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.
25 #include "byte-order.h"
26 #include "classifier.h"
29 #include "dynamic-string.h"
32 #include "meta-flow.h"
35 #include "ofp-actions.h"
36 #include "ofp-errors.h"
38 #include "ofp-print.h"
41 #include "ofproto-provider.h"
42 #include "openflow/nicira-ext.h"
43 #include "openflow/openflow.h"
47 #include "poll-loop.h"
53 #include "unaligned.h"
57 VLOG_DEFINE_THIS_MODULE(ofproto
);
59 COVERAGE_DEFINE(ofproto_error
);
60 COVERAGE_DEFINE(ofproto_flush
);
61 COVERAGE_DEFINE(ofproto_no_packet_in
);
62 COVERAGE_DEFINE(ofproto_packet_out
);
63 COVERAGE_DEFINE(ofproto_queue_req
);
64 COVERAGE_DEFINE(ofproto_recv_openflow
);
65 COVERAGE_DEFINE(ofproto_reinit_ports
);
66 COVERAGE_DEFINE(ofproto_uninstallable
);
67 COVERAGE_DEFINE(ofproto_update_port
);
70 S_OPENFLOW
, /* Processing OpenFlow commands. */
71 S_EVICT
, /* Evicting flows from over-limit tables. */
72 S_FLUSH
, /* Deleting all flow table rules. */
75 enum ofoperation_type
{
81 /* A single OpenFlow request can execute any number of operations. The
82 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
83 * ofconn to which an error reply should be sent if necessary.
85 * ofproto initiates some operations internally. These operations are still
86 * assigned to groups but will not have an associated ofconn. */
88 struct ofproto
*ofproto
; /* Owning ofproto. */
89 struct list ofproto_node
; /* In ofproto's "pending" list. */
90 struct list ops
; /* List of "struct ofoperation"s. */
91 int n_running
; /* Number of ops still pending. */
93 /* Data needed to send OpenFlow reply on failure or to send a buffered
96 * If list_is_empty(ofconn_node) then this ofopgroup never had an
97 * associated ofconn or its ofconn's connection dropped after it initiated
98 * the operation. In the latter case 'ofconn' is a wild pointer that
99 * refers to freed memory, so the 'ofconn' member must be used only if
100 * !list_is_empty(ofconn_node).
102 struct list ofconn_node
; /* In ofconn's list of pending opgroups. */
103 struct ofconn
*ofconn
; /* ofconn for reply (but see note above). */
104 struct ofp_header
*request
; /* Original request (truncated at 64 bytes). */
105 uint32_t buffer_id
; /* Buffer id from original request. */
108 static struct ofopgroup
*ofopgroup_create_unattached(struct ofproto
*);
109 static struct ofopgroup
*ofopgroup_create(struct ofproto
*, struct ofconn
*,
110 const struct ofp_header
*,
112 static void ofopgroup_submit(struct ofopgroup
*);
113 static void ofopgroup_complete(struct ofopgroup
*);
115 /* A single flow table operation. */
117 struct ofopgroup
*group
; /* Owning group. */
118 struct list group_node
; /* In ofopgroup's "ops" list. */
119 struct hmap_node hmap_node
; /* In ofproto's "deletions" hmap. */
120 struct rule
*rule
; /* Rule being operated upon. */
121 enum ofoperation_type type
; /* Type of operation. */
123 /* OFOPERATION_ADD. */
124 struct rule
*victim
; /* Rule being replaced, if any.. */
126 /* OFOPERATION_MODIFY: The old actions, if the actions are changing. */
127 struct ofpact
*ofpacts
;
130 /* OFOPERATION_DELETE. */
131 enum ofp_flow_removed_reason reason
; /* Reason flow was removed. */
133 ovs_be64 flow_cookie
; /* Rule's old flow cookie. */
134 enum ofperr error
; /* 0 if no error. */
137 static struct ofoperation
*ofoperation_create(struct ofopgroup
*,
139 enum ofoperation_type
,
140 enum ofp_flow_removed_reason
);
141 static void ofoperation_destroy(struct ofoperation
*);
144 static void oftable_init(struct oftable
*);
145 static void oftable_destroy(struct oftable
*);
147 static void oftable_set_name(struct oftable
*, const char *name
);
149 static void oftable_disable_eviction(struct oftable
*);
150 static void oftable_enable_eviction(struct oftable
*,
151 const struct mf_subfield
*fields
,
154 static void oftable_remove_rule(struct rule
*);
155 static struct rule
*oftable_replace_rule(struct rule
*);
156 static void oftable_substitute_rule(struct rule
*old
, struct rule
*new);
158 /* A set of rules within a single OpenFlow table (oftable) that have the same
159 * values for the oftable's eviction_fields. A rule to be evicted, when one is
160 * needed, is taken from the eviction group that contains the greatest number
163 * An oftable owns any number of eviction groups, each of which contains any
166 * Membership in an eviction group is imprecise, based on the hash of the
167 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
168 * That is, if two rules have different eviction_fields, but those
169 * eviction_fields hash to the same value, then they will belong to the same
170 * eviction_group anyway.
172 * (When eviction is not enabled on an oftable, we don't track any eviction
173 * groups, to save time and space.) */
174 struct eviction_group
{
175 struct hmap_node id_node
; /* In oftable's "eviction_groups_by_id". */
176 struct heap_node size_node
; /* In oftable's "eviction_groups_by_size". */
177 struct heap rules
; /* Contains "struct rule"s. */
180 static struct rule
*choose_rule_to_evict(struct oftable
*);
181 static void ofproto_evict(struct ofproto
*);
182 static uint32_t rule_eviction_priority(struct rule
*);
185 static void ofport_destroy__(struct ofport
*);
186 static void ofport_destroy(struct ofport
*);
188 static void update_port(struct ofproto
*, const char *devname
);
189 static int init_ports(struct ofproto
*);
190 static void reinit_ports(struct ofproto
*);
193 static void ofproto_rule_destroy__(struct rule
*);
194 static void ofproto_rule_send_removed(struct rule
*, uint8_t reason
);
195 static bool rule_is_modifiable(const struct rule
*);
198 static enum ofperr
add_flow(struct ofproto
*, struct ofconn
*,
199 const struct ofputil_flow_mod
*,
200 const struct ofp_header
*);
201 static void delete_flow__(struct rule
*, struct ofopgroup
*);
202 static bool handle_openflow(struct ofconn
*, struct ofpbuf
*);
203 static enum ofperr
handle_flow_mod__(struct ofproto
*, struct ofconn
*,
204 const struct ofputil_flow_mod
*,
205 const struct ofp_header
*);
208 static uint64_t pick_datapath_id(const struct ofproto
*);
209 static uint64_t pick_fallback_dpid(void);
210 static void ofproto_destroy__(struct ofproto
*);
211 static void update_mtu(struct ofproto
*, struct ofport
*);
214 static void ofproto_unixctl_init(void);
216 /* All registered ofproto classes, in probe order. */
217 static const struct ofproto_class
**ofproto_classes
;
218 static size_t n_ofproto_classes
;
219 static size_t allocated_ofproto_classes
;
221 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
222 static struct hmap all_ofprotos
= HMAP_INITIALIZER(&all_ofprotos
);
224 /* Initial mappings of port to OpenFlow number mappings. */
225 static struct shash init_ofp_ports
= SHASH_INITIALIZER(&init_ofp_ports
);
227 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
229 /* Must be called to initialize the ofproto library.
231 * The caller may pass in 'iface_hints', which contains an shash of
232 * "iface_hint" elements indexed by the interface's name. The provider
233 * may use these hints to describe the startup configuration in order to
234 * reinitialize its state. The caller owns the provided data, so a
235 * provider will make copies of anything required. An ofproto provider
236 * will remove any existing state that is not described by the hint, and
237 * may choose to remove it all. */
239 ofproto_init(const struct shash
*iface_hints
)
241 struct shash_node
*node
;
244 ofproto_class_register(&ofproto_dpif_class
);
246 /* Make a local copy, since we don't own 'iface_hints' elements. */
247 SHASH_FOR_EACH(node
, iface_hints
) {
248 const struct iface_hint
*orig_hint
= node
->data
;
249 struct iface_hint
*new_hint
= xmalloc(sizeof *new_hint
);
250 const char *br_type
= ofproto_normalize_type(orig_hint
->br_type
);
252 new_hint
->br_name
= xstrdup(orig_hint
->br_name
);
253 new_hint
->br_type
= xstrdup(br_type
);
254 new_hint
->ofp_port
= orig_hint
->ofp_port
;
256 shash_add(&init_ofp_ports
, node
->name
, new_hint
);
259 for (i
= 0; i
< n_ofproto_classes
; i
++) {
260 ofproto_classes
[i
]->init(&init_ofp_ports
);
264 /* 'type' should be a normalized datapath type, as returned by
265 * ofproto_normalize_type(). Returns the corresponding ofproto_class
266 * structure, or a null pointer if there is none registered for 'type'. */
267 static const struct ofproto_class
*
268 ofproto_class_find__(const char *type
)
272 for (i
= 0; i
< n_ofproto_classes
; i
++) {
273 const struct ofproto_class
*class = ofproto_classes
[i
];
278 class->enumerate_types(&types
);
279 found
= sset_contains(&types
, type
);
280 sset_destroy(&types
);
286 VLOG_WARN("unknown datapath type %s", type
);
290 /* Registers a new ofproto class. After successful registration, new ofprotos
291 * of that type can be created using ofproto_create(). */
293 ofproto_class_register(const struct ofproto_class
*new_class
)
297 for (i
= 0; i
< n_ofproto_classes
; i
++) {
298 if (ofproto_classes
[i
] == new_class
) {
303 if (n_ofproto_classes
>= allocated_ofproto_classes
) {
304 ofproto_classes
= x2nrealloc(ofproto_classes
,
305 &allocated_ofproto_classes
,
306 sizeof *ofproto_classes
);
308 ofproto_classes
[n_ofproto_classes
++] = new_class
;
312 /* Unregisters a datapath provider. 'type' must have been previously
313 * registered and not currently be in use by any ofprotos. After
314 * unregistration new datapaths of that type cannot be opened using
315 * ofproto_create(). */
317 ofproto_class_unregister(const struct ofproto_class
*class)
321 for (i
= 0; i
< n_ofproto_classes
; i
++) {
322 if (ofproto_classes
[i
] == class) {
323 for (i
++; i
< n_ofproto_classes
; i
++) {
324 ofproto_classes
[i
- 1] = ofproto_classes
[i
];
330 VLOG_WARN("attempted to unregister an ofproto class that is not "
335 /* Clears 'types' and enumerates all registered ofproto types into it. The
336 * caller must first initialize the sset. */
338 ofproto_enumerate_types(struct sset
*types
)
342 for (i
= 0; i
< n_ofproto_classes
; i
++) {
343 ofproto_classes
[i
]->enumerate_types(types
);
347 /* Returns the fully spelled out name for the given ofproto 'type'.
349 * Normalized type string can be compared with strcmp(). Unnormalized type
350 * string might be the same even if they have different spellings. */
352 ofproto_normalize_type(const char *type
)
354 return type
&& type
[0] ? type
: "system";
357 /* Clears 'names' and enumerates the names of all known created ofprotos with
358 * the given 'type'. The caller must first initialize the sset. Returns 0 if
359 * successful, otherwise a positive errno value.
361 * Some kinds of datapaths might not be practically enumerable. This is not
362 * considered an error. */
364 ofproto_enumerate_names(const char *type
, struct sset
*names
)
366 const struct ofproto_class
*class = ofproto_class_find__(type
);
367 return class ? class->enumerate_names(type
, names
) : EAFNOSUPPORT
;
371 ofproto_create(const char *datapath_name
, const char *datapath_type
,
372 struct ofproto
**ofprotop
)
374 const struct ofproto_class
*class;
375 struct ofproto
*ofproto
;
381 ofproto_unixctl_init();
383 datapath_type
= ofproto_normalize_type(datapath_type
);
384 class = ofproto_class_find__(datapath_type
);
386 VLOG_WARN("could not create datapath %s of unknown type %s",
387 datapath_name
, datapath_type
);
391 ofproto
= class->alloc();
393 VLOG_ERR("failed to allocate datapath %s of type %s",
394 datapath_name
, datapath_type
);
399 memset(ofproto
, 0, sizeof *ofproto
);
400 ofproto
->ofproto_class
= class;
401 ofproto
->name
= xstrdup(datapath_name
);
402 ofproto
->type
= xstrdup(datapath_type
);
403 hmap_insert(&all_ofprotos
, &ofproto
->hmap_node
,
404 hash_string(ofproto
->name
, 0));
405 ofproto
->datapath_id
= 0;
406 ofproto_set_flow_eviction_threshold(ofproto
,
407 OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT
);
408 ofproto
->forward_bpdu
= false;
409 ofproto
->fallback_dpid
= pick_fallback_dpid();
410 ofproto
->mfr_desc
= NULL
;
411 ofproto
->hw_desc
= NULL
;
412 ofproto
->sw_desc
= NULL
;
413 ofproto
->serial_desc
= NULL
;
414 ofproto
->dp_desc
= NULL
;
415 ofproto
->frag_handling
= OFPC_FRAG_NORMAL
;
416 hmap_init(&ofproto
->ports
);
417 shash_init(&ofproto
->port_by_name
);
418 simap_init(&ofproto
->ofp_requests
);
419 ofproto
->max_ports
= OFPP_MAX
;
420 ofproto
->tables
= NULL
;
421 ofproto
->n_tables
= 0;
422 list_init(&ofproto
->expirable
);
423 ofproto
->connmgr
= connmgr_create(ofproto
, datapath_name
, datapath_name
);
424 ofproto
->state
= S_OPENFLOW
;
425 list_init(&ofproto
->pending
);
426 ofproto
->n_pending
= 0;
427 hmap_init(&ofproto
->deletions
);
428 ofproto
->n_add
= ofproto
->n_delete
= ofproto
->n_modify
= 0;
429 ofproto
->first_op
= ofproto
->last_op
= LLONG_MIN
;
430 ofproto
->next_op_report
= LLONG_MAX
;
431 ofproto
->op_backoff
= LLONG_MIN
;
432 ofproto
->vlan_bitmap
= NULL
;
433 ofproto
->vlans_changed
= false;
434 ofproto
->min_mtu
= INT_MAX
;
436 error
= ofproto
->ofproto_class
->construct(ofproto
);
438 VLOG_ERR("failed to open datapath %s: %s",
439 datapath_name
, strerror(error
));
440 ofproto_destroy__(ofproto
);
444 /* The "max_ports" member should have been set by ->construct(ofproto).
445 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
446 ofproto
->ofp_port_ids
= bitmap_allocate(ofproto
->max_ports
);
447 bitmap_set1(ofproto
->ofp_port_ids
, 0);
449 /* Check that hidden tables, if any, are at the end. */
450 ovs_assert(ofproto
->n_tables
);
451 for (i
= 0; i
+ 1 < ofproto
->n_tables
; i
++) {
452 enum oftable_flags flags
= ofproto
->tables
[i
].flags
;
453 enum oftable_flags next_flags
= ofproto
->tables
[i
+ 1].flags
;
455 ovs_assert(!(flags
& OFTABLE_HIDDEN
) || next_flags
& OFTABLE_HIDDEN
);
458 ofproto
->datapath_id
= pick_datapath_id(ofproto
);
465 /* Must be called (only) by an ofproto implementation in its constructor
466 * function. See the large comment on 'construct' in struct ofproto_class for
469 ofproto_init_tables(struct ofproto
*ofproto
, int n_tables
)
471 struct oftable
*table
;
473 ovs_assert(!ofproto
->n_tables
);
474 ovs_assert(n_tables
>= 1 && n_tables
<= 255);
476 ofproto
->n_tables
= n_tables
;
477 ofproto
->tables
= xmalloc(n_tables
* sizeof *ofproto
->tables
);
478 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
483 /* To be optionally called (only) by an ofproto implementation in its
484 * constructor function. See the large comment on 'construct' in struct
485 * ofproto_class for details.
487 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
488 * will then ensure that actions passed into the ofproto implementation will
489 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
490 * function is not called, there will be no such restriction.
492 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
493 * the 'max_ports' restriction. */
495 ofproto_init_max_ports(struct ofproto
*ofproto
, uint16_t max_ports
)
497 ovs_assert(max_ports
<= OFPP_MAX
);
498 ofproto
->max_ports
= max_ports
;
502 ofproto_get_datapath_id(const struct ofproto
*ofproto
)
504 return ofproto
->datapath_id
;
508 ofproto_set_datapath_id(struct ofproto
*p
, uint64_t datapath_id
)
510 uint64_t old_dpid
= p
->datapath_id
;
511 p
->datapath_id
= datapath_id
? datapath_id
: pick_datapath_id(p
);
512 if (p
->datapath_id
!= old_dpid
) {
513 /* Force all active connections to reconnect, since there is no way to
514 * notify a controller that the datapath ID has changed. */
515 ofproto_reconnect_controllers(p
);
520 ofproto_set_controllers(struct ofproto
*p
,
521 const struct ofproto_controller
*controllers
,
522 size_t n_controllers
, uint32_t allowed_versions
)
524 connmgr_set_controllers(p
->connmgr
, controllers
, n_controllers
,
529 ofproto_set_fail_mode(struct ofproto
*p
, enum ofproto_fail_mode fail_mode
)
531 connmgr_set_fail_mode(p
->connmgr
, fail_mode
);
534 /* Drops the connections between 'ofproto' and all of its controllers, forcing
535 * them to reconnect. */
537 ofproto_reconnect_controllers(struct ofproto
*ofproto
)
539 connmgr_reconnect(ofproto
->connmgr
);
542 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
543 * in-band control should guarantee access, in the same way that in-band
544 * control guarantees access to OpenFlow controllers. */
546 ofproto_set_extra_in_band_remotes(struct ofproto
*ofproto
,
547 const struct sockaddr_in
*extras
, size_t n
)
549 connmgr_set_extra_in_band_remotes(ofproto
->connmgr
, extras
, n
);
552 /* Sets the OpenFlow queue used by flows set up by in-band control on
553 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
554 * flows will use the default queue. */
556 ofproto_set_in_band_queue(struct ofproto
*ofproto
, int queue_id
)
558 connmgr_set_in_band_queue(ofproto
->connmgr
, queue_id
);
561 /* Sets the number of flows at which eviction from the kernel flow table
564 ofproto_set_flow_eviction_threshold(struct ofproto
*ofproto
, unsigned threshold
)
566 if (threshold
< OFPROTO_FLOW_EVICTION_THRESHOLD_MIN
) {
567 ofproto
->flow_eviction_threshold
= OFPROTO_FLOW_EVICTION_THRESHOLD_MIN
;
569 ofproto
->flow_eviction_threshold
= threshold
;
573 /* If forward_bpdu is true, the NORMAL action will forward frames with
574 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
575 * the NORMAL action will drop these frames. */
577 ofproto_set_forward_bpdu(struct ofproto
*ofproto
, bool forward_bpdu
)
579 bool old_val
= ofproto
->forward_bpdu
;
580 ofproto
->forward_bpdu
= forward_bpdu
;
581 if (old_val
!= ofproto
->forward_bpdu
) {
582 if (ofproto
->ofproto_class
->forward_bpdu_changed
) {
583 ofproto
->ofproto_class
->forward_bpdu_changed(ofproto
);
588 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
589 * 'idle_time', in seconds, and the maximum number of MAC table entries to
592 ofproto_set_mac_table_config(struct ofproto
*ofproto
, unsigned idle_time
,
595 if (ofproto
->ofproto_class
->set_mac_table_config
) {
596 ofproto
->ofproto_class
->set_mac_table_config(ofproto
, idle_time
,
602 ofproto_set_dp_desc(struct ofproto
*p
, const char *dp_desc
)
605 p
->dp_desc
= dp_desc
? xstrdup(dp_desc
) : NULL
;
609 ofproto_set_snoops(struct ofproto
*ofproto
, const struct sset
*snoops
)
611 return connmgr_set_snoops(ofproto
->connmgr
, snoops
);
615 ofproto_set_netflow(struct ofproto
*ofproto
,
616 const struct netflow_options
*nf_options
)
618 if (nf_options
&& sset_is_empty(&nf_options
->collectors
)) {
622 if (ofproto
->ofproto_class
->set_netflow
) {
623 return ofproto
->ofproto_class
->set_netflow(ofproto
, nf_options
);
625 return nf_options
? EOPNOTSUPP
: 0;
630 ofproto_set_sflow(struct ofproto
*ofproto
,
631 const struct ofproto_sflow_options
*oso
)
633 if (oso
&& sset_is_empty(&oso
->targets
)) {
637 if (ofproto
->ofproto_class
->set_sflow
) {
638 return ofproto
->ofproto_class
->set_sflow(ofproto
, oso
);
640 return oso
? EOPNOTSUPP
: 0;
645 ofproto_set_ipfix(struct ofproto
*ofproto
,
646 const struct ofproto_ipfix_bridge_exporter_options
*bo
,
647 const struct ofproto_ipfix_flow_exporter_options
*fo
,
650 if (ofproto
->ofproto_class
->set_ipfix
) {
651 return ofproto
->ofproto_class
->set_ipfix(ofproto
, bo
, fo
, n_fo
);
653 return (bo
|| fo
) ? EOPNOTSUPP
: 0;
657 /* Spanning Tree Protocol (STP) configuration. */
659 /* Configures STP on 'ofproto' using the settings defined in 's'. If
660 * 's' is NULL, disables STP.
662 * Returns 0 if successful, otherwise a positive errno value. */
664 ofproto_set_stp(struct ofproto
*ofproto
,
665 const struct ofproto_stp_settings
*s
)
667 return (ofproto
->ofproto_class
->set_stp
668 ? ofproto
->ofproto_class
->set_stp(ofproto
, s
)
672 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
673 * 'enabled' member of 's' is false, then the other members are not
676 * Returns 0 if successful, otherwise a positive errno value. */
678 ofproto_get_stp_status(struct ofproto
*ofproto
,
679 struct ofproto_stp_status
*s
)
681 return (ofproto
->ofproto_class
->get_stp_status
682 ? ofproto
->ofproto_class
->get_stp_status(ofproto
, s
)
686 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
687 * in 's'. The caller is responsible for assigning STP port numbers
688 * (using the 'port_num' member in the range of 1 through 255, inclusive)
689 * and ensuring there are no duplicates. If the 's' is NULL, then STP
690 * is disabled on the port.
692 * Returns 0 if successful, otherwise a positive errno value.*/
694 ofproto_port_set_stp(struct ofproto
*ofproto
, uint16_t ofp_port
,
695 const struct ofproto_port_stp_settings
*s
)
697 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
699 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16
,
700 ofproto
->name
, ofp_port
);
704 return (ofproto
->ofproto_class
->set_stp_port
705 ? ofproto
->ofproto_class
->set_stp_port(ofport
, s
)
709 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
710 * 's'. If the 'enabled' member in 's' is false, then the other members
711 * are not meaningful.
713 * Returns 0 if successful, otherwise a positive errno value.*/
715 ofproto_port_get_stp_status(struct ofproto
*ofproto
, uint16_t ofp_port
,
716 struct ofproto_port_stp_status
*s
)
718 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
720 VLOG_WARN_RL(&rl
, "%s: cannot get STP status on nonexistent "
721 "port %"PRIu16
, ofproto
->name
, ofp_port
);
725 return (ofproto
->ofproto_class
->get_stp_port_status
726 ? ofproto
->ofproto_class
->get_stp_port_status(ofport
, s
)
730 /* Queue DSCP configuration. */
732 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
733 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
734 * to implement QoS. Instead, it is used to implement features which require
735 * knowledge of what queues exist on a port, and some basic information about
738 * Returns 0 if successful, otherwise a positive errno value. */
740 ofproto_port_set_queues(struct ofproto
*ofproto
, uint16_t ofp_port
,
741 const struct ofproto_port_queue
*queues
,
744 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
747 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16
,
748 ofproto
->name
, ofp_port
);
752 return (ofproto
->ofproto_class
->set_queues
753 ? ofproto
->ofproto_class
->set_queues(ofport
, queues
, n_queues
)
757 /* Connectivity Fault Management configuration. */
759 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
761 ofproto_port_clear_cfm(struct ofproto
*ofproto
, uint16_t ofp_port
)
763 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
764 if (ofport
&& ofproto
->ofproto_class
->set_cfm
) {
765 ofproto
->ofproto_class
->set_cfm(ofport
, NULL
);
769 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
770 * basic configuration from the configuration members in 'cfm', and the remote
771 * maintenance point ID from remote_mpid. Ignores the statistics members of
774 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
776 ofproto_port_set_cfm(struct ofproto
*ofproto
, uint16_t ofp_port
,
777 const struct cfm_settings
*s
)
779 struct ofport
*ofport
;
782 ofport
= ofproto_get_port(ofproto
, ofp_port
);
784 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16
,
785 ofproto
->name
, ofp_port
);
789 /* XXX: For configuration simplicity, we only support one remote_mpid
790 * outside of the CFM module. It's not clear if this is the correct long
791 * term solution or not. */
792 error
= (ofproto
->ofproto_class
->set_cfm
793 ? ofproto
->ofproto_class
->set_cfm(ofport
, s
)
796 VLOG_WARN("%s: CFM configuration on port %"PRIu16
" (%s) failed (%s)",
797 ofproto
->name
, ofp_port
, netdev_get_name(ofport
->netdev
),
802 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
803 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
804 * 0 if LACP partner information is not current (generally indicating a
805 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
807 ofproto_port_is_lacp_current(struct ofproto
*ofproto
, uint16_t ofp_port
)
809 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
810 return (ofport
&& ofproto
->ofproto_class
->port_is_lacp_current
811 ? ofproto
->ofproto_class
->port_is_lacp_current(ofport
)
817 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
818 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
819 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
820 * configuration plus, if there is more than one slave, a bonding
823 * If 'aux' is already registered then this function updates its configuration
824 * to 's'. Otherwise, this function registers a new bundle.
826 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
829 ofproto_bundle_register(struct ofproto
*ofproto
, void *aux
,
830 const struct ofproto_bundle_settings
*s
)
832 return (ofproto
->ofproto_class
->bundle_set
833 ? ofproto
->ofproto_class
->bundle_set(ofproto
, aux
, s
)
837 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
838 * If no such bundle has been registered, this has no effect. */
840 ofproto_bundle_unregister(struct ofproto
*ofproto
, void *aux
)
842 return ofproto_bundle_register(ofproto
, aux
, NULL
);
846 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
847 * If 'aux' is already registered then this function updates its configuration
848 * to 's'. Otherwise, this function registers a new mirror. */
850 ofproto_mirror_register(struct ofproto
*ofproto
, void *aux
,
851 const struct ofproto_mirror_settings
*s
)
853 return (ofproto
->ofproto_class
->mirror_set
854 ? ofproto
->ofproto_class
->mirror_set(ofproto
, aux
, s
)
858 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
859 * If no mirror has been registered, this has no effect. */
861 ofproto_mirror_unregister(struct ofproto
*ofproto
, void *aux
)
863 return ofproto_mirror_register(ofproto
, aux
, NULL
);
866 /* Retrieves statistics from mirror associated with client data pointer
867 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
868 * 'bytes', respectively. If a particular counters is not supported,
869 * the appropriate argument is set to UINT64_MAX. */
871 ofproto_mirror_get_stats(struct ofproto
*ofproto
, void *aux
,
872 uint64_t *packets
, uint64_t *bytes
)
874 if (!ofproto
->ofproto_class
->mirror_get_stats
) {
875 *packets
= *bytes
= UINT64_MAX
;
879 return ofproto
->ofproto_class
->mirror_get_stats(ofproto
, aux
,
883 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
884 * which all packets are flooded, instead of using MAC learning. If
885 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
887 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
890 ofproto_set_flood_vlans(struct ofproto
*ofproto
, unsigned long *flood_vlans
)
892 return (ofproto
->ofproto_class
->set_flood_vlans
893 ? ofproto
->ofproto_class
->set_flood_vlans(ofproto
, flood_vlans
)
897 /* Returns true if 'aux' is a registered bundle that is currently in use as the
898 * output for a mirror. */
900 ofproto_is_mirror_output_bundle(const struct ofproto
*ofproto
, void *aux
)
902 return (ofproto
->ofproto_class
->is_mirror_output_bundle
903 ? ofproto
->ofproto_class
->is_mirror_output_bundle(ofproto
, aux
)
907 /* Configuration of OpenFlow tables. */
909 /* Returns the number of OpenFlow tables in 'ofproto'. */
911 ofproto_get_n_tables(const struct ofproto
*ofproto
)
913 return ofproto
->n_tables
;
916 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
917 * settings from 's'. 'table_id' must be in the range 0 through the number of
918 * OpenFlow tables in 'ofproto' minus 1, inclusive.
920 * For read-only tables, only the name may be configured. */
922 ofproto_configure_table(struct ofproto
*ofproto
, int table_id
,
923 const struct ofproto_table_settings
*s
)
925 struct oftable
*table
;
927 ovs_assert(table_id
>= 0 && table_id
< ofproto
->n_tables
);
928 table
= &ofproto
->tables
[table_id
];
930 oftable_set_name(table
, s
->name
);
932 if (table
->flags
& OFTABLE_READONLY
) {
937 oftable_enable_eviction(table
, s
->groups
, s
->n_groups
);
939 oftable_disable_eviction(table
);
942 table
->max_flows
= s
->max_flows
;
943 if (classifier_count(&table
->cls
) > table
->max_flows
944 && table
->eviction_fields
) {
945 /* 'table' contains more flows than allowed. We might not be able to
946 * evict them right away because of the asynchronous nature of flow
947 * table changes. Schedule eviction for later. */
948 switch (ofproto
->state
) {
950 ofproto
->state
= S_EVICT
;
954 /* We're already deleting flows, nothing more to do. */
961 ofproto_has_snoops(const struct ofproto
*ofproto
)
963 return connmgr_has_snoops(ofproto
->connmgr
);
967 ofproto_get_snoops(const struct ofproto
*ofproto
, struct sset
*snoops
)
969 connmgr_get_snoops(ofproto
->connmgr
, snoops
);
973 ofproto_flush__(struct ofproto
*ofproto
)
975 struct ofopgroup
*group
;
976 struct oftable
*table
;
978 if (ofproto
->ofproto_class
->flush
) {
979 ofproto
->ofproto_class
->flush(ofproto
);
982 group
= ofopgroup_create_unattached(ofproto
);
983 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
984 struct rule
*rule
, *next_rule
;
985 struct cls_cursor cursor
;
987 if (table
->flags
& OFTABLE_HIDDEN
) {
991 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
992 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
993 if (!rule
->pending
) {
994 ofoperation_create(group
, rule
, OFOPERATION_DELETE
,
996 oftable_remove_rule(rule
);
997 ofproto
->ofproto_class
->rule_destruct(rule
);
1001 ofopgroup_submit(group
);
1005 ofproto_destroy__(struct ofproto
*ofproto
)
1007 struct oftable
*table
;
1009 ovs_assert(list_is_empty(&ofproto
->pending
));
1010 ovs_assert(!ofproto
->n_pending
);
1012 connmgr_destroy(ofproto
->connmgr
);
1014 hmap_remove(&all_ofprotos
, &ofproto
->hmap_node
);
1015 free(ofproto
->name
);
1016 free(ofproto
->type
);
1017 free(ofproto
->mfr_desc
);
1018 free(ofproto
->hw_desc
);
1019 free(ofproto
->sw_desc
);
1020 free(ofproto
->serial_desc
);
1021 free(ofproto
->dp_desc
);
1022 hmap_destroy(&ofproto
->ports
);
1023 shash_destroy(&ofproto
->port_by_name
);
1024 bitmap_free(ofproto
->ofp_port_ids
);
1025 simap_destroy(&ofproto
->ofp_requests
);
1027 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
1028 oftable_destroy(table
);
1030 free(ofproto
->tables
);
1032 hmap_destroy(&ofproto
->deletions
);
1034 free(ofproto
->vlan_bitmap
);
1036 ofproto
->ofproto_class
->dealloc(ofproto
);
1040 ofproto_destroy(struct ofproto
*p
)
1042 struct ofport
*ofport
, *next_ofport
;
1049 HMAP_FOR_EACH_SAFE (ofport
, next_ofport
, hmap_node
, &p
->ports
) {
1050 ofport_destroy(ofport
);
1053 p
->ofproto_class
->destruct(p
);
1054 ofproto_destroy__(p
);
1057 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1058 * kernel datapath, for example, this destroys the datapath in the kernel, and
1059 * with the netdev-based datapath, it tears down the data structures that
1060 * represent the datapath.
1062 * The datapath should not be currently open as an ofproto. */
1064 ofproto_delete(const char *name
, const char *type
)
1066 const struct ofproto_class
*class = ofproto_class_find__(type
);
1067 return (!class ? EAFNOSUPPORT
1068 : !class->del
? EACCES
1069 : class->del(type
, name
));
1073 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
1075 if (error
== ENOBUFS
) {
1076 reinit_ports(ofproto
);
1077 } else if (!error
) {
1078 update_port(ofproto
, devname
);
1084 ofproto_type_run(const char *datapath_type
)
1086 const struct ofproto_class
*class;
1089 datapath_type
= ofproto_normalize_type(datapath_type
);
1090 class = ofproto_class_find__(datapath_type
);
1092 error
= class->type_run
? class->type_run(datapath_type
) : 0;
1093 if (error
&& error
!= EAGAIN
) {
1094 VLOG_ERR_RL(&rl
, "%s: type_run failed (%s)",
1095 datapath_type
, strerror(error
));
1101 ofproto_type_run_fast(const char *datapath_type
)
1103 const struct ofproto_class
*class;
1106 datapath_type
= ofproto_normalize_type(datapath_type
);
1107 class = ofproto_class_find__(datapath_type
);
1109 error
= class->type_run_fast
? class->type_run_fast(datapath_type
) : 0;
1110 if (error
&& error
!= EAGAIN
) {
1111 VLOG_ERR_RL(&rl
, "%s: type_run_fast failed (%s)",
1112 datapath_type
, strerror(error
));
1118 ofproto_type_wait(const char *datapath_type
)
1120 const struct ofproto_class
*class;
1122 datapath_type
= ofproto_normalize_type(datapath_type
);
1123 class = ofproto_class_find__(datapath_type
);
1125 if (class->type_wait
) {
1126 class->type_wait(datapath_type
);
1131 ofproto_run(struct ofproto
*p
)
1133 struct sset changed_netdevs
;
1134 const char *changed_netdev
;
1135 struct ofport
*ofport
;
1138 error
= p
->ofproto_class
->run(p
);
1139 if (error
&& error
!= EAGAIN
) {
1140 VLOG_ERR_RL(&rl
, "%s: run failed (%s)", p
->name
, strerror(error
));
1143 if (p
->ofproto_class
->port_poll
) {
1146 while ((error
= p
->ofproto_class
->port_poll(p
, &devname
)) != EAGAIN
) {
1147 process_port_change(p
, error
, devname
);
1151 /* Update OpenFlow port status for any port whose netdev has changed.
1153 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1154 * destroyed, so it's not safe to update ports directly from the
1155 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1156 * need this two-phase approach. */
1157 sset_init(&changed_netdevs
);
1158 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1159 unsigned int change_seq
= netdev_change_seq(ofport
->netdev
);
1160 if (ofport
->change_seq
!= change_seq
) {
1161 ofport
->change_seq
= change_seq
;
1162 sset_add(&changed_netdevs
, netdev_get_name(ofport
->netdev
));
1165 SSET_FOR_EACH (changed_netdev
, &changed_netdevs
) {
1166 update_port(p
, changed_netdev
);
1168 sset_destroy(&changed_netdevs
);
1172 connmgr_run(p
->connmgr
, handle_openflow
);
1176 connmgr_run(p
->connmgr
, NULL
);
1178 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1179 p
->state
= S_OPENFLOW
;
1184 connmgr_run(p
->connmgr
, NULL
);
1186 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1187 connmgr_flushed(p
->connmgr
);
1188 p
->state
= S_OPENFLOW
;
1196 if (time_msec() >= p
->next_op_report
) {
1197 long long int ago
= (time_msec() - p
->first_op
) / 1000;
1198 long long int interval
= (p
->last_op
- p
->first_op
) / 1000;
1202 ds_put_format(&s
, "%d flow_mods ",
1203 p
->n_add
+ p
->n_delete
+ p
->n_modify
);
1204 if (interval
== ago
) {
1205 ds_put_format(&s
, "in the last %lld s", ago
);
1206 } else if (interval
) {
1207 ds_put_format(&s
, "in the %lld s starting %lld s ago",
1210 ds_put_format(&s
, "%lld s ago", ago
);
1213 ds_put_cstr(&s
, " (");
1215 ds_put_format(&s
, "%d adds, ", p
->n_add
);
1218 ds_put_format(&s
, "%d deletes, ", p
->n_delete
);
1221 ds_put_format(&s
, "%d modifications, ", p
->n_modify
);
1224 ds_put_char(&s
, ')');
1226 VLOG_INFO("%s: %s", p
->name
, ds_cstr(&s
));
1229 p
->n_add
= p
->n_delete
= p
->n_modify
= 0;
1230 p
->next_op_report
= LLONG_MAX
;
1236 /* Performs periodic activity required by 'ofproto' that needs to be done
1237 * with the least possible latency.
1239 * It makes sense to call this function a couple of times per poll loop, to
1240 * provide a significant performance boost on some benchmarks with the
1241 * ofproto-dpif implementation. */
1243 ofproto_run_fast(struct ofproto
*p
)
1247 error
= p
->ofproto_class
->run_fast
? p
->ofproto_class
->run_fast(p
) : 0;
1248 if (error
&& error
!= EAGAIN
) {
1249 VLOG_ERR_RL(&rl
, "%s: fastpath run failed (%s)",
1250 p
->name
, strerror(error
));
1256 ofproto_wait(struct ofproto
*p
)
1258 struct ofport
*ofport
;
1260 p
->ofproto_class
->wait(p
);
1261 if (p
->ofproto_class
->port_poll_wait
) {
1262 p
->ofproto_class
->port_poll_wait(p
);
1265 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1266 if (ofport
->change_seq
!= netdev_change_seq(ofport
->netdev
)) {
1267 poll_immediate_wake();
1273 connmgr_wait(p
->connmgr
, true);
1278 connmgr_wait(p
->connmgr
, false);
1279 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1280 poll_immediate_wake();
1287 ofproto_is_alive(const struct ofproto
*p
)
1289 return connmgr_has_controllers(p
->connmgr
);
1292 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1293 * memory_report(). */
1295 ofproto_get_memory_usage(const struct ofproto
*ofproto
, struct simap
*usage
)
1297 const struct oftable
*table
;
1298 unsigned int n_rules
;
1300 simap_increase(usage
, "ports", hmap_count(&ofproto
->ports
));
1301 simap_increase(usage
, "ops",
1302 ofproto
->n_pending
+ hmap_count(&ofproto
->deletions
));
1305 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
1306 n_rules
+= classifier_count(&table
->cls
);
1308 simap_increase(usage
, "rules", n_rules
);
1310 if (ofproto
->ofproto_class
->get_memory_usage
) {
1311 ofproto
->ofproto_class
->get_memory_usage(ofproto
, usage
);
1314 connmgr_get_memory_usage(ofproto
->connmgr
, usage
);
1318 ofproto_get_ofproto_controller_info(const struct ofproto
*ofproto
,
1321 connmgr_get_controller_info(ofproto
->connmgr
, info
);
1325 ofproto_free_ofproto_controller_info(struct shash
*info
)
1327 connmgr_free_controller_info(info
);
1330 /* Makes a deep copy of 'old' into 'port'. */
1332 ofproto_port_clone(struct ofproto_port
*port
, const struct ofproto_port
*old
)
1334 port
->name
= xstrdup(old
->name
);
1335 port
->type
= xstrdup(old
->type
);
1336 port
->ofp_port
= old
->ofp_port
;
1339 /* Frees memory allocated to members of 'ofproto_port'.
1341 * Do not call this function on an ofproto_port obtained from
1342 * ofproto_port_dump_next(): that function retains ownership of the data in the
1345 ofproto_port_destroy(struct ofproto_port
*ofproto_port
)
1347 free(ofproto_port
->name
);
1348 free(ofproto_port
->type
);
1351 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1353 * This function provides no status indication. An error status for the entire
1354 * dump operation is provided when it is completed by calling
1355 * ofproto_port_dump_done().
1358 ofproto_port_dump_start(struct ofproto_port_dump
*dump
,
1359 const struct ofproto
*ofproto
)
1361 dump
->ofproto
= ofproto
;
1362 dump
->error
= ofproto
->ofproto_class
->port_dump_start(ofproto
,
1366 /* Attempts to retrieve another port from 'dump', which must have been created
1367 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1368 * 'port' and returns true. On failure, returns false.
1370 * Failure might indicate an actual error or merely that the last port has been
1371 * dumped. An error status for the entire dump operation is provided when it
1372 * is completed by calling ofproto_port_dump_done().
1374 * The ofproto owns the data stored in 'port'. It will remain valid until at
1375 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1376 * ofproto_port_dump_done(). */
1378 ofproto_port_dump_next(struct ofproto_port_dump
*dump
,
1379 struct ofproto_port
*port
)
1381 const struct ofproto
*ofproto
= dump
->ofproto
;
1387 dump
->error
= ofproto
->ofproto_class
->port_dump_next(ofproto
, dump
->state
,
1390 ofproto
->ofproto_class
->port_dump_done(ofproto
, dump
->state
);
1396 /* Completes port table dump operation 'dump', which must have been created
1397 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1398 * error-free, otherwise a positive errno value describing the problem. */
1400 ofproto_port_dump_done(struct ofproto_port_dump
*dump
)
1402 const struct ofproto
*ofproto
= dump
->ofproto
;
1404 dump
->error
= ofproto
->ofproto_class
->port_dump_done(ofproto
,
1407 return dump
->error
== EOF
? 0 : dump
->error
;
1410 /* Returns the type to pass to netdev_open() when a datapath of type
1411 * 'datapath_type' has a port of type 'port_type', for a few special
1412 * cases when a netdev type differs from a port type. For example, when
1413 * using the userspace datapath, a port of type "internal" needs to be
1416 * Returns either 'type' itself or a string literal, which must not be
1419 ofproto_port_open_type(const char *datapath_type
, const char *port_type
)
1421 const struct ofproto_class
*class;
1423 datapath_type
= ofproto_normalize_type(datapath_type
);
1424 class = ofproto_class_find__(datapath_type
);
1429 return (class->port_open_type
1430 ? class->port_open_type(datapath_type
, port_type
)
1434 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1435 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1436 * the port's OpenFlow port number.
1438 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1439 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1440 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1441 * 'ofp_portp' is non-null). */
1443 ofproto_port_add(struct ofproto
*ofproto
, struct netdev
*netdev
,
1444 uint16_t *ofp_portp
)
1446 uint16_t ofp_port
= ofp_portp
? *ofp_portp
: OFPP_NONE
;
1449 error
= ofproto
->ofproto_class
->port_add(ofproto
, netdev
);
1451 const char *netdev_name
= netdev_get_name(netdev
);
1453 simap_put(&ofproto
->ofp_requests
, netdev_name
, ofp_port
);
1454 update_port(ofproto
, netdev_name
);
1457 struct ofproto_port ofproto_port
;
1459 ofproto_port_query_by_name(ofproto
, netdev_get_name(netdev
),
1461 *ofp_portp
= error
? OFPP_NONE
: ofproto_port
.ofp_port
;
1462 ofproto_port_destroy(&ofproto_port
);
1467 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1468 * initializes '*port' appropriately; on failure, returns a positive errno
1471 * The caller owns the data in 'ofproto_port' and must free it with
1472 * ofproto_port_destroy() when it is no longer needed. */
1474 ofproto_port_query_by_name(const struct ofproto
*ofproto
, const char *devname
,
1475 struct ofproto_port
*port
)
1479 error
= ofproto
->ofproto_class
->port_query_by_name(ofproto
, devname
, port
);
1481 memset(port
, 0, sizeof *port
);
1486 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1487 * Returns 0 if successful, otherwise a positive errno. */
1489 ofproto_port_del(struct ofproto
*ofproto
, uint16_t ofp_port
)
1491 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
1492 const char *name
= ofport
? netdev_get_name(ofport
->netdev
) : "<unknown>";
1493 struct simap_node
*ofp_request_node
;
1496 ofp_request_node
= simap_find(&ofproto
->ofp_requests
, name
);
1497 if (ofp_request_node
) {
1498 simap_delete(&ofproto
->ofp_requests
, ofp_request_node
);
1501 error
= ofproto
->ofproto_class
->port_del(ofproto
, ofp_port
);
1502 if (!error
&& ofport
) {
1503 /* 'name' is the netdev's name and update_port() is going to close the
1504 * netdev. Just in case update_port() refers to 'name' after it
1505 * destroys 'ofport', make a copy of it around the update_port()
1507 char *devname
= xstrdup(name
);
1508 update_port(ofproto
, devname
);
1514 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1515 * performs the 'n_actions' actions in 'actions'. The new flow will not
1518 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1519 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1520 * controllers; otherwise, it will be hidden.
1522 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1524 * This is a helper function for in-band control and fail-open. */
1526 ofproto_add_flow(struct ofproto
*ofproto
, const struct match
*match
,
1527 unsigned int priority
,
1528 const struct ofpact
*ofpacts
, size_t ofpacts_len
)
1530 const struct rule
*rule
;
1532 rule
= rule_from_cls_rule(classifier_find_match_exactly(
1533 &ofproto
->tables
[0].cls
, match
, priority
));
1534 if (!rule
|| !ofpacts_equal(rule
->ofpacts
, rule
->ofpacts_len
,
1535 ofpacts
, ofpacts_len
)) {
1536 struct ofputil_flow_mod fm
;
1538 memset(&fm
, 0, sizeof fm
);
1540 fm
.priority
= priority
;
1541 fm
.buffer_id
= UINT32_MAX
;
1542 fm
.ofpacts
= xmemdup(ofpacts
, ofpacts_len
);
1543 fm
.ofpacts_len
= ofpacts_len
;
1544 add_flow(ofproto
, NULL
, &fm
, NULL
);
1549 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1550 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1551 * operation cannot be initiated now but may be retried later.
1553 * This is a helper function for in-band control and fail-open. */
1555 ofproto_flow_mod(struct ofproto
*ofproto
, const struct ofputil_flow_mod
*fm
)
1557 return handle_flow_mod__(ofproto
, NULL
, fm
, NULL
);
1560 /* Searches for a rule with matching criteria exactly equal to 'target' in
1561 * ofproto's table 0 and, if it finds one, deletes it.
1563 * This is a helper function for in-band control and fail-open. */
1565 ofproto_delete_flow(struct ofproto
*ofproto
,
1566 const struct match
*target
, unsigned int priority
)
1570 rule
= rule_from_cls_rule(classifier_find_match_exactly(
1571 &ofproto
->tables
[0].cls
, target
, priority
));
1573 /* No such rule -> success. */
1575 } else if (rule
->pending
) {
1576 /* An operation on the rule is already pending -> failure.
1577 * Caller must retry later if it's important. */
1580 /* Initiate deletion -> success. */
1581 struct ofopgroup
*group
= ofopgroup_create_unattached(ofproto
);
1582 ofoperation_create(group
, rule
, OFOPERATION_DELETE
, OFPRR_DELETE
);
1583 oftable_remove_rule(rule
);
1584 ofproto
->ofproto_class
->rule_destruct(rule
);
1585 ofopgroup_submit(group
);
1591 /* Starts the process of deleting all of the flows from all of ofproto's flow
1592 * tables and then reintroducing the flows required by in-band control and
1593 * fail-open. The process will complete in a later call to ofproto_run(). */
1595 ofproto_flush_flows(struct ofproto
*ofproto
)
1597 COVERAGE_INC(ofproto_flush
);
1598 ofproto
->state
= S_FLUSH
;
1602 reinit_ports(struct ofproto
*p
)
1604 struct ofproto_port_dump dump
;
1605 struct sset devnames
;
1606 struct ofport
*ofport
;
1607 struct ofproto_port ofproto_port
;
1608 const char *devname
;
1610 COVERAGE_INC(ofproto_reinit_ports
);
1612 sset_init(&devnames
);
1613 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1614 sset_add(&devnames
, netdev_get_name(ofport
->netdev
));
1616 OFPROTO_PORT_FOR_EACH (&ofproto_port
, &dump
, p
) {
1617 sset_add(&devnames
, ofproto_port
.name
);
1620 SSET_FOR_EACH (devname
, &devnames
) {
1621 update_port(p
, devname
);
1623 sset_destroy(&devnames
);
1627 alloc_ofp_port(struct ofproto
*ofproto
, const char *netdev_name
)
1630 uint16_t end_port_no
= ofproto
->alloc_port_no
;
1632 ofp_port
= simap_get(&ofproto
->ofp_requests
, netdev_name
);
1633 ofp_port
= ofp_port
? ofp_port
: OFPP_NONE
;
1635 if (ofp_port
>= ofproto
->max_ports
1636 || bitmap_is_set(ofproto
->ofp_port_ids
, ofp_port
)) {
1637 /* Search for a free OpenFlow port number. We try not to
1638 * immediately reuse them to prevent problems due to old
1641 if (++ofproto
->alloc_port_no
>= ofproto
->max_ports
) {
1642 ofproto
->alloc_port_no
= 0;
1644 if (!bitmap_is_set(ofproto
->ofp_port_ids
,
1645 ofproto
->alloc_port_no
)) {
1646 ofp_port
= ofproto
->alloc_port_no
;
1649 if (ofproto
->alloc_port_no
== end_port_no
) {
1654 bitmap_set1(ofproto
->ofp_port_ids
, ofp_port
);
1659 dealloc_ofp_port(const struct ofproto
*ofproto
, uint16_t ofp_port
)
1661 if (ofp_port
< ofproto
->max_ports
) {
1662 bitmap_set0(ofproto
->ofp_port_ids
, ofp_port
);
1666 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1667 * pointer if the netdev cannot be opened. On success, also fills in
1669 static struct netdev
*
1670 ofport_open(struct ofproto
*ofproto
,
1671 struct ofproto_port
*ofproto_port
,
1672 struct ofputil_phy_port
*pp
)
1674 enum netdev_flags flags
;
1675 struct netdev
*netdev
;
1678 error
= netdev_open(ofproto_port
->name
, ofproto_port
->type
, &netdev
);
1680 VLOG_WARN_RL(&rl
, "%s: ignoring port %s (%"PRIu16
") because netdev %s "
1681 "cannot be opened (%s)",
1683 ofproto_port
->name
, ofproto_port
->ofp_port
,
1684 ofproto_port
->name
, strerror(error
));
1688 if (ofproto_port
->ofp_port
== OFPP_NONE
) {
1689 if (!strcmp(ofproto
->name
, ofproto_port
->name
)) {
1690 ofproto_port
->ofp_port
= OFPP_LOCAL
;
1692 ofproto_port
->ofp_port
= alloc_ofp_port(ofproto
,
1693 ofproto_port
->name
);
1696 pp
->port_no
= ofproto_port
->ofp_port
;
1697 netdev_get_etheraddr(netdev
, pp
->hw_addr
);
1698 ovs_strlcpy(pp
->name
, ofproto_port
->name
, sizeof pp
->name
);
1699 netdev_get_flags(netdev
, &flags
);
1700 pp
->config
= flags
& NETDEV_UP
? 0 : OFPUTIL_PC_PORT_DOWN
;
1701 pp
->state
= netdev_get_carrier(netdev
) ? 0 : OFPUTIL_PS_LINK_DOWN
;
1702 netdev_get_features(netdev
, &pp
->curr
, &pp
->advertised
,
1703 &pp
->supported
, &pp
->peer
);
1704 pp
->curr_speed
= netdev_features_to_bps(pp
->curr
, 0);
1705 pp
->max_speed
= netdev_features_to_bps(pp
->supported
, 0);
1710 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1711 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1714 ofport_equal(const struct ofputil_phy_port
*a
,
1715 const struct ofputil_phy_port
*b
)
1717 return (eth_addr_equals(a
->hw_addr
, b
->hw_addr
)
1718 && a
->state
== b
->state
1719 && !((a
->config
^ b
->config
) & OFPUTIL_PC_PORT_DOWN
)
1720 && a
->curr
== b
->curr
1721 && a
->advertised
== b
->advertised
1722 && a
->supported
== b
->supported
1723 && a
->peer
== b
->peer
1724 && a
->curr_speed
== b
->curr_speed
1725 && a
->max_speed
== b
->max_speed
);
1728 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1729 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1730 * one with the same name or port number). */
1732 ofport_install(struct ofproto
*p
,
1733 struct netdev
*netdev
, const struct ofputil_phy_port
*pp
)
1735 const char *netdev_name
= netdev_get_name(netdev
);
1736 struct ofport
*ofport
;
1739 /* Create ofport. */
1740 ofport
= p
->ofproto_class
->port_alloc();
1745 ofport
->ofproto
= p
;
1746 ofport
->netdev
= netdev
;
1747 ofport
->change_seq
= netdev_change_seq(netdev
);
1749 ofport
->ofp_port
= pp
->port_no
;
1751 /* Add port to 'p'. */
1752 hmap_insert(&p
->ports
, &ofport
->hmap_node
, hash_int(ofport
->ofp_port
, 0));
1753 shash_add(&p
->port_by_name
, netdev_name
, ofport
);
1755 update_mtu(p
, ofport
);
1757 /* Let the ofproto_class initialize its private data. */
1758 error
= p
->ofproto_class
->port_construct(ofport
);
1762 connmgr_send_port_status(p
->connmgr
, pp
, OFPPR_ADD
);
1766 VLOG_WARN_RL(&rl
, "%s: could not add port %s (%s)",
1767 p
->name
, netdev_name
, strerror(error
));
1769 ofport_destroy__(ofport
);
1771 netdev_close(netdev
);
1775 /* Removes 'ofport' from 'p' and destroys it. */
1777 ofport_remove(struct ofport
*ofport
)
1779 connmgr_send_port_status(ofport
->ofproto
->connmgr
, &ofport
->pp
,
1781 ofport_destroy(ofport
);
1784 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1787 ofport_remove_with_name(struct ofproto
*ofproto
, const char *name
)
1789 struct ofport
*port
= shash_find_data(&ofproto
->port_by_name
, name
);
1791 ofport_remove(port
);
1795 /* Updates 'port' with new 'pp' description.
1797 * Does not handle a name or port number change. The caller must implement
1798 * such a change as a delete followed by an add. */
1800 ofport_modified(struct ofport
*port
, struct ofputil_phy_port
*pp
)
1802 memcpy(port
->pp
.hw_addr
, pp
->hw_addr
, ETH_ADDR_LEN
);
1803 port
->pp
.config
= ((port
->pp
.config
& ~OFPUTIL_PC_PORT_DOWN
)
1804 | (pp
->config
& OFPUTIL_PC_PORT_DOWN
));
1805 port
->pp
.state
= pp
->state
;
1806 port
->pp
.curr
= pp
->curr
;
1807 port
->pp
.advertised
= pp
->advertised
;
1808 port
->pp
.supported
= pp
->supported
;
1809 port
->pp
.peer
= pp
->peer
;
1810 port
->pp
.curr_speed
= pp
->curr_speed
;
1811 port
->pp
.max_speed
= pp
->max_speed
;
1813 connmgr_send_port_status(port
->ofproto
->connmgr
, &port
->pp
, OFPPR_MODIFY
);
1816 /* Update OpenFlow 'state' in 'port' and notify controller. */
1818 ofproto_port_set_state(struct ofport
*port
, enum ofputil_port_state state
)
1820 if (port
->pp
.state
!= state
) {
1821 port
->pp
.state
= state
;
1822 connmgr_send_port_status(port
->ofproto
->connmgr
, &port
->pp
,
1828 ofproto_port_unregister(struct ofproto
*ofproto
, uint16_t ofp_port
)
1830 struct ofport
*port
= ofproto_get_port(ofproto
, ofp_port
);
1832 if (port
->ofproto
->ofproto_class
->set_realdev
) {
1833 port
->ofproto
->ofproto_class
->set_realdev(port
, 0, 0);
1835 if (port
->ofproto
->ofproto_class
->set_stp_port
) {
1836 port
->ofproto
->ofproto_class
->set_stp_port(port
, NULL
);
1838 if (port
->ofproto
->ofproto_class
->set_cfm
) {
1839 port
->ofproto
->ofproto_class
->set_cfm(port
, NULL
);
1841 if (port
->ofproto
->ofproto_class
->bundle_remove
) {
1842 port
->ofproto
->ofproto_class
->bundle_remove(port
);
1848 ofport_destroy__(struct ofport
*port
)
1850 struct ofproto
*ofproto
= port
->ofproto
;
1851 const char *name
= netdev_get_name(port
->netdev
);
1853 hmap_remove(&ofproto
->ports
, &port
->hmap_node
);
1854 shash_delete(&ofproto
->port_by_name
,
1855 shash_find(&ofproto
->port_by_name
, name
));
1857 netdev_close(port
->netdev
);
1858 ofproto
->ofproto_class
->port_dealloc(port
);
1862 ofport_destroy(struct ofport
*port
)
1865 dealloc_ofp_port(port
->ofproto
, port
->ofp_port
);
1866 port
->ofproto
->ofproto_class
->port_destruct(port
);
1867 ofport_destroy__(port
);
1872 ofproto_get_port(const struct ofproto
*ofproto
, uint16_t ofp_port
)
1874 struct ofport
*port
;
1876 HMAP_FOR_EACH_IN_BUCKET (port
, hmap_node
,
1877 hash_int(ofp_port
, 0), &ofproto
->ports
) {
1878 if (port
->ofp_port
== ofp_port
) {
1886 ofproto_port_get_stats(const struct ofport
*port
, struct netdev_stats
*stats
)
1888 struct ofproto
*ofproto
= port
->ofproto
;
1891 if (ofproto
->ofproto_class
->port_get_stats
) {
1892 error
= ofproto
->ofproto_class
->port_get_stats(port
, stats
);
1901 update_port(struct ofproto
*ofproto
, const char *name
)
1903 struct ofproto_port ofproto_port
;
1904 struct ofputil_phy_port pp
;
1905 struct netdev
*netdev
;
1906 struct ofport
*port
;
1908 COVERAGE_INC(ofproto_update_port
);
1910 /* Fetch 'name''s location and properties from the datapath. */
1911 netdev
= (!ofproto_port_query_by_name(ofproto
, name
, &ofproto_port
)
1912 ? ofport_open(ofproto
, &ofproto_port
, &pp
)
1915 port
= ofproto_get_port(ofproto
, ofproto_port
.ofp_port
);
1916 if (port
&& !strcmp(netdev_get_name(port
->netdev
), name
)) {
1917 struct netdev
*old_netdev
= port
->netdev
;
1919 /* 'name' hasn't changed location. Any properties changed? */
1920 if (!ofport_equal(&port
->pp
, &pp
)) {
1921 ofport_modified(port
, &pp
);
1924 update_mtu(ofproto
, port
);
1926 /* Install the newly opened netdev in case it has changed.
1927 * Don't close the old netdev yet in case port_modified has to
1928 * remove a retained reference to it.*/
1929 port
->netdev
= netdev
;
1930 port
->change_seq
= netdev_change_seq(netdev
);
1932 if (port
->ofproto
->ofproto_class
->port_modified
) {
1933 port
->ofproto
->ofproto_class
->port_modified(port
);
1936 netdev_close(old_netdev
);
1938 /* If 'port' is nonnull then its name differs from 'name' and thus
1939 * we should delete it. If we think there's a port named 'name'
1940 * then its port number must be wrong now so delete it too. */
1942 ofport_remove(port
);
1944 ofport_remove_with_name(ofproto
, name
);
1945 ofport_install(ofproto
, netdev
, &pp
);
1948 /* Any port named 'name' is gone now. */
1949 ofport_remove_with_name(ofproto
, name
);
1951 ofproto_port_destroy(&ofproto_port
);
1955 init_ports(struct ofproto
*p
)
1957 struct ofproto_port_dump dump
;
1958 struct ofproto_port ofproto_port
;
1959 struct shash_node
*node
, *next
;
1961 OFPROTO_PORT_FOR_EACH (&ofproto_port
, &dump
, p
) {
1962 const char *name
= ofproto_port
.name
;
1964 if (shash_find(&p
->port_by_name
, name
)) {
1965 VLOG_WARN_RL(&rl
, "%s: ignoring duplicate device %s in datapath",
1968 struct ofputil_phy_port pp
;
1969 struct netdev
*netdev
;
1971 /* Check if an OpenFlow port number had been requested. */
1972 node
= shash_find(&init_ofp_ports
, name
);
1974 const struct iface_hint
*iface_hint
= node
->data
;
1975 simap_put(&p
->ofp_requests
, name
, iface_hint
->ofp_port
);
1978 netdev
= ofport_open(p
, &ofproto_port
, &pp
);
1980 ofport_install(p
, netdev
, &pp
);
1985 SHASH_FOR_EACH_SAFE(node
, next
, &init_ofp_ports
) {
1986 struct iface_hint
*iface_hint
= node
->data
;
1988 if (!strcmp(iface_hint
->br_name
, p
->name
)) {
1989 free(iface_hint
->br_name
);
1990 free(iface_hint
->br_type
);
1992 shash_delete(&init_ofp_ports
, node
);
1999 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2000 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2002 find_min_mtu(struct ofproto
*p
)
2004 struct ofport
*ofport
;
2007 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
2008 struct netdev
*netdev
= ofport
->netdev
;
2011 /* Skip any internal ports, since that's what we're trying to
2013 if (!strcmp(netdev_get_type(netdev
), "internal")) {
2017 if (netdev_get_mtu(netdev
, &dev_mtu
)) {
2020 if (!mtu
|| dev_mtu
< mtu
) {
2025 return mtu
? mtu
: ETH_PAYLOAD_MAX
;
2028 /* Update MTU of all datapath devices on 'p' to the minimum of the
2029 * non-datapath ports in event of 'port' added or changed. */
2031 update_mtu(struct ofproto
*p
, struct ofport
*port
)
2033 struct ofport
*ofport
;
2034 struct netdev
*netdev
= port
->netdev
;
2035 int dev_mtu
, old_min
;
2037 if (netdev_get_mtu(netdev
, &dev_mtu
)) {
2041 if (!strcmp(netdev_get_type(port
->netdev
), "internal")) {
2042 if (dev_mtu
> p
->min_mtu
) {
2043 if (!netdev_set_mtu(port
->netdev
, p
->min_mtu
)) {
2044 dev_mtu
= p
->min_mtu
;
2047 port
->mtu
= dev_mtu
;
2051 /* For non-internal port find new min mtu. */
2052 old_min
= p
->min_mtu
;
2053 port
->mtu
= dev_mtu
;
2054 p
->min_mtu
= find_min_mtu(p
);
2055 if (p
->min_mtu
== old_min
) {
2059 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
2060 struct netdev
*netdev
= ofport
->netdev
;
2062 if (!strcmp(netdev_get_type(netdev
), "internal")) {
2063 if (!netdev_set_mtu(netdev
, p
->min_mtu
)) {
2064 ofport
->mtu
= p
->min_mtu
;
2071 ofproto_rule_destroy__(struct rule
*rule
)
2074 cls_rule_destroy(&rule
->cr
);
2075 free(rule
->ofpacts
);
2076 rule
->ofproto
->ofproto_class
->rule_dealloc(rule
);
2080 /* This function allows an ofproto implementation to destroy any rules that
2081 * remain when its ->destruct() function is called. The caller must have
2082 * already uninitialized any derived members of 'rule' (step 5 described in the
2083 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
2084 * This function implements steps 6 and 7.
2086 * This function should only be called from an ofproto implementation's
2087 * ->destruct() function. It is not suitable elsewhere. */
2089 ofproto_rule_destroy(struct rule
*rule
)
2091 ovs_assert(!rule
->pending
);
2092 oftable_remove_rule(rule
);
2093 ofproto_rule_destroy__(rule
);
2096 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2097 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2099 ofproto_rule_has_out_port(const struct rule
*rule
, uint16_t port
)
2101 return (port
== OFPP_ANY
2102 || ofpacts_output_to_port(rule
->ofpacts
, rule
->ofpacts_len
, port
));
2105 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2106 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2108 ofoperation_has_out_port(const struct ofoperation
*op
, uint16_t out_port
)
2110 if (ofproto_rule_has_out_port(op
->rule
, out_port
)) {
2115 case OFOPERATION_ADD
:
2116 return op
->victim
&& ofproto_rule_has_out_port(op
->victim
, out_port
);
2118 case OFOPERATION_DELETE
:
2121 case OFOPERATION_MODIFY
:
2122 return ofpacts_output_to_port(op
->ofpacts
, op
->ofpacts_len
, out_port
);
2128 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2129 * statistics appropriately. 'packet' must have at least sizeof(struct
2130 * ofp10_packet_in) bytes of headroom.
2132 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2133 * with statistics for 'packet' either way.
2135 * Takes ownership of 'packet'. */
2137 rule_execute(struct rule
*rule
, uint16_t in_port
, struct ofpbuf
*packet
)
2141 ovs_assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp10_packet_in
));
2143 flow_extract(packet
, 0, 0, NULL
, in_port
, &flow
);
2144 return rule
->ofproto
->ofproto_class
->rule_execute(rule
, &flow
, packet
);
2147 /* Returns true if 'rule' should be hidden from the controller.
2149 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2150 * (e.g. by in-band control) and are intentionally hidden from the
2153 ofproto_rule_is_hidden(const struct rule
*rule
)
2155 return rule
->cr
.priority
> UINT16_MAX
;
2158 static enum oftable_flags
2159 rule_get_flags(const struct rule
*rule
)
2161 return rule
->ofproto
->tables
[rule
->table_id
].flags
;
2165 rule_is_modifiable(const struct rule
*rule
)
2167 return !(rule_get_flags(rule
) & OFTABLE_READONLY
);
2171 handle_echo_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2173 ofconn_send_reply(ofconn
, make_echo_reply(oh
));
2178 handle_features_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2180 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2181 struct ofputil_switch_features features
;
2182 struct ofport
*port
;
2188 ofproto
->ofproto_class
->get_features(ofproto
, &arp_match_ip
,
2190 ovs_assert(features
.actions
& OFPUTIL_A_OUTPUT
); /* sanity check */
2192 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2193 * if present, are always at the end.) */
2194 n_tables
= ofproto
->n_tables
;
2195 for (i
= 0; i
< ofproto
->n_tables
; i
++) {
2196 if (ofproto
->tables
[i
].flags
& OFTABLE_HIDDEN
) {
2202 features
.datapath_id
= ofproto
->datapath_id
;
2203 features
.n_buffers
= pktbuf_capacity();
2204 features
.n_tables
= n_tables
;
2205 features
.capabilities
= (OFPUTIL_C_FLOW_STATS
| OFPUTIL_C_TABLE_STATS
|
2206 OFPUTIL_C_PORT_STATS
| OFPUTIL_C_QUEUE_STATS
);
2208 features
.capabilities
|= OFPUTIL_C_ARP_MATCH_IP
;
2210 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2211 features
.auxiliary_id
= 0;
2212 b
= ofputil_encode_switch_features(&features
, ofconn_get_protocol(ofconn
),
2214 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
2215 ofputil_put_switch_features_port(&port
->pp
, b
);
2218 ofconn_send_reply(ofconn
, b
);
2223 handle_get_config_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2225 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2226 struct ofp_switch_config
*osc
;
2227 enum ofp_config_flags flags
;
2231 buf
= ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY
, oh
, 0);
2232 osc
= ofpbuf_put_uninit(buf
, sizeof *osc
);
2233 flags
= ofproto
->frag_handling
;
2234 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2235 if (oh
->version
< OFP13_VERSION
2236 && ofconn_get_invalid_ttl_to_controller(ofconn
)) {
2237 flags
|= OFPC_INVALID_TTL_TO_CONTROLLER
;
2239 osc
->flags
= htons(flags
);
2240 osc
->miss_send_len
= htons(ofconn_get_miss_send_len(ofconn
));
2241 ofconn_send_reply(ofconn
, buf
);
2247 handle_set_config(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2249 const struct ofp_switch_config
*osc
= ofpmsg_body(oh
);
2250 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2251 uint16_t flags
= ntohs(osc
->flags
);
2253 if (ofconn_get_type(ofconn
) != OFCONN_PRIMARY
2254 || ofconn_get_role(ofconn
) != OFPCR12_ROLE_SLAVE
) {
2255 enum ofp_config_flags cur
= ofproto
->frag_handling
;
2256 enum ofp_config_flags next
= flags
& OFPC_FRAG_MASK
;
2258 ovs_assert((cur
& OFPC_FRAG_MASK
) == cur
);
2260 if (ofproto
->ofproto_class
->set_frag_handling(ofproto
, next
)) {
2261 ofproto
->frag_handling
= next
;
2263 VLOG_WARN_RL(&rl
, "%s: unsupported fragment handling mode %s",
2265 ofputil_frag_handling_to_string(next
));
2269 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2270 ofconn_set_invalid_ttl_to_controller(ofconn
,
2271 (oh
->version
< OFP13_VERSION
2272 && flags
& OFPC_INVALID_TTL_TO_CONTROLLER
));
2274 ofconn_set_miss_send_len(ofconn
, ntohs(osc
->miss_send_len
));
2279 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2280 * error message code for the caller to propagate upward. Otherwise, returns
2283 * The log message mentions 'msg_type'. */
2285 reject_slave_controller(struct ofconn
*ofconn
)
2287 if (ofconn_get_type(ofconn
) == OFCONN_PRIMARY
2288 && ofconn_get_role(ofconn
) == OFPCR12_ROLE_SLAVE
) {
2289 return OFPERR_OFPBRC_EPERM
;
2296 handle_packet_out(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2298 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2299 struct ofputil_packet_out po
;
2300 struct ofpbuf
*payload
;
2301 uint64_t ofpacts_stub
[1024 / 8];
2302 struct ofpbuf ofpacts
;
2306 COVERAGE_INC(ofproto_packet_out
);
2308 error
= reject_slave_controller(ofconn
);
2313 /* Decode message. */
2314 ofpbuf_use_stub(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
2315 error
= ofputil_decode_packet_out(&po
, oh
, &ofpacts
);
2317 goto exit_free_ofpacts
;
2319 if (po
.in_port
>= p
->max_ports
&& po
.in_port
< OFPP_MAX
) {
2320 error
= OFPERR_OFPBRC_BAD_PORT
;
2321 goto exit_free_ofpacts
;
2326 if (po
.buffer_id
!= UINT32_MAX
) {
2327 error
= ofconn_pktbuf_retrieve(ofconn
, po
.buffer_id
, &payload
, NULL
);
2328 if (error
|| !payload
) {
2329 goto exit_free_ofpacts
;
2332 payload
= xmalloc(sizeof *payload
);
2333 ofpbuf_use_const(payload
, po
.packet
, po
.packet_len
);
2336 /* Verify actions against packet, then send packet if successful. */
2337 flow_extract(payload
, 0, 0, NULL
, po
.in_port
, &flow
);
2338 error
= ofpacts_check(po
.ofpacts
, po
.ofpacts_len
, &flow
, p
->max_ports
);
2340 error
= p
->ofproto_class
->packet_out(p
, payload
, &flow
,
2341 po
.ofpacts
, po
.ofpacts_len
);
2343 ofpbuf_delete(payload
);
2346 ofpbuf_uninit(&ofpacts
);
2352 update_port_config(struct ofport
*port
,
2353 enum ofputil_port_config config
,
2354 enum ofputil_port_config mask
)
2356 enum ofputil_port_config old_config
= port
->pp
.config
;
2357 enum ofputil_port_config toggle
;
2359 toggle
= (config
^ port
->pp
.config
) & mask
;
2360 if (toggle
& OFPUTIL_PC_PORT_DOWN
) {
2361 if (config
& OFPUTIL_PC_PORT_DOWN
) {
2362 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, true);
2364 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, true);
2366 toggle
&= ~OFPUTIL_PC_PORT_DOWN
;
2369 port
->pp
.config
^= toggle
;
2370 if (port
->pp
.config
!= old_config
) {
2371 port
->ofproto
->ofproto_class
->port_reconfigured(port
, old_config
);
2376 handle_port_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2378 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2379 struct ofputil_port_mod pm
;
2380 struct ofport
*port
;
2383 error
= reject_slave_controller(ofconn
);
2388 error
= ofputil_decode_port_mod(oh
, &pm
);
2393 port
= ofproto_get_port(p
, pm
.port_no
);
2395 return OFPERR_OFPPMFC_BAD_PORT
;
2396 } else if (!eth_addr_equals(port
->pp
.hw_addr
, pm
.hw_addr
)) {
2397 return OFPERR_OFPPMFC_BAD_HW_ADDR
;
2399 update_port_config(port
, pm
.config
, pm
.mask
);
2401 netdev_set_advertisements(port
->netdev
, pm
.advertise
);
2408 handle_desc_stats_request(struct ofconn
*ofconn
,
2409 const struct ofp_header
*request
)
2411 static const char *default_mfr_desc
= "Nicira, Inc.";
2412 static const char *default_hw_desc
= "Open vSwitch";
2413 static const char *default_sw_desc
= VERSION
;
2414 static const char *default_serial_desc
= "None";
2415 static const char *default_dp_desc
= "None";
2417 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2418 struct ofp_desc_stats
*ods
;
2421 msg
= ofpraw_alloc_stats_reply(request
, 0);
2422 ods
= ofpbuf_put_zeros(msg
, sizeof *ods
);
2423 ovs_strlcpy(ods
->mfr_desc
, p
->mfr_desc
? p
->mfr_desc
: default_mfr_desc
,
2424 sizeof ods
->mfr_desc
);
2425 ovs_strlcpy(ods
->hw_desc
, p
->hw_desc
? p
->hw_desc
: default_hw_desc
,
2426 sizeof ods
->hw_desc
);
2427 ovs_strlcpy(ods
->sw_desc
, p
->sw_desc
? p
->sw_desc
: default_sw_desc
,
2428 sizeof ods
->sw_desc
);
2429 ovs_strlcpy(ods
->serial_num
,
2430 p
->serial_desc
? p
->serial_desc
: default_serial_desc
,
2431 sizeof ods
->serial_num
);
2432 ovs_strlcpy(ods
->dp_desc
, p
->dp_desc
? p
->dp_desc
: default_dp_desc
,
2433 sizeof ods
->dp_desc
);
2434 ofconn_send_reply(ofconn
, msg
);
2440 handle_table_stats_request(struct ofconn
*ofconn
,
2441 const struct ofp_header
*request
)
2443 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2444 struct ofp12_table_stats
*ots
;
2449 /* Set up default values.
2451 * ofp12_table_stats is used as a generic structure as
2452 * it is able to hold all the fields for ofp10_table_stats
2453 * and ofp11_table_stats (and of course itself).
2455 ots
= xcalloc(p
->n_tables
, sizeof *ots
);
2456 for (i
= 0; i
< p
->n_tables
; i
++) {
2457 ots
[i
].table_id
= i
;
2458 sprintf(ots
[i
].name
, "table%zu", i
);
2459 ots
[i
].match
= htonll(OFPXMT12_MASK
);
2460 ots
[i
].wildcards
= htonll(OFPXMT12_MASK
);
2461 ots
[i
].write_actions
= htonl(OFPAT11_OUTPUT
);
2462 ots
[i
].apply_actions
= htonl(OFPAT11_OUTPUT
);
2463 ots
[i
].write_setfields
= htonll(OFPXMT12_MASK
);
2464 ots
[i
].apply_setfields
= htonll(OFPXMT12_MASK
);
2465 ots
[i
].metadata_match
= htonll(UINT64_MAX
);
2466 ots
[i
].metadata_write
= htonll(UINT64_MAX
);
2467 ots
[i
].instructions
= htonl(OFPIT11_ALL
);
2468 ots
[i
].config
= htonl(OFPTC11_TABLE_MISS_MASK
);
2469 ots
[i
].max_entries
= htonl(1000000); /* An arbitrary big number. */
2470 ots
[i
].active_count
= htonl(classifier_count(&p
->tables
[i
].cls
));
2473 p
->ofproto_class
->get_tables(p
, ots
);
2475 /* Post-process the tables, dropping hidden tables. */
2476 n_tables
= p
->n_tables
;
2477 for (i
= 0; i
< p
->n_tables
; i
++) {
2478 const struct oftable
*table
= &p
->tables
[i
];
2480 if (table
->flags
& OFTABLE_HIDDEN
) {
2486 ovs_strzcpy(ots
[i
].name
, table
->name
, sizeof ots
[i
].name
);
2489 if (table
->max_flows
< ntohl(ots
[i
].max_entries
)) {
2490 ots
[i
].max_entries
= htonl(table
->max_flows
);
2494 msg
= ofputil_encode_table_stats_reply(ots
, n_tables
, request
);
2495 ofconn_send_reply(ofconn
, msg
);
2503 append_port_stat(struct ofport
*port
, struct list
*replies
)
2505 struct ofputil_port_stats ops
= { .port_no
= port
->pp
.port_no
};
2507 /* Intentionally ignore return value, since errors will set
2508 * 'stats' to all-1s, which is correct for OpenFlow, and
2509 * netdev_get_stats() will log errors. */
2510 ofproto_port_get_stats(port
, &ops
.stats
);
2512 ofputil_append_port_stat(replies
, &ops
);
2516 handle_port_stats_request(struct ofconn
*ofconn
,
2517 const struct ofp_header
*request
)
2519 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2520 struct ofport
*port
;
2521 struct list replies
;
2525 error
= ofputil_decode_port_stats_request(request
, &port_no
);
2530 ofpmp_init(&replies
, request
);
2531 if (port_no
!= OFPP_ANY
) {
2532 port
= ofproto_get_port(p
, port_no
);
2534 append_port_stat(port
, &replies
);
2537 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
2538 append_port_stat(port
, &replies
);
2542 ofconn_send_replies(ofconn
, &replies
);
2547 handle_port_desc_stats_request(struct ofconn
*ofconn
,
2548 const struct ofp_header
*request
)
2550 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2551 enum ofp_version version
;
2552 struct ofport
*port
;
2553 struct list replies
;
2555 ofpmp_init(&replies
, request
);
2557 version
= ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn
));
2558 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
2559 ofputil_append_port_desc_stats_reply(version
, &port
->pp
, &replies
);
2562 ofconn_send_replies(ofconn
, &replies
);
2567 calc_flow_duration__(long long int start
, long long int now
,
2568 uint32_t *sec
, uint32_t *nsec
)
2570 long long int msecs
= now
- start
;
2571 *sec
= msecs
/ 1000;
2572 *nsec
= (msecs
% 1000) * (1000 * 1000);
2575 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2576 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2578 check_table_id(const struct ofproto
*ofproto
, uint8_t table_id
)
2580 return (table_id
== 0xff || table_id
< ofproto
->n_tables
2582 : OFPERR_OFPBRC_BAD_TABLE_ID
);
2586 static struct oftable
*
2587 next_visible_table(const struct ofproto
*ofproto
, uint8_t table_id
)
2589 struct oftable
*table
;
2591 for (table
= &ofproto
->tables
[table_id
];
2592 table
< &ofproto
->tables
[ofproto
->n_tables
];
2594 if (!(table
->flags
& OFTABLE_HIDDEN
)) {
2602 static struct oftable
*
2603 first_matching_table(const struct ofproto
*ofproto
, uint8_t table_id
)
2605 if (table_id
== 0xff) {
2606 return next_visible_table(ofproto
, 0);
2607 } else if (table_id
< ofproto
->n_tables
) {
2608 return &ofproto
->tables
[table_id
];
2614 static struct oftable
*
2615 next_matching_table(const struct ofproto
*ofproto
,
2616 const struct oftable
*table
, uint8_t table_id
)
2618 return (table_id
== 0xff
2619 ? next_visible_table(ofproto
, (table
- ofproto
->tables
) + 1)
2623 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2625 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2626 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2628 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2629 * only once, for that table. (This can be used to access tables marked
2632 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2633 * entered at all. (Perhaps you should have validated TABLE_ID with
2634 * check_table_id().)
2636 * All parameters are evaluated multiple times.
2638 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2639 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2641 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2643 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2644 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2645 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2648 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2649 * to 'out_port' are included.
2651 * Hidden rules are always omitted.
2653 * Returns 0 on success, otherwise an OpenFlow error code. */
2655 collect_rules_loose(struct ofproto
*ofproto
, uint8_t table_id
,
2656 const struct match
*match
,
2657 ovs_be64 cookie
, ovs_be64 cookie_mask
,
2658 uint16_t out_port
, struct list
*rules
)
2660 struct oftable
*table
;
2664 error
= check_table_id(ofproto
, table_id
);
2670 cls_rule_init(&cr
, match
, 0);
2671 FOR_EACH_MATCHING_TABLE (table
, table_id
, ofproto
) {
2672 struct cls_cursor cursor
;
2675 cls_cursor_init(&cursor
, &table
->cls
, &cr
);
2676 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
2677 if (rule
->pending
) {
2678 error
= OFPROTO_POSTPONE
;
2681 if (!ofproto_rule_is_hidden(rule
)
2682 && ofproto_rule_has_out_port(rule
, out_port
)
2683 && !((rule
->flow_cookie
^ cookie
) & cookie_mask
)) {
2684 list_push_back(rules
, &rule
->ofproto_node
);
2690 cls_rule_destroy(&cr
);
2694 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2695 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2696 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2699 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2700 * to 'out_port' are included.
2702 * Hidden rules are always omitted.
2704 * Returns 0 on success, otherwise an OpenFlow error code. */
2706 collect_rules_strict(struct ofproto
*ofproto
, uint8_t table_id
,
2707 const struct match
*match
, unsigned int priority
,
2708 ovs_be64 cookie
, ovs_be64 cookie_mask
,
2709 uint16_t out_port
, struct list
*rules
)
2711 struct oftable
*table
;
2715 error
= check_table_id(ofproto
, table_id
);
2721 cls_rule_init(&cr
, match
, priority
);
2722 FOR_EACH_MATCHING_TABLE (table
, table_id
, ofproto
) {
2725 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&table
->cls
,
2728 if (rule
->pending
) {
2729 error
= OFPROTO_POSTPONE
;
2732 if (!ofproto_rule_is_hidden(rule
)
2733 && ofproto_rule_has_out_port(rule
, out_port
)
2734 && !((rule
->flow_cookie
^ cookie
) & cookie_mask
)) {
2735 list_push_back(rules
, &rule
->ofproto_node
);
2741 cls_rule_destroy(&cr
);
2745 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2746 * forced into the range of a uint16_t. */
2748 age_secs(long long int age_ms
)
2750 return (age_ms
< 0 ? 0
2751 : age_ms
>= UINT16_MAX
* 1000 ? UINT16_MAX
2752 : (unsigned int) age_ms
/ 1000);
2756 handle_flow_stats_request(struct ofconn
*ofconn
,
2757 const struct ofp_header
*request
)
2759 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2760 struct ofputil_flow_stats_request fsr
;
2761 struct list replies
;
2766 error
= ofputil_decode_flow_stats_request(&fsr
, request
);
2771 error
= collect_rules_loose(ofproto
, fsr
.table_id
, &fsr
.match
,
2772 fsr
.cookie
, fsr
.cookie_mask
,
2773 fsr
.out_port
, &rules
);
2778 ofpmp_init(&replies
, request
);
2779 LIST_FOR_EACH (rule
, ofproto_node
, &rules
) {
2780 long long int now
= time_msec();
2781 struct ofputil_flow_stats fs
;
2783 minimatch_expand(&rule
->cr
.match
, &fs
.match
);
2784 fs
.priority
= rule
->cr
.priority
;
2785 fs
.cookie
= rule
->flow_cookie
;
2786 fs
.table_id
= rule
->table_id
;
2787 calc_flow_duration__(rule
->created
, now
, &fs
.duration_sec
,
2789 fs
.idle_timeout
= rule
->idle_timeout
;
2790 fs
.hard_timeout
= rule
->hard_timeout
;
2791 fs
.idle_age
= age_secs(now
- rule
->used
);
2792 fs
.hard_age
= age_secs(now
- rule
->modified
);
2793 ofproto
->ofproto_class
->rule_get_stats(rule
, &fs
.packet_count
,
2795 fs
.ofpacts
= rule
->ofpacts
;
2796 fs
.ofpacts_len
= rule
->ofpacts_len
;
2798 if (rule
->send_flow_removed
) {
2799 fs
.flags
|= OFPFF_SEND_FLOW_REM
;
2800 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
2801 and OFPFF13_NO_BYT_COUNTS */
2803 ofputil_append_flow_stats_reply(&fs
, &replies
);
2805 ofconn_send_replies(ofconn
, &replies
);
2811 flow_stats_ds(struct rule
*rule
, struct ds
*results
)
2813 uint64_t packet_count
, byte_count
;
2815 rule
->ofproto
->ofproto_class
->rule_get_stats(rule
,
2816 &packet_count
, &byte_count
);
2818 if (rule
->table_id
!= 0) {
2819 ds_put_format(results
, "table_id=%"PRIu8
", ", rule
->table_id
);
2821 ds_put_format(results
, "duration=%llds, ",
2822 (time_msec() - rule
->created
) / 1000);
2823 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
2824 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
2825 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
2826 cls_rule_format(&rule
->cr
, results
);
2827 ds_put_char(results
, ',');
2828 ofpacts_format(rule
->ofpacts
, rule
->ofpacts_len
, results
);
2829 ds_put_cstr(results
, "\n");
2832 /* Adds a pretty-printed description of all flows to 'results', including
2833 * hidden flows (e.g., set up by in-band control). */
2835 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
2837 struct oftable
*table
;
2839 OFPROTO_FOR_EACH_TABLE (table
, p
) {
2840 struct cls_cursor cursor
;
2843 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
2844 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
2845 flow_stats_ds(rule
, results
);
2850 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2851 * '*engine_type' and '*engine_id', respectively. */
2853 ofproto_get_netflow_ids(const struct ofproto
*ofproto
,
2854 uint8_t *engine_type
, uint8_t *engine_id
)
2856 ofproto
->ofproto_class
->get_netflow_ids(ofproto
, engine_type
, engine_id
);
2859 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
2860 * true if the port's CFM status was successfully stored into '*status'.
2861 * Returns false if the port did not have CFM configured, in which case
2862 * '*status' is indeterminate.
2864 * The caller must provide and owns '*status', but it does not own and must not
2865 * modify or free the array returned in 'status->rmps'. */
2867 ofproto_port_get_cfm_status(const struct ofproto
*ofproto
, uint16_t ofp_port
,
2868 struct ofproto_cfm_status
*status
)
2870 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
2872 && ofproto
->ofproto_class
->get_cfm_status
2873 && ofproto
->ofproto_class
->get_cfm_status(ofport
, status
));
2877 handle_aggregate_stats_request(struct ofconn
*ofconn
,
2878 const struct ofp_header
*oh
)
2880 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2881 struct ofputil_flow_stats_request request
;
2882 struct ofputil_aggregate_stats stats
;
2883 bool unknown_packets
, unknown_bytes
;
2884 struct ofpbuf
*reply
;
2889 error
= ofputil_decode_flow_stats_request(&request
, oh
);
2894 error
= collect_rules_loose(ofproto
, request
.table_id
, &request
.match
,
2895 request
.cookie
, request
.cookie_mask
,
2896 request
.out_port
, &rules
);
2901 memset(&stats
, 0, sizeof stats
);
2902 unknown_packets
= unknown_bytes
= false;
2903 LIST_FOR_EACH (rule
, ofproto_node
, &rules
) {
2904 uint64_t packet_count
;
2905 uint64_t byte_count
;
2907 ofproto
->ofproto_class
->rule_get_stats(rule
, &packet_count
,
2910 if (packet_count
== UINT64_MAX
) {
2911 unknown_packets
= true;
2913 stats
.packet_count
+= packet_count
;
2916 if (byte_count
== UINT64_MAX
) {
2917 unknown_bytes
= true;
2919 stats
.byte_count
+= byte_count
;
2924 if (unknown_packets
) {
2925 stats
.packet_count
= UINT64_MAX
;
2927 if (unknown_bytes
) {
2928 stats
.byte_count
= UINT64_MAX
;
2931 reply
= ofputil_encode_aggregate_stats_reply(&stats
, oh
);
2932 ofconn_send_reply(ofconn
, reply
);
2937 struct queue_stats_cbdata
{
2938 struct ofport
*ofport
;
2939 struct list replies
;
2943 put_queue_stats(struct queue_stats_cbdata
*cbdata
, uint32_t queue_id
,
2944 const struct netdev_queue_stats
*stats
)
2947 struct ofputil_queue_stats oqs
= {
2948 .port_no
= cbdata
->ofport
->pp
.port_no
,
2949 .queue_id
= queue_id
,
2952 ofputil_append_queue_stat(&cbdata
->replies
, &oqs
);
2956 handle_queue_stats_dump_cb(uint32_t queue_id
,
2957 struct netdev_queue_stats
*stats
,
2960 struct queue_stats_cbdata
*cbdata
= cbdata_
;
2962 put_queue_stats(cbdata
, queue_id
, stats
);
2966 handle_queue_stats_for_port(struct ofport
*port
, uint32_t queue_id
,
2967 struct queue_stats_cbdata
*cbdata
)
2969 cbdata
->ofport
= port
;
2970 if (queue_id
== OFPQ_ALL
) {
2971 netdev_dump_queue_stats(port
->netdev
,
2972 handle_queue_stats_dump_cb
, cbdata
);
2974 struct netdev_queue_stats stats
;
2976 if (!netdev_get_queue_stats(port
->netdev
, queue_id
, &stats
)) {
2977 put_queue_stats(cbdata
, queue_id
, &stats
);
2979 return OFPERR_OFPQOFC_BAD_QUEUE
;
2986 handle_queue_stats_request(struct ofconn
*ofconn
,
2987 const struct ofp_header
*rq
)
2989 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2990 struct queue_stats_cbdata cbdata
;
2991 struct ofport
*port
;
2993 struct ofputil_queue_stats_request oqsr
;
2995 COVERAGE_INC(ofproto_queue_req
);
2997 ofpmp_init(&cbdata
.replies
, rq
);
2999 error
= ofputil_decode_queue_stats_request(rq
, &oqsr
);
3004 if (oqsr
.port_no
== OFPP_ANY
) {
3005 error
= OFPERR_OFPQOFC_BAD_QUEUE
;
3006 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
3007 if (!handle_queue_stats_for_port(port
, oqsr
.queue_id
, &cbdata
)) {
3012 port
= ofproto_get_port(ofproto
, oqsr
.port_no
);
3014 ? handle_queue_stats_for_port(port
, oqsr
.queue_id
, &cbdata
)
3015 : OFPERR_OFPQOFC_BAD_PORT
);
3018 ofconn_send_replies(ofconn
, &cbdata
.replies
);
3020 ofpbuf_list_delete(&cbdata
.replies
);
3027 is_flow_deletion_pending(const struct ofproto
*ofproto
,
3028 const struct cls_rule
*cls_rule
,
3031 if (!hmap_is_empty(&ofproto
->deletions
)) {
3032 struct ofoperation
*op
;
3034 HMAP_FOR_EACH_WITH_HASH (op
, hmap_node
,
3035 cls_rule_hash(cls_rule
, table_id
),
3036 &ofproto
->deletions
) {
3037 if (cls_rule_equal(cls_rule
, &op
->rule
->cr
)) {
3046 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3047 * in which no matching flow already exists in the flow table.
3049 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3050 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3051 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3052 * initiated now but may be retried later.
3054 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3055 * ownership remains with the caller.
3057 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3060 add_flow(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3061 const struct ofputil_flow_mod
*fm
, const struct ofp_header
*request
)
3063 struct oftable
*table
;
3064 struct ofopgroup
*group
;
3065 struct rule
*victim
;
3070 error
= check_table_id(ofproto
, fm
->table_id
);
3076 if (fm
->table_id
== 0xff) {
3078 if (ofproto
->ofproto_class
->rule_choose_table
) {
3079 error
= ofproto
->ofproto_class
->rule_choose_table(ofproto
,
3085 ovs_assert(table_id
< ofproto
->n_tables
);
3086 table
= &ofproto
->tables
[table_id
];
3088 table
= &ofproto
->tables
[0];
3090 } else if (fm
->table_id
< ofproto
->n_tables
) {
3091 table
= &ofproto
->tables
[fm
->table_id
];
3093 return OFPERR_OFPBRC_BAD_TABLE_ID
;
3096 if (table
->flags
& OFTABLE_READONLY
) {
3097 return OFPERR_OFPBRC_EPERM
;
3100 /* Allocate new rule and initialize classifier rule. */
3101 rule
= ofproto
->ofproto_class
->rule_alloc();
3103 VLOG_WARN_RL(&rl
, "%s: failed to create rule (%s)",
3104 ofproto
->name
, strerror(error
));
3107 cls_rule_init(&rule
->cr
, &fm
->match
, fm
->priority
);
3109 /* Serialize against pending deletion. */
3110 if (is_flow_deletion_pending(ofproto
, &cr
, table
- ofproto
->tables
)) {
3111 cls_rule_destroy(&rule
->cr
);
3112 ofproto
->ofproto_class
->rule_dealloc(rule
);
3113 return OFPROTO_POSTPONE
;
3116 /* Check for overlap, if requested. */
3117 if (fm
->flags
& OFPFF_CHECK_OVERLAP
3118 && classifier_rule_overlaps(&table
->cls
, &rule
->cr
)) {
3119 cls_rule_destroy(&rule
->cr
);
3120 ofproto
->ofproto_class
->rule_dealloc(rule
);
3121 return OFPERR_OFPFMFC_OVERLAP
;
3124 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3126 rule
->ofproto
= ofproto
;
3127 rule
->pending
= NULL
;
3128 rule
->flow_cookie
= fm
->new_cookie
;
3129 rule
->created
= rule
->modified
= rule
->used
= time_msec();
3130 rule
->idle_timeout
= fm
->idle_timeout
;
3131 rule
->hard_timeout
= fm
->hard_timeout
;
3132 rule
->table_id
= table
- ofproto
->tables
;
3133 rule
->send_flow_removed
= (fm
->flags
& OFPFF_SEND_FLOW_REM
) != 0;
3134 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3135 and OFPFF13_NO_BYT_COUNTS */
3136 rule
->ofpacts
= xmemdup(fm
->ofpacts
, fm
->ofpacts_len
);
3137 rule
->ofpacts_len
= fm
->ofpacts_len
;
3138 rule
->evictable
= true;
3139 rule
->eviction_group
= NULL
;
3140 list_init(&rule
->expirable
);
3141 rule
->monitor_flags
= 0;
3142 rule
->add_seqno
= 0;
3143 rule
->modify_seqno
= 0;
3145 /* Insert new rule. */
3146 victim
= oftable_replace_rule(rule
);
3147 if (victim
&& !rule_is_modifiable(victim
)) {
3148 error
= OFPERR_OFPBRC_EPERM
;
3149 } else if (victim
&& victim
->pending
) {
3150 error
= OFPROTO_POSTPONE
;
3152 struct ofoperation
*op
;
3155 if (classifier_count(&table
->cls
) > table
->max_flows
) {
3158 was_evictable
= rule
->evictable
;
3159 rule
->evictable
= false;
3160 evict
= choose_rule_to_evict(table
);
3161 rule
->evictable
= was_evictable
;
3164 error
= OFPERR_OFPFMFC_TABLE_FULL
;
3166 } else if (evict
->pending
) {
3167 error
= OFPROTO_POSTPONE
;
3174 group
= ofopgroup_create(ofproto
, ofconn
, request
, fm
->buffer_id
);
3175 op
= ofoperation_create(group
, rule
, OFOPERATION_ADD
, 0);
3176 op
->victim
= victim
;
3178 error
= ofproto
->ofproto_class
->rule_construct(rule
);
3180 op
->group
->n_running
--;
3181 ofoperation_destroy(rule
->pending
);
3183 delete_flow__(evict
, group
);
3185 ofopgroup_submit(group
);
3189 /* Back out if an error occurred. */
3191 oftable_substitute_rule(rule
, victim
);
3192 ofproto_rule_destroy__(rule
);
3197 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3199 /* Modifies the rules listed in 'rules', changing their actions to match those
3202 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3205 * Returns 0 on success, otherwise an OpenFlow error code. */
3207 modify_flows__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3208 const struct ofputil_flow_mod
*fm
,
3209 const struct ofp_header
*request
, struct list
*rules
)
3211 struct ofopgroup
*group
;
3215 group
= ofopgroup_create(ofproto
, ofconn
, request
, fm
->buffer_id
);
3216 error
= OFPERR_OFPBRC_EPERM
;
3217 LIST_FOR_EACH (rule
, ofproto_node
, rules
) {
3218 struct ofoperation
*op
;
3219 bool actions_changed
;
3220 ovs_be64 new_cookie
;
3222 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3224 if (rule_is_modifiable(rule
)) {
3225 /* At least one rule is modifiable, don't report EPERM error. */
3231 actions_changed
= !ofpacts_equal(fm
->ofpacts
, fm
->ofpacts_len
,
3232 rule
->ofpacts
, rule
->ofpacts_len
);
3233 new_cookie
= (fm
->new_cookie
!= htonll(UINT64_MAX
)
3235 : rule
->flow_cookie
);
3237 op
= ofoperation_create(group
, rule
, OFOPERATION_MODIFY
, 0);
3238 rule
->flow_cookie
= new_cookie
;
3239 if (actions_changed
) {
3240 op
->ofpacts
= rule
->ofpacts
;
3241 op
->ofpacts_len
= rule
->ofpacts_len
;
3242 rule
->ofpacts
= xmemdup(fm
->ofpacts
, fm
->ofpacts_len
);
3243 rule
->ofpacts_len
= fm
->ofpacts_len
;
3244 rule
->ofproto
->ofproto_class
->rule_modify_actions(rule
);
3246 ofoperation_complete(op
, 0);
3249 ofopgroup_submit(group
);
3255 modify_flows_add(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3256 const struct ofputil_flow_mod
*fm
,
3257 const struct ofp_header
*request
)
3259 if (fm
->cookie_mask
!= htonll(0) || fm
->new_cookie
== htonll(UINT64_MAX
)) {
3262 return add_flow(ofproto
, ofconn
, fm
, request
);
3265 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3268 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3271 modify_flows_loose(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3272 const struct ofputil_flow_mod
*fm
,
3273 const struct ofp_header
*request
)
3278 error
= collect_rules_loose(ofproto
, fm
->table_id
, &fm
->match
,
3279 fm
->cookie
, fm
->cookie_mask
,
3283 } else if (list_is_empty(&rules
)) {
3284 return modify_flows_add(ofproto
, ofconn
, fm
, request
);
3286 return modify_flows__(ofproto
, ofconn
, fm
, request
, &rules
);
3290 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3293 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3296 modify_flow_strict(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3297 const struct ofputil_flow_mod
*fm
,
3298 const struct ofp_header
*request
)
3303 error
= collect_rules_strict(ofproto
, fm
->table_id
, &fm
->match
,
3304 fm
->priority
, fm
->cookie
, fm
->cookie_mask
,
3309 } else if (list_is_empty(&rules
)) {
3310 return modify_flows_add(ofproto
, ofconn
, fm
, request
);
3312 return list_is_singleton(&rules
) ? modify_flows__(ofproto
, ofconn
,
3313 fm
, request
, &rules
)
3318 /* OFPFC_DELETE implementation. */
3321 delete_flow__(struct rule
*rule
, struct ofopgroup
*group
)
3323 struct ofproto
*ofproto
= rule
->ofproto
;
3325 ofproto_rule_send_removed(rule
, OFPRR_DELETE
);
3327 ofoperation_create(group
, rule
, OFOPERATION_DELETE
, OFPRR_DELETE
);
3328 oftable_remove_rule(rule
);
3329 ofproto
->ofproto_class
->rule_destruct(rule
);
3332 /* Deletes the rules listed in 'rules'.
3334 * Returns 0 on success, otherwise an OpenFlow error code. */
3336 delete_flows__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3337 const struct ofp_header
*request
, struct list
*rules
)
3339 struct rule
*rule
, *next
;
3340 struct ofopgroup
*group
;
3342 group
= ofopgroup_create(ofproto
, ofconn
, request
, UINT32_MAX
);
3343 LIST_FOR_EACH_SAFE (rule
, next
, ofproto_node
, rules
) {
3344 delete_flow__(rule
, group
);
3346 ofopgroup_submit(group
);
3351 /* Implements OFPFC_DELETE. */
3353 delete_flows_loose(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3354 const struct ofputil_flow_mod
*fm
,
3355 const struct ofp_header
*request
)
3360 error
= collect_rules_loose(ofproto
, fm
->table_id
, &fm
->match
,
3361 fm
->cookie
, fm
->cookie_mask
,
3362 fm
->out_port
, &rules
);
3363 return (error
? error
3364 : !list_is_empty(&rules
) ? delete_flows__(ofproto
, ofconn
, request
,
3369 /* Implements OFPFC_DELETE_STRICT. */
3371 delete_flow_strict(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3372 const struct ofputil_flow_mod
*fm
,
3373 const struct ofp_header
*request
)
3378 error
= collect_rules_strict(ofproto
, fm
->table_id
, &fm
->match
,
3379 fm
->priority
, fm
->cookie
, fm
->cookie_mask
,
3380 fm
->out_port
, &rules
);
3381 return (error
? error
3382 : list_is_singleton(&rules
) ? delete_flows__(ofproto
, ofconn
,
3388 ofproto_rule_send_removed(struct rule
*rule
, uint8_t reason
)
3390 struct ofputil_flow_removed fr
;
3392 if (ofproto_rule_is_hidden(rule
) || !rule
->send_flow_removed
) {
3396 minimatch_expand(&rule
->cr
.match
, &fr
.match
);
3397 fr
.priority
= rule
->cr
.priority
;
3398 fr
.cookie
= rule
->flow_cookie
;
3400 fr
.table_id
= rule
->table_id
;
3401 calc_flow_duration__(rule
->created
, time_msec(),
3402 &fr
.duration_sec
, &fr
.duration_nsec
);
3403 fr
.idle_timeout
= rule
->idle_timeout
;
3404 fr
.hard_timeout
= rule
->hard_timeout
;
3405 rule
->ofproto
->ofproto_class
->rule_get_stats(rule
, &fr
.packet_count
,
3408 connmgr_send_flow_removed(rule
->ofproto
->connmgr
, &fr
);
3412 ofproto_rule_update_used(struct rule
*rule
, long long int used
)
3414 if (used
> rule
->used
) {
3415 struct eviction_group
*evg
= rule
->eviction_group
;
3419 heap_change(&evg
->rules
, &rule
->evg_node
,
3420 rule_eviction_priority(rule
));
3425 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3426 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3429 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3432 * ofproto implementation ->run() functions should use this function to expire
3433 * OpenFlow flows. */
3435 ofproto_rule_expire(struct rule
*rule
, uint8_t reason
)
3437 struct ofproto
*ofproto
= rule
->ofproto
;
3438 struct ofopgroup
*group
;
3440 ovs_assert(reason
== OFPRR_HARD_TIMEOUT
|| reason
== OFPRR_IDLE_TIMEOUT
);
3442 ofproto_rule_send_removed(rule
, reason
);
3444 group
= ofopgroup_create_unattached(ofproto
);
3445 ofoperation_create(group
, rule
, OFOPERATION_DELETE
, reason
);
3446 oftable_remove_rule(rule
);
3447 ofproto
->ofproto_class
->rule_destruct(rule
);
3448 ofopgroup_submit(group
);
3452 handle_flow_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3454 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
3455 struct ofputil_flow_mod fm
;
3456 uint64_t ofpacts_stub
[1024 / 8];
3457 struct ofpbuf ofpacts
;
3461 error
= reject_slave_controller(ofconn
);
3466 ofpbuf_use_stub(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
3467 error
= ofputil_decode_flow_mod(&fm
, oh
, ofconn_get_protocol(ofconn
),
3470 error
= ofpacts_check(fm
.ofpacts
, fm
.ofpacts_len
,
3471 &fm
.match
.flow
, ofproto
->max_ports
);
3474 error
= handle_flow_mod__(ofproto
, ofconn
, &fm
, oh
);
3477 goto exit_free_ofpacts
;
3480 /* Record the operation for logging a summary report. */
3481 switch (fm
.command
) {
3487 case OFPFC_MODIFY_STRICT
:
3488 ofproto
->n_modify
++;
3492 case OFPFC_DELETE_STRICT
:
3493 ofproto
->n_delete
++;
3498 if (ofproto
->next_op_report
== LLONG_MAX
) {
3499 ofproto
->first_op
= now
;
3500 ofproto
->next_op_report
= MAX(now
+ 10 * 1000,
3501 ofproto
->op_backoff
);
3502 ofproto
->op_backoff
= ofproto
->next_op_report
+ 60 * 1000;
3504 ofproto
->last_op
= now
;
3507 ofpbuf_uninit(&ofpacts
);
3513 handle_flow_mod__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3514 const struct ofputil_flow_mod
*fm
,
3515 const struct ofp_header
*oh
)
3517 if (ofproto
->n_pending
>= 50) {
3518 ovs_assert(!list_is_empty(&ofproto
->pending
));
3519 return OFPROTO_POSTPONE
;
3522 switch (fm
->command
) {
3524 return add_flow(ofproto
, ofconn
, fm
, oh
);
3527 return modify_flows_loose(ofproto
, ofconn
, fm
, oh
);
3529 case OFPFC_MODIFY_STRICT
:
3530 return modify_flow_strict(ofproto
, ofconn
, fm
, oh
);
3533 return delete_flows_loose(ofproto
, ofconn
, fm
, oh
);
3535 case OFPFC_DELETE_STRICT
:
3536 return delete_flow_strict(ofproto
, ofconn
, fm
, oh
);
3539 if (fm
->command
> 0xff) {
3540 VLOG_WARN_RL(&rl
, "%s: flow_mod has explicit table_id but "
3541 "flow_mod_table_id extension is not enabled",
3544 return OFPERR_OFPFMFC_BAD_COMMAND
;
3549 handle_role_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3551 struct ofputil_role_request request
;
3552 struct ofputil_role_request reply
;
3556 error
= ofputil_decode_role_message(oh
, &request
);
3561 if (request
.role
!= OFPCR12_ROLE_NOCHANGE
) {
3562 if (ofconn_get_role(ofconn
) != request
.role
3563 && ofconn_has_pending_opgroups(ofconn
)) {
3564 return OFPROTO_POSTPONE
;
3567 if (request
.have_generation_id
3568 && !ofconn_set_master_election_id(ofconn
, request
.generation_id
)) {
3569 return OFPERR_OFPRRFC_STALE
;
3572 ofconn_set_role(ofconn
, request
.role
);
3575 reply
.role
= ofconn_get_role(ofconn
);
3576 reply
.have_generation_id
= ofconn_get_master_election_id(
3577 ofconn
, &reply
.generation_id
);
3578 buf
= ofputil_encode_role_reply(oh
, &reply
);
3579 ofconn_send_reply(ofconn
, buf
);
3585 handle_nxt_flow_mod_table_id(struct ofconn
*ofconn
,
3586 const struct ofp_header
*oh
)
3588 const struct nx_flow_mod_table_id
*msg
= ofpmsg_body(oh
);
3589 enum ofputil_protocol cur
, next
;
3591 cur
= ofconn_get_protocol(ofconn
);
3592 next
= ofputil_protocol_set_tid(cur
, msg
->set
!= 0);
3593 ofconn_set_protocol(ofconn
, next
);
3599 handle_nxt_set_flow_format(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3601 const struct nx_set_flow_format
*msg
= ofpmsg_body(oh
);
3602 enum ofputil_protocol cur
, next
;
3603 enum ofputil_protocol next_base
;
3605 next_base
= ofputil_nx_flow_format_to_protocol(ntohl(msg
->format
));
3607 return OFPERR_OFPBRC_EPERM
;
3610 cur
= ofconn_get_protocol(ofconn
);
3611 next
= ofputil_protocol_set_base(cur
, next_base
);
3612 if (cur
!= next
&& ofconn_has_pending_opgroups(ofconn
)) {
3613 /* Avoid sending async messages in surprising protocol. */
3614 return OFPROTO_POSTPONE
;
3617 ofconn_set_protocol(ofconn
, next
);
3622 handle_nxt_set_packet_in_format(struct ofconn
*ofconn
,
3623 const struct ofp_header
*oh
)
3625 const struct nx_set_packet_in_format
*msg
= ofpmsg_body(oh
);
3628 format
= ntohl(msg
->format
);
3629 if (format
!= NXPIF_OPENFLOW10
&& format
!= NXPIF_NXM
) {
3630 return OFPERR_OFPBRC_EPERM
;
3633 if (format
!= ofconn_get_packet_in_format(ofconn
)
3634 && ofconn_has_pending_opgroups(ofconn
)) {
3635 /* Avoid sending async message in surprsing packet in format. */
3636 return OFPROTO_POSTPONE
;
3639 ofconn_set_packet_in_format(ofconn
, format
);
3644 handle_nxt_set_async_config(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3646 const struct nx_async_config
*msg
= ofpmsg_body(oh
);
3647 uint32_t master
[OAM_N_TYPES
];
3648 uint32_t slave
[OAM_N_TYPES
];
3650 master
[OAM_PACKET_IN
] = ntohl(msg
->packet_in_mask
[0]);
3651 master
[OAM_PORT_STATUS
] = ntohl(msg
->port_status_mask
[0]);
3652 master
[OAM_FLOW_REMOVED
] = ntohl(msg
->flow_removed_mask
[0]);
3654 slave
[OAM_PACKET_IN
] = ntohl(msg
->packet_in_mask
[1]);
3655 slave
[OAM_PORT_STATUS
] = ntohl(msg
->port_status_mask
[1]);
3656 slave
[OAM_FLOW_REMOVED
] = ntohl(msg
->flow_removed_mask
[1]);
3658 ofconn_set_async_config(ofconn
, master
, slave
);
3659 if (ofconn_get_type(ofconn
) == OFCONN_SERVICE
&&
3660 !ofconn_get_miss_send_len(ofconn
)) {
3661 ofconn_set_miss_send_len(ofconn
, OFP_DEFAULT_MISS_SEND_LEN
);
3668 handle_nxt_set_controller_id(struct ofconn
*ofconn
,
3669 const struct ofp_header
*oh
)
3671 const struct nx_controller_id
*nci
= ofpmsg_body(oh
);
3673 if (!is_all_zeros(nci
->zero
, sizeof nci
->zero
)) {
3674 return OFPERR_NXBRC_MUST_BE_ZERO
;
3677 ofconn_set_controller_id(ofconn
, ntohs(nci
->controller_id
));
3682 handle_barrier_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3686 if (ofconn_has_pending_opgroups(ofconn
)) {
3687 return OFPROTO_POSTPONE
;
3690 buf
= ofpraw_alloc_reply((oh
->version
== OFP10_VERSION
3691 ? OFPRAW_OFPT10_BARRIER_REPLY
3692 : OFPRAW_OFPT11_BARRIER_REPLY
), oh
, 0);
3693 ofconn_send_reply(ofconn
, buf
);
3698 ofproto_compose_flow_refresh_update(const struct rule
*rule
,
3699 enum nx_flow_monitor_flags flags
,
3702 struct ofoperation
*op
= rule
->pending
;
3703 struct ofputil_flow_update fu
;
3706 if (op
&& op
->type
== OFOPERATION_ADD
&& !op
->victim
) {
3707 /* We'll report the final flow when the operation completes. Reporting
3708 * it now would cause a duplicate report later. */
3712 fu
.event
= (flags
& (NXFMF_INITIAL
| NXFMF_ADD
)
3713 ? NXFME_ADDED
: NXFME_MODIFIED
);
3715 fu
.idle_timeout
= rule
->idle_timeout
;
3716 fu
.hard_timeout
= rule
->hard_timeout
;
3717 fu
.table_id
= rule
->table_id
;
3718 fu
.cookie
= rule
->flow_cookie
;
3719 minimatch_expand(&rule
->cr
.match
, &match
);
3721 fu
.priority
= rule
->cr
.priority
;
3722 if (!(flags
& NXFMF_ACTIONS
)) {
3726 fu
.ofpacts
= rule
->ofpacts
;
3727 fu
.ofpacts_len
= rule
->ofpacts_len
;
3729 /* An operation is in progress. Use the previous version of the flow's
3730 * actions, so that when the operation commits we report the change. */
3732 case OFOPERATION_ADD
:
3733 /* We already verified that there was a victim. */
3734 fu
.ofpacts
= op
->victim
->ofpacts
;
3735 fu
.ofpacts_len
= op
->victim
->ofpacts_len
;
3738 case OFOPERATION_MODIFY
:
3740 fu
.ofpacts
= op
->ofpacts
;
3741 fu
.ofpacts_len
= op
->ofpacts_len
;
3743 fu
.ofpacts
= rule
->ofpacts
;
3744 fu
.ofpacts_len
= rule
->ofpacts_len
;
3748 case OFOPERATION_DELETE
:
3749 fu
.ofpacts
= rule
->ofpacts
;
3750 fu
.ofpacts_len
= rule
->ofpacts_len
;
3758 if (list_is_empty(msgs
)) {
3759 ofputil_start_flow_update(msgs
);
3761 ofputil_append_flow_update(&fu
, msgs
);
3765 ofmonitor_compose_refresh_updates(struct list
*rules
, struct list
*msgs
)
3769 LIST_FOR_EACH (rule
, ofproto_node
, rules
) {
3770 enum nx_flow_monitor_flags flags
= rule
->monitor_flags
;
3771 rule
->monitor_flags
= 0;
3773 ofproto_compose_flow_refresh_update(rule
, flags
, msgs
);
3778 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor
*m
,
3779 struct rule
*rule
, uint64_t seqno
,
3782 enum nx_flow_monitor_flags update
;
3784 if (ofproto_rule_is_hidden(rule
)) {
3789 ? ofoperation_has_out_port(rule
->pending
, m
->out_port
)
3790 : ofproto_rule_has_out_port(rule
, m
->out_port
))) {
3795 if (rule
->add_seqno
> seqno
) {
3796 update
= NXFMF_ADD
| NXFMF_MODIFY
;
3797 } else if (rule
->modify_seqno
> seqno
) {
3798 update
= NXFMF_MODIFY
;
3803 if (!(m
->flags
& update
)) {
3807 update
= NXFMF_INITIAL
;
3810 if (!rule
->monitor_flags
) {
3811 list_push_back(rules
, &rule
->ofproto_node
);
3813 rule
->monitor_flags
|= update
| (m
->flags
& NXFMF_ACTIONS
);
3817 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor
*m
,
3821 const struct ofproto
*ofproto
= ofconn_get_ofproto(m
->ofconn
);
3822 const struct ofoperation
*op
;
3823 const struct oftable
*table
;
3824 struct cls_rule target
;
3826 cls_rule_init_from_minimatch(&target
, &m
->match
, 0);
3827 FOR_EACH_MATCHING_TABLE (table
, m
->table_id
, ofproto
) {
3828 struct cls_cursor cursor
;
3831 cls_cursor_init(&cursor
, &table
->cls
, &target
);
3832 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3833 ovs_assert(!rule
->pending
); /* XXX */
3834 ofproto_collect_ofmonitor_refresh_rule(m
, rule
, seqno
, rules
);
3838 HMAP_FOR_EACH (op
, hmap_node
, &ofproto
->deletions
) {
3839 struct rule
*rule
= op
->rule
;
3841 if (((m
->table_id
== 0xff
3842 ? !(ofproto
->tables
[rule
->table_id
].flags
& OFTABLE_HIDDEN
)
3843 : m
->table_id
== rule
->table_id
))
3844 && cls_rule_is_loose_match(&rule
->cr
, &target
.match
)) {
3845 ofproto_collect_ofmonitor_refresh_rule(m
, rule
, seqno
, rules
);
3848 cls_rule_destroy(&target
);
3852 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor
*m
,
3855 if (m
->flags
& NXFMF_INITIAL
) {
3856 ofproto_collect_ofmonitor_refresh_rules(m
, 0, rules
);
3861 ofmonitor_collect_resume_rules(struct ofmonitor
*m
,
3862 uint64_t seqno
, struct list
*rules
)
3864 ofproto_collect_ofmonitor_refresh_rules(m
, seqno
, rules
);
3868 handle_flow_monitor_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3870 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
3871 struct ofmonitor
**monitors
;
3872 size_t n_monitors
, allocated_monitors
;
3873 struct list replies
;
3880 ofpbuf_use_const(&b
, oh
, ntohs(oh
->length
));
3882 n_monitors
= allocated_monitors
= 0;
3884 struct ofputil_flow_monitor_request request
;
3885 struct ofmonitor
*m
;
3888 retval
= ofputil_decode_flow_monitor_request(&request
, &b
);
3889 if (retval
== EOF
) {
3891 } else if (retval
) {
3896 if (request
.table_id
!= 0xff
3897 && request
.table_id
>= ofproto
->n_tables
) {
3898 error
= OFPERR_OFPBRC_BAD_TABLE_ID
;
3902 error
= ofmonitor_create(&request
, ofconn
, &m
);
3907 if (n_monitors
>= allocated_monitors
) {
3908 monitors
= x2nrealloc(monitors
, &allocated_monitors
,
3911 monitors
[n_monitors
++] = m
;
3915 for (i
= 0; i
< n_monitors
; i
++) {
3916 ofproto_collect_ofmonitor_initial_rules(monitors
[i
], &rules
);
3919 ofpmp_init(&replies
, oh
);
3920 ofmonitor_compose_refresh_updates(&rules
, &replies
);
3921 ofconn_send_replies(ofconn
, &replies
);
3928 for (i
= 0; i
< n_monitors
; i
++) {
3929 ofmonitor_destroy(monitors
[i
]);
3936 handle_flow_monitor_cancel(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3938 struct ofmonitor
*m
;
3941 id
= ofputil_decode_flow_monitor_cancel(oh
);
3942 m
= ofmonitor_lookup(ofconn
, id
);
3944 return OFPERR_NXBRC_FM_BAD_ID
;
3947 ofmonitor_destroy(m
);
3952 handle_openflow__(struct ofconn
*ofconn
, const struct ofpbuf
*msg
)
3954 const struct ofp_header
*oh
= msg
->data
;
3958 error
= ofptype_decode(&type
, oh
);
3964 /* OpenFlow requests. */
3965 case OFPTYPE_ECHO_REQUEST
:
3966 return handle_echo_request(ofconn
, oh
);
3968 case OFPTYPE_FEATURES_REQUEST
:
3969 return handle_features_request(ofconn
, oh
);
3971 case OFPTYPE_GET_CONFIG_REQUEST
:
3972 return handle_get_config_request(ofconn
, oh
);
3974 case OFPTYPE_SET_CONFIG
:
3975 return handle_set_config(ofconn
, oh
);
3977 case OFPTYPE_PACKET_OUT
:
3978 return handle_packet_out(ofconn
, oh
);
3980 case OFPTYPE_PORT_MOD
:
3981 return handle_port_mod(ofconn
, oh
);
3983 case OFPTYPE_FLOW_MOD
:
3984 return handle_flow_mod(ofconn
, oh
);
3986 case OFPTYPE_BARRIER_REQUEST
:
3987 return handle_barrier_request(ofconn
, oh
);
3989 case OFPTYPE_ROLE_REQUEST
:
3990 return handle_role_request(ofconn
, oh
);
3992 /* OpenFlow replies. */
3993 case OFPTYPE_ECHO_REPLY
:
3996 /* Nicira extension requests. */
3997 case OFPTYPE_FLOW_MOD_TABLE_ID
:
3998 return handle_nxt_flow_mod_table_id(ofconn
, oh
);
4000 case OFPTYPE_SET_FLOW_FORMAT
:
4001 return handle_nxt_set_flow_format(ofconn
, oh
);
4003 case OFPTYPE_SET_PACKET_IN_FORMAT
:
4004 return handle_nxt_set_packet_in_format(ofconn
, oh
);
4006 case OFPTYPE_SET_CONTROLLER_ID
:
4007 return handle_nxt_set_controller_id(ofconn
, oh
);
4009 case OFPTYPE_FLOW_AGE
:
4010 /* Nothing to do. */
4013 case OFPTYPE_FLOW_MONITOR_CANCEL
:
4014 return handle_flow_monitor_cancel(ofconn
, oh
);
4016 case OFPTYPE_SET_ASYNC_CONFIG
:
4017 return handle_nxt_set_async_config(ofconn
, oh
);
4019 /* Statistics requests. */
4020 case OFPTYPE_DESC_STATS_REQUEST
:
4021 return handle_desc_stats_request(ofconn
, oh
);
4023 case OFPTYPE_FLOW_STATS_REQUEST
:
4024 return handle_flow_stats_request(ofconn
, oh
);
4026 case OFPTYPE_AGGREGATE_STATS_REQUEST
:
4027 return handle_aggregate_stats_request(ofconn
, oh
);
4029 case OFPTYPE_TABLE_STATS_REQUEST
:
4030 return handle_table_stats_request(ofconn
, oh
);
4032 case OFPTYPE_PORT_STATS_REQUEST
:
4033 return handle_port_stats_request(ofconn
, oh
);
4035 case OFPTYPE_QUEUE_STATS_REQUEST
:
4036 return handle_queue_stats_request(ofconn
, oh
);
4038 case OFPTYPE_PORT_DESC_STATS_REQUEST
:
4039 return handle_port_desc_stats_request(ofconn
, oh
);
4041 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST
:
4042 return handle_flow_monitor_request(ofconn
, oh
);
4044 /* FIXME: Change the following once they are implemented: */
4045 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST
:
4046 case OFPTYPE_GET_ASYNC_REQUEST
:
4047 case OFPTYPE_METER_MOD
:
4048 case OFPTYPE_GROUP_REQUEST
:
4049 case OFPTYPE_GROUP_DESC_REQUEST
:
4050 case OFPTYPE_GROUP_FEATURES_REQUEST
:
4051 case OFPTYPE_METER_REQUEST
:
4052 case OFPTYPE_METER_CONFIG_REQUEST
:
4053 case OFPTYPE_METER_FEATURES_REQUEST
:
4054 case OFPTYPE_TABLE_FEATURES_REQUEST
:
4055 return OFPERR_OFPBRC_BAD_TYPE
;
4059 case OFPTYPE_FEATURES_REPLY
:
4060 case OFPTYPE_GET_CONFIG_REPLY
:
4061 case OFPTYPE_PACKET_IN
:
4062 case OFPTYPE_FLOW_REMOVED
:
4063 case OFPTYPE_PORT_STATUS
:
4064 case OFPTYPE_BARRIER_REPLY
:
4065 case OFPTYPE_QUEUE_GET_CONFIG_REPLY
:
4066 case OFPTYPE_DESC_STATS_REPLY
:
4067 case OFPTYPE_FLOW_STATS_REPLY
:
4068 case OFPTYPE_QUEUE_STATS_REPLY
:
4069 case OFPTYPE_PORT_STATS_REPLY
:
4070 case OFPTYPE_TABLE_STATS_REPLY
:
4071 case OFPTYPE_AGGREGATE_STATS_REPLY
:
4072 case OFPTYPE_PORT_DESC_STATS_REPLY
:
4073 case OFPTYPE_ROLE_REPLY
:
4074 case OFPTYPE_FLOW_MONITOR_PAUSED
:
4075 case OFPTYPE_FLOW_MONITOR_RESUMED
:
4076 case OFPTYPE_FLOW_MONITOR_STATS_REPLY
:
4077 case OFPTYPE_GET_ASYNC_REPLY
:
4078 case OFPTYPE_GROUP_REPLY
:
4079 case OFPTYPE_GROUP_DESC_REPLY
:
4080 case OFPTYPE_GROUP_FEATURES_REPLY
:
4081 case OFPTYPE_METER_REPLY
:
4082 case OFPTYPE_METER_CONFIG_REPLY
:
4083 case OFPTYPE_METER_FEATURES_REPLY
:
4084 case OFPTYPE_TABLE_FEATURES_REPLY
:
4086 return OFPERR_OFPBRC_BAD_TYPE
;
4091 handle_openflow(struct ofconn
*ofconn
, struct ofpbuf
*ofp_msg
)
4093 int error
= handle_openflow__(ofconn
, ofp_msg
);
4094 if (error
&& error
!= OFPROTO_POSTPONE
) {
4095 ofconn_send_error(ofconn
, ofp_msg
->data
, error
);
4097 COVERAGE_INC(ofproto_recv_openflow
);
4098 return error
!= OFPROTO_POSTPONE
;
4101 /* Asynchronous operations. */
4103 /* Creates and returns a new ofopgroup that is not associated with any
4104 * OpenFlow connection.
4106 * The caller should add operations to the returned group with
4107 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4108 static struct ofopgroup
*
4109 ofopgroup_create_unattached(struct ofproto
*ofproto
)
4111 struct ofopgroup
*group
= xzalloc(sizeof *group
);
4112 group
->ofproto
= ofproto
;
4113 list_init(&group
->ofproto_node
);
4114 list_init(&group
->ops
);
4115 list_init(&group
->ofconn_node
);
4119 /* Creates and returns a new ofopgroup for 'ofproto'.
4121 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
4122 * connection. The 'request' and 'buffer_id' arguments are ignored.
4124 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
4125 * If the ofopgroup eventually fails, then the error reply will include
4126 * 'request'. If the ofopgroup eventually succeeds, then the packet with
4127 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
4129 * The caller should add operations to the returned group with
4130 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4131 static struct ofopgroup
*
4132 ofopgroup_create(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
4133 const struct ofp_header
*request
, uint32_t buffer_id
)
4135 struct ofopgroup
*group
= ofopgroup_create_unattached(ofproto
);
4137 size_t request_len
= ntohs(request
->length
);
4139 ovs_assert(ofconn_get_ofproto(ofconn
) == ofproto
);
4141 ofconn_add_opgroup(ofconn
, &group
->ofconn_node
);
4142 group
->ofconn
= ofconn
;
4143 group
->request
= xmemdup(request
, MIN(request_len
, 64));
4144 group
->buffer_id
= buffer_id
;
4149 /* Submits 'group' for processing.
4151 * If 'group' contains no operations (e.g. none were ever added, or all of the
4152 * ones that were added completed synchronously), then it is destroyed
4153 * immediately. Otherwise it is added to the ofproto's list of pending
4156 ofopgroup_submit(struct ofopgroup
*group
)
4158 if (!group
->n_running
) {
4159 ofopgroup_complete(group
);
4161 list_push_back(&group
->ofproto
->pending
, &group
->ofproto_node
);
4162 group
->ofproto
->n_pending
++;
4167 ofopgroup_complete(struct ofopgroup
*group
)
4169 struct ofproto
*ofproto
= group
->ofproto
;
4171 struct ofconn
*abbrev_ofconn
;
4172 ovs_be32 abbrev_xid
;
4174 struct ofoperation
*op
, *next_op
;
4177 ovs_assert(!group
->n_running
);
4180 LIST_FOR_EACH (op
, group_node
, &group
->ops
) {
4187 if (!error
&& group
->ofconn
&& group
->buffer_id
!= UINT32_MAX
) {
4188 LIST_FOR_EACH (op
, group_node
, &group
->ops
) {
4189 if (op
->type
!= OFOPERATION_DELETE
) {
4190 struct ofpbuf
*packet
;
4193 error
= ofconn_pktbuf_retrieve(group
->ofconn
, group
->buffer_id
,
4197 error
= rule_execute(op
->rule
, in_port
, packet
);
4204 if (!error
&& !list_is_empty(&group
->ofconn_node
)) {
4205 abbrev_ofconn
= group
->ofconn
;
4206 abbrev_xid
= group
->request
->xid
;
4208 abbrev_ofconn
= NULL
;
4209 abbrev_xid
= htonl(0);
4211 LIST_FOR_EACH_SAFE (op
, next_op
, group_node
, &group
->ops
) {
4212 struct rule
*rule
= op
->rule
;
4214 /* We generally want to report the change to active OpenFlow flow
4215 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
4217 - The operation failed.
4219 - The affected rule is not visible to controllers.
4221 - The operation's only effect was to update rule->modified. */
4223 || ofproto_rule_is_hidden(rule
)
4224 || (op
->type
== OFOPERATION_MODIFY
4226 && rule
->flow_cookie
== op
->flow_cookie
))) {
4227 /* Check that we can just cast from ofoperation_type to
4228 * nx_flow_update_event. */
4229 BUILD_ASSERT_DECL((enum nx_flow_update_event
) OFOPERATION_ADD
4231 BUILD_ASSERT_DECL((enum nx_flow_update_event
) OFOPERATION_DELETE
4233 BUILD_ASSERT_DECL((enum nx_flow_update_event
) OFOPERATION_MODIFY
4236 ofmonitor_report(ofproto
->connmgr
, rule
,
4237 (enum nx_flow_update_event
) op
->type
,
4238 op
->reason
, abbrev_ofconn
, abbrev_xid
);
4241 rule
->pending
= NULL
;
4244 case OFOPERATION_ADD
:
4248 ofproto_rule_destroy__(op
->victim
);
4249 vid_mask
= minimask_get_vid_mask(&rule
->cr
.match
.mask
);
4250 if (vid_mask
== VLAN_VID_MASK
) {
4251 if (ofproto
->vlan_bitmap
) {
4252 uint16_t vid
= miniflow_get_vid(&rule
->cr
.match
.flow
);
4253 if (!bitmap_is_set(ofproto
->vlan_bitmap
, vid
)) {
4254 bitmap_set1(ofproto
->vlan_bitmap
, vid
);
4255 ofproto
->vlans_changed
= true;
4258 ofproto
->vlans_changed
= true;
4262 oftable_substitute_rule(rule
, op
->victim
);
4263 ofproto_rule_destroy__(rule
);
4267 case OFOPERATION_DELETE
:
4268 ovs_assert(!op
->error
);
4269 ofproto_rule_destroy__(rule
);
4273 case OFOPERATION_MODIFY
:
4275 rule
->modified
= time_msec();
4277 rule
->flow_cookie
= op
->flow_cookie
;
4279 free(rule
->ofpacts
);
4280 rule
->ofpacts
= op
->ofpacts
;
4281 rule
->ofpacts_len
= op
->ofpacts_len
;
4283 op
->ofpacts_len
= 0;
4292 ofoperation_destroy(op
);
4295 ofmonitor_flush(ofproto
->connmgr
);
4297 if (!list_is_empty(&group
->ofproto_node
)) {
4298 ovs_assert(ofproto
->n_pending
> 0);
4299 ofproto
->n_pending
--;
4300 list_remove(&group
->ofproto_node
);
4302 if (!list_is_empty(&group
->ofconn_node
)) {
4303 list_remove(&group
->ofconn_node
);
4305 ofconn_send_error(group
->ofconn
, group
->request
, error
);
4307 connmgr_retry(ofproto
->connmgr
);
4309 free(group
->request
);
4313 /* Initiates a new operation on 'rule', of the specified 'type', within
4314 * 'group'. Prior to calling, 'rule' must not have any pending operation.
4316 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
4317 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
4319 * Returns the newly created ofoperation (which is also available as
4320 * rule->pending). */
4321 static struct ofoperation
*
4322 ofoperation_create(struct ofopgroup
*group
, struct rule
*rule
,
4323 enum ofoperation_type type
,
4324 enum ofp_flow_removed_reason reason
)
4326 struct ofproto
*ofproto
= group
->ofproto
;
4327 struct ofoperation
*op
;
4329 ovs_assert(!rule
->pending
);
4331 op
= rule
->pending
= xzalloc(sizeof *op
);
4333 list_push_back(&group
->ops
, &op
->group_node
);
4336 op
->reason
= reason
;
4337 op
->flow_cookie
= rule
->flow_cookie
;
4341 if (type
== OFOPERATION_DELETE
) {
4342 hmap_insert(&ofproto
->deletions
, &op
->hmap_node
,
4343 cls_rule_hash(&rule
->cr
, rule
->table_id
));
4350 ofoperation_destroy(struct ofoperation
*op
)
4352 struct ofopgroup
*group
= op
->group
;
4355 op
->rule
->pending
= NULL
;
4357 if (op
->type
== OFOPERATION_DELETE
) {
4358 hmap_remove(&group
->ofproto
->deletions
, &op
->hmap_node
);
4360 list_remove(&op
->group_node
);
4365 /* Indicates that 'op' completed with status 'error', which is either 0 to
4366 * indicate success or an OpenFlow error code on failure.
4368 * If 'error' is 0, indicating success, the operation will be committed
4369 * permanently to the flow table. There is one interesting subcase:
4371 * - If 'op' is an "add flow" operation that is replacing an existing rule in
4372 * the flow table (the "victim" rule) by a new one, then the caller must
4373 * have uninitialized any derived state in the victim rule, as in step 5 in
4374 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
4375 * performs steps 6 and 7 for the victim rule, most notably by calling its
4376 * ->rule_dealloc() function.
4378 * If 'error' is nonzero, then generally the operation will be rolled back:
4380 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
4381 * restores the original rule. The caller must have uninitialized any
4382 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
4383 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
4384 * and 7 for the new rule, calling its ->rule_dealloc() function.
4386 * - If 'op' is a "modify flow" operation, ofproto restores the original
4389 * - 'op' must not be a "delete flow" operation. Removing a rule is not
4390 * allowed to fail. It must always succeed.
4392 * Please see the large comment in ofproto/ofproto-provider.h titled
4393 * "Asynchronous Operation Support" for more information. */
4395 ofoperation_complete(struct ofoperation
*op
, enum ofperr error
)
4397 struct ofopgroup
*group
= op
->group
;
4399 ovs_assert(op
->rule
->pending
== op
);
4400 ovs_assert(group
->n_running
> 0);
4401 ovs_assert(!error
|| op
->type
!= OFOPERATION_DELETE
);
4404 if (!--group
->n_running
&& !list_is_empty(&group
->ofproto_node
)) {
4405 ofopgroup_complete(group
);
4410 ofoperation_get_victim(struct ofoperation
*op
)
4412 ovs_assert(op
->type
== OFOPERATION_ADD
);
4417 pick_datapath_id(const struct ofproto
*ofproto
)
4419 const struct ofport
*port
;
4421 port
= ofproto_get_port(ofproto
, OFPP_LOCAL
);
4423 uint8_t ea
[ETH_ADDR_LEN
];
4426 error
= netdev_get_etheraddr(port
->netdev
, ea
);
4428 return eth_addr_to_uint64(ea
);
4430 VLOG_WARN("%s: could not get MAC address for %s (%s)",
4431 ofproto
->name
, netdev_get_name(port
->netdev
),
4434 return ofproto
->fallback_dpid
;
4438 pick_fallback_dpid(void)
4440 uint8_t ea
[ETH_ADDR_LEN
];
4441 eth_addr_nicira_random(ea
);
4442 return eth_addr_to_uint64(ea
);
4445 /* Table overflow policy. */
4447 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
4448 * is not configured to evict rules or if the table contains no evictable
4449 * rules. (Rules with 'evictable' set to false or with no timeouts are not
4451 static struct rule
*
4452 choose_rule_to_evict(struct oftable
*table
)
4454 struct eviction_group
*evg
;
4456 if (!table
->eviction_fields
) {
4460 /* In the common case, the outer and inner loops here will each be entered
4463 * - The inner loop normally "return"s in its first iteration. If the
4464 * eviction group has any evictable rules, then it always returns in
4467 * - The outer loop only iterates more than once if the largest eviction
4468 * group has no evictable rules.
4470 * - The outer loop can exit only if table's 'max_flows' is all filled up
4471 * by unevictable rules'. */
4472 HEAP_FOR_EACH (evg
, size_node
, &table
->eviction_groups_by_size
) {
4475 HEAP_FOR_EACH (rule
, evg_node
, &evg
->rules
) {
4476 if (rule
->evictable
) {
4485 /* Searches 'ofproto' for tables that have more flows than their configured
4486 * maximum and that have flow eviction enabled, and evicts as many flows as
4487 * necessary and currently feasible from them.
4489 * This triggers only when an OpenFlow table has N flows in it and then the
4490 * client configures a maximum number of flows less than N. */
4492 ofproto_evict(struct ofproto
*ofproto
)
4494 struct ofopgroup
*group
;
4495 struct oftable
*table
;
4497 group
= ofopgroup_create_unattached(ofproto
);
4498 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
4499 while (classifier_count(&table
->cls
) > table
->max_flows
4500 && table
->eviction_fields
) {
4503 rule
= choose_rule_to_evict(table
);
4504 if (!rule
|| rule
->pending
) {
4508 ofoperation_create(group
, rule
,
4509 OFOPERATION_DELETE
, OFPRR_EVICTION
);
4510 oftable_remove_rule(rule
);
4511 ofproto
->ofproto_class
->rule_destruct(rule
);
4514 ofopgroup_submit(group
);
4517 /* Eviction groups. */
4519 /* Returns the priority to use for an eviction_group that contains 'n_rules'
4520 * rules. The priority contains low-order random bits to ensure that eviction
4521 * groups with the same number of rules are prioritized randomly. */
4523 eviction_group_priority(size_t n_rules
)
4525 uint16_t size
= MIN(UINT16_MAX
, n_rules
);
4526 return (size
<< 16) | random_uint16();
4529 /* Updates 'evg', an eviction_group within 'table', following a change that
4530 * adds or removes rules in 'evg'. */
4532 eviction_group_resized(struct oftable
*table
, struct eviction_group
*evg
)
4534 heap_change(&table
->eviction_groups_by_size
, &evg
->size_node
,
4535 eviction_group_priority(heap_count(&evg
->rules
)));
4538 /* Destroys 'evg', an eviction_group within 'table':
4540 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
4541 * rules themselves, just removes them from the eviction group.)
4543 * - Removes 'evg' from 'table'.
4547 eviction_group_destroy(struct oftable
*table
, struct eviction_group
*evg
)
4549 while (!heap_is_empty(&evg
->rules
)) {
4552 rule
= CONTAINER_OF(heap_pop(&evg
->rules
), struct rule
, evg_node
);
4553 rule
->eviction_group
= NULL
;
4555 hmap_remove(&table
->eviction_groups_by_id
, &evg
->id_node
);
4556 heap_remove(&table
->eviction_groups_by_size
, &evg
->size_node
);
4557 heap_destroy(&evg
->rules
);
4561 /* Removes 'rule' from its eviction group, if any. */
4563 eviction_group_remove_rule(struct rule
*rule
)
4565 if (rule
->eviction_group
) {
4566 struct oftable
*table
= &rule
->ofproto
->tables
[rule
->table_id
];
4567 struct eviction_group
*evg
= rule
->eviction_group
;
4569 rule
->eviction_group
= NULL
;
4570 heap_remove(&evg
->rules
, &rule
->evg_node
);
4571 if (heap_is_empty(&evg
->rules
)) {
4572 eviction_group_destroy(table
, evg
);
4574 eviction_group_resized(table
, evg
);
4579 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
4580 * returns the hash value. */
4582 eviction_group_hash_rule(struct rule
*rule
)
4584 struct oftable
*table
= &rule
->ofproto
->tables
[rule
->table_id
];
4585 const struct mf_subfield
*sf
;
4589 hash
= table
->eviction_group_id_basis
;
4590 miniflow_expand(&rule
->cr
.match
.flow
, &flow
);
4591 for (sf
= table
->eviction_fields
;
4592 sf
< &table
->eviction_fields
[table
->n_eviction_fields
];
4595 if (mf_are_prereqs_ok(sf
->field
, &flow
)) {
4596 union mf_value value
;
4598 mf_get_value(sf
->field
, &flow
, &value
);
4600 bitwise_zero(&value
, sf
->field
->n_bytes
, 0, sf
->ofs
);
4602 if (sf
->ofs
+ sf
->n_bits
< sf
->field
->n_bytes
* 8) {
4603 unsigned int start
= sf
->ofs
+ sf
->n_bits
;
4604 bitwise_zero(&value
, sf
->field
->n_bytes
, start
,
4605 sf
->field
->n_bytes
* 8 - start
);
4607 hash
= hash_bytes(&value
, sf
->field
->n_bytes
, hash
);
4609 hash
= hash_int(hash
, 0);
4616 /* Returns an eviction group within 'table' with the given 'id', creating one
4618 static struct eviction_group
*
4619 eviction_group_find(struct oftable
*table
, uint32_t id
)
4621 struct eviction_group
*evg
;
4623 HMAP_FOR_EACH_WITH_HASH (evg
, id_node
, id
, &table
->eviction_groups_by_id
) {
4627 evg
= xmalloc(sizeof *evg
);
4628 hmap_insert(&table
->eviction_groups_by_id
, &evg
->id_node
, id
);
4629 heap_insert(&table
->eviction_groups_by_size
, &evg
->size_node
,
4630 eviction_group_priority(0));
4631 heap_init(&evg
->rules
);
4636 /* Returns an eviction priority for 'rule'. The return value should be
4637 * interpreted so that higher priorities make a rule more attractive candidates
4640 rule_eviction_priority(struct rule
*rule
)
4642 long long int hard_expiration
;
4643 long long int idle_expiration
;
4644 long long int expiration
;
4645 uint32_t expiration_offset
;
4647 /* Calculate time of expiration. */
4648 hard_expiration
= (rule
->hard_timeout
4649 ? rule
->modified
+ rule
->hard_timeout
* 1000
4651 idle_expiration
= (rule
->idle_timeout
4652 ? rule
->used
+ rule
->idle_timeout
* 1000
4654 expiration
= MIN(hard_expiration
, idle_expiration
);
4655 if (expiration
== LLONG_MAX
) {
4659 /* Calculate the time of expiration as a number of (approximate) seconds
4660 * after program startup.
4662 * This should work OK for program runs that last UINT32_MAX seconds or
4663 * less. Therefore, please restart OVS at least once every 136 years. */
4664 expiration_offset
= (expiration
>> 10) - (time_boot_msec() >> 10);
4666 /* Invert the expiration offset because we're using a max-heap. */
4667 return UINT32_MAX
- expiration_offset
;
4670 /* Adds 'rule' to an appropriate eviction group for its oftable's
4671 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
4672 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4675 * The caller must ensure that 'rule' is not already in an eviction group. */
4677 eviction_group_add_rule(struct rule
*rule
)
4679 struct ofproto
*ofproto
= rule
->ofproto
;
4680 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
4682 if (table
->eviction_fields
4683 && (rule
->hard_timeout
|| rule
->idle_timeout
)) {
4684 struct eviction_group
*evg
;
4686 evg
= eviction_group_find(table
, eviction_group_hash_rule(rule
));
4688 rule
->eviction_group
= evg
;
4689 heap_insert(&evg
->rules
, &rule
->evg_node
,
4690 rule_eviction_priority(rule
));
4691 eviction_group_resized(table
, evg
);
4697 /* Initializes 'table'. */
4699 oftable_init(struct oftable
*table
)
4701 memset(table
, 0, sizeof *table
);
4702 classifier_init(&table
->cls
);
4703 table
->max_flows
= UINT_MAX
;
4706 /* Destroys 'table', including its classifier and eviction groups.
4708 * The caller is responsible for freeing 'table' itself. */
4710 oftable_destroy(struct oftable
*table
)
4712 ovs_assert(classifier_is_empty(&table
->cls
));
4713 oftable_disable_eviction(table
);
4714 classifier_destroy(&table
->cls
);
4718 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4719 * string, then 'table' will use its default name.
4721 * This only affects the name exposed for a table exposed through the OpenFlow
4722 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4724 oftable_set_name(struct oftable
*table
, const char *name
)
4726 if (name
&& name
[0]) {
4727 int len
= strnlen(name
, OFP_MAX_TABLE_NAME_LEN
);
4728 if (!table
->name
|| strncmp(name
, table
->name
, len
)) {
4730 table
->name
= xmemdup0(name
, len
);
4738 /* oftables support a choice of two policies when adding a rule would cause the
4739 * number of flows in the table to exceed the configured maximum number: either
4740 * they can refuse to add the new flow or they can evict some existing flow.
4741 * This function configures the former policy on 'table'. */
4743 oftable_disable_eviction(struct oftable
*table
)
4745 if (table
->eviction_fields
) {
4746 struct eviction_group
*evg
, *next
;
4748 HMAP_FOR_EACH_SAFE (evg
, next
, id_node
,
4749 &table
->eviction_groups_by_id
) {
4750 eviction_group_destroy(table
, evg
);
4752 hmap_destroy(&table
->eviction_groups_by_id
);
4753 heap_destroy(&table
->eviction_groups_by_size
);
4755 free(table
->eviction_fields
);
4756 table
->eviction_fields
= NULL
;
4757 table
->n_eviction_fields
= 0;
4761 /* oftables support a choice of two policies when adding a rule would cause the
4762 * number of flows in the table to exceed the configured maximum number: either
4763 * they can refuse to add the new flow or they can evict some existing flow.
4764 * This function configures the latter policy on 'table', with fairness based
4765 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4766 * 'n_fields' as 0 disables fairness.) */
4768 oftable_enable_eviction(struct oftable
*table
,
4769 const struct mf_subfield
*fields
, size_t n_fields
)
4771 struct cls_cursor cursor
;
4774 if (table
->eviction_fields
4775 && n_fields
== table
->n_eviction_fields
4777 || !memcmp(fields
, table
->eviction_fields
,
4778 n_fields
* sizeof *fields
))) {
4783 oftable_disable_eviction(table
);
4785 table
->n_eviction_fields
= n_fields
;
4786 table
->eviction_fields
= xmemdup(fields
, n_fields
* sizeof *fields
);
4788 table
->eviction_group_id_basis
= random_uint32();
4789 hmap_init(&table
->eviction_groups_by_id
);
4790 heap_init(&table
->eviction_groups_by_size
);
4792 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
4793 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
4794 eviction_group_add_rule(rule
);
4798 /* Removes 'rule' from the oftable that contains it. */
4800 oftable_remove_rule(struct rule
*rule
)
4802 struct ofproto
*ofproto
= rule
->ofproto
;
4803 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
4805 classifier_remove(&table
->cls
, &rule
->cr
);
4806 eviction_group_remove_rule(rule
);
4807 if (!list_is_empty(&rule
->expirable
)) {
4808 list_remove(&rule
->expirable
);
4812 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4813 * oftable that has an identical cls_rule. Returns the rule that was removed,
4814 * if any, and otherwise NULL. */
4815 static struct rule
*
4816 oftable_replace_rule(struct rule
*rule
)
4818 struct ofproto
*ofproto
= rule
->ofproto
;
4819 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
4820 struct rule
*victim
;
4821 bool may_expire
= rule
->hard_timeout
|| rule
->idle_timeout
;
4824 list_insert(&ofproto
->expirable
, &rule
->expirable
);
4827 victim
= rule_from_cls_rule(classifier_replace(&table
->cls
, &rule
->cr
));
4829 if (!list_is_empty(&victim
->expirable
)) {
4830 list_remove(&victim
->expirable
);
4832 eviction_group_remove_rule(victim
);
4834 eviction_group_add_rule(rule
);
4838 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4840 oftable_substitute_rule(struct rule
*old
, struct rule
*new)
4843 oftable_replace_rule(new);
4845 oftable_remove_rule(old
);
4849 /* unixctl commands. */
4852 ofproto_lookup(const char *name
)
4854 struct ofproto
*ofproto
;
4856 HMAP_FOR_EACH_WITH_HASH (ofproto
, hmap_node
, hash_string(name
, 0),
4858 if (!strcmp(ofproto
->name
, name
)) {
4866 ofproto_unixctl_list(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
4867 const char *argv
[] OVS_UNUSED
, void *aux OVS_UNUSED
)
4869 struct ofproto
*ofproto
;
4873 HMAP_FOR_EACH (ofproto
, hmap_node
, &all_ofprotos
) {
4874 ds_put_format(&results
, "%s\n", ofproto
->name
);
4876 unixctl_command_reply(conn
, ds_cstr(&results
));
4877 ds_destroy(&results
);
4881 ofproto_unixctl_init(void)
4883 static bool registered
;
4889 unixctl_command_register("ofproto/list", "", 0, 0,
4890 ofproto_unixctl_list
, NULL
);
4893 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4895 * This is deprecated. It is only for compatibility with broken device drivers
4896 * in old versions of Linux that do not properly support VLANs when VLAN
4897 * devices are not used. When broken device drivers are no longer in
4898 * widespread use, we will delete these interfaces. */
4900 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4901 * (exactly) by an OpenFlow rule in 'ofproto'. */
4903 ofproto_get_vlan_usage(struct ofproto
*ofproto
, unsigned long int *vlan_bitmap
)
4905 const struct oftable
*oftable
;
4907 free(ofproto
->vlan_bitmap
);
4908 ofproto
->vlan_bitmap
= bitmap_allocate(4096);
4909 ofproto
->vlans_changed
= false;
4911 OFPROTO_FOR_EACH_TABLE (oftable
, ofproto
) {
4912 const struct cls_table
*table
;
4914 HMAP_FOR_EACH (table
, hmap_node
, &oftable
->cls
.tables
) {
4915 if (minimask_get_vid_mask(&table
->mask
) == VLAN_VID_MASK
) {
4916 const struct cls_rule
*rule
;
4918 HMAP_FOR_EACH (rule
, hmap_node
, &table
->rules
) {
4919 uint16_t vid
= miniflow_get_vid(&rule
->match
.flow
);
4920 bitmap_set1(vlan_bitmap
, vid
);
4921 bitmap_set1(ofproto
->vlan_bitmap
, vid
);
4928 /* Returns true if new VLANs have come into use by the flow table since the
4929 * last call to ofproto_get_vlan_usage().
4931 * We don't track when old VLANs stop being used. */
4933 ofproto_has_vlan_usage_changed(const struct ofproto
*ofproto
)
4935 return ofproto
->vlans_changed
;
4938 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4939 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4940 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4941 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4942 * then the VLAN device is un-enslaved. */
4944 ofproto_port_set_realdev(struct ofproto
*ofproto
, uint16_t vlandev_ofp_port
,
4945 uint16_t realdev_ofp_port
, int vid
)
4947 struct ofport
*ofport
;
4950 ovs_assert(vlandev_ofp_port
!= realdev_ofp_port
);
4952 ofport
= ofproto_get_port(ofproto
, vlandev_ofp_port
);
4954 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16
,
4955 ofproto
->name
, vlandev_ofp_port
);
4959 if (!ofproto
->ofproto_class
->set_realdev
) {
4960 if (!vlandev_ofp_port
) {
4963 VLOG_WARN("%s: vlan splinters not supported", ofproto
->name
);
4967 error
= ofproto
->ofproto_class
->set_realdev(ofport
, realdev_ofp_port
, vid
);
4969 VLOG_WARN("%s: setting realdev on port %"PRIu16
" (%s) failed (%s)",
4970 ofproto
->name
, vlandev_ofp_port
,
4971 netdev_get_name(ofport
->netdev
), strerror(error
));