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
{
82 /* A single OpenFlow request can execute any number of operations. The
83 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
84 * ofconn to which an error reply should be sent if necessary.
86 * ofproto initiates some operations internally. These operations are still
87 * assigned to groups but will not have an associated ofconn. */
89 struct ofproto
*ofproto
; /* Owning ofproto. */
90 struct list ofproto_node
; /* In ofproto's "pending" list. */
91 struct list ops
; /* List of "struct ofoperation"s. */
92 int n_running
; /* Number of ops still pending. */
94 /* Data needed to send OpenFlow reply on failure or to send a buffered
97 * If list_is_empty(ofconn_node) then this ofopgroup never had an
98 * associated ofconn or its ofconn's connection dropped after it initiated
99 * the operation. In the latter case 'ofconn' is a wild pointer that
100 * refers to freed memory, so the 'ofconn' member must be used only if
101 * !list_is_empty(ofconn_node).
103 struct list ofconn_node
; /* In ofconn's list of pending opgroups. */
104 struct ofconn
*ofconn
; /* ofconn for reply (but see note above). */
105 struct ofp_header
*request
; /* Original request (truncated at 64 bytes). */
106 uint32_t buffer_id
; /* Buffer id from original request. */
109 static struct ofopgroup
*ofopgroup_create_unattached(struct ofproto
*);
110 static struct ofopgroup
*ofopgroup_create(struct ofproto
*, struct ofconn
*,
111 const struct ofp_header
*,
113 static void ofopgroup_submit(struct ofopgroup
*);
114 static void ofopgroup_complete(struct ofopgroup
*);
116 /* A single flow table operation. */
118 struct ofopgroup
*group
; /* Owning group. */
119 struct list group_node
; /* In ofopgroup's "ops" list. */
120 struct hmap_node hmap_node
; /* In ofproto's "deletions" hmap. */
121 struct rule
*rule
; /* Rule being operated upon. */
122 enum ofoperation_type type
; /* Type of operation. */
124 /* OFOPERATION_ADD. */
125 struct rule
*victim
; /* Rule being replaced, if any.. */
127 /* OFOPERATION_MODIFY: The old actions, if the actions are changing. */
128 struct ofpact
*ofpacts
;
132 /* OFOPERATION_DELETE. */
133 enum ofp_flow_removed_reason reason
; /* Reason flow was removed. */
135 ovs_be64 flow_cookie
; /* Rule's old flow cookie. */
136 enum ofperr error
; /* 0 if no error. */
139 static struct ofoperation
*ofoperation_create(struct ofopgroup
*,
141 enum ofoperation_type
,
142 enum ofp_flow_removed_reason
);
143 static void ofoperation_destroy(struct ofoperation
*);
146 static void oftable_init(struct oftable
*);
147 static void oftable_destroy(struct oftable
*);
149 static void oftable_set_name(struct oftable
*, const char *name
);
151 static void oftable_disable_eviction(struct oftable
*);
152 static void oftable_enable_eviction(struct oftable
*,
153 const struct mf_subfield
*fields
,
156 static void oftable_remove_rule(struct rule
*rule
) OVS_RELEASES(rule
->evict
);
157 static void oftable_remove_rule__(struct ofproto
*ofproto
,
158 struct classifier
*cls
, struct rule
*rule
)
159 OVS_REQ_WRLOCK(cls
->rwlock
) OVS_RELEASES(rule
->evict
);
160 static struct rule
*oftable_replace_rule(struct rule
*);
161 static void oftable_substitute_rule(struct rule
*old
, struct rule
*new);
163 /* A set of rules within a single OpenFlow table (oftable) that have the same
164 * values for the oftable's eviction_fields. A rule to be evicted, when one is
165 * needed, is taken from the eviction group that contains the greatest number
168 * An oftable owns any number of eviction groups, each of which contains any
171 * Membership in an eviction group is imprecise, based on the hash of the
172 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
173 * That is, if two rules have different eviction_fields, but those
174 * eviction_fields hash to the same value, then they will belong to the same
175 * eviction_group anyway.
177 * (When eviction is not enabled on an oftable, we don't track any eviction
178 * groups, to save time and space.) */
179 struct eviction_group
{
180 struct hmap_node id_node
; /* In oftable's "eviction_groups_by_id". */
181 struct heap_node size_node
; /* In oftable's "eviction_groups_by_size". */
182 struct heap rules
; /* Contains "struct rule"s. */
185 static bool choose_rule_to_evict(struct oftable
*table
, struct rule
**rulep
)
186 OVS_TRY_WRLOCK(true, (*rulep
)->evict
);
187 static void ofproto_evict(struct ofproto
*);
188 static uint32_t rule_eviction_priority(struct rule
*);
191 static void ofport_destroy__(struct ofport
*);
192 static void ofport_destroy(struct ofport
*);
194 static void update_port(struct ofproto
*, const char *devname
);
195 static int init_ports(struct ofproto
*);
196 static void reinit_ports(struct ofproto
*);
199 static void ofproto_rule_destroy__(struct rule
*);
200 static void ofproto_rule_send_removed(struct rule
*, uint8_t reason
);
201 static bool rule_is_modifiable(const struct rule
*);
204 static enum ofperr
add_flow(struct ofproto
*, struct ofconn
*,
205 struct ofputil_flow_mod
*,
206 const struct ofp_header
*);
207 static void delete_flow__(struct rule
*rule
, struct ofopgroup
*,
208 enum ofp_flow_removed_reason
)
209 OVS_RELEASES(rule
->evict
);
210 static bool handle_openflow(struct ofconn
*, const struct ofpbuf
*);
211 static enum ofperr
handle_flow_mod__(struct ofproto
*, struct ofconn
*,
212 struct ofputil_flow_mod
*,
213 const struct ofp_header
*);
214 static void calc_duration(long long int start
, long long int now
,
215 uint32_t *sec
, uint32_t *nsec
);
218 static uint64_t pick_datapath_id(const struct ofproto
*);
219 static uint64_t pick_fallback_dpid(void);
220 static void ofproto_destroy__(struct ofproto
*);
221 static void update_mtu(struct ofproto
*, struct ofport
*);
222 static void meter_delete(struct ofproto
*, uint32_t first
, uint32_t last
);
225 static void ofproto_unixctl_init(void);
227 /* All registered ofproto classes, in probe order. */
228 static const struct ofproto_class
**ofproto_classes
;
229 static size_t n_ofproto_classes
;
230 static size_t allocated_ofproto_classes
;
232 unsigned flow_eviction_threshold
= OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT
;
233 unsigned n_handler_threads
;
234 enum ofproto_flow_miss_model flow_miss_model
= OFPROTO_HANDLE_MISS_AUTO
;
236 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
237 static struct hmap all_ofprotos
= HMAP_INITIALIZER(&all_ofprotos
);
239 /* Initial mappings of port to OpenFlow number mappings. */
240 static struct shash init_ofp_ports
= SHASH_INITIALIZER(&init_ofp_ports
);
242 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
244 /* The default value of true waits for flow restore. */
245 static bool flow_restore_wait
= true;
247 /* Must be called to initialize the ofproto library.
249 * The caller may pass in 'iface_hints', which contains an shash of
250 * "iface_hint" elements indexed by the interface's name. The provider
251 * may use these hints to describe the startup configuration in order to
252 * reinitialize its state. The caller owns the provided data, so a
253 * provider will make copies of anything required. An ofproto provider
254 * will remove any existing state that is not described by the hint, and
255 * may choose to remove it all. */
257 ofproto_init(const struct shash
*iface_hints
)
259 struct shash_node
*node
;
262 ofproto_class_register(&ofproto_dpif_class
);
264 /* Make a local copy, since we don't own 'iface_hints' elements. */
265 SHASH_FOR_EACH(node
, iface_hints
) {
266 const struct iface_hint
*orig_hint
= node
->data
;
267 struct iface_hint
*new_hint
= xmalloc(sizeof *new_hint
);
268 const char *br_type
= ofproto_normalize_type(orig_hint
->br_type
);
270 new_hint
->br_name
= xstrdup(orig_hint
->br_name
);
271 new_hint
->br_type
= xstrdup(br_type
);
272 new_hint
->ofp_port
= orig_hint
->ofp_port
;
274 shash_add(&init_ofp_ports
, node
->name
, new_hint
);
277 for (i
= 0; i
< n_ofproto_classes
; i
++) {
278 ofproto_classes
[i
]->init(&init_ofp_ports
);
282 /* 'type' should be a normalized datapath type, as returned by
283 * ofproto_normalize_type(). Returns the corresponding ofproto_class
284 * structure, or a null pointer if there is none registered for 'type'. */
285 static const struct ofproto_class
*
286 ofproto_class_find__(const char *type
)
290 for (i
= 0; i
< n_ofproto_classes
; i
++) {
291 const struct ofproto_class
*class = ofproto_classes
[i
];
296 class->enumerate_types(&types
);
297 found
= sset_contains(&types
, type
);
298 sset_destroy(&types
);
304 VLOG_WARN("unknown datapath type %s", type
);
308 /* Registers a new ofproto class. After successful registration, new ofprotos
309 * of that type can be created using ofproto_create(). */
311 ofproto_class_register(const struct ofproto_class
*new_class
)
315 for (i
= 0; i
< n_ofproto_classes
; i
++) {
316 if (ofproto_classes
[i
] == new_class
) {
321 if (n_ofproto_classes
>= allocated_ofproto_classes
) {
322 ofproto_classes
= x2nrealloc(ofproto_classes
,
323 &allocated_ofproto_classes
,
324 sizeof *ofproto_classes
);
326 ofproto_classes
[n_ofproto_classes
++] = new_class
;
330 /* Unregisters a datapath provider. 'type' must have been previously
331 * registered and not currently be in use by any ofprotos. After
332 * unregistration new datapaths of that type cannot be opened using
333 * ofproto_create(). */
335 ofproto_class_unregister(const struct ofproto_class
*class)
339 for (i
= 0; i
< n_ofproto_classes
; i
++) {
340 if (ofproto_classes
[i
] == class) {
341 for (i
++; i
< n_ofproto_classes
; i
++) {
342 ofproto_classes
[i
- 1] = ofproto_classes
[i
];
348 VLOG_WARN("attempted to unregister an ofproto class that is not "
353 /* Clears 'types' and enumerates all registered ofproto types into it. The
354 * caller must first initialize the sset. */
356 ofproto_enumerate_types(struct sset
*types
)
360 for (i
= 0; i
< n_ofproto_classes
; i
++) {
361 ofproto_classes
[i
]->enumerate_types(types
);
365 /* Returns the fully spelled out name for the given ofproto 'type'.
367 * Normalized type string can be compared with strcmp(). Unnormalized type
368 * string might be the same even if they have different spellings. */
370 ofproto_normalize_type(const char *type
)
372 return type
&& type
[0] ? type
: "system";
375 /* Clears 'names' and enumerates the names of all known created ofprotos with
376 * the given 'type'. The caller must first initialize the sset. Returns 0 if
377 * successful, otherwise a positive errno value.
379 * Some kinds of datapaths might not be practically enumerable. This is not
380 * considered an error. */
382 ofproto_enumerate_names(const char *type
, struct sset
*names
)
384 const struct ofproto_class
*class = ofproto_class_find__(type
);
385 return class ? class->enumerate_names(type
, names
) : EAFNOSUPPORT
;
389 ofproto_create(const char *datapath_name
, const char *datapath_type
,
390 struct ofproto
**ofprotop
)
392 const struct ofproto_class
*class;
393 struct ofproto
*ofproto
;
399 ofproto_unixctl_init();
401 datapath_type
= ofproto_normalize_type(datapath_type
);
402 class = ofproto_class_find__(datapath_type
);
404 VLOG_WARN("could not create datapath %s of unknown type %s",
405 datapath_name
, datapath_type
);
409 ofproto
= class->alloc();
411 VLOG_ERR("failed to allocate datapath %s of type %s",
412 datapath_name
, datapath_type
);
417 memset(ofproto
, 0, sizeof *ofproto
);
418 ofproto
->ofproto_class
= class;
419 ofproto
->name
= xstrdup(datapath_name
);
420 ofproto
->type
= xstrdup(datapath_type
);
421 hmap_insert(&all_ofprotos
, &ofproto
->hmap_node
,
422 hash_string(ofproto
->name
, 0));
423 ofproto
->datapath_id
= 0;
424 ofproto
->forward_bpdu
= false;
425 ofproto
->fallback_dpid
= pick_fallback_dpid();
426 ofproto
->mfr_desc
= NULL
;
427 ofproto
->hw_desc
= NULL
;
428 ofproto
->sw_desc
= NULL
;
429 ofproto
->serial_desc
= NULL
;
430 ofproto
->dp_desc
= NULL
;
431 ofproto
->frag_handling
= OFPC_FRAG_NORMAL
;
432 hmap_init(&ofproto
->ports
);
433 shash_init(&ofproto
->port_by_name
);
434 simap_init(&ofproto
->ofp_requests
);
435 ofproto
->max_ports
= OFPP_MAX
;
436 ofproto
->tables
= NULL
;
437 ofproto
->n_tables
= 0;
438 hindex_init(&ofproto
->cookies
);
439 list_init(&ofproto
->expirable
);
440 ovs_mutex_init_recursive(&ofproto
->expirable_mutex
);
441 ofproto
->connmgr
= connmgr_create(ofproto
, datapath_name
, datapath_name
);
442 ofproto
->state
= S_OPENFLOW
;
443 list_init(&ofproto
->pending
);
444 ofproto
->n_pending
= 0;
445 hmap_init(&ofproto
->deletions
);
446 ofproto
->n_add
= ofproto
->n_delete
= ofproto
->n_modify
= 0;
447 ofproto
->first_op
= ofproto
->last_op
= LLONG_MIN
;
448 ofproto
->next_op_report
= LLONG_MAX
;
449 ofproto
->op_backoff
= LLONG_MIN
;
450 ofproto
->vlan_bitmap
= NULL
;
451 ofproto
->vlans_changed
= false;
452 ofproto
->min_mtu
= INT_MAX
;
454 error
= ofproto
->ofproto_class
->construct(ofproto
);
456 VLOG_ERR("failed to open datapath %s: %s",
457 datapath_name
, ovs_strerror(error
));
458 ofproto_destroy__(ofproto
);
462 /* The "max_ports" member should have been set by ->construct(ofproto).
463 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
464 ofproto
->ofp_port_ids
= bitmap_allocate(ofp_to_u16(ofproto
->max_ports
));
465 bitmap_set1(ofproto
->ofp_port_ids
, 0);
467 /* Check that hidden tables, if any, are at the end. */
468 ovs_assert(ofproto
->n_tables
);
469 for (i
= 0; i
+ 1 < ofproto
->n_tables
; i
++) {
470 enum oftable_flags flags
= ofproto
->tables
[i
].flags
;
471 enum oftable_flags next_flags
= ofproto
->tables
[i
+ 1].flags
;
473 ovs_assert(!(flags
& OFTABLE_HIDDEN
) || next_flags
& OFTABLE_HIDDEN
);
476 ofproto
->datapath_id
= pick_datapath_id(ofproto
);
479 /* Initialize meters table. */
480 if (ofproto
->ofproto_class
->meter_get_features
) {
481 ofproto
->ofproto_class
->meter_get_features(ofproto
,
482 &ofproto
->meter_features
);
484 memset(&ofproto
->meter_features
, 0, sizeof ofproto
->meter_features
);
486 ofproto
->meters
= xzalloc((ofproto
->meter_features
.max_meters
+ 1)
487 * sizeof(struct meter
*));
493 /* Must be called (only) by an ofproto implementation in its constructor
494 * function. See the large comment on 'construct' in struct ofproto_class for
497 ofproto_init_tables(struct ofproto
*ofproto
, int n_tables
)
499 struct oftable
*table
;
501 ovs_assert(!ofproto
->n_tables
);
502 ovs_assert(n_tables
>= 1 && n_tables
<= 255);
504 ofproto
->n_tables
= n_tables
;
505 ofproto
->tables
= xmalloc(n_tables
* sizeof *ofproto
->tables
);
506 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
511 /* To be optionally called (only) by an ofproto implementation in its
512 * constructor function. See the large comment on 'construct' in struct
513 * ofproto_class for details.
515 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
516 * will then ensure that actions passed into the ofproto implementation will
517 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
518 * function is not called, there will be no such restriction.
520 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
521 * the 'max_ports' restriction. */
523 ofproto_init_max_ports(struct ofproto
*ofproto
, ofp_port_t max_ports
)
525 ovs_assert(ofp_to_u16(max_ports
) <= ofp_to_u16(OFPP_MAX
));
526 ofproto
->max_ports
= max_ports
;
530 ofproto_get_datapath_id(const struct ofproto
*ofproto
)
532 return ofproto
->datapath_id
;
536 ofproto_set_datapath_id(struct ofproto
*p
, uint64_t datapath_id
)
538 uint64_t old_dpid
= p
->datapath_id
;
539 p
->datapath_id
= datapath_id
? datapath_id
: pick_datapath_id(p
);
540 if (p
->datapath_id
!= old_dpid
) {
541 /* Force all active connections to reconnect, since there is no way to
542 * notify a controller that the datapath ID has changed. */
543 ofproto_reconnect_controllers(p
);
548 ofproto_set_controllers(struct ofproto
*p
,
549 const struct ofproto_controller
*controllers
,
550 size_t n_controllers
, uint32_t allowed_versions
)
552 connmgr_set_controllers(p
->connmgr
, controllers
, n_controllers
,
557 ofproto_set_fail_mode(struct ofproto
*p
, enum ofproto_fail_mode fail_mode
)
559 connmgr_set_fail_mode(p
->connmgr
, fail_mode
);
562 /* Drops the connections between 'ofproto' and all of its controllers, forcing
563 * them to reconnect. */
565 ofproto_reconnect_controllers(struct ofproto
*ofproto
)
567 connmgr_reconnect(ofproto
->connmgr
);
570 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
571 * in-band control should guarantee access, in the same way that in-band
572 * control guarantees access to OpenFlow controllers. */
574 ofproto_set_extra_in_band_remotes(struct ofproto
*ofproto
,
575 const struct sockaddr_in
*extras
, size_t n
)
577 connmgr_set_extra_in_band_remotes(ofproto
->connmgr
, extras
, n
);
580 /* Sets the OpenFlow queue used by flows set up by in-band control on
581 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
582 * flows will use the default queue. */
584 ofproto_set_in_band_queue(struct ofproto
*ofproto
, int queue_id
)
586 connmgr_set_in_band_queue(ofproto
->connmgr
, queue_id
);
589 /* Sets the number of flows at which eviction from the kernel flow table
592 ofproto_set_flow_eviction_threshold(unsigned threshold
)
594 flow_eviction_threshold
= MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN
,
598 /* Sets the path for handling flow misses. */
600 ofproto_set_flow_miss_model(unsigned model
)
602 flow_miss_model
= model
;
605 /* If forward_bpdu is true, the NORMAL action will forward frames with
606 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
607 * the NORMAL action will drop these frames. */
609 ofproto_set_forward_bpdu(struct ofproto
*ofproto
, bool forward_bpdu
)
611 bool old_val
= ofproto
->forward_bpdu
;
612 ofproto
->forward_bpdu
= forward_bpdu
;
613 if (old_val
!= ofproto
->forward_bpdu
) {
614 if (ofproto
->ofproto_class
->forward_bpdu_changed
) {
615 ofproto
->ofproto_class
->forward_bpdu_changed(ofproto
);
620 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
621 * 'idle_time', in seconds, and the maximum number of MAC table entries to
624 ofproto_set_mac_table_config(struct ofproto
*ofproto
, unsigned idle_time
,
627 if (ofproto
->ofproto_class
->set_mac_table_config
) {
628 ofproto
->ofproto_class
->set_mac_table_config(ofproto
, idle_time
,
633 /* Sets number of upcall handler threads. The default is
634 * (number of online cores - 1). */
636 ofproto_set_n_handler_threads(unsigned limit
)
639 n_handler_threads
= limit
;
641 n_handler_threads
= MAX(1, sysconf(_SC_NPROCESSORS_ONLN
) - 1);
646 ofproto_set_dp_desc(struct ofproto
*p
, const char *dp_desc
)
649 p
->dp_desc
= dp_desc
? xstrdup(dp_desc
) : NULL
;
653 ofproto_set_snoops(struct ofproto
*ofproto
, const struct sset
*snoops
)
655 return connmgr_set_snoops(ofproto
->connmgr
, snoops
);
659 ofproto_set_netflow(struct ofproto
*ofproto
,
660 const struct netflow_options
*nf_options
)
662 if (nf_options
&& sset_is_empty(&nf_options
->collectors
)) {
666 if (ofproto
->ofproto_class
->set_netflow
) {
667 return ofproto
->ofproto_class
->set_netflow(ofproto
, nf_options
);
669 return nf_options
? EOPNOTSUPP
: 0;
674 ofproto_set_sflow(struct ofproto
*ofproto
,
675 const struct ofproto_sflow_options
*oso
)
677 if (oso
&& sset_is_empty(&oso
->targets
)) {
681 if (ofproto
->ofproto_class
->set_sflow
) {
682 return ofproto
->ofproto_class
->set_sflow(ofproto
, oso
);
684 return oso
? EOPNOTSUPP
: 0;
689 ofproto_set_ipfix(struct ofproto
*ofproto
,
690 const struct ofproto_ipfix_bridge_exporter_options
*bo
,
691 const struct ofproto_ipfix_flow_exporter_options
*fo
,
694 if (ofproto
->ofproto_class
->set_ipfix
) {
695 return ofproto
->ofproto_class
->set_ipfix(ofproto
, bo
, fo
, n_fo
);
697 return (bo
|| fo
) ? EOPNOTSUPP
: 0;
702 ofproto_set_flow_restore_wait(bool flow_restore_wait_db
)
704 flow_restore_wait
= flow_restore_wait_db
;
708 ofproto_get_flow_restore_wait(void)
710 return flow_restore_wait
;
714 /* Spanning Tree Protocol (STP) configuration. */
716 /* Configures STP on 'ofproto' using the settings defined in 's'. If
717 * 's' is NULL, disables STP.
719 * Returns 0 if successful, otherwise a positive errno value. */
721 ofproto_set_stp(struct ofproto
*ofproto
,
722 const struct ofproto_stp_settings
*s
)
724 return (ofproto
->ofproto_class
->set_stp
725 ? ofproto
->ofproto_class
->set_stp(ofproto
, s
)
729 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
730 * 'enabled' member of 's' is false, then the other members are not
733 * Returns 0 if successful, otherwise a positive errno value. */
735 ofproto_get_stp_status(struct ofproto
*ofproto
,
736 struct ofproto_stp_status
*s
)
738 return (ofproto
->ofproto_class
->get_stp_status
739 ? ofproto
->ofproto_class
->get_stp_status(ofproto
, s
)
743 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
744 * in 's'. The caller is responsible for assigning STP port numbers
745 * (using the 'port_num' member in the range of 1 through 255, inclusive)
746 * and ensuring there are no duplicates. If the 's' is NULL, then STP
747 * is disabled on the port.
749 * Returns 0 if successful, otherwise a positive errno value.*/
751 ofproto_port_set_stp(struct ofproto
*ofproto
, ofp_port_t ofp_port
,
752 const struct ofproto_port_stp_settings
*s
)
754 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
756 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16
,
757 ofproto
->name
, ofp_port
);
761 return (ofproto
->ofproto_class
->set_stp_port
762 ? ofproto
->ofproto_class
->set_stp_port(ofport
, s
)
766 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
767 * 's'. If the 'enabled' member in 's' is false, then the other members
768 * are not meaningful.
770 * Returns 0 if successful, otherwise a positive errno value.*/
772 ofproto_port_get_stp_status(struct ofproto
*ofproto
, ofp_port_t ofp_port
,
773 struct ofproto_port_stp_status
*s
)
775 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
777 VLOG_WARN_RL(&rl
, "%s: cannot get STP status on nonexistent "
778 "port %"PRIu16
, ofproto
->name
, ofp_port
);
782 return (ofproto
->ofproto_class
->get_stp_port_status
783 ? ofproto
->ofproto_class
->get_stp_port_status(ofport
, s
)
787 /* Queue DSCP configuration. */
789 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
790 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
791 * to implement QoS. Instead, it is used to implement features which require
792 * knowledge of what queues exist on a port, and some basic information about
795 * Returns 0 if successful, otherwise a positive errno value. */
797 ofproto_port_set_queues(struct ofproto
*ofproto
, ofp_port_t ofp_port
,
798 const struct ofproto_port_queue
*queues
,
801 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
804 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16
,
805 ofproto
->name
, ofp_port
);
809 return (ofproto
->ofproto_class
->set_queues
810 ? ofproto
->ofproto_class
->set_queues(ofport
, queues
, n_queues
)
814 /* Connectivity Fault Management configuration. */
816 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
818 ofproto_port_clear_cfm(struct ofproto
*ofproto
, ofp_port_t ofp_port
)
820 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
821 if (ofport
&& ofproto
->ofproto_class
->set_cfm
) {
822 ofproto
->ofproto_class
->set_cfm(ofport
, NULL
);
826 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
827 * basic configuration from the configuration members in 'cfm', and the remote
828 * maintenance point ID from remote_mpid. Ignores the statistics members of
831 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
833 ofproto_port_set_cfm(struct ofproto
*ofproto
, ofp_port_t ofp_port
,
834 const struct cfm_settings
*s
)
836 struct ofport
*ofport
;
839 ofport
= ofproto_get_port(ofproto
, ofp_port
);
841 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16
,
842 ofproto
->name
, ofp_port
);
846 /* XXX: For configuration simplicity, we only support one remote_mpid
847 * outside of the CFM module. It's not clear if this is the correct long
848 * term solution or not. */
849 error
= (ofproto
->ofproto_class
->set_cfm
850 ? ofproto
->ofproto_class
->set_cfm(ofport
, s
)
853 VLOG_WARN("%s: CFM configuration on port %"PRIu16
" (%s) failed (%s)",
854 ofproto
->name
, ofp_port
, netdev_get_name(ofport
->netdev
),
855 ovs_strerror(error
));
859 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
860 * 'ofproto' does not have a port 'ofp_port'. */
862 ofproto_port_set_bfd(struct ofproto
*ofproto
, ofp_port_t ofp_port
,
863 const struct smap
*cfg
)
865 struct ofport
*ofport
;
868 ofport
= ofproto_get_port(ofproto
, ofp_port
);
870 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16
,
871 ofproto
->name
, ofp_port
);
875 error
= (ofproto
->ofproto_class
->set_bfd
876 ? ofproto
->ofproto_class
->set_bfd(ofport
, cfg
)
879 VLOG_WARN("%s: bfd configuration on port %"PRIu16
" (%s) failed (%s)",
880 ofproto
->name
, ofp_port
, netdev_get_name(ofport
->netdev
),
881 ovs_strerror(error
));
885 /* Populates 'status' with key value pairs indicating the status of the BFD
886 * session on 'ofp_port'. This information is intended to be populated in the
887 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
890 ofproto_port_get_bfd_status(struct ofproto
*ofproto
, ofp_port_t ofp_port
,
893 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
894 return (ofport
&& ofproto
->ofproto_class
->get_bfd_status
895 ? ofproto
->ofproto_class
->get_bfd_status(ofport
, status
)
899 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
900 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
901 * 0 if LACP partner information is not current (generally indicating a
902 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
904 ofproto_port_is_lacp_current(struct ofproto
*ofproto
, ofp_port_t ofp_port
)
906 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
907 return (ofport
&& ofproto
->ofproto_class
->port_is_lacp_current
908 ? ofproto
->ofproto_class
->port_is_lacp_current(ofport
)
914 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
915 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
916 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
917 * configuration plus, if there is more than one slave, a bonding
920 * If 'aux' is already registered then this function updates its configuration
921 * to 's'. Otherwise, this function registers a new bundle.
923 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
926 ofproto_bundle_register(struct ofproto
*ofproto
, void *aux
,
927 const struct ofproto_bundle_settings
*s
)
929 return (ofproto
->ofproto_class
->bundle_set
930 ? ofproto
->ofproto_class
->bundle_set(ofproto
, aux
, s
)
934 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
935 * If no such bundle has been registered, this has no effect. */
937 ofproto_bundle_unregister(struct ofproto
*ofproto
, void *aux
)
939 return ofproto_bundle_register(ofproto
, aux
, NULL
);
943 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
944 * If 'aux' is already registered then this function updates its configuration
945 * to 's'. Otherwise, this function registers a new mirror. */
947 ofproto_mirror_register(struct ofproto
*ofproto
, void *aux
,
948 const struct ofproto_mirror_settings
*s
)
950 return (ofproto
->ofproto_class
->mirror_set
951 ? ofproto
->ofproto_class
->mirror_set(ofproto
, aux
, s
)
955 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
956 * If no mirror has been registered, this has no effect. */
958 ofproto_mirror_unregister(struct ofproto
*ofproto
, void *aux
)
960 return ofproto_mirror_register(ofproto
, aux
, NULL
);
963 /* Retrieves statistics from mirror associated with client data pointer
964 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
965 * 'bytes', respectively. If a particular counters is not supported,
966 * the appropriate argument is set to UINT64_MAX. */
968 ofproto_mirror_get_stats(struct ofproto
*ofproto
, void *aux
,
969 uint64_t *packets
, uint64_t *bytes
)
971 if (!ofproto
->ofproto_class
->mirror_get_stats
) {
972 *packets
= *bytes
= UINT64_MAX
;
976 return ofproto
->ofproto_class
->mirror_get_stats(ofproto
, aux
,
980 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
981 * which all packets are flooded, instead of using MAC learning. If
982 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
984 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
987 ofproto_set_flood_vlans(struct ofproto
*ofproto
, unsigned long *flood_vlans
)
989 return (ofproto
->ofproto_class
->set_flood_vlans
990 ? ofproto
->ofproto_class
->set_flood_vlans(ofproto
, flood_vlans
)
994 /* Returns true if 'aux' is a registered bundle that is currently in use as the
995 * output for a mirror. */
997 ofproto_is_mirror_output_bundle(const struct ofproto
*ofproto
, void *aux
)
999 return (ofproto
->ofproto_class
->is_mirror_output_bundle
1000 ? ofproto
->ofproto_class
->is_mirror_output_bundle(ofproto
, aux
)
1004 /* Configuration of OpenFlow tables. */
1006 /* Returns the number of OpenFlow tables in 'ofproto'. */
1008 ofproto_get_n_tables(const struct ofproto
*ofproto
)
1010 return ofproto
->n_tables
;
1013 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1014 * settings from 's'. 'table_id' must be in the range 0 through the number of
1015 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1017 * For read-only tables, only the name may be configured. */
1019 ofproto_configure_table(struct ofproto
*ofproto
, int table_id
,
1020 const struct ofproto_table_settings
*s
)
1022 struct oftable
*table
;
1024 ovs_assert(table_id
>= 0 && table_id
< ofproto
->n_tables
);
1025 table
= &ofproto
->tables
[table_id
];
1027 oftable_set_name(table
, s
->name
);
1029 if (table
->flags
& OFTABLE_READONLY
) {
1034 oftable_enable_eviction(table
, s
->groups
, s
->n_groups
);
1036 oftable_disable_eviction(table
);
1039 table
->max_flows
= s
->max_flows
;
1040 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
1041 if (classifier_count(&table
->cls
) > table
->max_flows
1042 && table
->eviction_fields
) {
1043 /* 'table' contains more flows than allowed. We might not be able to
1044 * evict them right away because of the asynchronous nature of flow
1045 * table changes. Schedule eviction for later. */
1046 switch (ofproto
->state
) {
1048 ofproto
->state
= S_EVICT
;
1052 /* We're already deleting flows, nothing more to do. */
1056 ovs_rwlock_unlock(&table
->cls
.rwlock
);
1060 ofproto_has_snoops(const struct ofproto
*ofproto
)
1062 return connmgr_has_snoops(ofproto
->connmgr
);
1066 ofproto_get_snoops(const struct ofproto
*ofproto
, struct sset
*snoops
)
1068 connmgr_get_snoops(ofproto
->connmgr
, snoops
);
1072 ofproto_flush__(struct ofproto
*ofproto
)
1074 struct ofopgroup
*group
;
1075 struct oftable
*table
;
1077 if (ofproto
->ofproto_class
->flush
) {
1078 ofproto
->ofproto_class
->flush(ofproto
);
1081 group
= ofopgroup_create_unattached(ofproto
);
1082 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
1083 struct rule
*rule
, *next_rule
;
1084 struct cls_cursor cursor
;
1086 if (table
->flags
& OFTABLE_HIDDEN
) {
1090 ovs_rwlock_wrlock(&table
->cls
.rwlock
);
1091 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
1092 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
1093 if (!rule
->pending
) {
1094 ofoperation_create(group
, rule
, OFOPERATION_DELETE
,
1096 ovs_rwlock_wrlock(&rule
->evict
);
1097 oftable_remove_rule__(ofproto
, &table
->cls
, rule
);
1098 ofproto
->ofproto_class
->rule_destruct(rule
);
1101 ovs_rwlock_unlock(&table
->cls
.rwlock
);
1103 ofopgroup_submit(group
);
1107 ofproto_destroy__(struct ofproto
*ofproto
)
1109 struct oftable
*table
;
1111 ovs_assert(list_is_empty(&ofproto
->pending
));
1112 ovs_assert(!ofproto
->n_pending
);
1114 if (ofproto
->meters
) {
1115 meter_delete(ofproto
, 1, ofproto
->meter_features
.max_meters
);
1116 free(ofproto
->meters
);
1119 connmgr_destroy(ofproto
->connmgr
);
1121 hmap_remove(&all_ofprotos
, &ofproto
->hmap_node
);
1122 free(ofproto
->name
);
1123 free(ofproto
->type
);
1124 free(ofproto
->mfr_desc
);
1125 free(ofproto
->hw_desc
);
1126 free(ofproto
->sw_desc
);
1127 free(ofproto
->serial_desc
);
1128 free(ofproto
->dp_desc
);
1129 hmap_destroy(&ofproto
->ports
);
1130 shash_destroy(&ofproto
->port_by_name
);
1131 bitmap_free(ofproto
->ofp_port_ids
);
1132 simap_destroy(&ofproto
->ofp_requests
);
1134 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
1135 oftable_destroy(table
);
1137 free(ofproto
->tables
);
1139 hmap_destroy(&ofproto
->deletions
);
1141 free(ofproto
->vlan_bitmap
);
1143 ovs_mutex_destroy(&ofproto
->expirable_mutex
);
1144 ofproto
->ofproto_class
->dealloc(ofproto
);
1148 ofproto_destroy(struct ofproto
*p
)
1150 struct ofport
*ofport
, *next_ofport
;
1157 HMAP_FOR_EACH_SAFE (ofport
, next_ofport
, hmap_node
, &p
->ports
) {
1158 ofport_destroy(ofport
);
1161 p
->ofproto_class
->destruct(p
);
1162 ofproto_destroy__(p
);
1165 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1166 * kernel datapath, for example, this destroys the datapath in the kernel, and
1167 * with the netdev-based datapath, it tears down the data structures that
1168 * represent the datapath.
1170 * The datapath should not be currently open as an ofproto. */
1172 ofproto_delete(const char *name
, const char *type
)
1174 const struct ofproto_class
*class = ofproto_class_find__(type
);
1175 return (!class ? EAFNOSUPPORT
1176 : !class->del
? EACCES
1177 : class->del(type
, name
));
1181 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
1183 if (error
== ENOBUFS
) {
1184 reinit_ports(ofproto
);
1185 } else if (!error
) {
1186 update_port(ofproto
, devname
);
1192 ofproto_type_run(const char *datapath_type
)
1194 const struct ofproto_class
*class;
1197 datapath_type
= ofproto_normalize_type(datapath_type
);
1198 class = ofproto_class_find__(datapath_type
);
1200 error
= class->type_run
? class->type_run(datapath_type
) : 0;
1201 if (error
&& error
!= EAGAIN
) {
1202 VLOG_ERR_RL(&rl
, "%s: type_run failed (%s)",
1203 datapath_type
, ovs_strerror(error
));
1209 ofproto_type_run_fast(const char *datapath_type
)
1211 const struct ofproto_class
*class;
1214 datapath_type
= ofproto_normalize_type(datapath_type
);
1215 class = ofproto_class_find__(datapath_type
);
1217 error
= class->type_run_fast
? class->type_run_fast(datapath_type
) : 0;
1218 if (error
&& error
!= EAGAIN
) {
1219 VLOG_ERR_RL(&rl
, "%s: type_run_fast failed (%s)",
1220 datapath_type
, ovs_strerror(error
));
1226 ofproto_type_wait(const char *datapath_type
)
1228 const struct ofproto_class
*class;
1230 datapath_type
= ofproto_normalize_type(datapath_type
);
1231 class = ofproto_class_find__(datapath_type
);
1233 if (class->type_wait
) {
1234 class->type_wait(datapath_type
);
1239 ofproto_run(struct ofproto
*p
)
1241 struct sset changed_netdevs
;
1242 const char *changed_netdev
;
1243 struct ofport
*ofport
;
1246 error
= p
->ofproto_class
->run(p
);
1247 if (error
&& error
!= EAGAIN
) {
1248 VLOG_ERR_RL(&rl
, "%s: run failed (%s)", p
->name
, ovs_strerror(error
));
1251 if (p
->ofproto_class
->port_poll
) {
1254 while ((error
= p
->ofproto_class
->port_poll(p
, &devname
)) != EAGAIN
) {
1255 process_port_change(p
, error
, devname
);
1259 /* Update OpenFlow port status for any port whose netdev has changed.
1261 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1262 * destroyed, so it's not safe to update ports directly from the
1263 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1264 * need this two-phase approach. */
1265 sset_init(&changed_netdevs
);
1266 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1267 unsigned int change_seq
= netdev_change_seq(ofport
->netdev
);
1268 if (ofport
->change_seq
!= change_seq
) {
1269 ofport
->change_seq
= change_seq
;
1270 sset_add(&changed_netdevs
, netdev_get_name(ofport
->netdev
));
1273 SSET_FOR_EACH (changed_netdev
, &changed_netdevs
) {
1274 update_port(p
, changed_netdev
);
1276 sset_destroy(&changed_netdevs
);
1280 connmgr_run(p
->connmgr
, handle_openflow
);
1284 connmgr_run(p
->connmgr
, NULL
);
1286 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1287 p
->state
= S_OPENFLOW
;
1292 connmgr_run(p
->connmgr
, NULL
);
1294 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1295 connmgr_flushed(p
->connmgr
);
1296 p
->state
= S_OPENFLOW
;
1304 if (time_msec() >= p
->next_op_report
) {
1305 long long int ago
= (time_msec() - p
->first_op
) / 1000;
1306 long long int interval
= (p
->last_op
- p
->first_op
) / 1000;
1310 ds_put_format(&s
, "%d flow_mods ",
1311 p
->n_add
+ p
->n_delete
+ p
->n_modify
);
1312 if (interval
== ago
) {
1313 ds_put_format(&s
, "in the last %lld s", ago
);
1314 } else if (interval
) {
1315 ds_put_format(&s
, "in the %lld s starting %lld s ago",
1318 ds_put_format(&s
, "%lld s ago", ago
);
1321 ds_put_cstr(&s
, " (");
1323 ds_put_format(&s
, "%d adds, ", p
->n_add
);
1326 ds_put_format(&s
, "%d deletes, ", p
->n_delete
);
1329 ds_put_format(&s
, "%d modifications, ", p
->n_modify
);
1332 ds_put_char(&s
, ')');
1334 VLOG_INFO("%s: %s", p
->name
, ds_cstr(&s
));
1337 p
->n_add
= p
->n_delete
= p
->n_modify
= 0;
1338 p
->next_op_report
= LLONG_MAX
;
1344 /* Performs periodic activity required by 'ofproto' that needs to be done
1345 * with the least possible latency.
1347 * It makes sense to call this function a couple of times per poll loop, to
1348 * provide a significant performance boost on some benchmarks with the
1349 * ofproto-dpif implementation. */
1351 ofproto_run_fast(struct ofproto
*p
)
1355 error
= p
->ofproto_class
->run_fast
? p
->ofproto_class
->run_fast(p
) : 0;
1356 if (error
&& error
!= EAGAIN
) {
1357 VLOG_ERR_RL(&rl
, "%s: fastpath run failed (%s)",
1358 p
->name
, ovs_strerror(error
));
1364 ofproto_wait(struct ofproto
*p
)
1366 struct ofport
*ofport
;
1368 p
->ofproto_class
->wait(p
);
1369 if (p
->ofproto_class
->port_poll_wait
) {
1370 p
->ofproto_class
->port_poll_wait(p
);
1373 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1374 if (ofport
->change_seq
!= netdev_change_seq(ofport
->netdev
)) {
1375 poll_immediate_wake();
1381 connmgr_wait(p
->connmgr
, true);
1386 connmgr_wait(p
->connmgr
, false);
1387 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1388 poll_immediate_wake();
1395 ofproto_is_alive(const struct ofproto
*p
)
1397 return connmgr_has_controllers(p
->connmgr
);
1400 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1401 * memory_report(). */
1403 ofproto_get_memory_usage(const struct ofproto
*ofproto
, struct simap
*usage
)
1405 const struct oftable
*table
;
1406 unsigned int n_rules
;
1408 simap_increase(usage
, "ports", hmap_count(&ofproto
->ports
));
1409 simap_increase(usage
, "ops",
1410 ofproto
->n_pending
+ hmap_count(&ofproto
->deletions
));
1413 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
1414 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
1415 n_rules
+= classifier_count(&table
->cls
);
1416 ovs_rwlock_unlock(&table
->cls
.rwlock
);
1418 simap_increase(usage
, "rules", n_rules
);
1420 if (ofproto
->ofproto_class
->get_memory_usage
) {
1421 ofproto
->ofproto_class
->get_memory_usage(ofproto
, usage
);
1424 connmgr_get_memory_usage(ofproto
->connmgr
, usage
);
1428 ofproto_get_ofproto_controller_info(const struct ofproto
*ofproto
,
1431 connmgr_get_controller_info(ofproto
->connmgr
, info
);
1435 ofproto_free_ofproto_controller_info(struct shash
*info
)
1437 connmgr_free_controller_info(info
);
1440 /* Makes a deep copy of 'old' into 'port'. */
1442 ofproto_port_clone(struct ofproto_port
*port
, const struct ofproto_port
*old
)
1444 port
->name
= xstrdup(old
->name
);
1445 port
->type
= xstrdup(old
->type
);
1446 port
->ofp_port
= old
->ofp_port
;
1449 /* Frees memory allocated to members of 'ofproto_port'.
1451 * Do not call this function on an ofproto_port obtained from
1452 * ofproto_port_dump_next(): that function retains ownership of the data in the
1455 ofproto_port_destroy(struct ofproto_port
*ofproto_port
)
1457 free(ofproto_port
->name
);
1458 free(ofproto_port
->type
);
1461 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1463 * This function provides no status indication. An error status for the entire
1464 * dump operation is provided when it is completed by calling
1465 * ofproto_port_dump_done().
1468 ofproto_port_dump_start(struct ofproto_port_dump
*dump
,
1469 const struct ofproto
*ofproto
)
1471 dump
->ofproto
= ofproto
;
1472 dump
->error
= ofproto
->ofproto_class
->port_dump_start(ofproto
,
1476 /* Attempts to retrieve another port from 'dump', which must have been created
1477 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1478 * 'port' and returns true. On failure, returns false.
1480 * Failure might indicate an actual error or merely that the last port has been
1481 * dumped. An error status for the entire dump operation is provided when it
1482 * is completed by calling ofproto_port_dump_done().
1484 * The ofproto owns the data stored in 'port'. It will remain valid until at
1485 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1486 * ofproto_port_dump_done(). */
1488 ofproto_port_dump_next(struct ofproto_port_dump
*dump
,
1489 struct ofproto_port
*port
)
1491 const struct ofproto
*ofproto
= dump
->ofproto
;
1497 dump
->error
= ofproto
->ofproto_class
->port_dump_next(ofproto
, dump
->state
,
1500 ofproto
->ofproto_class
->port_dump_done(ofproto
, dump
->state
);
1506 /* Completes port table dump operation 'dump', which must have been created
1507 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1508 * error-free, otherwise a positive errno value describing the problem. */
1510 ofproto_port_dump_done(struct ofproto_port_dump
*dump
)
1512 const struct ofproto
*ofproto
= dump
->ofproto
;
1514 dump
->error
= ofproto
->ofproto_class
->port_dump_done(ofproto
,
1517 return dump
->error
== EOF
? 0 : dump
->error
;
1520 /* Returns the type to pass to netdev_open() when a datapath of type
1521 * 'datapath_type' has a port of type 'port_type', for a few special
1522 * cases when a netdev type differs from a port type. For example, when
1523 * using the userspace datapath, a port of type "internal" needs to be
1526 * Returns either 'type' itself or a string literal, which must not be
1529 ofproto_port_open_type(const char *datapath_type
, const char *port_type
)
1531 const struct ofproto_class
*class;
1533 datapath_type
= ofproto_normalize_type(datapath_type
);
1534 class = ofproto_class_find__(datapath_type
);
1539 return (class->port_open_type
1540 ? class->port_open_type(datapath_type
, port_type
)
1544 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1545 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1546 * the port's OpenFlow port number.
1548 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1549 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1550 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1551 * 'ofp_portp' is non-null). */
1553 ofproto_port_add(struct ofproto
*ofproto
, struct netdev
*netdev
,
1554 ofp_port_t
*ofp_portp
)
1556 ofp_port_t ofp_port
= ofp_portp
? *ofp_portp
: OFPP_NONE
;
1559 error
= ofproto
->ofproto_class
->port_add(ofproto
, netdev
);
1561 const char *netdev_name
= netdev_get_name(netdev
);
1563 simap_put(&ofproto
->ofp_requests
, netdev_name
,
1564 ofp_to_u16(ofp_port
));
1565 update_port(ofproto
, netdev_name
);
1568 struct ofproto_port ofproto_port
;
1570 ofproto_port_query_by_name(ofproto
, netdev_get_name(netdev
),
1572 *ofp_portp
= error
? OFPP_NONE
: ofproto_port
.ofp_port
;
1573 ofproto_port_destroy(&ofproto_port
);
1578 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1579 * initializes '*port' appropriately; on failure, returns a positive errno
1582 * The caller owns the data in 'ofproto_port' and must free it with
1583 * ofproto_port_destroy() when it is no longer needed. */
1585 ofproto_port_query_by_name(const struct ofproto
*ofproto
, const char *devname
,
1586 struct ofproto_port
*port
)
1590 error
= ofproto
->ofproto_class
->port_query_by_name(ofproto
, devname
, port
);
1592 memset(port
, 0, sizeof *port
);
1597 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1598 * Returns 0 if successful, otherwise a positive errno. */
1600 ofproto_port_del(struct ofproto
*ofproto
, ofp_port_t ofp_port
)
1602 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
1603 const char *name
= ofport
? netdev_get_name(ofport
->netdev
) : "<unknown>";
1604 struct simap_node
*ofp_request_node
;
1607 ofp_request_node
= simap_find(&ofproto
->ofp_requests
, name
);
1608 if (ofp_request_node
) {
1609 simap_delete(&ofproto
->ofp_requests
, ofp_request_node
);
1612 error
= ofproto
->ofproto_class
->port_del(ofproto
, ofp_port
);
1613 if (!error
&& ofport
) {
1614 /* 'name' is the netdev's name and update_port() is going to close the
1615 * netdev. Just in case update_port() refers to 'name' after it
1616 * destroys 'ofport', make a copy of it around the update_port()
1618 char *devname
= xstrdup(name
);
1619 update_port(ofproto
, devname
);
1625 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1626 * performs the 'n_actions' actions in 'actions'. The new flow will not
1629 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1630 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1631 * controllers; otherwise, it will be hidden.
1633 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1635 * This is a helper function for in-band control and fail-open. */
1637 ofproto_add_flow(struct ofproto
*ofproto
, const struct match
*match
,
1638 unsigned int priority
,
1639 const struct ofpact
*ofpacts
, size_t ofpacts_len
)
1641 const struct rule
*rule
;
1643 ovs_rwlock_rdlock(&ofproto
->tables
[0].cls
.rwlock
);
1644 rule
= rule_from_cls_rule(classifier_find_match_exactly(
1645 &ofproto
->tables
[0].cls
, match
, priority
));
1646 ovs_rwlock_unlock(&ofproto
->tables
[0].cls
.rwlock
);
1647 if (!rule
|| !ofpacts_equal(rule
->ofpacts
, rule
->ofpacts_len
,
1648 ofpacts
, ofpacts_len
)) {
1649 struct ofputil_flow_mod fm
;
1651 memset(&fm
, 0, sizeof fm
);
1653 fm
.priority
= priority
;
1654 fm
.buffer_id
= UINT32_MAX
;
1655 fm
.ofpacts
= xmemdup(ofpacts
, ofpacts_len
);
1656 fm
.ofpacts_len
= ofpacts_len
;
1657 add_flow(ofproto
, NULL
, &fm
, NULL
);
1662 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1663 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1664 * operation cannot be initiated now but may be retried later.
1666 * This is a helper function for in-band control and fail-open. */
1668 ofproto_flow_mod(struct ofproto
*ofproto
, struct ofputil_flow_mod
*fm
)
1670 return handle_flow_mod__(ofproto
, NULL
, fm
, NULL
);
1673 /* Searches for a rule with matching criteria exactly equal to 'target' in
1674 * ofproto's table 0 and, if it finds one, deletes it.
1676 * This is a helper function for in-band control and fail-open. */
1678 ofproto_delete_flow(struct ofproto
*ofproto
,
1679 const struct match
*target
, unsigned int priority
)
1683 ovs_rwlock_rdlock(&ofproto
->tables
[0].cls
.rwlock
);
1684 rule
= rule_from_cls_rule(classifier_find_match_exactly(
1685 &ofproto
->tables
[0].cls
, target
, priority
));
1686 ovs_rwlock_unlock(&ofproto
->tables
[0].cls
.rwlock
);
1688 /* No such rule -> success. */
1690 } else if (rule
->pending
) {
1691 /* An operation on the rule is already pending -> failure.
1692 * Caller must retry later if it's important. */
1695 /* Initiate deletion -> success. */
1696 struct ofopgroup
*group
= ofopgroup_create_unattached(ofproto
);
1697 ofoperation_create(group
, rule
, OFOPERATION_DELETE
, OFPRR_DELETE
);
1698 ovs_rwlock_wrlock(&rule
->evict
);
1699 oftable_remove_rule(rule
);
1700 ofproto
->ofproto_class
->rule_destruct(rule
);
1701 ofopgroup_submit(group
);
1707 /* Starts the process of deleting all of the flows from all of ofproto's flow
1708 * tables and then reintroducing the flows required by in-band control and
1709 * fail-open. The process will complete in a later call to ofproto_run(). */
1711 ofproto_flush_flows(struct ofproto
*ofproto
)
1713 COVERAGE_INC(ofproto_flush
);
1714 ofproto
->state
= S_FLUSH
;
1718 reinit_ports(struct ofproto
*p
)
1720 struct ofproto_port_dump dump
;
1721 struct sset devnames
;
1722 struct ofport
*ofport
;
1723 struct ofproto_port ofproto_port
;
1724 const char *devname
;
1726 COVERAGE_INC(ofproto_reinit_ports
);
1728 sset_init(&devnames
);
1729 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1730 sset_add(&devnames
, netdev_get_name(ofport
->netdev
));
1732 OFPROTO_PORT_FOR_EACH (&ofproto_port
, &dump
, p
) {
1733 sset_add(&devnames
, ofproto_port
.name
);
1736 SSET_FOR_EACH (devname
, &devnames
) {
1737 update_port(p
, devname
);
1739 sset_destroy(&devnames
);
1743 alloc_ofp_port(struct ofproto
*ofproto
, const char *netdev_name
)
1745 uint16_t max_ports
= ofp_to_u16(ofproto
->max_ports
);
1748 port_idx
= simap_get(&ofproto
->ofp_requests
, netdev_name
);
1750 port_idx
= UINT16_MAX
;
1753 if (port_idx
>= max_ports
1754 || bitmap_is_set(ofproto
->ofp_port_ids
, port_idx
)) {
1755 uint16_t end_port_no
= ofp_to_u16(ofproto
->alloc_port_no
);
1756 uint16_t alloc_port_no
= end_port_no
;
1758 /* Search for a free OpenFlow port number. We try not to
1759 * immediately reuse them to prevent problems due to old
1762 if (++alloc_port_no
>= max_ports
) {
1765 if (!bitmap_is_set(ofproto
->ofp_port_ids
, alloc_port_no
)) {
1766 port_idx
= alloc_port_no
;
1767 ofproto
->alloc_port_no
= u16_to_ofp(alloc_port_no
);
1770 if (alloc_port_no
== end_port_no
) {
1775 bitmap_set1(ofproto
->ofp_port_ids
, port_idx
);
1776 return u16_to_ofp(port_idx
);
1780 dealloc_ofp_port(const struct ofproto
*ofproto
, ofp_port_t ofp_port
)
1782 if (ofp_to_u16(ofp_port
) < ofp_to_u16(ofproto
->max_ports
)) {
1783 bitmap_set0(ofproto
->ofp_port_ids
, ofp_to_u16(ofp_port
));
1787 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1788 * pointer if the netdev cannot be opened. On success, also fills in
1790 static struct netdev
*
1791 ofport_open(struct ofproto
*ofproto
,
1792 struct ofproto_port
*ofproto_port
,
1793 struct ofputil_phy_port
*pp
)
1795 enum netdev_flags flags
;
1796 struct netdev
*netdev
;
1799 error
= netdev_open(ofproto_port
->name
, ofproto_port
->type
, &netdev
);
1801 VLOG_WARN_RL(&rl
, "%s: ignoring port %s (%"PRIu16
") because netdev %s "
1802 "cannot be opened (%s)",
1804 ofproto_port
->name
, ofproto_port
->ofp_port
,
1805 ofproto_port
->name
, ovs_strerror(error
));
1809 if (ofproto_port
->ofp_port
== OFPP_NONE
) {
1810 if (!strcmp(ofproto
->name
, ofproto_port
->name
)) {
1811 ofproto_port
->ofp_port
= OFPP_LOCAL
;
1813 ofproto_port
->ofp_port
= alloc_ofp_port(ofproto
,
1814 ofproto_port
->name
);
1817 pp
->port_no
= ofproto_port
->ofp_port
;
1818 netdev_get_etheraddr(netdev
, pp
->hw_addr
);
1819 ovs_strlcpy(pp
->name
, ofproto_port
->name
, sizeof pp
->name
);
1820 netdev_get_flags(netdev
, &flags
);
1821 pp
->config
= flags
& NETDEV_UP
? 0 : OFPUTIL_PC_PORT_DOWN
;
1822 pp
->state
= netdev_get_carrier(netdev
) ? 0 : OFPUTIL_PS_LINK_DOWN
;
1823 netdev_get_features(netdev
, &pp
->curr
, &pp
->advertised
,
1824 &pp
->supported
, &pp
->peer
);
1825 pp
->curr_speed
= netdev_features_to_bps(pp
->curr
, 0);
1826 pp
->max_speed
= netdev_features_to_bps(pp
->supported
, 0);
1831 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1832 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1835 ofport_equal(const struct ofputil_phy_port
*a
,
1836 const struct ofputil_phy_port
*b
)
1838 return (eth_addr_equals(a
->hw_addr
, b
->hw_addr
)
1839 && a
->state
== b
->state
1840 && !((a
->config
^ b
->config
) & OFPUTIL_PC_PORT_DOWN
)
1841 && a
->curr
== b
->curr
1842 && a
->advertised
== b
->advertised
1843 && a
->supported
== b
->supported
1844 && a
->peer
== b
->peer
1845 && a
->curr_speed
== b
->curr_speed
1846 && a
->max_speed
== b
->max_speed
);
1849 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1850 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1851 * one with the same name or port number). */
1853 ofport_install(struct ofproto
*p
,
1854 struct netdev
*netdev
, const struct ofputil_phy_port
*pp
)
1856 const char *netdev_name
= netdev_get_name(netdev
);
1857 struct ofport
*ofport
;
1860 /* Create ofport. */
1861 ofport
= p
->ofproto_class
->port_alloc();
1866 ofport
->ofproto
= p
;
1867 ofport
->netdev
= netdev
;
1868 ofport
->change_seq
= netdev_change_seq(netdev
);
1870 ofport
->ofp_port
= pp
->port_no
;
1871 ofport
->created
= time_msec();
1873 /* Add port to 'p'. */
1874 hmap_insert(&p
->ports
, &ofport
->hmap_node
,
1875 hash_ofp_port(ofport
->ofp_port
));
1876 shash_add(&p
->port_by_name
, netdev_name
, ofport
);
1878 update_mtu(p
, ofport
);
1880 /* Let the ofproto_class initialize its private data. */
1881 error
= p
->ofproto_class
->port_construct(ofport
);
1885 connmgr_send_port_status(p
->connmgr
, pp
, OFPPR_ADD
);
1889 VLOG_WARN_RL(&rl
, "%s: could not add port %s (%s)",
1890 p
->name
, netdev_name
, ovs_strerror(error
));
1892 ofport_destroy__(ofport
);
1894 netdev_close(netdev
);
1898 /* Removes 'ofport' from 'p' and destroys it. */
1900 ofport_remove(struct ofport
*ofport
)
1902 connmgr_send_port_status(ofport
->ofproto
->connmgr
, &ofport
->pp
,
1904 ofport_destroy(ofport
);
1907 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1910 ofport_remove_with_name(struct ofproto
*ofproto
, const char *name
)
1912 struct ofport
*port
= shash_find_data(&ofproto
->port_by_name
, name
);
1914 ofport_remove(port
);
1918 /* Updates 'port' with new 'pp' description.
1920 * Does not handle a name or port number change. The caller must implement
1921 * such a change as a delete followed by an add. */
1923 ofport_modified(struct ofport
*port
, struct ofputil_phy_port
*pp
)
1925 memcpy(port
->pp
.hw_addr
, pp
->hw_addr
, ETH_ADDR_LEN
);
1926 port
->pp
.config
= ((port
->pp
.config
& ~OFPUTIL_PC_PORT_DOWN
)
1927 | (pp
->config
& OFPUTIL_PC_PORT_DOWN
));
1928 port
->pp
.state
= pp
->state
;
1929 port
->pp
.curr
= pp
->curr
;
1930 port
->pp
.advertised
= pp
->advertised
;
1931 port
->pp
.supported
= pp
->supported
;
1932 port
->pp
.peer
= pp
->peer
;
1933 port
->pp
.curr_speed
= pp
->curr_speed
;
1934 port
->pp
.max_speed
= pp
->max_speed
;
1936 connmgr_send_port_status(port
->ofproto
->connmgr
, &port
->pp
, OFPPR_MODIFY
);
1939 /* Update OpenFlow 'state' in 'port' and notify controller. */
1941 ofproto_port_set_state(struct ofport
*port
, enum ofputil_port_state state
)
1943 if (port
->pp
.state
!= state
) {
1944 port
->pp
.state
= state
;
1945 connmgr_send_port_status(port
->ofproto
->connmgr
, &port
->pp
,
1951 ofproto_port_unregister(struct ofproto
*ofproto
, ofp_port_t ofp_port
)
1953 struct ofport
*port
= ofproto_get_port(ofproto
, ofp_port
);
1955 if (port
->ofproto
->ofproto_class
->set_realdev
) {
1956 port
->ofproto
->ofproto_class
->set_realdev(port
, 0, 0);
1958 if (port
->ofproto
->ofproto_class
->set_stp_port
) {
1959 port
->ofproto
->ofproto_class
->set_stp_port(port
, NULL
);
1961 if (port
->ofproto
->ofproto_class
->set_cfm
) {
1962 port
->ofproto
->ofproto_class
->set_cfm(port
, NULL
);
1964 if (port
->ofproto
->ofproto_class
->bundle_remove
) {
1965 port
->ofproto
->ofproto_class
->bundle_remove(port
);
1971 ofport_destroy__(struct ofport
*port
)
1973 struct ofproto
*ofproto
= port
->ofproto
;
1974 const char *name
= netdev_get_name(port
->netdev
);
1976 hmap_remove(&ofproto
->ports
, &port
->hmap_node
);
1977 shash_delete(&ofproto
->port_by_name
,
1978 shash_find(&ofproto
->port_by_name
, name
));
1980 netdev_close(port
->netdev
);
1981 ofproto
->ofproto_class
->port_dealloc(port
);
1985 ofport_destroy(struct ofport
*port
)
1988 dealloc_ofp_port(port
->ofproto
, port
->ofp_port
);
1989 port
->ofproto
->ofproto_class
->port_destruct(port
);
1990 ofport_destroy__(port
);
1995 ofproto_get_port(const struct ofproto
*ofproto
, ofp_port_t ofp_port
)
1997 struct ofport
*port
;
1999 HMAP_FOR_EACH_IN_BUCKET (port
, hmap_node
, hash_ofp_port(ofp_port
),
2001 if (port
->ofp_port
== ofp_port
) {
2009 ofproto_port_get_stats(const struct ofport
*port
, struct netdev_stats
*stats
)
2011 struct ofproto
*ofproto
= port
->ofproto
;
2014 if (ofproto
->ofproto_class
->port_get_stats
) {
2015 error
= ofproto
->ofproto_class
->port_get_stats(port
, stats
);
2024 update_port(struct ofproto
*ofproto
, const char *name
)
2026 struct ofproto_port ofproto_port
;
2027 struct ofputil_phy_port pp
;
2028 struct netdev
*netdev
;
2029 struct ofport
*port
;
2031 COVERAGE_INC(ofproto_update_port
);
2033 /* Fetch 'name''s location and properties from the datapath. */
2034 netdev
= (!ofproto_port_query_by_name(ofproto
, name
, &ofproto_port
)
2035 ? ofport_open(ofproto
, &ofproto_port
, &pp
)
2039 port
= ofproto_get_port(ofproto
, ofproto_port
.ofp_port
);
2040 if (port
&& !strcmp(netdev_get_name(port
->netdev
), name
)) {
2041 struct netdev
*old_netdev
= port
->netdev
;
2043 /* 'name' hasn't changed location. Any properties changed? */
2044 if (!ofport_equal(&port
->pp
, &pp
)) {
2045 ofport_modified(port
, &pp
);
2048 update_mtu(ofproto
, port
);
2050 /* Install the newly opened netdev in case it has changed.
2051 * Don't close the old netdev yet in case port_modified has to
2052 * remove a retained reference to it.*/
2053 port
->netdev
= netdev
;
2054 port
->change_seq
= netdev_change_seq(netdev
);
2056 if (port
->ofproto
->ofproto_class
->port_modified
) {
2057 port
->ofproto
->ofproto_class
->port_modified(port
);
2060 netdev_close(old_netdev
);
2062 /* If 'port' is nonnull then its name differs from 'name' and thus
2063 * we should delete it. If we think there's a port named 'name'
2064 * then its port number must be wrong now so delete it too. */
2066 ofport_remove(port
);
2068 ofport_remove_with_name(ofproto
, name
);
2069 ofport_install(ofproto
, netdev
, &pp
);
2072 /* Any port named 'name' is gone now. */
2073 ofport_remove_with_name(ofproto
, name
);
2075 ofproto_port_destroy(&ofproto_port
);
2079 init_ports(struct ofproto
*p
)
2081 struct ofproto_port_dump dump
;
2082 struct ofproto_port ofproto_port
;
2083 struct shash_node
*node
, *next
;
2085 OFPROTO_PORT_FOR_EACH (&ofproto_port
, &dump
, p
) {
2086 const char *name
= ofproto_port
.name
;
2088 if (shash_find(&p
->port_by_name
, name
)) {
2089 VLOG_WARN_RL(&rl
, "%s: ignoring duplicate device %s in datapath",
2092 struct ofputil_phy_port pp
;
2093 struct netdev
*netdev
;
2095 /* Check if an OpenFlow port number had been requested. */
2096 node
= shash_find(&init_ofp_ports
, name
);
2098 const struct iface_hint
*iface_hint
= node
->data
;
2099 simap_put(&p
->ofp_requests
, name
,
2100 ofp_to_u16(iface_hint
->ofp_port
));
2103 netdev
= ofport_open(p
, &ofproto_port
, &pp
);
2105 ofport_install(p
, netdev
, &pp
);
2106 if (ofproto_port
.ofp_port
< p
->max_ports
) {
2107 p
->alloc_port_no
= MAX(p
->alloc_port_no
,
2108 ofproto_port
.ofp_port
);
2114 SHASH_FOR_EACH_SAFE(node
, next
, &init_ofp_ports
) {
2115 struct iface_hint
*iface_hint
= node
->data
;
2117 if (!strcmp(iface_hint
->br_name
, p
->name
)) {
2118 free(iface_hint
->br_name
);
2119 free(iface_hint
->br_type
);
2121 shash_delete(&init_ofp_ports
, node
);
2128 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2129 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2131 find_min_mtu(struct ofproto
*p
)
2133 struct ofport
*ofport
;
2136 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
2137 struct netdev
*netdev
= ofport
->netdev
;
2140 /* Skip any internal ports, since that's what we're trying to
2142 if (!strcmp(netdev_get_type(netdev
), "internal")) {
2146 if (netdev_get_mtu(netdev
, &dev_mtu
)) {
2149 if (!mtu
|| dev_mtu
< mtu
) {
2154 return mtu
? mtu
: ETH_PAYLOAD_MAX
;
2157 /* Update MTU of all datapath devices on 'p' to the minimum of the
2158 * non-datapath ports in event of 'port' added or changed. */
2160 update_mtu(struct ofproto
*p
, struct ofport
*port
)
2162 struct ofport
*ofport
;
2163 struct netdev
*netdev
= port
->netdev
;
2164 int dev_mtu
, old_min
;
2166 if (netdev_get_mtu(netdev
, &dev_mtu
)) {
2170 if (!strcmp(netdev_get_type(port
->netdev
), "internal")) {
2171 if (dev_mtu
> p
->min_mtu
) {
2172 if (!netdev_set_mtu(port
->netdev
, p
->min_mtu
)) {
2173 dev_mtu
= p
->min_mtu
;
2176 port
->mtu
= dev_mtu
;
2180 /* For non-internal port find new min mtu. */
2181 old_min
= p
->min_mtu
;
2182 port
->mtu
= dev_mtu
;
2183 p
->min_mtu
= find_min_mtu(p
);
2184 if (p
->min_mtu
== old_min
) {
2188 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
2189 struct netdev
*netdev
= ofport
->netdev
;
2191 if (!strcmp(netdev_get_type(netdev
), "internal")) {
2192 if (!netdev_set_mtu(netdev
, p
->min_mtu
)) {
2193 ofport
->mtu
= p
->min_mtu
;
2200 ofproto_rule_destroy__(struct rule
*rule
)
2203 cls_rule_destroy(&rule
->cr
);
2204 free(rule
->ofpacts
);
2205 ovs_mutex_destroy(&rule
->timeout_mutex
);
2206 ovs_rwlock_destroy(&rule
->evict
);
2207 rule
->ofproto
->ofproto_class
->rule_dealloc(rule
);
2211 /* This function allows an ofproto implementation to destroy any rules that
2212 * remain when its ->destruct() function is called. The caller must have
2213 * already uninitialized any derived members of 'rule' (step 5 described in the
2214 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
2215 * This function implements steps 6 and 7.
2217 * This function should only be called from an ofproto implementation's
2218 * ->destruct() function. It is not suitable elsewhere. */
2220 ofproto_rule_destroy(struct ofproto
*ofproto
, struct classifier
*cls
,
2221 struct rule
*rule
) OVS_REQ_WRLOCK(cls
->rwlock
)
2223 ovs_assert(!rule
->pending
);
2224 if (!ovs_rwlock_trywrlock(&rule
->evict
)) {
2225 oftable_remove_rule__(ofproto
, cls
, rule
);
2229 ofproto_rule_destroy__(rule
);
2232 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2233 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2235 ofproto_rule_has_out_port(const struct rule
*rule
, ofp_port_t port
)
2237 return (port
== OFPP_ANY
2238 || ofpacts_output_to_port(rule
->ofpacts
, rule
->ofpacts_len
, port
));
2241 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2242 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2244 ofoperation_has_out_port(const struct ofoperation
*op
, ofp_port_t out_port
)
2246 if (ofproto_rule_has_out_port(op
->rule
, out_port
)) {
2251 case OFOPERATION_ADD
:
2252 return op
->victim
&& ofproto_rule_has_out_port(op
->victim
, out_port
);
2254 case OFOPERATION_DELETE
:
2257 case OFOPERATION_MODIFY
:
2258 return ofpacts_output_to_port(op
->ofpacts
, op
->ofpacts_len
, out_port
);
2264 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2265 * statistics appropriately. 'packet' must have at least sizeof(struct
2266 * ofp10_packet_in) bytes of headroom.
2268 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2269 * with statistics for 'packet' either way.
2271 * Takes ownership of 'packet'. */
2273 rule_execute(struct rule
*rule
, ofp_port_t in_port
, struct ofpbuf
*packet
)
2276 union flow_in_port in_port_
;
2278 ovs_assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp10_packet_in
));
2280 in_port_
.ofp_port
= in_port
;
2281 flow_extract(packet
, 0, 0, NULL
, &in_port_
, &flow
);
2282 return rule
->ofproto
->ofproto_class
->rule_execute(rule
, &flow
, packet
);
2285 /* Returns true if 'rule' should be hidden from the controller.
2287 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2288 * (e.g. by in-band control) and are intentionally hidden from the
2291 ofproto_rule_is_hidden(const struct rule
*rule
)
2293 return rule
->cr
.priority
> UINT16_MAX
;
2296 static enum oftable_flags
2297 rule_get_flags(const struct rule
*rule
)
2299 return rule
->ofproto
->tables
[rule
->table_id
].flags
;
2303 rule_is_modifiable(const struct rule
*rule
)
2305 return !(rule_get_flags(rule
) & OFTABLE_READONLY
);
2309 handle_echo_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2311 ofconn_send_reply(ofconn
, make_echo_reply(oh
));
2316 handle_features_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2318 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2319 struct ofputil_switch_features features
;
2320 struct ofport
*port
;
2326 ofproto
->ofproto_class
->get_features(ofproto
, &arp_match_ip
,
2328 ovs_assert(features
.actions
& OFPUTIL_A_OUTPUT
); /* sanity check */
2330 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2331 * if present, are always at the end.) */
2332 n_tables
= ofproto
->n_tables
;
2333 for (i
= 0; i
< ofproto
->n_tables
; i
++) {
2334 if (ofproto
->tables
[i
].flags
& OFTABLE_HIDDEN
) {
2340 features
.datapath_id
= ofproto
->datapath_id
;
2341 features
.n_buffers
= pktbuf_capacity();
2342 features
.n_tables
= n_tables
;
2343 features
.capabilities
= (OFPUTIL_C_FLOW_STATS
| OFPUTIL_C_TABLE_STATS
|
2344 OFPUTIL_C_PORT_STATS
| OFPUTIL_C_QUEUE_STATS
);
2346 features
.capabilities
|= OFPUTIL_C_ARP_MATCH_IP
;
2348 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2349 features
.auxiliary_id
= 0;
2350 b
= ofputil_encode_switch_features(&features
, ofconn_get_protocol(ofconn
),
2352 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
2353 ofputil_put_switch_features_port(&port
->pp
, b
);
2356 ofconn_send_reply(ofconn
, b
);
2361 handle_get_config_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2363 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2364 struct ofp_switch_config
*osc
;
2365 enum ofp_config_flags flags
;
2369 buf
= ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY
, oh
, 0);
2370 osc
= ofpbuf_put_uninit(buf
, sizeof *osc
);
2371 flags
= ofproto
->frag_handling
;
2372 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2373 if (oh
->version
< OFP13_VERSION
2374 && ofconn_get_invalid_ttl_to_controller(ofconn
)) {
2375 flags
|= OFPC_INVALID_TTL_TO_CONTROLLER
;
2377 osc
->flags
= htons(flags
);
2378 osc
->miss_send_len
= htons(ofconn_get_miss_send_len(ofconn
));
2379 ofconn_send_reply(ofconn
, buf
);
2385 handle_set_config(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2387 const struct ofp_switch_config
*osc
= ofpmsg_body(oh
);
2388 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2389 uint16_t flags
= ntohs(osc
->flags
);
2391 if (ofconn_get_type(ofconn
) != OFCONN_PRIMARY
2392 || ofconn_get_role(ofconn
) != OFPCR12_ROLE_SLAVE
) {
2393 enum ofp_config_flags cur
= ofproto
->frag_handling
;
2394 enum ofp_config_flags next
= flags
& OFPC_FRAG_MASK
;
2396 ovs_assert((cur
& OFPC_FRAG_MASK
) == cur
);
2398 if (ofproto
->ofproto_class
->set_frag_handling(ofproto
, next
)) {
2399 ofproto
->frag_handling
= next
;
2401 VLOG_WARN_RL(&rl
, "%s: unsupported fragment handling mode %s",
2403 ofputil_frag_handling_to_string(next
));
2407 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2408 ofconn_set_invalid_ttl_to_controller(ofconn
,
2409 (oh
->version
< OFP13_VERSION
2410 && flags
& OFPC_INVALID_TTL_TO_CONTROLLER
));
2412 ofconn_set_miss_send_len(ofconn
, ntohs(osc
->miss_send_len
));
2417 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2418 * error message code for the caller to propagate upward. Otherwise, returns
2421 * The log message mentions 'msg_type'. */
2423 reject_slave_controller(struct ofconn
*ofconn
)
2425 if (ofconn_get_type(ofconn
) == OFCONN_PRIMARY
2426 && ofconn_get_role(ofconn
) == OFPCR12_ROLE_SLAVE
) {
2427 return OFPERR_OFPBRC_EPERM
;
2433 /* Finds the OFPACT_METER action, if any, in the 'ofpacts_len' bytes of
2434 * 'ofpacts'. If found, returns its meter ID; if not, returns 0.
2436 * This function relies on the order of 'ofpacts' being correct (as checked by
2437 * ofpacts_verify()). */
2439 find_meter(const struct ofpact ofpacts
[], size_t ofpacts_len
)
2441 const struct ofpact
*a
;
2443 OFPACT_FOR_EACH (a
, ofpacts
, ofpacts_len
) {
2444 enum ovs_instruction_type inst
;
2446 inst
= ovs_instruction_type_from_ofpact_type(a
->type
);
2447 if (a
->type
== OFPACT_METER
) {
2448 return ofpact_get_METER(a
)->meter_id
;
2449 } else if (inst
> OVSINST_OFPIT13_METER
) {
2457 /* Checks that the 'ofpacts_len' bytes of actions in 'ofpacts' are appropriate
2458 * for a packet with the prerequisites satisfied by 'flow' in table 'table_id'.
2459 * 'flow' may be temporarily modified, but is restored at return.
2462 ofproto_check_ofpacts(struct ofproto
*ofproto
,
2463 const struct ofpact ofpacts
[], size_t ofpacts_len
,
2464 struct flow
*flow
, uint8_t table_id
)
2469 error
= ofpacts_check(ofpacts
, ofpacts_len
, flow
, ofproto
->max_ports
,
2475 mid
= find_meter(ofpacts
, ofpacts_len
);
2476 if (mid
&& ofproto_get_provider_meter_id(ofproto
, mid
) == UINT32_MAX
) {
2477 return OFPERR_OFPMMFC_INVALID_METER
;
2483 handle_packet_out(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2485 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2486 struct ofputil_packet_out po
;
2487 struct ofpbuf
*payload
;
2488 uint64_t ofpacts_stub
[1024 / 8];
2489 struct ofpbuf ofpacts
;
2491 union flow_in_port in_port_
;
2494 COVERAGE_INC(ofproto_packet_out
);
2496 error
= reject_slave_controller(ofconn
);
2501 /* Decode message. */
2502 ofpbuf_use_stub(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
2503 error
= ofputil_decode_packet_out(&po
, oh
, &ofpacts
);
2505 goto exit_free_ofpacts
;
2507 if (ofp_to_u16(po
.in_port
) >= ofp_to_u16(p
->max_ports
)
2508 && ofp_to_u16(po
.in_port
) < ofp_to_u16(OFPP_MAX
)) {
2509 error
= OFPERR_OFPBRC_BAD_PORT
;
2510 goto exit_free_ofpacts
;
2515 if (po
.buffer_id
!= UINT32_MAX
) {
2516 error
= ofconn_pktbuf_retrieve(ofconn
, po
.buffer_id
, &payload
, NULL
);
2517 if (error
|| !payload
) {
2518 goto exit_free_ofpacts
;
2521 payload
= xmalloc(sizeof *payload
);
2522 ofpbuf_use_const(payload
, po
.packet
, po
.packet_len
);
2525 /* Verify actions against packet, then send packet if successful. */
2526 in_port_
.ofp_port
= po
.in_port
;
2527 flow_extract(payload
, 0, 0, NULL
, &in_port_
, &flow
);
2528 error
= ofproto_check_ofpacts(p
, po
.ofpacts
, po
.ofpacts_len
, &flow
, 0);
2530 error
= p
->ofproto_class
->packet_out(p
, payload
, &flow
,
2531 po
.ofpacts
, po
.ofpacts_len
);
2533 ofpbuf_delete(payload
);
2536 ofpbuf_uninit(&ofpacts
);
2542 update_port_config(struct ofport
*port
,
2543 enum ofputil_port_config config
,
2544 enum ofputil_port_config mask
)
2546 enum ofputil_port_config old_config
= port
->pp
.config
;
2547 enum ofputil_port_config toggle
;
2549 toggle
= (config
^ port
->pp
.config
) & mask
;
2550 if (toggle
& OFPUTIL_PC_PORT_DOWN
) {
2551 if (config
& OFPUTIL_PC_PORT_DOWN
) {
2552 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, NULL
);
2554 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, NULL
);
2556 toggle
&= ~OFPUTIL_PC_PORT_DOWN
;
2559 port
->pp
.config
^= toggle
;
2560 if (port
->pp
.config
!= old_config
) {
2561 port
->ofproto
->ofproto_class
->port_reconfigured(port
, old_config
);
2566 handle_port_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2568 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2569 struct ofputil_port_mod pm
;
2570 struct ofport
*port
;
2573 error
= reject_slave_controller(ofconn
);
2578 error
= ofputil_decode_port_mod(oh
, &pm
);
2583 port
= ofproto_get_port(p
, pm
.port_no
);
2585 return OFPERR_OFPPMFC_BAD_PORT
;
2586 } else if (!eth_addr_equals(port
->pp
.hw_addr
, pm
.hw_addr
)) {
2587 return OFPERR_OFPPMFC_BAD_HW_ADDR
;
2589 update_port_config(port
, pm
.config
, pm
.mask
);
2591 netdev_set_advertisements(port
->netdev
, pm
.advertise
);
2598 handle_desc_stats_request(struct ofconn
*ofconn
,
2599 const struct ofp_header
*request
)
2601 static const char *default_mfr_desc
= "Nicira, Inc.";
2602 static const char *default_hw_desc
= "Open vSwitch";
2603 static const char *default_sw_desc
= VERSION
;
2604 static const char *default_serial_desc
= "None";
2605 static const char *default_dp_desc
= "None";
2607 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2608 struct ofp_desc_stats
*ods
;
2611 msg
= ofpraw_alloc_stats_reply(request
, 0);
2612 ods
= ofpbuf_put_zeros(msg
, sizeof *ods
);
2613 ovs_strlcpy(ods
->mfr_desc
, p
->mfr_desc
? p
->mfr_desc
: default_mfr_desc
,
2614 sizeof ods
->mfr_desc
);
2615 ovs_strlcpy(ods
->hw_desc
, p
->hw_desc
? p
->hw_desc
: default_hw_desc
,
2616 sizeof ods
->hw_desc
);
2617 ovs_strlcpy(ods
->sw_desc
, p
->sw_desc
? p
->sw_desc
: default_sw_desc
,
2618 sizeof ods
->sw_desc
);
2619 ovs_strlcpy(ods
->serial_num
,
2620 p
->serial_desc
? p
->serial_desc
: default_serial_desc
,
2621 sizeof ods
->serial_num
);
2622 ovs_strlcpy(ods
->dp_desc
, p
->dp_desc
? p
->dp_desc
: default_dp_desc
,
2623 sizeof ods
->dp_desc
);
2624 ofconn_send_reply(ofconn
, msg
);
2630 handle_table_stats_request(struct ofconn
*ofconn
,
2631 const struct ofp_header
*request
)
2633 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2634 struct ofp12_table_stats
*ots
;
2639 /* Set up default values.
2641 * ofp12_table_stats is used as a generic structure as
2642 * it is able to hold all the fields for ofp10_table_stats
2643 * and ofp11_table_stats (and of course itself).
2645 ots
= xcalloc(p
->n_tables
, sizeof *ots
);
2646 for (i
= 0; i
< p
->n_tables
; i
++) {
2647 ots
[i
].table_id
= i
;
2648 sprintf(ots
[i
].name
, "table%zu", i
);
2649 ots
[i
].match
= htonll(OFPXMT12_MASK
);
2650 ots
[i
].wildcards
= htonll(OFPXMT12_MASK
);
2651 ots
[i
].write_actions
= htonl(OFPAT11_OUTPUT
);
2652 ots
[i
].apply_actions
= htonl(OFPAT11_OUTPUT
);
2653 ots
[i
].write_setfields
= htonll(OFPXMT12_MASK
);
2654 ots
[i
].apply_setfields
= htonll(OFPXMT12_MASK
);
2655 ots
[i
].metadata_match
= htonll(UINT64_MAX
);
2656 ots
[i
].metadata_write
= htonll(UINT64_MAX
);
2657 ots
[i
].instructions
= htonl(OFPIT11_ALL
);
2658 ots
[i
].config
= htonl(OFPTC11_TABLE_MISS_MASK
);
2659 ots
[i
].max_entries
= htonl(1000000); /* An arbitrary big number. */
2660 ovs_rwlock_rdlock(&p
->tables
[i
].cls
.rwlock
);
2661 ots
[i
].active_count
= htonl(classifier_count(&p
->tables
[i
].cls
));
2662 ovs_rwlock_unlock(&p
->tables
[i
].cls
.rwlock
);
2665 p
->ofproto_class
->get_tables(p
, ots
);
2667 /* Post-process the tables, dropping hidden tables. */
2668 n_tables
= p
->n_tables
;
2669 for (i
= 0; i
< p
->n_tables
; i
++) {
2670 const struct oftable
*table
= &p
->tables
[i
];
2672 if (table
->flags
& OFTABLE_HIDDEN
) {
2678 ovs_strzcpy(ots
[i
].name
, table
->name
, sizeof ots
[i
].name
);
2681 if (table
->max_flows
< ntohl(ots
[i
].max_entries
)) {
2682 ots
[i
].max_entries
= htonl(table
->max_flows
);
2686 msg
= ofputil_encode_table_stats_reply(ots
, n_tables
, request
);
2687 ofconn_send_reply(ofconn
, msg
);
2695 append_port_stat(struct ofport
*port
, struct list
*replies
)
2697 struct ofputil_port_stats ops
= { .port_no
= port
->pp
.port_no
};
2699 calc_duration(port
->created
, time_msec(),
2700 &ops
.duration_sec
, &ops
.duration_nsec
);
2702 /* Intentionally ignore return value, since errors will set
2703 * 'stats' to all-1s, which is correct for OpenFlow, and
2704 * netdev_get_stats() will log errors. */
2705 ofproto_port_get_stats(port
, &ops
.stats
);
2707 ofputil_append_port_stat(replies
, &ops
);
2711 handle_port_stats_request(struct ofconn
*ofconn
,
2712 const struct ofp_header
*request
)
2714 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2715 struct ofport
*port
;
2716 struct list replies
;
2720 error
= ofputil_decode_port_stats_request(request
, &port_no
);
2725 ofpmp_init(&replies
, request
);
2726 if (port_no
!= OFPP_ANY
) {
2727 port
= ofproto_get_port(p
, port_no
);
2729 append_port_stat(port
, &replies
);
2732 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
2733 append_port_stat(port
, &replies
);
2737 ofconn_send_replies(ofconn
, &replies
);
2742 handle_port_desc_stats_request(struct ofconn
*ofconn
,
2743 const struct ofp_header
*request
)
2745 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2746 enum ofp_version version
;
2747 struct ofport
*port
;
2748 struct list replies
;
2750 ofpmp_init(&replies
, request
);
2752 version
= ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn
));
2753 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
2754 ofputil_append_port_desc_stats_reply(version
, &port
->pp
, &replies
);
2757 ofconn_send_replies(ofconn
, &replies
);
2762 hash_cookie(ovs_be64 cookie
)
2764 return hash_2words((OVS_FORCE
uint64_t)cookie
>> 32,
2765 (OVS_FORCE
uint64_t)cookie
);
2769 cookies_insert(struct ofproto
*ofproto
, struct rule
*rule
)
2771 hindex_insert(&ofproto
->cookies
, &rule
->cookie_node
,
2772 hash_cookie(rule
->flow_cookie
));
2776 cookies_remove(struct ofproto
*ofproto
, struct rule
*rule
)
2778 hindex_remove(&ofproto
->cookies
, &rule
->cookie_node
);
2782 ofproto_rule_change_cookie(struct ofproto
*ofproto
, struct rule
*rule
,
2783 ovs_be64 new_cookie
)
2785 if (new_cookie
!= rule
->flow_cookie
) {
2786 cookies_remove(ofproto
, rule
);
2788 rule
->flow_cookie
= new_cookie
;
2790 cookies_insert(ofproto
, rule
);
2795 calc_duration(long long int start
, long long int now
,
2796 uint32_t *sec
, uint32_t *nsec
)
2798 long long int msecs
= now
- start
;
2799 *sec
= msecs
/ 1000;
2800 *nsec
= (msecs
% 1000) * (1000 * 1000);
2803 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2804 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2806 check_table_id(const struct ofproto
*ofproto
, uint8_t table_id
)
2808 return (table_id
== 0xff || table_id
< ofproto
->n_tables
2810 : OFPERR_OFPBRC_BAD_TABLE_ID
);
2814 static struct oftable
*
2815 next_visible_table(const struct ofproto
*ofproto
, uint8_t table_id
)
2817 struct oftable
*table
;
2819 for (table
= &ofproto
->tables
[table_id
];
2820 table
< &ofproto
->tables
[ofproto
->n_tables
];
2822 if (!(table
->flags
& OFTABLE_HIDDEN
)) {
2830 static struct oftable
*
2831 first_matching_table(const struct ofproto
*ofproto
, uint8_t table_id
)
2833 if (table_id
== 0xff) {
2834 return next_visible_table(ofproto
, 0);
2835 } else if (table_id
< ofproto
->n_tables
) {
2836 return &ofproto
->tables
[table_id
];
2842 static struct oftable
*
2843 next_matching_table(const struct ofproto
*ofproto
,
2844 const struct oftable
*table
, uint8_t table_id
)
2846 return (table_id
== 0xff
2847 ? next_visible_table(ofproto
, (table
- ofproto
->tables
) + 1)
2851 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2853 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2854 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2856 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2857 * only once, for that table. (This can be used to access tables marked
2860 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2861 * entered at all. (Perhaps you should have validated TABLE_ID with
2862 * check_table_id().)
2864 * All parameters are evaluated multiple times.
2866 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2867 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2869 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2871 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2872 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2873 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2876 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2877 * to 'out_port' are included.
2879 * Hidden rules are always omitted.
2881 * Returns 0 on success, otherwise an OpenFlow error code. */
2883 collect_rules_loose(struct ofproto
*ofproto
, uint8_t table_id
,
2884 const struct match
*match
,
2885 ovs_be64 cookie
, ovs_be64 cookie_mask
,
2886 ofp_port_t out_port
, struct list
*rules
)
2888 struct oftable
*table
;
2892 error
= check_table_id(ofproto
, table_id
);
2898 cls_rule_init(&cr
, match
, 0);
2900 if (cookie_mask
== htonll(UINT64_MAX
)) {
2903 HINDEX_FOR_EACH_WITH_HASH (rule
, cookie_node
, hash_cookie(cookie
),
2904 &ofproto
->cookies
) {
2905 if (table_id
!= rule
->table_id
&& table_id
!= 0xff) {
2908 if (ofproto_rule_is_hidden(rule
)) {
2911 if (cls_rule_is_loose_match(&rule
->cr
, &cr
.match
)) {
2912 if (rule
->pending
) {
2913 error
= OFPROTO_POSTPONE
;
2916 if (rule
->flow_cookie
== cookie
/* Hash collisions possible. */
2917 && ofproto_rule_has_out_port(rule
, out_port
)) {
2918 list_push_back(rules
, &rule
->ofproto_node
);
2925 FOR_EACH_MATCHING_TABLE (table
, table_id
, ofproto
) {
2926 struct cls_cursor cursor
;
2929 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
2930 cls_cursor_init(&cursor
, &table
->cls
, &cr
);
2931 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
2932 if (rule
->pending
) {
2933 ovs_rwlock_unlock(&table
->cls
.rwlock
);
2934 error
= OFPROTO_POSTPONE
;
2937 if (!ofproto_rule_is_hidden(rule
)
2938 && ofproto_rule_has_out_port(rule
, out_port
)
2939 && !((rule
->flow_cookie
^ cookie
) & cookie_mask
)) {
2940 list_push_back(rules
, &rule
->ofproto_node
);
2943 ovs_rwlock_unlock(&table
->cls
.rwlock
);
2947 cls_rule_destroy(&cr
);
2951 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2952 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2953 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2956 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2957 * to 'out_port' are included.
2959 * Hidden rules are always omitted.
2961 * Returns 0 on success, otherwise an OpenFlow error code. */
2963 collect_rules_strict(struct ofproto
*ofproto
, uint8_t table_id
,
2964 const struct match
*match
, unsigned int priority
,
2965 ovs_be64 cookie
, ovs_be64 cookie_mask
,
2966 ofp_port_t out_port
, struct list
*rules
)
2968 struct oftable
*table
;
2972 error
= check_table_id(ofproto
, table_id
);
2978 cls_rule_init(&cr
, match
, priority
);
2980 if (cookie_mask
== htonll(UINT64_MAX
)) {
2983 HINDEX_FOR_EACH_WITH_HASH (rule
, cookie_node
, hash_cookie(cookie
),
2984 &ofproto
->cookies
) {
2985 if (table_id
!= rule
->table_id
&& table_id
!= 0xff) {
2988 if (ofproto_rule_is_hidden(rule
)) {
2991 if (cls_rule_equal(&rule
->cr
, &cr
)) {
2992 if (rule
->pending
) {
2993 error
= OFPROTO_POSTPONE
;
2996 if (rule
->flow_cookie
== cookie
/* Hash collisions possible. */
2997 && ofproto_rule_has_out_port(rule
, out_port
)) {
2998 list_push_back(rules
, &rule
->ofproto_node
);
3005 FOR_EACH_MATCHING_TABLE (table
, table_id
, ofproto
) {
3008 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
3009 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&table
->cls
,
3011 ovs_rwlock_unlock(&table
->cls
.rwlock
);
3013 if (rule
->pending
) {
3014 error
= OFPROTO_POSTPONE
;
3017 if (!ofproto_rule_is_hidden(rule
)
3018 && ofproto_rule_has_out_port(rule
, out_port
)
3019 && !((rule
->flow_cookie
^ cookie
) & cookie_mask
)) {
3020 list_push_back(rules
, &rule
->ofproto_node
);
3026 cls_rule_destroy(&cr
);
3030 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
3031 * forced into the range of a uint16_t. */
3033 age_secs(long long int age_ms
)
3035 return (age_ms
< 0 ? 0
3036 : age_ms
>= UINT16_MAX
* 1000 ? UINT16_MAX
3037 : (unsigned int) age_ms
/ 1000);
3041 handle_flow_stats_request(struct ofconn
*ofconn
,
3042 const struct ofp_header
*request
)
3044 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
3045 struct ofputil_flow_stats_request fsr
;
3046 struct list replies
;
3051 error
= ofputil_decode_flow_stats_request(&fsr
, request
);
3056 error
= collect_rules_loose(ofproto
, fsr
.table_id
, &fsr
.match
,
3057 fsr
.cookie
, fsr
.cookie_mask
,
3058 fsr
.out_port
, &rules
);
3063 ofpmp_init(&replies
, request
);
3064 LIST_FOR_EACH (rule
, ofproto_node
, &rules
) {
3065 long long int now
= time_msec();
3066 struct ofputil_flow_stats fs
;
3068 minimatch_expand(&rule
->cr
.match
, &fs
.match
);
3069 fs
.priority
= rule
->cr
.priority
;
3070 fs
.cookie
= rule
->flow_cookie
;
3071 fs
.table_id
= rule
->table_id
;
3072 calc_duration(rule
->created
, now
, &fs
.duration_sec
, &fs
.duration_nsec
);
3073 fs
.idle_age
= age_secs(now
- rule
->used
);
3074 fs
.hard_age
= age_secs(now
- rule
->modified
);
3075 ofproto
->ofproto_class
->rule_get_stats(rule
, &fs
.packet_count
,
3077 fs
.ofpacts
= rule
->ofpacts
;
3078 fs
.ofpacts_len
= rule
->ofpacts_len
;
3081 ovs_mutex_lock(&rule
->timeout_mutex
);
3082 fs
.idle_timeout
= rule
->idle_timeout
;
3083 fs
.hard_timeout
= rule
->hard_timeout
;
3084 ovs_mutex_unlock(&rule
->timeout_mutex
);
3086 if (rule
->send_flow_removed
) {
3087 fs
.flags
|= OFPFF_SEND_FLOW_REM
;
3088 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3089 and OFPFF13_NO_BYT_COUNTS */
3091 ofputil_append_flow_stats_reply(&fs
, &replies
);
3093 ofconn_send_replies(ofconn
, &replies
);
3099 flow_stats_ds(struct rule
*rule
, struct ds
*results
)
3101 uint64_t packet_count
, byte_count
;
3103 rule
->ofproto
->ofproto_class
->rule_get_stats(rule
,
3104 &packet_count
, &byte_count
);
3106 if (rule
->table_id
!= 0) {
3107 ds_put_format(results
, "table_id=%"PRIu8
", ", rule
->table_id
);
3109 ds_put_format(results
, "duration=%llds, ",
3110 (time_msec() - rule
->created
) / 1000);
3111 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
3112 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
3113 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
3114 cls_rule_format(&rule
->cr
, results
);
3115 ds_put_char(results
, ',');
3116 ofpacts_format(rule
->ofpacts
, rule
->ofpacts_len
, results
);
3117 ds_put_cstr(results
, "\n");
3120 /* Adds a pretty-printed description of all flows to 'results', including
3121 * hidden flows (e.g., set up by in-band control). */
3123 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
3125 struct oftable
*table
;
3127 OFPROTO_FOR_EACH_TABLE (table
, p
) {
3128 struct cls_cursor cursor
;
3131 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
3132 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
3133 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3134 flow_stats_ds(rule
, results
);
3136 ovs_rwlock_unlock(&table
->cls
.rwlock
);
3140 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3141 * '*engine_type' and '*engine_id', respectively. */
3143 ofproto_get_netflow_ids(const struct ofproto
*ofproto
,
3144 uint8_t *engine_type
, uint8_t *engine_id
)
3146 ofproto
->ofproto_class
->get_netflow_ids(ofproto
, engine_type
, engine_id
);
3149 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
3150 * true if the port's CFM status was successfully stored into '*status'.
3151 * Returns false if the port did not have CFM configured, in which case
3152 * '*status' is indeterminate.
3154 * The caller must provide and owns '*status', and must free 'status->rmps'. */
3156 ofproto_port_get_cfm_status(const struct ofproto
*ofproto
, ofp_port_t ofp_port
,
3157 struct ofproto_cfm_status
*status
)
3159 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
3161 && ofproto
->ofproto_class
->get_cfm_status
3162 && ofproto
->ofproto_class
->get_cfm_status(ofport
, status
));
3166 handle_aggregate_stats_request(struct ofconn
*ofconn
,
3167 const struct ofp_header
*oh
)
3169 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
3170 struct ofputil_flow_stats_request request
;
3171 struct ofputil_aggregate_stats stats
;
3172 bool unknown_packets
, unknown_bytes
;
3173 struct ofpbuf
*reply
;
3178 error
= ofputil_decode_flow_stats_request(&request
, oh
);
3183 error
= collect_rules_loose(ofproto
, request
.table_id
, &request
.match
,
3184 request
.cookie
, request
.cookie_mask
,
3185 request
.out_port
, &rules
);
3190 memset(&stats
, 0, sizeof stats
);
3191 unknown_packets
= unknown_bytes
= false;
3192 LIST_FOR_EACH (rule
, ofproto_node
, &rules
) {
3193 uint64_t packet_count
;
3194 uint64_t byte_count
;
3196 ofproto
->ofproto_class
->rule_get_stats(rule
, &packet_count
,
3199 if (packet_count
== UINT64_MAX
) {
3200 unknown_packets
= true;
3202 stats
.packet_count
+= packet_count
;
3205 if (byte_count
== UINT64_MAX
) {
3206 unknown_bytes
= true;
3208 stats
.byte_count
+= byte_count
;
3213 if (unknown_packets
) {
3214 stats
.packet_count
= UINT64_MAX
;
3216 if (unknown_bytes
) {
3217 stats
.byte_count
= UINT64_MAX
;
3220 reply
= ofputil_encode_aggregate_stats_reply(&stats
, oh
);
3221 ofconn_send_reply(ofconn
, reply
);
3226 struct queue_stats_cbdata
{
3227 struct ofport
*ofport
;
3228 struct list replies
;
3233 put_queue_stats(struct queue_stats_cbdata
*cbdata
, uint32_t queue_id
,
3234 const struct netdev_queue_stats
*stats
)
3236 struct ofputil_queue_stats oqs
;
3238 oqs
.port_no
= cbdata
->ofport
->pp
.port_no
;
3239 oqs
.queue_id
= queue_id
;
3240 oqs
.tx_bytes
= stats
->tx_bytes
;
3241 oqs
.tx_packets
= stats
->tx_packets
;
3242 oqs
.tx_errors
= stats
->tx_errors
;
3243 if (stats
->created
!= LLONG_MIN
) {
3244 calc_duration(stats
->created
, cbdata
->now
,
3245 &oqs
.duration_sec
, &oqs
.duration_nsec
);
3247 oqs
.duration_sec
= oqs
.duration_nsec
= UINT32_MAX
;
3249 ofputil_append_queue_stat(&cbdata
->replies
, &oqs
);
3253 handle_queue_stats_dump_cb(uint32_t queue_id
,
3254 struct netdev_queue_stats
*stats
,
3257 struct queue_stats_cbdata
*cbdata
= cbdata_
;
3259 put_queue_stats(cbdata
, queue_id
, stats
);
3263 handle_queue_stats_for_port(struct ofport
*port
, uint32_t queue_id
,
3264 struct queue_stats_cbdata
*cbdata
)
3266 cbdata
->ofport
= port
;
3267 if (queue_id
== OFPQ_ALL
) {
3268 netdev_dump_queue_stats(port
->netdev
,
3269 handle_queue_stats_dump_cb
, cbdata
);
3271 struct netdev_queue_stats stats
;
3273 if (!netdev_get_queue_stats(port
->netdev
, queue_id
, &stats
)) {
3274 put_queue_stats(cbdata
, queue_id
, &stats
);
3276 return OFPERR_OFPQOFC_BAD_QUEUE
;
3283 handle_queue_stats_request(struct ofconn
*ofconn
,
3284 const struct ofp_header
*rq
)
3286 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
3287 struct queue_stats_cbdata cbdata
;
3288 struct ofport
*port
;
3290 struct ofputil_queue_stats_request oqsr
;
3292 COVERAGE_INC(ofproto_queue_req
);
3294 ofpmp_init(&cbdata
.replies
, rq
);
3295 cbdata
.now
= time_msec();
3297 error
= ofputil_decode_queue_stats_request(rq
, &oqsr
);
3302 if (oqsr
.port_no
== OFPP_ANY
) {
3303 error
= OFPERR_OFPQOFC_BAD_QUEUE
;
3304 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
3305 if (!handle_queue_stats_for_port(port
, oqsr
.queue_id
, &cbdata
)) {
3310 port
= ofproto_get_port(ofproto
, oqsr
.port_no
);
3312 ? handle_queue_stats_for_port(port
, oqsr
.queue_id
, &cbdata
)
3313 : OFPERR_OFPQOFC_BAD_PORT
);
3316 ofconn_send_replies(ofconn
, &cbdata
.replies
);
3318 ofpbuf_list_delete(&cbdata
.replies
);
3325 is_flow_deletion_pending(const struct ofproto
*ofproto
,
3326 const struct cls_rule
*cls_rule
,
3329 if (!hmap_is_empty(&ofproto
->deletions
)) {
3330 struct ofoperation
*op
;
3332 HMAP_FOR_EACH_WITH_HASH (op
, hmap_node
,
3333 cls_rule_hash(cls_rule
, table_id
),
3334 &ofproto
->deletions
) {
3335 if (cls_rule_equal(cls_rule
, &op
->rule
->cr
)) {
3344 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3345 * in which no matching flow already exists in the flow table.
3347 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3348 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3349 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3350 * initiated now but may be retried later.
3352 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3353 * ownership remains with the caller.
3355 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3358 add_flow(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3359 struct ofputil_flow_mod
*fm
, const struct ofp_header
*request
)
3361 struct oftable
*table
;
3362 struct ofopgroup
*group
;
3363 struct rule
*victim
;
3368 error
= check_table_id(ofproto
, fm
->table_id
);
3374 if (fm
->table_id
== 0xff) {
3375 if (ofproto
->ofproto_class
->rule_choose_table
) {
3376 error
= ofproto
->ofproto_class
->rule_choose_table(ofproto
,
3382 ovs_assert(table_id
< ofproto
->n_tables
);
3386 } else if (fm
->table_id
< ofproto
->n_tables
) {
3387 table_id
= fm
->table_id
;
3389 return OFPERR_OFPBRC_BAD_TABLE_ID
;
3392 table
= &ofproto
->tables
[table_id
];
3394 if (table
->flags
& OFTABLE_READONLY
) {
3395 return OFPERR_OFPBRC_EPERM
;
3398 /* Verify actions. */
3399 error
= ofproto_check_ofpacts(ofproto
, fm
->ofpacts
, fm
->ofpacts_len
,
3400 &fm
->match
.flow
, table_id
);
3405 /* Allocate new rule and initialize classifier rule. */
3406 rule
= ofproto
->ofproto_class
->rule_alloc();
3408 VLOG_WARN_RL(&rl
, "%s: failed to create rule (%s)",
3409 ofproto
->name
, ovs_strerror(error
));
3412 cls_rule_init(&rule
->cr
, &fm
->match
, fm
->priority
);
3414 /* Serialize against pending deletion. */
3415 if (is_flow_deletion_pending(ofproto
, &rule
->cr
, table_id
)) {
3416 cls_rule_destroy(&rule
->cr
);
3417 ofproto
->ofproto_class
->rule_dealloc(rule
);
3418 return OFPROTO_POSTPONE
;
3421 /* Check for overlap, if requested. */
3422 if (fm
->flags
& OFPFF_CHECK_OVERLAP
) {
3425 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
3426 overlaps
= classifier_rule_overlaps(&table
->cls
, &rule
->cr
);
3427 ovs_rwlock_unlock(&table
->cls
.rwlock
);
3430 cls_rule_destroy(&rule
->cr
);
3431 ofproto
->ofproto_class
->rule_dealloc(rule
);
3432 return OFPERR_OFPFMFC_OVERLAP
;
3436 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3438 rule
->ofproto
= ofproto
;
3439 rule
->pending
= NULL
;
3440 rule
->flow_cookie
= fm
->new_cookie
;
3441 rule
->created
= rule
->modified
= rule
->used
= time_msec();
3443 ovs_mutex_init(&rule
->timeout_mutex
);
3444 ovs_mutex_lock(&rule
->timeout_mutex
);
3445 rule
->idle_timeout
= fm
->idle_timeout
;
3446 rule
->hard_timeout
= fm
->hard_timeout
;
3447 ovs_mutex_unlock(&rule
->timeout_mutex
);
3449 rule
->table_id
= table
- ofproto
->tables
;
3450 rule
->send_flow_removed
= (fm
->flags
& OFPFF_SEND_FLOW_REM
) != 0;
3451 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3452 and OFPFF13_NO_BYT_COUNTS */
3453 rule
->ofpacts
= xmemdup(fm
->ofpacts
, fm
->ofpacts_len
);
3454 rule
->ofpacts_len
= fm
->ofpacts_len
;
3455 rule
->meter_id
= find_meter(rule
->ofpacts
, rule
->ofpacts_len
);
3456 list_init(&rule
->meter_list_node
);
3457 rule
->eviction_group
= NULL
;
3458 list_init(&rule
->expirable
);
3459 rule
->monitor_flags
= 0;
3460 rule
->add_seqno
= 0;
3461 rule
->modify_seqno
= 0;
3462 ovs_rwlock_init(&rule
->evict
);
3464 /* Insert new rule. */
3465 victim
= oftable_replace_rule(rule
);
3466 if (victim
&& !rule_is_modifiable(victim
)) {
3467 error
= OFPERR_OFPBRC_EPERM
;
3468 } else if (victim
&& victim
->pending
) {
3469 error
= OFPROTO_POSTPONE
;
3471 struct ofoperation
*op
;
3475 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
3476 n_rules
= classifier_count(&table
->cls
);
3477 ovs_rwlock_unlock(&table
->cls
.rwlock
);
3478 if (n_rules
> table
->max_flows
) {
3479 ovs_rwlock_rdlock(&rule
->evict
);
3480 if (choose_rule_to_evict(table
, &evict
)) {
3481 ovs_rwlock_unlock(&rule
->evict
);
3482 ovs_rwlock_unlock(&evict
->evict
);
3483 if (evict
->pending
) {
3484 error
= OFPROTO_POSTPONE
;
3488 ovs_rwlock_unlock(&rule
->evict
);
3489 error
= OFPERR_OFPFMFC_TABLE_FULL
;
3496 group
= ofopgroup_create(ofproto
, ofconn
, request
, fm
->buffer_id
);
3497 op
= ofoperation_create(group
, rule
, OFOPERATION_ADD
, 0);
3498 op
->victim
= victim
;
3500 error
= ofproto
->ofproto_class
->rule_construct(rule
);
3502 op
->group
->n_running
--;
3503 ofoperation_destroy(rule
->pending
);
3505 /* It would be better if we maintained the lock we took in
3506 * choose_rule_to_evict() earlier, but that confuses the thread
3507 * safety analysis, and this code is fragile enough that we really
3508 * need it. In the worst case, we'll have to block a little while
3509 * before we perform the eviction, which doesn't seem like a big
3511 ovs_rwlock_wrlock(&evict
->evict
);
3512 delete_flow__(evict
, group
, OFPRR_EVICTION
);
3514 ofopgroup_submit(group
);
3518 /* Back out if an error occurred. */
3520 oftable_substitute_rule(rule
, victim
);
3521 ofproto_rule_destroy__(rule
);
3526 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3528 /* Modifies the rules listed in 'rules', changing their actions to match those
3531 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3534 * Returns 0 on success, otherwise an OpenFlow error code. */
3536 modify_flows__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3537 struct ofputil_flow_mod
*fm
, const struct ofp_header
*request
,
3540 struct ofopgroup
*group
;
3544 group
= ofopgroup_create(ofproto
, ofconn
, request
, fm
->buffer_id
);
3545 error
= OFPERR_OFPBRC_EPERM
;
3546 LIST_FOR_EACH (rule
, ofproto_node
, rules
) {
3547 struct ofoperation
*op
;
3548 bool actions_changed
;
3550 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3552 if (rule_is_modifiable(rule
)) {
3553 /* At least one rule is modifiable, don't report EPERM error. */
3559 /* Verify actions. */
3560 error
= ofpacts_check(fm
->ofpacts
, fm
->ofpacts_len
, &fm
->match
.flow
,
3561 ofproto
->max_ports
, rule
->table_id
);
3566 actions_changed
= !ofpacts_equal(fm
->ofpacts
, fm
->ofpacts_len
,
3567 rule
->ofpacts
, rule
->ofpacts_len
);
3569 op
= ofoperation_create(group
, rule
, OFOPERATION_MODIFY
, 0);
3571 if (fm
->modify_cookie
&& fm
->new_cookie
!= htonll(UINT64_MAX
)) {
3572 ofproto_rule_change_cookie(ofproto
, rule
, fm
->new_cookie
);
3574 if (actions_changed
) {
3575 op
->ofpacts
= rule
->ofpacts
;
3576 op
->ofpacts_len
= rule
->ofpacts_len
;
3577 op
->meter_id
= rule
->meter_id
;
3578 rule
->ofpacts
= xmemdup(fm
->ofpacts
, fm
->ofpacts_len
);
3579 rule
->ofpacts_len
= fm
->ofpacts_len
;
3580 rule
->meter_id
= find_meter(rule
->ofpacts
, rule
->ofpacts_len
);
3581 rule
->ofproto
->ofproto_class
->rule_modify_actions(rule
);
3583 ofoperation_complete(op
, 0);
3586 ofopgroup_submit(group
);
3592 modify_flows_add(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3593 struct ofputil_flow_mod
*fm
, const struct ofp_header
*request
)
3595 if (fm
->cookie_mask
!= htonll(0) || fm
->new_cookie
== htonll(UINT64_MAX
)) {
3598 return add_flow(ofproto
, ofconn
, fm
, request
);
3601 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3604 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3607 modify_flows_loose(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3608 struct ofputil_flow_mod
*fm
,
3609 const struct ofp_header
*request
)
3614 error
= collect_rules_loose(ofproto
, fm
->table_id
, &fm
->match
,
3615 fm
->cookie
, fm
->cookie_mask
,
3619 } else if (list_is_empty(&rules
)) {
3620 return modify_flows_add(ofproto
, ofconn
, fm
, request
);
3622 return modify_flows__(ofproto
, ofconn
, fm
, request
, &rules
);
3626 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3629 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3632 modify_flow_strict(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3633 struct ofputil_flow_mod
*fm
,
3634 const struct ofp_header
*request
)
3639 error
= collect_rules_strict(ofproto
, fm
->table_id
, &fm
->match
,
3640 fm
->priority
, fm
->cookie
, fm
->cookie_mask
,
3645 } else if (list_is_empty(&rules
)) {
3646 return modify_flows_add(ofproto
, ofconn
, fm
, request
);
3648 return list_is_singleton(&rules
) ? modify_flows__(ofproto
, ofconn
,
3649 fm
, request
, &rules
)
3654 /* OFPFC_DELETE implementation. */
3657 delete_flow__(struct rule
*rule
, struct ofopgroup
*group
,
3658 enum ofp_flow_removed_reason reason
)
3660 struct ofproto
*ofproto
= rule
->ofproto
;
3662 ofproto_rule_send_removed(rule
, reason
);
3664 ofoperation_create(group
, rule
, OFOPERATION_DELETE
, reason
);
3665 oftable_remove_rule(rule
);
3666 ofproto
->ofproto_class
->rule_destruct(rule
);
3669 /* Deletes the rules listed in 'rules'.
3671 * Returns 0 on success, otherwise an OpenFlow error code. */
3673 delete_flows__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3674 const struct ofp_header
*request
, struct list
*rules
,
3675 enum ofp_flow_removed_reason reason
)
3677 struct rule
*rule
, *next
;
3678 struct ofopgroup
*group
;
3680 group
= ofopgroup_create(ofproto
, ofconn
, request
, UINT32_MAX
);
3681 LIST_FOR_EACH_SAFE (rule
, next
, ofproto_node
, rules
) {
3682 ovs_rwlock_wrlock(&rule
->evict
);
3683 delete_flow__(rule
, group
, reason
);
3685 ofopgroup_submit(group
);
3690 /* Implements OFPFC_DELETE. */
3692 delete_flows_loose(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3693 const struct ofputil_flow_mod
*fm
,
3694 const struct ofp_header
*request
)
3699 error
= collect_rules_loose(ofproto
, fm
->table_id
, &fm
->match
,
3700 fm
->cookie
, fm
->cookie_mask
,
3701 fm
->out_port
, &rules
);
3702 return (error
? error
3703 : !list_is_empty(&rules
) ? delete_flows__(ofproto
, ofconn
, request
,
3704 &rules
, OFPRR_DELETE
)
3708 /* Implements OFPFC_DELETE_STRICT. */
3710 delete_flow_strict(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3711 const struct ofputil_flow_mod
*fm
,
3712 const struct ofp_header
*request
)
3717 error
= collect_rules_strict(ofproto
, fm
->table_id
, &fm
->match
,
3718 fm
->priority
, fm
->cookie
, fm
->cookie_mask
,
3719 fm
->out_port
, &rules
);
3720 return (error
? error
3721 : list_is_singleton(&rules
) ? delete_flows__(ofproto
, ofconn
,
3728 ofproto_rule_send_removed(struct rule
*rule
, uint8_t reason
)
3730 struct ofputil_flow_removed fr
;
3732 if (ofproto_rule_is_hidden(rule
) || !rule
->send_flow_removed
) {
3736 minimatch_expand(&rule
->cr
.match
, &fr
.match
);
3737 fr
.priority
= rule
->cr
.priority
;
3738 fr
.cookie
= rule
->flow_cookie
;
3740 fr
.table_id
= rule
->table_id
;
3741 calc_duration(rule
->created
, time_msec(),
3742 &fr
.duration_sec
, &fr
.duration_nsec
);
3743 ovs_mutex_lock(&rule
->timeout_mutex
);
3744 fr
.idle_timeout
= rule
->idle_timeout
;
3745 fr
.hard_timeout
= rule
->hard_timeout
;
3746 ovs_mutex_unlock(&rule
->timeout_mutex
);
3747 rule
->ofproto
->ofproto_class
->rule_get_stats(rule
, &fr
.packet_count
,
3750 connmgr_send_flow_removed(rule
->ofproto
->connmgr
, &fr
);
3754 ofproto_rule_update_used(struct rule
*rule
, long long int used
)
3756 if (used
> rule
->used
) {
3757 struct eviction_group
*evg
= rule
->eviction_group
;
3761 heap_change(&evg
->rules
, &rule
->evg_node
,
3762 rule_eviction_priority(rule
));
3767 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3768 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3771 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3774 * ofproto implementation ->run() functions should use this function to expire
3775 * OpenFlow flows. */
3777 ofproto_rule_expire(struct rule
*rule
, uint8_t reason
)
3779 struct ofproto
*ofproto
= rule
->ofproto
;
3780 struct ofopgroup
*group
;
3782 ovs_assert(reason
== OFPRR_HARD_TIMEOUT
|| reason
== OFPRR_IDLE_TIMEOUT
);
3784 ofproto_rule_send_removed(rule
, reason
);
3786 group
= ofopgroup_create_unattached(ofproto
);
3787 ofoperation_create(group
, rule
, OFOPERATION_DELETE
, reason
);
3788 oftable_remove_rule(rule
);
3789 ofproto
->ofproto_class
->rule_destruct(rule
);
3790 ofopgroup_submit(group
);
3794 handle_flow_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3796 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
3797 struct ofputil_flow_mod fm
;
3798 uint64_t ofpacts_stub
[1024 / 8];
3799 struct ofpbuf ofpacts
;
3803 error
= reject_slave_controller(ofconn
);
3808 ofpbuf_use_stub(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
3809 error
= ofputil_decode_flow_mod(&fm
, oh
, ofconn_get_protocol(ofconn
),
3812 error
= handle_flow_mod__(ofproto
, ofconn
, &fm
, oh
);
3815 goto exit_free_ofpacts
;
3818 /* Record the operation for logging a summary report. */
3819 switch (fm
.command
) {
3825 case OFPFC_MODIFY_STRICT
:
3826 ofproto
->n_modify
++;
3830 case OFPFC_DELETE_STRICT
:
3831 ofproto
->n_delete
++;
3836 if (ofproto
->next_op_report
== LLONG_MAX
) {
3837 ofproto
->first_op
= now
;
3838 ofproto
->next_op_report
= MAX(now
+ 10 * 1000,
3839 ofproto
->op_backoff
);
3840 ofproto
->op_backoff
= ofproto
->next_op_report
+ 60 * 1000;
3842 ofproto
->last_op
= now
;
3845 ofpbuf_uninit(&ofpacts
);
3851 handle_flow_mod__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3852 struct ofputil_flow_mod
*fm
, const struct ofp_header
*oh
)
3854 if (ofproto
->n_pending
>= 50) {
3855 ovs_assert(!list_is_empty(&ofproto
->pending
));
3856 return OFPROTO_POSTPONE
;
3859 switch (fm
->command
) {
3861 return add_flow(ofproto
, ofconn
, fm
, oh
);
3864 return modify_flows_loose(ofproto
, ofconn
, fm
, oh
);
3866 case OFPFC_MODIFY_STRICT
:
3867 return modify_flow_strict(ofproto
, ofconn
, fm
, oh
);
3870 return delete_flows_loose(ofproto
, ofconn
, fm
, oh
);
3872 case OFPFC_DELETE_STRICT
:
3873 return delete_flow_strict(ofproto
, ofconn
, fm
, oh
);
3876 if (fm
->command
> 0xff) {
3877 VLOG_WARN_RL(&rl
, "%s: flow_mod has explicit table_id but "
3878 "flow_mod_table_id extension is not enabled",
3881 return OFPERR_OFPFMFC_BAD_COMMAND
;
3886 handle_role_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3888 struct ofputil_role_request request
;
3889 struct ofputil_role_request reply
;
3893 error
= ofputil_decode_role_message(oh
, &request
);
3898 if (request
.role
!= OFPCR12_ROLE_NOCHANGE
) {
3899 if (ofconn_get_role(ofconn
) != request
.role
3900 && ofconn_has_pending_opgroups(ofconn
)) {
3901 return OFPROTO_POSTPONE
;
3904 if (request
.have_generation_id
3905 && !ofconn_set_master_election_id(ofconn
, request
.generation_id
)) {
3906 return OFPERR_OFPRRFC_STALE
;
3909 ofconn_set_role(ofconn
, request
.role
);
3912 reply
.role
= ofconn_get_role(ofconn
);
3913 reply
.have_generation_id
= ofconn_get_master_election_id(
3914 ofconn
, &reply
.generation_id
);
3915 buf
= ofputil_encode_role_reply(oh
, &reply
);
3916 ofconn_send_reply(ofconn
, buf
);
3922 handle_nxt_flow_mod_table_id(struct ofconn
*ofconn
,
3923 const struct ofp_header
*oh
)
3925 const struct nx_flow_mod_table_id
*msg
= ofpmsg_body(oh
);
3926 enum ofputil_protocol cur
, next
;
3928 cur
= ofconn_get_protocol(ofconn
);
3929 next
= ofputil_protocol_set_tid(cur
, msg
->set
!= 0);
3930 ofconn_set_protocol(ofconn
, next
);
3936 handle_nxt_set_flow_format(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3938 const struct nx_set_flow_format
*msg
= ofpmsg_body(oh
);
3939 enum ofputil_protocol cur
, next
;
3940 enum ofputil_protocol next_base
;
3942 next_base
= ofputil_nx_flow_format_to_protocol(ntohl(msg
->format
));
3944 return OFPERR_OFPBRC_EPERM
;
3947 cur
= ofconn_get_protocol(ofconn
);
3948 next
= ofputil_protocol_set_base(cur
, next_base
);
3949 if (cur
!= next
&& ofconn_has_pending_opgroups(ofconn
)) {
3950 /* Avoid sending async messages in surprising protocol. */
3951 return OFPROTO_POSTPONE
;
3954 ofconn_set_protocol(ofconn
, next
);
3959 handle_nxt_set_packet_in_format(struct ofconn
*ofconn
,
3960 const struct ofp_header
*oh
)
3962 const struct nx_set_packet_in_format
*msg
= ofpmsg_body(oh
);
3965 format
= ntohl(msg
->format
);
3966 if (format
!= NXPIF_OPENFLOW10
&& format
!= NXPIF_NXM
) {
3967 return OFPERR_OFPBRC_EPERM
;
3970 if (format
!= ofconn_get_packet_in_format(ofconn
)
3971 && ofconn_has_pending_opgroups(ofconn
)) {
3972 /* Avoid sending async message in surprsing packet in format. */
3973 return OFPROTO_POSTPONE
;
3976 ofconn_set_packet_in_format(ofconn
, format
);
3981 handle_nxt_set_async_config(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3983 const struct nx_async_config
*msg
= ofpmsg_body(oh
);
3984 uint32_t master
[OAM_N_TYPES
];
3985 uint32_t slave
[OAM_N_TYPES
];
3987 master
[OAM_PACKET_IN
] = ntohl(msg
->packet_in_mask
[0]);
3988 master
[OAM_PORT_STATUS
] = ntohl(msg
->port_status_mask
[0]);
3989 master
[OAM_FLOW_REMOVED
] = ntohl(msg
->flow_removed_mask
[0]);
3991 slave
[OAM_PACKET_IN
] = ntohl(msg
->packet_in_mask
[1]);
3992 slave
[OAM_PORT_STATUS
] = ntohl(msg
->port_status_mask
[1]);
3993 slave
[OAM_FLOW_REMOVED
] = ntohl(msg
->flow_removed_mask
[1]);
3995 ofconn_set_async_config(ofconn
, master
, slave
);
3996 if (ofconn_get_type(ofconn
) == OFCONN_SERVICE
&&
3997 !ofconn_get_miss_send_len(ofconn
)) {
3998 ofconn_set_miss_send_len(ofconn
, OFP_DEFAULT_MISS_SEND_LEN
);
4005 handle_nxt_set_controller_id(struct ofconn
*ofconn
,
4006 const struct ofp_header
*oh
)
4008 const struct nx_controller_id
*nci
= ofpmsg_body(oh
);
4010 if (!is_all_zeros(nci
->zero
, sizeof nci
->zero
)) {
4011 return OFPERR_NXBRC_MUST_BE_ZERO
;
4014 ofconn_set_controller_id(ofconn
, ntohs(nci
->controller_id
));
4019 handle_barrier_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4023 if (ofconn_has_pending_opgroups(ofconn
)) {
4024 return OFPROTO_POSTPONE
;
4027 buf
= ofpraw_alloc_reply((oh
->version
== OFP10_VERSION
4028 ? OFPRAW_OFPT10_BARRIER_REPLY
4029 : OFPRAW_OFPT11_BARRIER_REPLY
), oh
, 0);
4030 ofconn_send_reply(ofconn
, buf
);
4035 ofproto_compose_flow_refresh_update(const struct rule
*rule
,
4036 enum nx_flow_monitor_flags flags
,
4039 struct ofoperation
*op
= rule
->pending
;
4040 struct ofputil_flow_update fu
;
4043 if (op
&& op
->type
== OFOPERATION_ADD
&& !op
->victim
) {
4044 /* We'll report the final flow when the operation completes. Reporting
4045 * it now would cause a duplicate report later. */
4049 fu
.event
= (flags
& (NXFMF_INITIAL
| NXFMF_ADD
)
4050 ? NXFME_ADDED
: NXFME_MODIFIED
);
4052 ovs_mutex_lock(&rule
->timeout_mutex
);
4053 fu
.idle_timeout
= rule
->idle_timeout
;
4054 fu
.hard_timeout
= rule
->hard_timeout
;
4055 ovs_mutex_unlock(&rule
->timeout_mutex
);
4056 fu
.table_id
= rule
->table_id
;
4057 fu
.cookie
= rule
->flow_cookie
;
4058 minimatch_expand(&rule
->cr
.match
, &match
);
4060 fu
.priority
= rule
->cr
.priority
;
4061 if (!(flags
& NXFMF_ACTIONS
)) {
4065 fu
.ofpacts
= rule
->ofpacts
;
4066 fu
.ofpacts_len
= rule
->ofpacts_len
;
4068 /* An operation is in progress. Use the previous version of the flow's
4069 * actions, so that when the operation commits we report the change. */
4071 case OFOPERATION_ADD
:
4072 /* We already verified that there was a victim. */
4073 fu
.ofpacts
= op
->victim
->ofpacts
;
4074 fu
.ofpacts_len
= op
->victim
->ofpacts_len
;
4077 case OFOPERATION_MODIFY
:
4079 fu
.ofpacts
= op
->ofpacts
;
4080 fu
.ofpacts_len
= op
->ofpacts_len
;
4082 fu
.ofpacts
= rule
->ofpacts
;
4083 fu
.ofpacts_len
= rule
->ofpacts_len
;
4087 case OFOPERATION_DELETE
:
4088 fu
.ofpacts
= rule
->ofpacts
;
4089 fu
.ofpacts_len
= rule
->ofpacts_len
;
4097 if (list_is_empty(msgs
)) {
4098 ofputil_start_flow_update(msgs
);
4100 ofputil_append_flow_update(&fu
, msgs
);
4104 ofmonitor_compose_refresh_updates(struct list
*rules
, struct list
*msgs
)
4108 LIST_FOR_EACH (rule
, ofproto_node
, rules
) {
4109 enum nx_flow_monitor_flags flags
= rule
->monitor_flags
;
4110 rule
->monitor_flags
= 0;
4112 ofproto_compose_flow_refresh_update(rule
, flags
, msgs
);
4117 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor
*m
,
4118 struct rule
*rule
, uint64_t seqno
,
4121 enum nx_flow_monitor_flags update
;
4123 if (ofproto_rule_is_hidden(rule
)) {
4128 ? ofoperation_has_out_port(rule
->pending
, m
->out_port
)
4129 : ofproto_rule_has_out_port(rule
, m
->out_port
))) {
4134 if (rule
->add_seqno
> seqno
) {
4135 update
= NXFMF_ADD
| NXFMF_MODIFY
;
4136 } else if (rule
->modify_seqno
> seqno
) {
4137 update
= NXFMF_MODIFY
;
4142 if (!(m
->flags
& update
)) {
4146 update
= NXFMF_INITIAL
;
4149 if (!rule
->monitor_flags
) {
4150 list_push_back(rules
, &rule
->ofproto_node
);
4152 rule
->monitor_flags
|= update
| (m
->flags
& NXFMF_ACTIONS
);
4156 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor
*m
,
4160 const struct ofproto
*ofproto
= ofconn_get_ofproto(m
->ofconn
);
4161 const struct ofoperation
*op
;
4162 const struct oftable
*table
;
4163 struct cls_rule target
;
4165 cls_rule_init_from_minimatch(&target
, &m
->match
, 0);
4166 FOR_EACH_MATCHING_TABLE (table
, m
->table_id
, ofproto
) {
4167 struct cls_cursor cursor
;
4170 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
4171 cls_cursor_init(&cursor
, &table
->cls
, &target
);
4172 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
4173 ovs_assert(!rule
->pending
); /* XXX */
4174 ofproto_collect_ofmonitor_refresh_rule(m
, rule
, seqno
, rules
);
4176 ovs_rwlock_unlock(&table
->cls
.rwlock
);
4179 HMAP_FOR_EACH (op
, hmap_node
, &ofproto
->deletions
) {
4180 struct rule
*rule
= op
->rule
;
4182 if (((m
->table_id
== 0xff
4183 ? !(ofproto
->tables
[rule
->table_id
].flags
& OFTABLE_HIDDEN
)
4184 : m
->table_id
== rule
->table_id
))
4185 && cls_rule_is_loose_match(&rule
->cr
, &target
.match
)) {
4186 ofproto_collect_ofmonitor_refresh_rule(m
, rule
, seqno
, rules
);
4189 cls_rule_destroy(&target
);
4193 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor
*m
,
4196 if (m
->flags
& NXFMF_INITIAL
) {
4197 ofproto_collect_ofmonitor_refresh_rules(m
, 0, rules
);
4202 ofmonitor_collect_resume_rules(struct ofmonitor
*m
,
4203 uint64_t seqno
, struct list
*rules
)
4205 ofproto_collect_ofmonitor_refresh_rules(m
, seqno
, rules
);
4209 handle_flow_monitor_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4211 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
4212 struct ofmonitor
**monitors
;
4213 size_t n_monitors
, allocated_monitors
;
4214 struct list replies
;
4221 ofpbuf_use_const(&b
, oh
, ntohs(oh
->length
));
4223 n_monitors
= allocated_monitors
= 0;
4225 struct ofputil_flow_monitor_request request
;
4226 struct ofmonitor
*m
;
4229 retval
= ofputil_decode_flow_monitor_request(&request
, &b
);
4230 if (retval
== EOF
) {
4232 } else if (retval
) {
4237 if (request
.table_id
!= 0xff
4238 && request
.table_id
>= ofproto
->n_tables
) {
4239 error
= OFPERR_OFPBRC_BAD_TABLE_ID
;
4243 error
= ofmonitor_create(&request
, ofconn
, &m
);
4248 if (n_monitors
>= allocated_monitors
) {
4249 monitors
= x2nrealloc(monitors
, &allocated_monitors
,
4252 monitors
[n_monitors
++] = m
;
4256 for (i
= 0; i
< n_monitors
; i
++) {
4257 ofproto_collect_ofmonitor_initial_rules(monitors
[i
], &rules
);
4260 ofpmp_init(&replies
, oh
);
4261 ofmonitor_compose_refresh_updates(&rules
, &replies
);
4262 ofconn_send_replies(ofconn
, &replies
);
4269 for (i
= 0; i
< n_monitors
; i
++) {
4270 ofmonitor_destroy(monitors
[i
]);
4277 handle_flow_monitor_cancel(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4279 struct ofmonitor
*m
;
4282 id
= ofputil_decode_flow_monitor_cancel(oh
);
4283 m
= ofmonitor_lookup(ofconn
, id
);
4285 return OFPERR_NXBRC_FM_BAD_ID
;
4288 ofmonitor_destroy(m
);
4292 /* Meters implementation.
4294 * Meter table entry, indexed by the OpenFlow meter_id.
4295 * These are always dynamically allocated to allocate enough space for
4297 * 'created' is used to compute the duration for meter stats.
4298 * 'list rules' is needed so that we can delete the dependent rules when the
4299 * meter table entry is deleted.
4300 * 'provider_meter_id' is for the provider's private use.
4303 long long int created
; /* Time created. */
4304 struct list rules
; /* List of "struct rule_dpif"s. */
4305 ofproto_meter_id provider_meter_id
;
4306 uint16_t flags
; /* Meter flags. */
4307 uint16_t n_bands
; /* Number of meter bands. */
4308 struct ofputil_meter_band
*bands
;
4312 * This is used in instruction validation at flow set-up time,
4313 * as flows may not use non-existing meters.
4314 * This is also used by ofproto-providers to translate OpenFlow meter_ids
4315 * in METER instructions to the corresponding provider meter IDs.
4316 * Return value of UINT32_MAX signifies an invalid meter.
4319 ofproto_get_provider_meter_id(const struct ofproto
* ofproto
,
4320 uint32_t of_meter_id
)
4322 if (of_meter_id
&& of_meter_id
<= ofproto
->meter_features
.max_meters
) {
4323 const struct meter
*meter
= ofproto
->meters
[of_meter_id
];
4325 return meter
->provider_meter_id
.uint32
;
4332 meter_update(struct meter
*meter
, const struct ofputil_meter_config
*config
)
4336 meter
->flags
= config
->flags
;
4337 meter
->n_bands
= config
->n_bands
;
4338 meter
->bands
= xmemdup(config
->bands
,
4339 config
->n_bands
* sizeof *meter
->bands
);
4342 static struct meter
*
4343 meter_create(const struct ofputil_meter_config
*config
,
4344 ofproto_meter_id provider_meter_id
)
4346 struct meter
*meter
;
4348 meter
= xzalloc(sizeof *meter
);
4349 meter
->provider_meter_id
= provider_meter_id
;
4350 meter
->created
= time_msec();
4351 list_init(&meter
->rules
);
4353 meter_update(meter
, config
);
4359 meter_delete(struct ofproto
*ofproto
, uint32_t first
, uint32_t last
)
4362 for (mid
= first
; mid
<= last
; ++mid
) {
4363 struct meter
*meter
= ofproto
->meters
[mid
];
4365 ofproto
->meters
[mid
] = NULL
;
4366 ofproto
->ofproto_class
->meter_del(ofproto
,
4367 meter
->provider_meter_id
);
4375 handle_add_meter(struct ofproto
*ofproto
, struct ofputil_meter_mod
*mm
)
4377 ofproto_meter_id provider_meter_id
= { UINT32_MAX
};
4378 struct meter
**meterp
= &ofproto
->meters
[mm
->meter
.meter_id
];
4382 return OFPERR_OFPMMFC_METER_EXISTS
;
4385 error
= ofproto
->ofproto_class
->meter_set(ofproto
, &provider_meter_id
,
4388 ovs_assert(provider_meter_id
.uint32
!= UINT32_MAX
);
4389 *meterp
= meter_create(&mm
->meter
, provider_meter_id
);
4395 handle_modify_meter(struct ofproto
*ofproto
, struct ofputil_meter_mod
*mm
)
4397 struct meter
*meter
= ofproto
->meters
[mm
->meter
.meter_id
];
4401 return OFPERR_OFPMMFC_UNKNOWN_METER
;
4404 error
= ofproto
->ofproto_class
->meter_set(ofproto
,
4405 &meter
->provider_meter_id
,
4407 ovs_assert(meter
->provider_meter_id
.uint32
!= UINT32_MAX
);
4409 meter_update(meter
, &mm
->meter
);
4415 handle_delete_meter(struct ofconn
*ofconn
, const struct ofp_header
*oh
,
4416 struct ofputil_meter_mod
*mm
)
4418 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
4419 uint32_t meter_id
= mm
->meter
.meter_id
;
4420 uint32_t first
, last
;
4423 if (meter_id
== OFPM13_ALL
) {
4425 last
= ofproto
->meter_features
.max_meters
;
4427 if (!meter_id
|| meter_id
> ofproto
->meter_features
.max_meters
) {
4430 first
= last
= meter_id
;
4433 /* First delete the rules that use this meter. If any of those rules are
4434 * currently being modified, postpone the whole operation until later. */
4436 for (meter_id
= first
; meter_id
<= last
; ++meter_id
) {
4437 struct meter
*meter
= ofproto
->meters
[meter_id
];
4438 if (meter
&& !list_is_empty(&meter
->rules
)) {
4441 LIST_FOR_EACH (rule
, meter_list_node
, &meter
->rules
) {
4442 if (rule
->pending
) {
4443 return OFPROTO_POSTPONE
;
4445 list_push_back(&rules
, &rule
->ofproto_node
);
4449 if (!list_is_empty(&rules
)) {
4450 delete_flows__(ofproto
, ofconn
, oh
, &rules
, OFPRR_METER_DELETE
);
4453 /* Delete the meters. */
4454 meter_delete(ofproto
, first
, last
);
4460 handle_meter_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4462 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
4463 struct ofputil_meter_mod mm
;
4464 uint64_t bands_stub
[256 / 8];
4465 struct ofpbuf bands
;
4469 error
= reject_slave_controller(ofconn
);
4474 ofpbuf_use_stub(&bands
, bands_stub
, sizeof bands_stub
);
4476 error
= ofputil_decode_meter_mod(oh
, &mm
, &bands
);
4478 goto exit_free_bands
;
4481 meter_id
= mm
.meter
.meter_id
;
4483 if (mm
.command
!= OFPMC13_DELETE
) {
4484 /* Fails also when meters are not implemented by the provider. */
4485 if (!meter_id
|| meter_id
> ofproto
->meter_features
.max_meters
) {
4486 error
= OFPERR_OFPMMFC_INVALID_METER
;
4487 goto exit_free_bands
;
4489 if (mm
.meter
.n_bands
> ofproto
->meter_features
.max_bands
) {
4490 error
= OFPERR_OFPMMFC_OUT_OF_BANDS
;
4491 goto exit_free_bands
;
4495 switch (mm
.command
) {
4497 error
= handle_add_meter(ofproto
, &mm
);
4500 case OFPMC13_MODIFY
:
4501 error
= handle_modify_meter(ofproto
, &mm
);
4504 case OFPMC13_DELETE
:
4505 error
= handle_delete_meter(ofconn
, oh
, &mm
);
4509 error
= OFPERR_OFPMMFC_BAD_COMMAND
;
4514 ofpbuf_uninit(&bands
);
4519 handle_meter_features_request(struct ofconn
*ofconn
,
4520 const struct ofp_header
*request
)
4522 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
4523 struct ofputil_meter_features features
;
4526 if (ofproto
->ofproto_class
->meter_get_features
) {
4527 ofproto
->ofproto_class
->meter_get_features(ofproto
, &features
);
4529 memset(&features
, 0, sizeof features
);
4531 b
= ofputil_encode_meter_features_reply(&features
, request
);
4533 ofconn_send_reply(ofconn
, b
);
4538 handle_meter_request(struct ofconn
*ofconn
, const struct ofp_header
*request
,
4541 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
4542 struct list replies
;
4543 uint64_t bands_stub
[256 / 8];
4544 struct ofpbuf bands
;
4545 uint32_t meter_id
, first
, last
;
4547 ofputil_decode_meter_request(request
, &meter_id
);
4549 if (meter_id
== OFPM13_ALL
) {
4551 last
= ofproto
->meter_features
.max_meters
;
4553 if (!meter_id
|| meter_id
> ofproto
->meter_features
.max_meters
||
4554 !ofproto
->meters
[meter_id
]) {
4555 return OFPERR_OFPMMFC_UNKNOWN_METER
;
4557 first
= last
= meter_id
;
4560 ofpbuf_use_stub(&bands
, bands_stub
, sizeof bands_stub
);
4561 ofpmp_init(&replies
, request
);
4563 for (meter_id
= first
; meter_id
<= last
; ++meter_id
) {
4564 struct meter
*meter
= ofproto
->meters
[meter_id
];
4566 continue; /* Skip non-existing meters. */
4568 if (type
== OFPTYPE_METER_STATS_REQUEST
) {
4569 struct ofputil_meter_stats stats
;
4571 stats
.meter_id
= meter_id
;
4573 /* Provider sets the packet and byte counts, we do the rest. */
4574 stats
.flow_count
= list_size(&meter
->rules
);
4575 calc_duration(meter
->created
, time_msec(),
4576 &stats
.duration_sec
, &stats
.duration_nsec
);
4577 stats
.n_bands
= meter
->n_bands
;
4578 ofpbuf_clear(&bands
);
4580 = ofpbuf_put_uninit(&bands
,
4581 meter
->n_bands
* sizeof *stats
.bands
);
4583 if (!ofproto
->ofproto_class
->meter_get(ofproto
,
4584 meter
->provider_meter_id
,
4586 ofputil_append_meter_stats(&replies
, &stats
);
4588 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
4589 struct ofputil_meter_config config
;
4591 config
.meter_id
= meter_id
;
4592 config
.flags
= meter
->flags
;
4593 config
.n_bands
= meter
->n_bands
;
4594 config
.bands
= meter
->bands
;
4595 ofputil_append_meter_config(&replies
, &config
);
4599 ofconn_send_replies(ofconn
, &replies
);
4600 ofpbuf_uninit(&bands
);
4605 handle_openflow__(struct ofconn
*ofconn
, const struct ofpbuf
*msg
)
4607 const struct ofp_header
*oh
= msg
->data
;
4611 error
= ofptype_decode(&type
, oh
);
4617 /* OpenFlow requests. */
4618 case OFPTYPE_ECHO_REQUEST
:
4619 return handle_echo_request(ofconn
, oh
);
4621 case OFPTYPE_FEATURES_REQUEST
:
4622 return handle_features_request(ofconn
, oh
);
4624 case OFPTYPE_GET_CONFIG_REQUEST
:
4625 return handle_get_config_request(ofconn
, oh
);
4627 case OFPTYPE_SET_CONFIG
:
4628 return handle_set_config(ofconn
, oh
);
4630 case OFPTYPE_PACKET_OUT
:
4631 return handle_packet_out(ofconn
, oh
);
4633 case OFPTYPE_PORT_MOD
:
4634 return handle_port_mod(ofconn
, oh
);
4636 case OFPTYPE_FLOW_MOD
:
4637 return handle_flow_mod(ofconn
, oh
);
4639 case OFPTYPE_METER_MOD
:
4640 return handle_meter_mod(ofconn
, oh
);
4642 case OFPTYPE_BARRIER_REQUEST
:
4643 return handle_barrier_request(ofconn
, oh
);
4645 case OFPTYPE_ROLE_REQUEST
:
4646 return handle_role_request(ofconn
, oh
);
4648 /* OpenFlow replies. */
4649 case OFPTYPE_ECHO_REPLY
:
4652 /* Nicira extension requests. */
4653 case OFPTYPE_FLOW_MOD_TABLE_ID
:
4654 return handle_nxt_flow_mod_table_id(ofconn
, oh
);
4656 case OFPTYPE_SET_FLOW_FORMAT
:
4657 return handle_nxt_set_flow_format(ofconn
, oh
);
4659 case OFPTYPE_SET_PACKET_IN_FORMAT
:
4660 return handle_nxt_set_packet_in_format(ofconn
, oh
);
4662 case OFPTYPE_SET_CONTROLLER_ID
:
4663 return handle_nxt_set_controller_id(ofconn
, oh
);
4665 case OFPTYPE_FLOW_AGE
:
4666 /* Nothing to do. */
4669 case OFPTYPE_FLOW_MONITOR_CANCEL
:
4670 return handle_flow_monitor_cancel(ofconn
, oh
);
4672 case OFPTYPE_SET_ASYNC_CONFIG
:
4673 return handle_nxt_set_async_config(ofconn
, oh
);
4675 /* Statistics requests. */
4676 case OFPTYPE_DESC_STATS_REQUEST
:
4677 return handle_desc_stats_request(ofconn
, oh
);
4679 case OFPTYPE_FLOW_STATS_REQUEST
:
4680 return handle_flow_stats_request(ofconn
, oh
);
4682 case OFPTYPE_AGGREGATE_STATS_REQUEST
:
4683 return handle_aggregate_stats_request(ofconn
, oh
);
4685 case OFPTYPE_TABLE_STATS_REQUEST
:
4686 return handle_table_stats_request(ofconn
, oh
);
4688 case OFPTYPE_PORT_STATS_REQUEST
:
4689 return handle_port_stats_request(ofconn
, oh
);
4691 case OFPTYPE_QUEUE_STATS_REQUEST
:
4692 return handle_queue_stats_request(ofconn
, oh
);
4694 case OFPTYPE_PORT_DESC_STATS_REQUEST
:
4695 return handle_port_desc_stats_request(ofconn
, oh
);
4697 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST
:
4698 return handle_flow_monitor_request(ofconn
, oh
);
4700 case OFPTYPE_METER_STATS_REQUEST
:
4701 case OFPTYPE_METER_CONFIG_STATS_REQUEST
:
4702 return handle_meter_request(ofconn
, oh
, type
);
4704 case OFPTYPE_METER_FEATURES_STATS_REQUEST
:
4705 return handle_meter_features_request(ofconn
, oh
);
4707 /* FIXME: Change the following once they are implemented: */
4708 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST
:
4709 case OFPTYPE_GET_ASYNC_REQUEST
:
4710 case OFPTYPE_GROUP_STATS_REQUEST
:
4711 case OFPTYPE_GROUP_DESC_STATS_REQUEST
:
4712 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST
:
4713 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST
:
4714 return OFPERR_OFPBRC_BAD_TYPE
;
4718 case OFPTYPE_FEATURES_REPLY
:
4719 case OFPTYPE_GET_CONFIG_REPLY
:
4720 case OFPTYPE_PACKET_IN
:
4721 case OFPTYPE_FLOW_REMOVED
:
4722 case OFPTYPE_PORT_STATUS
:
4723 case OFPTYPE_BARRIER_REPLY
:
4724 case OFPTYPE_QUEUE_GET_CONFIG_REPLY
:
4725 case OFPTYPE_DESC_STATS_REPLY
:
4726 case OFPTYPE_FLOW_STATS_REPLY
:
4727 case OFPTYPE_QUEUE_STATS_REPLY
:
4728 case OFPTYPE_PORT_STATS_REPLY
:
4729 case OFPTYPE_TABLE_STATS_REPLY
:
4730 case OFPTYPE_AGGREGATE_STATS_REPLY
:
4731 case OFPTYPE_PORT_DESC_STATS_REPLY
:
4732 case OFPTYPE_ROLE_REPLY
:
4733 case OFPTYPE_FLOW_MONITOR_PAUSED
:
4734 case OFPTYPE_FLOW_MONITOR_RESUMED
:
4735 case OFPTYPE_FLOW_MONITOR_STATS_REPLY
:
4736 case OFPTYPE_GET_ASYNC_REPLY
:
4737 case OFPTYPE_GROUP_STATS_REPLY
:
4738 case OFPTYPE_GROUP_DESC_STATS_REPLY
:
4739 case OFPTYPE_GROUP_FEATURES_STATS_REPLY
:
4740 case OFPTYPE_METER_STATS_REPLY
:
4741 case OFPTYPE_METER_CONFIG_STATS_REPLY
:
4742 case OFPTYPE_METER_FEATURES_STATS_REPLY
:
4743 case OFPTYPE_TABLE_FEATURES_STATS_REPLY
:
4745 return OFPERR_OFPBRC_BAD_TYPE
;
4750 handle_openflow(struct ofconn
*ofconn
, const struct ofpbuf
*ofp_msg
)
4752 int error
= handle_openflow__(ofconn
, ofp_msg
);
4753 if (error
&& error
!= OFPROTO_POSTPONE
) {
4754 ofconn_send_error(ofconn
, ofp_msg
->data
, error
);
4756 COVERAGE_INC(ofproto_recv_openflow
);
4757 return error
!= OFPROTO_POSTPONE
;
4760 /* Asynchronous operations. */
4762 /* Creates and returns a new ofopgroup that is not associated with any
4763 * OpenFlow connection.
4765 * The caller should add operations to the returned group with
4766 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4767 static struct ofopgroup
*
4768 ofopgroup_create_unattached(struct ofproto
*ofproto
)
4770 struct ofopgroup
*group
= xzalloc(sizeof *group
);
4771 group
->ofproto
= ofproto
;
4772 list_init(&group
->ofproto_node
);
4773 list_init(&group
->ops
);
4774 list_init(&group
->ofconn_node
);
4778 /* Creates and returns a new ofopgroup for 'ofproto'.
4780 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
4781 * connection. The 'request' and 'buffer_id' arguments are ignored.
4783 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
4784 * If the ofopgroup eventually fails, then the error reply will include
4785 * 'request'. If the ofopgroup eventually succeeds, then the packet with
4786 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
4788 * The caller should add operations to the returned group with
4789 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4790 static struct ofopgroup
*
4791 ofopgroup_create(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
4792 const struct ofp_header
*request
, uint32_t buffer_id
)
4794 struct ofopgroup
*group
= ofopgroup_create_unattached(ofproto
);
4796 size_t request_len
= ntohs(request
->length
);
4798 ovs_assert(ofconn_get_ofproto(ofconn
) == ofproto
);
4800 ofconn_add_opgroup(ofconn
, &group
->ofconn_node
);
4801 group
->ofconn
= ofconn
;
4802 group
->request
= xmemdup(request
, MIN(request_len
, 64));
4803 group
->buffer_id
= buffer_id
;
4808 /* Submits 'group' for processing.
4810 * If 'group' contains no operations (e.g. none were ever added, or all of the
4811 * ones that were added completed synchronously), then it is destroyed
4812 * immediately. Otherwise it is added to the ofproto's list of pending
4815 ofopgroup_submit(struct ofopgroup
*group
)
4817 if (!group
->n_running
) {
4818 ofopgroup_complete(group
);
4820 list_push_back(&group
->ofproto
->pending
, &group
->ofproto_node
);
4821 group
->ofproto
->n_pending
++;
4826 ofopgroup_complete(struct ofopgroup
*group
)
4828 struct ofproto
*ofproto
= group
->ofproto
;
4830 struct ofconn
*abbrev_ofconn
;
4831 ovs_be32 abbrev_xid
;
4833 struct ofoperation
*op
, *next_op
;
4836 ovs_assert(!group
->n_running
);
4839 LIST_FOR_EACH (op
, group_node
, &group
->ops
) {
4846 if (!error
&& group
->ofconn
&& group
->buffer_id
!= UINT32_MAX
) {
4847 LIST_FOR_EACH (op
, group_node
, &group
->ops
) {
4848 if (op
->type
!= OFOPERATION_DELETE
) {
4849 struct ofpbuf
*packet
;
4852 error
= ofconn_pktbuf_retrieve(group
->ofconn
, group
->buffer_id
,
4856 error
= rule_execute(op
->rule
, in_port
, packet
);
4863 if (!error
&& !list_is_empty(&group
->ofconn_node
)) {
4864 abbrev_ofconn
= group
->ofconn
;
4865 abbrev_xid
= group
->request
->xid
;
4867 abbrev_ofconn
= NULL
;
4868 abbrev_xid
= htonl(0);
4870 LIST_FOR_EACH_SAFE (op
, next_op
, group_node
, &group
->ops
) {
4871 struct rule
*rule
= op
->rule
;
4873 /* We generally want to report the change to active OpenFlow flow
4874 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
4876 - The operation failed.
4878 - The affected rule is not visible to controllers.
4880 - The operation's only effect was to update rule->modified. */
4882 || ofproto_rule_is_hidden(rule
)
4883 || (op
->type
== OFOPERATION_MODIFY
4885 && rule
->flow_cookie
== op
->flow_cookie
))) {
4886 /* Check that we can just cast from ofoperation_type to
4887 * nx_flow_update_event. */
4888 BUILD_ASSERT_DECL((enum nx_flow_update_event
) OFOPERATION_ADD
4890 BUILD_ASSERT_DECL((enum nx_flow_update_event
) OFOPERATION_DELETE
4892 BUILD_ASSERT_DECL((enum nx_flow_update_event
) OFOPERATION_MODIFY
4895 ofmonitor_report(ofproto
->connmgr
, rule
,
4896 (enum nx_flow_update_event
) op
->type
,
4897 op
->reason
, abbrev_ofconn
, abbrev_xid
);
4900 rule
->pending
= NULL
;
4903 case OFOPERATION_ADD
:
4907 ofproto_rule_destroy__(op
->victim
);
4908 vid_mask
= minimask_get_vid_mask(&rule
->cr
.match
.mask
);
4909 if (vid_mask
== VLAN_VID_MASK
) {
4910 if (ofproto
->vlan_bitmap
) {
4911 uint16_t vid
= miniflow_get_vid(&rule
->cr
.match
.flow
);
4912 if (!bitmap_is_set(ofproto
->vlan_bitmap
, vid
)) {
4913 bitmap_set1(ofproto
->vlan_bitmap
, vid
);
4914 ofproto
->vlans_changed
= true;
4917 ofproto
->vlans_changed
= true;
4921 oftable_substitute_rule(rule
, op
->victim
);
4922 ofproto_rule_destroy__(rule
);
4926 case OFOPERATION_DELETE
:
4927 ovs_assert(!op
->error
);
4928 ofproto_rule_destroy__(rule
);
4932 case OFOPERATION_MODIFY
:
4934 rule
->modified
= time_msec();
4936 ofproto_rule_change_cookie(ofproto
, rule
, op
->flow_cookie
);
4938 free(rule
->ofpacts
);
4939 rule
->ofpacts
= op
->ofpacts
;
4940 rule
->ofpacts_len
= op
->ofpacts_len
;
4942 op
->ofpacts_len
= 0;
4951 ofoperation_destroy(op
);
4954 ofmonitor_flush(ofproto
->connmgr
);
4956 if (!list_is_empty(&group
->ofproto_node
)) {
4957 ovs_assert(ofproto
->n_pending
> 0);
4958 ofproto
->n_pending
--;
4959 list_remove(&group
->ofproto_node
);
4961 if (!list_is_empty(&group
->ofconn_node
)) {
4962 list_remove(&group
->ofconn_node
);
4964 ofconn_send_error(group
->ofconn
, group
->request
, error
);
4966 connmgr_retry(ofproto
->connmgr
);
4968 free(group
->request
);
4972 /* Initiates a new operation on 'rule', of the specified 'type', within
4973 * 'group'. Prior to calling, 'rule' must not have any pending operation.
4975 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
4976 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
4978 * Returns the newly created ofoperation (which is also available as
4979 * rule->pending). */
4980 static struct ofoperation
*
4981 ofoperation_create(struct ofopgroup
*group
, struct rule
*rule
,
4982 enum ofoperation_type type
,
4983 enum ofp_flow_removed_reason reason
)
4985 struct ofproto
*ofproto
= group
->ofproto
;
4986 struct ofoperation
*op
;
4988 ovs_assert(!rule
->pending
);
4990 op
= rule
->pending
= xzalloc(sizeof *op
);
4992 list_push_back(&group
->ops
, &op
->group_node
);
4995 op
->reason
= reason
;
4996 op
->flow_cookie
= rule
->flow_cookie
;
5000 if (type
== OFOPERATION_DELETE
) {
5001 hmap_insert(&ofproto
->deletions
, &op
->hmap_node
,
5002 cls_rule_hash(&rule
->cr
, rule
->table_id
));
5009 ofoperation_destroy(struct ofoperation
*op
)
5011 struct ofopgroup
*group
= op
->group
;
5014 op
->rule
->pending
= NULL
;
5016 if (op
->type
== OFOPERATION_DELETE
) {
5017 hmap_remove(&group
->ofproto
->deletions
, &op
->hmap_node
);
5019 list_remove(&op
->group_node
);
5024 /* Indicates that 'op' completed with status 'error', which is either 0 to
5025 * indicate success or an OpenFlow error code on failure.
5027 * If 'error' is 0, indicating success, the operation will be committed
5028 * permanently to the flow table. There is one interesting subcase:
5030 * - If 'op' is an "add flow" operation that is replacing an existing rule in
5031 * the flow table (the "victim" rule) by a new one, then the caller must
5032 * have uninitialized any derived state in the victim rule, as in step 5 in
5033 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
5034 * performs steps 6 and 7 for the victim rule, most notably by calling its
5035 * ->rule_dealloc() function.
5037 * If 'error' is nonzero, then generally the operation will be rolled back:
5039 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
5040 * restores the original rule. The caller must have uninitialized any
5041 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
5042 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
5043 * and 7 for the new rule, calling its ->rule_dealloc() function.
5045 * - If 'op' is a "modify flow" operation, ofproto restores the original
5048 * - 'op' must not be a "delete flow" operation. Removing a rule is not
5049 * allowed to fail. It must always succeed.
5051 * Please see the large comment in ofproto/ofproto-provider.h titled
5052 * "Asynchronous Operation Support" for more information. */
5054 ofoperation_complete(struct ofoperation
*op
, enum ofperr error
)
5056 struct ofopgroup
*group
= op
->group
;
5058 ovs_assert(op
->rule
->pending
== op
);
5059 ovs_assert(group
->n_running
> 0);
5060 ovs_assert(!error
|| op
->type
!= OFOPERATION_DELETE
);
5063 if (!--group
->n_running
&& !list_is_empty(&group
->ofproto_node
)) {
5064 ofopgroup_complete(group
);
5069 ofoperation_get_victim(struct ofoperation
*op
)
5071 ovs_assert(op
->type
== OFOPERATION_ADD
);
5076 pick_datapath_id(const struct ofproto
*ofproto
)
5078 const struct ofport
*port
;
5080 port
= ofproto_get_port(ofproto
, OFPP_LOCAL
);
5082 uint8_t ea
[ETH_ADDR_LEN
];
5085 error
= netdev_get_etheraddr(port
->netdev
, ea
);
5087 return eth_addr_to_uint64(ea
);
5089 VLOG_WARN("%s: could not get MAC address for %s (%s)",
5090 ofproto
->name
, netdev_get_name(port
->netdev
),
5091 ovs_strerror(error
));
5093 return ofproto
->fallback_dpid
;
5097 pick_fallback_dpid(void)
5099 uint8_t ea
[ETH_ADDR_LEN
];
5100 eth_addr_nicira_random(ea
);
5101 return eth_addr_to_uint64(ea
);
5104 /* Table overflow policy. */
5106 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
5107 * to NULL if the table is not configured to evict rules or if the table
5108 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
5109 * or with no timeouts are not evictable.) */
5111 choose_rule_to_evict(struct oftable
*table
, struct rule
**rulep
)
5113 struct eviction_group
*evg
;
5116 if (!table
->eviction_fields
) {
5120 /* In the common case, the outer and inner loops here will each be entered
5123 * - The inner loop normally "return"s in its first iteration. If the
5124 * eviction group has any evictable rules, then it always returns in
5127 * - The outer loop only iterates more than once if the largest eviction
5128 * group has no evictable rules.
5130 * - The outer loop can exit only if table's 'max_flows' is all filled up
5131 * by unevictable rules. */
5132 HEAP_FOR_EACH (evg
, size_node
, &table
->eviction_groups_by_size
) {
5135 HEAP_FOR_EACH (rule
, evg_node
, &evg
->rules
) {
5136 if (!ovs_rwlock_trywrlock(&rule
->evict
)) {
5146 /* Searches 'ofproto' for tables that have more flows than their configured
5147 * maximum and that have flow eviction enabled, and evicts as many flows as
5148 * necessary and currently feasible from them.
5150 * This triggers only when an OpenFlow table has N flows in it and then the
5151 * client configures a maximum number of flows less than N. */
5153 ofproto_evict(struct ofproto
*ofproto
)
5155 struct ofopgroup
*group
;
5156 struct oftable
*table
;
5158 group
= ofopgroup_create_unattached(ofproto
);
5159 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
5160 while (table
->eviction_fields
) {
5164 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
5165 n_rules
= classifier_count(&table
->cls
);
5166 ovs_rwlock_unlock(&table
->cls
.rwlock
);
5168 if (n_rules
<= table
->max_flows
) {
5172 if (!choose_rule_to_evict(table
, &rule
)) {
5176 if (rule
->pending
) {
5177 ovs_rwlock_unlock(&rule
->evict
);
5181 ofoperation_create(group
, rule
,
5182 OFOPERATION_DELETE
, OFPRR_EVICTION
);
5183 oftable_remove_rule(rule
);
5184 ofproto
->ofproto_class
->rule_destruct(rule
);
5187 ofopgroup_submit(group
);
5190 /* Eviction groups. */
5192 /* Returns the priority to use for an eviction_group that contains 'n_rules'
5193 * rules. The priority contains low-order random bits to ensure that eviction
5194 * groups with the same number of rules are prioritized randomly. */
5196 eviction_group_priority(size_t n_rules
)
5198 uint16_t size
= MIN(UINT16_MAX
, n_rules
);
5199 return (size
<< 16) | random_uint16();
5202 /* Updates 'evg', an eviction_group within 'table', following a change that
5203 * adds or removes rules in 'evg'. */
5205 eviction_group_resized(struct oftable
*table
, struct eviction_group
*evg
)
5207 heap_change(&table
->eviction_groups_by_size
, &evg
->size_node
,
5208 eviction_group_priority(heap_count(&evg
->rules
)));
5211 /* Destroys 'evg', an eviction_group within 'table':
5213 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
5214 * rules themselves, just removes them from the eviction group.)
5216 * - Removes 'evg' from 'table'.
5220 eviction_group_destroy(struct oftable
*table
, struct eviction_group
*evg
)
5222 while (!heap_is_empty(&evg
->rules
)) {
5225 rule
= CONTAINER_OF(heap_pop(&evg
->rules
), struct rule
, evg_node
);
5226 rule
->eviction_group
= NULL
;
5228 hmap_remove(&table
->eviction_groups_by_id
, &evg
->id_node
);
5229 heap_remove(&table
->eviction_groups_by_size
, &evg
->size_node
);
5230 heap_destroy(&evg
->rules
);
5234 /* Removes 'rule' from its eviction group, if any. */
5236 eviction_group_remove_rule(struct rule
*rule
)
5238 if (rule
->eviction_group
) {
5239 struct oftable
*table
= &rule
->ofproto
->tables
[rule
->table_id
];
5240 struct eviction_group
*evg
= rule
->eviction_group
;
5242 rule
->eviction_group
= NULL
;
5243 heap_remove(&evg
->rules
, &rule
->evg_node
);
5244 if (heap_is_empty(&evg
->rules
)) {
5245 eviction_group_destroy(table
, evg
);
5247 eviction_group_resized(table
, evg
);
5252 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
5253 * returns the hash value. */
5255 eviction_group_hash_rule(struct rule
*rule
)
5257 struct oftable
*table
= &rule
->ofproto
->tables
[rule
->table_id
];
5258 const struct mf_subfield
*sf
;
5262 hash
= table
->eviction_group_id_basis
;
5263 miniflow_expand(&rule
->cr
.match
.flow
, &flow
);
5264 for (sf
= table
->eviction_fields
;
5265 sf
< &table
->eviction_fields
[table
->n_eviction_fields
];
5268 if (mf_are_prereqs_ok(sf
->field
, &flow
)) {
5269 union mf_value value
;
5271 mf_get_value(sf
->field
, &flow
, &value
);
5273 bitwise_zero(&value
, sf
->field
->n_bytes
, 0, sf
->ofs
);
5275 if (sf
->ofs
+ sf
->n_bits
< sf
->field
->n_bytes
* 8) {
5276 unsigned int start
= sf
->ofs
+ sf
->n_bits
;
5277 bitwise_zero(&value
, sf
->field
->n_bytes
, start
,
5278 sf
->field
->n_bytes
* 8 - start
);
5280 hash
= hash_bytes(&value
, sf
->field
->n_bytes
, hash
);
5282 hash
= hash_int(hash
, 0);
5289 /* Returns an eviction group within 'table' with the given 'id', creating one
5291 static struct eviction_group
*
5292 eviction_group_find(struct oftable
*table
, uint32_t id
)
5294 struct eviction_group
*evg
;
5296 HMAP_FOR_EACH_WITH_HASH (evg
, id_node
, id
, &table
->eviction_groups_by_id
) {
5300 evg
= xmalloc(sizeof *evg
);
5301 hmap_insert(&table
->eviction_groups_by_id
, &evg
->id_node
, id
);
5302 heap_insert(&table
->eviction_groups_by_size
, &evg
->size_node
,
5303 eviction_group_priority(0));
5304 heap_init(&evg
->rules
);
5309 /* Returns an eviction priority for 'rule'. The return value should be
5310 * interpreted so that higher priorities make a rule more attractive candidates
5313 rule_eviction_priority(struct rule
*rule
)
5315 long long int hard_expiration
;
5316 long long int idle_expiration
;
5317 long long int expiration
;
5318 uint32_t expiration_offset
;
5320 /* Calculate time of expiration. */
5321 ovs_mutex_lock(&rule
->timeout_mutex
);
5322 hard_expiration
= (rule
->hard_timeout
5323 ? rule
->modified
+ rule
->hard_timeout
* 1000
5325 idle_expiration
= (rule
->idle_timeout
5326 ? rule
->used
+ rule
->idle_timeout
* 1000
5328 expiration
= MIN(hard_expiration
, idle_expiration
);
5329 ovs_mutex_unlock(&rule
->timeout_mutex
);
5330 if (expiration
== LLONG_MAX
) {
5334 /* Calculate the time of expiration as a number of (approximate) seconds
5335 * after program startup.
5337 * This should work OK for program runs that last UINT32_MAX seconds or
5338 * less. Therefore, please restart OVS at least once every 136 years. */
5339 expiration_offset
= (expiration
>> 10) - (time_boot_msec() >> 10);
5341 /* Invert the expiration offset because we're using a max-heap. */
5342 return UINT32_MAX
- expiration_offset
;
5345 /* Adds 'rule' to an appropriate eviction group for its oftable's
5346 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
5347 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
5350 * The caller must ensure that 'rule' is not already in an eviction group. */
5352 eviction_group_add_rule(struct rule
*rule
)
5354 struct ofproto
*ofproto
= rule
->ofproto
;
5355 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
5358 ovs_mutex_lock(&rule
->timeout_mutex
);
5359 has_timeout
= rule
->hard_timeout
|| rule
->idle_timeout
;
5360 ovs_mutex_unlock(&rule
->timeout_mutex
);
5362 if (table
->eviction_fields
&& has_timeout
) {
5363 struct eviction_group
*evg
;
5365 evg
= eviction_group_find(table
, eviction_group_hash_rule(rule
));
5367 rule
->eviction_group
= evg
;
5368 heap_insert(&evg
->rules
, &rule
->evg_node
,
5369 rule_eviction_priority(rule
));
5370 eviction_group_resized(table
, evg
);
5376 /* Initializes 'table'. */
5378 oftable_init(struct oftable
*table
)
5380 memset(table
, 0, sizeof *table
);
5381 classifier_init(&table
->cls
);
5382 table
->max_flows
= UINT_MAX
;
5385 /* Destroys 'table', including its classifier and eviction groups.
5387 * The caller is responsible for freeing 'table' itself. */
5389 oftable_destroy(struct oftable
*table
)
5391 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
5392 ovs_assert(classifier_is_empty(&table
->cls
));
5393 ovs_rwlock_unlock(&table
->cls
.rwlock
);
5394 oftable_disable_eviction(table
);
5395 classifier_destroy(&table
->cls
);
5399 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
5400 * string, then 'table' will use its default name.
5402 * This only affects the name exposed for a table exposed through the OpenFlow
5403 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
5405 oftable_set_name(struct oftable
*table
, const char *name
)
5407 if (name
&& name
[0]) {
5408 int len
= strnlen(name
, OFP_MAX_TABLE_NAME_LEN
);
5409 if (!table
->name
|| strncmp(name
, table
->name
, len
)) {
5411 table
->name
= xmemdup0(name
, len
);
5419 /* oftables support a choice of two policies when adding a rule would cause the
5420 * number of flows in the table to exceed the configured maximum number: either
5421 * they can refuse to add the new flow or they can evict some existing flow.
5422 * This function configures the former policy on 'table'. */
5424 oftable_disable_eviction(struct oftable
*table
)
5426 if (table
->eviction_fields
) {
5427 struct eviction_group
*evg
, *next
;
5429 HMAP_FOR_EACH_SAFE (evg
, next
, id_node
,
5430 &table
->eviction_groups_by_id
) {
5431 eviction_group_destroy(table
, evg
);
5433 hmap_destroy(&table
->eviction_groups_by_id
);
5434 heap_destroy(&table
->eviction_groups_by_size
);
5436 free(table
->eviction_fields
);
5437 table
->eviction_fields
= NULL
;
5438 table
->n_eviction_fields
= 0;
5442 /* oftables support a choice of two policies when adding a rule would cause the
5443 * number of flows in the table to exceed the configured maximum number: either
5444 * they can refuse to add the new flow or they can evict some existing flow.
5445 * This function configures the latter policy on 'table', with fairness based
5446 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
5447 * 'n_fields' as 0 disables fairness.) */
5449 oftable_enable_eviction(struct oftable
*table
,
5450 const struct mf_subfield
*fields
, size_t n_fields
)
5452 struct cls_cursor cursor
;
5455 if (table
->eviction_fields
5456 && n_fields
== table
->n_eviction_fields
5458 || !memcmp(fields
, table
->eviction_fields
,
5459 n_fields
* sizeof *fields
))) {
5464 oftable_disable_eviction(table
);
5466 table
->n_eviction_fields
= n_fields
;
5467 table
->eviction_fields
= xmemdup(fields
, n_fields
* sizeof *fields
);
5469 table
->eviction_group_id_basis
= random_uint32();
5470 hmap_init(&table
->eviction_groups_by_id
);
5471 heap_init(&table
->eviction_groups_by_size
);
5473 ovs_rwlock_rdlock(&table
->cls
.rwlock
);
5474 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
5475 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
5476 eviction_group_add_rule(rule
);
5478 ovs_rwlock_unlock(&table
->cls
.rwlock
);
5481 /* Removes 'rule' from the oftable that contains it. */
5483 oftable_remove_rule__(struct ofproto
*ofproto
, struct classifier
*cls
,
5485 OVS_REQ_WRLOCK(cls
->rwlock
) OVS_RELEASES(rule
->evict
)
5487 classifier_remove(cls
, &rule
->cr
);
5488 if (rule
->meter_id
) {
5489 list_remove(&rule
->meter_list_node
);
5491 cookies_remove(ofproto
, rule
);
5492 eviction_group_remove_rule(rule
);
5493 ovs_mutex_lock(&ofproto
->expirable_mutex
);
5494 if (!list_is_empty(&rule
->expirable
)) {
5495 list_remove(&rule
->expirable
);
5497 ovs_mutex_unlock(&ofproto
->expirable_mutex
);
5498 if (!list_is_empty(&rule
->meter_list_node
)) {
5499 list_remove(&rule
->meter_list_node
);
5501 ovs_rwlock_unlock(&rule
->evict
);
5505 oftable_remove_rule(struct rule
*rule
)
5507 struct ofproto
*ofproto
= rule
->ofproto
;
5508 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
5510 ovs_rwlock_wrlock(&table
->cls
.rwlock
);
5511 oftable_remove_rule__(ofproto
, &table
->cls
, rule
);
5512 ovs_rwlock_unlock(&table
->cls
.rwlock
);
5515 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
5516 * oftable that has an identical cls_rule. Returns the rule that was removed,
5517 * if any, and otherwise NULL. */
5518 static struct rule
*
5519 oftable_replace_rule(struct rule
*rule
)
5521 struct ofproto
*ofproto
= rule
->ofproto
;
5522 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
5523 struct rule
*victim
;
5526 ovs_mutex_lock(&rule
->timeout_mutex
);
5527 may_expire
= rule
->hard_timeout
|| rule
->idle_timeout
;
5528 ovs_mutex_unlock(&rule
->timeout_mutex
);
5531 ovs_mutex_lock(&ofproto
->expirable_mutex
);
5532 list_insert(&ofproto
->expirable
, &rule
->expirable
);
5533 ovs_mutex_unlock(&ofproto
->expirable_mutex
);
5535 cookies_insert(ofproto
, rule
);
5536 if (rule
->meter_id
) {
5537 struct meter
*meter
= ofproto
->meters
[rule
->meter_id
];
5538 list_insert(&meter
->rules
, &rule
->meter_list_node
);
5540 ovs_rwlock_wrlock(&table
->cls
.rwlock
);
5541 victim
= rule_from_cls_rule(classifier_replace(&table
->cls
, &rule
->cr
));
5542 ovs_rwlock_unlock(&table
->cls
.rwlock
);
5544 if (victim
->meter_id
) {
5545 list_remove(&victim
->meter_list_node
);
5547 cookies_remove(ofproto
, victim
);
5549 ovs_mutex_lock(&ofproto
->expirable_mutex
);
5550 if (!list_is_empty(&victim
->expirable
)) {
5551 list_remove(&victim
->expirable
);
5553 ovs_mutex_unlock(&ofproto
->expirable_mutex
);
5554 eviction_group_remove_rule(victim
);
5556 eviction_group_add_rule(rule
);
5560 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
5562 oftable_substitute_rule(struct rule
*old
, struct rule
*new)
5565 oftable_replace_rule(new);
5567 ovs_rwlock_wrlock(&old
->evict
);
5568 oftable_remove_rule(old
);
5572 /* unixctl commands. */
5575 ofproto_lookup(const char *name
)
5577 struct ofproto
*ofproto
;
5579 HMAP_FOR_EACH_WITH_HASH (ofproto
, hmap_node
, hash_string(name
, 0),
5581 if (!strcmp(ofproto
->name
, name
)) {
5589 ofproto_unixctl_list(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
5590 const char *argv
[] OVS_UNUSED
, void *aux OVS_UNUSED
)
5592 struct ofproto
*ofproto
;
5596 HMAP_FOR_EACH (ofproto
, hmap_node
, &all_ofprotos
) {
5597 ds_put_format(&results
, "%s\n", ofproto
->name
);
5599 unixctl_command_reply(conn
, ds_cstr(&results
));
5600 ds_destroy(&results
);
5604 ofproto_unixctl_init(void)
5606 static bool registered
;
5612 unixctl_command_register("ofproto/list", "", 0, 0,
5613 ofproto_unixctl_list
, NULL
);
5616 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5618 * This is deprecated. It is only for compatibility with broken device drivers
5619 * in old versions of Linux that do not properly support VLANs when VLAN
5620 * devices are not used. When broken device drivers are no longer in
5621 * widespread use, we will delete these interfaces. */
5623 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
5624 * (exactly) by an OpenFlow rule in 'ofproto'. */
5626 ofproto_get_vlan_usage(struct ofproto
*ofproto
, unsigned long int *vlan_bitmap
)
5628 const struct oftable
*oftable
;
5630 free(ofproto
->vlan_bitmap
);
5631 ofproto
->vlan_bitmap
= bitmap_allocate(4096);
5632 ofproto
->vlans_changed
= false;
5634 OFPROTO_FOR_EACH_TABLE (oftable
, ofproto
) {
5635 const struct cls_table
*table
;
5637 HMAP_FOR_EACH (table
, hmap_node
, &oftable
->cls
.tables
) {
5638 if (minimask_get_vid_mask(&table
->mask
) == VLAN_VID_MASK
) {
5639 const struct cls_rule
*rule
;
5641 HMAP_FOR_EACH (rule
, hmap_node
, &table
->rules
) {
5642 uint16_t vid
= miniflow_get_vid(&rule
->match
.flow
);
5643 bitmap_set1(vlan_bitmap
, vid
);
5644 bitmap_set1(ofproto
->vlan_bitmap
, vid
);
5651 /* Returns true if new VLANs have come into use by the flow table since the
5652 * last call to ofproto_get_vlan_usage().
5654 * We don't track when old VLANs stop being used. */
5656 ofproto_has_vlan_usage_changed(const struct ofproto
*ofproto
)
5658 return ofproto
->vlans_changed
;
5661 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
5662 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
5663 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
5664 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
5665 * then the VLAN device is un-enslaved. */
5667 ofproto_port_set_realdev(struct ofproto
*ofproto
, ofp_port_t vlandev_ofp_port
,
5668 ofp_port_t realdev_ofp_port
, int vid
)
5670 struct ofport
*ofport
;
5673 ovs_assert(vlandev_ofp_port
!= realdev_ofp_port
);
5675 ofport
= ofproto_get_port(ofproto
, vlandev_ofp_port
);
5677 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16
,
5678 ofproto
->name
, vlandev_ofp_port
);
5682 if (!ofproto
->ofproto_class
->set_realdev
) {
5683 if (!vlandev_ofp_port
) {
5686 VLOG_WARN("%s: vlan splinters not supported", ofproto
->name
);
5690 error
= ofproto
->ofproto_class
->set_realdev(ofport
, realdev_ofp_port
, vid
);
5692 VLOG_WARN("%s: setting realdev on port %"PRIu16
" (%s) failed (%s)",
5693 ofproto
->name
, vlandev_ofp_port
,
5694 netdev_get_name(ofport
->netdev
), ovs_strerror(error
));