote
837 * maintenance point ID from remote_mpid. Ignores the statistics members of
840 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
842 ofproto_port_set_cfm(struct ofproto
*ofproto
, ofp_port_t ofp_port
,
843 const struct cfm_settings
*s
)
845 struct ofport
*ofport
;
848 ofport
= ofproto_get_port(ofproto
, ofp_port
);
850 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16
,
851 ofproto
->name
, ofp_port
);
855 /* XXX: For configuration simplicity, we only support one remote_mpid
856 * outside of the CFM module. It's not clear if this is the correct long
857 * term solution or not. */
858 error
= (ofproto
->ofproto_class
->set_cfm
859 ? ofproto
->ofproto_class
->set_cfm(ofport
, s
)
862 VLOG_WARN("%s: CFM configuration on port %"PRIu16
" (%s) failed (%s)",
863 ofproto
->name
, ofp_port
, netdev_get_name(ofport
->netdev
),
864 ovs_strerror(error
));
868 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
869 * 'ofproto' does not have a port 'ofp_port'. */
871 ofproto_port_set_bfd(struct ofproto
*ofproto
, ofp_port_t ofp_port
,
872 const struct smap
*cfg
)
874 struct ofport
*ofport
;
877 ofport
= ofproto_get_port(ofproto
, ofp_port
);
879 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16
,
880 ofproto
->name
, ofp_port
);
884 error
= (ofproto
->ofproto_class
->set_bfd
885 ? ofproto
->ofproto_class
->set_bfd(ofport
, cfg
)
888 VLOG_WARN("%s: bfd configuration on port %"PRIu16
" (%s) failed (%s)",
889 ofproto
->name
, ofp_port
, netdev_get_name(ofport
->netdev
),
890 ovs_strerror(error
));
894 /* Populates 'status' with key value pairs indicating the status of the BFD
895 * session on 'ofp_port'. This information is intended to be populated in the
896 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
899 ofproto_port_get_bfd_status(struct ofproto
*ofproto
, ofp_port_t ofp_port
,
902 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
903 return (ofport
&& ofproto
->ofproto_class
->get_bfd_status
904 ? ofproto
->ofproto_class
->get_bfd_status(ofport
, status
)
908 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
909 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
910 * 0 if LACP partner information is not current (generally indicating a
911 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
913 ofproto_port_is_lacp_current(struct ofproto
*ofproto
, ofp_port_t ofp_port
)
915 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
916 return (ofport
&& ofproto
->ofproto_class
->port_is_lacp_current
917 ? ofproto
->ofproto_class
->port_is_lacp_current(ofport
)
923 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
924 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
925 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
926 * configuration plus, if there is more than one slave, a bonding
929 * If 'aux' is already registered then this function updates its configuration
930 * to 's'. Otherwise, this function registers a new bundle.
932 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
935 ofproto_bundle_register(struct ofproto
*ofproto
, void *aux
,
936 const struct ofproto_bundle_settings
*s
)
938 return (ofproto
->ofproto_class
->bundle_set
939 ? ofproto
->ofproto_class
->bundle_set(ofproto
, aux
, s
)
943 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
944 * If no such bundle has been registered, this has no effect. */
946 ofproto_bundle_unregister(struct ofproto
*ofproto
, void *aux
)
948 return ofproto_bundle_register(ofproto
, aux
, NULL
);
952 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
953 * If 'aux' is already registered then this function updates its configuration
954 * to 's'. Otherwise, this function registers a new mirror. */
956 ofproto_mirror_register(struct ofproto
*ofproto
, void *aux
,
957 const struct ofproto_mirror_settings
*s
)
959 return (ofproto
->ofproto_class
->mirror_set
960 ? ofproto
->ofproto_class
->mirror_set(ofproto
, aux
, s
)
964 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
965 * If no mirror has been registered, this has no effect. */
967 ofproto_mirror_unregister(struct ofproto
*ofproto
, void *aux
)
969 return ofproto_mirror_register(ofproto
, aux
, NULL
);
972 /* Retrieves statistics from mirror associated with client data pointer
973 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
974 * 'bytes', respectively. If a particular counters is not supported,
975 * the appropriate argument is set to UINT64_MAX. */
977 ofproto_mirror_get_stats(struct ofproto
*ofproto
, void *aux
,
978 uint64_t *packets
, uint64_t *bytes
)
980 if (!ofproto
->ofproto_class
->mirror_get_stats
) {
981 *packets
= *bytes
= UINT64_MAX
;
985 return ofproto
->ofproto_class
->mirror_get_stats(ofproto
, aux
,
989 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
990 * which all packets are flooded, instead of using MAC learning. If
991 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
993 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
996 ofproto_set_flood_vlans(struct ofproto
*ofproto
, unsigned long *flood_vlans
)
998 return (ofproto
->ofproto_class
->set_flood_vlans
999 ? ofproto
->ofproto_class
->set_flood_vlans(ofproto
, flood_vlans
)
1003 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1004 * output for a mirror. */
1006 ofproto_is_mirror_output_bundle(const struct ofproto
*ofproto
, void *aux
)
1008 return (ofproto
->ofproto_class
->is_mirror_output_bundle
1009 ? ofproto
->ofproto_class
->is_mirror_output_bundle(ofproto
, aux
)
1013 /* Configuration of OpenFlow tables. */
1015 /* Returns the number of OpenFlow tables in 'ofproto'. */
1017 ofproto_get_n_tables(const struct ofproto
*ofproto
)
1019 return ofproto
->n_tables
;
1022 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1023 * settings from 's'. 'table_id' must be in the range 0 through the number of
1024 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1026 * For read-only tables, only the name may be configured. */
1028 ofproto_configure_table(struct ofproto
*ofproto
, int table_id
,
1029 const struct ofproto_table_settings
*s
)
1031 struct oftable
*table
;
1033 ovs_assert(table_id
>= 0 && table_id
< ofproto
->n_tables
);
1034 table
= &ofproto
->tables
[table_id
];
1036 oftable_set_name(table
, s
->name
);
1038 if (table
->flags
& OFTABLE_READONLY
) {
1043 oftable_enable_eviction(table
, s
->groups
, s
->n_groups
);
1045 oftable_disable_eviction(table
);
1048 table
->max_flows
= s
->max_flows
;
1049 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
1050 if (classifier_count(&table
->cls
) > table
->max_flows
1051 && table
->eviction_fields
) {
1052 /* 'table' contains more flows than allowed. We might not be able to
1053 * evict them right away because of the asynchronous nature of flow
1054 * table changes. Schedule eviction for later. */
1055 switch (ofproto
->state
) {
1057 ofproto
->state
= S_EVICT
;
1061 /* We're already deleting flows, nothing more to do. */
1065 ovs_rwlock_unlock(&table
->cls
.rwlock
);
1069 ofproto_has_snoops(const struct ofproto
*ofproto
)
1071 return connmgr_has_snoops(ofproto
->connmgr
);
1075 ofproto_get_snoops(const struct ofproto
*ofproto
, struct sset
*snoops
)
1077 connmgr_get_snoops(ofproto
->connmgr
, snoops
);
1080 /* Deletes 'rule' from 'cls' within 'ofproto'.
1082 * The 'cls' argument is redundant (it is &ofproto->tables[rule->table_id].cls)
1083 * but it allows Clang to do better checking. */
1085 ofproto_delete_rule(struct ofproto
*ofproto
, struct classifier
*cls
,
1087 OVS_REQ_WRLOCK(cls
->rwlock
)
1089 struct ofopgroup
*group
;
1091 ovs_assert(!rule
->pending
);
1092 ovs_assert(cls
== &ofproto
->tables
[rule
->table_id
].cls
);
1094 group
= ofopgroup_create_unattached(ofproto
);
1095 ofoperation_create(group
, rule
, OFOPERATION_DELETE
, OFPRR_DELETE
);
1096 ovs_rwlock_wrlock(&rule
->evict
);
1097 oftable_remove_rule__(ofproto
, cls
, rule
);
1098 ofproto
->ofproto_class
->rule_delete(rule
);
1099 ofopgroup_submit(group
);
1103 ofproto_flush__(struct ofproto
*ofproto
)
1105 struct oftable
*table
;
1107 if (ofproto
->ofproto_class
->flush
) {
1108 ofproto
->ofproto_class
->flush(ofproto
);
1111 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
1112 struct rule
*rule
, *next_rule
;
1113 struct cls_cursor cursor
;
1115 if (table
->flags
& OFTABLE_HIDDEN
) {
1119 ovs_rwlock_wrlock(&table
->cls
.rwlock
);
1120 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
1121 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
1122 if (!rule
->pending
) {
1123 ofproto_delete_rule(ofproto
, &table
->cls
, rule
);
1126 ovs_rwlock_unlock(&table
->cls
.rwlock
);
1130 static void delete_group(struct ofproto
*ofproto
, uint32_t group_id
);
1133 ofproto_destroy__(struct ofproto
*ofproto
)
1135 struct oftable
*table
;
1137 ovs_assert(list_is_empty(&ofproto
->pending
));
1138 ovs_assert(!ofproto
->n_pending
);
1140 if (ofproto
->meters
) {
1141 meter_delete(ofproto
, 1, ofproto
->meter_features
.max_meters
);
1142 free(ofproto
->meters
);
1145 delete_group(ofproto
, OFPG_ALL
);
1146 ovs_rwlock_destroy(&ofproto
->groups_rwlock
);
1147 hmap_destroy(&ofproto
->groups
);
1149 connmgr_destroy(ofproto
->connmgr
);
1151 hmap_remove(&all_ofprotos
, &ofproto
->hmap_node
);
1152 free(ofproto
->name
);
1153 free(ofproto
->type
);
1154 free(ofproto
->mfr_desc
);
1155 free(ofproto
->hw_desc
);
1156 free(ofproto
->sw_desc
);
1157 free(ofproto
->serial_desc
);
1158 free(ofproto
->dp_desc
);
1159 hmap_destroy(&ofproto
->ports
);
1160 shash_destroy(&ofproto
->port_by_name
);
1161 bitmap_free(ofproto
->ofp_port_ids
);
1162 simap_destroy(&ofproto
->ofp_requests
);
1164 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
1165 oftable_destroy(table
);
1167 free(ofproto
->tables
);
1169 hmap_destroy(&ofproto
->deletions
);
1171 free(ofproto
->vlan_bitmap
);
1173 ovs_mutex_destroy(&ofproto
->expirable_mutex
);
1174 ofproto
->ofproto_class
->dealloc(ofproto
);
1178 ofproto_destroy(struct ofproto
*p
)
1180 struct ofport
*ofport
, *next_ofport
;
1187 HMAP_FOR_EACH_SAFE (ofport
, next_ofport
, hmap_node
, &p
->ports
) {
1188 ofport_destroy(ofport
);
1191 p
->ofproto_class
->destruct(p
);
1192 ofproto_destroy__(p
);
1195 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1196 * kernel datapath, for example, this destroys the datapath in the kernel, and
1197 * with the netdev-based datapath, it tears down the data structures that
1198 * represent the datapath.
1200 * The datapath should not be currently open as an ofproto. */
1202 ofproto_delete(const char *name
, const char *type
)
1204 const struct ofproto_class
*class = ofproto_class_find__(type
);
1205 return (!class ? EAFNOSUPPORT
1206 : !class->del
? EACCES
1207 : class->del(type
, name
));
1211 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
1213 if (error
== ENOBUFS
) {
1214 reinit_ports(ofproto
);
1215 } else if (!error
) {
1216 update_port(ofproto
, devname
);
1222 ofproto_type_run(const char *datapath_type
)
1224 const struct ofproto_class
*class;
1227 datapath_type
= ofproto_normalize_type(datapath_type
);
1228 class = ofproto_class_find__(datapath_type
);
1230 error
= class->type_run
? class->type_run(datapath_type
) : 0;
1231 if (error
&& error
!= EAGAIN
) {
1232 VLOG_ERR_RL(&rl
, "%s: type_run failed (%s)",
1233 datapath_type
, ovs_strerror(error
));
1239 ofproto_type_run_fast(const char *datapath_type
)
1241 const struct ofproto_class
*class;
1244 datapath_type
= ofproto_normalize_type(datapath_type
);
1245 class = ofproto_class_find__(datapath_type
);
1247 error
= class->type_run_fast
? class->type_run_fast(datapath_type
) : 0;
1248 if (error
&& error
!= EAGAIN
) {
1249 VLOG_ERR_RL(&rl
, "%s: type_run_fast failed (%s)",
1250 datapath_type
, ovs_strerror(error
));
1256 ofproto_type_wait(const char *datapath_type
)
1258 const struct ofproto_class
*class;
1260 datapath_type
= ofproto_normalize_type(datapath_type
);
1261 class = ofproto_class_find__(datapath_type
);
1263 if (class->type_wait
) {
1264 class->type_wait(datapath_type
);
1269 ofproto_run(struct ofproto
*p
)
1271 struct sset changed_netdevs
;
1272 const char *changed_netdev
;
1273 struct ofport
*ofport
;
1276 error
= p
->ofproto_class
->run(p
);
1277 if (error
&& error
!= EAGAIN
) {
1278 VLOG_ERR_RL(&rl
, "%s: run failed (%s)", p
->name
, ovs_strerror(error
));
1281 if (p
->ofproto_class
->port_poll
) {
1284 while ((error
= p
->ofproto_class
->port_poll(p
, &devname
)) != EAGAIN
) {
1285 process_port_change(p
, error
, devname
);
1289 /* Update OpenFlow port status for any port whose netdev has changed.
1291 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1292 * destroyed, so it's not safe to update ports directly from the
1293 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1294 * need this two-phase approach. */
1295 sset_init(&changed_netdevs
);
1296 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1297 unsigned int change_seq
= netdev_change_seq(ofport
->netdev
);
1298 if (ofport
->change_seq
!= change_seq
) {
1299 ofport
->change_seq
= change_seq
;
1300 sset_add(&changed_netdevs
, netdev_get_name(ofport
->netdev
));
1303 SSET_FOR_EACH (changed_netdev
, &changed_netdevs
) {
1304 update_port(p
, changed_netdev
);
1306 sset_destroy(&changed_netdevs
);
1310 connmgr_run(p
->connmgr
, handle_openflow
);
1314 connmgr_run(p
->connmgr
, NULL
);
1316 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1317 p
->state
= S_OPENFLOW
;
1322 connmgr_run(p
->connmgr
, NULL
);
1324 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1325 connmgr_flushed(p
->connmgr
);
1326 p
->state
= S_OPENFLOW
;
1334 if (time_msec() >= p
->next_op_report
) {
1335 long long int ago
= (time_msec() - p
->first_op
) / 1000;
1336 long long int interval
= (p
->last_op
- p
->first_op
) / 1000;
1340 ds_put_format(&s
, "%d flow_mods ",
1341 p
->n_add
+ p
->n_delete
+ p
->n_modify
);
1342 if (interval
== ago
) {
1343 ds_put_format(&s
, "in the last %lld s", ago
);
1344 } else if (interval
) {
1345 ds_put_format(&s
, "in the %lld s starting %lld s ago",
1348 ds_put_format(&s
, "%lld s ago", ago
);
1351 ds_put_cstr(&s
, " (");
1353 ds_put_format(&s
, "%d adds, ", p
->n_add
);
1356 ds_put_format(&s
, "%d deletes, ", p
->n_delete
);
1359 ds_put_format(&s
, "%d modifications, ", p
->n_modify
);
1362 ds_put_char(&s
, ')');
1364 VLOG_INFO("%s: %s", p
->name
, ds_cstr(&s
));
1367 p
->n_add
= p
->n_delete
= p
->n_modify
= 0;
1368 p
->next_op_report
= LLONG_MAX
;
1374 /* Performs periodic activity required by 'ofproto' that needs to be done
1375 * with the least possible latency.
1377 * It makes sense to call this function a couple of times per poll loop, to
1378 * provide a significant performance boost on some benchmarks with the
1379 * ofproto-dpif implementation. */
1381 ofproto_run_fast(struct ofproto
*p
)
1385 error
= p
->ofproto_class
->run_fast
? p
->ofproto_class
->run_fast(p
) : 0;
1386 if (error
&& error
!= EAGAIN
) {
1387 VLOG_ERR_RL(&rl
, "%s: fastpath run failed (%s)",
1388 p
->name
, ovs_strerror(error
));
1394 ofproto_wait(struct ofproto
*p
)
1396 struct ofport
*ofport
;
1398 p
->ofproto_class
->wait(p
);
1399 if (p
->ofproto_class
->port_poll_wait
) {
1400 p
->ofproto_class
->port_poll_wait(p
);
1403 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1404 if (ofport
->change_seq
!= netdev_change_seq(ofport
->netdev
)) {
1405 poll_immediate_wake();
1411 connmgr_wait(p
->connmgr
, true);
1416 connmgr_wait(p
->connmgr
, false);
1417 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1418 poll_immediate_wake();
1425 ofproto_is_alive(const struct ofproto
*p
)
1427 return connmgr_has_controllers(p
->connmgr
);
1430 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1431 * memory_report(). */
1433 ofproto_get_memory_usage(const struct ofproto
*ofproto
, struct simap
*usage
)
1435 const struct oftable
*table
;
1436 unsigned int n_rules
;
1438 simap_increase(usage
, "ports", hmap_count(&ofproto
->ports
));
1439 simap_increase(usage
, "ops",
1440 ofproto
->n_pending
+ hmap_count(&ofproto
->deletions
));
1443 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
1444 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
1445 n_rules
+= classifier_count(&table
->cls
);
1446 ovs_rwlock_unlock(&table
->cls
.rwlock
);
1448 simap_increase(usage
, "rules", n_rules
);
1450 if (ofproto
->ofproto_class
->get_memory_usage
) {
1451 ofproto
->ofproto_class
->get_memory_usage(ofproto
, usage
);
1454 connmgr_get_memory_usage(ofproto
->connmgr
, usage
);
1458 ofproto_get_ofproto_controller_info(const struct ofproto
*ofproto
,
1461 connmgr_get_controller_info(ofproto
->connmgr
, info
);
1465 ofproto_free_ofproto_controller_info(struct shash
*info
)
1467 connmgr_free_controller_info(info
);
1470 /* Makes a deep copy of 'old' into 'port'. */
1472 ofproto_port_clone(struct ofproto_port
*port
, const struct ofproto_port
*old
)
1474 port
->name
= xstrdup(old
->name
);
1475 port
->type
= xstrdup(old
->type
);
1476 port
->ofp_port
= old
->ofp_port
;
1479 /* Frees memory allocated to members of 'ofproto_port'.
1481 * Do not call this function on an ofproto_port obtained from
1482 * ofproto_port_dump_next(): that function retains ownership of the data in the
1485 ofproto_port_destroy(struct ofproto_port
*ofproto_port
)
1487 free(ofproto_port
->name
);
1488 free(ofproto_port
->type
);
1491 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1493 * This function provides no status indication. An error status for the entire
1494 * dump operation is provided when it is completed by calling
1495 * ofproto_port_dump_done().
1498 ofproto_port_dump_start(struct ofproto_port_dump
*dump
,
1499 const struct ofproto
*ofproto
)
1501 dump
->ofproto
= ofproto
;
1502 dump
->error
= ofproto
->ofproto_class
->port_dump_start(ofproto
,
1506 /* Attempts to retrieve another port from 'dump', which must have been created
1507 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1508 * 'port' and returns true. On failure, returns false.
1510 * Failure might indicate an actual error or merely that the last port has been
1511 * dumped. An error status for the entire dump operation is provided when it
1512 * is completed by calling ofproto_port_dump_done().
1514 * The ofproto owns the data stored in 'port'. It will remain valid until at
1515 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1516 * ofproto_port_dump_done(). */
1518 ofproto_port_dump_next(struct ofproto_port_dump
*dump
,
1519 struct ofproto_port
*port
)
1521 const struct ofproto
*ofproto
= dump
->ofproto
;
1527 dump
->error
= ofproto
->ofproto_class
->port_dump_next(ofproto
, dump
->state
,
1530 ofproto
->ofproto_class
->port_dump_done(ofproto
, dump
->state
);
1536 /* Completes port table dump operation 'dump', which must have been created
1537 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1538 * error-free, otherwise a positive errno value describing the problem. */
1540 ofproto_port_dump_done(struct ofproto_port_dump
*dump
)
1542 const struct ofproto
*ofproto
= dump
->ofproto
;
1544 dump
->error
= ofproto
->ofproto_class
->port_dump_done(ofproto
,
1547 return dump
->error
== EOF
? 0 : dump
->error
;
1550 /* Returns the type to pass to netdev_open() when a datapath of type
1551 * 'datapath_type' has a port of type 'port_type', for a few special
1552 * cases when a netdev type differs from a port type. For example, when
1553 * using the userspace datapath, a port of type "internal" needs to be
1556 * Returns either 'type' itself or a string literal, which must not be
1559 ofproto_port_open_type(const char *datapath_type
, const char *port_type
)
1561 const struct ofproto_class
*class;
1563 datapath_type
= ofproto_normalize_type(datapath_type
);
1564 class = ofproto_class_find__(datapath_type
);
1569 return (class->port_open_type
1570 ? class->port_open_type(datapath_type
, port_type
)
1574 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1575 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1576 * the port's OpenFlow port number.
1578 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1579 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1580 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1581 * 'ofp_portp' is non-null). */
1583 ofproto_port_add(struct ofproto
*ofproto
, struct netdev
*netdev
,
1584 ofp_port_t
*ofp_portp
)
1586 ofp_port_t ofp_port
= ofp_portp
? *ofp_portp
: OFPP_NONE
;
1589 error
= ofproto
->ofproto_class
->port_add(ofproto
, netdev
);
1591 const char *netdev_name
= netdev_get_name(netdev
);
1593 simap_put(&ofproto
->ofp_requests
, netdev_name
,
1594 ofp_to_u16(ofp_port
));
1595 update_port(ofproto
, netdev_name
);
1598 struct ofproto_port ofproto_port
;
1600 ofproto_port_query_by_name(ofproto
, netdev_get_name(netdev
),
1602 *ofp_portp
= error
? OFPP_NONE
: ofproto_port
.ofp_port
;
1603 ofproto_port_destroy(&ofproto_port
);
1608 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1609 * initializes '*port' appropriately; on failure, returns a positive errno
1612 * The caller owns the data in 'ofproto_port' and must free it with
1613 * ofproto_port_destroy() when it is no longer needed. */
1615 ofproto_port_query_by_name(const struct ofproto
*ofproto
, const char *devname
,
1616 struct ofproto_port
*port
)
1620 error
= ofproto
->ofproto_class
->port_query_by_name(ofproto
, devname
, port
);
1622 memset(port
, 0, sizeof *port
);
1627 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1628 * Returns 0 if successful, otherwise a positive errno. */
1630 ofproto_port_del(struct ofproto
*ofproto
, ofp_port_t ofp_port
)
1632 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
1633 const char *name
= ofport
? netdev_get_name(ofport
->netdev
) : "<unknown>";
1634 struct simap_node
*ofp_request_node
;
1637 ofp_request_node
= simap_find(&ofproto
->ofp_requests
, name
);
1638 if (ofp_request_node
) {
1639 simap_delete(&ofproto
->ofp_requests
, ofp_request_node
);
1642 error
= ofproto
->ofproto_class
->port_del(ofproto
, ofp_port
);
1643 if (!error
&& ofport
) {
1644 /* 'name' is the netdev's name and update_port() is going to close the
1645 * netdev. Just in case update_port() refers to 'name' after it
1646 * destroys 'ofport', make a copy of it around the update_port()
1648 char *devname
= xstrdup(name
);
1649 update_port(ofproto
, devname
);
1655 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1656 * performs the 'n_actions' actions in 'actions'. The new flow will not
1659 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1660 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1661 * controllers; otherwise, it will be hidden.
1663 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1665 * This is a helper function for in-band control and fail-open. */
1667 ofproto_add_flow(struct ofproto
*ofproto
, const struct match
*match
,
1668 unsigned int priority
,
1669 const struct ofpact
*ofpacts
, size_t ofpacts_len
)
1671 const struct rule
*rule
;
1673 ovs_rwlock_rdlock(&ofproto
->tables
[0].cls
.rwlock
);
1674 rule
= rule_from_cls_rule(classifier_find_match_exactly(
1675 &ofproto
->tables
[0].cls
, match
, priority
));
1676 ovs_rwlock_unlock(&ofproto
->tables
[0].cls
.rwlock
);
1677 if (!rule
|| !ofpacts_equal(rule
->ofpacts
, rule
->ofpacts_len
,
1678 ofpacts
, ofpacts_len
)) {
1679 struct ofputil_flow_mod fm
;
1681 memset(&fm
, 0, sizeof fm
);
1683 fm
.priority
= priority
;
1684 fm
.buffer_id
= UINT32_MAX
;
1685 fm
.ofpacts
= xmemdup(ofpacts
, ofpacts_len
);
1686 fm
.ofpacts_len
= ofpacts_len
;
1687 add_flow(ofproto
, NULL
, &fm
, NULL
);
1692 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1693 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1694 * operation cannot be initiated now but may be retried later.
1696 * This is a helper function for in-band control and fail-open. */
1698 ofproto_flow_mod(struct ofproto
*ofproto
, struct ofputil_flow_mod
*fm
)
1700 return handle_flow_mod__(ofproto
, NULL
, fm
, NULL
);
1703 /* Searches for a rule with matching criteria exactly equal to 'target' in
1704 * ofproto's table 0 and, if it finds one, deletes it.
1706 * This is a helper function for in-band control and fail-open. */
1708 ofproto_delete_flow(struct ofproto
*ofproto
,
1709 const struct match
*target
, unsigned int priority
)
1711 struct classifier
*cls
= &ofproto
->tables
[0].cls
;
1714 ovs_rwlock_rdlock(&cls
->rwlock
);
1715 rule
= rule_from_cls_rule(classifier_find_match_exactly(cls
, target
,
1717 ovs_rwlock_unlock(&cls
->rwlock
);
1719 /* No such rule -> success. */
1721 } else if (rule
->pending
) {
1722 /* An operation on the rule is already pending -> failure.
1723 * Caller must retry later if it's important. */
1726 /* Initiate deletion -> success. */
1727 ovs_rwlock_wrlock(&cls
->rwlock
);
1728 ofproto_delete_rule(ofproto
, cls
, rule
);
1729 ovs_rwlock_unlock(&cls
->rwlock
);
1736 /* Starts the process of deleting all of the flows from all of ofproto's flow
1737 * tables and then reintroducing the flows required by in-band control and
1738 * fail-open. The process will complete in a later call to ofproto_run(). */
1740 ofproto_flush_flows(struct ofproto
*ofproto
)
1742 COVERAGE_INC(ofproto_flush
);
1743 ofproto
->state
= S_FLUSH
;
1747 reinit_ports(struct ofproto
*p
)
1749 struct ofproto_port_dump dump
;
1750 struct sset devnames
;
1751 struct ofport
*ofport
;
1752 struct ofproto_port ofproto_port
;
1753 const char *devname
;
1755 COVERAGE_INC(ofproto_reinit_ports
);
1757 sset_init(&devnames
);
1758 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1759 sset_add(&devnames
, netdev_get_name(ofport
->netdev
));
1761 OFPROTO_PORT_FOR_EACH (&ofproto_port
, &dump
, p
) {
1762 sset_add(&devnames
, ofproto_port
.name
);
1765 SSET_FOR_EACH (devname
, &devnames
) {
1766 update_port(p
, devname
);
1768 sset_destroy(&devnames
);
1772 alloc_ofp_port(struct ofproto
*ofproto
, const char *netdev_name
)
1776 port_idx
= simap_get(&ofproto
->ofp_requests
, netdev_name
);
1777 port_idx
= port_idx
? port_idx
: UINT16_MAX
;
1779 if (port_idx
>= ofproto
->max_ports
1780 || bitmap_is_set(ofproto
->ofp_port_ids
, port_idx
)) {
1781 uint16_t end_port_no
= ofproto
->alloc_port_no
;
1783 /* Search for a free OpenFlow port number. We try not to
1784 * immediately reuse them to prevent problems due to old
1787 if (++ofproto
->alloc_port_no
>= ofproto
->max_ports
) {
1788 ofproto
->alloc_port_no
= 0;
1790 if (!bitmap_is_set(ofproto
->ofp_port_ids
,
1791 ofproto
->alloc_port_no
)) {
1792 port_idx
= ofproto
->alloc_port_no
;
1795 if (ofproto
->alloc_port_no
== end_port_no
) {
1800 bitmap_set1(ofproto
->ofp_port_ids
, port_idx
);
1801 return u16_to_ofp(port_idx
);
1805 dealloc_ofp_port(const struct ofproto
*ofproto
, ofp_port_t ofp_port
)
1807 if (ofp_to_u16(ofp_port
) < ofproto
->max_ports
) {
1808 bitmap_set0(ofproto
->ofp_port_ids
, ofp_to_u16(ofp_port
));
1812 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1813 * pointer if the netdev cannot be opened. On success, also fills in
1815 static struct netdev
*
1816 ofport_open(struct ofproto
*ofproto
,
1817 struct ofproto_port
*ofproto_port
,
1818 struct ofputil_phy_port
*pp
)
1820 enum netdev_flags flags
;
1821 struct netdev
*netdev
;
1824 error
= netdev_open(ofproto_port
->name
, ofproto_port
->type
, &netdev
);
1826 VLOG_WARN_RL(&rl
, "%s: ignoring port %s (%"PRIu16
") because netdev %s "
1827 "cannot be opened (%s)",
1829 ofproto_port
->name
, ofproto_port
->ofp_port
,
1830 ofproto_port
->name
, ovs_strerror(error
));
1834 if (ofproto_port
->ofp_port
== OFPP_NONE
) {
1835 if (!strcmp(ofproto
->name
, ofproto_port
->name
)) {
1836 ofproto_port
->ofp_port
= OFPP_LOCAL
;
1838 ofproto_port
->ofp_port
= alloc_ofp_port(ofproto
,
1839 ofproto_port
->name
);
1842 pp
->port_no
= ofproto_port
->ofp_port
;
1843 netdev_get_etheraddr(netdev
, pp
->hw_addr
);
1844 ovs_strlcpy(pp
->name
, ofproto_port
->name
, sizeof pp
->name
);
1845 netdev_get_flags(netdev
, &flags
);
1846 pp
->config
= flags
& NETDEV_UP
? 0 : OFPUTIL_PC_PORT_DOWN
;
1847 pp
->state
= netdev_get_carrier(netdev
) ? 0 : OFPUTIL_PS_LINK_DOWN
;
1848 netdev_get_features(netdev
, &pp
->curr
, &pp
->advertised
,
1849 &pp
->supported
, &pp
->peer
);
1850 pp
->curr_speed
= netdev_features_to_bps(pp
->curr
, 0) / 1000;
1851 pp
->max_speed
= netdev_features_to_bps(pp
->supported
, 0) / 1000;
1856 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1857 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1860 ofport_equal(const struct ofputil_phy_port
*a
,
1861 const struct ofputil_phy_port
*b
)
1863 return (eth_addr_equals(a
->hw_addr
, b
->hw_addr
)
1864 && a
->state
== b
->state
1865 && !((a
->config
^ b
->config
) & OFPUTIL_PC_PORT_DOWN
)
1866 && a
->curr
== b
->curr
1867 && a
->advertised
== b
->advertised
1868 && a
->supported
== b
->supported
1869 && a
->peer
== b
->peer
1870 && a
->curr_speed
== b
->curr_speed
1871 && a
->max_speed
== b
->max_speed
);
1874 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1875 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1876 * one with the same name or port number). */
1878 ofport_install(struct ofproto
*p
,
1879 struct netdev
*netdev
, const struct ofputil_phy_port
*pp
)
1881 const char *netdev_name
= netdev_get_name(netdev
);
1882 struct ofport
*ofport
;
1885 /* Create ofport. */
1886 ofport
= p
->ofproto_class
->port_alloc();
1891 ofport
->ofproto
= p
;
1892 ofport
->netdev
= netdev
;
1893 ofport
->change_seq
= netdev_change_seq(netdev
);
1895 ofport
->ofp_port
= pp
->port_no
;
1896 ofport
->created
= time_msec();
1898 /* Add port to 'p'. */
1899 hmap_insert(&p
->ports
, &ofport
->hmap_node
,
1900 hash_ofp_port(ofport
->ofp_port
));
1901 shash_add(&p
->port_by_name
, netdev_name
, ofport
);
1903 update_mtu(p
, ofport
);
1905 /* Let the ofproto_class initialize its private data. */
1906 error
= p
->ofproto_class
->port_construct(ofport
);
1910 connmgr_send_port_status(p
->connmgr
, pp
, OFPPR_ADD
);
1914 VLOG_WARN_RL(&rl
, "%s: could not add port %s (%s)",
1915 p
->name
, netdev_name
, ovs_strerror(error
));
1917 ofport_destroy__(ofport
);
1919 netdev_close(netdev
);
1923 /* Removes 'ofport' from 'p' and destroys it. */
1925 ofport_remove(struct ofport
*ofport
)
1927 connmgr_send_port_status(ofport
->ofproto
->connmgr
, &ofport
->pp
,
1929 ofport_destroy(ofport
);
1932 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1935 ofport_remove_with_name(struct ofproto
*ofproto
, const char *name
)
1937 struct ofport
*port
= shash_find_data(&ofproto
->port_by_name
, name
);
1939 ofport_remove(port
);
1943 /* Updates 'port' with new 'pp' description.
1945 * Does not handle a name or port number change. The caller must implement
1946 * such a change as a delete followed by an add. */
1948 ofport_modified(struct ofport
*port
, struct ofputil_phy_port
*pp
)
1950 memcpy(port
->pp
.hw_addr
, pp
->hw_addr
, ETH_ADDR_LEN
);
1951 port
->pp
.config
= ((port
->pp
.config
& ~OFPUTIL_PC_PORT_DOWN
)
1952 | (pp
->config
& OFPUTIL_PC_PORT_DOWN
));
1953 port
->pp
.state
= pp
->state
;
1954 port
->pp
.curr
= pp
->curr
;
1955 port
->pp
.advertised
= pp
->advertised
;
1956 port
->pp
.supported
= pp
->supported
;
1957 port
->pp
.peer
= pp
->peer
;
1958 port
->pp
.curr_speed
= pp
->curr_speed
;
1959 port
->pp
.max_speed
= pp
->max_speed
;
1961 connmgr_send_port_status(port
->ofproto
->connmgr
, &port
->pp
, OFPPR_MODIFY
);
1964 /* Update OpenFlow 'state' in 'port' and notify controller. */
1966 ofproto_port_set_state(struct ofport
*port
, enum ofputil_port_state state
)
1968 if (port
->pp
.state
!= state
) {
1969 port
->pp
.state
= state
;
1970 connmgr_send_port_status(port
->ofproto
->connmgr
, &port
->pp
,
1976 ofproto_port_unregister(struct ofproto
*ofproto
, ofp_port_t ofp_port
)
1978 struct ofport
*port
= ofproto_get_port(ofproto
, ofp_port
);
1980 if (port
->ofproto
->ofproto_class
->set_realdev
) {
1981 port
->ofproto
->ofproto_class
->set_realdev(port
, 0, 0);
1983 if (port
->ofproto
->ofproto_class
->set_stp_port
) {
1984 port
->ofproto
->ofproto_class
->set_stp_port(port
, NULL
);
1986 if (port
->ofproto
->ofproto_class
->set_cfm
) {
1987 port
->ofproto
->ofproto_class
->set_cfm(port
, NULL
);
1989 if (port
->ofproto
->ofproto_class
->bundle_remove
) {
1990 port
->ofproto
->ofproto_class
->bundle_remove(port
);
1996 ofport_destroy__(struct ofport
*port
)
1998 struct ofproto
*ofproto
= port
->ofproto
;
1999 const char *name
= netdev_get_name(port
->netdev
);
2001 hmap_remove(&ofproto
->ports
, &port
->hmap_node
);
2002 shash_delete(&ofproto
->port_by_name
,
2003 shash_find(&ofproto
->port_by_name
, name
));
2005 netdev_close(port
->netdev
);
2006 ofproto
->ofproto_class
->port_dealloc(port
);
2010 ofport_destroy(struct ofport
*port
)
2013 dealloc_ofp_port(port
->ofproto
, port
->ofp_port
);
2014 port
->ofproto
->ofproto_class
->port_destruct(port
);
2015 ofport_destroy__(port
);
2020 ofproto_get_port(const struct ofproto
*ofproto
, ofp_port_t ofp_port
)
2022 struct ofport
*port
;
2024 HMAP_FOR_EACH_IN_BUCKET (port
, hmap_node
, hash_ofp_port(ofp_port
),
2026 if (port
->ofp_port
== ofp_port
) {
2034 ofproto_port_get_stats(const struct ofport
*port
, struct netdev_stats
*stats
)
2036 struct ofproto
*ofproto
= port
->ofproto
;
2039 if (ofproto
->ofproto_class
->port_get_stats
) {
2040 error
= ofproto
->ofproto_class
->port_get_stats(port
, stats
);
2049 update_port(struct ofproto
*ofproto
, const char *name
)
2051 struct ofproto_port ofproto_port
;
2052 struct ofputil_phy_port pp
;
2053 struct netdev
*netdev
;
2054 struct ofport
*port
;
2056 COVERAGE_INC(ofproto_update_port
);
2058 /* Fetch 'name''s location and properties from the datapath. */
2059 netdev
= (!ofproto_port_query_by_name(ofproto
, name
, &ofproto_port
)
2060 ? ofport_open(ofproto
, &ofproto_port
, &pp
)
2064 port
= ofproto_get_port(ofproto
, ofproto_port
.ofp_port
);
2065 if (port
&& !strcmp(netdev_get_name(port
->netdev
), name
)) {
2066 struct netdev
*old_netdev
= port
->netdev
;
2068 /* 'name' hasn't changed location. Any properties changed? */
2069 if (!ofport_equal(&port
->pp
, &pp
)) {
2070 ofport_modified(port
, &pp
);
2073 update_mtu(ofproto
, port
);
2075 /* Install the newly opened netdev in case it has changed.
2076 * Don't close the old netdev yet in case port_modified has to
2077 * remove a retained reference to it.*/
2078 port
->netdev
= netdev
;
2079 port
->change_seq
= netdev_change_seq(netdev
);
2081 if (port
->ofproto
->ofproto_class
->port_modified
) {
2082 port
->ofproto
->ofproto_class
->port_modified(port
);
2085 netdev_close(old_netdev
);
2087 /* If 'port' is nonnull then its name differs from 'name' and thus
2088 * we should delete it. If we think there's a port named 'name'
2089 * then its port number must be wrong now so delete it too. */
2091 ofport_remove(port
);
2093 ofport_remove_with_name(ofproto
, name
);
2094 ofport_install(ofproto
, netdev
, &pp
);
2097 /* Any port named 'name' is gone now. */
2098 ofport_remove_with_name(ofproto
, name
);
2100 ofproto_port_destroy(&ofproto_port
);
2104 init_ports(struct ofproto
*p
)
2106 struct ofproto_port_dump dump
;
2107 struct ofproto_port ofproto_port
;
2108 struct shash_node
*node
, *next
;
2110 OFPROTO_PORT_FOR_EACH (&ofproto_port
, &dump
, p
) {
2111 const char *name
= ofproto_port
.name
;
2113 if (shash_find(&p
->port_by_name
, name
)) {
2114 VLOG_WARN_RL(&rl
, "%s: ignoring duplicate device %s in datapath",
2117 struct ofputil_phy_port pp
;
2118 struct netdev
*netdev
;
2120 /* Check if an OpenFlow port number had been requested. */
2121 node
= shash_find(&init_ofp_ports
, name
);
2123 const struct iface_hint
*iface_hint
= node
->data
;
2124 simap_put(&p
->ofp_requests
, name
,
2125 ofp_to_u16(iface_hint
->ofp_port
));
2128 netdev
= ofport_open(p
, &ofproto_port
, &pp
);
2130 ofport_install(p
, netdev
, &pp
);
2131 if (ofp_to_u16(ofproto_port
.ofp_port
) < p
->max_ports
) {
2132 p
->alloc_port_no
= MAX(p
->alloc_port_no
,
2133 ofp_to_u16(ofproto_port
.ofp_port
));
2139 SHASH_FOR_EACH_SAFE(node
, next
, &init_ofp_ports
) {
2140 struct iface_hint
*iface_hint
= node
->data
;
2142 if (!strcmp(iface_hint
->br_name
, p
->name
)) {
2143 free(iface_hint
->br_name
);
2144 free(iface_hint
->br_type
);
2146 shash_delete(&init_ofp_ports
, node
);
2153 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2154 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2156 find_min_mtu(struct ofproto
*p
)
2158 struct ofport
*ofport
;
2161 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
2162 struct netdev
*netdev
= ofport
->netdev
;
2165 /* Skip any internal ports, since that's what we're trying to
2167 if (!strcmp(netdev_get_type(netdev
), "internal")) {
2171 if (netdev_get_mtu(netdev
, &dev_mtu
)) {
2174 if (!mtu
|| dev_mtu
< mtu
) {
2179 return mtu
? mtu
: ETH_PAYLOAD_MAX
;
2182 /* Update MTU of all datapath devices on 'p' to the minimum of the
2183 * non-datapath ports in event of 'port' added or changed. */
2185 update_mtu(struct ofproto
*p
, struct ofport
*port
)
2187 struct ofport
*ofport
;
2188 struct netdev
*netdev
= port
->netdev
;
2189 int dev_mtu
, old_min
;
2191 if (netdev_get_mtu(netdev
, &dev_mtu
)) {
2195 if (!strcmp(netdev_get_type(port
->netdev
), "internal")) {
2196 if (dev_mtu
> p
->min_mtu
) {
2197 if (!netdev_set_mtu(port
->netdev
, p
->min_mtu
)) {
2198 dev_mtu
= p
->min_mtu
;
2201 port
->mtu
= dev_mtu
;
2205 /* For non-internal port find new min mtu. */
2206 old_min
= p
->min_mtu
;
2207 port
->mtu
= dev_mtu
;
2208 p
->min_mtu
= find_min_mtu(p
);
2209 if (p
->min_mtu
== old_min
) {
2213 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
2214 struct netdev
*netdev
= ofport
->netdev
;
2216 if (!strcmp(netdev_get_type(netdev
), "internal")) {
2217 if (!netdev_set_mtu(netdev
, p
->min_mtu
)) {
2218 ofport
->mtu
= p
->min_mtu
;
2225 ofproto_rule_destroy__(struct rule
*rule
)
2228 rule
->ofproto
->ofproto_class
->rule_destruct(rule
);
2229 cls_rule_destroy(&rule
->cr
);
2230 free(rule
->ofpacts
);
2231 ovs_mutex_destroy(&rule
->timeout_mutex
);
2232 ovs_rwlock_destroy(&rule
->evict
);
2233 rule
->ofproto
->ofproto_class
->rule_dealloc(rule
);
2237 /* This function allows an ofproto implementation to destroy any rules that
2238 * remain when its ->destruct() function is called.. This function implements
2239 * steps 4.4 and 4.5 in the section titled "Rule Life Cycle" in
2240 * ofproto-provider.h.
2242 * This function should only be called from an ofproto implementation's
2243 * ->destruct() function. It is not suitable elsewhere. */
2245 ofproto_rule_delete(struct ofproto
*ofproto
, struct classifier
*cls
,
2247 OVS_REQ_WRLOCK(cls
->rwlock
)
2249 ofproto_delete_rule(ofproto
, cls
, rule
);
2252 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2253 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2255 ofproto_rule_has_out_port(const struct rule
*rule
, ofp_port_t port
)
2257 return (port
== OFPP_ANY
2258 || ofpacts_output_to_port(rule
->ofpacts
, rule
->ofpacts_len
, port
));
2261 /* Returns true if 'rule' has group and equals group_id. */
2263 ofproto_rule_has_out_group(const struct rule
*rule
, uint32_t group_id
)
2265 return (group_id
== OFPG11_ANY
2266 || ofpacts_output_to_group(rule
->ofpacts
, rule
->ofpacts_len
, group_id
));
2269 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2270 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2272 ofoperation_has_out_port(const struct ofoperation
*op
, ofp_port_t out_port
)
2274 if (ofproto_rule_has_out_port(op
->rule
, out_port
)) {
2279 case OFOPERATION_ADD
:
2280 case OFOPERATION_DELETE
:
2283 case OFOPERATION_MODIFY
:
2284 case OFOPERATION_REPLACE
:
2285 return ofpacts_output_to_port(op
->ofpacts
, op
->ofpacts_len
, out_port
);
2291 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2292 * statistics appropriately.
2294 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2295 * with statistics for 'packet' either way.
2297 * Takes ownership of 'packet'. */
2299 rule_execute(struct rule
*rule
, ofp_port_t in_port
, struct ofpbuf
*packet
)
2302 union flow_in_port in_port_
;
2304 in_port_
.ofp_port
= in_port
;
2305 flow_extract(packet
, 0, 0, NULL
, &in_port_
, &flow
);
2306 return rule
->ofproto
->ofproto_class
->rule_execute(rule
, &flow
, packet
);
2309 /* Returns true if 'rule' should be hidden from the controller.
2311 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2312 * (e.g. by in-band control) and are intentionally hidden from the
2315 ofproto_rule_is_hidden(const struct rule
*rule
)
2317 return rule
->cr
.priority
> UINT16_MAX
;
2320 static enum oftable_flags
2321 rule_get_flags(const struct rule
*rule
)
2323 return rule
->ofproto
->tables
[rule
->table_id
].flags
;
2327 rule_is_modifiable(const struct rule
*rule
)
2329 return !(rule_get_flags(rule
) & OFTABLE_READONLY
);
2333 handle_echo_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2335 ofconn_send_reply(ofconn
, make_echo_reply(oh
));
2340 handle_features_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2342 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2343 struct ofputil_switch_features features
;
2344 struct ofport
*port
;
2350 ofproto
->ofproto_class
->get_features(ofproto
, &arp_match_ip
,
2352 ovs_assert(features
.actions
& OFPUTIL_A_OUTPUT
); /* sanity check */
2354 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2355 * if present, are always at the end.) */
2356 n_tables
= ofproto
->n_tables
;
2357 for (i
= 0; i
< ofproto
->n_tables
; i
++) {
2358 if (ofproto
->tables
[i
].flags
& OFTABLE_HIDDEN
) {
2364 features
.datapath_id
= ofproto
->datapath_id
;
2365 features
.n_buffers
= pktbuf_capacity();
2366 features
.n_tables
= n_tables
;
2367 features
.capabilities
= (OFPUTIL_C_FLOW_STATS
| OFPUTIL_C_TABLE_STATS
|
2368 OFPUTIL_C_PORT_STATS
| OFPUTIL_C_QUEUE_STATS
);
2370 features
.capabilities
|= OFPUTIL_C_ARP_MATCH_IP
;
2372 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2373 features
.auxiliary_id
= 0;
2374 b
= ofputil_encode_switch_features(&features
, ofconn_get_protocol(ofconn
),
2376 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
2377 ofputil_put_switch_features_port(&port
->pp
, b
);
2380 ofconn_send_reply(ofconn
, b
);
2385 handle_get_config_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2387 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2388 struct ofp_switch_config
*osc
;
2389 enum ofp_config_flags flags
;
2393 buf
= ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY
, oh
, 0);
2394 osc
= ofpbuf_put_uninit(buf
, sizeof *osc
);
2395 flags
= ofproto
->frag_handling
;
2396 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2397 if (oh
->version
< OFP13_VERSION
2398 && ofconn_get_invalid_ttl_to_controller(ofconn
)) {
2399 flags
|= OFPC_INVALID_TTL_TO_CONTROLLER
;
2401 osc
->flags
= htons(flags
);
2402 osc
->miss_send_len
= htons(ofconn_get_miss_send_len(ofconn
));
2403 ofconn_send_reply(ofconn
, buf
);
2409 handle_set_config(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2411 const struct ofp_switch_config
*osc
= ofpmsg_body(oh
);
2412 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2413 uint16_t flags
= ntohs(osc
->flags
);
2415 if (ofconn_get_type(ofconn
) != OFCONN_PRIMARY
2416 || ofconn_get_role(ofconn
) != OFPCR12_ROLE_SLAVE
) {
2417 enum ofp_config_flags cur
= ofproto
->frag_handling
;
2418 enum ofp_config_flags next
= flags
& OFPC_FRAG_MASK
;
2420 ovs_assert((cur
& OFPC_FRAG_MASK
) == cur
);
2422 if (ofproto
->ofproto_class
->set_frag_handling(ofproto
, next
)) {
2423 ofproto
->frag_handling
= next
;
2425 VLOG_WARN_RL(&rl
, "%s: unsupported fragment handling mode %s",
2427 ofputil_frag_handling_to_string(next
));
2431 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2432 ofconn_set_invalid_ttl_to_controller(ofconn
,
2433 (oh
->version
< OFP13_VERSION
2434 && flags
& OFPC_INVALID_TTL_TO_CONTROLLER
));
2436 ofconn_set_miss_send_len(ofconn
, ntohs(osc
->miss_send_len
));
2441 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2442 * error message code for the caller to propagate upward. Otherwise, returns
2445 * The log message mentions 'msg_type'. */
2447 reject_slave_controller(struct ofconn
*ofconn
)
2449 if (ofconn_get_type(ofconn
) == OFCONN_PRIMARY
2450 && ofconn_get_role(ofconn
) == OFPCR12_ROLE_SLAVE
) {
2451 return OFPERR_OFPBRC_EPERM
;
2457 /* Finds the OFPACT_METER action, if any, in the 'ofpacts_len' bytes of
2458 * 'ofpacts'. If found, returns its meter ID; if not, returns 0.
2460 * This function relies on the order of 'ofpacts' being correct (as checked by
2461 * ofpacts_verify()). */
2463 find_meter(const struct ofpact ofpacts
[], size_t ofpacts_len
)
2465 const struct ofpact
*a
;
2467 OFPACT_FOR_EACH (a
, ofpacts
, ofpacts_len
) {
2468 enum ovs_instruction_type inst
;
2470 inst
= ovs_instruction_type_from_ofpact_type(a
->type
);
2471 if (a
->type
== OFPACT_METER
) {
2472 return ofpact_get_METER(a
)->meter_id
;
2473 } else if (inst
> OVSINST_OFPIT13_METER
) {
2481 /* Checks that the 'ofpacts_len' bytes of actions in 'ofpacts' are appropriate
2482 * for a packet with the prerequisites satisfied by 'flow' in table 'table_id'.
2483 * 'flow' may be temporarily modified, but is restored at return.
2486 ofproto_check_ofpacts(struct ofproto
*ofproto
,
2487 const struct ofpact ofpacts
[], size_t ofpacts_len
,
2488 struct flow
*flow
, uint8_t table_id
)
2493 error
= ofpacts_check(ofpacts
, ofpacts_len
, flow
,
2494 u16_to_ofp(ofproto
->max_ports
), table_id
);
2499 mid
= find_meter(ofpacts
, ofpacts_len
);
2500 if (mid
&& ofproto_get_provider_meter_id(ofproto
, mid
) == UINT32_MAX
) {
2501 return OFPERR_OFPMMFC_INVALID_METER
;
2507 handle_packet_out(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2509 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2510 struct ofputil_packet_out po
;
2511 struct ofpbuf
*payload
;
2512 uint64_t ofpacts_stub
[1024 / 8];
2513 struct ofpbuf ofpacts
;
2515 union flow_in_port in_port_
;
2518 COVERAGE_INC(ofproto_packet_out
);
2520 error
= reject_slave_controller(ofconn
);
2525 /* Decode message. */
2526 ofpbuf_use_stub(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
2527 error
= ofputil_decode_packet_out(&po
, oh
, &ofpacts
);
2529 goto exit_free_ofpacts
;
2531 if (ofp_to_u16(po
.in_port
) >= p
->max_ports
2532 && ofp_to_u16(po
.in_port
) < ofp_to_u16(OFPP_MAX
)) {
2533 error
= OFPERR_OFPBRC_BAD_PORT
;
2534 goto exit_free_ofpacts
;
2539 if (po
.buffer_id
!= UINT32_MAX
) {
2540 error
= ofconn_pktbuf_retrieve(ofconn
, po
.buffer_id
, &payload
, NULL
);
2541 if (error
|| !payload
) {
2542 goto exit_free_ofpacts
;
2545 /* Ensure that the L3 header is 32-bit aligned. */
2546 payload
= ofpbuf_clone_data_with_headroom(po
.packet
, po
.packet_len
, 2);
2549 /* Verify actions against packet, then send packet if successful. */
2550 in_port_
.ofp_port
= po
.in_port
;
2551 flow_extract(payload
, 0, 0, NULL
, &in_port_
, &flow
);
2552 error
= ofproto_check_ofpacts(p
, po
.ofpacts
, po
.ofpacts_len
, &flow
, 0);
2554 error
= p
->ofproto_class
->packet_out(p
, payload
, &flow
,
2555 po
.ofpacts
, po
.ofpacts_len
);
2557 ofpbuf_delete(payload
);
2560 ofpbuf_uninit(&ofpacts
);
2566 update_port_config(struct ofport
*port
,
2567 enum ofputil_port_config config
,
2568 enum ofputil_port_config mask
)
2570 enum ofputil_port_config old_config
= port
->pp
.config
;
2571 enum ofputil_port_config toggle
;
2573 toggle
= (config
^ port
->pp
.config
) & mask
;
2574 if (toggle
& OFPUTIL_PC_PORT_DOWN
) {
2575 if (config
& OFPUTIL_PC_PORT_DOWN
) {
2576 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, NULL
);
2578 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, NULL
);
2580 toggle
&= ~OFPUTIL_PC_PORT_DOWN
;
2583 port
->pp
.config
^= toggle
;
2584 if (port
->pp
.config
!= old_config
) {
2585 port
->ofproto
->ofproto_class
->port_reconfigured(port
, old_config
);
2590 handle_port_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2592 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2593 struct ofputil_port_mod pm
;
2594 struct ofport
*port
;
2597 error
= reject_slave_controller(ofconn
);
2602 error
= ofputil_decode_port_mod(oh
, &pm
);
2607 port
= ofproto_get_port(p
, pm
.port_no
);
2609 return OFPERR_OFPPMFC_BAD_PORT
;
2610 } else if (!eth_addr_equals(port
->pp
.hw_addr
, pm
.hw_addr
)) {
2611 return OFPERR_OFPPMFC_BAD_HW_ADDR
;
2613 update_port_config(port
, pm
.config
, pm
.mask
);
2615 netdev_set_advertisements(port
->netdev
, pm
.advertise
);
2622 handle_desc_stats_request(struct ofconn
*ofconn
,
2623 const struct ofp_header
*request
)
2625 static const char *default_mfr_desc
= "Nicira, Inc.";
2626 static const char *default_hw_desc
= "Open vSwitch";
2627 static const char *default_sw_desc
= VERSION
;
2628 static const char *default_serial_desc
= "None";
2629 static const char *default_dp_desc
= "None";
2631 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2632 struct ofp_desc_stats
*ods
;
2635 msg
= ofpraw_alloc_stats_reply(request
, 0);
2636 ods
= ofpbuf_put_zeros(msg
, sizeof *ods
);
2637 ovs_strlcpy(ods
->mfr_desc
, p
->mfr_desc
? p
->mfr_desc
: default_mfr_desc
,
2638 sizeof ods
->mfr_desc
);
2639 ovs_strlcpy(ods
->hw_desc
, p
->hw_desc
? p
->hw_desc
: default_hw_desc
,
2640 sizeof ods
->hw_desc
);
2641 ovs_strlcpy(ods
->sw_desc
, p
->sw_desc
? p
->sw_desc
: default_sw_desc
,
2642 sizeof ods
->sw_desc
);
2643 ovs_strlcpy(ods
->serial_num
,
2644 p
->serial_desc
? p
->serial_desc
: default_serial_desc
,
2645 sizeof ods
->serial_num
);
2646 ovs_strlcpy(ods
->dp_desc
, p
->dp_desc
? p
->dp_desc
: default_dp_desc
,
2647 sizeof ods
->dp_desc
);
2648 ofconn_send_reply(ofconn
, msg
);
2654 handle_table_stats_request(struct ofconn
*ofconn
,
2655 const struct ofp_header
*request
)
2657 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2658 struct ofp12_table_stats
*ots
;
2663 /* Set up default values.
2665 * ofp12_table_stats is used as a generic structure as
2666 * it is able to hold all the fields for ofp10_table_stats
2667 * and ofp11_table_stats (and of course itself).
2669 ots
= xcalloc(p
->n_tables
, sizeof *ots
);
2670 for (i
= 0; i
< p
->n_tables
; i
++) {
2671 ots
[i
].table_id
= i
;
2672 sprintf(ots
[i
].name
, "table%zu", i
);
2673 ots
[i
].match
= htonll(OFPXMT12_MASK
);
2674 ots
[i
].wildcards
= htonll(OFPXMT12_MASK
);
2675 ots
[i
].write_actions
= htonl(OFPAT11_OUTPUT
);
2676 ots
[i
].apply_actions
= htonl(OFPAT11_OUTPUT
);
2677 ots
[i
].write_setfields
= htonll(OFPXMT12_MASK
);
2678 ots
[i
].apply_setfields
= htonll(OFPXMT12_MASK
);
2679 ots
[i
].metadata_match
= htonll(UINT64_MAX
);
2680 ots
[i
].metadata_write
= htonll(UINT64_MAX
);
2681 ots
[i
].instructions
= htonl(OFPIT11_ALL
);
2682 ots
[i
].config
= htonl(OFPTC11_TABLE_MISS_MASK
);
2683 ots
[i
].max_entries
= htonl(1000000); /* An arbitrary big number. */
2684 ovs_rwlock_rdlock(&p
->tables
[i
].cls
.rwlock
);
2685 ots
[i
].active_count
= htonl(classifier_count(&p
->tables
[i
].cls
));
2686 ovs_rwlock_unlock(&p
->tables
[i
].cls
.rwlock
);
2689 p
->ofproto_class
->get_tables(p
, ots
);
2691 /* Post-process the tables, dropping hidden tables. */
2692 n_tables
= p
->n_tables
;
2693 for (i
= 0; i
< p
->n_tables
; i
++) {
2694 const struct oftable
*table
= &p
->tables
[i
];
2696 if (table
->flags
& OFTABLE_HIDDEN
) {
2702 ovs_strzcpy(ots
[i
].name
, table
->name
, sizeof ots
[i
].name
);
2705 if (table
->max_flows
< ntohl(ots
[i
].max_entries
)) {
2706 ots
[i
].max_entries
= htonl(table
->max_flows
);
2710 msg
= ofputil_encode_table_stats_reply(ots
, n_tables
, request
);
2711 ofconn_send_reply(ofconn
, msg
);
2719 append_port_stat(struct ofport
*port
, struct list
*replies
)
2721 struct ofputil_port_stats ops
= { .port_no
= port
->pp
.port_no
};
2723 calc_duration(port
->created
, time_msec(),
2724 &ops
.duration_sec
, &ops
.duration_nsec
);
2726 /* Intentionally ignore return value, since errors will set
2727 * 'stats' to all-1s, which is correct for OpenFlow, and
2728 * netdev_get_stats() will log errors. */
2729 ofproto_port_get_stats(port
, &ops
.stats
);
2731 ofputil_append_port_stat(replies
, &ops
);
2735 handle_port_stats_request(struct ofconn
*ofconn
,
2736 const struct ofp_header
*request
)
2738 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2739 struct ofport
*port
;
2740 struct list replies
;
2744 error
= ofputil_decode_port_stats_request(request
, &port_no
);
2749 ofpmp_init(&replies
, request
);
2750 if (port_no
!= OFPP_ANY
) {
2751 port
= ofproto_get_port(p
, port_no
);
2753 append_port_stat(port
, &replies
);
2756 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
2757 append_port_stat(port
, &replies
);
2761 ofconn_send_replies(ofconn
, &replies
);
2766 handle_port_desc_stats_request(struct ofconn
*ofconn
,
2767 const struct ofp_header
*request
)
2769 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2770 enum ofp_version version
;
2771 struct ofport
*port
;
2772 struct list replies
;
2774 ofpmp_init(&replies
, request
);
2776 version
= ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn
));
2777 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
2778 ofputil_append_port_desc_stats_reply(version
, &port
->pp
, &replies
);
2781 ofconn_send_replies(ofconn
, &replies
);
2786 hash_cookie(ovs_be64 cookie
)
2788 return hash_2words((OVS_FORCE
uint64_t)cookie
>> 32,
2789 (OVS_FORCE
uint64_t)cookie
);
2793 cookies_insert(struct ofproto
*ofproto
, struct rule
*rule
)
2795 hindex_insert(&ofproto
->cookies
, &rule
->cookie_node
,
2796 hash_cookie(rule
->flow_cookie
));
2800 cookies_remove(struct ofproto
*ofproto
, struct rule
*rule
)
2802 hindex_remove(&ofproto
->cookies
, &rule
->cookie_node
);
2806 ofproto_rule_change_cookie(struct ofproto
*ofproto
, struct rule
*rule
,
2807 ovs_be64 new_cookie
)
2809 if (new_cookie
!= rule
->flow_cookie
) {
2810 cookies_remove(ofproto
, rule
);
2812 rule
->flow_cookie
= new_cookie
;
2814 cookies_insert(ofproto
, rule
);
2819 calc_duration(long long int start
, long long int now
,
2820 uint32_t *sec
, uint32_t *nsec
)
2822 long long int msecs
= now
- start
;
2823 *sec
= msecs
/ 1000;
2824 *nsec
= (msecs
% 1000) * (1000 * 1000);
2827 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2828 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2830 check_table_id(const struct ofproto
*ofproto
, uint8_t table_id
)
2832 return (table_id
== 0xff || table_id
< ofproto
->n_tables
2834 : OFPERR_OFPBRC_BAD_TABLE_ID
);
2838 static struct oftable
*
2839 next_visible_table(const struct ofproto
*ofproto
, uint8_t table_id
)
2841 struct oftable
*table
;
2843 for (table
= &ofproto
->tables
[table_id
];
2844 table
< &ofproto
->tables
[ofproto
->n_tables
];
2846 if (!(table
->flags
& OFTABLE_HIDDEN
)) {
2854 static struct oftable
*
2855 first_matching_table(const struct ofproto
*ofproto
, uint8_t table_id
)
2857 if (table_id
== 0xff) {
2858 return next_visible_table(ofproto
, 0);
2859 } else if (table_id
< ofproto
->n_tables
) {
2860 return &ofproto
->tables
[table_id
];
2866 static struct oftable
*
2867 next_matching_table(const struct ofproto
*ofproto
,
2868 const struct oftable
*table
, uint8_t table_id
)
2870 return (table_id
== 0xff
2871 ? next_visible_table(ofproto
, (table
- ofproto
->tables
) + 1)
2875 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2877 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2878 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2880 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2881 * only once, for that table. (This can be used to access tables marked
2884 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2885 * entered at all. (Perhaps you should have validated TABLE_ID with
2886 * check_table_id().)
2888 * All parameters are evaluated multiple times.
2890 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2891 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2893 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2895 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2896 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2897 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2900 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2901 * to 'out_port' are included.
2903 * Hidden rules are always omitted.
2905 * Returns 0 on success, otherwise an OpenFlow error code. */
2907 collect_rules_loose(struct ofproto
*ofproto
, uint8_t table_id
,
2908 const struct match
*match
,
2909 ovs_be64 cookie
, ovs_be64 cookie_mask
,
2910 ofp_port_t out_port
, uint32_t out_group
,
2913 struct oftable
*table
;
2917 error
= check_table_id(ofproto
, table_id
);
2923 cls_rule_init(&cr
, match
, 0);
2925 if (cookie_mask
== htonll(UINT64_MAX
)) {
2928 HINDEX_FOR_EACH_WITH_HASH (rule
, cookie_node
, hash_cookie(cookie
),
2929 &ofproto
->cookies
) {
2930 if (table_id
!= rule
->table_id
&& table_id
!= 0xff) {
2933 if (ofproto_rule_is_hidden(rule
)) {
2936 if (cls_rule_is_loose_match(&rule
->cr
, &cr
.match
)) {
2937 if (rule
->pending
) {
2938 error
= OFPROTO_POSTPONE
;
2941 if (rule
->flow_cookie
== cookie
/* Hash collisions possible. */
2942 && ofproto_rule_has_out_port(rule
, out_port
)) {
2943 list_push_back(rules
, &rule
->ofproto_node
);
2950 FOR_EACH_MATCHING_TABLE (table
, table_id
, ofproto
) {
2951 struct cls_cursor cursor
;
2954 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
2955 cls_cursor_init(&cursor
, &table
->cls
, &cr
);
2956 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
2957 if (rule
->pending
) {
2958 ovs_rwlock_unlock(&table
->cls
.rwlock
);
2959 error
= OFPROTO_POSTPONE
;
2962 if (!ofproto_rule_is_hidden(rule
)
2963 && ofproto_rule_has_out_port(rule
, out_port
)
2964 && ofproto_rule_has_out_group(rule
, out_group
)
2965 && !((rule
->flow_cookie
^ cookie
) & cookie_mask
)) {
2966 list_push_back(rules
, &rule
->ofproto_node
);
2969 ovs_rwlock_unlock(&table
->cls
.rwlock
);
2973 cls_rule_destroy(&cr
);
2977 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2978 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2979 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2982 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2983 * to 'out_port' are included.
2985 * Hidden rules are always omitted.
2987 * Returns 0 on success, otherwise an OpenFlow error code. */
2989 collect_rules_strict(struct ofproto
*ofproto
, uint8_t table_id
,
2990 const struct match
*match
, unsigned int priority
,
2991 ovs_be64 cookie
, ovs_be64 cookie_mask
,
2992 ofp_port_t out_port
, uint32_t out_group
,
2995 struct oftable
*table
;
2999 error
= check_table_id(ofproto
, table_id
);
3005 cls_rule_init(&cr
, match
, priority
);
3007 if (cookie_mask
== htonll(UINT64_MAX
)) {
3010 HINDEX_FOR_EACH_WITH_HASH (rule
, cookie_node
, hash_cookie(cookie
),
3011 &ofproto
->cookies
) {
3012 if (table_id
!= rule
->table_id
&& table_id
!= 0xff) {
3015 if (ofproto_rule_is_hidden(rule
)) {
3018 if (cls_rule_equal(&rule
->cr
, &cr
)) {
3019 if (rule
->pending
) {
3020 error
= OFPROTO_POSTPONE
;
3023 if (rule
->flow_cookie
== cookie
/* Hash collisions possible. */
3024 && ofproto_rule_has_out_port(rule
, out_port
)) {
3025 list_push_back(rules
, &rule
->ofproto_node
);
3032 FOR_EACH_MATCHING_TABLE (table
, table_id
, ofproto
) {
3035 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
3036 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&table
->cls
,
3038 ovs_rwlock_unlock(&table
->cls
.rwlock
);
3040 if (rule
->pending
) {
3041 error
= OFPROTO_POSTPONE
;
3044 if (!ofproto_rule_is_hidden(rule
)
3045 && ofproto_rule_has_out_port(rule
, out_port
)
3046 && ofproto_rule_has_out_group(rule
, out_group
)
3047 && !((rule
->flow_cookie
^ cookie
) & cookie_mask
)) {
3048 list_push_back(rules
, &rule
->ofproto_node
);
3054 cls_rule_destroy(&cr
);
3058 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
3059 * forced into the range of a uint16_t. */
3061 age_secs(long long int age_ms
)
3063 return (age_ms
< 0 ? 0
3064 : age_ms
>= UINT16_MAX
* 1000 ? UINT16_MAX
3065 : (unsigned int) age_ms
/ 1000);
3069 handle_flow_stats_request(struct ofconn
*ofconn
,
3070 const struct ofp_header
*request
)
3072 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
3073 struct ofputil_flow_stats_request fsr
;
3074 struct list replies
;
3079 error
= ofputil_decode_flow_stats_request(&fsr
, request
);
3084 error
= collect_rules_loose(ofproto
, fsr
.table_id
, &fsr
.match
,
3085 fsr
.cookie
, fsr
.cookie_mask
,
3086 fsr
.out_port
, fsr
.out_group
, &rules
);
3091 ofpmp_init(&replies
, request
);
3092 LIST_FOR_EACH (rule
, ofproto_node
, &rules
) {
3093 long long int now
= time_msec();
3094 struct ofputil_flow_stats fs
;
3096 minimatch_expand(&rule
->cr
.match
, &fs
.match
);
3097 fs
.priority
= rule
->cr
.priority
;
3098 fs
.cookie
= rule
->flow_cookie
;
3099 fs
.table_id
= rule
->table_id
;
3100 calc_duration(rule
->created
, now
, &fs
.duration_sec
, &fs
.duration_nsec
);
3101 fs
.idle_age
= age_secs(now
- rule
->used
);
3102 fs
.hard_age
= age_secs(now
- rule
->modified
);
3103 ofproto
->ofproto_class
->rule_get_stats(rule
, &fs
.packet_count
,
3105 fs
.ofpacts
= rule
->ofpacts
;
3106 fs
.ofpacts_len
= rule
->ofpacts_len
;
3108 ovs_mutex_lock(&rule
->timeout_mutex
);
3109 fs
.idle_timeout
= rule
->idle_timeout
;
3110 fs
.hard_timeout
= rule
->hard_timeout
;
3111 ovs_mutex_unlock(&rule
->timeout_mutex
);
3114 if (rule
->send_flow_removed
) {
3115 fs
.flags
|= OFPUTIL_FF_SEND_FLOW_REM
;
3116 /* FIXME: Implement OFPUTIL_FF_NO_PKT_COUNTS and
3117 OFPUTIL_FF_NO_BYT_COUNTS. */
3119 ofputil_append_flow_stats_reply(&fs
, &replies
);
3121 ofconn_send_replies(ofconn
, &replies
);
3127 flow_stats_ds(struct rule
*rule
, struct ds
*results
)
3129 uint64_t packet_count
, byte_count
;
3131 rule
->ofproto
->ofproto_class
->rule_get_stats(rule
,
3132 &packet_count
, &byte_count
);
3134 if (rule
->table_id
!= 0) {
3135 ds_put_format(results
, "table_id=%"PRIu8
", ", rule
->table_id
);
3137 ds_put_format(results
, "duration=%llds, ",
3138 (time_msec() - rule
->created
) / 1000);
3139 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
3140 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
3141 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
3142 cls_rule_format(&rule
->cr
, results
);
3143 ds_put_char(results
, ',');
3144 ofpacts_format(rule
->ofpacts
, rule
->ofpacts_len
, results
);
3145 ds_put_cstr(results
, "\n");
3148 /* Adds a pretty-printed description of all flows to 'results', including
3149 * hidden flows (e.g., set up by in-band control). */
3151 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
3153 struct oftable
*table
;
3155 OFPROTO_FOR_EACH_TABLE (table
, p
) {
3156 struct cls_cursor cursor
;
3159 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
3160 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
3161 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3162 flow_stats_ds(rule
, results
);
3164 ovs_rwlock_unlock(&table
->cls
.rwlock
);
3168 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3169 * '*engine_type' and '*engine_id', respectively. */
3171 ofproto_get_netflow_ids(const struct ofproto
*ofproto
,
3172 uint8_t *engine_type
, uint8_t *engine_id
)
3174 ofproto
->ofproto_class
->get_netflow_ids(ofproto
, engine_type
, engine_id
);
3177 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
3178 * true if the port's CFM status was successfully stored into '*status'.
3179 * Returns false if the port did not have CFM configured, in which case
3180 * '*status' is indeterminate.
3182 * The caller must provide and owns '*status', and must free 'status->rmps'. */
3184 ofproto_port_get_cfm_status(const struct ofproto
*ofproto
, ofp_port_t ofp_port
,
3185 struct ofproto_cfm_status
*status
)
3187 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
3189 && ofproto
->ofproto_class
->get_cfm_status
3190 && ofproto
->ofproto_class
->get_cfm_status(ofport
, status
));
3194 handle_aggregate_stats_request(struct ofconn
*ofconn
,
3195 const struct ofp_header
*oh
)
3197 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
3198 struct ofputil_flow_stats_request request
;
3199 struct ofputil_aggregate_stats stats
;
3200 bool unknown_packets
, unknown_bytes
;
3201 struct ofpbuf
*reply
;
3206 error
= ofputil_decode_flow_stats_request(&request
, oh
);
3211 error
= collect_rules_loose(ofproto
, request
.table_id
, &request
.match
,
3212 request
.cookie
, request
.cookie_mask
,
3213 request
.out_port
, request
.out_group
, &rules
);
3218 memset(&stats
, 0, sizeof stats
);
3219 unknown_packets
= unknown_bytes
= false;
3220 LIST_FOR_EACH (rule
, ofproto_node
, &rules
) {
3221 uint64_t packet_count
;
3222 uint64_t byte_count
;
3224 ofproto
->ofproto_class
->rule_get_stats(rule
, &packet_count
,
3227 if (packet_count
== UINT64_MAX
) {
3228 unknown_packets
= true;
3230 stats
.packet_count
+= packet_count
;
3233 if (byte_count
== UINT64_MAX
) {
3234 unknown_bytes
= true;
3236 stats
.byte_count
+= byte_count
;
3241 if (unknown_packets
) {
3242 stats
.packet_count
= UINT64_MAX
;
3244 if (unknown_bytes
) {
3245 stats
.byte_count
= UINT64_MAX
;
3248 reply
= ofputil_encode_aggregate_stats_reply(&stats
, oh
);
3249 ofconn_send_reply(ofconn
, reply
);
3254 struct queue_stats_cbdata
{
3255 struct ofport
*ofport
;
3256 struct list replies
;
3261 put_queue_stats(struct queue_stats_cbdata
*cbdata
, uint32_t queue_id
,
3262 const struct netdev_queue_stats
*stats
)
3264 struct ofputil_queue_stats oqs
;
3266 oqs
.port_no
= cbdata
->ofport
->pp
.port_no
;
3267 oqs
.queue_id
= queue_id
;
3268 oqs
.tx_bytes
= stats
->tx_bytes
;
3269 oqs
.tx_packets
= stats
->tx_packets
;
3270 oqs
.tx_errors
= stats
->tx_errors
;
3271 if (stats
->created
!= LLONG_MIN
) {
3272 calc_duration(stats
->created
, cbdata
->now
,
3273 &oqs
.duration_sec
, &oqs
.duration_nsec
);
3275 oqs
.duration_sec
= oqs
.duration_nsec
= UINT32_MAX
;
3277 ofputil_append_queue_stat(&cbdata
->replies
, &oqs
);
3281 handle_queue_stats_dump_cb(uint32_t queue_id
,
3282 struct netdev_queue_stats
*stats
,
3285 struct queue_stats_cbdata
*cbdata
= cbdata_
;
3287 put_queue_stats(cbdata
, queue_id
, stats
);
3291 handle_queue_stats_for_port(struct ofport
*port
, uint32_t queue_id
,
3292 struct queue_stats_cbdata
*cbdata
)
3294 cbdata
->ofport
= port
;
3295 if (queue_id
== OFPQ_ALL
) {
3296 netdev_dump_queue_stats(port
->netdev
,
3297 handle_queue_stats_dump_cb
, cbdata
);
3299 struct netdev_queue_stats stats
;
3301 if (!netdev_get_queue_stats(port
->netdev
, queue_id
, &stats
)) {
3302 put_queue_stats(cbdata
, queue_id
, &stats
);
3304 return OFPERR_OFPQOFC_BAD_QUEUE
;
3311 handle_queue_stats_request(struct ofconn
*ofconn
,
3312 const struct ofp_header
*rq
)
3314 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
3315 struct queue_stats_cbdata cbdata
;
3316 struct ofport
*port
;
3318 struct ofputil_queue_stats_request oqsr
;
3320 COVERAGE_INC(ofproto_queue_req
);
3322 ofpmp_init(&cbdata
.replies
, rq
);
3323 cbdata
.now
= time_msec();
3325 error
= ofputil_decode_queue_stats_request(rq
, &oqsr
);
3330 if (oqsr
.port_no
== OFPP_ANY
) {
3331 error
= OFPERR_OFPQOFC_BAD_QUEUE
;
3332 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
3333 if (!handle_queue_stats_for_port(port
, oqsr
.queue_id
, &cbdata
)) {
3338 port
= ofproto_get_port(ofproto
, oqsr
.port_no
);
3340 ? handle_queue_stats_for_port(port
, oqsr
.queue_id
, &cbdata
)
3341 : OFPERR_OFPQOFC_BAD_PORT
);
3344 ofconn_send_replies(ofconn
, &cbdata
.replies
);
3346 ofpbuf_list_delete(&cbdata
.replies
);
3353 is_flow_deletion_pending(const struct ofproto
*ofproto
,
3354 const struct cls_rule
*cls_rule
,
3357 if (!hmap_is_empty(&ofproto
->deletions
)) {
3358 struct ofoperation
*op
;
3360 HMAP_FOR_EACH_WITH_HASH (op
, hmap_node
,
3361 cls_rule_hash(cls_rule
, table_id
),
3362 &ofproto
->deletions
) {
3363 if (cls_rule_equal(cls_rule
, &op
->rule
->cr
)) {
3373 evict_rule_from_table(struct ofproto
*ofproto
, struct oftable
*table
)
3378 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
3379 n_rules
= classifier_count(&table
->cls
);
3380 ovs_rwlock_unlock(&table
->cls
.rwlock
);
3382 if (n_rules
< table
->max_flows
) {
3384 } else if (!choose_rule_to_evict(table
, &rule
)) {
3385 return OFPERR_OFPFMFC_TABLE_FULL
;
3386 } else if (rule
->pending
) {
3387 ovs_rwlock_unlock(&rule
->evict
);
3388 return OFPROTO_POSTPONE
;
3390 struct ofopgroup
*group
;
3392 group
= ofopgroup_create_unattached(ofproto
);
3393 delete_flow__(rule
, group
, OFPRR_EVICTION
);
3394 ofopgroup_submit(group
);
3400 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3401 * in which no matching flow already exists in the flow table.
3403 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3404 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3405 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3406 * initiated now but may be retried later.
3408 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3409 * ownership remains with the caller.
3411 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3414 add_flow(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3415 struct ofputil_flow_mod
*fm
, const struct ofp_header
*request
)
3417 struct oftable
*table
;
3418 struct ofopgroup
*group
;
3424 error
= check_table_id(ofproto
, fm
->table_id
);
3430 if (fm
->table_id
== 0xff) {
3431 if (ofproto
->ofproto_class
->rule_choose_table
) {
3432 error
= ofproto
->ofproto_class
->rule_choose_table(ofproto
,
3438 ovs_assert(table_id
< ofproto
->n_tables
);
3442 } else if (fm
->table_id
< ofproto
->n_tables
) {
3443 table_id
= fm
->table_id
;
3445 return OFPERR_OFPBRC_BAD_TABLE_ID
;
3448 table
= &ofproto
->tables
[table_id
];
3450 if (table
->flags
& OFTABLE_READONLY
) {
3451 return OFPERR_OFPBRC_EPERM
;
3454 cls_rule_init(&cr
, &fm
->match
, fm
->priority
);
3456 /* Transform "add" into "modify" if there's an existing identical flow. */
3457 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
3458 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&table
->cls
, &cr
));
3459 ovs_rwlock_unlock(&table
->cls
.rwlock
);
3461 cls_rule_destroy(&cr
);
3462 if (!rule_is_modifiable(rule
)) {
3463 return OFPERR_OFPBRC_EPERM
;
3464 } else if (rule
->pending
) {
3465 return OFPROTO_POSTPONE
;
3470 list_push_back(&rules
, &rule
->ofproto_node
);
3471 fm
->modify_cookie
= true;
3472 return modify_flows__(ofproto
, ofconn
, fm
, request
, &rules
);
3476 /* Verify actions. */
3477 error
= ofproto_check_ofpacts(ofproto
, fm
->ofpacts
, fm
->ofpacts_len
,
3478 &fm
->match
.flow
, table_id
);
3480 cls_rule_destroy(&cr
);
3484 /* Serialize against pending deletion. */
3485 if (is_flow_deletion_pending(ofproto
, &cr
, table_id
)) {
3486 cls_rule_destroy(&cr
);
3487 return OFPROTO_POSTPONE
;
3490 /* Check for overlap, if requested. */
3491 if (fm
->flags
& OFPUTIL_FF_CHECK_OVERLAP
) {
3494 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
3495 overlaps
= classifier_rule_overlaps(&table
->cls
, &cr
);
3496 ovs_rwlock_unlock(&table
->cls
.rwlock
);
3499 cls_rule_destroy(&cr
);
3500 return OFPERR_OFPFMFC_OVERLAP
;
3504 /* If necessary, evict an existing rule to clear out space. */
3505 error
= evict_rule_from_table(ofproto
, table
);
3507 cls_rule_destroy(&cr
);
3511 /* Allocate new rule. */
3512 rule
= ofproto
->ofproto_class
->rule_alloc();
3514 cls_rule_destroy(&cr
);
3515 VLOG_WARN_RL(&rl
, "%s: failed to create rule (%s)",
3516 ofproto
->name
, ovs_strerror(error
));
3520 /* Initialize base state. */
3521 rule
->ofproto
= ofproto
;
3522 cls_rule_move(&rule
->cr
, &cr
);
3523 rule
->pending
= NULL
;
3524 rule
->flow_cookie
= fm
->new_cookie
;
3525 rule
->created
= rule
->modified
= rule
->used
= time_msec();
3527 ovs_mutex_init(&rule
->timeout_mutex
);
3528 ovs_mutex_lock(&rule
->timeout_mutex
);
3529 rule
->idle_timeout
= fm
->idle_timeout
;
3530 rule
->hard_timeout
= fm
->hard_timeout
;
3531 ovs_mutex_unlock(&rule
->timeout_mutex
);
3533 rule
->table_id
= table
- ofproto
->tables
;
3534 rule
->send_flow_removed
= (fm
->flags
& OFPUTIL_FF_SEND_FLOW_REM
) != 0;
3535 rule
->ofpacts
= xmemdup(fm
->ofpacts
, fm
->ofpacts_len
);
3536 rule
->ofpacts_len
= fm
->ofpacts_len
;
3537 rule
->meter_id
= find_meter(rule
->ofpacts
, rule
->ofpacts_len
);
3538 list_init(&rule
->meter_list_node
);
3539 rule
->eviction_group
= NULL
;
3540 list_init(&rule
->expirable
);
3541 rule
->monitor_flags
= 0;
3542 rule
->add_seqno
= 0;
3543 rule
->modify_seqno
= 0;
3544 ovs_rwlock_init(&rule
->evict
);
3546 /* Construct rule, initializing derived state. */
3547 error
= ofproto
->ofproto_class
->rule_construct(rule
);
3549 ofproto_rule_destroy__(rule
);
3554 oftable_insert_rule(rule
);
3556 group
= ofopgroup_create(ofproto
, ofconn
, request
, fm
->buffer_id
);
3557 ofoperation_create(group
, rule
, OFOPERATION_ADD
, 0);
3558 ofproto
->ofproto_class
->rule_insert(rule
);
3559 ofopgroup_submit(group
);
3564 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3566 /* Modifies the rules listed in 'rules', changing their actions to match those
3569 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3572 * Returns 0 on success, otherwise an OpenFlow error code. */
3574 modify_flows__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3575 struct ofputil_flow_mod
*fm
, const struct ofp_header
*request
,
3578 enum ofoperation_type type
;
3579 struct ofopgroup
*group
;
3583 type
= fm
->command
== OFPFC_ADD
? OFOPERATION_REPLACE
: OFOPERATION_MODIFY
;
3584 group
= ofopgroup_create(ofproto
, ofconn
, request
, fm
->buffer_id
);
3585 error
= OFPERR_OFPBRC_EPERM
;
3586 LIST_FOR_EACH (rule
, ofproto_node
, rules
) {
3587 struct ofoperation
*op
;
3588 bool actions_changed
;
3589 bool reset_counters
;
3591 /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */
3593 if (rule_is_modifiable(rule
)) {
3594 /* At least one rule is modifiable, don't report EPERM error. */
3600 /* Verify actions. */
3601 error
= ofpacts_check(fm
->ofpacts
, fm
->ofpacts_len
, &fm
->match
.flow
,
3602 u16_to_ofp(ofproto
->max_ports
), rule
->table_id
);
3607 actions_changed
= !ofpacts_equal(fm
->ofpacts
, fm
->ofpacts_len
,
3608 rule
->ofpacts
, rule
->ofpacts_len
);
3610 op
= ofoperation_create(group
, rule
, type
, 0);
3612 if (fm
->modify_cookie
&& fm
->new_cookie
!= htonll(UINT64_MAX
)) {
3613 ofproto_rule_change_cookie(ofproto
, rule
, fm
->new_cookie
);
3615 if (type
== OFOPERATION_REPLACE
) {
3616 ovs_mutex_lock(&rule
->timeout_mutex
);
3617 rule
->idle_timeout
= fm
->idle_timeout
;
3618 rule
->hard_timeout
= fm
->hard_timeout
;
3619 ovs_mutex_unlock(&rule
->timeout_mutex
);
3621 rule
->send_flow_removed
= (fm
->flags
3622 & OFPUTIL_FF_SEND_FLOW_REM
) != 0;
3624 if (fm
->idle_timeout
|| fm
->hard_timeout
) {
3625 if (!rule
->eviction_group
) {
3626 eviction_group_add_rule(rule
);
3629 eviction_group_remove_rule(rule
);
3633 reset_counters
= (fm
->flags
& OFPUTIL_FF_RESET_COUNTS
) != 0;
3634 if (actions_changed
|| reset_counters
) {
3635 op
->ofpacts
= rule
->ofpacts
;
3636 op
->ofpacts_len
= rule
->ofpacts_len
;
3637 op
->meter_id
= rule
->meter_id
;
3638 rule
->ofpacts
= xmemdup(fm
->ofpacts
, fm
->ofpacts_len
);
3639 rule
->ofpacts_len
= fm
->ofpacts_len
;
3640 rule
->meter_id
= find_meter(rule
->ofpacts
, rule
->ofpacts_len
);
3641 rule
->ofproto
->ofproto_class
->rule_modify_actions(rule
,
3644 ofoperation_complete(op
, 0);
3647 ofopgroup_submit(group
);
3653 modify_flows_add(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3654 struct ofputil_flow_mod
*fm
, const struct ofp_header
*request
)
3656 if (fm
->cookie_mask
!= htonll(0) || fm
->new_cookie
== htonll(UINT64_MAX
)) {
3659 return add_flow(ofproto
, ofconn
, fm
, request
);
3662 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3665 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3668 modify_flows_loose(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3669 struct ofputil_flow_mod
*fm
,
3670 const struct ofp_header
*request
)
3675 error
= collect_rules_loose(ofproto
, fm
->table_id
, &fm
->match
,
3676 fm
->cookie
, fm
->cookie_mask
,
3677 OFPP_ANY
, OFPG11_ANY
, &rules
);
3680 } else if (list_is_empty(&rules
)) {
3681 return modify_flows_add(ofproto
, ofconn
, fm
, request
);
3683 return modify_flows__(ofproto
, ofconn
, fm
, request
, &rules
);
3687 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3690 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3693 modify_flow_strict(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3694 struct ofputil_flow_mod
*fm
,
3695 const struct ofp_header
*request
)
3700 error
= collect_rules_strict(ofproto
, fm
->table_id
, &fm
->match
,
3701 fm
->priority
, fm
->cookie
, fm
->cookie_mask
,
3702 OFPP_ANY
, OFPG11_ANY
, &rules
);
3705 } else if (list_is_empty(&rules
)) {
3706 return modify_flows_add(ofproto
, ofconn
, fm
, request
);
3708 return list_is_singleton(&rules
) ? modify_flows__(ofproto
, ofconn
,
3709 fm
, request
, &rules
)
3714 /* OFPFC_DELETE implementation. */
3717 delete_flow__(struct rule
*rule
, struct ofopgroup
*group
,
3718 enum ofp_flow_removed_reason reason
)
3720 struct ofproto
*ofproto
= rule
->ofproto
;
3722 ofproto_rule_send_removed(rule
, reason
);
3724 ofoperation_create(group
, rule
, OFOPERATION_DELETE
, reason
);
3725 oftable_remove_rule(rule
);
3726 ofproto
->ofproto_class
->rule_delete(rule
);
3729 /* Deletes the rules listed in 'rules'.
3731 * Returns 0 on success, otherwise an OpenFlow error code. */
3733 delete_flows__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3734 const struct ofp_header
*request
, struct list
*rules
,
3735 enum ofp_flow_removed_reason reason
)
3737 struct rule
*rule
, *next
;
3738 struct ofopgroup
*group
;
3740 group
= ofopgroup_create(ofproto
, ofconn
, request
, UINT32_MAX
);
3741 LIST_FOR_EACH_SAFE (rule
, next
, ofproto_node
, rules
) {
3742 ovs_rwlock_wrlock(&rule
->evict
);
3743 delete_flow__(rule
, group
, reason
);
3745 ofopgroup_submit(group
);
3750 /* Implements OFPFC_DELETE. */
3752 delete_flows_loose(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3753 const struct ofputil_flow_mod
*fm
,
3754 const struct ofp_header
*request
)
3759 error
= collect_rules_loose(ofproto
, fm
->table_id
, &fm
->match
,
3760 fm
->cookie
, fm
->cookie_mask
,
3761 fm
->out_port
, fm
->out_group
, &rules
);
3762 return (error
? error
3763 : !list_is_empty(&rules
) ? delete_flows__(ofproto
, ofconn
, request
,
3764 &rules
, OFPRR_DELETE
)
3768 /* Implements OFPFC_DELETE_STRICT. */
3770 delete_flow_strict(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3771 const struct ofputil_flow_mod
*fm
,
3772 const struct ofp_header
*request
)
3777 error
= collect_rules_strict(ofproto
, fm
->table_id
, &fm
->match
,
3778 fm
->priority
, fm
->cookie
, fm
->cookie_mask
,
3779 fm
->out_port
, fm
->out_group
, &rules
);
3780 return (error
? error
3781 : list_is_singleton(&rules
) ? delete_flows__(ofproto
, ofconn
,
3788 ofproto_rule_send_removed(struct rule
*rule
, uint8_t reason
)
3790 struct ofputil_flow_removed fr
;
3792 if (ofproto_rule_is_hidden(rule
) || !rule
->send_flow_removed
) {
3796 minimatch_expand(&rule
->cr
.match
, &fr
.match
);
3797 fr
.priority
= rule
->cr
.priority
;
3798 fr
.cookie
= rule
->flow_cookie
;
3800 fr
.table_id
= rule
->table_id
;
3801 calc_duration(rule
->created
, time_msec(),
3802 &fr
.duration_sec
, &fr
.duration_nsec
);
3803 ovs_mutex_lock(&rule
->timeout_mutex
);
3804 fr
.idle_timeout
= rule
->idle_timeout
;
3805 fr
.hard_timeout
= rule
->hard_timeout
;
3806 ovs_mutex_unlock(&rule
->timeout_mutex
);
3807 rule
->ofproto
->ofproto_class
->rule_get_stats(rule
, &fr
.packet_count
,
3810 connmgr_send_flow_removed(rule
->ofproto
->connmgr
, &fr
);
3814 ofproto_rule_update_used(struct rule
*rule
, long long int used
)
3816 if (used
> rule
->used
) {
3817 struct eviction_group
*evg
= rule
->eviction_group
;
3821 heap_change(&evg
->rules
, &rule
->evg_node
,
3822 rule_eviction_priority(rule
));
3827 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3828 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3831 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3834 * ofproto implementation ->run() functions should use this function to expire
3835 * OpenFlow flows. */
3837 ofproto_rule_expire(struct rule
*rule
, uint8_t reason
)
3839 struct ofproto
*ofproto
= rule
->ofproto
;
3840 struct classifier
*cls
= &ofproto
->tables
[rule
->table_id
].cls
;
3842 ovs_assert(reason
== OFPRR_HARD_TIMEOUT
|| reason
== OFPRR_IDLE_TIMEOUT
3843 || reason
== OFPRR_DELETE
|| reason
== OFPRR_GROUP_DELETE
);
3844 ofproto_rule_send_removed(rule
, reason
);
3846 ovs_rwlock_wrlock(&cls
->rwlock
);
3847 ofproto_delete_rule(ofproto
, cls
, rule
);
3848 ovs_rwlock_unlock(&cls
->rwlock
);
3851 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
3852 * means "infinite". */
3854 reduce_timeout(uint16_t max
, uint16_t *timeout
)
3856 if (max
&& (!*timeout
|| *timeout
> max
)) {
3861 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
3862 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
3863 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
3865 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
3867 ofproto_rule_reduce_timeouts(struct rule
*rule
,
3868 uint16_t idle_timeout
, uint16_t hard_timeout
)
3869 OVS_EXCLUDED(rule
->ofproto
->expirable_mutex
, rule
->timeout_mutex
)
3871 if (!idle_timeout
&& !hard_timeout
) {
3875 ovs_mutex_lock(&rule
->ofproto
->expirable_mutex
);
3876 if (list_is_empty(&rule
->expirable
)) {
3877 list_insert(&rule
->ofproto
->expirable
, &rule
->expirable
);
3879 ovs_mutex_unlock(&rule
->ofproto
->expirable_mutex
);
3881 ovs_mutex_lock(&rule
->timeout_mutex
);
3882 reduce_timeout(idle_timeout
, &rule
->idle_timeout
);
3883 reduce_timeout(hard_timeout
, &rule
->hard_timeout
);
3884 ovs_mutex_unlock(&rule
->timeout_mutex
);
3886 if (!rule
->eviction_group
) {
3887 eviction_group_add_rule(rule
);
3892 handle_flow_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3894 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
3895 struct ofputil_flow_mod fm
;
3896 uint64_t ofpacts_stub
[1024 / 8];
3897 struct ofpbuf ofpacts
;
3901 error
= reject_slave_controller(ofconn
);
3906 ofpbuf_use_stub(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
3907 error
= ofputil_decode_flow_mod(&fm
, oh
, ofconn_get_protocol(ofconn
),
3910 error
= handle_flow_mod__(ofproto
, ofconn
, &fm
, oh
);
3913 goto exit_free_ofpacts
;
3916 /* Record the operation for logging a summary report. */
3917 switch (fm
.command
) {
3923 case OFPFC_MODIFY_STRICT
:
3924 ofproto
->n_modify
++;
3928 case OFPFC_DELETE_STRICT
:
3929 ofproto
->n_delete
++;
3934 if (ofproto
->next_op_report
== LLONG_MAX
) {
3935 ofproto
->first_op
= now
;
3936 ofproto
->next_op_report
= MAX(now
+ 10 * 1000,
3937 ofproto
->op_backoff
);
3938 ofproto
->op_backoff
= ofproto
->next_op_report
+ 60 * 1000;
3940 ofproto
->last_op
= now
;
3943 ofpbuf_uninit(&ofpacts
);
3949 handle_flow_mod__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3950 struct ofputil_flow_mod
*fm
, const struct ofp_header
*oh
)
3952 if (ofproto
->n_pending
>= 50) {
3953 ovs_assert(!list_is_empty(&ofproto
->pending
));
3954 return OFPROTO_POSTPONE
;
3957 switch (fm
->command
) {
3959 return add_flow(ofproto
, ofconn
, fm
, oh
);
3962 return modify_flows_loose(ofproto
, ofconn
, fm
, oh
);
3964 case OFPFC_MODIFY_STRICT
:
3965 return modify_flow_strict(ofproto
, ofconn
, fm
, oh
);
3968 return delete_flows_loose(ofproto
, ofconn
, fm
, oh
);
3970 case OFPFC_DELETE_STRICT
:
3971 return delete_flow_strict(ofproto
, ofconn
, fm
, oh
);
3974 if (fm
->command
> 0xff) {
3975 VLOG_WARN_RL(&rl
, "%s: flow_mod has explicit table_id but "
3976 "flow_mod_table_id extension is not enabled",
3979 return OFPERR_OFPFMFC_BAD_COMMAND
;
3984 handle_role_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3986 struct ofputil_role_request request
;
3987 struct ofputil_role_request reply
;
3991 error
= ofputil_decode_role_message(oh
, &request
);
3996 if (request
.role
!= OFPCR12_ROLE_NOCHANGE
) {
3997 if (ofconn_get_role(ofconn
) != request
.role
3998 && ofconn_has_pending_opgroups(ofconn
)) {
3999 return OFPROTO_POSTPONE
;
4002 if (request
.have_generation_id
4003 && !ofconn_set_master_election_id(ofconn
, request
.generation_id
)) {
4004 return OFPERR_OFPRRFC_STALE
;
4007 ofconn_set_role(ofconn
, request
.role
);
4010 reply
.role
= ofconn_get_role(ofconn
);
4011 reply
.have_generation_id
= ofconn_get_master_election_id(
4012 ofconn
, &reply
.generation_id
);
4013 buf
= ofputil_encode_role_reply(oh
, &reply
);
4014 ofconn_send_reply(ofconn
, buf
);
4020 handle_nxt_flow_mod_table_id(struct ofconn
*ofconn
,
4021 const struct ofp_header
*oh
)
4023 const struct nx_flow_mod_table_id
*msg
= ofpmsg_body(oh
);
4024 enum ofputil_protocol cur
, next
;
4026 cur
= ofconn_get_protocol(ofconn
);
4027 next
= ofputil_protocol_set_tid(cur
, msg
->set
!= 0);
4028 ofconn_set_protocol(ofconn
, next
);
4034 handle_nxt_set_flow_format(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4036 const struct nx_set_flow_format
*msg
= ofpmsg_body(oh
);
4037 enum ofputil_protocol cur
, next
;
4038 enum ofputil_protocol next_base
;
4040 next_base
= ofputil_nx_flow_format_to_protocol(ntohl(msg
->format
));
4042 return OFPERR_OFPBRC_EPERM
;
4045 cur
= ofconn_get_protocol(ofconn
);
4046 next
= ofputil_protocol_set_base(cur
, next_base
);
4047 if (cur
!= next
&& ofconn_has_pending_opgroups(ofconn
)) {
4048 /* Avoid sending async messages in surprising protocol. */
4049 return OFPROTO_POSTPONE
;
4052 ofconn_set_protocol(ofconn
, next
);
4057 handle_nxt_set_packet_in_format(struct ofconn
*ofconn
,
4058 const struct ofp_header
*oh
)
4060 const struct nx_set_packet_in_format
*msg
= ofpmsg_body(oh
);
4063 format
= ntohl(msg
->format
);
4064 if (format
!= NXPIF_OPENFLOW10
&& format
!= NXPIF_NXM
) {
4065 return OFPERR_OFPBRC_EPERM
;
4068 if (format
!= ofconn_get_packet_in_format(ofconn
)
4069 && ofconn_has_pending_opgroups(ofconn
)) {
4070 /* Avoid sending async message in surprsing packet in format. */
4071 return OFPROTO_POSTPONE
;
4074 ofconn_set_packet_in_format(ofconn
, format
);
4079 handle_nxt_set_async_config(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4081 const struct nx_async_config
*msg
= ofpmsg_body(oh
);
4082 uint32_t master
[OAM_N_TYPES
];
4083 uint32_t slave
[OAM_N_TYPES
];
4085 master
[OAM_PACKET_IN
] = ntohl(msg
->packet_in_mask
[0]);
4086 master
[OAM_PORT_STATUS
] = ntohl(msg
->port_status_mask
[0]);
4087 master
[OAM_FLOW_REMOVED
] = ntohl(msg
->flow_removed_mask
[0]);
4089 slave
[OAM_PACKET_IN
] = ntohl(msg
->packet_in_mask
[1]);
4090 slave
[OAM_PORT_STATUS
] = ntohl(msg
->port_status_mask
[1]);
4091 slave
[OAM_FLOW_REMOVED
] = ntohl(msg
->flow_removed_mask
[1]);
4093 ofconn_set_async_config(ofconn
, master
, slave
);
4094 if (ofconn_get_type(ofconn
) == OFCONN_SERVICE
&&
4095 !ofconn_get_miss_send_len(ofconn
)) {
4096 ofconn_set_miss_send_len(ofconn
, OFP_DEFAULT_MISS_SEND_LEN
);
4103 handle_nxt_set_controller_id(struct ofconn
*ofconn
,
4104 const struct ofp_header
*oh
)
4106 const struct nx_controller_id
*nci
= ofpmsg_body(oh
);
4108 if (!is_all_zeros(nci
->zero
, sizeof nci
->zero
)) {
4109 return OFPERR_NXBRC_MUST_BE_ZERO
;
4112 ofconn_set_controller_id(ofconn
, ntohs(nci
->controller_id
));
4117 handle_barrier_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4121 if (ofconn_has_pending_opgroups(ofconn
)) {
4122 return OFPROTO_POSTPONE
;
4125 buf
= ofpraw_alloc_reply((oh
->version
== OFP10_VERSION
4126 ? OFPRAW_OFPT10_BARRIER_REPLY
4127 : OFPRAW_OFPT11_BARRIER_REPLY
), oh
, 0);
4128 ofconn_send_reply(ofconn
, buf
);
4133 ofproto_compose_flow_refresh_update(const struct rule
*rule
,
4134 enum nx_flow_monitor_flags flags
,
4137 struct ofoperation
*op
= rule
->pending
;
4138 struct ofputil_flow_update fu
;
4141 if (op
&& op
->type
== OFOPERATION_ADD
) {
4142 /* We'll report the final flow when the operation completes. Reporting
4143 * it now would cause a duplicate report later. */
4147 fu
.event
= (flags
& (NXFMF_INITIAL
| NXFMF_ADD
)
4148 ? NXFME_ADDED
: NXFME_MODIFIED
);
4150 ovs_mutex_lock(&rule
->timeout_mutex
);
4151 fu
.idle_timeout
= rule
->idle_timeout
;
4152 fu
.hard_timeout
= rule
->hard_timeout
;
4153 ovs_mutex_unlock(&rule
->timeout_mutex
);
4154 fu
.table_id
= rule
->table_id
;
4155 fu
.cookie
= rule
->flow_cookie
;
4156 minimatch_expand(&rule
->cr
.match
, &match
);
4158 fu
.priority
= rule
->cr
.priority
;
4159 if (!(flags
& NXFMF_ACTIONS
)) {
4163 fu
.ofpacts
= rule
->ofpacts
;
4164 fu
.ofpacts_len
= rule
->ofpacts_len
;
4166 /* An operation is in progress. Use the previous version of the flow's
4167 * actions, so that when the operation commits we report the change. */
4169 case OFOPERATION_ADD
:
4172 case OFOPERATION_MODIFY
:
4173 case OFOPERATION_REPLACE
:
4175 fu
.ofpacts
= op
->ofpacts
;
4176 fu
.ofpacts_len
= op
->ofpacts_len
;
4178 fu
.ofpacts
= rule
->ofpacts
;
4179 fu
.ofpacts_len
= rule
->ofpacts_len
;
4183 case OFOPERATION_DELETE
:
4184 fu
.ofpacts
= rule
->ofpacts
;
4185 fu
.ofpacts_len
= rule
->ofpacts_len
;
4193 if (list_is_empty(msgs
)) {
4194 ofputil_start_flow_update(msgs
);
4196 ofputil_append_flow_update(&fu
, msgs
);
4200 ofmonitor_compose_refresh_updates(struct list
*rules
, struct list
*msgs
)
4204 LIST_FOR_EACH (rule
, ofproto_node
, rules
) {
4205 enum nx_flow_monitor_flags flags
= rule
->monitor_flags
;
4206 rule
->monitor_flags
= 0;
4208 ofproto_compose_flow_refresh_update(rule
, flags
, msgs
);
4213 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor
*m
,
4214 struct rule
*rule
, uint64_t seqno
,
4217 enum nx_flow_monitor_flags update
;
4219 if (ofproto_rule_is_hidden(rule
)) {
4224 ? ofoperation_has_out_port(rule
->pending
, m
->out_port
)
4225 : ofproto_rule_has_out_port(rule
, m
->out_port
))) {
4230 if (rule
->add_seqno
> seqno
) {
4231 update
= NXFMF_ADD
| NXFMF_MODIFY
;
4232 } else if (rule
->modify_seqno
> seqno
) {
4233 update
= NXFMF_MODIFY
;
4238 if (!(m
->flags
& update
)) {
4242 update
= NXFMF_INITIAL
;
4245 if (!rule
->monitor_flags
) {
4246 list_push_back(rules
, &rule
->ofproto_node
);
4248 rule
->monitor_flags
|= update
| (m
->flags
& NXFMF_ACTIONS
);
4252 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor
*m
,
4256 const struct ofproto
*ofproto
= ofconn_get_ofproto(m
->ofconn
);
4257 const struct ofoperation
*op
;
4258 const struct oftable
*table
;
4259 struct cls_rule target
;
4261 cls_rule_init_from_minimatch(&target
, &m
->match
, 0);
4262 FOR_EACH_MATCHING_TABLE (table
, m
->table_id
, ofproto
) {
4263 struct cls_cursor cursor
;
4266 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
4267 cls_cursor_init(&cursor
, &table
->cls
, &target
);
4268 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
4269 ovs_assert(!rule
->pending
); /* XXX */
4270 ofproto_collect_ofmonitor_refresh_rule(m
, rule
, seqno
, rules
);
4272 ovs_rwlock_unlock(&table
->cls
.rwlock
);
4275 HMAP_FOR_EACH (op
, hmap_node
, &ofproto
->deletions
) {
4276 struct rule
*rule
= op
->rule
;
4278 if (((m
->table_id
== 0xff
4279 ? !(ofproto
->tables
[rule
->table_id
].flags
& OFTABLE_HIDDEN
)
4280 : m
->table_id
== rule
->table_id
))
4281 && cls_rule_is_loose_match(&rule
->cr
, &target
.match
)) {
4282 ofproto_collect_ofmonitor_refresh_rule(m
, rule
, seqno
, rules
);
4285 cls_rule_destroy(&target
);
4289 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor
*m
,
4292 if (m
->flags
& NXFMF_INITIAL
) {
4293 ofproto_collect_ofmonitor_refresh_rules(m
, 0, rules
);
4298 ofmonitor_collect_resume_rules(struct ofmonitor
*m
,
4299 uint64_t seqno
, struct list
*rules
)
4301 ofproto_collect_ofmonitor_refresh_rules(m
, seqno
, rules
);
4305 handle_flow_monitor_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4307 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
4308 struct ofmonitor
**monitors
;
4309 size_t n_monitors
, allocated_monitors
;
4310 struct list replies
;
4317 ofpbuf_use_const(&b
, oh
, ntohs(oh
->length
));
4319 n_monitors
= allocated_monitors
= 0;
4321 struct ofputil_flow_monitor_request request
;
4322 struct ofmonitor
*m
;
4325 retval
= ofputil_decode_flow_monitor_request(&request
, &b
);
4326 if (retval
== EOF
) {
4328 } else if (retval
) {
4333 if (request
.table_id
!= 0xff
4334 && request
.table_id
>= ofproto
->n_tables
) {
4335 error
= OFPERR_OFPBRC_BAD_TABLE_ID
;
4339 error
= ofmonitor_create(&request
, ofconn
, &m
);
4344 if (n_monitors
>= allocated_monitors
) {
4345 monitors
= x2nrealloc(monitors
, &allocated_monitors
,
4348 monitors
[n_monitors
++] = m
;
4352 for (i
= 0; i
< n_monitors
; i
++) {
4353 ofproto_collect_ofmonitor_initial_rules(monitors
[i
], &rules
);
4356 ofpmp_init(&replies
, oh
);
4357 ofmonitor_compose_refresh_updates(&rules
, &replies
);
4358 ofconn_send_replies(ofconn
, &replies
);
4365 for (i
= 0; i
< n_monitors
; i
++) {
4366 ofmonitor_destroy(monitors
[i
]);
4373 handle_flow_monitor_cancel(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4375 struct ofmonitor
*m
;
4378 id
= ofputil_decode_flow_monitor_cancel(oh
);
4379 m
= ofmonitor_lookup(ofconn
, id
);
4381 return OFPERR_NXBRC_FM_BAD_ID
;
4384 ofmonitor_destroy(m
);
4388 /* Meters implementation.
4390 * Meter table entry, indexed by the OpenFlow meter_id.
4391 * These are always dynamically allocated to allocate enough space for
4393 * 'created' is used to compute the duration for meter stats.
4394 * 'list rules' is needed so that we can delete the dependent rules when the
4395 * meter table entry is deleted.
4396 * 'provider_meter_id' is for the provider's private use.
4399 long long int created
; /* Time created. */
4400 struct list rules
; /* List of "struct rule_dpif"s. */
4401 ofproto_meter_id provider_meter_id
;
4402 uint16_t flags
; /* Meter flags. */
4403 uint16_t n_bands
; /* Number of meter bands. */
4404 struct ofputil_meter_band
*bands
;
4408 * This is used in instruction validation at flow set-up time,
4409 * as flows may not use non-existing meters.
4410 * This is also used by ofproto-providers to translate OpenFlow meter_ids
4411 * in METER instructions to the corresponding provider meter IDs.
4412 * Return value of UINT32_MAX signifies an invalid meter.
4415 ofproto_get_provider_meter_id(const struct ofproto
* ofproto
,
4416 uint32_t of_meter_id
)
4418 if (of_meter_id
&& of_meter_id
<= ofproto
->meter_features
.max_meters
) {
4419 const struct meter
*meter
= ofproto
->meters
[of_meter_id
];
4421 return meter
->provider_meter_id
.uint32
;
4428 meter_update(struct meter
*meter
, const struct ofputil_meter_config
*config
)
4432 meter
->flags
= config
->flags
;
4433 meter
->n_bands
= config
->n_bands
;
4434 meter
->bands
= xmemdup(config
->bands
,
4435 config
->n_bands
* sizeof *meter
->bands
);
4438 static struct meter
*
4439 meter_create(const struct ofputil_meter_config
*config
,
4440 ofproto_meter_id provider_meter_id
)
4442 struct meter
*meter
;
4444 meter
= xzalloc(sizeof *meter
);
4445 meter
->provider_meter_id
= provider_meter_id
;
4446 meter
->created
= time_msec();
4447 list_init(&meter
->rules
);
4449 meter_update(meter
, config
);
4455 meter_delete(struct ofproto
*ofproto
, uint32_t first
, uint32_t last
)
4458 for (mid
= first
; mid
<= last
; ++mid
) {
4459 struct meter
*meter
= ofproto
->meters
[mid
];
4461 ofproto
->meters
[mid
] = NULL
;
4462 ofproto
->ofproto_class
->meter_del(ofproto
,
4463 meter
->provider_meter_id
);
4471 handle_add_meter(struct ofproto
*ofproto
, struct ofputil_meter_mod
*mm
)
4473 ofproto_meter_id provider_meter_id
= { UINT32_MAX
};
4474 struct meter
**meterp
= &ofproto
->meters
[mm
->meter
.meter_id
];
4478 return OFPERR_OFPMMFC_METER_EXISTS
;
4481 error
= ofproto
->ofproto_class
->meter_set(ofproto
, &provider_meter_id
,
4484 ovs_assert(provider_meter_id
.uint32
!= UINT32_MAX
);
4485 *meterp
= meter_create(&mm
->meter
, provider_meter_id
);
4491 handle_modify_meter(struct ofproto
*ofproto
, struct ofputil_meter_mod
*mm
)
4493 struct meter
*meter
= ofproto
->meters
[mm
->meter
.meter_id
];
4497 return OFPERR_OFPMMFC_UNKNOWN_METER
;
4500 error
= ofproto
->ofproto_class
->meter_set(ofproto
,
4501 &meter
->provider_meter_id
,
4503 ovs_assert(meter
->provider_meter_id
.uint32
!= UINT32_MAX
);
4505 meter_update(meter
, &mm
->meter
);
4511 handle_delete_meter(struct ofconn
*ofconn
, const struct ofp_header
*oh
,
4512 struct ofputil_meter_mod
*mm
)
4514 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
4515 uint32_t meter_id
= mm
->meter
.meter_id
;
4516 uint32_t first
, last
;
4519 if (meter_id
== OFPM13_ALL
) {
4521 last
= ofproto
->meter_features
.max_meters
;
4523 if (!meter_id
|| meter_id
> ofproto
->meter_features
.max_meters
) {
4526 first
= last
= meter_id
;
4529 /* First delete the rules that use this meter. If any of those rules are
4530 * currently being modified, postpone the whole operation until later. */
4532 for (meter_id
= first
; meter_id
<= last
; ++meter_id
) {
4533 struct meter
*meter
= ofproto
->meters
[meter_id
];
4534 if (meter
&& !list_is_empty(&meter
->rules
)) {
4537 LIST_FOR_EACH (rule
, meter_list_node
, &meter
->rules
) {
4538 if (rule
->pending
) {
4539 return OFPROTO_POSTPONE
;
4541 list_push_back(&rules
, &rule
->ofproto_node
);
4545 if (!list_is_empty(&rules
)) {
4546 delete_flows__(ofproto
, ofconn
, oh
, &rules
, OFPRR_METER_DELETE
);
4549 /* Delete the meters. */
4550 meter_delete(ofproto
, first
, last
);
4556 handle_meter_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4558 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
4559 struct ofputil_meter_mod mm
;
4560 uint64_t bands_stub
[256 / 8];
4561 struct ofpbuf bands
;
4565 error
= reject_slave_controller(ofconn
);
4570 ofpbuf_use_stub(&bands
, bands_stub
, sizeof bands_stub
);
4572 error
= ofputil_decode_meter_mod(oh
, &mm
, &bands
);
4574 goto exit_free_bands
;
4577 meter_id
= mm
.meter
.meter_id
;
4579 if (mm
.command
!= OFPMC13_DELETE
) {
4580 /* Fails also when meters are not implemented by the provider. */
4581 if (!meter_id
|| meter_id
> ofproto
->meter_features
.max_meters
) {
4582 error
= OFPERR_OFPMMFC_INVALID_METER
;
4583 goto exit_free_bands
;
4585 if (mm
.meter
.n_bands
> ofproto
->meter_features
.max_bands
) {
4586 error
= OFPERR_OFPMMFC_OUT_OF_BANDS
;
4587 goto exit_free_bands
;
4591 switch (mm
.command
) {
4593 error
= handle_add_meter(ofproto
, &mm
);
4596 case OFPMC13_MODIFY
:
4597 error
= handle_modify_meter(ofproto
, &mm
);
4600 case OFPMC13_DELETE
:
4601 error
= handle_delete_meter(ofconn
, oh
, &mm
);
4605 error
= OFPERR_OFPMMFC_BAD_COMMAND
;
4610 ofpbuf_uninit(&bands
);
4615 handle_meter_features_request(struct ofconn
*ofconn
,
4616 const struct ofp_header
*request
)
4618 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
4619 struct ofputil_meter_features features
;
4622 if (ofproto
->ofproto_class
->meter_get_features
) {
4623 ofproto
->ofproto_class
->meter_get_features(ofproto
, &features
);
4625 memset(&features
, 0, sizeof features
);
4627 b
= ofputil_encode_meter_features_reply(&features
, request
);
4629 ofconn_send_reply(ofconn
, b
);
4634 handle_meter_request(struct ofconn
*ofconn
, const struct ofp_header
*request
,
4637 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
4638 struct list replies
;
4639 uint64_t bands_stub
[256 / 8];
4640 struct ofpbuf bands
;
4641 uint32_t meter_id
, first
, last
;
4643 ofputil_decode_meter_request(request
, &meter_id
);
4645 if (meter_id
== OFPM13_ALL
) {
4647 last
= ofproto
->meter_features
.max_meters
;
4649 if (!meter_id
|| meter_id
> ofproto
->meter_features
.max_meters
||
4650 !ofproto
->meters
[meter_id
]) {
4651 return OFPERR_OFPMMFC_UNKNOWN_METER
;
4653 first
= last
= meter_id
;
4656 ofpbuf_use_stub(&bands
, bands_stub
, sizeof bands_stub
);
4657 ofpmp_init(&replies
, request
);
4659 for (meter_id
= first
; meter_id
<= last
; ++meter_id
) {
4660 struct meter
*meter
= ofproto
->meters
[meter_id
];
4662 continue; /* Skip non-existing meters. */
4664 if (type
== OFPTYPE_METER_STATS_REQUEST
) {
4665 struct ofputil_meter_stats stats
;
4667 stats
.meter_id
= meter_id
;
4669 /* Provider sets the packet and byte counts, we do the rest. */
4670 stats
.flow_count
= list_size(&meter
->rules
);
4671 calc_duration(meter
->created
, time_msec(),
4672 &stats
.duration_sec
, &stats
.duration_nsec
);
4673 stats
.n_bands
= meter
->n_bands
;
4674 ofpbuf_clear(&bands
);
4676 = ofpbuf_put_uninit(&bands
,
4677 meter
->n_bands
* sizeof *stats
.bands
);
4679 if (!ofproto
->ofproto_class
->meter_get(ofproto
,
4680 meter
->provider_meter_id
,
4682 ofputil_append_meter_stats(&replies
, &stats
);
4684 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
4685 struct ofputil_meter_config config
;
4687 config
.meter_id
= meter_id
;
4688 config
.flags
= meter
->flags
;
4689 config
.n_bands
= meter
->n_bands
;
4690 config
.bands
= meter
->bands
;
4691 ofputil_append_meter_config(&replies
, &config
);
4695 ofconn_send_replies(ofconn
, &replies
);
4696 ofpbuf_uninit(&bands
);
4701 ofproto_group_lookup(const struct ofproto
*ofproto
, uint32_t group_id
,
4702 struct ofgroup
**group
)
4703 OVS_TRY_RDLOCK(true, (*group
)->rwlock
)
4705 ovs_rwlock_rdlock(&ofproto
->groups_rwlock
);
4706 HMAP_FOR_EACH_IN_BUCKET (*group
, hmap_node
,
4707 hash_int(group_id
, 0), &ofproto
->groups
) {
4708 if ((*group
)->group_id
== group_id
) {
4709 ovs_rwlock_rdlock(&(*group
)->rwlock
);
4710 ovs_rwlock_unlock(&ofproto
->groups_rwlock
);
4714 ovs_rwlock_unlock(&ofproto
->groups_rwlock
);
4719 ofproto_group_release(struct ofgroup
*group
)
4720 OVS_RELEASES(group
->rwlock
)
4722 ovs_rwlock_unlock(&group
->rwlock
);
4726 ofproto_group_write_lookup(const struct ofproto
*ofproto
, uint32_t group_id
,
4727 struct ofgroup
**group
)
4728 OVS_TRY_WRLOCK(true, ofproto
->groups_rwlock
)
4729 OVS_TRY_WRLOCK(true, (*group
)->rwlock
)
4731 ovs_rwlock_wrlock(&ofproto
->groups_rwlock
);
4732 HMAP_FOR_EACH_IN_BUCKET (*group
, hmap_node
,
4733 hash_int(group_id
, 0), &ofproto
->groups
) {
4734 if ((*group
)->group_id
== group_id
) {
4735 ovs_rwlock_wrlock(&(*group
)->rwlock
);
4739 ovs_rwlock_unlock(&ofproto
->groups_rwlock
);
4744 ofproto_group_exists(const struct ofproto
*ofproto
, uint32_t group_id
)
4745 OVS_REQ_RDLOCK(ofproto
->groups_rwlock
)
4747 struct ofgroup
*grp
;
4749 HMAP_FOR_EACH_IN_BUCKET (grp
, hmap_node
,
4750 hash_int(group_id
, 0), &ofproto
->groups
) {
4751 if (grp
->group_id
== group_id
) {
4759 append_group_stats(struct ofgroup
*group
, struct list
*replies
)
4760 OVS_REQ_RDLOCK(group
->rwlock
)
4762 struct ofputil_group_stats ogs
;
4763 struct ofproto
*ofproto
= group
->ofproto
;
4764 long long int now
= time_msec();
4767 error
= (ofproto
->ofproto_class
->group_get_stats
4768 ? ofproto
->ofproto_class
->group_get_stats(group
, &ogs
)
4771 ogs
.ref_count
= UINT32_MAX
;
4772 ogs
.packet_count
= UINT64_MAX
;
4773 ogs
.byte_count
= UINT64_MAX
;
4774 ogs
.n_buckets
= group
->n_buckets
;
4775 memset(ogs
.bucket_stats
, 0xff,
4776 ogs
.n_buckets
* sizeof *ogs
.bucket_stats
);
4779 ogs
.group_id
= group
->group_id
;
4780 calc_duration(group
->created
, now
, &ogs
.duration_sec
, &ogs
.duration_nsec
);
4782 ofputil_append_group_stats(replies
, &ogs
);
4786 handle_group_stats_request(struct ofconn
*ofconn
,
4787 const struct ofp_header
*request
)
4789 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
4790 struct list replies
;
4792 struct ofgroup
*group
;
4795 error
= ofputil_decode_group_stats_request(request
, &group_id
);
4800 ofpmp_init(&replies
, request
);
4802 if (group_id
== OFPG_ALL
) {
4803 ovs_rwlock_rdlock(&ofproto
->groups_rwlock
);
4804 HMAP_FOR_EACH (group
, hmap_node
, &ofproto
->groups
) {
4805 ovs_rwlock_rdlock(&group
->rwlock
);
4806 append_group_stats(group
, &replies
);
4807 ovs_rwlock_unlock(&group
->rwlock
);
4809 ovs_rwlock_unlock(&ofproto
->groups_rwlock
);
4811 if (ofproto_group_lookup(ofproto
, group_id
, &group
)) {
4812 append_group_stats(group
, &replies
);
4813 ofproto_group_release(group
);
4817 ofconn_send_replies(ofconn
, &replies
);
4823 handle_group_desc_stats_request(struct ofconn
*ofconn
,
4824 const struct ofp_header
*request
)
4826 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
4827 struct list replies
;
4828 struct ofputil_group_desc gds
;
4829 struct ofgroup
*group
;
4831 ofpmp_init(&replies
, request
);
4833 ovs_rwlock_rdlock(&ofproto
->groups_rwlock
);
4834 HMAP_FOR_EACH (group
, hmap_node
, &ofproto
->groups
) {
4835 gds
.group_id
= group
->group_id
;
4836 gds
.type
= group
->type
;
4837 ofputil_append_group_desc_reply(&gds
, &group
->buckets
, &replies
);
4839 ovs_rwlock_unlock(&ofproto
->groups_rwlock
);
4841 ofconn_send_replies(ofconn
, &replies
);
4847 handle_group_features_stats_request(struct ofconn
*ofconn
,
4848 const struct ofp_header
*request
)
4850 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
4853 msg
= ofputil_encode_group_features_reply(&p
->ogf
, request
);
4855 ofconn_send_reply(ofconn
, msg
);
4861 /* Implements OFPGC11_ADD
4862 * in which no matching flow already exists in the flow table.
4864 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
4865 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
4866 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
4867 * initiated now but may be retried later.
4869 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
4870 * ownership remains with the caller.
4872 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4875 add_group(struct ofproto
*ofproto
, struct ofputil_group_mod
*gm
)
4877 struct ofgroup
*ofgroup
;
4880 if (gm
->group_id
> OFPG_MAX
) {
4881 return OFPERR_OFPGMFC_INVALID_GROUP
;
4883 if (gm
->type
> OFPGT11_FF
) {
4884 return OFPERR_OFPGMFC_BAD_TYPE
;
4887 /* Allocate new group and initialize it. */
4888 ofgroup
= ofproto
->ofproto_class
->group_alloc();
4890 VLOG_WARN_RL(&rl
, "%s: failed to create group", ofproto
->name
);
4891 return OFPERR_OFPGMFC_OUT_OF_GROUPS
;
4894 ovs_rwlock_init(&ofgroup
->rwlock
);
4895 ofgroup
->ofproto
= ofproto
;
4896 ofgroup
->group_id
= gm
->group_id
;
4897 ofgroup
->type
= gm
->type
;
4898 ofgroup
->created
= ofgroup
->modified
= time_msec();
4900 list_move(&ofgroup
->buckets
, &gm
->buckets
);
4901 ofgroup
->n_buckets
= list_size(&ofgroup
->buckets
);
4903 /* Construct called BEFORE any locks are held. */
4904 error
= ofproto
->ofproto_class
->group_construct(ofgroup
);
4909 /* We wrlock as late as possible to minimize the time we jam any other
4910 * threads: No visible state changes before acquiring the lock. */
4911 ovs_rwlock_wrlock(&ofproto
->groups_rwlock
);
4913 if (ofproto
->n_groups
[gm
->type
] >= ofproto
->ogf
.max_groups
[gm
->type
]) {
4914 error
= OFPERR_OFPGMFC_OUT_OF_GROUPS
;
4918 if (ofproto_group_exists(ofproto
, gm
->group_id
)) {
4919 error
= OFPERR_OFPGMFC_GROUP_EXISTS
;
4924 /* Insert new group. */
4925 hmap_insert(&ofproto
->groups
, &ofgroup
->hmap_node
,
4926 hash_int(ofgroup
->group_id
, 0));
4927 ofproto
->n_groups
[ofgroup
->type
]++;
4929 ovs_rwlock_unlock(&ofproto
->groups_rwlock
);
4934 ovs_rwlock_unlock(&ofproto
->groups_rwlock
);
4935 ofproto
->ofproto_class
->group_destruct(ofgroup
);
4937 ofputil_bucket_list_destroy(&ofgroup
->buckets
);
4938 ofproto
->ofproto_class
->group_dealloc(ofgroup
);
4943 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
4946 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4949 modify_group(struct ofproto
*ofproto
, struct ofputil_group_mod
*gm
)
4951 struct ofgroup
*ofgroup
;
4952 struct ofgroup
*victim
;
4955 if (gm
->group_id
> OFPG_MAX
) {
4956 return OFPERR_OFPGMFC_INVALID_GROUP
;
4959 if (gm
->type
> OFPGT11_FF
) {
4960 return OFPERR_OFPGMFC_BAD_TYPE
;
4963 victim
= ofproto
->ofproto_class
->group_alloc();
4965 VLOG_WARN_RL(&rl
, "%s: failed to allocate group", ofproto
->name
);
4966 return OFPERR_OFPGMFC_OUT_OF_GROUPS
;
4969 if (!ofproto_group_write_lookup(ofproto
, gm
->group_id
, &ofgroup
)) {
4970 error
= OFPERR_OFPGMFC_UNKNOWN_GROUP
;
4973 /* Both group's and its container's write locks held now.
4974 * Also, n_groups[] is protected by ofproto->groups_rwlock. */
4975 if (ofgroup
->type
!= gm
->type
4976 && ofproto
->n_groups
[gm
->type
] >= ofproto
->ogf
.max_groups
[gm
->type
]) {
4977 error
= OFPERR_OFPGMFC_OUT_OF_GROUPS
;
4982 list_move(&victim
->buckets
, &ofgroup
->buckets
);
4984 ofgroup
->type
= gm
->type
;
4985 list_move(&ofgroup
->buckets
, &gm
->buckets
);
4986 ofgroup
->n_buckets
= list_size(&ofgroup
->buckets
);
4988 error
= ofproto
->ofproto_class
->group_modify(ofgroup
, victim
);
4990 ofputil_bucket_list_destroy(&victim
->buckets
);
4991 ofproto
->n_groups
[victim
->type
]--;
4992 ofproto
->n_groups
[ofgroup
->type
]++;
4993 ofgroup
->modified
= time_msec();
4995 ofputil_bucket_list_destroy(&ofgroup
->buckets
);
4998 list_move(&ofgroup
->buckets
, &victim
->buckets
);
5002 ovs_rwlock_unlock(&ofgroup
->rwlock
);
5003 ovs_rwlock_unlock(&ofproto
->groups_rwlock
);
5005 ofproto
->ofproto_class
->group_dealloc(victim
);
5010 delete_group__(struct ofproto
*ofproto
, struct ofgroup
*ofgroup
)
5011 OVS_RELEASES(ofproto
->groups_rwlock
)
5013 /* Must wait until existing readers are done,
5014 * while holding the container's write lock at the same time. */
5015 ovs_rwlock_wrlock(&ofgroup
->rwlock
);
5016 hmap_remove(&ofproto
->groups
, &ofgroup
->hmap_node
);
5017 /* No-one can find this group any more. */
5018 ofproto
->n_groups
[ofgroup
->type
]--;
5019 ovs_rwlock_unlock(&ofproto
->groups_rwlock
);
5021 ofproto
->ofproto_class
->group_destruct(ofgroup
);
5022 ofputil_bucket_list_destroy(&ofgroup
->buckets
);
5023 ovs_rwlock_unlock(&ofgroup
->rwlock
);
5024 ovs_rwlock_destroy(&ofgroup
->rwlock
);
5025 ofproto
->ofproto_class
->group_dealloc(ofgroup
);
5028 /* Implements OFPGC_DELETE. */
5030 delete_group(struct ofproto
*ofproto
, uint32_t group_id
)
5032 struct ofgroup
*ofgroup
;
5034 ovs_rwlock_wrlock(&ofproto
->groups_rwlock
);
5035 if (group_id
== OFPG_ALL
) {
5037 struct hmap_node
*node
= hmap_first(&ofproto
->groups
);
5041 ofgroup
= CONTAINER_OF(node
, struct ofgroup
, hmap_node
);
5042 delete_group__(ofproto
, ofgroup
);
5043 /* Lock for each node separately, so that we will not jam the
5044 * other threads for too long time. */
5045 ovs_rwlock_wrlock(&ofproto
->groups_rwlock
);
5048 HMAP_FOR_EACH_IN_BUCKET (ofgroup
, hmap_node
,
5049 hash_int(group_id
, 0), &ofproto
->groups
) {
5050 if (ofgroup
->group_id
== group_id
) {
5051 delete_group__(ofproto
, ofgroup
);
5056 ovs_rwlock_unlock(&ofproto
->groups_rwlock
);
5060 handle_group_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
5062 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
5063 struct ofputil_group_mod gm
;
5066 error
= reject_slave_controller(ofconn
);
5071 error
= ofputil_decode_group_mod(oh
, &gm
);
5076 switch (gm
.command
) {
5078 return add_group(ofproto
, &gm
);
5080 case OFPGC11_MODIFY
:
5081 return modify_group(ofproto
, &gm
);
5083 case OFPGC11_DELETE
:
5084 delete_group(ofproto
, gm
.group_id
);
5088 if (gm
.command
> OFPGC11_DELETE
) {
5089 VLOG_WARN_RL(&rl
, "%s: Invalid group_mod command type %d",
5090 ofproto
->name
, gm
.command
);
5092 return OFPERR_OFPGMFC_BAD_COMMAND
;
5097 handle_openflow__(struct ofconn
*ofconn
, const struct ofpbuf
*msg
)
5099 const struct ofp_header
*oh
= msg
->data
;
5103 error
= ofptype_decode(&type
, oh
);
5109 /* OpenFlow requests. */
5110 case OFPTYPE_ECHO_REQUEST
:
5111 return handle_echo_request(ofconn
, oh
);
5113 case OFPTYPE_FEATURES_REQUEST
:
5114 return handle_features_request(ofconn
, oh
);
5116 case OFPTYPE_GET_CONFIG_REQUEST
:
5117 return handle_get_config_request(ofconn
, oh
);
5119 case OFPTYPE_SET_CONFIG
:
5120 return handle_set_config(ofconn
, oh
);
5122 case OFPTYPE_PACKET_OUT
:
5123 return handle_packet_out(ofconn
, oh
);
5125 case OFPTYPE_PORT_MOD
:
5126 return handle_port_mod(ofconn
, oh
);
5128 case OFPTYPE_FLOW_MOD
:
5129 return handle_flow_mod(ofconn
, oh
);
5131 case OFPTYPE_GROUP_MOD
:
5132 return handle_group_mod(ofconn
, oh
);
5134 case OFPTYPE_METER_MOD
:
5135 return handle_meter_mod(ofconn
, oh
);
5137 case OFPTYPE_BARRIER_REQUEST
:
5138 return handle_barrier_request(ofconn
, oh
);
5140 case OFPTYPE_ROLE_REQUEST
:
5141 return handle_role_request(ofconn
, oh
);
5143 /* OpenFlow replies. */
5144 case OFPTYPE_ECHO_REPLY
:
5147 /* Nicira extension requests. */
5148 case OFPTYPE_FLOW_MOD_TABLE_ID
:
5149 return handle_nxt_flow_mod_table_id(ofconn
, oh
);
5151 case OFPTYPE_SET_FLOW_FORMAT
:
5152 return handle_nxt_set_flow_format(ofconn
, oh
);
5154 case OFPTYPE_SET_PACKET_IN_FORMAT
:
5155 return handle_nxt_set_packet_in_format(ofconn
, oh
);
5157 case OFPTYPE_SET_CONTROLLER_ID
:
5158 return handle_nxt_set_controller_id(ofconn
, oh
);
5160 case OFPTYPE_FLOW_AGE
:
5161 /* Nothing to do. */
5164 case OFPTYPE_FLOW_MONITOR_CANCEL
:
5165 return handle_flow_monitor_cancel(ofconn
, oh
);
5167 case OFPTYPE_SET_ASYNC_CONFIG
:
5168 return handle_nxt_set_async_config(ofconn
, oh
);
5170 /* Statistics requests. */
5171 case OFPTYPE_DESC_STATS_REQUEST
:
5172 return handle_desc_stats_request(ofconn
, oh
);
5174 case OFPTYPE_FLOW_STATS_REQUEST
:
5175 return handle_flow_stats_request(ofconn
, oh
);
5177 case OFPTYPE_AGGREGATE_STATS_REQUEST
:
5178 return handle_aggregate_stats_request(ofconn
, oh
);
5180 case OFPTYPE_TABLE_STATS_REQUEST
:
5181 return handle_table_stats_request(ofconn
, oh
);
5183 case OFPTYPE_PORT_STATS_REQUEST
:
5184 return handle_port_stats_request(ofconn
, oh
);
5186 case OFPTYPE_QUEUE_STATS_REQUEST
:
5187 return handle_queue_stats_request(ofconn
, oh
);
5189 case OFPTYPE_PORT_DESC_STATS_REQUEST
:
5190 return handle_port_desc_stats_request(ofconn
, oh
);
5192 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST
:
5193 return handle_flow_monitor_request(ofconn
, oh
);
5195 case OFPTYPE_METER_STATS_REQUEST
:
5196 case OFPTYPE_METER_CONFIG_STATS_REQUEST
:
5197 return handle_meter_request(ofconn
, oh
, type
);
5199 case OFPTYPE_METER_FEATURES_STATS_REQUEST
:
5200 return handle_meter_features_request(ofconn
, oh
);
5202 case OFPTYPE_GROUP_STATS_REQUEST
:
5203 return handle_group_stats_request(ofconn
, oh
);
5205 case OFPTYPE_GROUP_DESC_STATS_REQUEST
:
5206 return handle_group_desc_stats_request(ofconn
, oh
);
5208 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST
:
5209 return handle_group_features_stats_request(ofconn
, oh
);
5211 /* FIXME: Change the following once they are implemented: */
5212 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST
:
5213 case OFPTYPE_GET_ASYNC_REQUEST
:
5214 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST
:
5215 return OFPERR_OFPBRC_BAD_TYPE
;
5219 case OFPTYPE_FEATURES_REPLY
:
5220 case OFPTYPE_GET_CONFIG_REPLY
:
5221 case OFPTYPE_PACKET_IN
:
5222 case OFPTYPE_FLOW_REMOVED
:
5223 case OFPTYPE_PORT_STATUS
:
5224 case OFPTYPE_BARRIER_REPLY
:
5225 case OFPTYPE_QUEUE_GET_CONFIG_REPLY
:
5226 case OFPTYPE_DESC_STATS_REPLY
:
5227 case OFPTYPE_FLOW_STATS_REPLY
:
5228 case OFPTYPE_QUEUE_STATS_REPLY
:
5229 case OFPTYPE_PORT_STATS_REPLY
:
5230 case OFPTYPE_TABLE_STATS_REPLY
:
5231 case OFPTYPE_AGGREGATE_STATS_REPLY
:
5232 case OFPTYPE_PORT_DESC_STATS_REPLY
:
5233 case OFPTYPE_ROLE_REPLY
:
5234 case OFPTYPE_FLOW_MONITOR_PAUSED
:
5235 case OFPTYPE_FLOW_MONITOR_RESUMED
:
5236 case OFPTYPE_FLOW_MONITOR_STATS_REPLY
:
5237 case OFPTYPE_GET_ASYNC_REPLY
:
5238 case OFPTYPE_GROUP_STATS_REPLY
:
5239 case OFPTYPE_GROUP_DESC_STATS_REPLY
:
5240 case OFPTYPE_GROUP_FEATURES_STATS_REPLY
:
5241 case OFPTYPE_METER_STATS_REPLY
:
5242 case OFPTYPE_METER_CONFIG_STATS_REPLY
:
5243 case OFPTYPE_METER_FEATURES_STATS_REPLY
:
5244 case OFPTYPE_TABLE_FEATURES_STATS_REPLY
:
5246 return OFPERR_OFPBRC_BAD_TYPE
;
5251 handle_openflow(struct ofconn
*ofconn
, const struct ofpbuf
*ofp_msg
)
5253 int error
= handle_openflow__(ofconn
, ofp_msg
);
5254 if (error
&& error
!= OFPROTO_POSTPONE
) {
5255 ofconn_send_error(ofconn
, ofp_msg
->data
, error
);
5257 COVERAGE_INC(ofproto_recv_openflow
);
5258 return error
!= OFPROTO_POSTPONE
;
5261 /* Asynchronous operations. */
5263 /* Creates and returns a new ofopgroup that is not associated with any
5264 * OpenFlow connection.
5266 * The caller should add operations to the returned group with
5267 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5268 static struct ofopgroup
*
5269 ofopgroup_create_unattached(struct ofproto
*ofproto
)
5271 struct ofopgroup
*group
= xzalloc(sizeof *group
);
5272 group
->ofproto
= ofproto
;
5273 list_init(&group
->ofproto_node
);
5274 list_init(&group
->ops
);
5275 list_init(&group
->ofconn_node
);
5279 /* Creates and returns a new ofopgroup for 'ofproto'.
5281 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
5282 * connection. The 'request' and 'buffer_id' arguments are ignored.
5284 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
5285 * If the ofopgroup eventually fails, then the error reply will include
5286 * 'request'. If the ofopgroup eventually succeeds, then the packet with
5287 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
5289 * The caller should add operations to the returned group with
5290 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5291 static struct ofopgroup
*
5292 ofopgroup_create(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
5293 const struct ofp_header
*request
, uint32_t buffer_id
)
5295 struct ofopgroup
*group
= ofopgroup_create_unattached(ofproto
);
5297 size_t request_len
= ntohs(request
->length
);
5299 ovs_assert(ofconn_get_ofproto(ofconn
) == ofproto
);
5301 ofconn_add_opgroup(ofconn
, &group
->ofconn_node
);
5302 group
->ofconn
= ofconn
;
5303 group
->request
= xmemdup(request
, MIN(request_len
, 64));
5304 group
->buffer_id
= buffer_id
;
5309 /* Submits 'group' for processing.
5311 * If 'group' contains no operations (e.g. none were ever added, or all of the
5312 * ones that were added completed synchronously), then it is destroyed
5313 * immediately. Otherwise it is added to the ofproto's list of pending
5316 ofopgroup_submit(struct ofopgroup
*group
)
5318 if (!group
->n_running
) {
5319 ofopgroup_complete(group
);
5321 list_push_back(&group
->ofproto
->pending
, &group
->ofproto_node
);
5322 group
->ofproto
->n_pending
++;
5327 ofopgroup_complete(struct ofopgroup
*group
)
5329 struct ofproto
*ofproto
= group
->ofproto
;
5331 struct ofconn
*abbrev_ofconn
;
5332 ovs_be32 abbrev_xid
;
5334 struct ofoperation
*op
, *next_op
;
5337 ovs_assert(!group
->n_running
);
5340 LIST_FOR_EACH (op
, group_node
, &group
->ops
) {
5347 if (!error
&& group
->ofconn
&& group
->buffer_id
!= UINT32_MAX
) {
5348 LIST_FOR_EACH (op
, group_node
, &group
->ops
) {
5349 if (op
->type
!= OFOPERATION_DELETE
) {
5350 struct ofpbuf
*packet
;
5353 error
= ofconn_pktbuf_retrieve(group
->ofconn
, group
->buffer_id
,
5357 error
= rule_execute(op
->rule
, in_port
, packet
);
5364 if (!error
&& !list_is_empty(&group
->ofconn_node
)) {
5365 abbrev_ofconn
= group
->ofconn
;
5366 abbrev_xid
= group
->request
->xid
;
5368 abbrev_ofconn
= NULL
;
5369 abbrev_xid
= htonl(0);
5371 LIST_FOR_EACH_SAFE (op
, next_op
, group_node
, &group
->ops
) {
5372 struct rule
*rule
= op
->rule
;
5374 /* We generally want to report the change to active OpenFlow flow
5375 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
5377 - The operation failed.
5379 - The affected rule is not visible to controllers.
5381 - The operation's only effect was to update rule->modified. */
5383 || ofproto_rule_is_hidden(rule
)
5384 || (op
->type
== OFOPERATION_MODIFY
5386 && rule
->flow_cookie
== op
->flow_cookie
))) {
5387 /* Check that we can just cast from ofoperation_type to
5388 * nx_flow_update_event. */
5389 enum nx_flow_update_event event_type
;
5392 case OFOPERATION_ADD
:
5393 case OFOPERATION_REPLACE
:
5394 event_type
= NXFME_ADDED
;
5397 case OFOPERATION_DELETE
:
5398 event_type
= NXFME_DELETED
;
5401 case OFOPERATION_MODIFY
:
5402 event_type
= NXFME_MODIFIED
;
5409 ofmonitor_report(ofproto
->connmgr
, rule
, event_type
,
5410 op
->reason
, abbrev_ofconn
, abbrev_xid
);
5413 rule
->pending
= NULL
;
5416 case OFOPERATION_ADD
:
5420 vid_mask
= minimask_get_vid_mask(&rule
->cr
.match
.mask
);
5421 if (vid_mask
== VLAN_VID_MASK
) {
5422 if (ofproto
->vlan_bitmap
) {
5423 uint16_t vid
= miniflow_get_vid(&rule
->cr
.match
.flow
);
5424 if (!bitmap_is_set(ofproto
->vlan_bitmap
, vid
)) {
5425 bitmap_set1(ofproto
->vlan_bitmap
, vid
);
5426 ofproto
->vlans_changed
= true;
5429 ofproto
->vlans_changed
= true;
5433 ovs_rwlock_wrlock(&rule
->evict
);
5434 oftable_remove_rule(rule
);
5435 ofproto_rule_destroy__(rule
);
5439 case OFOPERATION_DELETE
:
5440 ovs_assert(!op
->error
);
5441 ofproto_rule_destroy__(rule
);
5445 case OFOPERATION_MODIFY
:
5446 case OFOPERATION_REPLACE
:
5448 long long int now
= time_msec();
5450 rule
->modified
= now
;
5451 if (op
->type
== OFOPERATION_REPLACE
) {
5452 rule
->created
= rule
->used
= now
;
5455 ofproto_rule_change_cookie(ofproto
, rule
, op
->flow_cookie
);
5456 ovs_mutex_lock(&rule
->timeout_mutex
);
5457 rule
->idle_timeout
= op
->idle_timeout
;
5458 rule
->hard_timeout
= op
->hard_timeout
;
5459 ovs_mutex_unlock(&rule
->timeout_mutex
);
5461 free(rule
->ofpacts
);
5462 rule
->ofpacts
= op
->ofpacts
;
5463 rule
->ofpacts_len
= op
->ofpacts_len
;
5465 op
->ofpacts_len
= 0;
5467 rule
->send_flow_removed
= op
->send_flow_removed
;
5475 ofoperation_destroy(op
);
5478 ofmonitor_flush(ofproto
->connmgr
);
5480 if (!list_is_empty(&group
->ofproto_node
)) {
5481 ovs_assert(ofproto
->n_pending
> 0);
5482 ofproto
->n_pending
--;
5483 list_remove(&group
->ofproto_node
);
5485 if (!list_is_empty(&group
->ofconn_node
)) {
5486 list_remove(&group
->ofconn_node
);
5488 ofconn_send_error(group
->ofconn
, group
->request
, error
);
5490 connmgr_retry(ofproto
->connmgr
);
5492 free(group
->request
);
5496 /* Initiates a new operation on 'rule', of the specified 'type', within
5497 * 'group'. Prior to calling, 'rule' must not have any pending operation.
5499 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
5500 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
5502 * Returns the newly created ofoperation (which is also available as
5503 * rule->pending). */
5504 static struct ofoperation
*
5505 ofoperation_create(struct ofopgroup
*group
, struct rule
*rule
,
5506 enum ofoperation_type type
,
5507 enum ofp_flow_removed_reason reason
)
5509 struct ofproto
*ofproto
= group
->ofproto
;
5510 struct ofoperation
*op
;
5512 ovs_assert(!rule
->pending
);
5514 op
= rule
->pending
= xzalloc(sizeof *op
);
5516 list_push_back(&group
->ops
, &op
->group_node
);
5519 op
->reason
= reason
;
5520 op
->flow_cookie
= rule
->flow_cookie
;
5521 ovs_mutex_lock(&rule
->timeout_mutex
);
5522 op
->idle_timeout
= rule
->idle_timeout
;
5523 op
->hard_timeout
= rule
->hard_timeout
;
5524 ovs_mutex_unlock(&rule
->timeout_mutex
);
5525 op
->send_flow_removed
= rule
->send_flow_removed
;
5529 if (type
== OFOPERATION_DELETE
) {
5530 hmap_insert(&ofproto
->deletions
, &op
->hmap_node
,
5531 cls_rule_hash(&rule
->cr
, rule
->table_id
));
5538 ofoperation_destroy(struct ofoperation
*op
)
5540 struct ofopgroup
*group
= op
->group
;
5543 op
->rule
->pending
= NULL
;
5545 if (op
->type
== OFOPERATION_DELETE
) {
5546 hmap_remove(&group
->ofproto
->deletions
, &op
->hmap_node
);
5548 list_remove(&op
->group_node
);
5553 /* Indicates that 'op' completed with status 'error', which is either 0 to
5554 * indicate success or an OpenFlow error code on failure.
5556 * If 'error' is 0, indicating success, the operation will be committed
5557 * permanently to the flow table.
5559 * If 'error' is nonzero, then generally the operation will be rolled back:
5561 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
5562 * restores the original rule. The caller must have uninitialized any
5563 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
5564 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
5565 * and 7 for the new rule, calling its ->rule_dealloc() function.
5567 * - If 'op' is a "modify flow" operation, ofproto restores the original
5570 * - 'op' must not be a "delete flow" operation. Removing a rule is not
5571 * allowed to fail. It must always succeed.
5573 * Please see the large comment in ofproto/ofproto-provider.h titled
5574 * "Asynchronous Operation Support" for more information. */
5576 ofoperation_complete(struct ofoperation
*op
, enum ofperr error
)
5578 struct ofopgroup
*group
= op
->group
;
5580 ovs_assert(op
->rule
->pending
== op
);
5581 ovs_assert(group
->n_running
> 0);
5582 ovs_assert(!error
|| op
->type
!= OFOPERATION_DELETE
);
5585 if (!--group
->n_running
&& !list_is_empty(&group
->ofproto_node
)) {
5586 ofopgroup_complete(group
);
5591 pick_datapath_id(const struct ofproto
*ofproto
)
5593 const struct ofport
*port
;
5595 port
= ofproto_get_port(ofproto
, OFPP_LOCAL
);
5597 uint8_t ea
[ETH_ADDR_LEN
];
5600 error
= netdev_get_etheraddr(port
->netdev
, ea
);
5602 return eth_addr_to_uint64(ea
);
5604 VLOG_WARN("%s: could not get MAC address for %s (%s)",
5605 ofproto
->name
, netdev_get_name(port
->netdev
),
5606 ovs_strerror(error
));
5608 return ofproto
->fallback_dpid
;
5612 pick_fallback_dpid(void)
5614 uint8_t ea
[ETH_ADDR_LEN
];
5615 eth_addr_nicira_random(ea
);
5616 return eth_addr_to_uint64(ea
);
5619 /* Table overflow policy. */
5621 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
5622 * to NULL if the table is not configured to evict rules or if the table
5623 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
5624 * or with no timeouts are not evictable.) */
5626 choose_rule_to_evict(struct oftable
*table
, struct rule
**rulep
)
5628 struct eviction_group
*evg
;
5631 if (!table
->eviction_fields
) {
5635 /* In the common case, the outer and inner loops here will each be entered
5638 * - The inner loop normally "return"s in its first iteration. If the
5639 * eviction group has any evictable rules, then it always returns in
5642 * - The outer loop only iterates more than once if the largest eviction
5643 * group has no evictable rules.
5645 * - The outer loop can exit only if table's 'max_flows' is all filled up
5646 * by unevictable rules. */
5647 HEAP_FOR_EACH (evg
, size_node
, &table
->eviction_groups_by_size
) {
5650 HEAP_FOR_EACH (rule
, evg_node
, &evg
->rules
) {
5651 if (!ovs_rwlock_trywrlock(&rule
->evict
)) {
5661 /* Searches 'ofproto' for tables that have more flows than their configured
5662 * maximum and that have flow eviction enabled, and evicts as many flows as
5663 * necessary and currently feasible from them.
5665 * This triggers only when an OpenFlow table has N flows in it and then the
5666 * client configures a maximum number of flows less than N. */
5668 ofproto_evict(struct ofproto
*ofproto
)
5670 struct ofopgroup
*group
;
5671 struct oftable
*table
;
5673 group
= ofopgroup_create_unattached(ofproto
);
5674 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
5675 while (table
->eviction_fields
) {
5679 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
5680 n_rules
= classifier_count(&table
->cls
);
5681 ovs_rwlock_unlock(&table
->cls
.rwlock
);
5683 if (n_rules
<= table
->max_flows
) {
5687 if (!choose_rule_to_evict(table
, &rule
)) {
5691 if (rule
->pending
) {
5692 ovs_rwlock_unlock(&rule
->evict
);
5696 ofoperation_create(group
, rule
,
5697 OFOPERATION_DELETE
, OFPRR_EVICTION
);
5698 oftable_remove_rule(rule
);
5699 ofproto
->ofproto_class
->rule_delete(rule
);
5702 ofopgroup_submit(group
);
5705 /* Eviction groups. */
5707 /* Returns the priority to use for an eviction_group that contains 'n_rules'
5708 * rules. The priority contains low-order random bits to ensure that eviction
5709 * groups with the same number of rules are prioritized randomly. */
5711 eviction_group_priority(size_t n_rules
)
5713 uint16_t size
= MIN(UINT16_MAX
, n_rules
);
5714 return (size
<< 16) | random_uint16();
5717 /* Updates 'evg', an eviction_group within 'table', following a change that
5718 * adds or removes rules in 'evg'. */
5720 eviction_group_resized(struct oftable
*table
, struct eviction_group
*evg
)
5722 heap_change(&table
->eviction_groups_by_size
, &evg
->size_node
,
5723 eviction_group_priority(heap_count(&evg
->rules
)));
5726 /* Destroys 'evg', an eviction_group within 'table':
5728 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
5729 * rules themselves, just removes them from the eviction group.)
5731 * - Removes 'evg' from 'table'.
5735 eviction_group_destroy(struct oftable
*table
, struct eviction_group
*evg
)
5737 while (!heap_is_empty(&evg
->rules
)) {
5740 rule
= CONTAINER_OF(heap_pop(&evg
->rules
), struct rule
, evg_node
);
5741 rule
->eviction_group
= NULL
;
5743 hmap_remove(&table
->eviction_groups_by_id
, &evg
->id_node
);
5744 heap_remove(&table
->eviction_groups_by_size
, &evg
->size_node
);
5745 heap_destroy(&evg
->rules
);
5749 /* Removes 'rule' from its eviction group, if any. */
5751 eviction_group_remove_rule(struct rule
*rule
)
5753 if (rule
->eviction_group
) {
5754 struct oftable
*table
= &rule
->ofproto
->tables
[rule
->table_id
];
5755 struct eviction_group
*evg
= rule
->eviction_group
;
5757 rule
->eviction_group
= NULL
;
5758 heap_remove(&evg
->rules
, &rule
->evg_node
);
5759 if (heap_is_empty(&evg
->rules
)) {
5760 eviction_group_destroy(table
, evg
);
5762 eviction_group_resized(table
, evg
);
5767 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
5768 * returns the hash value. */
5770 eviction_group_hash_rule(struct rule
*rule
)
5772 struct oftable
*table
= &rule
->ofproto
->tables
[rule
->table_id
];
5773 const struct mf_subfield
*sf
;
5777 hash
= table
->eviction_group_id_basis
;
5778 miniflow_expand(&rule
->cr
.match
.flow
, &flow
);
5779 for (sf
= table
->eviction_fields
;
5780 sf
< &table
->eviction_fields
[table
->n_eviction_fields
];
5783 if (mf_are_prereqs_ok(sf
->field
, &flow
)) {
5784 union mf_value value
;
5786 mf_get_value(sf
->field
, &flow
, &value
);
5788 bitwise_zero(&value
, sf
->field
->n_bytes
, 0, sf
->ofs
);
5790 if (sf
->ofs
+ sf
->n_bits
< sf
->field
->n_bytes
* 8) {
5791 unsigned int start
= sf
->ofs
+ sf
->n_bits
;
5792 bitwise_zero(&value
, sf
->field
->n_bytes
, start
,
5793 sf
->field
->n_bytes
* 8 - start
);
5795 hash
= hash_bytes(&value
, sf
->field
->n_bytes
, hash
);
5797 hash
= hash_int(hash
, 0);
5804 /* Returns an eviction group within 'table' with the given 'id', creating one
5806 static struct eviction_group
*
5807 eviction_group_find(struct oftable
*table
, uint32_t id
)
5809 struct eviction_group
*evg
;
5811 HMAP_FOR_EACH_WITH_HASH (evg
, id_node
, id
, &table
->eviction_groups_by_id
) {
5815 evg
= xmalloc(sizeof *evg
);
5816 hmap_insert(&table
->eviction_groups_by_id
, &evg
->id_node
, id
);
5817 heap_insert(&table
->eviction_groups_by_size
, &evg
->size_node
,
5818 eviction_group_priority(0));
5819 heap_init(&evg
->rules
);
5824 /* Returns an eviction priority for 'rule'. The return value should be
5825 * interpreted so that higher priorities make a rule more attractive candidates
5828 rule_eviction_priority(struct rule
*rule
)
5830 long long int hard_expiration
;
5831 long long int idle_expiration
;
5832 long long int expiration
;
5833 uint32_t expiration_offset
;
5835 /* Calculate time of expiration. */
5836 ovs_mutex_lock(&rule
->timeout_mutex
);
5837 hard_expiration
= (rule
->hard_timeout
5838 ? rule
->modified
+ rule
->hard_timeout
* 1000
5840 idle_expiration
= (rule
->idle_timeout
5841 ? rule
->used
+ rule
->idle_timeout
* 1000
5843 expiration
= MIN(hard_expiration
, idle_expiration
);
5844 ovs_mutex_unlock(&rule
->timeout_mutex
);
5845 if (expiration
== LLONG_MAX
) {
5849 /* Calculate the time of expiration as a number of (approximate) seconds
5850 * after program startup.
5852 * This should work OK for program runs that last UINT32_MAX seconds or
5853 * less. Therefore, please restart OVS at least once every 136 years. */
5854 expiration_offset
= (expiration
>> 10) - (time_boot_msec() >> 10);
5856 /* Invert the expiration offset because we're using a max-heap. */
5857 return UINT32_MAX
- expiration_offset
;
5860 /* Adds 'rule' to an appropriate eviction group for its oftable's
5861 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
5862 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
5865 * The caller must ensure that 'rule' is not already in an eviction group. */
5867 eviction_group_add_rule(struct rule
*rule
)
5869 struct ofproto
*ofproto
= rule
->ofproto
;
5870 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
5873 ovs_mutex_lock(&rule
->timeout_mutex
);
5874 has_timeout
= rule
->hard_timeout
|| rule
->idle_timeout
;
5875 ovs_mutex_unlock(&rule
->timeout_mutex
);
5877 if (table
->eviction_fields
&& has_timeout
) {
5878 struct eviction_group
*evg
;
5880 evg
= eviction_group_find(table
, eviction_group_hash_rule(rule
));
5882 rule
->eviction_group
= evg
;
5883 heap_insert(&evg
->rules
, &rule
->evg_node
,
5884 rule_eviction_priority(rule
));
5885 eviction_group_resized(table
, evg
);
5891 /* Initializes 'table'. */
5893 oftable_init(struct oftable
*table
)
5895 memset(table
, 0, sizeof *table
);
5896 classifier_init(&table
->cls
);
5897 table
->max_flows
= UINT_MAX
;
5900 /* Destroys 'table', including its classifier and eviction groups.
5902 * The caller is responsible for freeing 'table' itself. */
5904 oftable_destroy(struct oftable
*table
)
5906 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
5907 ovs_assert(classifier_is_empty(&table
->cls
));
5908 ovs_rwlock_unlock(&table
->cls
.rwlock
);
5909 oftable_disable_eviction(table
);
5910 classifier_destroy(&table
->cls
);
5914 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
5915 * string, then 'table' will use its default name.
5917 * This only affects the name exposed for a table exposed through the OpenFlow
5918 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
5920 oftable_set_name(struct oftable
*table
, const char *name
)
5922 if (name
&& name
[0]) {
5923 int len
= strnlen(name
, OFP_MAX_TABLE_NAME_LEN
);
5924 if (!table
->name
|| strncmp(name
, table
->name
, len
)) {
5926 table
->name
= xmemdup0(name
, len
);
5934 /* oftables support a choice of two policies when adding a rule would cause the
5935 * number of flows in the table to exceed the configured maximum number: either
5936 * they can refuse to add the new flow or they can evict some existing flow.
5937 * This function configures the former policy on 'table'. */
5939 oftable_disable_eviction(struct oftable
*table
)
5941 if (table
->eviction_fields
) {
5942 struct eviction_group
*evg
, *next
;
5944 HMAP_FOR_EACH_SAFE (evg
, next
, id_node
,
5945 &table
->eviction_groups_by_id
) {
5946 eviction_group_destroy(table
, evg
);
5948 hmap_destroy(&table
->eviction_groups_by_id
);
5949 heap_destroy(&table
->eviction_groups_by_size
);
5951 free(table
->eviction_fields
);
5952 table
->eviction_fields
= NULL
;
5953 table
->n_eviction_fields
= 0;
5957 /* oftables support a choice of two policies when adding a rule would cause the
5958 * number of flows in the table to exceed the configured maximum number: either
5959 * they can refuse to add the new flow or they can evict some existing flow.
5960 * This function configures the latter policy on 'table', with fairness based
5961 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
5962 * 'n_fields' as 0 disables fairness.) */
5964 oftable_enable_eviction(struct oftable
*table
,
5965 const struct mf_subfield
*fields
, size_t n_fields
)
5967 struct cls_cursor cursor
;
5970 if (table
->eviction_fields
5971 && n_fields
== table
->n_eviction_fields
5973 || !memcmp(fields
, table
->eviction_fields
,
5974 n_fields
* sizeof *fields
))) {
5979 oftable_disable_eviction(table
);
5981 table
->n_eviction_fields
= n_fields
;
5982 table
->eviction_fields
= xmemdup(fields
, n_fields
* sizeof *fields
);
5984 table
->eviction_group_id_basis
= random_uint32();
5985 hmap_init(&table
->eviction_groups_by_id
);
5986 heap_init(&table
->eviction_groups_by_size
);
5988 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
5989 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
5990 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
5991 eviction_group_add_rule(rule
);
5993 ovs_rwlock_unlock(&table
->cls
.rwlock
);
5996 /* Removes 'rule' from the oftable that contains it. */
5998 oftable_remove_rule__(struct ofproto
*ofproto
, struct classifier
*cls
,
6000 OVS_REQ_WRLOCK(cls
->rwlock
) OVS_RELEASES(rule
->evict
)
6002 classifier_remove(cls
, &rule
->cr
);
6003 cookies_remove(ofproto
, rule
);
6004 eviction_group_remove_rule(rule
);
6005 ovs_mutex_lock(&ofproto
->expirable_mutex
);
6006 if (!list_is_empty(&rule
->expirable
)) {
6007 list_remove(&rule
->expirable
);
6009 ovs_mutex_unlock(&ofproto
->expirable_mutex
);
6010 if (!list_is_empty(&rule
->meter_list_node
)) {
6011 list_remove(&rule
->meter_list_node
);
6012 list_init(&rule
->meter_list_node
);
6014 ovs_rwlock_unlock(&rule
->evict
);
6018 oftable_remove_rule(struct rule
*rule
)
6020 struct ofproto
*ofproto
= rule
->ofproto
;
6021 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
6023 ovs_rwlock_wrlock(&table
->cls
.rwlock
);
6024 oftable_remove_rule__(ofproto
, &table
->cls
, rule
);
6025 ovs_rwlock_unlock(&table
->cls
.rwlock
);
6028 /* Inserts 'rule' into its oftable, which must not already contain any rule for
6029 * the same cls_rule. */
6031 oftable_insert_rule(struct rule
*rule
)
6033 struct ofproto
*ofproto
= rule
->ofproto
;
6034 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
6037 ovs_mutex_lock(&rule
->timeout_mutex
);
6038 may_expire
= rule
->hard_timeout
|| rule
->idle_timeout
;
6039 ovs_mutex_unlock(&rule
->timeout_mutex
);
6042 ovs_mutex_lock(&ofproto
->expirable_mutex
);
6043 list_insert(&ofproto
->expirable
, &rule
->expirable
);
6044 ovs_mutex_unlock(&ofproto
->expirable_mutex
);
6046 cookies_insert(ofproto
, rule
);
6047 if (rule
->meter_id
) {
6048 struct meter
*meter
= ofproto
->meters
[rule
->meter_id
];
6049 list_insert(&meter
->rules
, &rule
->meter_list_node
);
6051 ovs_rwlock_wrlock(&table
->cls
.rwlock
);
6052 classifier_insert(&table
->cls
, &rule
->cr
);
6053 ovs_rwlock_unlock(&table
->cls
.rwlock
);
6054 eviction_group_add_rule(rule
);
6057 /* unixctl commands. */
6060 ofproto_lookup(const char *name
)
6062 struct ofproto
*ofproto
;
6064 HMAP_FOR_EACH_WITH_HASH (ofproto
, hmap_node
, hash_string(name
, 0),
6066 if (!strcmp(ofproto
->name
, name
)) {
6074 ofproto_unixctl_list(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
6075 const char *argv
[] OVS_UNUSED
, void *aux OVS_UNUSED
)
6077 struct ofproto
*ofproto
;
6081 HMAP_FOR_EACH (ofproto
, hmap_node
, &all_ofprotos
) {
6082 ds_put_format(&results
, "%s\n", ofproto
->name
);
6084 unixctl_command_reply(conn
, ds_cstr(&results
));
6085 ds_destroy(&results
);
6089 ofproto_unixctl_init(void)
6091 static bool registered
;
6097 unixctl_command_register("ofproto/list", "", 0, 0,
6098 ofproto_unixctl_list
, NULL
);
6101 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6103 * This is deprecated. It is only for compatibility with broken device drivers
6104 * in old versions of Linux that do not properly support VLANs when VLAN
6105 * devices are not used. When broken device drivers are no longer in
6106 * widespread use, we will delete these interfaces. */
6108 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
6109 * (exactly) by an OpenFlow rule in 'ofproto'. */
6111 ofproto_get_vlan_usage(struct ofproto
*ofproto
, unsigned long int *vlan_bitmap
)
6113 const struct oftable
*oftable
;
6115 free(ofproto
->vlan_bitmap
);
6116 ofproto
->vlan_bitmap
= bitmap_allocate(4096);
6117 ofproto
->vlans_changed
= false;
6119 OFPROTO_FOR_EACH_TABLE (oftable
, ofproto
) {
6120 const struct cls_table
*table
;
6122 HMAP_FOR_EACH (table
, hmap_node
, &oftable
->cls
.tables
) {
6123 if (minimask_get_vid_mask(&table
->mask
) == VLAN_VID_MASK
) {
6124 const struct cls_rule
*rule
;
6126 HMAP_FOR_EACH (rule
, hmap_node
, &table
->rules
) {
6127 uint16_t vid
= miniflow_get_vid(&rule
->match
.flow
);
6128 bitmap_set1(vlan_bitmap
, vid
);
6129 bitmap_set1(ofproto
->vlan_bitmap
, vid
);
6136 /* Returns true if new VLANs have come into use by the flow table since the
6137 * last call to ofproto_get_vlan_usage().
6139 * We don't track when old VLANs stop being used. */
6141 ofproto_has_vlan_usage_changed(const struct ofproto
*ofproto
)
6143 return ofproto
->vlans_changed
;
6146 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
6147 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
6148 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
6149 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
6150 * then the VLAN device is un-enslaved. */
6152 ofproto_port_set_realdev(struct ofproto
*ofproto
, ofp_port_t vlandev_ofp_port
,
6153 ofp_port_t realdev_ofp_port
, int vid
)
6155 struct ofport
*ofport
;
6158 ovs_assert(vlandev_ofp_port
!= realdev_ofp_port
);
6160 ofport
= ofproto_get_port(ofproto
, vlandev_ofp_port
);
6162 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16
,
6163 ofproto
->name
, vlandev_ofp_port
);
6167 if (!ofproto
->ofproto_class
->set_realdev
) {
6168 if (!vlandev_ofp_port
) {
6171 VLOG_WARN("%s: vlan splinters not supported", ofproto
->name
);
6175 error
= ofproto
->ofproto_class
->set_realdev(ofport
, realdev_ofp_port
, vid
);
6177 VLOG_WARN("%s: setting realdev on port %"PRIu16
" (%s) failed (%s)",
6178 ofproto
->name
, vlandev_ofp_port
,
6179 netdev_get_name(ofport
->netdev
), ovs_strerror(error
));