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 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
803 * 'ofproto' does not have a port 'ofp_port'. */
805 ofproto_port_set_bfd(struct ofproto
*ofproto
, uint16_t ofp_port
,
806 const struct smap
*cfg
)
808 struct ofport
*ofport
;
811 ofport
= ofproto_get_port(ofproto
, ofp_port
);
813 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16
,
814 ofproto
->name
, ofp_port
);
817 error
= (ofproto
->ofproto_class
->set_bfd
818 ? ofproto
->ofproto_class
->set_bfd(ofport
, cfg
)
821 VLOG_WARN("%s: bfd configuration on port %"PRIu16
" (%s) failed (%s)",
822 ofproto
->name
, ofp_port
, netdev_get_name(ofport
->netdev
),
827 /* Populates 'status' with key value pairs indicating the status of the BFD
828 * session on 'ofp_port'. This information is intended to be populated in the
829 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
832 ofproto_port_get_bfd_status(struct ofproto
*ofproto
, uint16_t ofp_port
,
835 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
836 return (ofport
&& ofproto
->ofproto_class
->get_bfd_status
837 ? ofproto
->ofproto_class
->get_bfd_status(ofport
, status
)
841 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
842 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
843 * 0 if LACP partner information is not current (generally indicating a
844 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
846 ofproto_port_is_lacp_current(struct ofproto
*ofproto
, uint16_t ofp_port
)
848 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
849 return (ofport
&& ofproto
->ofproto_class
->port_is_lacp_current
850 ? ofproto
->ofproto_class
->port_is_lacp_current(ofport
)
856 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
857 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
858 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
859 * configuration plus, if there is more than one slave, a bonding
862 * If 'aux' is already registered then this function updates its configuration
863 * to 's'. Otherwise, this function registers a new bundle.
865 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
868 ofproto_bundle_register(struct ofproto
*ofproto
, void *aux
,
869 const struct ofproto_bundle_settings
*s
)
871 return (ofproto
->ofproto_class
->bundle_set
872 ? ofproto
->ofproto_class
->bundle_set(ofproto
, aux
, s
)
876 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
877 * If no such bundle has been registered, this has no effect. */
879 ofproto_bundle_unregister(struct ofproto
*ofproto
, void *aux
)
881 return ofproto_bundle_register(ofproto
, aux
, NULL
);
885 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
886 * If 'aux' is already registered then this function updates its configuration
887 * to 's'. Otherwise, this function registers a new mirror. */
889 ofproto_mirror_register(struct ofproto
*ofproto
, void *aux
,
890 const struct ofproto_mirror_settings
*s
)
892 return (ofproto
->ofproto_class
->mirror_set
893 ? ofproto
->ofproto_class
->mirror_set(ofproto
, aux
, s
)
897 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
898 * If no mirror has been registered, this has no effect. */
900 ofproto_mirror_unregister(struct ofproto
*ofproto
, void *aux
)
902 return ofproto_mirror_register(ofproto
, aux
, NULL
);
905 /* Retrieves statistics from mirror associated with client data pointer
906 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
907 * 'bytes', respectively. If a particular counters is not supported,
908 * the appropriate argument is set to UINT64_MAX. */
910 ofproto_mirror_get_stats(struct ofproto
*ofproto
, void *aux
,
911 uint64_t *packets
, uint64_t *bytes
)
913 if (!ofproto
->ofproto_class
->mirror_get_stats
) {
914 *packets
= *bytes
= UINT64_MAX
;
918 return ofproto
->ofproto_class
->mirror_get_stats(ofproto
, aux
,
922 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
923 * which all packets are flooded, instead of using MAC learning. If
924 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
926 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
929 ofproto_set_flood_vlans(struct ofproto
*ofproto
, unsigned long *flood_vlans
)
931 return (ofproto
->ofproto_class
->set_flood_vlans
932 ? ofproto
->ofproto_class
->set_flood_vlans(ofproto
, flood_vlans
)
936 /* Returns true if 'aux' is a registered bundle that is currently in use as the
937 * output for a mirror. */
939 ofproto_is_mirror_output_bundle(const struct ofproto
*ofproto
, void *aux
)
941 return (ofproto
->ofproto_class
->is_mirror_output_bundle
942 ? ofproto
->ofproto_class
->is_mirror_output_bundle(ofproto
, aux
)
946 /* Configuration of OpenFlow tables. */
948 /* Returns the number of OpenFlow tables in 'ofproto'. */
950 ofproto_get_n_tables(const struct ofproto
*ofproto
)
952 return ofproto
->n_tables
;
955 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
956 * settings from 's'. 'table_id' must be in the range 0 through the number of
957 * OpenFlow tables in 'ofproto' minus 1, inclusive.
959 * For read-only tables, only the name may be configured. */
961 ofproto_configure_table(struct ofproto
*ofproto
, int table_id
,
962 const struct ofproto_table_settings
*s
)
964 struct oftable
*table
;
966 ovs_assert(table_id
>= 0 && table_id
< ofproto
->n_tables
);
967 table
= &ofproto
->tables
[table_id
];
969 oftable_set_name(table
, s
->name
);
971 if (table
->flags
& OFTABLE_READONLY
) {
976 oftable_enable_eviction(table
, s
->groups
, s
->n_groups
);
978 oftable_disable_eviction(table
);
981 table
->max_flows
= s
->max_flows
;
982 if (classifier_count(&table
->cls
) > table
->max_flows
983 && table
->eviction_fields
) {
984 /* 'table' contains more flows than allowed. We might not be able to
985 * evict them right away because of the asynchronous nature of flow
986 * table changes. Schedule eviction for later. */
987 switch (ofproto
->state
) {
989 ofproto
->state
= S_EVICT
;
993 /* We're already deleting flows, nothing more to do. */
1000 ofproto_has_snoops(const struct ofproto
*ofproto
)
1002 return connmgr_has_snoops(ofproto
->connmgr
);
1006 ofproto_get_snoops(const struct ofproto
*ofproto
, struct sset
*snoops
)
1008 connmgr_get_snoops(ofproto
->connmgr
, snoops
);
1012 ofproto_flush__(struct ofproto
*ofproto
)
1014 struct ofopgroup
*group
;
1015 struct oftable
*table
;
1017 if (ofproto
->ofproto_class
->flush
) {
1018 ofproto
->ofproto_class
->flush(ofproto
);
1021 group
= ofopgroup_create_unattached(ofproto
);
1022 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
1023 struct rule
*rule
, *next_rule
;
1024 struct cls_cursor cursor
;
1026 if (table
->flags
& OFTABLE_HIDDEN
) {
1030 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
1031 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
1032 if (!rule
->pending
) {
1033 ofoperation_create(group
, rule
, OFOPERATION_DELETE
,
1035 oftable_remove_rule(rule
);
1036 ofproto
->ofproto_class
->rule_destruct(rule
);
1040 ofopgroup_submit(group
);
1044 ofproto_destroy__(struct ofproto
*ofproto
)
1046 struct oftable
*table
;
1048 ovs_assert(list_is_empty(&ofproto
->pending
));
1049 ovs_assert(!ofproto
->n_pending
);
1051 connmgr_destroy(ofproto
->connmgr
);
1053 hmap_remove(&all_ofprotos
, &ofproto
->hmap_node
);
1054 free(ofproto
->name
);
1055 free(ofproto
->type
);
1056 free(ofproto
->mfr_desc
);
1057 free(ofproto
->hw_desc
);
1058 free(ofproto
->sw_desc
);
1059 free(ofproto
->serial_desc
);
1060 free(ofproto
->dp_desc
);
1061 hmap_destroy(&ofproto
->ports
);
1062 shash_destroy(&ofproto
->port_by_name
);
1063 bitmap_free(ofproto
->ofp_port_ids
);
1064 simap_destroy(&ofproto
->ofp_requests
);
1066 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
1067 oftable_destroy(table
);
1069 free(ofproto
->tables
);
1071 hmap_destroy(&ofproto
->deletions
);
1073 free(ofproto
->vlan_bitmap
);
1075 ofproto
->ofproto_class
->dealloc(ofproto
);
1079 ofproto_destroy(struct ofproto
*p
)
1081 struct ofport
*ofport
, *next_ofport
;
1088 HMAP_FOR_EACH_SAFE (ofport
, next_ofport
, hmap_node
, &p
->ports
) {
1089 ofport_destroy(ofport
);
1092 p
->ofproto_class
->destruct(p
);
1093 ofproto_destroy__(p
);
1096 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1097 * kernel datapath, for example, this destroys the datapath in the kernel, and
1098 * with the netdev-based datapath, it tears down the data structures that
1099 * represent the datapath.
1101 * The datapath should not be currently open as an ofproto. */
1103 ofproto_delete(const char *name
, const char *type
)
1105 const struct ofproto_class
*class = ofproto_class_find__(type
);
1106 return (!class ? EAFNOSUPPORT
1107 : !class->del
? EACCES
1108 : class->del(type
, name
));
1112 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
1114 if (error
== ENOBUFS
) {
1115 reinit_ports(ofproto
);
1116 } else if (!error
) {
1117 update_port(ofproto
, devname
);
1123 ofproto_type_run(const char *datapath_type
)
1125 const struct ofproto_class
*class;
1128 datapath_type
= ofproto_normalize_type(datapath_type
);
1129 class = ofproto_class_find__(datapath_type
);
1131 error
= class->type_run
? class->type_run(datapath_type
) : 0;
1132 if (error
&& error
!= EAGAIN
) {
1133 VLOG_ERR_RL(&rl
, "%s: type_run failed (%s)",
1134 datapath_type
, strerror(error
));
1140 ofproto_type_run_fast(const char *datapath_type
)
1142 const struct ofproto_class
*class;
1145 datapath_type
= ofproto_normalize_type(datapath_type
);
1146 class = ofproto_class_find__(datapath_type
);
1148 error
= class->type_run_fast
? class->type_run_fast(datapath_type
) : 0;
1149 if (error
&& error
!= EAGAIN
) {
1150 VLOG_ERR_RL(&rl
, "%s: type_run_fast failed (%s)",
1151 datapath_type
, strerror(error
));
1157 ofproto_type_wait(const char *datapath_type
)
1159 const struct ofproto_class
*class;
1161 datapath_type
= ofproto_normalize_type(datapath_type
);
1162 class = ofproto_class_find__(datapath_type
);
1164 if (class->type_wait
) {
1165 class->type_wait(datapath_type
);
1170 ofproto_run(struct ofproto
*p
)
1172 struct sset changed_netdevs
;
1173 const char *changed_netdev
;
1174 struct ofport
*ofport
;
1177 error
= p
->ofproto_class
->run(p
);
1178 if (error
&& error
!= EAGAIN
) {
1179 VLOG_ERR_RL(&rl
, "%s: run failed (%s)", p
->name
, strerror(error
));
1182 if (p
->ofproto_class
->port_poll
) {
1185 while ((error
= p
->ofproto_class
->port_poll(p
, &devname
)) != EAGAIN
) {
1186 process_port_change(p
, error
, devname
);
1190 /* Update OpenFlow port status for any port whose netdev has changed.
1192 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1193 * destroyed, so it's not safe to update ports directly from the
1194 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1195 * need this two-phase approach. */
1196 sset_init(&changed_netdevs
);
1197 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1198 unsigned int change_seq
= netdev_change_seq(ofport
->netdev
);
1199 if (ofport
->change_seq
!= change_seq
) {
1200 ofport
->change_seq
= change_seq
;
1201 sset_add(&changed_netdevs
, netdev_get_name(ofport
->netdev
));
1204 SSET_FOR_EACH (changed_netdev
, &changed_netdevs
) {
1205 update_port(p
, changed_netdev
);
1207 sset_destroy(&changed_netdevs
);
1211 connmgr_run(p
->connmgr
, handle_openflow
);
1215 connmgr_run(p
->connmgr
, NULL
);
1217 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1218 p
->state
= S_OPENFLOW
;
1223 connmgr_run(p
->connmgr
, NULL
);
1225 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1226 connmgr_flushed(p
->connmgr
);
1227 p
->state
= S_OPENFLOW
;
1235 if (time_msec() >= p
->next_op_report
) {
1236 long long int ago
= (time_msec() - p
->first_op
) / 1000;
1237 long long int interval
= (p
->last_op
- p
->first_op
) / 1000;
1241 ds_put_format(&s
, "%d flow_mods ",
1242 p
->n_add
+ p
->n_delete
+ p
->n_modify
);
1243 if (interval
== ago
) {
1244 ds_put_format(&s
, "in the last %lld s", ago
);
1245 } else if (interval
) {
1246 ds_put_format(&s
, "in the %lld s starting %lld s ago",
1249 ds_put_format(&s
, "%lld s ago", ago
);
1252 ds_put_cstr(&s
, " (");
1254 ds_put_format(&s
, "%d adds, ", p
->n_add
);
1257 ds_put_format(&s
, "%d deletes, ", p
->n_delete
);
1260 ds_put_format(&s
, "%d modifications, ", p
->n_modify
);
1263 ds_put_char(&s
, ')');
1265 VLOG_INFO("%s: %s", p
->name
, ds_cstr(&s
));
1268 p
->n_add
= p
->n_delete
= p
->n_modify
= 0;
1269 p
->next_op_report
= LLONG_MAX
;
1275 /* Performs periodic activity required by 'ofproto' that needs to be done
1276 * with the least possible latency.
1278 * It makes sense to call this function a couple of times per poll loop, to
1279 * provide a significant performance boost on some benchmarks with the
1280 * ofproto-dpif implementation. */
1282 ofproto_run_fast(struct ofproto
*p
)
1286 error
= p
->ofproto_class
->run_fast
? p
->ofproto_class
->run_fast(p
) : 0;
1287 if (error
&& error
!= EAGAIN
) {
1288 VLOG_ERR_RL(&rl
, "%s: fastpath run failed (%s)",
1289 p
->name
, strerror(error
));
1295 ofproto_wait(struct ofproto
*p
)
1297 struct ofport
*ofport
;
1299 p
->ofproto_class
->wait(p
);
1300 if (p
->ofproto_class
->port_poll_wait
) {
1301 p
->ofproto_class
->port_poll_wait(p
);
1304 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1305 if (ofport
->change_seq
!= netdev_change_seq(ofport
->netdev
)) {
1306 poll_immediate_wake();
1312 connmgr_wait(p
->connmgr
, true);
1317 connmgr_wait(p
->connmgr
, false);
1318 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1319 poll_immediate_wake();
1326 ofproto_is_alive(const struct ofproto
*p
)
1328 return connmgr_has_controllers(p
->connmgr
);
1331 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1332 * memory_report(). */
1334 ofproto_get_memory_usage(const struct ofproto
*ofproto
, struct simap
*usage
)
1336 const struct oftable
*table
;
1337 unsigned int n_rules
;
1339 simap_increase(usage
, "ports", hmap_count(&ofproto
->ports
));
1340 simap_increase(usage
, "ops",
1341 ofproto
->n_pending
+ hmap_count(&ofproto
->deletions
));
1344 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
1345 n_rules
+= classifier_count(&table
->cls
);
1347 simap_increase(usage
, "rules", n_rules
);
1349 if (ofproto
->ofproto_class
->get_memory_usage
) {
1350 ofproto
->ofproto_class
->get_memory_usage(ofproto
, usage
);
1353 connmgr_get_memory_usage(ofproto
->connmgr
, usage
);
1357 ofproto_get_ofproto_controller_info(const struct ofproto
*ofproto
,
1360 connmgr_get_controller_info(ofproto
->connmgr
, info
);
1364 ofproto_free_ofproto_controller_info(struct shash
*info
)
1366 connmgr_free_controller_info(info
);
1369 /* Makes a deep copy of 'old' into 'port'. */
1371 ofproto_port_clone(struct ofproto_port
*port
, const struct ofproto_port
*old
)
1373 port
->name
= xstrdup(old
->name
);
1374 port
->type
= xstrdup(old
->type
);
1375 port
->ofp_port
= old
->ofp_port
;
1378 /* Frees memory allocated to members of 'ofproto_port'.
1380 * Do not call this function on an ofproto_port obtained from
1381 * ofproto_port_dump_next(): that function retains ownership of the data in the
1384 ofproto_port_destroy(struct ofproto_port
*ofproto_port
)
1386 free(ofproto_port
->name
);
1387 free(ofproto_port
->type
);
1390 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1392 * This function provides no status indication. An error status for the entire
1393 * dump operation is provided when it is completed by calling
1394 * ofproto_port_dump_done().
1397 ofproto_port_dump_start(struct ofproto_port_dump
*dump
,
1398 const struct ofproto
*ofproto
)
1400 dump
->ofproto
= ofproto
;
1401 dump
->error
= ofproto
->ofproto_class
->port_dump_start(ofproto
,
1405 /* Attempts to retrieve another port from 'dump', which must have been created
1406 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1407 * 'port' and returns true. On failure, returns false.
1409 * Failure might indicate an actual error or merely that the last port has been
1410 * dumped. An error status for the entire dump operation is provided when it
1411 * is completed by calling ofproto_port_dump_done().
1413 * The ofproto owns the data stored in 'port'. It will remain valid until at
1414 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1415 * ofproto_port_dump_done(). */
1417 ofproto_port_dump_next(struct ofproto_port_dump
*dump
,
1418 struct ofproto_port
*port
)
1420 const struct ofproto
*ofproto
= dump
->ofproto
;
1426 dump
->error
= ofproto
->ofproto_class
->port_dump_next(ofproto
, dump
->state
,
1429 ofproto
->ofproto_class
->port_dump_done(ofproto
, dump
->state
);
1435 /* Completes port table dump operation 'dump', which must have been created
1436 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1437 * error-free, otherwise a positive errno value describing the problem. */
1439 ofproto_port_dump_done(struct ofproto_port_dump
*dump
)
1441 const struct ofproto
*ofproto
= dump
->ofproto
;
1443 dump
->error
= ofproto
->ofproto_class
->port_dump_done(ofproto
,
1446 return dump
->error
== EOF
? 0 : dump
->error
;
1449 /* Returns the type to pass to netdev_open() when a datapath of type
1450 * 'datapath_type' has a port of type 'port_type', for a few special
1451 * cases when a netdev type differs from a port type. For example, when
1452 * using the userspace datapath, a port of type "internal" needs to be
1455 * Returns either 'type' itself or a string literal, which must not be
1458 ofproto_port_open_type(const char *datapath_type
, const char *port_type
)
1460 const struct ofproto_class
*class;
1462 datapath_type
= ofproto_normalize_type(datapath_type
);
1463 class = ofproto_class_find__(datapath_type
);
1468 return (class->port_open_type
1469 ? class->port_open_type(datapath_type
, port_type
)
1473 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1474 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1475 * the port's OpenFlow port number.
1477 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1478 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1479 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1480 * 'ofp_portp' is non-null). */
1482 ofproto_port_add(struct ofproto
*ofproto
, struct netdev
*netdev
,
1483 uint16_t *ofp_portp
)
1485 uint16_t ofp_port
= ofp_portp
? *ofp_portp
: OFPP_NONE
;
1488 error
= ofproto
->ofproto_class
->port_add(ofproto
, netdev
);
1490 const char *netdev_name
= netdev_get_name(netdev
);
1492 simap_put(&ofproto
->ofp_requests
, netdev_name
, ofp_port
);
1493 update_port(ofproto
, netdev_name
);
1496 struct ofproto_port ofproto_port
;
1498 ofproto_port_query_by_name(ofproto
, netdev_get_name(netdev
),
1500 *ofp_portp
= error
? OFPP_NONE
: ofproto_port
.ofp_port
;
1501 ofproto_port_destroy(&ofproto_port
);
1506 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1507 * initializes '*port' appropriately; on failure, returns a positive errno
1510 * The caller owns the data in 'ofproto_port' and must free it with
1511 * ofproto_port_destroy() when it is no longer needed. */
1513 ofproto_port_query_by_name(const struct ofproto
*ofproto
, const char *devname
,
1514 struct ofproto_port
*port
)
1518 error
= ofproto
->ofproto_class
->port_query_by_name(ofproto
, devname
, port
);
1520 memset(port
, 0, sizeof *port
);
1525 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1526 * Returns 0 if successful, otherwise a positive errno. */
1528 ofproto_port_del(struct ofproto
*ofproto
, uint16_t ofp_port
)
1530 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
1531 const char *name
= ofport
? netdev_get_name(ofport
->netdev
) : "<unknown>";
1532 struct simap_node
*ofp_request_node
;
1535 ofp_request_node
= simap_find(&ofproto
->ofp_requests
, name
);
1536 if (ofp_request_node
) {
1537 simap_delete(&ofproto
->ofp_requests
, ofp_request_node
);
1540 error
= ofproto
->ofproto_class
->port_del(ofproto
, ofp_port
);
1541 if (!error
&& ofport
) {
1542 /* 'name' is the netdev's name and update_port() is going to close the
1543 * netdev. Just in case update_port() refers to 'name' after it
1544 * destroys 'ofport', make a copy of it around the update_port()
1546 char *devname
= xstrdup(name
);
1547 update_port(ofproto
, devname
);
1553 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1554 * performs the 'n_actions' actions in 'actions'. The new flow will not
1557 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1558 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1559 * controllers; otherwise, it will be hidden.
1561 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1563 * This is a helper function for in-band control and fail-open. */
1565 ofproto_add_flow(struct ofproto
*ofproto
, const struct match
*match
,
1566 unsigned int priority
,
1567 const struct ofpact
*ofpacts
, size_t ofpacts_len
)
1569 const struct rule
*rule
;
1571 rule
= rule_from_cls_rule(classifier_find_match_exactly(
1572 &ofproto
->tables
[0].cls
, match
, priority
));
1573 if (!rule
|| !ofpacts_equal(rule
->ofpacts
, rule
->ofpacts_len
,
1574 ofpacts
, ofpacts_len
)) {
1575 struct ofputil_flow_mod fm
;
1577 memset(&fm
, 0, sizeof fm
);
1579 fm
.priority
= priority
;
1580 fm
.buffer_id
= UINT32_MAX
;
1581 fm
.ofpacts
= xmemdup(ofpacts
, ofpacts_len
);
1582 fm
.ofpacts_len
= ofpacts_len
;
1583 add_flow(ofproto
, NULL
, &fm
, NULL
);
1588 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1589 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1590 * operation cannot be initiated now but may be retried later.
1592 * This is a helper function for in-band control and fail-open. */
1594 ofproto_flow_mod(struct ofproto
*ofproto
, const struct ofputil_flow_mod
*fm
)
1596 return handle_flow_mod__(ofproto
, NULL
, fm
, NULL
);
1599 /* Searches for a rule with matching criteria exactly equal to 'target' in
1600 * ofproto's table 0 and, if it finds one, deletes it.
1602 * This is a helper function for in-band control and fail-open. */
1604 ofproto_delete_flow(struct ofproto
*ofproto
,
1605 const struct match
*target
, unsigned int priority
)
1609 rule
= rule_from_cls_rule(classifier_find_match_exactly(
1610 &ofproto
->tables
[0].cls
, target
, priority
));
1612 /* No such rule -> success. */
1614 } else if (rule
->pending
) {
1615 /* An operation on the rule is already pending -> failure.
1616 * Caller must retry later if it's important. */
1619 /* Initiate deletion -> success. */
1620 struct ofopgroup
*group
= ofopgroup_create_unattached(ofproto
);
1621 ofoperation_create(group
, rule
, OFOPERATION_DELETE
, OFPRR_DELETE
);
1622 oftable_remove_rule(rule
);
1623 ofproto
->ofproto_class
->rule_destruct(rule
);
1624 ofopgroup_submit(group
);
1630 /* Starts the process of deleting all of the flows from all of ofproto's flow
1631 * tables and then reintroducing the flows required by in-band control and
1632 * fail-open. The process will complete in a later call to ofproto_run(). */
1634 ofproto_flush_flows(struct ofproto
*ofproto
)
1636 COVERAGE_INC(ofproto_flush
);
1637 ofproto
->state
= S_FLUSH
;
1641 reinit_ports(struct ofproto
*p
)
1643 struct ofproto_port_dump dump
;
1644 struct sset devnames
;
1645 struct ofport
*ofport
;
1646 struct ofproto_port ofproto_port
;
1647 const char *devname
;
1649 COVERAGE_INC(ofproto_reinit_ports
);
1651 sset_init(&devnames
);
1652 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1653 sset_add(&devnames
, netdev_get_name(ofport
->netdev
));
1655 OFPROTO_PORT_FOR_EACH (&ofproto_port
, &dump
, p
) {
1656 sset_add(&devnames
, ofproto_port
.name
);
1659 SSET_FOR_EACH (devname
, &devnames
) {
1660 update_port(p
, devname
);
1662 sset_destroy(&devnames
);
1666 alloc_ofp_port(struct ofproto
*ofproto
, const char *netdev_name
)
1669 uint16_t end_port_no
= ofproto
->alloc_port_no
;
1671 ofp_port
= simap_get(&ofproto
->ofp_requests
, netdev_name
);
1672 ofp_port
= ofp_port
? ofp_port
: OFPP_NONE
;
1674 if (ofp_port
>= ofproto
->max_ports
1675 || bitmap_is_set(ofproto
->ofp_port_ids
, ofp_port
)) {
1676 /* Search for a free OpenFlow port number. We try not to
1677 * immediately reuse them to prevent problems due to old
1680 if (++ofproto
->alloc_port_no
>= ofproto
->max_ports
) {
1681 ofproto
->alloc_port_no
= 0;
1683 if (!bitmap_is_set(ofproto
->ofp_port_ids
,
1684 ofproto
->alloc_port_no
)) {
1685 ofp_port
= ofproto
->alloc_port_no
;
1688 if (ofproto
->alloc_port_no
== end_port_no
) {
1693 bitmap_set1(ofproto
->ofp_port_ids
, ofp_port
);
1698 dealloc_ofp_port(const struct ofproto
*ofproto
, uint16_t ofp_port
)
1700 if (ofp_port
< ofproto
->max_ports
) {
1701 bitmap_set0(ofproto
->ofp_port_ids
, ofp_port
);
1705 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1706 * pointer if the netdev cannot be opened. On success, also fills in
1708 static struct netdev
*
1709 ofport_open(struct ofproto
*ofproto
,
1710 struct ofproto_port
*ofproto_port
,
1711 struct ofputil_phy_port
*pp
)
1713 enum netdev_flags flags
;
1714 struct netdev
*netdev
;
1717 error
= netdev_open(ofproto_port
->name
, ofproto_port
->type
, &netdev
);
1719 VLOG_WARN_RL(&rl
, "%s: ignoring port %s (%"PRIu16
") because netdev %s "
1720 "cannot be opened (%s)",
1722 ofproto_port
->name
, ofproto_port
->ofp_port
,
1723 ofproto_port
->name
, strerror(error
));
1727 if (ofproto_port
->ofp_port
== OFPP_NONE
) {
1728 if (!strcmp(ofproto
->name
, ofproto_port
->name
)) {
1729 ofproto_port
->ofp_port
= OFPP_LOCAL
;
1731 ofproto_port
->ofp_port
= alloc_ofp_port(ofproto
,
1732 ofproto_port
->name
);
1735 pp
->port_no
= ofproto_port
->ofp_port
;
1736 netdev_get_etheraddr(netdev
, pp
->hw_addr
);
1737 ovs_strlcpy(pp
->name
, ofproto_port
->name
, sizeof pp
->name
);
1738 netdev_get_flags(netdev
, &flags
);
1739 pp
->config
= flags
& NETDEV_UP
? 0 : OFPUTIL_PC_PORT_DOWN
;
1740 pp
->state
= netdev_get_carrier(netdev
) ? 0 : OFPUTIL_PS_LINK_DOWN
;
1741 netdev_get_features(netdev
, &pp
->curr
, &pp
->advertised
,
1742 &pp
->supported
, &pp
->peer
);
1743 pp
->curr_speed
= netdev_features_to_bps(pp
->curr
, 0);
1744 pp
->max_speed
= netdev_features_to_bps(pp
->supported
, 0);
1749 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1750 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1753 ofport_equal(const struct ofputil_phy_port
*a
,
1754 const struct ofputil_phy_port
*b
)
1756 return (eth_addr_equals(a
->hw_addr
, b
->hw_addr
)
1757 && a
->state
== b
->state
1758 && !((a
->config
^ b
->config
) & OFPUTIL_PC_PORT_DOWN
)
1759 && a
->curr
== b
->curr
1760 && a
->advertised
== b
->advertised
1761 && a
->supported
== b
->supported
1762 && a
->peer
== b
->peer
1763 && a
->curr_speed
== b
->curr_speed
1764 && a
->max_speed
== b
->max_speed
);
1767 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1768 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1769 * one with the same name or port number). */
1771 ofport_install(struct ofproto
*p
,
1772 struct netdev
*netdev
, const struct ofputil_phy_port
*pp
)
1774 const char *netdev_name
= netdev_get_name(netdev
);
1775 struct ofport
*ofport
;
1778 /* Create ofport. */
1779 ofport
= p
->ofproto_class
->port_alloc();
1784 ofport
->ofproto
= p
;
1785 ofport
->netdev
= netdev
;
1786 ofport
->change_seq
= netdev_change_seq(netdev
);
1788 ofport
->ofp_port
= pp
->port_no
;
1790 /* Add port to 'p'. */
1791 hmap_insert(&p
->ports
, &ofport
->hmap_node
, hash_int(ofport
->ofp_port
, 0));
1792 shash_add(&p
->port_by_name
, netdev_name
, ofport
);
1794 update_mtu(p
, ofport
);
1796 /* Let the ofproto_class initialize its private data. */
1797 error
= p
->ofproto_class
->port_construct(ofport
);
1801 connmgr_send_port_status(p
->connmgr
, pp
, OFPPR_ADD
);
1805 VLOG_WARN_RL(&rl
, "%s: could not add port %s (%s)",
1806 p
->name
, netdev_name
, strerror(error
));
1808 ofport_destroy__(ofport
);
1810 netdev_close(netdev
);
1814 /* Removes 'ofport' from 'p' and destroys it. */
1816 ofport_remove(struct ofport
*ofport
)
1818 connmgr_send_port_status(ofport
->ofproto
->connmgr
, &ofport
->pp
,
1820 ofport_destroy(ofport
);
1823 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1826 ofport_remove_with_name(struct ofproto
*ofproto
, const char *name
)
1828 struct ofport
*port
= shash_find_data(&ofproto
->port_by_name
, name
);
1830 ofport_remove(port
);
1834 /* Updates 'port' with new 'pp' description.
1836 * Does not handle a name or port number change. The caller must implement
1837 * such a change as a delete followed by an add. */
1839 ofport_modified(struct ofport
*port
, struct ofputil_phy_port
*pp
)
1841 memcpy(port
->pp
.hw_addr
, pp
->hw_addr
, ETH_ADDR_LEN
);
1842 port
->pp
.config
= ((port
->pp
.config
& ~OFPUTIL_PC_PORT_DOWN
)
1843 | (pp
->config
& OFPUTIL_PC_PORT_DOWN
));
1844 port
->pp
.state
= pp
->state
;
1845 port
->pp
.curr
= pp
->curr
;
1846 port
->pp
.advertised
= pp
->advertised
;
1847 port
->pp
.supported
= pp
->supported
;
1848 port
->pp
.peer
= pp
->peer
;
1849 port
->pp
.curr_speed
= pp
->curr_speed
;
1850 port
->pp
.max_speed
= pp
->max_speed
;
1852 connmgr_send_port_status(port
->ofproto
->connmgr
, &port
->pp
, OFPPR_MODIFY
);
1855 /* Update OpenFlow 'state' in 'port' and notify controller. */
1857 ofproto_port_set_state(struct ofport
*port
, enum ofputil_port_state state
)
1859 if (port
->pp
.state
!= state
) {
1860 port
->pp
.state
= state
;
1861 connmgr_send_port_status(port
->ofproto
->connmgr
, &port
->pp
,
1867 ofproto_port_unregister(struct ofproto
*ofproto
, uint16_t ofp_port
)
1869 struct ofport
*port
= ofproto_get_port(ofproto
, ofp_port
);
1871 if (port
->ofproto
->ofproto_class
->set_realdev
) {
1872 port
->ofproto
->ofproto_class
->set_realdev(port
, 0, 0);
1874 if (port
->ofproto
->ofproto_class
->set_stp_port
) {
1875 port
->ofproto
->ofproto_class
->set_stp_port(port
, NULL
);
1877 if (port
->ofproto
->ofproto_class
->set_cfm
) {
1878 port
->ofproto
->ofproto_class
->set_cfm(port
, NULL
);
1880 if (port
->ofproto
->ofproto_class
->bundle_remove
) {
1881 port
->ofproto
->ofproto_class
->bundle_remove(port
);
1887 ofport_destroy__(struct ofport
*port
)
1889 struct ofproto
*ofproto
= port
->ofproto
;
1890 const char *name
= netdev_get_name(port
->netdev
);
1892 hmap_remove(&ofproto
->ports
, &port
->hmap_node
);
1893 shash_delete(&ofproto
->port_by_name
,
1894 shash_find(&ofproto
->port_by_name
, name
));
1896 netdev_close(port
->netdev
);
1897 ofproto
->ofproto_class
->port_dealloc(port
);
1901 ofport_destroy(struct ofport
*port
)
1904 dealloc_ofp_port(port
->ofproto
, port
->ofp_port
);
1905 port
->ofproto
->ofproto_class
->port_destruct(port
);
1906 ofport_destroy__(port
);
1911 ofproto_get_port(const struct ofproto
*ofproto
, uint16_t ofp_port
)
1913 struct ofport
*port
;
1915 HMAP_FOR_EACH_IN_BUCKET (port
, hmap_node
,
1916 hash_int(ofp_port
, 0), &ofproto
->ports
) {
1917 if (port
->ofp_port
== ofp_port
) {
1925 ofproto_port_get_stats(const struct ofport
*port
, struct netdev_stats
*stats
)
1927 struct ofproto
*ofproto
= port
->ofproto
;
1930 if (ofproto
->ofproto_class
->port_get_stats
) {
1931 error
= ofproto
->ofproto_class
->port_get_stats(port
, stats
);
1940 update_port(struct ofproto
*ofproto
, const char *name
)
1942 struct ofproto_port ofproto_port
;
1943 struct ofputil_phy_port pp
;
1944 struct netdev
*netdev
;
1945 struct ofport
*port
;
1947 COVERAGE_INC(ofproto_update_port
);
1949 /* Fetch 'name''s location and properties from the datapath. */
1950 netdev
= (!ofproto_port_query_by_name(ofproto
, name
, &ofproto_port
)
1951 ? ofport_open(ofproto
, &ofproto_port
, &pp
)
1954 port
= ofproto_get_port(ofproto
, ofproto_port
.ofp_port
);
1955 if (port
&& !strcmp(netdev_get_name(port
->netdev
), name
)) {
1956 struct netdev
*old_netdev
= port
->netdev
;
1958 /* 'name' hasn't changed location. Any properties changed? */
1959 if (!ofport_equal(&port
->pp
, &pp
)) {
1960 ofport_modified(port
, &pp
);
1963 update_mtu(ofproto
, port
);
1965 /* Install the newly opened netdev in case it has changed.
1966 * Don't close the old netdev yet in case port_modified has to
1967 * remove a retained reference to it.*/
1968 port
->netdev
= netdev
;
1969 port
->change_seq
= netdev_change_seq(netdev
);
1971 if (port
->ofproto
->ofproto_class
->port_modified
) {
1972 port
->ofproto
->ofproto_class
->port_modified(port
);
1975 netdev_close(old_netdev
);
1977 /* If 'port' is nonnull then its name differs from 'name' and thus
1978 * we should delete it. If we think there's a port named 'name'
1979 * then its port number must be wrong now so delete it too. */
1981 ofport_remove(port
);
1983 ofport_remove_with_name(ofproto
, name
);
1984 ofport_install(ofproto
, netdev
, &pp
);
1987 /* Any port named 'name' is gone now. */
1988 ofport_remove_with_name(ofproto
, name
);
1990 ofproto_port_destroy(&ofproto_port
);
1994 init_ports(struct ofproto
*p
)
1996 struct ofproto_port_dump dump
;
1997 struct ofproto_port ofproto_port
;
1998 struct shash_node
*node
, *next
;
2000 OFPROTO_PORT_FOR_EACH (&ofproto_port
, &dump
, p
) {
2001 const char *name
= ofproto_port
.name
;
2003 if (shash_find(&p
->port_by_name
, name
)) {
2004 VLOG_WARN_RL(&rl
, "%s: ignoring duplicate device %s in datapath",
2007 struct ofputil_phy_port pp
;
2008 struct netdev
*netdev
;
2010 /* Check if an OpenFlow port number had been requested. */
2011 node
= shash_find(&init_ofp_ports
, name
);
2013 const struct iface_hint
*iface_hint
= node
->data
;
2014 simap_put(&p
->ofp_requests
, name
, iface_hint
->ofp_port
);
2017 netdev
= ofport_open(p
, &ofproto_port
, &pp
);
2019 ofport_install(p
, netdev
, &pp
);
2024 SHASH_FOR_EACH_SAFE(node
, next
, &init_ofp_ports
) {
2025 struct iface_hint
*iface_hint
= node
->data
;
2027 if (!strcmp(iface_hint
->br_name
, p
->name
)) {
2028 free(iface_hint
->br_name
);
2029 free(iface_hint
->br_type
);
2031 shash_delete(&init_ofp_ports
, node
);
2038 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2039 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2041 find_min_mtu(struct ofproto
*p
)
2043 struct ofport
*ofport
;
2046 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
2047 struct netdev
*netdev
= ofport
->netdev
;
2050 /* Skip any internal ports, since that's what we're trying to
2052 if (!strcmp(netdev_get_type(netdev
), "internal")) {
2056 if (netdev_get_mtu(netdev
, &dev_mtu
)) {
2059 if (!mtu
|| dev_mtu
< mtu
) {
2064 return mtu
? mtu
: ETH_PAYLOAD_MAX
;
2067 /* Update MTU of all datapath devices on 'p' to the minimum of the
2068 * non-datapath ports in event of 'port' added or changed. */
2070 update_mtu(struct ofproto
*p
, struct ofport
*port
)
2072 struct ofport
*ofport
;
2073 struct netdev
*netdev
= port
->netdev
;
2074 int dev_mtu
, old_min
;
2076 if (netdev_get_mtu(netdev
, &dev_mtu
)) {
2080 if (!strcmp(netdev_get_type(port
->netdev
), "internal")) {
2081 if (dev_mtu
> p
->min_mtu
) {
2082 if (!netdev_set_mtu(port
->netdev
, p
->min_mtu
)) {
2083 dev_mtu
= p
->min_mtu
;
2086 port
->mtu
= dev_mtu
;
2090 /* For non-internal port find new min mtu. */
2091 old_min
= p
->min_mtu
;
2092 port
->mtu
= dev_mtu
;
2093 p
->min_mtu
= find_min_mtu(p
);
2094 if (p
->min_mtu
== old_min
) {
2098 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
2099 struct netdev
*netdev
= ofport
->netdev
;
2101 if (!strcmp(netdev_get_type(netdev
), "internal")) {
2102 if (!netdev_set_mtu(netdev
, p
->min_mtu
)) {
2103 ofport
->mtu
= p
->min_mtu
;
2110 ofproto_rule_destroy__(struct rule
*rule
)
2113 cls_rule_destroy(&rule
->cr
);
2114 free(rule
->ofpacts
);
2115 rule
->ofproto
->ofproto_class
->rule_dealloc(rule
);
2119 /* This function allows an ofproto implementation to destroy any rules that
2120 * remain when its ->destruct() function is called. The caller must have
2121 * already uninitialized any derived members of 'rule' (step 5 described in the
2122 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
2123 * This function implements steps 6 and 7.
2125 * This function should only be called from an ofproto implementation's
2126 * ->destruct() function. It is not suitable elsewhere. */
2128 ofproto_rule_destroy(struct rule
*rule
)
2130 ovs_assert(!rule
->pending
);
2131 oftable_remove_rule(rule
);
2132 ofproto_rule_destroy__(rule
);
2135 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2136 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2138 ofproto_rule_has_out_port(const struct rule
*rule
, uint16_t port
)
2140 return (port
== OFPP_ANY
2141 || ofpacts_output_to_port(rule
->ofpacts
, rule
->ofpacts_len
, port
));
2144 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2145 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2147 ofoperation_has_out_port(const struct ofoperation
*op
, uint16_t out_port
)
2149 if (ofproto_rule_has_out_port(op
->rule
, out_port
)) {
2154 case OFOPERATION_ADD
:
2155 return op
->victim
&& ofproto_rule_has_out_port(op
->victim
, out_port
);
2157 case OFOPERATION_DELETE
:
2160 case OFOPERATION_MODIFY
:
2161 return ofpacts_output_to_port(op
->ofpacts
, op
->ofpacts_len
, out_port
);
2167 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2168 * statistics appropriately. 'packet' must have at least sizeof(struct
2169 * ofp10_packet_in) bytes of headroom.
2171 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2172 * with statistics for 'packet' either way.
2174 * Takes ownership of 'packet'. */
2176 rule_execute(struct rule
*rule
, uint16_t in_port
, struct ofpbuf
*packet
)
2180 ovs_assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp10_packet_in
));
2182 flow_extract(packet
, 0, 0, NULL
, in_port
, &flow
);
2183 return rule
->ofproto
->ofproto_class
->rule_execute(rule
, &flow
, packet
);
2186 /* Returns true if 'rule' should be hidden from the controller.
2188 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2189 * (e.g. by in-band control) and are intentionally hidden from the
2192 ofproto_rule_is_hidden(const struct rule
*rule
)
2194 return rule
->cr
.priority
> UINT16_MAX
;
2197 static enum oftable_flags
2198 rule_get_flags(const struct rule
*rule
)
2200 return rule
->ofproto
->tables
[rule
->table_id
].flags
;
2204 rule_is_modifiable(const struct rule
*rule
)
2206 return !(rule_get_flags(rule
) & OFTABLE_READONLY
);
2210 handle_echo_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2212 ofconn_send_reply(ofconn
, make_echo_reply(oh
));
2217 handle_features_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2219 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2220 struct ofputil_switch_features features
;
2221 struct ofport
*port
;
2227 ofproto
->ofproto_class
->get_features(ofproto
, &arp_match_ip
,
2229 ovs_assert(features
.actions
& OFPUTIL_A_OUTPUT
); /* sanity check */
2231 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2232 * if present, are always at the end.) */
2233 n_tables
= ofproto
->n_tables
;
2234 for (i
= 0; i
< ofproto
->n_tables
; i
++) {
2235 if (ofproto
->tables
[i
].flags
& OFTABLE_HIDDEN
) {
2241 features
.datapath_id
= ofproto
->datapath_id
;
2242 features
.n_buffers
= pktbuf_capacity();
2243 features
.n_tables
= n_tables
;
2244 features
.capabilities
= (OFPUTIL_C_FLOW_STATS
| OFPUTIL_C_TABLE_STATS
|
2245 OFPUTIL_C_PORT_STATS
| OFPUTIL_C_QUEUE_STATS
);
2247 features
.capabilities
|= OFPUTIL_C_ARP_MATCH_IP
;
2249 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2250 features
.auxiliary_id
= 0;
2251 b
= ofputil_encode_switch_features(&features
, ofconn_get_protocol(ofconn
),
2253 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
2254 ofputil_put_switch_features_port(&port
->pp
, b
);
2257 ofconn_send_reply(ofconn
, b
);
2262 handle_get_config_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2264 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2265 struct ofp_switch_config
*osc
;
2266 enum ofp_config_flags flags
;
2270 buf
= ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY
, oh
, 0);
2271 osc
= ofpbuf_put_uninit(buf
, sizeof *osc
);
2272 flags
= ofproto
->frag_handling
;
2273 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2274 if (oh
->version
< OFP13_VERSION
2275 && ofconn_get_invalid_ttl_to_controller(ofconn
)) {
2276 flags
|= OFPC_INVALID_TTL_TO_CONTROLLER
;
2278 osc
->flags
= htons(flags
);
2279 osc
->miss_send_len
= htons(ofconn_get_miss_send_len(ofconn
));
2280 ofconn_send_reply(ofconn
, buf
);
2286 handle_set_config(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2288 const struct ofp_switch_config
*osc
= ofpmsg_body(oh
);
2289 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2290 uint16_t flags
= ntohs(osc
->flags
);
2292 if (ofconn_get_type(ofconn
) != OFCONN_PRIMARY
2293 || ofconn_get_role(ofconn
) != OFPCR12_ROLE_SLAVE
) {
2294 enum ofp_config_flags cur
= ofproto
->frag_handling
;
2295 enum ofp_config_flags next
= flags
& OFPC_FRAG_MASK
;
2297 ovs_assert((cur
& OFPC_FRAG_MASK
) == cur
);
2299 if (ofproto
->ofproto_class
->set_frag_handling(ofproto
, next
)) {
2300 ofproto
->frag_handling
= next
;
2302 VLOG_WARN_RL(&rl
, "%s: unsupported fragment handling mode %s",
2304 ofputil_frag_handling_to_string(next
));
2308 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2309 ofconn_set_invalid_ttl_to_controller(ofconn
,
2310 (oh
->version
< OFP13_VERSION
2311 && flags
& OFPC_INVALID_TTL_TO_CONTROLLER
));
2313 ofconn_set_miss_send_len(ofconn
, ntohs(osc
->miss_send_len
));
2318 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2319 * error message code for the caller to propagate upward. Otherwise, returns
2322 * The log message mentions 'msg_type'. */
2324 reject_slave_controller(struct ofconn
*ofconn
)
2326 if (ofconn_get_type(ofconn
) == OFCONN_PRIMARY
2327 && ofconn_get_role(ofconn
) == OFPCR12_ROLE_SLAVE
) {
2328 return OFPERR_OFPBRC_EPERM
;
2335 handle_packet_out(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2337 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2338 struct ofputil_packet_out po
;
2339 struct ofpbuf
*payload
;
2340 uint64_t ofpacts_stub
[1024 / 8];
2341 struct ofpbuf ofpacts
;
2345 COVERAGE_INC(ofproto_packet_out
);
2347 error
= reject_slave_controller(ofconn
);
2352 /* Decode message. */
2353 ofpbuf_use_stub(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
2354 error
= ofputil_decode_packet_out(&po
, oh
, &ofpacts
);
2356 goto exit_free_ofpacts
;
2358 if (po
.in_port
>= p
->max_ports
&& po
.in_port
< OFPP_MAX
) {
2359 error
= OFPERR_OFPBRC_BAD_PORT
;
2360 goto exit_free_ofpacts
;
2365 if (po
.buffer_id
!= UINT32_MAX
) {
2366 error
= ofconn_pktbuf_retrieve(ofconn
, po
.buffer_id
, &payload
, NULL
);
2367 if (error
|| !payload
) {
2368 goto exit_free_ofpacts
;
2371 payload
= xmalloc(sizeof *payload
);
2372 ofpbuf_use_const(payload
, po
.packet
, po
.packet_len
);
2375 /* Verify actions against packet, then send packet if successful. */
2376 flow_extract(payload
, 0, 0, NULL
, po
.in_port
, &flow
);
2377 error
= ofpacts_check(po
.ofpacts
, po
.ofpacts_len
, &flow
, p
->max_ports
);
2379 error
= p
->ofproto_class
->packet_out(p
, payload
, &flow
,
2380 po
.ofpacts
, po
.ofpacts_len
);
2382 ofpbuf_delete(payload
);
2385 ofpbuf_uninit(&ofpacts
);
2391 update_port_config(struct ofport
*port
,
2392 enum ofputil_port_config config
,
2393 enum ofputil_port_config mask
)
2395 enum ofputil_port_config old_config
= port
->pp
.config
;
2396 enum ofputil_port_config toggle
;
2398 toggle
= (config
^ port
->pp
.config
) & mask
;
2399 if (toggle
& OFPUTIL_PC_PORT_DOWN
) {
2400 if (config
& OFPUTIL_PC_PORT_DOWN
) {
2401 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, NULL
);
2403 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, NULL
);
2405 toggle
&= ~OFPUTIL_PC_PORT_DOWN
;
2408 port
->pp
.config
^= toggle
;
2409 if (port
->pp
.config
!= old_config
) {
2410 port
->ofproto
->ofproto_class
->port_reconfigured(port
, old_config
);
2415 handle_port_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2417 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2418 struct ofputil_port_mod pm
;
2419 struct ofport
*port
;
2422 error
= reject_slave_controller(ofconn
);
2427 error
= ofputil_decode_port_mod(oh
, &pm
);
2432 port
= ofproto_get_port(p
, pm
.port_no
);
2434 return OFPERR_OFPPMFC_BAD_PORT
;
2435 } else if (!eth_addr_equals(port
->pp
.hw_addr
, pm
.hw_addr
)) {
2436 return OFPERR_OFPPMFC_BAD_HW_ADDR
;
2438 update_port_config(port
, pm
.config
, pm
.mask
);
2440 netdev_set_advertisements(port
->netdev
, pm
.advertise
);
2447 handle_desc_stats_request(struct ofconn
*ofconn
,
2448 const struct ofp_header
*request
)
2450 static const char *default_mfr_desc
= "Nicira, Inc.";
2451 static const char *default_hw_desc
= "Open vSwitch";
2452 static const char *default_sw_desc
= VERSION
;
2453 static const char *default_serial_desc
= "None";
2454 static const char *default_dp_desc
= "None";
2456 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2457 struct ofp_desc_stats
*ods
;
2460 msg
= ofpraw_alloc_stats_reply(request
, 0);
2461 ods
= ofpbuf_put_zeros(msg
, sizeof *ods
);
2462 ovs_strlcpy(ods
->mfr_desc
, p
->mfr_desc
? p
->mfr_desc
: default_mfr_desc
,
2463 sizeof ods
->mfr_desc
);
2464 ovs_strlcpy(ods
->hw_desc
, p
->hw_desc
? p
->hw_desc
: default_hw_desc
,
2465 sizeof ods
->hw_desc
);
2466 ovs_strlcpy(ods
->sw_desc
, p
->sw_desc
? p
->sw_desc
: default_sw_desc
,
2467 sizeof ods
->sw_desc
);
2468 ovs_strlcpy(ods
->serial_num
,
2469 p
->serial_desc
? p
->serial_desc
: default_serial_desc
,
2470 sizeof ods
->serial_num
);
2471 ovs_strlcpy(ods
->dp_desc
, p
->dp_desc
? p
->dp_desc
: default_dp_desc
,
2472 sizeof ods
->dp_desc
);
2473 ofconn_send_reply(ofconn
, msg
);
2479 handle_table_stats_request(struct ofconn
*ofconn
,
2480 const struct ofp_header
*request
)
2482 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2483 struct ofp12_table_stats
*ots
;
2488 /* Set up default values.
2490 * ofp12_table_stats is used as a generic structure as
2491 * it is able to hold all the fields for ofp10_table_stats
2492 * and ofp11_table_stats (and of course itself).
2494 ots
= xcalloc(p
->n_tables
, sizeof *ots
);
2495 for (i
= 0; i
< p
->n_tables
; i
++) {
2496 ots
[i
].table_id
= i
;
2497 sprintf(ots
[i
].name
, "table%zu", i
);
2498 ots
[i
].match
= htonll(OFPXMT12_MASK
);
2499 ots
[i
].wildcards
= htonll(OFPXMT12_MASK
);
2500 ots
[i
].write_actions
= htonl(OFPAT11_OUTPUT
);
2501 ots
[i
].apply_actions
= htonl(OFPAT11_OUTPUT
);
2502 ots
[i
].write_setfields
= htonll(OFPXMT12_MASK
);
2503 ots
[i
].apply_setfields
= htonll(OFPXMT12_MASK
);
2504 ots
[i
].metadata_match
= htonll(UINT64_MAX
);
2505 ots
[i
].metadata_write
= htonll(UINT64_MAX
);
2506 ots
[i
].instructions
= htonl(OFPIT11_ALL
);
2507 ots
[i
].config
= htonl(OFPTC11_TABLE_MISS_MASK
);
2508 ots
[i
].max_entries
= htonl(1000000); /* An arbitrary big number. */
2509 ots
[i
].active_count
= htonl(classifier_count(&p
->tables
[i
].cls
));
2512 p
->ofproto_class
->get_tables(p
, ots
);
2514 /* Post-process the tables, dropping hidden tables. */
2515 n_tables
= p
->n_tables
;
2516 for (i
= 0; i
< p
->n_tables
; i
++) {
2517 const struct oftable
*table
= &p
->tables
[i
];
2519 if (table
->flags
& OFTABLE_HIDDEN
) {
2525 ovs_strzcpy(ots
[i
].name
, table
->name
, sizeof ots
[i
].name
);
2528 if (table
->max_flows
< ntohl(ots
[i
].max_entries
)) {
2529 ots
[i
].max_entries
= htonl(table
->max_flows
);
2533 msg
= ofputil_encode_table_stats_reply(ots
, n_tables
, request
);
2534 ofconn_send_reply(ofconn
, msg
);
2542 append_port_stat(struct ofport
*port
, struct list
*replies
)
2544 struct ofputil_port_stats ops
= { .port_no
= port
->pp
.port_no
};
2546 /* Intentionally ignore return value, since errors will set
2547 * 'stats' to all-1s, which is correct for OpenFlow, and
2548 * netdev_get_stats() will log errors. */
2549 ofproto_port_get_stats(port
, &ops
.stats
);
2551 ofputil_append_port_stat(replies
, &ops
);
2555 handle_port_stats_request(struct ofconn
*ofconn
,
2556 const struct ofp_header
*request
)
2558 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2559 struct ofport
*port
;
2560 struct list replies
;
2564 error
= ofputil_decode_port_stats_request(request
, &port_no
);
2569 ofpmp_init(&replies
, request
);
2570 if (port_no
!= OFPP_ANY
) {
2571 port
= ofproto_get_port(p
, port_no
);
2573 append_port_stat(port
, &replies
);
2576 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
2577 append_port_stat(port
, &replies
);
2581 ofconn_send_replies(ofconn
, &replies
);
2586 handle_port_desc_stats_request(struct ofconn
*ofconn
,
2587 const struct ofp_header
*request
)
2589 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2590 enum ofp_version version
;
2591 struct ofport
*port
;
2592 struct list replies
;
2594 ofpmp_init(&replies
, request
);
2596 version
= ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn
));
2597 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
2598 ofputil_append_port_desc_stats_reply(version
, &port
->pp
, &replies
);
2601 ofconn_send_replies(ofconn
, &replies
);
2606 calc_flow_duration__(long long int start
, long long int now
,
2607 uint32_t *sec
, uint32_t *nsec
)
2609 long long int msecs
= now
- start
;
2610 *sec
= msecs
/ 1000;
2611 *nsec
= (msecs
% 1000) * (1000 * 1000);
2614 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2615 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2617 check_table_id(const struct ofproto
*ofproto
, uint8_t table_id
)
2619 return (table_id
== 0xff || table_id
< ofproto
->n_tables
2621 : OFPERR_OFPBRC_BAD_TABLE_ID
);
2625 static struct oftable
*
2626 next_visible_table(const struct ofproto
*ofproto
, uint8_t table_id
)
2628 struct oftable
*table
;
2630 for (table
= &ofproto
->tables
[table_id
];
2631 table
< &ofproto
->tables
[ofproto
->n_tables
];
2633 if (!(table
->flags
& OFTABLE_HIDDEN
)) {
2641 static struct oftable
*
2642 first_matching_table(const struct ofproto
*ofproto
, uint8_t table_id
)
2644 if (table_id
== 0xff) {
2645 return next_visible_table(ofproto
, 0);
2646 } else if (table_id
< ofproto
->n_tables
) {
2647 return &ofproto
->tables
[table_id
];
2653 static struct oftable
*
2654 next_matching_table(const struct ofproto
*ofproto
,
2655 const struct oftable
*table
, uint8_t table_id
)
2657 return (table_id
== 0xff
2658 ? next_visible_table(ofproto
, (table
- ofproto
->tables
) + 1)
2662 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2664 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2665 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2667 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2668 * only once, for that table. (This can be used to access tables marked
2671 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2672 * entered at all. (Perhaps you should have validated TABLE_ID with
2673 * check_table_id().)
2675 * All parameters are evaluated multiple times.
2677 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2678 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2680 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2682 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2683 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2684 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2687 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2688 * to 'out_port' are included.
2690 * Hidden rules are always omitted.
2692 * Returns 0 on success, otherwise an OpenFlow error code. */
2694 collect_rules_loose(struct ofproto
*ofproto
, uint8_t table_id
,
2695 const struct match
*match
,
2696 ovs_be64 cookie
, ovs_be64 cookie_mask
,
2697 uint16_t out_port
, struct list
*rules
)
2699 struct oftable
*table
;
2703 error
= check_table_id(ofproto
, table_id
);
2709 cls_rule_init(&cr
, match
, 0);
2710 FOR_EACH_MATCHING_TABLE (table
, table_id
, ofproto
) {
2711 struct cls_cursor cursor
;
2714 cls_cursor_init(&cursor
, &table
->cls
, &cr
);
2715 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
2716 if (rule
->pending
) {
2717 error
= OFPROTO_POSTPONE
;
2720 if (!ofproto_rule_is_hidden(rule
)
2721 && ofproto_rule_has_out_port(rule
, out_port
)
2722 && !((rule
->flow_cookie
^ cookie
) & cookie_mask
)) {
2723 list_push_back(rules
, &rule
->ofproto_node
);
2729 cls_rule_destroy(&cr
);
2733 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2734 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2735 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2738 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2739 * to 'out_port' are included.
2741 * Hidden rules are always omitted.
2743 * Returns 0 on success, otherwise an OpenFlow error code. */
2745 collect_rules_strict(struct ofproto
*ofproto
, uint8_t table_id
,
2746 const struct match
*match
, unsigned int priority
,
2747 ovs_be64 cookie
, ovs_be64 cookie_mask
,
2748 uint16_t out_port
, struct list
*rules
)
2750 struct oftable
*table
;
2754 error
= check_table_id(ofproto
, table_id
);
2760 cls_rule_init(&cr
, match
, priority
);
2761 FOR_EACH_MATCHING_TABLE (table
, table_id
, ofproto
) {
2764 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&table
->cls
,
2767 if (rule
->pending
) {
2768 error
= OFPROTO_POSTPONE
;
2771 if (!ofproto_rule_is_hidden(rule
)
2772 && ofproto_rule_has_out_port(rule
, out_port
)
2773 && !((rule
->flow_cookie
^ cookie
) & cookie_mask
)) {
2774 list_push_back(rules
, &rule
->ofproto_node
);
2780 cls_rule_destroy(&cr
);
2784 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2785 * forced into the range of a uint16_t. */
2787 age_secs(long long int age_ms
)
2789 return (age_ms
< 0 ? 0
2790 : age_ms
>= UINT16_MAX
* 1000 ? UINT16_MAX
2791 : (unsigned int) age_ms
/ 1000);
2795 handle_flow_stats_request(struct ofconn
*ofconn
,
2796 const struct ofp_header
*request
)
2798 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2799 struct ofputil_flow_stats_request fsr
;
2800 struct list replies
;
2805 error
= ofputil_decode_flow_stats_request(&fsr
, request
);
2810 error
= collect_rules_loose(ofproto
, fsr
.table_id
, &fsr
.match
,
2811 fsr
.cookie
, fsr
.cookie_mask
,
2812 fsr
.out_port
, &rules
);
2817 ofpmp_init(&replies
, request
);
2818 LIST_FOR_EACH (rule
, ofproto_node
, &rules
) {
2819 long long int now
= time_msec();
2820 struct ofputil_flow_stats fs
;
2822 minimatch_expand(&rule
->cr
.match
, &fs
.match
);
2823 fs
.priority
= rule
->cr
.priority
;
2824 fs
.cookie
= rule
->flow_cookie
;
2825 fs
.table_id
= rule
->table_id
;
2826 calc_flow_duration__(rule
->created
, now
, &fs
.duration_sec
,
2828 fs
.idle_timeout
= rule
->idle_timeout
;
2829 fs
.hard_timeout
= rule
->hard_timeout
;
2830 fs
.idle_age
= age_secs(now
- rule
->used
);
2831 fs
.hard_age
= age_secs(now
- rule
->modified
);
2832 ofproto
->ofproto_class
->rule_get_stats(rule
, &fs
.packet_count
,
2834 fs
.ofpacts
= rule
->ofpacts
;
2835 fs
.ofpacts_len
= rule
->ofpacts_len
;
2837 if (rule
->send_flow_removed
) {
2838 fs
.flags
|= OFPFF_SEND_FLOW_REM
;
2839 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
2840 and OFPFF13_NO_BYT_COUNTS */
2842 ofputil_append_flow_stats_reply(&fs
, &replies
);
2844 ofconn_send_replies(ofconn
, &replies
);
2850 flow_stats_ds(struct rule
*rule
, struct ds
*results
)
2852 uint64_t packet_count
, byte_count
;
2854 rule
->ofproto
->ofproto_class
->rule_get_stats(rule
,
2855 &packet_count
, &byte_count
);
2857 if (rule
->table_id
!= 0) {
2858 ds_put_format(results
, "table_id=%"PRIu8
", ", rule
->table_id
);
2860 ds_put_format(results
, "duration=%llds, ",
2861 (time_msec() - rule
->created
) / 1000);
2862 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
2863 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
2864 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
2865 cls_rule_format(&rule
->cr
, results
);
2866 ds_put_char(results
, ',');
2867 ofpacts_format(rule
->ofpacts
, rule
->ofpacts_len
, results
);
2868 ds_put_cstr(results
, "\n");
2871 /* Adds a pretty-printed description of all flows to 'results', including
2872 * hidden flows (e.g., set up by in-band control). */
2874 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
2876 struct oftable
*table
;
2878 OFPROTO_FOR_EACH_TABLE (table
, p
) {
2879 struct cls_cursor cursor
;
2882 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
2883 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
2884 flow_stats_ds(rule
, results
);
2889 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2890 * '*engine_type' and '*engine_id', respectively. */
2892 ofproto_get_netflow_ids(const struct ofproto
*ofproto
,
2893 uint8_t *engine_type
, uint8_t *engine_id
)
2895 ofproto
->ofproto_class
->get_netflow_ids(ofproto
, engine_type
, engine_id
);
2898 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
2899 * true if the port's CFM status was successfully stored into '*status'.
2900 * Returns false if the port did not have CFM configured, in which case
2901 * '*status' is indeterminate.
2903 * The caller must provide and owns '*status', but it does not own and must not
2904 * modify or free the array returned in 'status->rmps'. */
2906 ofproto_port_get_cfm_status(const struct ofproto
*ofproto
, uint16_t ofp_port
,
2907 struct ofproto_cfm_status
*status
)
2909 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
2911 && ofproto
->ofproto_class
->get_cfm_status
2912 && ofproto
->ofproto_class
->get_cfm_status(ofport
, status
));
2916 handle_aggregate_stats_request(struct ofconn
*ofconn
,
2917 const struct ofp_header
*oh
)
2919 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2920 struct ofputil_flow_stats_request request
;
2921 struct ofputil_aggregate_stats stats
;
2922 bool unknown_packets
, unknown_bytes
;
2923 struct ofpbuf
*reply
;
2928 error
= ofputil_decode_flow_stats_request(&request
, oh
);
2933 error
= collect_rules_loose(ofproto
, request
.table_id
, &request
.match
,
2934 request
.cookie
, request
.cookie_mask
,
2935 request
.out_port
, &rules
);
2940 memset(&stats
, 0, sizeof stats
);
2941 unknown_packets
= unknown_bytes
= false;
2942 LIST_FOR_EACH (rule
, ofproto_node
, &rules
) {
2943 uint64_t packet_count
;
2944 uint64_t byte_count
;
2946 ofproto
->ofproto_class
->rule_get_stats(rule
, &packet_count
,
2949 if (packet_count
== UINT64_MAX
) {
2950 unknown_packets
= true;
2952 stats
.packet_count
+= packet_count
;
2955 if (byte_count
== UINT64_MAX
) {
2956 unknown_bytes
= true;
2958 stats
.byte_count
+= byte_count
;
2963 if (unknown_packets
) {
2964 stats
.packet_count
= UINT64_MAX
;
2966 if (unknown_bytes
) {
2967 stats
.byte_count
= UINT64_MAX
;
2970 reply
= ofputil_encode_aggregate_stats_reply(&stats
, oh
);
2971 ofconn_send_reply(ofconn
, reply
);
2976 struct queue_stats_cbdata
{
2977 struct ofport
*ofport
;
2978 struct list replies
;
2982 put_queue_stats(struct queue_stats_cbdata
*cbdata
, uint32_t queue_id
,
2983 const struct netdev_queue_stats
*stats
)
2986 struct ofputil_queue_stats oqs
= {
2987 .port_no
= cbdata
->ofport
->pp
.port_no
,
2988 .queue_id
= queue_id
,
2991 ofputil_append_queue_stat(&cbdata
->replies
, &oqs
);
2995 handle_queue_stats_dump_cb(uint32_t queue_id
,
2996 struct netdev_queue_stats
*stats
,
2999 struct queue_stats_cbdata
*cbdata
= cbdata_
;
3001 put_queue_stats(cbdata
, queue_id
, stats
);
3005 handle_queue_stats_for_port(struct ofport
*port
, uint32_t queue_id
,
3006 struct queue_stats_cbdata
*cbdata
)
3008 cbdata
->ofport
= port
;
3009 if (queue_id
== OFPQ_ALL
) {
3010 netdev_dump_queue_stats(port
->netdev
,
3011 handle_queue_stats_dump_cb
, cbdata
);
3013 struct netdev_queue_stats stats
;
3015 if (!netdev_get_queue_stats(port
->netdev
, queue_id
, &stats
)) {
3016 put_queue_stats(cbdata
, queue_id
, &stats
);
3018 return OFPERR_OFPQOFC_BAD_QUEUE
;
3025 handle_queue_stats_request(struct ofconn
*ofconn
,
3026 const struct ofp_header
*rq
)
3028 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
3029 struct queue_stats_cbdata cbdata
;
3030 struct ofport
*port
;
3032 struct ofputil_queue_stats_request oqsr
;
3034 COVERAGE_INC(ofproto_queue_req
);
3036 ofpmp_init(&cbdata
.replies
, rq
);
3038 error
= ofputil_decode_queue_stats_request(rq
, &oqsr
);
3043 if (oqsr
.port_no
== OFPP_ANY
) {
3044 error
= OFPERR_OFPQOFC_BAD_QUEUE
;
3045 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
3046 if (!handle_queue_stats_for_port(port
, oqsr
.queue_id
, &cbdata
)) {
3051 port
= ofproto_get_port(ofproto
, oqsr
.port_no
);
3053 ? handle_queue_stats_for_port(port
, oqsr
.queue_id
, &cbdata
)
3054 : OFPERR_OFPQOFC_BAD_PORT
);
3057 ofconn_send_replies(ofconn
, &cbdata
.replies
);
3059 ofpbuf_list_delete(&cbdata
.replies
);
3066 is_flow_deletion_pending(const struct ofproto
*ofproto
,
3067 const struct cls_rule
*cls_rule
,
3070 if (!hmap_is_empty(&ofproto
->deletions
)) {
3071 struct ofoperation
*op
;
3073 HMAP_FOR_EACH_WITH_HASH (op
, hmap_node
,
3074 cls_rule_hash(cls_rule
, table_id
),
3075 &ofproto
->deletions
) {
3076 if (cls_rule_equal(cls_rule
, &op
->rule
->cr
)) {
3085 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3086 * in which no matching flow already exists in the flow table.
3088 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3089 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3090 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3091 * initiated now but may be retried later.
3093 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3094 * ownership remains with the caller.
3096 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3099 add_flow(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3100 const struct ofputil_flow_mod
*fm
, const struct ofp_header
*request
)
3102 struct oftable
*table
;
3103 struct ofopgroup
*group
;
3104 struct rule
*victim
;
3109 error
= check_table_id(ofproto
, fm
->table_id
);
3115 if (fm
->table_id
== 0xff) {
3117 if (ofproto
->ofproto_class
->rule_choose_table
) {
3118 error
= ofproto
->ofproto_class
->rule_choose_table(ofproto
,
3124 ovs_assert(table_id
< ofproto
->n_tables
);
3125 table
= &ofproto
->tables
[table_id
];
3127 table
= &ofproto
->tables
[0];
3129 } else if (fm
->table_id
< ofproto
->n_tables
) {
3130 table
= &ofproto
->tables
[fm
->table_id
];
3132 return OFPERR_OFPBRC_BAD_TABLE_ID
;
3135 if (table
->flags
& OFTABLE_READONLY
) {
3136 return OFPERR_OFPBRC_EPERM
;
3139 /* Allocate new rule and initialize classifier rule. */
3140 rule
= ofproto
->ofproto_class
->rule_alloc();
3142 VLOG_WARN_RL(&rl
, "%s: failed to create rule (%s)",
3143 ofproto
->name
, strerror(error
));
3146 cls_rule_init(&rule
->cr
, &fm
->match
, fm
->priority
);
3148 /* Serialize against pending deletion. */
3149 if (is_flow_deletion_pending(ofproto
, &cr
, table
- ofproto
->tables
)) {
3150 cls_rule_destroy(&rule
->cr
);
3151 ofproto
->ofproto_class
->rule_dealloc(rule
);
3152 return OFPROTO_POSTPONE
;
3155 /* Check for overlap, if requested. */
3156 if (fm
->flags
& OFPFF_CHECK_OVERLAP
3157 && classifier_rule_overlaps(&table
->cls
, &rule
->cr
)) {
3158 cls_rule_destroy(&rule
->cr
);
3159 ofproto
->ofproto_class
->rule_dealloc(rule
);
3160 return OFPERR_OFPFMFC_OVERLAP
;
3163 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3165 rule
->ofproto
= ofproto
;
3166 rule
->pending
= NULL
;
3167 rule
->flow_cookie
= fm
->new_cookie
;
3168 rule
->created
= rule
->modified
= rule
->used
= time_msec();
3169 rule
->idle_timeout
= fm
->idle_timeout
;
3170 rule
->hard_timeout
= fm
->hard_timeout
;
3171 rule
->table_id
= table
- ofproto
->tables
;
3172 rule
->send_flow_removed
= (fm
->flags
& OFPFF_SEND_FLOW_REM
) != 0;
3173 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3174 and OFPFF13_NO_BYT_COUNTS */
3175 rule
->ofpacts
= xmemdup(fm
->ofpacts
, fm
->ofpacts_len
);
3176 rule
->ofpacts_len
= fm
->ofpacts_len
;
3177 rule
->evictable
= true;
3178 rule
->eviction_group
= NULL
;
3179 list_init(&rule
->expirable
);
3180 rule
->monitor_flags
= 0;
3181 rule
->add_seqno
= 0;
3182 rule
->modify_seqno
= 0;
3184 /* Insert new rule. */
3185 victim
= oftable_replace_rule(rule
);
3186 if (victim
&& !rule_is_modifiable(victim
)) {
3187 error
= OFPERR_OFPBRC_EPERM
;
3188 } else if (victim
&& victim
->pending
) {
3189 error
= OFPROTO_POSTPONE
;
3191 struct ofoperation
*op
;
3194 if (classifier_count(&table
->cls
) > table
->max_flows
) {
3197 was_evictable
= rule
->evictable
;
3198 rule
->evictable
= false;
3199 evict
= choose_rule_to_evict(table
);
3200 rule
->evictable
= was_evictable
;
3203 error
= OFPERR_OFPFMFC_TABLE_FULL
;
3205 } else if (evict
->pending
) {
3206 error
= OFPROTO_POSTPONE
;
3213 group
= ofopgroup_create(ofproto
, ofconn
, request
, fm
->buffer_id
);
3214 op
= ofoperation_create(group
, rule
, OFOPERATION_ADD
, 0);
3215 op
->victim
= victim
;
3217 error
= ofproto
->ofproto_class
->rule_construct(rule
);
3219 op
->group
->n_running
--;
3220 ofoperation_destroy(rule
->pending
);
3222 delete_flow__(evict
, group
);
3224 ofopgroup_submit(group
);
3228 /* Back out if an error occurred. */
3230 oftable_substitute_rule(rule
, victim
);
3231 ofproto_rule_destroy__(rule
);
3236 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3238 /* Modifies the rules listed in 'rules', changing their actions to match those
3241 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3244 * Returns 0 on success, otherwise an OpenFlow error code. */
3246 modify_flows__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3247 const struct ofputil_flow_mod
*fm
,
3248 const struct ofp_header
*request
, struct list
*rules
)
3250 struct ofopgroup
*group
;
3254 group
= ofopgroup_create(ofproto
, ofconn
, request
, fm
->buffer_id
);
3255 error
= OFPERR_OFPBRC_EPERM
;
3256 LIST_FOR_EACH (rule
, ofproto_node
, rules
) {
3257 struct ofoperation
*op
;
3258 bool actions_changed
;
3259 ovs_be64 new_cookie
;
3261 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3263 if (rule_is_modifiable(rule
)) {
3264 /* At least one rule is modifiable, don't report EPERM error. */
3270 actions_changed
= !ofpacts_equal(fm
->ofpacts
, fm
->ofpacts_len
,
3271 rule
->ofpacts
, rule
->ofpacts_len
);
3272 new_cookie
= (fm
->new_cookie
!= htonll(UINT64_MAX
)
3274 : rule
->flow_cookie
);
3276 op
= ofoperation_create(group
, rule
, OFOPERATION_MODIFY
, 0);
3277 rule
->flow_cookie
= new_cookie
;
3278 if (actions_changed
) {
3279 op
->ofpacts
= rule
->ofpacts
;
3280 op
->ofpacts_len
= rule
->ofpacts_len
;
3281 rule
->ofpacts
= xmemdup(fm
->ofpacts
, fm
->ofpacts_len
);
3282 rule
->ofpacts_len
= fm
->ofpacts_len
;
3283 rule
->ofproto
->ofproto_class
->rule_modify_actions(rule
);
3285 ofoperation_complete(op
, 0);
3288 ofopgroup_submit(group
);
3294 modify_flows_add(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3295 const struct ofputil_flow_mod
*fm
,
3296 const struct ofp_header
*request
)
3298 if (fm
->cookie_mask
!= htonll(0) || fm
->new_cookie
== htonll(UINT64_MAX
)) {
3301 return add_flow(ofproto
, ofconn
, fm
, request
);
3304 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3307 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3310 modify_flows_loose(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3311 const struct ofputil_flow_mod
*fm
,
3312 const struct ofp_header
*request
)
3317 error
= collect_rules_loose(ofproto
, fm
->table_id
, &fm
->match
,
3318 fm
->cookie
, fm
->cookie_mask
,
3322 } else if (list_is_empty(&rules
)) {
3323 return modify_flows_add(ofproto
, ofconn
, fm
, request
);
3325 return modify_flows__(ofproto
, ofconn
, fm
, request
, &rules
);
3329 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3332 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3335 modify_flow_strict(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3336 const struct ofputil_flow_mod
*fm
,
3337 const struct ofp_header
*request
)
3342 error
= collect_rules_strict(ofproto
, fm
->table_id
, &fm
->match
,
3343 fm
->priority
, fm
->cookie
, fm
->cookie_mask
,
3348 } else if (list_is_empty(&rules
)) {
3349 return modify_flows_add(ofproto
, ofconn
, fm
, request
);
3351 return list_is_singleton(&rules
) ? modify_flows__(ofproto
, ofconn
,
3352 fm
, request
, &rules
)
3357 /* OFPFC_DELETE implementation. */
3360 delete_flow__(struct rule
*rule
, struct ofopgroup
*group
)
3362 struct ofproto
*ofproto
= rule
->ofproto
;
3364 ofproto_rule_send_removed(rule
, OFPRR_DELETE
);
3366 ofoperation_create(group
, rule
, OFOPERATION_DELETE
, OFPRR_DELETE
);
3367 oftable_remove_rule(rule
);
3368 ofproto
->ofproto_class
->rule_destruct(rule
);
3371 /* Deletes the rules listed in 'rules'.
3373 * Returns 0 on success, otherwise an OpenFlow error code. */
3375 delete_flows__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3376 const struct ofp_header
*request
, struct list
*rules
)
3378 struct rule
*rule
, *next
;
3379 struct ofopgroup
*group
;
3381 group
= ofopgroup_create(ofproto
, ofconn
, request
, UINT32_MAX
);
3382 LIST_FOR_EACH_SAFE (rule
, next
, ofproto_node
, rules
) {
3383 delete_flow__(rule
, group
);
3385 ofopgroup_submit(group
);
3390 /* Implements OFPFC_DELETE. */
3392 delete_flows_loose(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3393 const struct ofputil_flow_mod
*fm
,
3394 const struct ofp_header
*request
)
3399 error
= collect_rules_loose(ofproto
, fm
->table_id
, &fm
->match
,
3400 fm
->cookie
, fm
->cookie_mask
,
3401 fm
->out_port
, &rules
);
3402 return (error
? error
3403 : !list_is_empty(&rules
) ? delete_flows__(ofproto
, ofconn
, request
,
3408 /* Implements OFPFC_DELETE_STRICT. */
3410 delete_flow_strict(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3411 const struct ofputil_flow_mod
*fm
,
3412 const struct ofp_header
*request
)
3417 error
= collect_rules_strict(ofproto
, fm
->table_id
, &fm
->match
,
3418 fm
->priority
, fm
->cookie
, fm
->cookie_mask
,
3419 fm
->out_port
, &rules
);
3420 return (error
? error
3421 : list_is_singleton(&rules
) ? delete_flows__(ofproto
, ofconn
,
3427 ofproto_rule_send_removed(struct rule
*rule
, uint8_t reason
)
3429 struct ofputil_flow_removed fr
;
3431 if (ofproto_rule_is_hidden(rule
) || !rule
->send_flow_removed
) {
3435 minimatch_expand(&rule
->cr
.match
, &fr
.match
);
3436 fr
.priority
= rule
->cr
.priority
;
3437 fr
.cookie
= rule
->flow_cookie
;
3439 fr
.table_id
= rule
->table_id
;
3440 calc_flow_duration__(rule
->created
, time_msec(),
3441 &fr
.duration_sec
, &fr
.duration_nsec
);
3442 fr
.idle_timeout
= rule
->idle_timeout
;
3443 fr
.hard_timeout
= rule
->hard_timeout
;
3444 rule
->ofproto
->ofproto_class
->rule_get_stats(rule
, &fr
.packet_count
,
3447 connmgr_send_flow_removed(rule
->ofproto
->connmgr
, &fr
);
3451 ofproto_rule_update_used(struct rule
*rule
, long long int used
)
3453 if (used
> rule
->used
) {
3454 struct eviction_group
*evg
= rule
->eviction_group
;
3458 heap_change(&evg
->rules
, &rule
->evg_node
,
3459 rule_eviction_priority(rule
));
3464 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3465 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3468 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3471 * ofproto implementation ->run() functions should use this function to expire
3472 * OpenFlow flows. */
3474 ofproto_rule_expire(struct rule
*rule
, uint8_t reason
)
3476 struct ofproto
*ofproto
= rule
->ofproto
;
3477 struct ofopgroup
*group
;
3479 ovs_assert(reason
== OFPRR_HARD_TIMEOUT
|| reason
== OFPRR_IDLE_TIMEOUT
);
3481 ofproto_rule_send_removed(rule
, reason
);
3483 group
= ofopgroup_create_unattached(ofproto
);
3484 ofoperation_create(group
, rule
, OFOPERATION_DELETE
, reason
);
3485 oftable_remove_rule(rule
);
3486 ofproto
->ofproto_class
->rule_destruct(rule
);
3487 ofopgroup_submit(group
);
3491 handle_flow_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3493 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
3494 struct ofputil_flow_mod fm
;
3495 uint64_t ofpacts_stub
[1024 / 8];
3496 struct ofpbuf ofpacts
;
3500 error
= reject_slave_controller(ofconn
);
3505 ofpbuf_use_stub(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
3506 error
= ofputil_decode_flow_mod(&fm
, oh
, ofconn_get_protocol(ofconn
),
3509 error
= ofpacts_check(fm
.ofpacts
, fm
.ofpacts_len
,
3510 &fm
.match
.flow
, ofproto
->max_ports
);
3513 error
= handle_flow_mod__(ofproto
, ofconn
, &fm
, oh
);
3516 goto exit_free_ofpacts
;
3519 /* Record the operation for logging a summary report. */
3520 switch (fm
.command
) {
3526 case OFPFC_MODIFY_STRICT
:
3527 ofproto
->n_modify
++;
3531 case OFPFC_DELETE_STRICT
:
3532 ofproto
->n_delete
++;
3537 if (ofproto
->next_op_report
== LLONG_MAX
) {
3538 ofproto
->first_op
= now
;
3539 ofproto
->next_op_report
= MAX(now
+ 10 * 1000,
3540 ofproto
->op_backoff
);
3541 ofproto
->op_backoff
= ofproto
->next_op_report
+ 60 * 1000;
3543 ofproto
->last_op
= now
;
3546 ofpbuf_uninit(&ofpacts
);
3552 handle_flow_mod__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3553 const struct ofputil_flow_mod
*fm
,
3554 const struct ofp_header
*oh
)
3556 if (ofproto
->n_pending
>= 50) {
3557 ovs_assert(!list_is_empty(&ofproto
->pending
));
3558 return OFPROTO_POSTPONE
;
3561 switch (fm
->command
) {
3563 return add_flow(ofproto
, ofconn
, fm
, oh
);
3566 return modify_flows_loose(ofproto
, ofconn
, fm
, oh
);
3568 case OFPFC_MODIFY_STRICT
:
3569 return modify_flow_strict(ofproto
, ofconn
, fm
, oh
);
3572 return delete_flows_loose(ofproto
, ofconn
, fm
, oh
);
3574 case OFPFC_DELETE_STRICT
:
3575 return delete_flow_strict(ofproto
, ofconn
, fm
, oh
);
3578 if (fm
->command
> 0xff) {
3579 VLOG_WARN_RL(&rl
, "%s: flow_mod has explicit table_id but "
3580 "flow_mod_table_id extension is not enabled",
3583 return OFPERR_OFPFMFC_BAD_COMMAND
;
3588 handle_role_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3590 struct ofputil_role_request request
;
3591 struct ofputil_role_request reply
;
3595 error
= ofputil_decode_role_message(oh
, &request
);
3600 if (request
.role
!= OFPCR12_ROLE_NOCHANGE
) {
3601 if (ofconn_get_role(ofconn
) != request
.role
3602 && ofconn_has_pending_opgroups(ofconn
)) {
3603 return OFPROTO_POSTPONE
;
3606 if (request
.have_generation_id
3607 && !ofconn_set_master_election_id(ofconn
, request
.generation_id
)) {
3608 return OFPERR_OFPRRFC_STALE
;
3611 ofconn_set_role(ofconn
, request
.role
);
3614 reply
.role
= ofconn_get_role(ofconn
);
3615 reply
.have_generation_id
= ofconn_get_master_election_id(
3616 ofconn
, &reply
.generation_id
);
3617 buf
= ofputil_encode_role_reply(oh
, &reply
);
3618 ofconn_send_reply(ofconn
, buf
);
3624 handle_nxt_flow_mod_table_id(struct ofconn
*ofconn
,
3625 const struct ofp_header
*oh
)
3627 const struct nx_flow_mod_table_id
*msg
= ofpmsg_body(oh
);
3628 enum ofputil_protocol cur
, next
;
3630 cur
= ofconn_get_protocol(ofconn
);
3631 next
= ofputil_protocol_set_tid(cur
, msg
->set
!= 0);
3632 ofconn_set_protocol(ofconn
, next
);
3638 handle_nxt_set_flow_format(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3640 const struct nx_set_flow_format
*msg
= ofpmsg_body(oh
);
3641 enum ofputil_protocol cur
, next
;
3642 enum ofputil_protocol next_base
;
3644 next_base
= ofputil_nx_flow_format_to_protocol(ntohl(msg
->format
));
3646 return OFPERR_OFPBRC_EPERM
;
3649 cur
= ofconn_get_protocol(ofconn
);
3650 next
= ofputil_protocol_set_base(cur
, next_base
);
3651 if (cur
!= next
&& ofconn_has_pending_opgroups(ofconn
)) {
3652 /* Avoid sending async messages in surprising protocol. */
3653 return OFPROTO_POSTPONE
;
3656 ofconn_set_protocol(ofconn
, next
);
3661 handle_nxt_set_packet_in_format(struct ofconn
*ofconn
,
3662 const struct ofp_header
*oh
)
3664 const struct nx_set_packet_in_format
*msg
= ofpmsg_body(oh
);
3667 format
= ntohl(msg
->format
);
3668 if (format
!= NXPIF_OPENFLOW10
&& format
!= NXPIF_NXM
) {
3669 return OFPERR_OFPBRC_EPERM
;
3672 if (format
!= ofconn_get_packet_in_format(ofconn
)
3673 && ofconn_has_pending_opgroups(ofconn
)) {
3674 /* Avoid sending async message in surprsing packet in format. */
3675 return OFPROTO_POSTPONE
;
3678 ofconn_set_packet_in_format(ofconn
, format
);
3683 handle_nxt_set_async_config(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3685 const struct nx_async_config
*msg
= ofpmsg_body(oh
);
3686 uint32_t master
[OAM_N_TYPES
];
3687 uint32_t slave
[OAM_N_TYPES
];
3689 master
[OAM_PACKET_IN
] = ntohl(msg
->packet_in_mask
[0]);
3690 master
[OAM_PORT_STATUS
] = ntohl(msg
->port_status_mask
[0]);
3691 master
[OAM_FLOW_REMOVED
] = ntohl(msg
->flow_removed_mask
[0]);
3693 slave
[OAM_PACKET_IN
] = ntohl(msg
->packet_in_mask
[1]);
3694 slave
[OAM_PORT_STATUS
] = ntohl(msg
->port_status_mask
[1]);
3695 slave
[OAM_FLOW_REMOVED
] = ntohl(msg
->flow_removed_mask
[1]);
3697 ofconn_set_async_config(ofconn
, master
, slave
);
3698 if (ofconn_get_type(ofconn
) == OFCONN_SERVICE
&&
3699 !ofconn_get_miss_send_len(ofconn
)) {
3700 ofconn_set_miss_send_len(ofconn
, OFP_DEFAULT_MISS_SEND_LEN
);
3707 handle_nxt_set_controller_id(struct ofconn
*ofconn
,
3708 const struct ofp_header
*oh
)
3710 const struct nx_controller_id
*nci
= ofpmsg_body(oh
);
3712 if (!is_all_zeros(nci
->zero
, sizeof nci
->zero
)) {
3713 return OFPERR_NXBRC_MUST_BE_ZERO
;
3716 ofconn_set_controller_id(ofconn
, ntohs(nci
->controller_id
));
3721 handle_barrier_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3725 if (ofconn_has_pending_opgroups(ofconn
)) {
3726 return OFPROTO_POSTPONE
;
3729 buf
= ofpraw_alloc_reply((oh
->version
== OFP10_VERSION
3730 ? OFPRAW_OFPT10_BARRIER_REPLY
3731 : OFPRAW_OFPT11_BARRIER_REPLY
), oh
, 0);
3732 ofconn_send_reply(ofconn
, buf
);
3737 ofproto_compose_flow_refresh_update(const struct rule
*rule
,
3738 enum nx_flow_monitor_flags flags
,
3741 struct ofoperation
*op
= rule
->pending
;
3742 struct ofputil_flow_update fu
;
3745 if (op
&& op
->type
== OFOPERATION_ADD
&& !op
->victim
) {
3746 /* We'll report the final flow when the operation completes. Reporting
3747 * it now would cause a duplicate report later. */
3751 fu
.event
= (flags
& (NXFMF_INITIAL
| NXFMF_ADD
)
3752 ? NXFME_ADDED
: NXFME_MODIFIED
);
3754 fu
.idle_timeout
= rule
->idle_timeout
;
3755 fu
.hard_timeout
= rule
->hard_timeout
;
3756 fu
.table_id
= rule
->table_id
;
3757 fu
.cookie
= rule
->flow_cookie
;
3758 minimatch_expand(&rule
->cr
.match
, &match
);
3760 fu
.priority
= rule
->cr
.priority
;
3761 if (!(flags
& NXFMF_ACTIONS
)) {
3765 fu
.ofpacts
= rule
->ofpacts
;
3766 fu
.ofpacts_len
= rule
->ofpacts_len
;
3768 /* An operation is in progress. Use the previous version of the flow's
3769 * actions, so that when the operation commits we report the change. */
3771 case OFOPERATION_ADD
:
3772 /* We already verified that there was a victim. */
3773 fu
.ofpacts
= op
->victim
->ofpacts
;
3774 fu
.ofpacts_len
= op
->victim
->ofpacts_len
;
3777 case OFOPERATION_MODIFY
:
3779 fu
.ofpacts
= op
->ofpacts
;
3780 fu
.ofpacts_len
= op
->ofpacts_len
;
3782 fu
.ofpacts
= rule
->ofpacts
;
3783 fu
.ofpacts_len
= rule
->ofpacts_len
;
3787 case OFOPERATION_DELETE
:
3788 fu
.ofpacts
= rule
->ofpacts
;
3789 fu
.ofpacts_len
= rule
->ofpacts_len
;
3797 if (list_is_empty(msgs
)) {
3798 ofputil_start_flow_update(msgs
);
3800 ofputil_append_flow_update(&fu
, msgs
);
3804 ofmonitor_compose_refresh_updates(struct list
*rules
, struct list
*msgs
)
3808 LIST_FOR_EACH (rule
, ofproto_node
, rules
) {
3809 enum nx_flow_monitor_flags flags
= rule
->monitor_flags
;
3810 rule
->monitor_flags
= 0;
3812 ofproto_compose_flow_refresh_update(rule
, flags
, msgs
);
3817 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor
*m
,
3818 struct rule
*rule
, uint64_t seqno
,
3821 enum nx_flow_monitor_flags update
;
3823 if (ofproto_rule_is_hidden(rule
)) {
3828 ? ofoperation_has_out_port(rule
->pending
, m
->out_port
)
3829 : ofproto_rule_has_out_port(rule
, m
->out_port
))) {
3834 if (rule
->add_seqno
> seqno
) {
3835 update
= NXFMF_ADD
| NXFMF_MODIFY
;
3836 } else if (rule
->modify_seqno
> seqno
) {
3837 update
= NXFMF_MODIFY
;
3842 if (!(m
->flags
& update
)) {
3846 update
= NXFMF_INITIAL
;
3849 if (!rule
->monitor_flags
) {
3850 list_push_back(rules
, &rule
->ofproto_node
);
3852 rule
->monitor_flags
|= update
| (m
->flags
& NXFMF_ACTIONS
);
3856 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor
*m
,
3860 const struct ofproto
*ofproto
= ofconn_get_ofproto(m
->ofconn
);
3861 const struct ofoperation
*op
;
3862 const struct oftable
*table
;
3863 struct cls_rule target
;
3865 cls_rule_init_from_minimatch(&target
, &m
->match
, 0);
3866 FOR_EACH_MATCHING_TABLE (table
, m
->table_id
, ofproto
) {
3867 struct cls_cursor cursor
;
3870 cls_cursor_init(&cursor
, &table
->cls
, &target
);
3871 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3872 ovs_assert(!rule
->pending
); /* XXX */
3873 ofproto_collect_ofmonitor_refresh_rule(m
, rule
, seqno
, rules
);
3877 HMAP_FOR_EACH (op
, hmap_node
, &ofproto
->deletions
) {
3878 struct rule
*rule
= op
->rule
;
3880 if (((m
->table_id
== 0xff
3881 ? !(ofproto
->tables
[rule
->table_id
].flags
& OFTABLE_HIDDEN
)
3882 : m
->table_id
== rule
->table_id
))
3883 && cls_rule_is_loose_match(&rule
->cr
, &target
.match
)) {
3884 ofproto_collect_ofmonitor_refresh_rule(m
, rule
, seqno
, rules
);
3887 cls_rule_destroy(&target
);
3891 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor
*m
,
3894 if (m
->flags
& NXFMF_INITIAL
) {
3895 ofproto_collect_ofmonitor_refresh_rules(m
, 0, rules
);
3900 ofmonitor_collect_resume_rules(struct ofmonitor
*m
,
3901 uint64_t seqno
, struct list
*rules
)
3903 ofproto_collect_ofmonitor_refresh_rules(m
, seqno
, rules
);
3907 handle_flow_monitor_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3909 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
3910 struct ofmonitor
**monitors
;
3911 size_t n_monitors
, allocated_monitors
;
3912 struct list replies
;
3919 ofpbuf_use_const(&b
, oh
, ntohs(oh
->length
));
3921 n_monitors
= allocated_monitors
= 0;
3923 struct ofputil_flow_monitor_request request
;
3924 struct ofmonitor
*m
;
3927 retval
= ofputil_decode_flow_monitor_request(&request
, &b
);
3928 if (retval
== EOF
) {
3930 } else if (retval
) {
3935 if (request
.table_id
!= 0xff
3936 && request
.table_id
>= ofproto
->n_tables
) {
3937 error
= OFPERR_OFPBRC_BAD_TABLE_ID
;
3941 error
= ofmonitor_create(&request
, ofconn
, &m
);
3946 if (n_monitors
>= allocated_monitors
) {
3947 monitors
= x2nrealloc(monitors
, &allocated_monitors
,
3950 monitors
[n_monitors
++] = m
;
3954 for (i
= 0; i
< n_monitors
; i
++) {
3955 ofproto_collect_ofmonitor_initial_rules(monitors
[i
], &rules
);
3958 ofpmp_init(&replies
, oh
);
3959 ofmonitor_compose_refresh_updates(&rules
, &replies
);
3960 ofconn_send_replies(ofconn
, &replies
);
3967 for (i
= 0; i
< n_monitors
; i
++) {
3968 ofmonitor_destroy(monitors
[i
]);
3975 handle_flow_monitor_cancel(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3977 struct ofmonitor
*m
;
3980 id
= ofputil_decode_flow_monitor_cancel(oh
);
3981 m
= ofmonitor_lookup(ofconn
, id
);
3983 return OFPERR_NXBRC_FM_BAD_ID
;
3986 ofmonitor_destroy(m
);
3991 handle_openflow__(struct ofconn
*ofconn
, const struct ofpbuf
*msg
)
3993 const struct ofp_header
*oh
= msg
->data
;
3997 error
= ofptype_decode(&type
, oh
);
4003 /* OpenFlow requests. */
4004 case OFPTYPE_ECHO_REQUEST
:
4005 return handle_echo_request(ofconn
, oh
);
4007 case OFPTYPE_FEATURES_REQUEST
:
4008 return handle_features_request(ofconn
, oh
);
4010 case OFPTYPE_GET_CONFIG_REQUEST
:
4011 return handle_get_config_request(ofconn
, oh
);
4013 case OFPTYPE_SET_CONFIG
:
4014 return handle_set_config(ofconn
, oh
);
4016 case OFPTYPE_PACKET_OUT
:
4017 return handle_packet_out(ofconn
, oh
);
4019 case OFPTYPE_PORT_MOD
:
4020 return handle_port_mod(ofconn
, oh
);
4022 case OFPTYPE_FLOW_MOD
:
4023 return handle_flow_mod(ofconn
, oh
);
4025 case OFPTYPE_BARRIER_REQUEST
:
4026 return handle_barrier_request(ofconn
, oh
);
4028 case OFPTYPE_ROLE_REQUEST
:
4029 return handle_role_request(ofconn
, oh
);
4031 /* OpenFlow replies. */
4032 case OFPTYPE_ECHO_REPLY
:
4035 /* Nicira extension requests. */
4036 case OFPTYPE_FLOW_MOD_TABLE_ID
:
4037 return handle_nxt_flow_mod_table_id(ofconn
, oh
);
4039 case OFPTYPE_SET_FLOW_FORMAT
:
4040 return handle_nxt_set_flow_format(ofconn
, oh
);
4042 case OFPTYPE_SET_PACKET_IN_FORMAT
:
4043 return handle_nxt_set_packet_in_format(ofconn
, oh
);
4045 case OFPTYPE_SET_CONTROLLER_ID
:
4046 return handle_nxt_set_controller_id(ofconn
, oh
);
4048 case OFPTYPE_FLOW_AGE
:
4049 /* Nothing to do. */
4052 case OFPTYPE_FLOW_MONITOR_CANCEL
:
4053 return handle_flow_monitor_cancel(ofconn
, oh
);
4055 case OFPTYPE_SET_ASYNC_CONFIG
:
4056 return handle_nxt_set_async_config(ofconn
, oh
);
4058 /* Statistics requests. */
4059 case OFPTYPE_DESC_STATS_REQUEST
:
4060 return handle_desc_stats_request(ofconn
, oh
);
4062 case OFPTYPE_FLOW_STATS_REQUEST
:
4063 return handle_flow_stats_request(ofconn
, oh
);
4065 case OFPTYPE_AGGREGATE_STATS_REQUEST
:
4066 return handle_aggregate_stats_request(ofconn
, oh
);
4068 case OFPTYPE_TABLE_STATS_REQUEST
:
4069 return handle_table_stats_request(ofconn
, oh
);
4071 case OFPTYPE_PORT_STATS_REQUEST
:
4072 return handle_port_stats_request(ofconn
, oh
);
4074 case OFPTYPE_QUEUE_STATS_REQUEST
:
4075 return handle_queue_stats_request(ofconn
, oh
);
4077 case OFPTYPE_PORT_DESC_STATS_REQUEST
:
4078 return handle_port_desc_stats_request(ofconn
, oh
);
4080 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST
:
4081 return handle_flow_monitor_request(ofconn
, oh
);
4083 /* FIXME: Change the following once they are implemented: */
4084 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST
:
4085 case OFPTYPE_GET_ASYNC_REQUEST
:
4086 case OFPTYPE_METER_MOD
:
4087 case OFPTYPE_GROUP_REQUEST
:
4088 case OFPTYPE_GROUP_DESC_REQUEST
:
4089 case OFPTYPE_GROUP_FEATURES_REQUEST
:
4090 case OFPTYPE_METER_REQUEST
:
4091 case OFPTYPE_METER_CONFIG_REQUEST
:
4092 case OFPTYPE_METER_FEATURES_REQUEST
:
4093 case OFPTYPE_TABLE_FEATURES_REQUEST
:
4094 return OFPERR_OFPBRC_BAD_TYPE
;
4098 case OFPTYPE_FEATURES_REPLY
:
4099 case OFPTYPE_GET_CONFIG_REPLY
:
4100 case OFPTYPE_PACKET_IN
:
4101 case OFPTYPE_FLOW_REMOVED
:
4102 case OFPTYPE_PORT_STATUS
:
4103 case OFPTYPE_BARRIER_REPLY
:
4104 case OFPTYPE_QUEUE_GET_CONFIG_REPLY
:
4105 case OFPTYPE_DESC_STATS_REPLY
:
4106 case OFPTYPE_FLOW_STATS_REPLY
:
4107 case OFPTYPE_QUEUE_STATS_REPLY
:
4108 case OFPTYPE_PORT_STATS_REPLY
:
4109 case OFPTYPE_TABLE_STATS_REPLY
:
4110 case OFPTYPE_AGGREGATE_STATS_REPLY
:
4111 case OFPTYPE_PORT_DESC_STATS_REPLY
:
4112 case OFPTYPE_ROLE_REPLY
:
4113 case OFPTYPE_FLOW_MONITOR_PAUSED
:
4114 case OFPTYPE_FLOW_MONITOR_RESUMED
:
4115 case OFPTYPE_FLOW_MONITOR_STATS_REPLY
:
4116 case OFPTYPE_GET_ASYNC_REPLY
:
4117 case OFPTYPE_GROUP_REPLY
:
4118 case OFPTYPE_GROUP_DESC_REPLY
:
4119 case OFPTYPE_GROUP_FEATURES_REPLY
:
4120 case OFPTYPE_METER_REPLY
:
4121 case OFPTYPE_METER_CONFIG_REPLY
:
4122 case OFPTYPE_METER_FEATURES_REPLY
:
4123 case OFPTYPE_TABLE_FEATURES_REPLY
:
4125 return OFPERR_OFPBRC_BAD_TYPE
;
4130 handle_openflow(struct ofconn
*ofconn
, struct ofpbuf
*ofp_msg
)
4132 int error
= handle_openflow__(ofconn
, ofp_msg
);
4133 if (error
&& error
!= OFPROTO_POSTPONE
) {
4134 ofconn_send_error(ofconn
, ofp_msg
->data
, error
);
4136 COVERAGE_INC(ofproto_recv_openflow
);
4137 return error
!= OFPROTO_POSTPONE
;
4140 /* Asynchronous operations. */
4142 /* Creates and returns a new ofopgroup that is not associated with any
4143 * OpenFlow connection.
4145 * The caller should add operations to the returned group with
4146 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4147 static struct ofopgroup
*
4148 ofopgroup_create_unattached(struct ofproto
*ofproto
)
4150 struct ofopgroup
*group
= xzalloc(sizeof *group
);
4151 group
->ofproto
= ofproto
;
4152 list_init(&group
->ofproto_node
);
4153 list_init(&group
->ops
);
4154 list_init(&group
->ofconn_node
);
4158 /* Creates and returns a new ofopgroup for 'ofproto'.
4160 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
4161 * connection. The 'request' and 'buffer_id' arguments are ignored.
4163 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
4164 * If the ofopgroup eventually fails, then the error reply will include
4165 * 'request'. If the ofopgroup eventually succeeds, then the packet with
4166 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
4168 * The caller should add operations to the returned group with
4169 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4170 static struct ofopgroup
*
4171 ofopgroup_create(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
4172 const struct ofp_header
*request
, uint32_t buffer_id
)
4174 struct ofopgroup
*group
= ofopgroup_create_unattached(ofproto
);
4176 size_t request_len
= ntohs(request
->length
);
4178 ovs_assert(ofconn_get_ofproto(ofconn
) == ofproto
);
4180 ofconn_add_opgroup(ofconn
, &group
->ofconn_node
);
4181 group
->ofconn
= ofconn
;
4182 group
->request
= xmemdup(request
, MIN(request_len
, 64));
4183 group
->buffer_id
= buffer_id
;
4188 /* Submits 'group' for processing.
4190 * If 'group' contains no operations (e.g. none were ever added, or all of the
4191 * ones that were added completed synchronously), then it is destroyed
4192 * immediately. Otherwise it is added to the ofproto's list of pending
4195 ofopgroup_submit(struct ofopgroup
*group
)
4197 if (!group
->n_running
) {
4198 ofopgroup_complete(group
);
4200 list_push_back(&group
->ofproto
->pending
, &group
->ofproto_node
);
4201 group
->ofproto
->n_pending
++;
4206 ofopgroup_complete(struct ofopgroup
*group
)
4208 struct ofproto
*ofproto
= group
->ofproto
;
4210 struct ofconn
*abbrev_ofconn
;
4211 ovs_be32 abbrev_xid
;
4213 struct ofoperation
*op
, *next_op
;
4216 ovs_assert(!group
->n_running
);
4219 LIST_FOR_EACH (op
, group_node
, &group
->ops
) {
4226 if (!error
&& group
->ofconn
&& group
->buffer_id
!= UINT32_MAX
) {
4227 LIST_FOR_EACH (op
, group_node
, &group
->ops
) {
4228 if (op
->type
!= OFOPERATION_DELETE
) {
4229 struct ofpbuf
*packet
;
4232 error
= ofconn_pktbuf_retrieve(group
->ofconn
, group
->buffer_id
,
4236 error
= rule_execute(op
->rule
, in_port
, packet
);
4243 if (!error
&& !list_is_empty(&group
->ofconn_node
)) {
4244 abbrev_ofconn
= group
->ofconn
;
4245 abbrev_xid
= group
->request
->xid
;
4247 abbrev_ofconn
= NULL
;
4248 abbrev_xid
= htonl(0);
4250 LIST_FOR_EACH_SAFE (op
, next_op
, group_node
, &group
->ops
) {
4251 struct rule
*rule
= op
->rule
;
4253 /* We generally want to report the change to active OpenFlow flow
4254 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
4256 - The operation failed.
4258 - The affected rule is not visible to controllers.
4260 - The operation's only effect was to update rule->modified. */
4262 || ofproto_rule_is_hidden(rule
)
4263 || (op
->type
== OFOPERATION_MODIFY
4265 && rule
->flow_cookie
== op
->flow_cookie
))) {
4266 /* Check that we can just cast from ofoperation_type to
4267 * nx_flow_update_event. */
4268 BUILD_ASSERT_DECL((enum nx_flow_update_event
) OFOPERATION_ADD
4270 BUILD_ASSERT_DECL((enum nx_flow_update_event
) OFOPERATION_DELETE
4272 BUILD_ASSERT_DECL((enum nx_flow_update_event
) OFOPERATION_MODIFY
4275 ofmonitor_report(ofproto
->connmgr
, rule
,
4276 (enum nx_flow_update_event
) op
->type
,
4277 op
->reason
, abbrev_ofconn
, abbrev_xid
);
4280 rule
->pending
= NULL
;
4283 case OFOPERATION_ADD
:
4287 ofproto_rule_destroy__(op
->victim
);
4288 vid_mask
= minimask_get_vid_mask(&rule
->cr
.match
.mask
);
4289 if (vid_mask
== VLAN_VID_MASK
) {
4290 if (ofproto
->vlan_bitmap
) {
4291 uint16_t vid
= miniflow_get_vid(&rule
->cr
.match
.flow
);
4292 if (!bitmap_is_set(ofproto
->vlan_bitmap
, vid
)) {
4293 bitmap_set1(ofproto
->vlan_bitmap
, vid
);
4294 ofproto
->vlans_changed
= true;
4297 ofproto
->vlans_changed
= true;
4301 oftable_substitute_rule(rule
, op
->victim
);
4302 ofproto_rule_destroy__(rule
);
4306 case OFOPERATION_DELETE
:
4307 ovs_assert(!op
->error
);
4308 ofproto_rule_destroy__(rule
);
4312 case OFOPERATION_MODIFY
:
4314 rule
->modified
= time_msec();
4316 rule
->flow_cookie
= op
->flow_cookie
;
4318 free(rule
->ofpacts
);
4319 rule
->ofpacts
= op
->ofpacts
;
4320 rule
->ofpacts_len
= op
->ofpacts_len
;
4322 op
->ofpacts_len
= 0;
4331 ofoperation_destroy(op
);
4334 ofmonitor_flush(ofproto
->connmgr
);
4336 if (!list_is_empty(&group
->ofproto_node
)) {
4337 ovs_assert(ofproto
->n_pending
> 0);
4338 ofproto
->n_pending
--;
4339 list_remove(&group
->ofproto_node
);
4341 if (!list_is_empty(&group
->ofconn_node
)) {
4342 list_remove(&group
->ofconn_node
);
4344 ofconn_send_error(group
->ofconn
, group
->request
, error
);
4346 connmgr_retry(ofproto
->connmgr
);
4348 free(group
->request
);
4352 /* Initiates a new operation on 'rule', of the specified 'type', within
4353 * 'group'. Prior to calling, 'rule' must not have any pending operation.
4355 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
4356 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
4358 * Returns the newly created ofoperation (which is also available as
4359 * rule->pending). */
4360 static struct ofoperation
*
4361 ofoperation_create(struct ofopgroup
*group
, struct rule
*rule
,
4362 enum ofoperation_type type
,
4363 enum ofp_flow_removed_reason reason
)
4365 struct ofproto
*ofproto
= group
->ofproto
;
4366 struct ofoperation
*op
;
4368 ovs_assert(!rule
->pending
);
4370 op
= rule
->pending
= xzalloc(sizeof *op
);
4372 list_push_back(&group
->ops
, &op
->group_node
);
4375 op
->reason
= reason
;
4376 op
->flow_cookie
= rule
->flow_cookie
;
4380 if (type
== OFOPERATION_DELETE
) {
4381 hmap_insert(&ofproto
->deletions
, &op
->hmap_node
,
4382 cls_rule_hash(&rule
->cr
, rule
->table_id
));
4389 ofoperation_destroy(struct ofoperation
*op
)
4391 struct ofopgroup
*group
= op
->group
;
4394 op
->rule
->pending
= NULL
;
4396 if (op
->type
== OFOPERATION_DELETE
) {
4397 hmap_remove(&group
->ofproto
->deletions
, &op
->hmap_node
);
4399 list_remove(&op
->group_node
);
4404 /* Indicates that 'op' completed with status 'error', which is either 0 to
4405 * indicate success or an OpenFlow error code on failure.
4407 * If 'error' is 0, indicating success, the operation will be committed
4408 * permanently to the flow table. There is one interesting subcase:
4410 * - If 'op' is an "add flow" operation that is replacing an existing rule in
4411 * the flow table (the "victim" rule) by a new one, then the caller must
4412 * have uninitialized any derived state in the victim rule, as in step 5 in
4413 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
4414 * performs steps 6 and 7 for the victim rule, most notably by calling its
4415 * ->rule_dealloc() function.
4417 * If 'error' is nonzero, then generally the operation will be rolled back:
4419 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
4420 * restores the original rule. The caller must have uninitialized any
4421 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
4422 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
4423 * and 7 for the new rule, calling its ->rule_dealloc() function.
4425 * - If 'op' is a "modify flow" operation, ofproto restores the original
4428 * - 'op' must not be a "delete flow" operation. Removing a rule is not
4429 * allowed to fail. It must always succeed.
4431 * Please see the large comment in ofproto/ofproto-provider.h titled
4432 * "Asynchronous Operation Support" for more information. */
4434 ofoperation_complete(struct ofoperation
*op
, enum ofperr error
)
4436 struct ofopgroup
*group
= op
->group
;
4438 ovs_assert(op
->rule
->pending
== op
);
4439 ovs_assert(group
->n_running
> 0);
4440 ovs_assert(!error
|| op
->type
!= OFOPERATION_DELETE
);
4443 if (!--group
->n_running
&& !list_is_empty(&group
->ofproto_node
)) {
4444 ofopgroup_complete(group
);
4449 ofoperation_get_victim(struct ofoperation
*op
)
4451 ovs_assert(op
->type
== OFOPERATION_ADD
);
4456 pick_datapath_id(const struct ofproto
*ofproto
)
4458 const struct ofport
*port
;
4460 port
= ofproto_get_port(ofproto
, OFPP_LOCAL
);
4462 uint8_t ea
[ETH_ADDR_LEN
];
4465 error
= netdev_get_etheraddr(port
->netdev
, ea
);
4467 return eth_addr_to_uint64(ea
);
4469 VLOG_WARN("%s: could not get MAC address for %s (%s)",
4470 ofproto
->name
, netdev_get_name(port
->netdev
),
4473 return ofproto
->fallback_dpid
;
4477 pick_fallback_dpid(void)
4479 uint8_t ea
[ETH_ADDR_LEN
];
4480 eth_addr_nicira_random(ea
);
4481 return eth_addr_to_uint64(ea
);
4484 /* Table overflow policy. */
4486 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
4487 * is not configured to evict rules or if the table contains no evictable
4488 * rules. (Rules with 'evictable' set to false or with no timeouts are not
4490 static struct rule
*
4491 choose_rule_to_evict(struct oftable
*table
)
4493 struct eviction_group
*evg
;
4495 if (!table
->eviction_fields
) {
4499 /* In the common case, the outer and inner loops here will each be entered
4502 * - The inner loop normally "return"s in its first iteration. If the
4503 * eviction group has any evictable rules, then it always returns in
4506 * - The outer loop only iterates more than once if the largest eviction
4507 * group has no evictable rules.
4509 * - The outer loop can exit only if table's 'max_flows' is all filled up
4510 * by unevictable rules'. */
4511 HEAP_FOR_EACH (evg
, size_node
, &table
->eviction_groups_by_size
) {
4514 HEAP_FOR_EACH (rule
, evg_node
, &evg
->rules
) {
4515 if (rule
->evictable
) {
4524 /* Searches 'ofproto' for tables that have more flows than their configured
4525 * maximum and that have flow eviction enabled, and evicts as many flows as
4526 * necessary and currently feasible from them.
4528 * This triggers only when an OpenFlow table has N flows in it and then the
4529 * client configures a maximum number of flows less than N. */
4531 ofproto_evict(struct ofproto
*ofproto
)
4533 struct ofopgroup
*group
;
4534 struct oftable
*table
;
4536 group
= ofopgroup_create_unattached(ofproto
);
4537 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
4538 while (classifier_count(&table
->cls
) > table
->max_flows
4539 && table
->eviction_fields
) {
4542 rule
= choose_rule_to_evict(table
);
4543 if (!rule
|| rule
->pending
) {
4547 ofoperation_create(group
, rule
,
4548 OFOPERATION_DELETE
, OFPRR_EVICTION
);
4549 oftable_remove_rule(rule
);
4550 ofproto
->ofproto_class
->rule_destruct(rule
);
4553 ofopgroup_submit(group
);
4556 /* Eviction groups. */
4558 /* Returns the priority to use for an eviction_group that contains 'n_rules'
4559 * rules. The priority contains low-order random bits to ensure that eviction
4560 * groups with the same number of rules are prioritized randomly. */
4562 eviction_group_priority(size_t n_rules
)
4564 uint16_t size
= MIN(UINT16_MAX
, n_rules
);
4565 return (size
<< 16) | random_uint16();
4568 /* Updates 'evg', an eviction_group within 'table', following a change that
4569 * adds or removes rules in 'evg'. */
4571 eviction_group_resized(struct oftable
*table
, struct eviction_group
*evg
)
4573 heap_change(&table
->eviction_groups_by_size
, &evg
->size_node
,
4574 eviction_group_priority(heap_count(&evg
->rules
)));
4577 /* Destroys 'evg', an eviction_group within 'table':
4579 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
4580 * rules themselves, just removes them from the eviction group.)
4582 * - Removes 'evg' from 'table'.
4586 eviction_group_destroy(struct oftable
*table
, struct eviction_group
*evg
)
4588 while (!heap_is_empty(&evg
->rules
)) {
4591 rule
= CONTAINER_OF(heap_pop(&evg
->rules
), struct rule
, evg_node
);
4592 rule
->eviction_group
= NULL
;
4594 hmap_remove(&table
->eviction_groups_by_id
, &evg
->id_node
);
4595 heap_remove(&table
->eviction_groups_by_size
, &evg
->size_node
);
4596 heap_destroy(&evg
->rules
);
4600 /* Removes 'rule' from its eviction group, if any. */
4602 eviction_group_remove_rule(struct rule
*rule
)
4604 if (rule
->eviction_group
) {
4605 struct oftable
*table
= &rule
->ofproto
->tables
[rule
->table_id
];
4606 struct eviction_group
*evg
= rule
->eviction_group
;
4608 rule
->eviction_group
= NULL
;
4609 heap_remove(&evg
->rules
, &rule
->evg_node
);
4610 if (heap_is_empty(&evg
->rules
)) {
4611 eviction_group_destroy(table
, evg
);
4613 eviction_group_resized(table
, evg
);
4618 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
4619 * returns the hash value. */
4621 eviction_group_hash_rule(struct rule
*rule
)
4623 struct oftable
*table
= &rule
->ofproto
->tables
[rule
->table_id
];
4624 const struct mf_subfield
*sf
;
4628 hash
= table
->eviction_group_id_basis
;
4629 miniflow_expand(&rule
->cr
.match
.flow
, &flow
);
4630 for (sf
= table
->eviction_fields
;
4631 sf
< &table
->eviction_fields
[table
->n_eviction_fields
];
4634 if (mf_are_prereqs_ok(sf
->field
, &flow
)) {
4635 union mf_value value
;
4637 mf_get_value(sf
->field
, &flow
, &value
);
4639 bitwise_zero(&value
, sf
->field
->n_bytes
, 0, sf
->ofs
);
4641 if (sf
->ofs
+ sf
->n_bits
< sf
->field
->n_bytes
* 8) {
4642 unsigned int start
= sf
->ofs
+ sf
->n_bits
;
4643 bitwise_zero(&value
, sf
->field
->n_bytes
, start
,
4644 sf
->field
->n_bytes
* 8 - start
);
4646 hash
= hash_bytes(&value
, sf
->field
->n_bytes
, hash
);
4648 hash
= hash_int(hash
, 0);
4655 /* Returns an eviction group within 'table' with the given 'id', creating one
4657 static struct eviction_group
*
4658 eviction_group_find(struct oftable
*table
, uint32_t id
)
4660 struct eviction_group
*evg
;
4662 HMAP_FOR_EACH_WITH_HASH (evg
, id_node
, id
, &table
->eviction_groups_by_id
) {
4666 evg
= xmalloc(sizeof *evg
);
4667 hmap_insert(&table
->eviction_groups_by_id
, &evg
->id_node
, id
);
4668 heap_insert(&table
->eviction_groups_by_size
, &evg
->size_node
,
4669 eviction_group_priority(0));
4670 heap_init(&evg
->rules
);
4675 /* Returns an eviction priority for 'rule'. The return value should be
4676 * interpreted so that higher priorities make a rule more attractive candidates
4679 rule_eviction_priority(struct rule
*rule
)
4681 long long int hard_expiration
;
4682 long long int idle_expiration
;
4683 long long int expiration
;
4684 uint32_t expiration_offset
;
4686 /* Calculate time of expiration. */
4687 hard_expiration
= (rule
->hard_timeout
4688 ? rule
->modified
+ rule
->hard_timeout
* 1000
4690 idle_expiration
= (rule
->idle_timeout
4691 ? rule
->used
+ rule
->idle_timeout
* 1000
4693 expiration
= MIN(hard_expiration
, idle_expiration
);
4694 if (expiration
== LLONG_MAX
) {
4698 /* Calculate the time of expiration as a number of (approximate) seconds
4699 * after program startup.
4701 * This should work OK for program runs that last UINT32_MAX seconds or
4702 * less. Therefore, please restart OVS at least once every 136 years. */
4703 expiration_offset
= (expiration
>> 10) - (time_boot_msec() >> 10);
4705 /* Invert the expiration offset because we're using a max-heap. */
4706 return UINT32_MAX
- expiration_offset
;
4709 /* Adds 'rule' to an appropriate eviction group for its oftable's
4710 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
4711 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4714 * The caller must ensure that 'rule' is not already in an eviction group. */
4716 eviction_group_add_rule(struct rule
*rule
)
4718 struct ofproto
*ofproto
= rule
->ofproto
;
4719 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
4721 if (table
->eviction_fields
4722 && (rule
->hard_timeout
|| rule
->idle_timeout
)) {
4723 struct eviction_group
*evg
;
4725 evg
= eviction_group_find(table
, eviction_group_hash_rule(rule
));
4727 rule
->eviction_group
= evg
;
4728 heap_insert(&evg
->rules
, &rule
->evg_node
,
4729 rule_eviction_priority(rule
));
4730 eviction_group_resized(table
, evg
);
4736 /* Initializes 'table'. */
4738 oftable_init(struct oftable
*table
)
4740 memset(table
, 0, sizeof *table
);
4741 classifier_init(&table
->cls
);
4742 table
->max_flows
= UINT_MAX
;
4745 /* Destroys 'table', including its classifier and eviction groups.
4747 * The caller is responsible for freeing 'table' itself. */
4749 oftable_destroy(struct oftable
*table
)
4751 ovs_assert(classifier_is_empty(&table
->cls
));
4752 oftable_disable_eviction(table
);
4753 classifier_destroy(&table
->cls
);
4757 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4758 * string, then 'table' will use its default name.
4760 * This only affects the name exposed for a table exposed through the OpenFlow
4761 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4763 oftable_set_name(struct oftable
*table
, const char *name
)
4765 if (name
&& name
[0]) {
4766 int len
= strnlen(name
, OFP_MAX_TABLE_NAME_LEN
);
4767 if (!table
->name
|| strncmp(name
, table
->name
, len
)) {
4769 table
->name
= xmemdup0(name
, len
);
4777 /* oftables support a choice of two policies when adding a rule would cause the
4778 * number of flows in the table to exceed the configured maximum number: either
4779 * they can refuse to add the new flow or they can evict some existing flow.
4780 * This function configures the former policy on 'table'. */
4782 oftable_disable_eviction(struct oftable
*table
)
4784 if (table
->eviction_fields
) {
4785 struct eviction_group
*evg
, *next
;
4787 HMAP_FOR_EACH_SAFE (evg
, next
, id_node
,
4788 &table
->eviction_groups_by_id
) {
4789 eviction_group_destroy(table
, evg
);
4791 hmap_destroy(&table
->eviction_groups_by_id
);
4792 heap_destroy(&table
->eviction_groups_by_size
);
4794 free(table
->eviction_fields
);
4795 table
->eviction_fields
= NULL
;
4796 table
->n_eviction_fields
= 0;
4800 /* oftables support a choice of two policies when adding a rule would cause the
4801 * number of flows in the table to exceed the configured maximum number: either
4802 * they can refuse to add the new flow or they can evict some existing flow.
4803 * This function configures the latter policy on 'table', with fairness based
4804 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4805 * 'n_fields' as 0 disables fairness.) */
4807 oftable_enable_eviction(struct oftable
*table
,
4808 const struct mf_subfield
*fields
, size_t n_fields
)
4810 struct cls_cursor cursor
;
4813 if (table
->eviction_fields
4814 && n_fields
== table
->n_eviction_fields
4816 || !memcmp(fields
, table
->eviction_fields
,
4817 n_fields
* sizeof *fields
))) {
4822 oftable_disable_eviction(table
);
4824 table
->n_eviction_fields
= n_fields
;
4825 table
->eviction_fields
= xmemdup(fields
, n_fields
* sizeof *fields
);
4827 table
->eviction_group_id_basis
= random_uint32();
4828 hmap_init(&table
->eviction_groups_by_id
);
4829 heap_init(&table
->eviction_groups_by_size
);
4831 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
4832 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
4833 eviction_group_add_rule(rule
);
4837 /* Removes 'rule' from the oftable that contains it. */
4839 oftable_remove_rule(struct rule
*rule
)
4841 struct ofproto
*ofproto
= rule
->ofproto
;
4842 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
4844 classifier_remove(&table
->cls
, &rule
->cr
);
4845 eviction_group_remove_rule(rule
);
4846 if (!list_is_empty(&rule
->expirable
)) {
4847 list_remove(&rule
->expirable
);
4851 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4852 * oftable that has an identical cls_rule. Returns the rule that was removed,
4853 * if any, and otherwise NULL. */
4854 static struct rule
*
4855 oftable_replace_rule(struct rule
*rule
)
4857 struct ofproto
*ofproto
= rule
->ofproto
;
4858 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
4859 struct rule
*victim
;
4860 bool may_expire
= rule
->hard_timeout
|| rule
->idle_timeout
;
4863 list_insert(&ofproto
->expirable
, &rule
->expirable
);
4866 victim
= rule_from_cls_rule(classifier_replace(&table
->cls
, &rule
->cr
));
4868 if (!list_is_empty(&victim
->expirable
)) {
4869 list_remove(&victim
->expirable
);
4871 eviction_group_remove_rule(victim
);
4873 eviction_group_add_rule(rule
);
4877 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4879 oftable_substitute_rule(struct rule
*old
, struct rule
*new)
4882 oftable_replace_rule(new);
4884 oftable_remove_rule(old
);
4888 /* unixctl commands. */
4891 ofproto_lookup(const char *name
)
4893 struct ofproto
*ofproto
;
4895 HMAP_FOR_EACH_WITH_HASH (ofproto
, hmap_node
, hash_string(name
, 0),
4897 if (!strcmp(ofproto
->name
, name
)) {
4905 ofproto_unixctl_list(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
4906 const char *argv
[] OVS_UNUSED
, void *aux OVS_UNUSED
)
4908 struct ofproto
*ofproto
;
4912 HMAP_FOR_EACH (ofproto
, hmap_node
, &all_ofprotos
) {
4913 ds_put_format(&results
, "%s\n", ofproto
->name
);
4915 unixctl_command_reply(conn
, ds_cstr(&results
));
4916 ds_destroy(&results
);
4920 ofproto_unixctl_init(void)
4922 static bool registered
;
4928 unixctl_command_register("ofproto/list", "", 0, 0,
4929 ofproto_unixctl_list
, NULL
);
4932 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4934 * This is deprecated. It is only for compatibility with broken device drivers
4935 * in old versions of Linux that do not properly support VLANs when VLAN
4936 * devices are not used. When broken device drivers are no longer in
4937 * widespread use, we will delete these interfaces. */
4939 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4940 * (exactly) by an OpenFlow rule in 'ofproto'. */
4942 ofproto_get_vlan_usage(struct ofproto
*ofproto
, unsigned long int *vlan_bitmap
)
4944 const struct oftable
*oftable
;
4946 free(ofproto
->vlan_bitmap
);
4947 ofproto
->vlan_bitmap
= bitmap_allocate(4096);
4948 ofproto
->vlans_changed
= false;
4950 OFPROTO_FOR_EACH_TABLE (oftable
, ofproto
) {
4951 const struct cls_table
*table
;
4953 HMAP_FOR_EACH (table
, hmap_node
, &oftable
->cls
.tables
) {
4954 if (minimask_get_vid_mask(&table
->mask
) == VLAN_VID_MASK
) {
4955 const struct cls_rule
*rule
;
4957 HMAP_FOR_EACH (rule
, hmap_node
, &table
->rules
) {
4958 uint16_t vid
= miniflow_get_vid(&rule
->match
.flow
);
4959 bitmap_set1(vlan_bitmap
, vid
);
4960 bitmap_set1(ofproto
->vlan_bitmap
, vid
);
4967 /* Returns true if new VLANs have come into use by the flow table since the
4968 * last call to ofproto_get_vlan_usage().
4970 * We don't track when old VLANs stop being used. */
4972 ofproto_has_vlan_usage_changed(const struct ofproto
*ofproto
)
4974 return ofproto
->vlans_changed
;
4977 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4978 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4979 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4980 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4981 * then the VLAN device is un-enslaved. */
4983 ofproto_port_set_realdev(struct ofproto
*ofproto
, uint16_t vlandev_ofp_port
,
4984 uint16_t realdev_ofp_port
, int vid
)
4986 struct ofport
*ofport
;
4989 ovs_assert(vlandev_ofp_port
!= realdev_ofp_port
);
4991 ofport
= ofproto_get_port(ofproto
, vlandev_ofp_port
);
4993 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16
,
4994 ofproto
->name
, vlandev_ofp_port
);
4998 if (!ofproto
->ofproto_class
->set_realdev
) {
4999 if (!vlandev_ofp_port
) {
5002 VLOG_WARN("%s: vlan splinters not supported", ofproto
->name
);
5006 error
= ofproto
->ofproto_class
->set_realdev(ofport
, realdev_ofp_port
, vid
);
5008 VLOG_WARN("%s: setting realdev on port %"PRIu16
" (%s) failed (%s)",
5009 ofproto
->name
, vlandev_ofp_port
,
5010 netdev_get_name(ofport
->netdev
), strerror(error
));