2 * Copyright (c) 2009, 2010, 2011, 2012 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 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
227 ofproto_initialize(void)
233 ofproto_class_register(&ofproto_dpif_class
);
237 /* 'type' should be a normalized datapath type, as returned by
238 * ofproto_normalize_type(). Returns the corresponding ofproto_class
239 * structure, or a null pointer if there is none registered for 'type'. */
240 static const struct ofproto_class
*
241 ofproto_class_find__(const char *type
)
245 ofproto_initialize();
246 for (i
= 0; i
< n_ofproto_classes
; i
++) {
247 const struct ofproto_class
*class = ofproto_classes
[i
];
252 class->enumerate_types(&types
);
253 found
= sset_contains(&types
, type
);
254 sset_destroy(&types
);
260 VLOG_WARN("unknown datapath type %s", type
);
264 /* Registers a new ofproto class. After successful registration, new ofprotos
265 * of that type can be created using ofproto_create(). */
267 ofproto_class_register(const struct ofproto_class
*new_class
)
271 for (i
= 0; i
< n_ofproto_classes
; i
++) {
272 if (ofproto_classes
[i
] == new_class
) {
277 if (n_ofproto_classes
>= allocated_ofproto_classes
) {
278 ofproto_classes
= x2nrealloc(ofproto_classes
,
279 &allocated_ofproto_classes
,
280 sizeof *ofproto_classes
);
282 ofproto_classes
[n_ofproto_classes
++] = new_class
;
286 /* Unregisters a datapath provider. 'type' must have been previously
287 * registered and not currently be in use by any ofprotos. After
288 * unregistration new datapaths of that type cannot be opened using
289 * ofproto_create(). */
291 ofproto_class_unregister(const struct ofproto_class
*class)
295 for (i
= 0; i
< n_ofproto_classes
; i
++) {
296 if (ofproto_classes
[i
] == class) {
297 for (i
++; i
< n_ofproto_classes
; i
++) {
298 ofproto_classes
[i
- 1] = ofproto_classes
[i
];
304 VLOG_WARN("attempted to unregister an ofproto class that is not "
309 /* Clears 'types' and enumerates all registered ofproto types into it. The
310 * caller must first initialize the sset. */
312 ofproto_enumerate_types(struct sset
*types
)
316 ofproto_initialize();
317 for (i
= 0; i
< n_ofproto_classes
; i
++) {
318 ofproto_classes
[i
]->enumerate_types(types
);
322 /* Returns the fully spelled out name for the given ofproto 'type'.
324 * Normalized type string can be compared with strcmp(). Unnormalized type
325 * string might be the same even if they have different spellings. */
327 ofproto_normalize_type(const char *type
)
329 return type
&& type
[0] ? type
: "system";
332 /* Clears 'names' and enumerates the names of all known created ofprotos with
333 * the given 'type'. The caller must first initialize the sset. Returns 0 if
334 * successful, otherwise a positive errno value.
336 * Some kinds of datapaths might not be practically enumerable. This is not
337 * considered an error. */
339 ofproto_enumerate_names(const char *type
, struct sset
*names
)
341 const struct ofproto_class
*class = ofproto_class_find__(type
);
342 return class ? class->enumerate_names(type
, names
) : EAFNOSUPPORT
;
346 ofproto_create(const char *datapath_name
, const char *datapath_type
,
347 struct ofproto
**ofprotop
)
349 const struct ofproto_class
*class;
350 struct ofproto
*ofproto
;
355 ofproto_initialize();
356 ofproto_unixctl_init();
358 datapath_type
= ofproto_normalize_type(datapath_type
);
359 class = ofproto_class_find__(datapath_type
);
361 VLOG_WARN("could not create datapath %s of unknown type %s",
362 datapath_name
, datapath_type
);
366 ofproto
= class->alloc();
368 VLOG_ERR("failed to allocate datapath %s of type %s",
369 datapath_name
, datapath_type
);
374 memset(ofproto
, 0, sizeof *ofproto
);
375 ofproto
->ofproto_class
= class;
376 ofproto
->name
= xstrdup(datapath_name
);
377 ofproto
->type
= xstrdup(datapath_type
);
378 hmap_insert(&all_ofprotos
, &ofproto
->hmap_node
,
379 hash_string(ofproto
->name
, 0));
380 ofproto
->datapath_id
= 0;
381 ofproto_set_flow_eviction_threshold(ofproto
,
382 OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT
);
383 ofproto
->forward_bpdu
= false;
384 ofproto
->fallback_dpid
= pick_fallback_dpid();
385 ofproto
->mfr_desc
= xstrdup(DEFAULT_MFR_DESC
);
386 ofproto
->hw_desc
= xstrdup(DEFAULT_HW_DESC
);
387 ofproto
->sw_desc
= xstrdup(DEFAULT_SW_DESC
);
388 ofproto
->serial_desc
= xstrdup(DEFAULT_SERIAL_DESC
);
389 ofproto
->dp_desc
= xstrdup(DEFAULT_DP_DESC
);
390 ofproto
->frag_handling
= OFPC_FRAG_NORMAL
;
391 hmap_init(&ofproto
->ports
);
392 shash_init(&ofproto
->port_by_name
);
393 ofproto
->max_ports
= OFPP_MAX
;
394 ofproto
->tables
= NULL
;
395 ofproto
->n_tables
= 0;
396 ofproto
->connmgr
= connmgr_create(ofproto
, datapath_name
, datapath_name
);
397 ofproto
->state
= S_OPENFLOW
;
398 list_init(&ofproto
->pending
);
399 ofproto
->n_pending
= 0;
400 hmap_init(&ofproto
->deletions
);
401 ofproto
->n_add
= ofproto
->n_delete
= ofproto
->n_modify
= 0;
402 ofproto
->first_op
= ofproto
->last_op
= LLONG_MIN
;
403 ofproto
->next_op_report
= LLONG_MAX
;
404 ofproto
->op_backoff
= LLONG_MIN
;
405 ofproto
->vlan_bitmap
= NULL
;
406 ofproto
->vlans_changed
= false;
407 ofproto
->min_mtu
= INT_MAX
;
409 error
= ofproto
->ofproto_class
->construct(ofproto
);
411 VLOG_ERR("failed to open datapath %s: %s",
412 datapath_name
, strerror(error
));
413 ofproto_destroy__(ofproto
);
417 assert(ofproto
->n_tables
);
419 ofproto
->datapath_id
= pick_datapath_id(ofproto
);
426 /* Must be called (only) by an ofproto implementation in its constructor
427 * function. See the large comment on 'construct' in struct ofproto_class for
430 ofproto_init_tables(struct ofproto
*ofproto
, int n_tables
)
432 struct oftable
*table
;
434 assert(!ofproto
->n_tables
);
435 assert(n_tables
>= 1 && n_tables
<= 255);
437 ofproto
->n_tables
= n_tables
;
438 ofproto
->tables
= xmalloc(n_tables
* sizeof *ofproto
->tables
);
439 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
444 /* To be optionally called (only) by an ofproto implementation in its
445 * constructor function. See the large comment on 'construct' in struct
446 * ofproto_class for details.
448 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
449 * will then ensure that actions passed into the ofproto implementation will
450 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
451 * function is not called, there will be no such restriction.
453 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
454 * the 'max_ports' restriction. */
456 ofproto_init_max_ports(struct ofproto
*ofproto
, uint16_t max_ports
)
458 assert(max_ports
<= OFPP_MAX
);
459 ofproto
->max_ports
= max_ports
;
463 ofproto_get_datapath_id(const struct ofproto
*ofproto
)
465 return ofproto
->datapath_id
;
469 ofproto_set_datapath_id(struct ofproto
*p
, uint64_t datapath_id
)
471 uint64_t old_dpid
= p
->datapath_id
;
472 p
->datapath_id
= datapath_id
? datapath_id
: pick_datapath_id(p
);
473 if (p
->datapath_id
!= old_dpid
) {
474 /* Force all active connections to reconnect, since there is no way to
475 * notify a controller that the datapath ID has changed. */
476 ofproto_reconnect_controllers(p
);
481 ofproto_set_controllers(struct ofproto
*p
,
482 const struct ofproto_controller
*controllers
,
483 size_t n_controllers
)
485 connmgr_set_controllers(p
->connmgr
, controllers
, n_controllers
);
489 ofproto_set_fail_mode(struct ofproto
*p
, enum ofproto_fail_mode fail_mode
)
491 connmgr_set_fail_mode(p
->connmgr
, fail_mode
);
494 /* Drops the connections between 'ofproto' and all of its controllers, forcing
495 * them to reconnect. */
497 ofproto_reconnect_controllers(struct ofproto
*ofproto
)
499 connmgr_reconnect(ofproto
->connmgr
);
502 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
503 * in-band control should guarantee access, in the same way that in-band
504 * control guarantees access to OpenFlow controllers. */
506 ofproto_set_extra_in_band_remotes(struct ofproto
*ofproto
,
507 const struct sockaddr_in
*extras
, size_t n
)
509 connmgr_set_extra_in_band_remotes(ofproto
->connmgr
, extras
, n
);
512 /* Sets the OpenFlow queue used by flows set up by in-band control on
513 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
514 * flows will use the default queue. */
516 ofproto_set_in_band_queue(struct ofproto
*ofproto
, int queue_id
)
518 connmgr_set_in_band_queue(ofproto
->connmgr
, queue_id
);
521 /* Sets the number of flows at which eviction from the kernel flow table
524 ofproto_set_flow_eviction_threshold(struct ofproto
*ofproto
, unsigned threshold
)
526 if (threshold
< OFPROTO_FLOW_EVICTION_THRESHOLD_MIN
) {
527 ofproto
->flow_eviction_threshold
= OFPROTO_FLOW_EVICTION_THRESHOLD_MIN
;
529 ofproto
->flow_eviction_threshold
= threshold
;
533 /* If forward_bpdu is true, the NORMAL action will forward frames with
534 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
535 * the NORMAL action will drop these frames. */
537 ofproto_set_forward_bpdu(struct ofproto
*ofproto
, bool forward_bpdu
)
539 bool old_val
= ofproto
->forward_bpdu
;
540 ofproto
->forward_bpdu
= forward_bpdu
;
541 if (old_val
!= ofproto
->forward_bpdu
) {
542 if (ofproto
->ofproto_class
->forward_bpdu_changed
) {
543 ofproto
->ofproto_class
->forward_bpdu_changed(ofproto
);
548 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
549 * 'idle_time', in seconds. */
551 ofproto_set_mac_idle_time(struct ofproto
*ofproto
, unsigned idle_time
)
553 if (ofproto
->ofproto_class
->set_mac_idle_time
) {
554 ofproto
->ofproto_class
->set_mac_idle_time(ofproto
, idle_time
);
559 ofproto_set_desc(struct ofproto
*p
,
560 const char *mfr_desc
, const char *hw_desc
,
561 const char *sw_desc
, const char *serial_desc
,
564 struct ofp_desc_stats
*ods
;
567 if (strlen(mfr_desc
) >= sizeof ods
->mfr_desc
) {
568 VLOG_WARN("%s: truncating mfr_desc, must be less than %zu bytes",
569 p
->name
, sizeof ods
->mfr_desc
);
572 p
->mfr_desc
= xstrdup(mfr_desc
);
575 if (strlen(hw_desc
) >= sizeof ods
->hw_desc
) {
576 VLOG_WARN("%s: truncating hw_desc, must be less than %zu bytes",
577 p
->name
, sizeof ods
->hw_desc
);
580 p
->hw_desc
= xstrdup(hw_desc
);
583 if (strlen(sw_desc
) >= sizeof ods
->sw_desc
) {
584 VLOG_WARN("%s: truncating sw_desc, must be less than %zu bytes",
585 p
->name
, sizeof ods
->sw_desc
);
588 p
->sw_desc
= xstrdup(sw_desc
);
591 if (strlen(serial_desc
) >= sizeof ods
->serial_num
) {
592 VLOG_WARN("%s: truncating serial_desc, must be less than %zu "
593 "bytes", p
->name
, sizeof ods
->serial_num
);
595 free(p
->serial_desc
);
596 p
->serial_desc
= xstrdup(serial_desc
);
599 if (strlen(dp_desc
) >= sizeof ods
->dp_desc
) {
600 VLOG_WARN("%s: truncating dp_desc, must be less than %zu bytes",
601 p
->name
, sizeof ods
->dp_desc
);
604 p
->dp_desc
= xstrdup(dp_desc
);
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;
644 /* Spanning Tree Protocol (STP) configuration. */
646 /* Configures STP on 'ofproto' using the settings defined in 's'. If
647 * 's' is NULL, disables STP.
649 * Returns 0 if successful, otherwise a positive errno value. */
651 ofproto_set_stp(struct ofproto
*ofproto
,
652 const struct ofproto_stp_settings
*s
)
654 return (ofproto
->ofproto_class
->set_stp
655 ? ofproto
->ofproto_class
->set_stp(ofproto
, s
)
659 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
660 * 'enabled' member of 's' is false, then the other members are not
663 * Returns 0 if successful, otherwise a positive errno value. */
665 ofproto_get_stp_status(struct ofproto
*ofproto
,
666 struct ofproto_stp_status
*s
)
668 return (ofproto
->ofproto_class
->get_stp_status
669 ? ofproto
->ofproto_class
->get_stp_status(ofproto
, s
)
673 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
674 * in 's'. The caller is responsible for assigning STP port numbers
675 * (using the 'port_num' member in the range of 1 through 255, inclusive)
676 * and ensuring there are no duplicates. If the 's' is NULL, then STP
677 * is disabled on the port.
679 * Returns 0 if successful, otherwise a positive errno value.*/
681 ofproto_port_set_stp(struct ofproto
*ofproto
, uint16_t ofp_port
,
682 const struct ofproto_port_stp_settings
*s
)
684 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
686 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16
,
687 ofproto
->name
, ofp_port
);
691 return (ofproto
->ofproto_class
->set_stp_port
692 ? ofproto
->ofproto_class
->set_stp_port(ofport
, s
)
696 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
697 * 's'. If the 'enabled' member in 's' is false, then the other members
698 * are not meaningful.
700 * Returns 0 if successful, otherwise a positive errno value.*/
702 ofproto_port_get_stp_status(struct ofproto
*ofproto
, uint16_t ofp_port
,
703 struct ofproto_port_stp_status
*s
)
705 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
707 VLOG_WARN_RL(&rl
, "%s: cannot get STP status on nonexistent "
708 "port %"PRIu16
, ofproto
->name
, ofp_port
);
712 return (ofproto
->ofproto_class
->get_stp_port_status
713 ? ofproto
->ofproto_class
->get_stp_port_status(ofport
, s
)
717 /* Queue DSCP configuration. */
719 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
720 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
721 * to implement QoS. Instead, it is used to implement features which require
722 * knowledge of what queues exist on a port, and some basic information about
725 * Returns 0 if successful, otherwise a positive errno value. */
727 ofproto_port_set_queues(struct ofproto
*ofproto
, uint16_t ofp_port
,
728 const struct ofproto_port_queue
*queues
,
731 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
734 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16
,
735 ofproto
->name
, ofp_port
);
739 return (ofproto
->ofproto_class
->set_queues
740 ? ofproto
->ofproto_class
->set_queues(ofport
, queues
, n_queues
)
744 /* Connectivity Fault Management configuration. */
746 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
748 ofproto_port_clear_cfm(struct ofproto
*ofproto
, uint16_t ofp_port
)
750 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
751 if (ofport
&& ofproto
->ofproto_class
->set_cfm
) {
752 ofproto
->ofproto_class
->set_cfm(ofport
, NULL
);
756 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
757 * basic configuration from the configuration members in 'cfm', and the remote
758 * maintenance point ID from remote_mpid. Ignores the statistics members of
761 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
763 ofproto_port_set_cfm(struct ofproto
*ofproto
, uint16_t ofp_port
,
764 const struct cfm_settings
*s
)
766 struct ofport
*ofport
;
769 ofport
= ofproto_get_port(ofproto
, ofp_port
);
771 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16
,
772 ofproto
->name
, ofp_port
);
776 /* XXX: For configuration simplicity, we only support one remote_mpid
777 * outside of the CFM module. It's not clear if this is the correct long
778 * term solution or not. */
779 error
= (ofproto
->ofproto_class
->set_cfm
780 ? ofproto
->ofproto_class
->set_cfm(ofport
, s
)
783 VLOG_WARN("%s: CFM configuration on port %"PRIu16
" (%s) failed (%s)",
784 ofproto
->name
, ofp_port
, netdev_get_name(ofport
->netdev
),
789 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
790 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
791 * 0 if LACP partner information is not current (generally indicating a
792 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
794 ofproto_port_is_lacp_current(struct ofproto
*ofproto
, uint16_t ofp_port
)
796 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
797 return (ofport
&& ofproto
->ofproto_class
->port_is_lacp_current
798 ? ofproto
->ofproto_class
->port_is_lacp_current(ofport
)
804 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
805 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
806 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
807 * configuration plus, if there is more than one slave, a bonding
810 * If 'aux' is already registered then this function updates its configuration
811 * to 's'. Otherwise, this function registers a new bundle.
813 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
816 ofproto_bundle_register(struct ofproto
*ofproto
, void *aux
,
817 const struct ofproto_bundle_settings
*s
)
819 return (ofproto
->ofproto_class
->bundle_set
820 ? ofproto
->ofproto_class
->bundle_set(ofproto
, aux
, s
)
824 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
825 * If no such bundle has been registered, this has no effect. */
827 ofproto_bundle_unregister(struct ofproto
*ofproto
, void *aux
)
829 return ofproto_bundle_register(ofproto
, aux
, NULL
);
833 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
834 * If 'aux' is already registered then this function updates its configuration
835 * to 's'. Otherwise, this function registers a new mirror. */
837 ofproto_mirror_register(struct ofproto
*ofproto
, void *aux
,
838 const struct ofproto_mirror_settings
*s
)
840 return (ofproto
->ofproto_class
->mirror_set
841 ? ofproto
->ofproto_class
->mirror_set(ofproto
, aux
, s
)
845 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
846 * If no mirror has been registered, this has no effect. */
848 ofproto_mirror_unregister(struct ofproto
*ofproto
, void *aux
)
850 return ofproto_mirror_register(ofproto
, aux
, NULL
);
853 /* Retrieves statistics from mirror associated with client data pointer
854 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
855 * 'bytes', respectively. If a particular counters is not supported,
856 * the appropriate argument is set to UINT64_MAX. */
858 ofproto_mirror_get_stats(struct ofproto
*ofproto
, void *aux
,
859 uint64_t *packets
, uint64_t *bytes
)
861 if (!ofproto
->ofproto_class
->mirror_get_stats
) {
862 *packets
= *bytes
= UINT64_MAX
;
866 return ofproto
->ofproto_class
->mirror_get_stats(ofproto
, aux
,
870 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
871 * which all packets are flooded, instead of using MAC learning. If
872 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
874 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
877 ofproto_set_flood_vlans(struct ofproto
*ofproto
, unsigned long *flood_vlans
)
879 return (ofproto
->ofproto_class
->set_flood_vlans
880 ? ofproto
->ofproto_class
->set_flood_vlans(ofproto
, flood_vlans
)
884 /* Returns true if 'aux' is a registered bundle that is currently in use as the
885 * output for a mirror. */
887 ofproto_is_mirror_output_bundle(const struct ofproto
*ofproto
, void *aux
)
889 return (ofproto
->ofproto_class
->is_mirror_output_bundle
890 ? ofproto
->ofproto_class
->is_mirror_output_bundle(ofproto
, aux
)
894 /* Configuration of OpenFlow tables. */
896 /* Returns the number of OpenFlow tables in 'ofproto'. */
898 ofproto_get_n_tables(const struct ofproto
*ofproto
)
900 return ofproto
->n_tables
;
903 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
904 * settings from 's'. 'table_id' must be in the range 0 through the number of
905 * OpenFlow tables in 'ofproto' minus 1, inclusive.
907 * For read-only tables, only the name may be configured. */
909 ofproto_configure_table(struct ofproto
*ofproto
, int table_id
,
910 const struct ofproto_table_settings
*s
)
912 struct oftable
*table
;
914 assert(table_id
>= 0 && table_id
< ofproto
->n_tables
);
915 table
= &ofproto
->tables
[table_id
];
917 oftable_set_name(table
, s
->name
);
919 if (table
->flags
& OFTABLE_READONLY
) {
924 oftable_enable_eviction(table
, s
->groups
, s
->n_groups
);
926 oftable_disable_eviction(table
);
929 table
->max_flows
= s
->max_flows
;
930 if (classifier_count(&table
->cls
) > table
->max_flows
931 && table
->eviction_fields
) {
932 /* 'table' contains more flows than allowed. We might not be able to
933 * evict them right away because of the asynchronous nature of flow
934 * table changes. Schedule eviction for later. */
935 switch (ofproto
->state
) {
937 ofproto
->state
= S_EVICT
;
941 /* We're already deleting flows, nothing more to do. */
948 ofproto_has_snoops(const struct ofproto
*ofproto
)
950 return connmgr_has_snoops(ofproto
->connmgr
);
954 ofproto_get_snoops(const struct ofproto
*ofproto
, struct sset
*snoops
)
956 connmgr_get_snoops(ofproto
->connmgr
, snoops
);
960 ofproto_flush__(struct ofproto
*ofproto
)
962 struct ofopgroup
*group
;
963 struct oftable
*table
;
965 if (ofproto
->ofproto_class
->flush
) {
966 ofproto
->ofproto_class
->flush(ofproto
);
969 group
= ofopgroup_create_unattached(ofproto
);
970 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
971 struct rule
*rule
, *next_rule
;
972 struct cls_cursor cursor
;
974 if (table
->flags
& OFTABLE_HIDDEN
) {
978 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
979 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
980 if (!rule
->pending
) {
981 ofoperation_create(group
, rule
, OFOPERATION_DELETE
,
983 oftable_remove_rule(rule
);
984 ofproto
->ofproto_class
->rule_destruct(rule
);
988 ofopgroup_submit(group
);
992 ofproto_destroy__(struct ofproto
*ofproto
)
994 struct oftable
*table
;
996 assert(list_is_empty(&ofproto
->pending
));
997 assert(!ofproto
->n_pending
);
999 connmgr_destroy(ofproto
->connmgr
);
1001 hmap_remove(&all_ofprotos
, &ofproto
->hmap_node
);
1002 free(ofproto
->name
);
1003 free(ofproto
->type
);
1004 free(ofproto
->mfr_desc
);
1005 free(ofproto
->hw_desc
);
1006 free(ofproto
->sw_desc
);
1007 free(ofproto
->serial_desc
);
1008 free(ofproto
->dp_desc
);
1009 hmap_destroy(&ofproto
->ports
);
1010 shash_destroy(&ofproto
->port_by_name
);
1012 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
1013 oftable_destroy(table
);
1015 free(ofproto
->tables
);
1017 hmap_destroy(&ofproto
->deletions
);
1019 free(ofproto
->vlan_bitmap
);
1021 ofproto
->ofproto_class
->dealloc(ofproto
);
1025 ofproto_destroy(struct ofproto
*p
)
1027 struct ofport
*ofport
, *next_ofport
;
1034 HMAP_FOR_EACH_SAFE (ofport
, next_ofport
, hmap_node
, &p
->ports
) {
1035 ofport_destroy(ofport
);
1038 p
->ofproto_class
->destruct(p
);
1039 ofproto_destroy__(p
);
1042 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1043 * kernel datapath, for example, this destroys the datapath in the kernel, and
1044 * with the netdev-based datapath, it tears down the data structures that
1045 * represent the datapath.
1047 * The datapath should not be currently open as an ofproto. */
1049 ofproto_delete(const char *name
, const char *type
)
1051 const struct ofproto_class
*class = ofproto_class_find__(type
);
1052 return (!class ? EAFNOSUPPORT
1053 : !class->del
? EACCES
1054 : class->del(type
, name
));
1058 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
1060 if (error
== ENOBUFS
) {
1061 reinit_ports(ofproto
);
1062 } else if (!error
) {
1063 update_port(ofproto
, devname
);
1069 ofproto_run(struct ofproto
*p
)
1071 struct sset changed_netdevs
;
1072 const char *changed_netdev
;
1073 struct ofport
*ofport
;
1076 error
= p
->ofproto_class
->run(p
);
1077 if (error
&& error
!= EAGAIN
) {
1078 VLOG_ERR_RL(&rl
, "%s: run failed (%s)", p
->name
, strerror(error
));
1081 if (p
->ofproto_class
->port_poll
) {
1084 while ((error
= p
->ofproto_class
->port_poll(p
, &devname
)) != EAGAIN
) {
1085 process_port_change(p
, error
, devname
);
1089 /* Update OpenFlow port status for any port whose netdev has changed.
1091 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1092 * destroyed, so it's not safe to update ports directly from the
1093 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1094 * need this two-phase approach. */
1095 sset_init(&changed_netdevs
);
1096 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1097 unsigned int change_seq
= netdev_change_seq(ofport
->netdev
);
1098 if (ofport
->change_seq
!= change_seq
) {
1099 ofport
->change_seq
= change_seq
;
1100 sset_add(&changed_netdevs
, netdev_get_name(ofport
->netdev
));
1103 SSET_FOR_EACH (changed_netdev
, &changed_netdevs
) {
1104 update_port(p
, changed_netdev
);
1106 sset_destroy(&changed_netdevs
);
1110 connmgr_run(p
->connmgr
, handle_openflow
);
1114 connmgr_run(p
->connmgr
, NULL
);
1116 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1117 p
->state
= S_OPENFLOW
;
1122 connmgr_run(p
->connmgr
, NULL
);
1124 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1125 connmgr_flushed(p
->connmgr
);
1126 p
->state
= S_OPENFLOW
;
1134 if (time_msec() >= p
->next_op_report
) {
1135 long long int ago
= (time_msec() - p
->first_op
) / 1000;
1136 long long int interval
= (p
->last_op
- p
->first_op
) / 1000;
1140 ds_put_format(&s
, "%d flow_mods ",
1141 p
->n_add
+ p
->n_delete
+ p
->n_modify
);
1142 if (interval
== ago
) {
1143 ds_put_format(&s
, "in the last %lld s", ago
);
1144 } else if (interval
) {
1145 ds_put_format(&s
, "in the %lld s starting %lld s ago",
1148 ds_put_format(&s
, "%lld s ago", ago
);
1151 ds_put_cstr(&s
, " (");
1153 ds_put_format(&s
, "%d adds, ", p
->n_add
);
1156 ds_put_format(&s
, "%d deletes, ", p
->n_delete
);
1159 ds_put_format(&s
, "%d modifications, ", p
->n_modify
);
1162 ds_put_char(&s
, ')');
1164 VLOG_INFO("%s: %s", p
->name
, ds_cstr(&s
));
1167 p
->n_add
= p
->n_delete
= p
->n_modify
= 0;
1168 p
->next_op_report
= LLONG_MAX
;
1174 /* Performs periodic activity required by 'ofproto' that needs to be done
1175 * with the least possible latency.
1177 * It makes sense to call this function a couple of times per poll loop, to
1178 * provide a significant performance boost on some benchmarks with the
1179 * ofproto-dpif implementation. */
1181 ofproto_run_fast(struct ofproto
*p
)
1185 error
= p
->ofproto_class
->run_fast
? p
->ofproto_class
->run_fast(p
) : 0;
1186 if (error
&& error
!= EAGAIN
) {
1187 VLOG_ERR_RL(&rl
, "%s: fastpath run failed (%s)",
1188 p
->name
, strerror(error
));
1194 ofproto_wait(struct ofproto
*p
)
1196 struct ofport
*ofport
;
1198 p
->ofproto_class
->wait(p
);
1199 if (p
->ofproto_class
->port_poll_wait
) {
1200 p
->ofproto_class
->port_poll_wait(p
);
1203 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1204 if (ofport
->change_seq
!= netdev_change_seq(ofport
->netdev
)) {
1205 poll_immediate_wake();
1211 connmgr_wait(p
->connmgr
, true);
1216 connmgr_wait(p
->connmgr
, false);
1217 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1218 poll_immediate_wake();
1225 ofproto_is_alive(const struct ofproto
*p
)
1227 return connmgr_has_controllers(p
->connmgr
);
1230 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1231 * memory_report(). */
1233 ofproto_get_memory_usage(const struct ofproto
*ofproto
, struct simap
*usage
)
1235 const struct oftable
*table
;
1236 unsigned int n_rules
;
1238 simap_increase(usage
, "ports", hmap_count(&ofproto
->ports
));
1239 simap_increase(usage
, "ops",
1240 ofproto
->n_pending
+ hmap_count(&ofproto
->deletions
));
1243 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
1244 n_rules
+= classifier_count(&table
->cls
);
1246 simap_increase(usage
, "rules", n_rules
);
1248 if (ofproto
->ofproto_class
->get_memory_usage
) {
1249 ofproto
->ofproto_class
->get_memory_usage(ofproto
, usage
);
1252 connmgr_get_memory_usage(ofproto
->connmgr
, usage
);
1256 ofproto_get_ofproto_controller_info(const struct ofproto
*ofproto
,
1259 connmgr_get_controller_info(ofproto
->connmgr
, info
);
1263 ofproto_free_ofproto_controller_info(struct shash
*info
)
1265 connmgr_free_controller_info(info
);
1268 /* Makes a deep copy of 'old' into 'port'. */
1270 ofproto_port_clone(struct ofproto_port
*port
, const struct ofproto_port
*old
)
1272 port
->name
= xstrdup(old
->name
);
1273 port
->type
= xstrdup(old
->type
);
1274 port
->ofp_port
= old
->ofp_port
;
1277 /* Frees memory allocated to members of 'ofproto_port'.
1279 * Do not call this function on an ofproto_port obtained from
1280 * ofproto_port_dump_next(): that function retains ownership of the data in the
1283 ofproto_port_destroy(struct ofproto_port
*ofproto_port
)
1285 free(ofproto_port
->name
);
1286 free(ofproto_port
->type
);
1289 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1291 * This function provides no status indication. An error status for the entire
1292 * dump operation is provided when it is completed by calling
1293 * ofproto_port_dump_done().
1296 ofproto_port_dump_start(struct ofproto_port_dump
*dump
,
1297 const struct ofproto
*ofproto
)
1299 dump
->ofproto
= ofproto
;
1300 dump
->error
= ofproto
->ofproto_class
->port_dump_start(ofproto
,
1304 /* Attempts to retrieve another port from 'dump', which must have been created
1305 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1306 * 'port' and returns true. On failure, returns false.
1308 * Failure might indicate an actual error or merely that the last port has been
1309 * dumped. An error status for the entire dump operation is provided when it
1310 * is completed by calling ofproto_port_dump_done().
1312 * The ofproto owns the data stored in 'port'. It will remain valid until at
1313 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1314 * ofproto_port_dump_done(). */
1316 ofproto_port_dump_next(struct ofproto_port_dump
*dump
,
1317 struct ofproto_port
*port
)
1319 const struct ofproto
*ofproto
= dump
->ofproto
;
1325 dump
->error
= ofproto
->ofproto_class
->port_dump_next(ofproto
, dump
->state
,
1328 ofproto
->ofproto_class
->port_dump_done(ofproto
, dump
->state
);
1334 /* Completes port table dump operation 'dump', which must have been created
1335 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1336 * error-free, otherwise a positive errno value describing the problem. */
1338 ofproto_port_dump_done(struct ofproto_port_dump
*dump
)
1340 const struct ofproto
*ofproto
= dump
->ofproto
;
1342 dump
->error
= ofproto
->ofproto_class
->port_dump_done(ofproto
,
1345 return dump
->error
== EOF
? 0 : dump
->error
;
1348 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
1349 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
1350 * is non-null). On failure, returns a positive errno value and sets
1351 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
1353 ofproto_port_add(struct ofproto
*ofproto
, struct netdev
*netdev
,
1354 uint16_t *ofp_portp
)
1359 error
= ofproto
->ofproto_class
->port_add(ofproto
, netdev
, &ofp_port
);
1361 update_port(ofproto
, netdev_get_name(netdev
));
1364 *ofp_portp
= error
? OFPP_NONE
: ofp_port
;
1369 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1370 * initializes '*port' appropriately; on failure, returns a positive errno
1373 * The caller owns the data in 'ofproto_port' and must free it with
1374 * ofproto_port_destroy() when it is no longer needed. */
1376 ofproto_port_query_by_name(const struct ofproto
*ofproto
, const char *devname
,
1377 struct ofproto_port
*port
)
1381 error
= ofproto
->ofproto_class
->port_query_by_name(ofproto
, devname
, port
);
1383 memset(port
, 0, sizeof *port
);
1388 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1389 * Returns 0 if successful, otherwise a positive errno. */
1391 ofproto_port_del(struct ofproto
*ofproto
, uint16_t ofp_port
)
1393 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
1394 const char *name
= ofport
? netdev_get_name(ofport
->netdev
) : "<unknown>";
1397 error
= ofproto
->ofproto_class
->port_del(ofproto
, ofp_port
);
1398 if (!error
&& ofport
) {
1399 /* 'name' is the netdev's name and update_port() is going to close the
1400 * netdev. Just in case update_port() refers to 'name' after it
1401 * destroys 'ofport', make a copy of it around the update_port()
1403 char *devname
= xstrdup(name
);
1404 update_port(ofproto
, devname
);
1410 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1411 * performs the 'n_actions' actions in 'actions'. The new flow will not
1414 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1415 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1416 * controllers; otherwise, it will be hidden.
1418 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1420 * This is a helper function for in-band control and fail-open. */
1422 ofproto_add_flow(struct ofproto
*ofproto
, const struct match
*match
,
1423 unsigned int priority
,
1424 const struct ofpact
*ofpacts
, size_t ofpacts_len
)
1426 const struct rule
*rule
;
1428 rule
= rule_from_cls_rule(classifier_find_match_exactly(
1429 &ofproto
->tables
[0].cls
, match
, priority
));
1430 if (!rule
|| !ofpacts_equal(rule
->ofpacts
, rule
->ofpacts_len
,
1431 ofpacts
, ofpacts_len
)) {
1432 struct ofputil_flow_mod fm
;
1434 memset(&fm
, 0, sizeof fm
);
1436 fm
.priority
= priority
;
1437 fm
.buffer_id
= UINT32_MAX
;
1438 fm
.ofpacts
= xmemdup(ofpacts
, ofpacts_len
);
1439 fm
.ofpacts_len
= ofpacts_len
;
1440 add_flow(ofproto
, NULL
, &fm
, NULL
);
1445 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1446 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1447 * operation cannot be initiated now but may be retried later.
1449 * This is a helper function for in-band control and fail-open. */
1451 ofproto_flow_mod(struct ofproto
*ofproto
, const struct ofputil_flow_mod
*fm
)
1453 return handle_flow_mod__(ofproto
, NULL
, fm
, NULL
);
1456 /* Searches for a rule with matching criteria exactly equal to 'target' in
1457 * ofproto's table 0 and, if it finds one, deletes it.
1459 * This is a helper function for in-band control and fail-open. */
1461 ofproto_delete_flow(struct ofproto
*ofproto
,
1462 const struct match
*target
, unsigned int priority
)
1466 rule
= rule_from_cls_rule(classifier_find_match_exactly(
1467 &ofproto
->tables
[0].cls
, target
, priority
));
1469 /* No such rule -> success. */
1471 } else if (rule
->pending
) {
1472 /* An operation on the rule is already pending -> failure.
1473 * Caller must retry later if it's important. */
1476 /* Initiate deletion -> success. */
1477 struct ofopgroup
*group
= ofopgroup_create_unattached(ofproto
);
1478 ofoperation_create(group
, rule
, OFOPERATION_DELETE
, OFPRR_DELETE
);
1479 oftable_remove_rule(rule
);
1480 ofproto
->ofproto_class
->rule_destruct(rule
);
1481 ofopgroup_submit(group
);
1487 /* Starts the process of deleting all of the flows from all of ofproto's flow
1488 * tables and then reintroducing the flows required by in-band control and
1489 * fail-open. The process will complete in a later call to ofproto_run(). */
1491 ofproto_flush_flows(struct ofproto
*ofproto
)
1493 COVERAGE_INC(ofproto_flush
);
1494 ofproto
->state
= S_FLUSH
;
1498 reinit_ports(struct ofproto
*p
)
1500 struct ofproto_port_dump dump
;
1501 struct sset devnames
;
1502 struct ofport
*ofport
;
1503 struct ofproto_port ofproto_port
;
1504 const char *devname
;
1506 COVERAGE_INC(ofproto_reinit_ports
);
1508 sset_init(&devnames
);
1509 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1510 sset_add(&devnames
, netdev_get_name(ofport
->netdev
));
1512 OFPROTO_PORT_FOR_EACH (&ofproto_port
, &dump
, p
) {
1513 sset_add(&devnames
, ofproto_port
.name
);
1516 SSET_FOR_EACH (devname
, &devnames
) {
1517 update_port(p
, devname
);
1519 sset_destroy(&devnames
);
1522 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1523 * pointer if the netdev cannot be opened. On success, also fills in
1525 static struct netdev
*
1526 ofport_open(const struct ofproto
*ofproto
,
1527 const struct ofproto_port
*ofproto_port
,
1528 struct ofputil_phy_port
*pp
)
1530 enum netdev_flags flags
;
1531 struct netdev
*netdev
;
1534 error
= netdev_open(ofproto_port
->name
, ofproto_port
->type
, &netdev
);
1536 VLOG_WARN_RL(&rl
, "%s: ignoring port %s (%"PRIu16
") because netdev %s "
1537 "cannot be opened (%s)",
1539 ofproto_port
->name
, ofproto_port
->ofp_port
,
1540 ofproto_port
->name
, strerror(error
));
1544 pp
->port_no
= ofproto_port
->ofp_port
;
1545 netdev_get_etheraddr(netdev
, pp
->hw_addr
);
1546 ovs_strlcpy(pp
->name
, ofproto_port
->name
, sizeof pp
->name
);
1547 netdev_get_flags(netdev
, &flags
);
1548 pp
->config
= flags
& NETDEV_UP
? 0 : OFPUTIL_PC_PORT_DOWN
;
1549 pp
->state
= netdev_get_carrier(netdev
) ? 0 : OFPUTIL_PS_LINK_DOWN
;
1550 netdev_get_features(netdev
, &pp
->curr
, &pp
->advertised
,
1551 &pp
->supported
, &pp
->peer
);
1552 pp
->curr_speed
= netdev_features_to_bps(pp
->curr
);
1553 pp
->max_speed
= netdev_features_to_bps(pp
->supported
);
1558 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1559 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1562 ofport_equal(const struct ofputil_phy_port
*a
,
1563 const struct ofputil_phy_port
*b
)
1565 return (eth_addr_equals(a
->hw_addr
, b
->hw_addr
)
1566 && a
->state
== b
->state
1567 && !((a
->config
^ b
->config
) & OFPUTIL_PC_PORT_DOWN
)
1568 && a
->curr
== b
->curr
1569 && a
->advertised
== b
->advertised
1570 && a
->supported
== b
->supported
1571 && a
->peer
== b
->peer
1572 && a
->curr_speed
== b
->curr_speed
1573 && a
->max_speed
== b
->max_speed
);
1576 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1577 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1578 * one with the same name or port number). */
1580 ofport_install(struct ofproto
*p
,
1581 struct netdev
*netdev
, const struct ofputil_phy_port
*pp
)
1583 const char *netdev_name
= netdev_get_name(netdev
);
1584 struct ofport
*ofport
;
1587 /* Create ofport. */
1588 ofport
= p
->ofproto_class
->port_alloc();
1593 ofport
->ofproto
= p
;
1594 ofport
->netdev
= netdev
;
1595 ofport
->change_seq
= netdev_change_seq(netdev
);
1597 ofport
->ofp_port
= pp
->port_no
;
1599 /* Add port to 'p'. */
1600 hmap_insert(&p
->ports
, &ofport
->hmap_node
, hash_int(ofport
->ofp_port
, 0));
1601 shash_add(&p
->port_by_name
, netdev_name
, ofport
);
1603 update_mtu(p
, ofport
);
1605 /* Let the ofproto_class initialize its private data. */
1606 error
= p
->ofproto_class
->port_construct(ofport
);
1610 connmgr_send_port_status(p
->connmgr
, pp
, OFPPR_ADD
);
1614 VLOG_WARN_RL(&rl
, "%s: could not add port %s (%s)",
1615 p
->name
, netdev_name
, strerror(error
));
1617 ofport_destroy__(ofport
);
1619 netdev_close(netdev
);
1623 /* Removes 'ofport' from 'p' and destroys it. */
1625 ofport_remove(struct ofport
*ofport
)
1627 connmgr_send_port_status(ofport
->ofproto
->connmgr
, &ofport
->pp
,
1629 ofport_destroy(ofport
);
1632 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1635 ofport_remove_with_name(struct ofproto
*ofproto
, const char *name
)
1637 struct ofport
*port
= shash_find_data(&ofproto
->port_by_name
, name
);
1639 ofport_remove(port
);
1643 /* Updates 'port' with new 'pp' description.
1645 * Does not handle a name or port number change. The caller must implement
1646 * such a change as a delete followed by an add. */
1648 ofport_modified(struct ofport
*port
, struct ofputil_phy_port
*pp
)
1650 memcpy(port
->pp
.hw_addr
, pp
->hw_addr
, ETH_ADDR_LEN
);
1651 port
->pp
.config
= ((port
->pp
.config
& ~OFPUTIL_PC_PORT_DOWN
)
1652 | (pp
->config
& OFPUTIL_PC_PORT_DOWN
));
1653 port
->pp
.state
= pp
->state
;
1654 port
->pp
.curr
= pp
->curr
;
1655 port
->pp
.advertised
= pp
->advertised
;
1656 port
->pp
.supported
= pp
->supported
;
1657 port
->pp
.peer
= pp
->peer
;
1658 port
->pp
.curr_speed
= pp
->curr_speed
;
1659 port
->pp
.max_speed
= pp
->max_speed
;
1661 connmgr_send_port_status(port
->ofproto
->connmgr
, &port
->pp
, OFPPR_MODIFY
);
1664 /* Update OpenFlow 'state' in 'port' and notify controller. */
1666 ofproto_port_set_state(struct ofport
*port
, enum ofputil_port_state state
)
1668 if (port
->pp
.state
!= state
) {
1669 port
->pp
.state
= state
;
1670 connmgr_send_port_status(port
->ofproto
->connmgr
, &port
->pp
,
1676 ofproto_port_unregister(struct ofproto
*ofproto
, uint16_t ofp_port
)
1678 struct ofport
*port
= ofproto_get_port(ofproto
, ofp_port
);
1680 if (port
->ofproto
->ofproto_class
->set_realdev
) {
1681 port
->ofproto
->ofproto_class
->set_realdev(port
, 0, 0);
1683 if (port
->ofproto
->ofproto_class
->set_stp_port
) {
1684 port
->ofproto
->ofproto_class
->set_stp_port(port
, NULL
);
1686 if (port
->ofproto
->ofproto_class
->set_cfm
) {
1687 port
->ofproto
->ofproto_class
->set_cfm(port
, NULL
);
1689 if (port
->ofproto
->ofproto_class
->bundle_remove
) {
1690 port
->ofproto
->ofproto_class
->bundle_remove(port
);
1696 ofport_destroy__(struct ofport
*port
)
1698 struct ofproto
*ofproto
= port
->ofproto
;
1699 const char *name
= netdev_get_name(port
->netdev
);
1701 hmap_remove(&ofproto
->ports
, &port
->hmap_node
);
1702 shash_delete(&ofproto
->port_by_name
,
1703 shash_find(&ofproto
->port_by_name
, name
));
1705 netdev_close(port
->netdev
);
1706 ofproto
->ofproto_class
->port_dealloc(port
);
1710 ofport_destroy(struct ofport
*port
)
1713 port
->ofproto
->ofproto_class
->port_destruct(port
);
1714 ofport_destroy__(port
);
1719 ofproto_get_port(const struct ofproto
*ofproto
, uint16_t ofp_port
)
1721 struct ofport
*port
;
1723 HMAP_FOR_EACH_IN_BUCKET (port
, hmap_node
,
1724 hash_int(ofp_port
, 0), &ofproto
->ports
) {
1725 if (port
->ofp_port
== ofp_port
) {
1733 ofproto_port_get_stats(const struct ofport
*port
, struct netdev_stats
*stats
)
1735 struct ofproto
*ofproto
= port
->ofproto
;
1738 if (ofproto
->ofproto_class
->port_get_stats
) {
1739 error
= ofproto
->ofproto_class
->port_get_stats(port
, stats
);
1748 update_port(struct ofproto
*ofproto
, const char *name
)
1750 struct ofproto_port ofproto_port
;
1751 struct ofputil_phy_port pp
;
1752 struct netdev
*netdev
;
1753 struct ofport
*port
;
1755 COVERAGE_INC(ofproto_update_port
);
1757 /* Fetch 'name''s location and properties from the datapath. */
1758 netdev
= (!ofproto_port_query_by_name(ofproto
, name
, &ofproto_port
)
1759 ? ofport_open(ofproto
, &ofproto_port
, &pp
)
1762 port
= ofproto_get_port(ofproto
, ofproto_port
.ofp_port
);
1763 if (port
&& !strcmp(netdev_get_name(port
->netdev
), name
)) {
1764 struct netdev
*old_netdev
= port
->netdev
;
1766 /* 'name' hasn't changed location. Any properties changed? */
1767 if (!ofport_equal(&port
->pp
, &pp
)) {
1768 ofport_modified(port
, &pp
);
1771 update_mtu(ofproto
, port
);
1773 /* Install the newly opened netdev in case it has changed.
1774 * Don't close the old netdev yet in case port_modified has to
1775 * remove a retained reference to it.*/
1776 port
->netdev
= netdev
;
1777 port
->change_seq
= netdev_change_seq(netdev
);
1779 if (port
->ofproto
->ofproto_class
->port_modified
) {
1780 port
->ofproto
->ofproto_class
->port_modified(port
);
1783 netdev_close(old_netdev
);
1785 /* If 'port' is nonnull then its name differs from 'name' and thus
1786 * we should delete it. If we think there's a port named 'name'
1787 * then its port number must be wrong now so delete it too. */
1789 ofport_remove(port
);
1791 ofport_remove_with_name(ofproto
, name
);
1792 ofport_install(ofproto
, netdev
, &pp
);
1795 /* Any port named 'name' is gone now. */
1796 ofport_remove_with_name(ofproto
, name
);
1798 ofproto_port_destroy(&ofproto_port
);
1802 init_ports(struct ofproto
*p
)
1804 struct ofproto_port_dump dump
;
1805 struct ofproto_port ofproto_port
;
1807 OFPROTO_PORT_FOR_EACH (&ofproto_port
, &dump
, p
) {
1808 uint16_t ofp_port
= ofproto_port
.ofp_port
;
1809 if (ofproto_get_port(p
, ofp_port
)) {
1810 VLOG_WARN_RL(&rl
, "%s: ignoring duplicate port %"PRIu16
" "
1811 "in datapath", p
->name
, ofp_port
);
1812 } else if (shash_find(&p
->port_by_name
, ofproto_port
.name
)) {
1813 VLOG_WARN_RL(&rl
, "%s: ignoring duplicate device %s in datapath",
1814 p
->name
, ofproto_port
.name
);
1816 struct ofputil_phy_port pp
;
1817 struct netdev
*netdev
;
1819 netdev
= ofport_open(p
, &ofproto_port
, &pp
);
1821 ofport_install(p
, netdev
, &pp
);
1829 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
1830 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
1832 find_min_mtu(struct ofproto
*p
)
1834 struct ofport
*ofport
;
1837 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1838 struct netdev
*netdev
= ofport
->netdev
;
1841 /* Skip any internal ports, since that's what we're trying to
1843 if (!strcmp(netdev_get_type(netdev
), "internal")) {
1847 if (netdev_get_mtu(netdev
, &dev_mtu
)) {
1850 if (!mtu
|| dev_mtu
< mtu
) {
1855 return mtu
? mtu
: ETH_PAYLOAD_MAX
;
1858 /* Update MTU of all datapath devices on 'p' to the minimum of the
1859 * non-datapath ports in event of 'port' added or changed. */
1861 update_mtu(struct ofproto
*p
, struct ofport
*port
)
1863 struct ofport
*ofport
;
1864 struct netdev
*netdev
= port
->netdev
;
1865 int dev_mtu
, old_min
;
1867 if (netdev_get_mtu(netdev
, &dev_mtu
)) {
1871 if (!strcmp(netdev_get_type(port
->netdev
), "internal")) {
1872 if (dev_mtu
> p
->min_mtu
) {
1873 if (!netdev_set_mtu(port
->netdev
, p
->min_mtu
)) {
1874 dev_mtu
= p
->min_mtu
;
1877 port
->mtu
= dev_mtu
;
1881 /* For non-internal port find new min mtu. */
1882 old_min
= p
->min_mtu
;
1883 port
->mtu
= dev_mtu
;
1884 p
->min_mtu
= find_min_mtu(p
);
1885 if (p
->min_mtu
== old_min
) {
1889 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1890 struct netdev
*netdev
= ofport
->netdev
;
1892 if (!strcmp(netdev_get_type(netdev
), "internal")) {
1893 if (!netdev_set_mtu(netdev
, p
->min_mtu
)) {
1894 ofport
->mtu
= p
->min_mtu
;
1901 ofproto_rule_destroy__(struct rule
*rule
)
1904 cls_rule_destroy(&rule
->cr
);
1905 free(rule
->ofpacts
);
1906 rule
->ofproto
->ofproto_class
->rule_dealloc(rule
);
1910 /* This function allows an ofproto implementation to destroy any rules that
1911 * remain when its ->destruct() function is called. The caller must have
1912 * already uninitialized any derived members of 'rule' (step 5 described in the
1913 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
1914 * This function implements steps 6 and 7.
1916 * This function should only be called from an ofproto implementation's
1917 * ->destruct() function. It is not suitable elsewhere. */
1919 ofproto_rule_destroy(struct rule
*rule
)
1921 assert(!rule
->pending
);
1922 oftable_remove_rule(rule
);
1923 ofproto_rule_destroy__(rule
);
1926 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1927 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
1929 ofproto_rule_has_out_port(const struct rule
*rule
, uint16_t port
)
1931 return (port
== OFPP_NONE
1932 || ofpacts_output_to_port(rule
->ofpacts
, rule
->ofpacts_len
, port
));
1935 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
1936 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
1938 ofoperation_has_out_port(const struct ofoperation
*op
, uint16_t out_port
)
1940 if (ofproto_rule_has_out_port(op
->rule
, out_port
)) {
1945 case OFOPERATION_ADD
:
1946 return op
->victim
&& ofproto_rule_has_out_port(op
->victim
, out_port
);
1948 case OFOPERATION_DELETE
:
1951 case OFOPERATION_MODIFY
:
1952 return ofpacts_output_to_port(op
->ofpacts
, op
->ofpacts_len
, out_port
);
1958 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1959 * statistics appropriately. 'packet' must have at least sizeof(struct
1960 * ofp_packet_in) bytes of headroom.
1962 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1963 * with statistics for 'packet' either way.
1965 * Takes ownership of 'packet'. */
1967 rule_execute(struct rule
*rule
, uint16_t in_port
, struct ofpbuf
*packet
)
1971 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
1973 flow_extract(packet
, 0, NULL
, in_port
, &flow
);
1974 return rule
->ofproto
->ofproto_class
->rule_execute(rule
, &flow
, packet
);
1977 /* Returns true if 'rule' should be hidden from the controller.
1979 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1980 * (e.g. by in-band control) and are intentionally hidden from the
1983 ofproto_rule_is_hidden(const struct rule
*rule
)
1985 return rule
->cr
.priority
> UINT16_MAX
;
1988 static enum oftable_flags
1989 rule_get_flags(const struct rule
*rule
)
1991 return rule
->ofproto
->tables
[rule
->table_id
].flags
;
1995 rule_is_modifiable(const struct rule
*rule
)
1997 return !(rule_get_flags(rule
) & OFTABLE_READONLY
);
2001 handle_echo_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2003 ofconn_send_reply(ofconn
, make_echo_reply(oh
));
2008 handle_features_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2010 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2011 struct ofputil_switch_features features
;
2012 struct ofport
*port
;
2016 ofproto
->ofproto_class
->get_features(ofproto
, &arp_match_ip
,
2018 assert(features
.actions
& OFPUTIL_A_OUTPUT
); /* sanity check */
2020 features
.datapath_id
= ofproto
->datapath_id
;
2021 features
.n_buffers
= pktbuf_capacity();
2022 features
.n_tables
= ofproto
->n_tables
;
2023 features
.capabilities
= (OFPUTIL_C_FLOW_STATS
| OFPUTIL_C_TABLE_STATS
|
2024 OFPUTIL_C_PORT_STATS
| OFPUTIL_C_QUEUE_STATS
);
2026 features
.capabilities
|= OFPUTIL_C_ARP_MATCH_IP
;
2029 b
= ofputil_encode_switch_features(&features
, ofconn_get_protocol(ofconn
),
2031 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
2032 ofputil_put_switch_features_port(&port
->pp
, b
);
2035 ofconn_send_reply(ofconn
, b
);
2040 handle_get_config_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2042 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2043 struct ofp_switch_config
*osc
;
2044 enum ofp_config_flags flags
;
2048 buf
= ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY
, oh
, 0);
2049 osc
= ofpbuf_put_uninit(buf
, sizeof *osc
);
2050 flags
= ofproto
->frag_handling
;
2051 if (ofconn_get_invalid_ttl_to_controller(ofconn
)) {
2052 flags
|= OFPC_INVALID_TTL_TO_CONTROLLER
;
2054 osc
->flags
= htons(flags
);
2055 osc
->miss_send_len
= htons(ofconn_get_miss_send_len(ofconn
));
2056 ofconn_send_reply(ofconn
, buf
);
2062 handle_set_config(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2064 const struct ofp_switch_config
*osc
= ofpmsg_body(oh
);
2065 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2066 uint16_t flags
= ntohs(osc
->flags
);
2068 if (ofconn_get_type(ofconn
) != OFCONN_PRIMARY
2069 || ofconn_get_role(ofconn
) != NX_ROLE_SLAVE
) {
2070 enum ofp_config_flags cur
= ofproto
->frag_handling
;
2071 enum ofp_config_flags next
= flags
& OFPC_FRAG_MASK
;
2073 assert((cur
& OFPC_FRAG_MASK
) == cur
);
2075 if (ofproto
->ofproto_class
->set_frag_handling(ofproto
, next
)) {
2076 ofproto
->frag_handling
= next
;
2078 VLOG_WARN_RL(&rl
, "%s: unsupported fragment handling mode %s",
2080 ofputil_frag_handling_to_string(next
));
2084 ofconn_set_invalid_ttl_to_controller(ofconn
,
2085 (flags
& OFPC_INVALID_TTL_TO_CONTROLLER
));
2087 ofconn_set_miss_send_len(ofconn
, ntohs(osc
->miss_send_len
));
2092 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2093 * error message code for the caller to propagate upward. Otherwise, returns
2096 * The log message mentions 'msg_type'. */
2098 reject_slave_controller(struct ofconn
*ofconn
)
2100 if (ofconn_get_type(ofconn
) == OFCONN_PRIMARY
2101 && ofconn_get_role(ofconn
) == NX_ROLE_SLAVE
) {
2102 return OFPERR_OFPBRC_EPERM
;
2109 handle_packet_out(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2111 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2112 struct ofputil_packet_out po
;
2113 struct ofpbuf
*payload
;
2114 uint64_t ofpacts_stub
[1024 / 8];
2115 struct ofpbuf ofpacts
;
2119 COVERAGE_INC(ofproto_packet_out
);
2121 error
= reject_slave_controller(ofconn
);
2126 /* Decode message. */
2127 ofpbuf_use_stub(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
2128 error
= ofputil_decode_packet_out(&po
, oh
, &ofpacts
);
2130 goto exit_free_ofpacts
;
2132 if (po
.in_port
>= p
->max_ports
&& po
.in_port
< OFPP_MAX
) {
2133 error
= OFPERR_OFPBRC_BAD_PORT
;
2134 goto exit_free_ofpacts
;
2139 if (po
.buffer_id
!= UINT32_MAX
) {
2140 error
= ofconn_pktbuf_retrieve(ofconn
, po
.buffer_id
, &payload
, NULL
);
2141 if (error
|| !payload
) {
2142 goto exit_free_ofpacts
;
2145 payload
= xmalloc(sizeof *payload
);
2146 ofpbuf_use_const(payload
, po
.packet
, po
.packet_len
);
2149 /* Verify actions against packet, then send packet if successful. */
2150 flow_extract(payload
, 0, NULL
, po
.in_port
, &flow
);
2151 error
= ofpacts_check(po
.ofpacts
, po
.ofpacts_len
, &flow
, p
->max_ports
);
2153 error
= p
->ofproto_class
->packet_out(p
, payload
, &flow
,
2154 po
.ofpacts
, po
.ofpacts_len
);
2156 ofpbuf_delete(payload
);
2159 ofpbuf_uninit(&ofpacts
);
2165 update_port_config(struct ofport
*port
,
2166 enum ofputil_port_config config
,
2167 enum ofputil_port_config mask
)
2169 enum ofputil_port_config old_config
= port
->pp
.config
;
2170 enum ofputil_port_config toggle
;
2172 toggle
= (config
^ port
->pp
.config
) & mask
;
2173 if (toggle
& OFPUTIL_PC_PORT_DOWN
) {
2174 if (config
& OFPUTIL_PC_PORT_DOWN
) {
2175 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, true);
2177 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, true);
2179 toggle
&= ~OFPUTIL_PC_PORT_DOWN
;
2182 port
->pp
.config
^= toggle
;
2183 if (port
->pp
.config
!= old_config
) {
2184 port
->ofproto
->ofproto_class
->port_reconfigured(port
, old_config
);
2189 handle_port_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2191 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2192 struct ofputil_port_mod pm
;
2193 struct ofport
*port
;
2196 error
= reject_slave_controller(ofconn
);
2201 error
= ofputil_decode_port_mod(oh
, &pm
);
2206 port
= ofproto_get_port(p
, pm
.port_no
);
2208 return OFPERR_OFPPMFC_BAD_PORT
;
2209 } else if (!eth_addr_equals(port
->pp
.hw_addr
, pm
.hw_addr
)) {
2210 return OFPERR_OFPPMFC_BAD_HW_ADDR
;
2212 update_port_config(port
, pm
.config
, pm
.mask
);
2214 netdev_set_advertisements(port
->netdev
, pm
.advertise
);
2221 handle_desc_stats_request(struct ofconn
*ofconn
,
2222 const struct ofp_header
*request
)
2224 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2225 struct ofp_desc_stats
*ods
;
2228 msg
= ofpraw_alloc_stats_reply(request
, 0);
2229 ods
= ofpbuf_put_zeros(msg
, sizeof *ods
);
2230 ovs_strlcpy(ods
->mfr_desc
, p
->mfr_desc
, sizeof ods
->mfr_desc
);
2231 ovs_strlcpy(ods
->hw_desc
, p
->hw_desc
, sizeof ods
->hw_desc
);
2232 ovs_strlcpy(ods
->sw_desc
, p
->sw_desc
, sizeof ods
->sw_desc
);
2233 ovs_strlcpy(ods
->serial_num
, p
->serial_desc
, sizeof ods
->serial_num
);
2234 ovs_strlcpy(ods
->dp_desc
, p
->dp_desc
, sizeof ods
->dp_desc
);
2235 ofconn_send_reply(ofconn
, msg
);
2241 handle_table_stats_request(struct ofconn
*ofconn
,
2242 const struct ofp_header
*request
)
2244 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2245 struct ofp12_table_stats
*ots
;
2249 /* Set up default values.
2251 * ofp12_table_stats is used as a generic structure as
2252 * it is able to hold all the fields for ofp10_table_stats
2253 * and ofp11_table_stats (and of course itself).
2255 ots
= xcalloc(p
->n_tables
, sizeof *ots
);
2256 for (i
= 0; i
< p
->n_tables
; i
++) {
2257 ots
[i
].table_id
= i
;
2258 sprintf(ots
[i
].name
, "table%zu", i
);
2259 ots
[i
].match
= htonll(OFPXMT12_MASK
);
2260 ots
[i
].wildcards
= htonll(OFPXMT12_MASK
);
2261 ots
[i
].write_actions
= htonl(OFPAT11_OUTPUT
);
2262 ots
[i
].apply_actions
= htonl(OFPAT11_OUTPUT
);
2263 ots
[i
].write_setfields
= htonll(OFPXMT12_MASK
);
2264 ots
[i
].apply_setfields
= htonll(OFPXMT12_MASK
);
2265 ots
[i
].metadata_match
= htonll(UINT64_MAX
);
2266 ots
[i
].metadata_write
= htonll(UINT64_MAX
);
2267 ots
[i
].instructions
= htonl(OFPIT11_ALL
);
2268 ots
[i
].config
= htonl(OFPTC11_TABLE_MISS_MASK
);
2269 ots
[i
].max_entries
= htonl(1000000); /* An arbitrary big number. */
2270 ots
[i
].active_count
= htonl(classifier_count(&p
->tables
[i
].cls
));
2273 p
->ofproto_class
->get_tables(p
, ots
);
2275 for (i
= 0; i
< p
->n_tables
; i
++) {
2276 const struct oftable
*table
= &p
->tables
[i
];
2279 ovs_strzcpy(ots
[i
].name
, table
->name
, sizeof ots
[i
].name
);
2282 if (table
->max_flows
< ntohl(ots
[i
].max_entries
)) {
2283 ots
[i
].max_entries
= htonl(table
->max_flows
);
2287 msg
= ofputil_encode_table_stats_reply(ots
, p
->n_tables
, request
);
2288 ofconn_send_reply(ofconn
, msg
);
2296 append_port_stat(struct ofport
*port
, struct list
*replies
)
2298 struct ofputil_port_stats ops
= { .port_no
= port
->pp
.port_no
};
2300 /* Intentionally ignore return value, since errors will set
2301 * 'stats' to all-1s, which is correct for OpenFlow, and
2302 * netdev_get_stats() will log errors. */
2303 ofproto_port_get_stats(port
, &ops
.stats
);
2305 ofputil_append_port_stat(replies
, &ops
);
2309 handle_port_stats_request(struct ofconn
*ofconn
,
2310 const struct ofp_header
*request
)
2312 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2313 struct ofport
*port
;
2314 struct list replies
;
2318 error
= ofputil_decode_port_stats_request(request
, &port_no
);
2323 ofpmp_init(&replies
, request
);
2324 if (port_no
!= OFPP_NONE
) {
2325 port
= ofproto_get_port(p
, port_no
);
2327 append_port_stat(port
, &replies
);
2330 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
2331 append_port_stat(port
, &replies
);
2335 ofconn_send_replies(ofconn
, &replies
);
2340 handle_port_desc_stats_request(struct ofconn
*ofconn
,
2341 const struct ofp_header
*request
)
2343 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2344 enum ofp_version version
;
2345 struct ofport
*port
;
2346 struct list replies
;
2348 ofpmp_init(&replies
, request
);
2350 version
= ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn
));
2351 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
2352 ofputil_append_port_desc_stats_reply(version
, &port
->pp
, &replies
);
2355 ofconn_send_replies(ofconn
, &replies
);
2360 calc_flow_duration__(long long int start
, long long int now
,
2361 uint32_t *sec
, uint32_t *nsec
)
2363 long long int msecs
= now
- start
;
2364 *sec
= msecs
/ 1000;
2365 *nsec
= (msecs
% 1000) * (1000 * 1000);
2368 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2369 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2371 check_table_id(const struct ofproto
*ofproto
, uint8_t table_id
)
2373 return (table_id
== 0xff || table_id
< ofproto
->n_tables
2375 : OFPERR_OFPBRC_BAD_TABLE_ID
);
2379 static struct oftable
*
2380 next_visible_table(const struct ofproto
*ofproto
, uint8_t table_id
)
2382 struct oftable
*table
;
2384 for (table
= &ofproto
->tables
[table_id
];
2385 table
< &ofproto
->tables
[ofproto
->n_tables
];
2387 if (!(table
->flags
& OFTABLE_HIDDEN
)) {
2395 static struct oftable
*
2396 first_matching_table(const struct ofproto
*ofproto
, uint8_t table_id
)
2398 if (table_id
== 0xff) {
2399 return next_visible_table(ofproto
, 0);
2400 } else if (table_id
< ofproto
->n_tables
) {
2401 return &ofproto
->tables
[table_id
];
2407 static struct oftable
*
2408 next_matching_table(const struct ofproto
*ofproto
,
2409 const struct oftable
*table
, uint8_t table_id
)
2411 return (table_id
== 0xff
2412 ? next_visible_table(ofproto
, (table
- ofproto
->tables
) + 1)
2416 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2418 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2419 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2421 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2422 * only once, for that table. (This can be used to access tables marked
2425 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2426 * entered at all. (Perhaps you should have validated TABLE_ID with
2427 * check_table_id().)
2429 * All parameters are evaluated multiple times.
2431 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2432 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2434 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2436 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2437 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2438 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2441 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2442 * to 'out_port' are included.
2444 * Hidden rules are always omitted.
2446 * Returns 0 on success, otherwise an OpenFlow error code. */
2448 collect_rules_loose(struct ofproto
*ofproto
, uint8_t table_id
,
2449 const struct match
*match
,
2450 ovs_be64 cookie
, ovs_be64 cookie_mask
,
2451 uint16_t out_port
, struct list
*rules
)
2453 struct oftable
*table
;
2457 error
= check_table_id(ofproto
, table_id
);
2463 cls_rule_init(&cr
, match
, 0);
2464 FOR_EACH_MATCHING_TABLE (table
, table_id
, ofproto
) {
2465 struct cls_cursor cursor
;
2468 cls_cursor_init(&cursor
, &table
->cls
, &cr
);
2469 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
2470 if (rule
->pending
) {
2471 error
= OFPROTO_POSTPONE
;
2474 if (!ofproto_rule_is_hidden(rule
)
2475 && ofproto_rule_has_out_port(rule
, out_port
)
2476 && !((rule
->flow_cookie
^ cookie
) & cookie_mask
)) {
2477 list_push_back(rules
, &rule
->ofproto_node
);
2483 cls_rule_destroy(&cr
);
2487 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2488 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2489 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2492 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2493 * to 'out_port' are included.
2495 * Hidden rules are always omitted.
2497 * Returns 0 on success, otherwise an OpenFlow error code. */
2499 collect_rules_strict(struct ofproto
*ofproto
, uint8_t table_id
,
2500 const struct match
*match
, unsigned int priority
,
2501 ovs_be64 cookie
, ovs_be64 cookie_mask
,
2502 uint16_t out_port
, struct list
*rules
)
2504 struct oftable
*table
;
2508 error
= check_table_id(ofproto
, table_id
);
2514 cls_rule_init(&cr
, match
, priority
);
2515 FOR_EACH_MATCHING_TABLE (table
, table_id
, ofproto
) {
2518 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&table
->cls
,
2521 if (rule
->pending
) {
2522 error
= OFPROTO_POSTPONE
;
2525 if (!ofproto_rule_is_hidden(rule
)
2526 && ofproto_rule_has_out_port(rule
, out_port
)
2527 && !((rule
->flow_cookie
^ cookie
) & cookie_mask
)) {
2528 list_push_back(rules
, &rule
->ofproto_node
);
2534 cls_rule_destroy(&cr
);
2538 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2539 * forced into the range of a uint16_t. */
2541 age_secs(long long int age_ms
)
2543 return (age_ms
< 0 ? 0
2544 : age_ms
>= UINT16_MAX
* 1000 ? UINT16_MAX
2545 : (unsigned int) age_ms
/ 1000);
2549 handle_flow_stats_request(struct ofconn
*ofconn
,
2550 const struct ofp_header
*request
)
2552 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2553 struct ofputil_flow_stats_request fsr
;
2554 struct list replies
;
2559 error
= ofputil_decode_flow_stats_request(&fsr
, request
);
2564 error
= collect_rules_loose(ofproto
, fsr
.table_id
, &fsr
.match
,
2565 fsr
.cookie
, fsr
.cookie_mask
,
2566 fsr
.out_port
, &rules
);
2571 ofpmp_init(&replies
, request
);
2572 LIST_FOR_EACH (rule
, ofproto_node
, &rules
) {
2573 long long int now
= time_msec();
2574 struct ofputil_flow_stats fs
;
2576 minimatch_expand(&rule
->cr
.match
, &fs
.match
);
2577 fs
.priority
= rule
->cr
.priority
;
2578 fs
.cookie
= rule
->flow_cookie
;
2579 fs
.table_id
= rule
->table_id
;
2580 calc_flow_duration__(rule
->created
, now
, &fs
.duration_sec
,
2582 fs
.idle_timeout
= rule
->idle_timeout
;
2583 fs
.hard_timeout
= rule
->hard_timeout
;
2584 fs
.idle_age
= age_secs(now
- rule
->used
);
2585 fs
.hard_age
= age_secs(now
- rule
->modified
);
2586 ofproto
->ofproto_class
->rule_get_stats(rule
, &fs
.packet_count
,
2588 fs
.ofpacts
= rule
->ofpacts
;
2589 fs
.ofpacts_len
= rule
->ofpacts_len
;
2590 ofputil_append_flow_stats_reply(&fs
, &replies
);
2592 ofconn_send_replies(ofconn
, &replies
);
2598 flow_stats_ds(struct rule
*rule
, struct ds
*results
)
2600 uint64_t packet_count
, byte_count
;
2602 rule
->ofproto
->ofproto_class
->rule_get_stats(rule
,
2603 &packet_count
, &byte_count
);
2605 if (rule
->table_id
!= 0) {
2606 ds_put_format(results
, "table_id=%"PRIu8
", ", rule
->table_id
);
2608 ds_put_format(results
, "duration=%llds, ",
2609 (time_msec() - rule
->created
) / 1000);
2610 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
2611 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
2612 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
2613 cls_rule_format(&rule
->cr
, results
);
2614 ds_put_char(results
, ',');
2615 if (rule
->ofpacts_len
> 0) {
2616 ofpacts_format(rule
->ofpacts
, rule
->ofpacts_len
, results
);
2618 ds_put_cstr(results
, "drop");
2620 ds_put_cstr(results
, "\n");
2623 /* Adds a pretty-printed description of all flows to 'results', including
2624 * hidden flows (e.g., set up by in-band control). */
2626 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
2628 struct oftable
*table
;
2630 OFPROTO_FOR_EACH_TABLE (table
, p
) {
2631 struct cls_cursor cursor
;
2634 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
2635 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
2636 flow_stats_ds(rule
, results
);
2641 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2642 * '*engine_type' and '*engine_id', respectively. */
2644 ofproto_get_netflow_ids(const struct ofproto
*ofproto
,
2645 uint8_t *engine_type
, uint8_t *engine_id
)
2647 ofproto
->ofproto_class
->get_netflow_ids(ofproto
, engine_type
, engine_id
);
2650 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns a
2651 * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
2652 * indicating a connectivity problem). Returns zero if CFM is not faulted,
2653 * and -1 if CFM is not enabled on 'ofp_port'. */
2655 ofproto_port_get_cfm_fault(const struct ofproto
*ofproto
, uint16_t ofp_port
)
2657 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
2658 return (ofport
&& ofproto
->ofproto_class
->get_cfm_fault
2659 ? ofproto
->ofproto_class
->get_cfm_fault(ofport
)
2663 /* Checks the operational status reported by the remote CFM endpoint of
2664 * 'ofp_port' Returns 1 if operationally up, 0 if operationally down, and -1
2665 * if CFM is not enabled on 'ofp_port' or does not support operational status.
2668 ofproto_port_get_cfm_opup(const struct ofproto
*ofproto
, uint16_t ofp_port
)
2670 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
2671 return (ofport
&& ofproto
->ofproto_class
->get_cfm_opup
2672 ? ofproto
->ofproto_class
->get_cfm_opup(ofport
)
2676 /* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
2677 * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by
2678 * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a
2679 * number less than 0 if CFM is not enabled on 'ofp_port'. */
2681 ofproto_port_get_cfm_remote_mpids(const struct ofproto
*ofproto
,
2682 uint16_t ofp_port
, const uint64_t **rmps
,
2685 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
2689 return (ofport
&& ofproto
->ofproto_class
->get_cfm_remote_mpids
2690 ? ofproto
->ofproto_class
->get_cfm_remote_mpids(ofport
, rmps
,
2695 /* Checks the health of the CFM for 'ofp_port' within 'ofproto'. Returns an
2696 * integer value between 0 and 100 to indicate the health of the port as a
2697 * percentage which is the average of cfm health of all the remote_mpids or
2698 * returns -1 if CFM is not enabled on 'ofport'. */
2700 ofproto_port_get_cfm_health(const struct ofproto
*ofproto
, uint16_t ofp_port
)
2702 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
2703 return (ofport
&& ofproto
->ofproto_class
->get_cfm_health
2704 ? ofproto
->ofproto_class
->get_cfm_health(ofport
)
2709 handle_aggregate_stats_request(struct ofconn
*ofconn
,
2710 const struct ofp_header
*oh
)
2712 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2713 struct ofputil_flow_stats_request request
;
2714 struct ofputil_aggregate_stats stats
;
2715 bool unknown_packets
, unknown_bytes
;
2716 struct ofpbuf
*reply
;
2721 error
= ofputil_decode_flow_stats_request(&request
, oh
);
2726 error
= collect_rules_loose(ofproto
, request
.table_id
, &request
.match
,
2727 request
.cookie
, request
.cookie_mask
,
2728 request
.out_port
, &rules
);
2733 memset(&stats
, 0, sizeof stats
);
2734 unknown_packets
= unknown_bytes
= false;
2735 LIST_FOR_EACH (rule
, ofproto_node
, &rules
) {
2736 uint64_t packet_count
;
2737 uint64_t byte_count
;
2739 ofproto
->ofproto_class
->rule_get_stats(rule
, &packet_count
,
2742 if (packet_count
== UINT64_MAX
) {
2743 unknown_packets
= true;
2745 stats
.packet_count
+= packet_count
;
2748 if (byte_count
== UINT64_MAX
) {
2749 unknown_bytes
= true;
2751 stats
.byte_count
+= byte_count
;
2756 if (unknown_packets
) {
2757 stats
.packet_count
= UINT64_MAX
;
2759 if (unknown_bytes
) {
2760 stats
.byte_count
= UINT64_MAX
;
2763 reply
= ofputil_encode_aggregate_stats_reply(&stats
, oh
);
2764 ofconn_send_reply(ofconn
, reply
);
2769 struct queue_stats_cbdata
{
2770 struct ofport
*ofport
;
2771 struct list replies
;
2775 put_queue_stats(struct queue_stats_cbdata
*cbdata
, uint32_t queue_id
,
2776 const struct netdev_queue_stats
*stats
)
2779 struct ofputil_queue_stats oqs
= {
2780 .port_no
= cbdata
->ofport
->pp
.port_no
,
2781 .queue_id
= queue_id
,
2784 ofputil_append_queue_stat(&cbdata
->replies
, &oqs
);
2788 handle_queue_stats_dump_cb(uint32_t queue_id
,
2789 struct netdev_queue_stats
*stats
,
2792 struct queue_stats_cbdata
*cbdata
= cbdata_
;
2794 put_queue_stats(cbdata
, queue_id
, stats
);
2798 handle_queue_stats_for_port(struct ofport
*port
, uint32_t queue_id
,
2799 struct queue_stats_cbdata
*cbdata
)
2801 cbdata
->ofport
= port
;
2802 if (queue_id
== OFPQ_ALL
) {
2803 netdev_dump_queue_stats(port
->netdev
,
2804 handle_queue_stats_dump_cb
, cbdata
);
2806 struct netdev_queue_stats stats
;
2808 if (!netdev_get_queue_stats(port
->netdev
, queue_id
, &stats
)) {
2809 put_queue_stats(cbdata
, queue_id
, &stats
);
2811 return OFPERR_OFPQOFC_BAD_QUEUE
;
2818 handle_queue_stats_request(struct ofconn
*ofconn
,
2819 const struct ofp_header
*rq
)
2821 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2822 struct queue_stats_cbdata cbdata
;
2823 struct ofport
*port
;
2825 struct ofputil_queue_stats_request oqsr
;
2827 COVERAGE_INC(ofproto_queue_req
);
2829 ofpmp_init(&cbdata
.replies
, rq
);
2831 error
= ofputil_decode_queue_stats_request(rq
, &oqsr
);
2836 if (oqsr
.port_no
== OFPP_ALL
) {
2837 error
= OFPERR_OFPQOFC_BAD_QUEUE
;
2838 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
2839 if (!handle_queue_stats_for_port(port
, oqsr
.queue_id
, &cbdata
)) {
2844 port
= ofproto_get_port(ofproto
, oqsr
.port_no
);
2846 ? handle_queue_stats_for_port(port
, oqsr
.queue_id
, &cbdata
)
2847 : OFPERR_OFPQOFC_BAD_PORT
);
2850 ofconn_send_replies(ofconn
, &cbdata
.replies
);
2852 ofpbuf_list_delete(&cbdata
.replies
);
2859 is_flow_deletion_pending(const struct ofproto
*ofproto
,
2860 const struct cls_rule
*cls_rule
,
2863 if (!hmap_is_empty(&ofproto
->deletions
)) {
2864 struct ofoperation
*op
;
2866 HMAP_FOR_EACH_WITH_HASH (op
, hmap_node
,
2867 cls_rule_hash(cls_rule
, table_id
),
2868 &ofproto
->deletions
) {
2869 if (cls_rule_equal(cls_rule
, &op
->rule
->cr
)) {
2878 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2879 * in which no matching flow already exists in the flow table.
2881 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2882 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
2883 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
2884 * initiated now but may be retried later.
2886 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
2887 * ownership remains with the caller.
2889 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2892 add_flow(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
2893 const struct ofputil_flow_mod
*fm
, const struct ofp_header
*request
)
2895 struct oftable
*table
;
2896 struct ofopgroup
*group
;
2897 struct rule
*victim
;
2902 error
= check_table_id(ofproto
, fm
->table_id
);
2908 if (fm
->table_id
== 0xff) {
2910 if (ofproto
->ofproto_class
->rule_choose_table
) {
2911 error
= ofproto
->ofproto_class
->rule_choose_table(ofproto
,
2917 assert(table_id
< ofproto
->n_tables
);
2918 table
= &ofproto
->tables
[table_id
];
2920 table
= &ofproto
->tables
[0];
2922 } else if (fm
->table_id
< ofproto
->n_tables
) {
2923 table
= &ofproto
->tables
[fm
->table_id
];
2925 return OFPERR_OFPBRC_BAD_TABLE_ID
;
2928 if (table
->flags
& OFTABLE_READONLY
) {
2929 return OFPERR_OFPBRC_EPERM
;
2932 /* Allocate new rule and initialize classifier rule. */
2933 rule
= ofproto
->ofproto_class
->rule_alloc();
2935 VLOG_WARN_RL(&rl
, "%s: failed to create rule (%s)",
2936 ofproto
->name
, strerror(error
));
2939 cls_rule_init(&rule
->cr
, &fm
->match
, fm
->priority
);
2941 /* Serialize against pending deletion. */
2942 if (is_flow_deletion_pending(ofproto
, &cr
, table
- ofproto
->tables
)) {
2943 cls_rule_destroy(&rule
->cr
);
2944 ofproto
->ofproto_class
->rule_dealloc(rule
);
2945 return OFPROTO_POSTPONE
;
2948 /* Check for overlap, if requested. */
2949 if (fm
->flags
& OFPFF_CHECK_OVERLAP
2950 && classifier_rule_overlaps(&table
->cls
, &rule
->cr
)) {
2951 cls_rule_destroy(&rule
->cr
);
2952 ofproto
->ofproto_class
->rule_dealloc(rule
);
2953 return OFPERR_OFPFMFC_OVERLAP
;
2956 rule
->ofproto
= ofproto
;
2957 rule
->pending
= NULL
;
2958 rule
->flow_cookie
= fm
->new_cookie
;
2959 rule
->created
= rule
->modified
= rule
->used
= time_msec();
2960 rule
->idle_timeout
= fm
->idle_timeout
;
2961 rule
->hard_timeout
= fm
->hard_timeout
;
2962 rule
->table_id
= table
- ofproto
->tables
;
2963 rule
->send_flow_removed
= (fm
->flags
& OFPFF_SEND_FLOW_REM
) != 0;
2964 rule
->ofpacts
= xmemdup(fm
->ofpacts
, fm
->ofpacts_len
);
2965 rule
->ofpacts_len
= fm
->ofpacts_len
;
2966 rule
->evictable
= true;
2967 rule
->eviction_group
= NULL
;
2968 rule
->monitor_flags
= 0;
2969 rule
->add_seqno
= 0;
2970 rule
->modify_seqno
= 0;
2972 /* Insert new rule. */
2973 victim
= oftable_replace_rule(rule
);
2974 if (victim
&& !rule_is_modifiable(victim
)) {
2975 error
= OFPERR_OFPBRC_EPERM
;
2976 } else if (victim
&& victim
->pending
) {
2977 error
= OFPROTO_POSTPONE
;
2979 struct ofoperation
*op
;
2982 if (classifier_count(&table
->cls
) > table
->max_flows
) {
2985 was_evictable
= rule
->evictable
;
2986 rule
->evictable
= false;
2987 evict
= choose_rule_to_evict(table
);
2988 rule
->evictable
= was_evictable
;
2991 error
= OFPERR_OFPFMFC_ALL_TABLES_FULL
;
2993 } else if (evict
->pending
) {
2994 error
= OFPROTO_POSTPONE
;
3001 group
= ofopgroup_create(ofproto
, ofconn
, request
, fm
->buffer_id
);
3002 op
= ofoperation_create(group
, rule
, OFOPERATION_ADD
, 0);
3003 op
->victim
= victim
;
3005 error
= ofproto
->ofproto_class
->rule_construct(rule
);
3007 op
->group
->n_running
--;
3008 ofoperation_destroy(rule
->pending
);
3010 delete_flow__(evict
, group
);
3012 ofopgroup_submit(group
);
3016 /* Back out if an error occurred. */
3018 oftable_substitute_rule(rule
, victim
);
3019 ofproto_rule_destroy__(rule
);
3024 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3026 /* Modifies the rules listed in 'rules', changing their actions to match those
3029 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3032 * Returns 0 on success, otherwise an OpenFlow error code. */
3034 modify_flows__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3035 const struct ofputil_flow_mod
*fm
,
3036 const struct ofp_header
*request
, struct list
*rules
)
3038 struct ofopgroup
*group
;
3042 group
= ofopgroup_create(ofproto
, ofconn
, request
, fm
->buffer_id
);
3043 error
= OFPERR_OFPBRC_EPERM
;
3044 LIST_FOR_EACH (rule
, ofproto_node
, rules
) {
3045 struct ofoperation
*op
;
3046 bool actions_changed
;
3047 ovs_be64 new_cookie
;
3049 if (rule_is_modifiable(rule
)) {
3050 /* At least one rule is modifiable, don't report EPERM error. */
3056 actions_changed
= !ofpacts_equal(fm
->ofpacts
, fm
->ofpacts_len
,
3057 rule
->ofpacts
, rule
->ofpacts_len
);
3058 new_cookie
= (fm
->new_cookie
!= htonll(UINT64_MAX
)
3060 : rule
->flow_cookie
);
3061 if (!actions_changed
&& new_cookie
== rule
->flow_cookie
) {
3062 /* No change at all. */
3066 op
= ofoperation_create(group
, rule
, OFOPERATION_MODIFY
, 0);
3067 rule
->flow_cookie
= new_cookie
;
3068 if (actions_changed
) {
3069 op
->ofpacts
= rule
->ofpacts
;
3070 op
->ofpacts_len
= rule
->ofpacts_len
;
3071 rule
->ofpacts
= xmemdup(fm
->ofpacts
, fm
->ofpacts_len
);
3072 rule
->ofpacts_len
= fm
->ofpacts_len
;
3073 rule
->ofproto
->ofproto_class
->rule_modify_actions(rule
);
3075 ofoperation_complete(op
, 0);
3078 ofopgroup_submit(group
);
3084 modify_flows_add(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3085 const struct ofputil_flow_mod
*fm
,
3086 const struct ofp_header
*request
)
3088 if (fm
->cookie_mask
!= htonll(0) || fm
->new_cookie
== htonll(UINT64_MAX
)) {
3091 return add_flow(ofproto
, ofconn
, fm
, request
);
3094 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3097 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3100 modify_flows_loose(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3101 const struct ofputil_flow_mod
*fm
,
3102 const struct ofp_header
*request
)
3107 error
= collect_rules_loose(ofproto
, fm
->table_id
, &fm
->match
,
3108 fm
->cookie
, fm
->cookie_mask
,
3112 } else if (list_is_empty(&rules
)) {
3113 return modify_flows_add(ofproto
, ofconn
, fm
, request
);
3115 return modify_flows__(ofproto
, ofconn
, fm
, request
, &rules
);
3119 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3122 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3125 modify_flow_strict(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3126 const struct ofputil_flow_mod
*fm
,
3127 const struct ofp_header
*request
)
3132 error
= collect_rules_strict(ofproto
, fm
->table_id
, &fm
->match
,
3133 fm
->priority
, fm
->cookie
, fm
->cookie_mask
,
3138 } else if (list_is_empty(&rules
)) {
3139 return modify_flows_add(ofproto
, ofconn
, fm
, request
);
3141 return list_is_singleton(&rules
) ? modify_flows__(ofproto
, ofconn
,
3142 fm
, request
, &rules
)
3147 /* OFPFC_DELETE implementation. */
3150 delete_flow__(struct rule
*rule
, struct ofopgroup
*group
)
3152 struct ofproto
*ofproto
= rule
->ofproto
;
3154 ofproto_rule_send_removed(rule
, OFPRR_DELETE
);
3156 ofoperation_create(group
, rule
, OFOPERATION_DELETE
, OFPRR_DELETE
);
3157 oftable_remove_rule(rule
);
3158 ofproto
->ofproto_class
->rule_destruct(rule
);
3161 /* Deletes the rules listed in 'rules'.
3163 * Returns 0 on success, otherwise an OpenFlow error code. */
3165 delete_flows__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3166 const struct ofp_header
*request
, struct list
*rules
)
3168 struct rule
*rule
, *next
;
3169 struct ofopgroup
*group
;
3171 group
= ofopgroup_create(ofproto
, ofconn
, request
, UINT32_MAX
);
3172 LIST_FOR_EACH_SAFE (rule
, next
, ofproto_node
, rules
) {
3173 delete_flow__(rule
, group
);
3175 ofopgroup_submit(group
);
3180 /* Implements OFPFC_DELETE. */
3182 delete_flows_loose(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3183 const struct ofputil_flow_mod
*fm
,
3184 const struct ofp_header
*request
)
3189 error
= collect_rules_loose(ofproto
, fm
->table_id
, &fm
->match
,
3190 fm
->cookie
, fm
->cookie_mask
,
3191 fm
->out_port
, &rules
);
3192 return (error
? error
3193 : !list_is_empty(&rules
) ? delete_flows__(ofproto
, ofconn
, request
,
3198 /* Implements OFPFC_DELETE_STRICT. */
3200 delete_flow_strict(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3201 const struct ofputil_flow_mod
*fm
,
3202 const struct ofp_header
*request
)
3207 error
= collect_rules_strict(ofproto
, fm
->table_id
, &fm
->match
,
3208 fm
->priority
, fm
->cookie
, fm
->cookie_mask
,
3209 fm
->out_port
, &rules
);
3210 return (error
? error
3211 : list_is_singleton(&rules
) ? delete_flows__(ofproto
, ofconn
,
3217 ofproto_rule_send_removed(struct rule
*rule
, uint8_t reason
)
3219 struct ofputil_flow_removed fr
;
3221 if (ofproto_rule_is_hidden(rule
) || !rule
->send_flow_removed
) {
3225 minimatch_expand(&rule
->cr
.match
, &fr
.match
);
3226 fr
.priority
= rule
->cr
.priority
;
3227 fr
.cookie
= rule
->flow_cookie
;
3229 fr
.table_id
= rule
->table_id
;
3230 calc_flow_duration__(rule
->created
, time_msec(),
3231 &fr
.duration_sec
, &fr
.duration_nsec
);
3232 fr
.idle_timeout
= rule
->idle_timeout
;
3233 fr
.hard_timeout
= rule
->hard_timeout
;
3234 rule
->ofproto
->ofproto_class
->rule_get_stats(rule
, &fr
.packet_count
,
3237 connmgr_send_flow_removed(rule
->ofproto
->connmgr
, &fr
);
3241 ofproto_rule_update_used(struct rule
*rule
, long long int used
)
3243 if (used
> rule
->used
) {
3244 struct eviction_group
*evg
= rule
->eviction_group
;
3248 heap_change(&evg
->rules
, &rule
->evg_node
,
3249 rule_eviction_priority(rule
));
3254 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3255 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3258 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3261 * ofproto implementation ->run() functions should use this function to expire
3262 * OpenFlow flows. */
3264 ofproto_rule_expire(struct rule
*rule
, uint8_t reason
)
3266 struct ofproto
*ofproto
= rule
->ofproto
;
3267 struct ofopgroup
*group
;
3269 assert(reason
== OFPRR_HARD_TIMEOUT
|| reason
== OFPRR_IDLE_TIMEOUT
);
3271 ofproto_rule_send_removed(rule
, reason
);
3273 group
= ofopgroup_create_unattached(ofproto
);
3274 ofoperation_create(group
, rule
, OFOPERATION_DELETE
, reason
);
3275 oftable_remove_rule(rule
);
3276 ofproto
->ofproto_class
->rule_destruct(rule
);
3277 ofopgroup_submit(group
);
3281 handle_flow_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3283 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
3284 struct ofputil_flow_mod fm
;
3285 uint64_t ofpacts_stub
[1024 / 8];
3286 struct ofpbuf ofpacts
;
3290 error
= reject_slave_controller(ofconn
);
3295 ofpbuf_use_stub(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
3296 error
= ofputil_decode_flow_mod(&fm
, oh
, ofconn_get_protocol(ofconn
),
3299 goto exit_free_ofpacts
;
3302 if (fm
.flags
& OFPFF10_EMERG
) {
3303 /* We do not support the OpenFlow 1.0 emergency flow cache, which
3304 * is not required in OpenFlow 1.0.1 and removed from OpenFlow 1.1.
3305 * There is no good error code, so just state that the flow table
3307 error
= OFPERR_OFPFMFC_ALL_TABLES_FULL
;
3310 error
= ofpacts_check(fm
.ofpacts
, fm
.ofpacts_len
,
3311 &fm
.match
.flow
, ofproto
->max_ports
);
3314 error
= handle_flow_mod__(ofconn_get_ofproto(ofconn
), ofconn
, &fm
, oh
);
3317 goto exit_free_ofpacts
;
3320 /* Record the operation for logging a summary report. */
3321 switch (fm
.command
) {
3327 case OFPFC_MODIFY_STRICT
:
3328 ofproto
->n_modify
++;
3332 case OFPFC_DELETE_STRICT
:
3333 ofproto
->n_delete
++;
3338 if (ofproto
->next_op_report
== LLONG_MAX
) {
3339 ofproto
->first_op
= now
;
3340 ofproto
->next_op_report
= MAX(now
+ 10 * 1000,
3341 ofproto
->op_backoff
);
3342 ofproto
->op_backoff
= ofproto
->next_op_report
+ 60 * 1000;
3344 ofproto
->last_op
= now
;
3347 ofpbuf_uninit(&ofpacts
);
3353 handle_flow_mod__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3354 const struct ofputil_flow_mod
*fm
,
3355 const struct ofp_header
*oh
)
3357 if (ofproto
->n_pending
>= 50) {
3358 assert(!list_is_empty(&ofproto
->pending
));
3359 return OFPROTO_POSTPONE
;
3362 switch (fm
->command
) {
3364 return add_flow(ofproto
, ofconn
, fm
, oh
);
3367 return modify_flows_loose(ofproto
, ofconn
, fm
, oh
);
3369 case OFPFC_MODIFY_STRICT
:
3370 return modify_flow_strict(ofproto
, ofconn
, fm
, oh
);
3373 return delete_flows_loose(ofproto
, ofconn
, fm
, oh
);
3375 case OFPFC_DELETE_STRICT
:
3376 return delete_flow_strict(ofproto
, ofconn
, fm
, oh
);
3379 if (fm
->command
> 0xff) {
3380 VLOG_WARN_RL(&rl
, "%s: flow_mod has explicit table_id but "
3381 "flow_mod_table_id extension is not enabled",
3384 return OFPERR_OFPFMFC_BAD_COMMAND
;
3389 handle_role_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3391 const struct nx_role_request
*nrr
= ofpmsg_body(oh
);
3392 struct nx_role_request
*reply
;
3396 role
= ntohl(nrr
->role
);
3397 if (role
!= NX_ROLE_OTHER
&& role
!= NX_ROLE_MASTER
3398 && role
!= NX_ROLE_SLAVE
) {
3399 return OFPERR_OFPRRFC_BAD_ROLE
;
3402 if (ofconn_get_role(ofconn
) != role
3403 && ofconn_has_pending_opgroups(ofconn
)) {
3404 return OFPROTO_POSTPONE
;
3407 ofconn_set_role(ofconn
, role
);
3409 buf
= ofpraw_alloc_reply(OFPRAW_NXT_ROLE_REPLY
, oh
, 0);
3410 reply
= ofpbuf_put_zeros(buf
, sizeof *reply
);
3411 reply
->role
= htonl(role
);
3412 ofconn_send_reply(ofconn
, buf
);
3418 handle_nxt_flow_mod_table_id(struct ofconn
*ofconn
,
3419 const struct ofp_header
*oh
)
3421 const struct nx_flow_mod_table_id
*msg
= ofpmsg_body(oh
);
3422 enum ofputil_protocol cur
, next
;
3424 cur
= ofconn_get_protocol(ofconn
);
3425 next
= ofputil_protocol_set_tid(cur
, msg
->set
!= 0);
3426 ofconn_set_protocol(ofconn
, next
);
3432 handle_nxt_set_flow_format(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3434 const struct nx_set_flow_format
*msg
= ofpmsg_body(oh
);
3435 enum ofputil_protocol cur
, next
;
3436 enum ofputil_protocol next_base
;
3438 next_base
= ofputil_nx_flow_format_to_protocol(ntohl(msg
->format
));
3440 return OFPERR_OFPBRC_EPERM
;
3443 cur
= ofconn_get_protocol(ofconn
);
3444 next
= ofputil_protocol_set_base(cur
, next_base
);
3445 if (cur
!= next
&& ofconn_has_pending_opgroups(ofconn
)) {
3446 /* Avoid sending async messages in surprising protocol. */
3447 return OFPROTO_POSTPONE
;
3450 ofconn_set_protocol(ofconn
, next
);
3455 handle_nxt_set_packet_in_format(struct ofconn
*ofconn
,
3456 const struct ofp_header
*oh
)
3458 const struct nx_set_packet_in_format
*msg
= ofpmsg_body(oh
);
3461 format
= ntohl(msg
->format
);
3462 if (format
!= NXPIF_OPENFLOW10
&& format
!= NXPIF_NXM
) {
3463 return OFPERR_OFPBRC_EPERM
;
3466 if (format
!= ofconn_get_packet_in_format(ofconn
)
3467 && ofconn_has_pending_opgroups(ofconn
)) {
3468 /* Avoid sending async message in surprsing packet in format. */
3469 return OFPROTO_POSTPONE
;
3472 ofconn_set_packet_in_format(ofconn
, format
);
3477 handle_nxt_set_async_config(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3479 const struct nx_async_config
*msg
= ofpmsg_body(oh
);
3480 uint32_t master
[OAM_N_TYPES
];
3481 uint32_t slave
[OAM_N_TYPES
];
3483 master
[OAM_PACKET_IN
] = ntohl(msg
->packet_in_mask
[0]);
3484 master
[OAM_PORT_STATUS
] = ntohl(msg
->port_status_mask
[0]);
3485 master
[OAM_FLOW_REMOVED
] = ntohl(msg
->flow_removed_mask
[0]);
3487 slave
[OAM_PACKET_IN
] = ntohl(msg
->packet_in_mask
[1]);
3488 slave
[OAM_PORT_STATUS
] = ntohl(msg
->port_status_mask
[1]);
3489 slave
[OAM_FLOW_REMOVED
] = ntohl(msg
->flow_removed_mask
[1]);
3491 ofconn_set_async_config(ofconn
, master
, slave
);
3492 if (ofconn_get_type(ofconn
) == OFCONN_SERVICE
&&
3493 !ofconn_get_miss_send_len(ofconn
)) {
3494 ofconn_set_miss_send_len(ofconn
, OFP_DEFAULT_MISS_SEND_LEN
);
3501 handle_nxt_set_controller_id(struct ofconn
*ofconn
,
3502 const struct ofp_header
*oh
)
3504 const struct nx_controller_id
*nci
= ofpmsg_body(oh
);
3506 if (!is_all_zeros(nci
->zero
, sizeof nci
->zero
)) {
3507 return OFPERR_NXBRC_MUST_BE_ZERO
;
3510 ofconn_set_controller_id(ofconn
, ntohs(nci
->controller_id
));
3515 handle_barrier_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3519 if (ofconn_has_pending_opgroups(ofconn
)) {
3520 return OFPROTO_POSTPONE
;
3523 buf
= ofpraw_alloc_reply((oh
->version
== OFP10_VERSION
3524 ? OFPRAW_OFPT10_BARRIER_REPLY
3525 : OFPRAW_OFPT11_BARRIER_REPLY
), oh
, 0);
3526 ofconn_send_reply(ofconn
, buf
);
3531 ofproto_compose_flow_refresh_update(const struct rule
*rule
,
3532 enum nx_flow_monitor_flags flags
,
3535 struct ofoperation
*op
= rule
->pending
;
3536 struct ofputil_flow_update fu
;
3539 if (op
&& op
->type
== OFOPERATION_ADD
&& !op
->victim
) {
3540 /* We'll report the final flow when the operation completes. Reporting
3541 * it now would cause a duplicate report later. */
3545 fu
.event
= (flags
& (NXFMF_INITIAL
| NXFMF_ADD
)
3546 ? NXFME_ADDED
: NXFME_MODIFIED
);
3548 fu
.idle_timeout
= rule
->idle_timeout
;
3549 fu
.hard_timeout
= rule
->hard_timeout
;
3550 fu
.table_id
= rule
->table_id
;
3551 fu
.cookie
= rule
->flow_cookie
;
3552 minimatch_expand(&rule
->cr
.match
, &match
);
3554 fu
.priority
= rule
->cr
.priority
;
3555 if (!(flags
& NXFMF_ACTIONS
)) {
3559 fu
.ofpacts
= rule
->ofpacts
;
3560 fu
.ofpacts_len
= rule
->ofpacts_len
;
3562 /* An operation is in progress. Use the previous version of the flow's
3563 * actions, so that when the operation commits we report the change. */
3565 case OFOPERATION_ADD
:
3566 /* We already verified that there was a victim. */
3567 fu
.ofpacts
= op
->victim
->ofpacts
;
3568 fu
.ofpacts_len
= op
->victim
->ofpacts_len
;
3571 case OFOPERATION_MODIFY
:
3573 fu
.ofpacts
= op
->ofpacts
;
3574 fu
.ofpacts_len
= op
->ofpacts_len
;
3576 fu
.ofpacts
= rule
->ofpacts
;
3577 fu
.ofpacts_len
= rule
->ofpacts_len
;
3581 case OFOPERATION_DELETE
:
3582 fu
.ofpacts
= rule
->ofpacts
;
3583 fu
.ofpacts_len
= rule
->ofpacts_len
;
3591 if (list_is_empty(msgs
)) {
3592 ofputil_start_flow_update(msgs
);
3594 ofputil_append_flow_update(&fu
, msgs
);
3598 ofmonitor_compose_refresh_updates(struct list
*rules
, struct list
*msgs
)
3602 LIST_FOR_EACH (rule
, ofproto_node
, rules
) {
3603 enum nx_flow_monitor_flags flags
= rule
->monitor_flags
;
3604 rule
->monitor_flags
= 0;
3606 ofproto_compose_flow_refresh_update(rule
, flags
, msgs
);
3611 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor
*m
,
3612 struct rule
*rule
, uint64_t seqno
,
3615 enum nx_flow_monitor_flags update
;
3617 if (ofproto_rule_is_hidden(rule
)) {
3622 ? ofoperation_has_out_port(rule
->pending
, m
->out_port
)
3623 : ofproto_rule_has_out_port(rule
, m
->out_port
))) {
3628 if (rule
->add_seqno
> seqno
) {
3629 update
= NXFMF_ADD
| NXFMF_MODIFY
;
3630 } else if (rule
->modify_seqno
> seqno
) {
3631 update
= NXFMF_MODIFY
;
3636 if (!(m
->flags
& update
)) {
3640 update
= NXFMF_INITIAL
;
3643 if (!rule
->monitor_flags
) {
3644 list_push_back(rules
, &rule
->ofproto_node
);
3646 rule
->monitor_flags
|= update
| (m
->flags
& NXFMF_ACTIONS
);
3650 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor
*m
,
3654 const struct ofproto
*ofproto
= ofconn_get_ofproto(m
->ofconn
);
3655 const struct ofoperation
*op
;
3656 const struct oftable
*table
;
3657 struct cls_rule target
;
3659 cls_rule_init_from_minimatch(&target
, &m
->match
, 0);
3660 FOR_EACH_MATCHING_TABLE (table
, m
->table_id
, ofproto
) {
3661 struct cls_cursor cursor
;
3664 cls_cursor_init(&cursor
, &table
->cls
, &target
);
3665 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3666 assert(!rule
->pending
); /* XXX */
3667 ofproto_collect_ofmonitor_refresh_rule(m
, rule
, seqno
, rules
);
3671 HMAP_FOR_EACH (op
, hmap_node
, &ofproto
->deletions
) {
3672 struct rule
*rule
= op
->rule
;
3674 if (((m
->table_id
== 0xff
3675 ? !(ofproto
->tables
[rule
->table_id
].flags
& OFTABLE_HIDDEN
)
3676 : m
->table_id
== rule
->table_id
))
3677 && cls_rule_is_loose_match(&rule
->cr
, &target
.match
)) {
3678 ofproto_collect_ofmonitor_refresh_rule(m
, rule
, seqno
, rules
);
3681 cls_rule_destroy(&target
);
3685 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor
*m
,
3688 if (m
->flags
& NXFMF_INITIAL
) {
3689 ofproto_collect_ofmonitor_refresh_rules(m
, 0, rules
);
3694 ofmonitor_collect_resume_rules(struct ofmonitor
*m
,
3695 uint64_t seqno
, struct list
*rules
)
3697 ofproto_collect_ofmonitor_refresh_rules(m
, seqno
, rules
);
3701 handle_flow_monitor_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3703 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
3704 struct ofmonitor
**monitors
;
3705 size_t n_monitors
, allocated_monitors
;
3706 struct list replies
;
3713 ofpbuf_use_const(&b
, oh
, ntohs(oh
->length
));
3715 n_monitors
= allocated_monitors
= 0;
3717 struct ofputil_flow_monitor_request request
;
3718 struct ofmonitor
*m
;
3721 retval
= ofputil_decode_flow_monitor_request(&request
, &b
);
3722 if (retval
== EOF
) {
3724 } else if (retval
) {
3729 if (request
.table_id
!= 0xff
3730 && request
.table_id
>= ofproto
->n_tables
) {
3731 error
= OFPERR_OFPBRC_BAD_TABLE_ID
;
3735 error
= ofmonitor_create(&request
, ofconn
, &m
);
3740 if (n_monitors
>= allocated_monitors
) {
3741 monitors
= x2nrealloc(monitors
, &allocated_monitors
,
3744 monitors
[n_monitors
++] = m
;
3748 for (i
= 0; i
< n_monitors
; i
++) {
3749 ofproto_collect_ofmonitor_initial_rules(monitors
[i
], &rules
);
3752 ofpmp_init(&replies
, oh
);
3753 ofmonitor_compose_refresh_updates(&rules
, &replies
);
3754 ofconn_send_replies(ofconn
, &replies
);
3761 for (i
= 0; i
< n_monitors
; i
++) {
3762 ofmonitor_destroy(monitors
[i
]);
3769 handle_flow_monitor_cancel(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3771 struct ofmonitor
*m
;
3774 id
= ofputil_decode_flow_monitor_cancel(oh
);
3775 m
= ofmonitor_lookup(ofconn
, id
);
3777 return OFPERR_NXBRC_FM_BAD_ID
;
3780 ofmonitor_destroy(m
);
3785 handle_openflow__(struct ofconn
*ofconn
, const struct ofpbuf
*msg
)
3787 const struct ofp_header
*oh
= msg
->data
;
3791 error
= ofptype_decode(&type
, oh
);
3797 /* OpenFlow requests. */
3798 case OFPTYPE_ECHO_REQUEST
:
3799 return handle_echo_request(ofconn
, oh
);
3801 case OFPTYPE_FEATURES_REQUEST
:
3802 return handle_features_request(ofconn
, oh
);
3804 case OFPTYPE_GET_CONFIG_REQUEST
:
3805 return handle_get_config_request(ofconn
, oh
);
3807 case OFPTYPE_SET_CONFIG
:
3808 return handle_set_config(ofconn
, oh
);
3810 case OFPTYPE_PACKET_OUT
:
3811 return handle_packet_out(ofconn
, oh
);
3813 case OFPTYPE_PORT_MOD
:
3814 return handle_port_mod(ofconn
, oh
);
3816 case OFPTYPE_FLOW_MOD
:
3817 return handle_flow_mod(ofconn
, oh
);
3819 case OFPTYPE_BARRIER_REQUEST
:
3820 return handle_barrier_request(ofconn
, oh
);
3822 /* OpenFlow replies. */
3823 case OFPTYPE_ECHO_REPLY
:
3826 /* Nicira extension requests. */
3827 case OFPTYPE_ROLE_REQUEST
:
3828 return handle_role_request(ofconn
, oh
);
3830 case OFPTYPE_FLOW_MOD_TABLE_ID
:
3831 return handle_nxt_flow_mod_table_id(ofconn
, oh
);
3833 case OFPTYPE_SET_FLOW_FORMAT
:
3834 return handle_nxt_set_flow_format(ofconn
, oh
);
3836 case OFPTYPE_SET_PACKET_IN_FORMAT
:
3837 return handle_nxt_set_packet_in_format(ofconn
, oh
);
3839 case OFPTYPE_SET_CONTROLLER_ID
:
3840 return handle_nxt_set_controller_id(ofconn
, oh
);
3842 case OFPTYPE_FLOW_AGE
:
3843 /* Nothing to do. */
3846 case OFPTYPE_FLOW_MONITOR_CANCEL
:
3847 return handle_flow_monitor_cancel(ofconn
, oh
);
3849 case OFPTYPE_SET_ASYNC_CONFIG
:
3850 return handle_nxt_set_async_config(ofconn
, oh
);
3852 /* Statistics requests. */
3853 case OFPTYPE_DESC_STATS_REQUEST
:
3854 return handle_desc_stats_request(ofconn
, oh
);
3856 case OFPTYPE_FLOW_STATS_REQUEST
:
3857 return handle_flow_stats_request(ofconn
, oh
);
3859 case OFPTYPE_AGGREGATE_STATS_REQUEST
:
3860 return handle_aggregate_stats_request(ofconn
, oh
);
3862 case OFPTYPE_TABLE_STATS_REQUEST
:
3863 return handle_table_stats_request(ofconn
, oh
);
3865 case OFPTYPE_PORT_STATS_REQUEST
:
3866 return handle_port_stats_request(ofconn
, oh
);
3868 case OFPTYPE_QUEUE_STATS_REQUEST
:
3869 return handle_queue_stats_request(ofconn
, oh
);
3871 case OFPTYPE_PORT_DESC_STATS_REQUEST
:
3872 return handle_port_desc_stats_request(ofconn
, oh
);
3874 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST
:
3875 return handle_flow_monitor_request(ofconn
, oh
);
3879 case OFPTYPE_FEATURES_REPLY
:
3880 case OFPTYPE_GET_CONFIG_REPLY
:
3881 case OFPTYPE_PACKET_IN
:
3882 case OFPTYPE_FLOW_REMOVED
:
3883 case OFPTYPE_PORT_STATUS
:
3884 case OFPTYPE_BARRIER_REPLY
:
3885 case OFPTYPE_DESC_STATS_REPLY
:
3886 case OFPTYPE_FLOW_STATS_REPLY
:
3887 case OFPTYPE_QUEUE_STATS_REPLY
:
3888 case OFPTYPE_PORT_STATS_REPLY
:
3889 case OFPTYPE_TABLE_STATS_REPLY
:
3890 case OFPTYPE_AGGREGATE_STATS_REPLY
:
3891 case OFPTYPE_PORT_DESC_STATS_REPLY
:
3892 case OFPTYPE_ROLE_REPLY
:
3893 case OFPTYPE_FLOW_MONITOR_PAUSED
:
3894 case OFPTYPE_FLOW_MONITOR_RESUMED
:
3895 case OFPTYPE_FLOW_MONITOR_STATS_REPLY
:
3897 return OFPERR_OFPBRC_BAD_TYPE
;
3902 handle_openflow(struct ofconn
*ofconn
, struct ofpbuf
*ofp_msg
)
3904 int error
= handle_openflow__(ofconn
, ofp_msg
);
3905 if (error
&& error
!= OFPROTO_POSTPONE
) {
3906 ofconn_send_error(ofconn
, ofp_msg
->data
, error
);
3908 COVERAGE_INC(ofproto_recv_openflow
);
3909 return error
!= OFPROTO_POSTPONE
;
3912 /* Asynchronous operations. */
3914 /* Creates and returns a new ofopgroup that is not associated with any
3915 * OpenFlow connection.
3917 * The caller should add operations to the returned group with
3918 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3919 static struct ofopgroup
*
3920 ofopgroup_create_unattached(struct ofproto
*ofproto
)
3922 struct ofopgroup
*group
= xzalloc(sizeof *group
);
3923 group
->ofproto
= ofproto
;
3924 list_init(&group
->ofproto_node
);
3925 list_init(&group
->ops
);
3926 list_init(&group
->ofconn_node
);
3930 /* Creates and returns a new ofopgroup for 'ofproto'.
3932 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
3933 * connection. The 'request' and 'buffer_id' arguments are ignored.
3935 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
3936 * If the ofopgroup eventually fails, then the error reply will include
3937 * 'request'. If the ofopgroup eventually succeeds, then the packet with
3938 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
3940 * The caller should add operations to the returned group with
3941 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3942 static struct ofopgroup
*
3943 ofopgroup_create(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3944 const struct ofp_header
*request
, uint32_t buffer_id
)
3946 struct ofopgroup
*group
= ofopgroup_create_unattached(ofproto
);
3948 size_t request_len
= ntohs(request
->length
);
3950 assert(ofconn_get_ofproto(ofconn
) == ofproto
);
3952 ofconn_add_opgroup(ofconn
, &group
->ofconn_node
);
3953 group
->ofconn
= ofconn
;
3954 group
->request
= xmemdup(request
, MIN(request_len
, 64));
3955 group
->buffer_id
= buffer_id
;
3960 /* Submits 'group' for processing.
3962 * If 'group' contains no operations (e.g. none were ever added, or all of the
3963 * ones that were added completed synchronously), then it is destroyed
3964 * immediately. Otherwise it is added to the ofproto's list of pending
3967 ofopgroup_submit(struct ofopgroup
*group
)
3969 if (!group
->n_running
) {
3970 ofopgroup_complete(group
);
3972 list_push_back(&group
->ofproto
->pending
, &group
->ofproto_node
);
3973 group
->ofproto
->n_pending
++;
3978 ofopgroup_complete(struct ofopgroup
*group
)
3980 struct ofproto
*ofproto
= group
->ofproto
;
3982 struct ofconn
*abbrev_ofconn
;
3983 ovs_be32 abbrev_xid
;
3985 struct ofoperation
*op
, *next_op
;
3988 assert(!group
->n_running
);
3991 LIST_FOR_EACH (op
, group_node
, &group
->ops
) {
3998 if (!error
&& group
->ofconn
&& group
->buffer_id
!= UINT32_MAX
) {
3999 LIST_FOR_EACH (op
, group_node
, &group
->ops
) {
4000 if (op
->type
!= OFOPERATION_DELETE
) {
4001 struct ofpbuf
*packet
;
4004 error
= ofconn_pktbuf_retrieve(group
->ofconn
, group
->buffer_id
,
4008 error
= rule_execute(op
->rule
, in_port
, packet
);
4015 if (!error
&& !list_is_empty(&group
->ofconn_node
)) {
4016 abbrev_ofconn
= group
->ofconn
;
4017 abbrev_xid
= group
->request
->xid
;
4019 abbrev_ofconn
= NULL
;
4020 abbrev_xid
= htonl(0);
4022 LIST_FOR_EACH_SAFE (op
, next_op
, group_node
, &group
->ops
) {
4023 struct rule
*rule
= op
->rule
;
4025 if (!op
->error
&& !ofproto_rule_is_hidden(rule
)) {
4026 /* Check that we can just cast from ofoperation_type to
4027 * nx_flow_update_event. */
4028 BUILD_ASSERT_DECL((enum nx_flow_update_event
) OFOPERATION_ADD
4030 BUILD_ASSERT_DECL((enum nx_flow_update_event
) OFOPERATION_DELETE
4032 BUILD_ASSERT_DECL((enum nx_flow_update_event
) OFOPERATION_MODIFY
4035 ofmonitor_report(ofproto
->connmgr
, rule
,
4036 (enum nx_flow_update_event
) op
->type
,
4037 op
->reason
, abbrev_ofconn
, abbrev_xid
);
4040 rule
->pending
= NULL
;
4043 case OFOPERATION_ADD
:
4047 ofproto_rule_destroy__(op
->victim
);
4048 vid_mask
= minimask_get_vid_mask(&rule
->cr
.match
.mask
);
4049 if (vid_mask
== VLAN_VID_MASK
) {
4050 if (ofproto
->vlan_bitmap
) {
4051 uint16_t vid
= miniflow_get_vid(&rule
->cr
.match
.flow
);
4052 if (!bitmap_is_set(ofproto
->vlan_bitmap
, vid
)) {
4053 bitmap_set1(ofproto
->vlan_bitmap
, vid
);
4054 ofproto
->vlans_changed
= true;
4057 ofproto
->vlans_changed
= true;
4061 oftable_substitute_rule(rule
, op
->victim
);
4062 ofproto_rule_destroy__(rule
);
4066 case OFOPERATION_DELETE
:
4068 ofproto_rule_destroy__(rule
);
4072 case OFOPERATION_MODIFY
:
4074 rule
->modified
= time_msec();
4076 rule
->flow_cookie
= op
->flow_cookie
;
4078 free(rule
->ofpacts
);
4079 rule
->ofpacts
= op
->ofpacts
;
4080 rule
->ofpacts_len
= op
->ofpacts_len
;
4082 op
->ofpacts_len
= 0;
4091 ofoperation_destroy(op
);
4094 ofmonitor_flush(ofproto
->connmgr
);
4096 if (!list_is_empty(&group
->ofproto_node
)) {
4097 assert(ofproto
->n_pending
> 0);
4098 ofproto
->n_pending
--;
4099 list_remove(&group
->ofproto_node
);
4101 if (!list_is_empty(&group
->ofconn_node
)) {
4102 list_remove(&group
->ofconn_node
);
4104 ofconn_send_error(group
->ofconn
, group
->request
, error
);
4106 connmgr_retry(ofproto
->connmgr
);
4108 free(group
->request
);
4112 /* Initiates a new operation on 'rule', of the specified 'type', within
4113 * 'group'. Prior to calling, 'rule' must not have any pending operation.
4115 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
4116 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
4118 * Returns the newly created ofoperation (which is also available as
4119 * rule->pending). */
4120 static struct ofoperation
*
4121 ofoperation_create(struct ofopgroup
*group
, struct rule
*rule
,
4122 enum ofoperation_type type
,
4123 enum ofp_flow_removed_reason reason
)
4125 struct ofproto
*ofproto
= group
->ofproto
;
4126 struct ofoperation
*op
;
4128 assert(!rule
->pending
);
4130 op
= rule
->pending
= xzalloc(sizeof *op
);
4132 list_push_back(&group
->ops
, &op
->group_node
);
4135 op
->reason
= reason
;
4136 op
->flow_cookie
= rule
->flow_cookie
;
4140 if (type
== OFOPERATION_DELETE
) {
4141 hmap_insert(&ofproto
->deletions
, &op
->hmap_node
,
4142 cls_rule_hash(&rule
->cr
, rule
->table_id
));
4149 ofoperation_destroy(struct ofoperation
*op
)
4151 struct ofopgroup
*group
= op
->group
;
4154 op
->rule
->pending
= NULL
;
4156 if (op
->type
== OFOPERATION_DELETE
) {
4157 hmap_remove(&group
->ofproto
->deletions
, &op
->hmap_node
);
4159 list_remove(&op
->group_node
);
4164 /* Indicates that 'op' completed with status 'error', which is either 0 to
4165 * indicate success or an OpenFlow error code on failure.
4167 * If 'error' is 0, indicating success, the operation will be committed
4168 * permanently to the flow table. There is one interesting subcase:
4170 * - If 'op' is an "add flow" operation that is replacing an existing rule in
4171 * the flow table (the "victim" rule) by a new one, then the caller must
4172 * have uninitialized any derived state in the victim rule, as in step 5 in
4173 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
4174 * performs steps 6 and 7 for the victim rule, most notably by calling its
4175 * ->rule_dealloc() function.
4177 * If 'error' is nonzero, then generally the operation will be rolled back:
4179 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
4180 * restores the original rule. The caller must have uninitialized any
4181 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
4182 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
4183 * and 7 for the new rule, calling its ->rule_dealloc() function.
4185 * - If 'op' is a "modify flow" operation, ofproto restores the original
4188 * - 'op' must not be a "delete flow" operation. Removing a rule is not
4189 * allowed to fail. It must always succeed.
4191 * Please see the large comment in ofproto/ofproto-provider.h titled
4192 * "Asynchronous Operation Support" for more information. */
4194 ofoperation_complete(struct ofoperation
*op
, enum ofperr error
)
4196 struct ofopgroup
*group
= op
->group
;
4198 assert(op
->rule
->pending
== op
);
4199 assert(group
->n_running
> 0);
4200 assert(!error
|| op
->type
!= OFOPERATION_DELETE
);
4203 if (!--group
->n_running
&& !list_is_empty(&group
->ofproto_node
)) {
4204 ofopgroup_complete(group
);
4209 ofoperation_get_victim(struct ofoperation
*op
)
4211 assert(op
->type
== OFOPERATION_ADD
);
4216 pick_datapath_id(const struct ofproto
*ofproto
)
4218 const struct ofport
*port
;
4220 port
= ofproto_get_port(ofproto
, OFPP_LOCAL
);
4222 uint8_t ea
[ETH_ADDR_LEN
];
4225 error
= netdev_get_etheraddr(port
->netdev
, ea
);
4227 return eth_addr_to_uint64(ea
);
4229 VLOG_WARN("%s: could not get MAC address for %s (%s)",
4230 ofproto
->name
, netdev_get_name(port
->netdev
),
4233 return ofproto
->fallback_dpid
;
4237 pick_fallback_dpid(void)
4239 uint8_t ea
[ETH_ADDR_LEN
];
4240 eth_addr_nicira_random(ea
);
4241 return eth_addr_to_uint64(ea
);
4244 /* Table overflow policy. */
4246 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
4247 * is not configured to evict rules or if the table contains no evictable
4248 * rules. (Rules with 'evictable' set to false or with no timeouts are not
4250 static struct rule
*
4251 choose_rule_to_evict(struct oftable
*table
)
4253 struct eviction_group
*evg
;
4255 if (!table
->eviction_fields
) {
4259 /* In the common case, the outer and inner loops here will each be entered
4262 * - The inner loop normally "return"s in its first iteration. If the
4263 * eviction group has any evictable rules, then it always returns in
4266 * - The outer loop only iterates more than once if the largest eviction
4267 * group has no evictable rules.
4269 * - The outer loop can exit only if table's 'max_flows' is all filled up
4270 * by unevictable rules'. */
4271 HEAP_FOR_EACH (evg
, size_node
, &table
->eviction_groups_by_size
) {
4274 HEAP_FOR_EACH (rule
, evg_node
, &evg
->rules
) {
4275 if (rule
->evictable
) {
4284 /* Searches 'ofproto' for tables that have more flows than their configured
4285 * maximum and that have flow eviction enabled, and evicts as many flows as
4286 * necessary and currently feasible from them.
4288 * This triggers only when an OpenFlow table has N flows in it and then the
4289 * client configures a maximum number of flows less than N. */
4291 ofproto_evict(struct ofproto
*ofproto
)
4293 struct ofopgroup
*group
;
4294 struct oftable
*table
;
4296 group
= ofopgroup_create_unattached(ofproto
);
4297 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
4298 while (classifier_count(&table
->cls
) > table
->max_flows
4299 && table
->eviction_fields
) {
4302 rule
= choose_rule_to_evict(table
);
4303 if (!rule
|| rule
->pending
) {
4307 ofoperation_create(group
, rule
,
4308 OFOPERATION_DELETE
, OFPRR_EVICTION
);
4309 oftable_remove_rule(rule
);
4310 ofproto
->ofproto_class
->rule_destruct(rule
);
4313 ofopgroup_submit(group
);
4316 /* Eviction groups. */
4318 /* Returns the priority to use for an eviction_group that contains 'n_rules'
4319 * rules. The priority contains low-order random bits to ensure that eviction
4320 * groups with the same number of rules are prioritized randomly. */
4322 eviction_group_priority(size_t n_rules
)
4324 uint16_t size
= MIN(UINT16_MAX
, n_rules
);
4325 return (size
<< 16) | random_uint16();
4328 /* Updates 'evg', an eviction_group within 'table', following a change that
4329 * adds or removes rules in 'evg'. */
4331 eviction_group_resized(struct oftable
*table
, struct eviction_group
*evg
)
4333 heap_change(&table
->eviction_groups_by_size
, &evg
->size_node
,
4334 eviction_group_priority(heap_count(&evg
->rules
)));
4337 /* Destroys 'evg', an eviction_group within 'table':
4339 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
4340 * rules themselves, just removes them from the eviction group.)
4342 * - Removes 'evg' from 'table'.
4346 eviction_group_destroy(struct oftable
*table
, struct eviction_group
*evg
)
4348 while (!heap_is_empty(&evg
->rules
)) {
4351 rule
= CONTAINER_OF(heap_pop(&evg
->rules
), struct rule
, evg_node
);
4352 rule
->eviction_group
= NULL
;
4354 hmap_remove(&table
->eviction_groups_by_id
, &evg
->id_node
);
4355 heap_remove(&table
->eviction_groups_by_size
, &evg
->size_node
);
4356 heap_destroy(&evg
->rules
);
4360 /* Removes 'rule' from its eviction group, if any. */
4362 eviction_group_remove_rule(struct rule
*rule
)
4364 if (rule
->eviction_group
) {
4365 struct oftable
*table
= &rule
->ofproto
->tables
[rule
->table_id
];
4366 struct eviction_group
*evg
= rule
->eviction_group
;
4368 rule
->eviction_group
= NULL
;
4369 heap_remove(&evg
->rules
, &rule
->evg_node
);
4370 if (heap_is_empty(&evg
->rules
)) {
4371 eviction_group_destroy(table
, evg
);
4373 eviction_group_resized(table
, evg
);
4378 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
4379 * returns the hash value. */
4381 eviction_group_hash_rule(struct rule
*rule
)
4383 struct oftable
*table
= &rule
->ofproto
->tables
[rule
->table_id
];
4384 const struct mf_subfield
*sf
;
4388 hash
= table
->eviction_group_id_basis
;
4389 miniflow_expand(&rule
->cr
.match
.flow
, &flow
);
4390 for (sf
= table
->eviction_fields
;
4391 sf
< &table
->eviction_fields
[table
->n_eviction_fields
];
4394 if (mf_are_prereqs_ok(sf
->field
, &flow
)) {
4395 union mf_value value
;
4397 mf_get_value(sf
->field
, &flow
, &value
);
4399 bitwise_zero(&value
, sf
->field
->n_bytes
, 0, sf
->ofs
);
4401 if (sf
->ofs
+ sf
->n_bits
< sf
->field
->n_bytes
* 8) {
4402 unsigned int start
= sf
->ofs
+ sf
->n_bits
;
4403 bitwise_zero(&value
, sf
->field
->n_bytes
, start
,
4404 sf
->field
->n_bytes
* 8 - start
);
4406 hash
= hash_bytes(&value
, sf
->field
->n_bytes
, hash
);
4408 hash
= hash_int(hash
, 0);
4415 /* Returns an eviction group within 'table' with the given 'id', creating one
4417 static struct eviction_group
*
4418 eviction_group_find(struct oftable
*table
, uint32_t id
)
4420 struct eviction_group
*evg
;
4422 HMAP_FOR_EACH_WITH_HASH (evg
, id_node
, id
, &table
->eviction_groups_by_id
) {
4426 evg
= xmalloc(sizeof *evg
);
4427 hmap_insert(&table
->eviction_groups_by_id
, &evg
->id_node
, id
);
4428 heap_insert(&table
->eviction_groups_by_size
, &evg
->size_node
,
4429 eviction_group_priority(0));
4430 heap_init(&evg
->rules
);
4435 /* Returns an eviction priority for 'rule'. The return value should be
4436 * interpreted so that higher priorities make a rule more attractive candidates
4439 rule_eviction_priority(struct rule
*rule
)
4441 long long int hard_expiration
;
4442 long long int idle_expiration
;
4443 long long int expiration
;
4444 uint32_t expiration_offset
;
4446 /* Calculate time of expiration. */
4447 hard_expiration
= (rule
->hard_timeout
4448 ? rule
->modified
+ rule
->hard_timeout
* 1000
4450 idle_expiration
= (rule
->idle_timeout
4451 ? rule
->used
+ rule
->idle_timeout
* 1000
4453 expiration
= MIN(hard_expiration
, idle_expiration
);
4454 if (expiration
== LLONG_MAX
) {
4458 /* Calculate the time of expiration as a number of (approximate) seconds
4459 * after program startup.
4461 * This should work OK for program runs that last UINT32_MAX seconds or
4462 * less. Therefore, please restart OVS at least once every 136 years. */
4463 expiration_offset
= (expiration
>> 10) - (time_boot_msec() >> 10);
4465 /* Invert the expiration offset because we're using a max-heap. */
4466 return UINT32_MAX
- expiration_offset
;
4469 /* Adds 'rule' to an appropriate eviction group for its oftable's
4470 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
4471 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4474 * The caller must ensure that 'rule' is not already in an eviction group. */
4476 eviction_group_add_rule(struct rule
*rule
)
4478 struct ofproto
*ofproto
= rule
->ofproto
;
4479 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
4481 if (table
->eviction_fields
4482 && (rule
->hard_timeout
|| rule
->idle_timeout
)) {
4483 struct eviction_group
*evg
;
4485 evg
= eviction_group_find(table
, eviction_group_hash_rule(rule
));
4487 rule
->eviction_group
= evg
;
4488 heap_insert(&evg
->rules
, &rule
->evg_node
,
4489 rule_eviction_priority(rule
));
4490 eviction_group_resized(table
, evg
);
4496 /* Initializes 'table'. */
4498 oftable_init(struct oftable
*table
)
4500 memset(table
, 0, sizeof *table
);
4501 classifier_init(&table
->cls
);
4502 table
->max_flows
= UINT_MAX
;
4505 /* Destroys 'table', including its classifier and eviction groups.
4507 * The caller is responsible for freeing 'table' itself. */
4509 oftable_destroy(struct oftable
*table
)
4511 assert(classifier_is_empty(&table
->cls
));
4512 oftable_disable_eviction(table
);
4513 classifier_destroy(&table
->cls
);
4517 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4518 * string, then 'table' will use its default name.
4520 * This only affects the name exposed for a table exposed through the OpenFlow
4521 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4523 oftable_set_name(struct oftable
*table
, const char *name
)
4525 if (name
&& name
[0]) {
4526 int len
= strnlen(name
, OFP_MAX_TABLE_NAME_LEN
);
4527 if (!table
->name
|| strncmp(name
, table
->name
, len
)) {
4529 table
->name
= xmemdup0(name
, len
);
4537 /* oftables support a choice of two policies when adding a rule would cause the
4538 * number of flows in the table to exceed the configured maximum number: either
4539 * they can refuse to add the new flow or they can evict some existing flow.
4540 * This function configures the former policy on 'table'. */
4542 oftable_disable_eviction(struct oftable
*table
)
4544 if (table
->eviction_fields
) {
4545 struct eviction_group
*evg
, *next
;
4547 HMAP_FOR_EACH_SAFE (evg
, next
, id_node
,
4548 &table
->eviction_groups_by_id
) {
4549 eviction_group_destroy(table
, evg
);
4551 hmap_destroy(&table
->eviction_groups_by_id
);
4552 heap_destroy(&table
->eviction_groups_by_size
);
4554 free(table
->eviction_fields
);
4555 table
->eviction_fields
= NULL
;
4556 table
->n_eviction_fields
= 0;
4560 /* oftables support a choice of two policies when adding a rule would cause the
4561 * number of flows in the table to exceed the configured maximum number: either
4562 * they can refuse to add the new flow or they can evict some existing flow.
4563 * This function configures the latter policy on 'table', with fairness based
4564 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4565 * 'n_fields' as 0 disables fairness.) */
4567 oftable_enable_eviction(struct oftable
*table
,
4568 const struct mf_subfield
*fields
, size_t n_fields
)
4570 struct cls_cursor cursor
;
4573 if (table
->eviction_fields
4574 && n_fields
== table
->n_eviction_fields
4576 || !memcmp(fields
, table
->eviction_fields
,
4577 n_fields
* sizeof *fields
))) {
4582 oftable_disable_eviction(table
);
4584 table
->n_eviction_fields
= n_fields
;
4585 table
->eviction_fields
= xmemdup(fields
, n_fields
* sizeof *fields
);
4587 table
->eviction_group_id_basis
= random_uint32();
4588 hmap_init(&table
->eviction_groups_by_id
);
4589 heap_init(&table
->eviction_groups_by_size
);
4591 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
4592 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
4593 eviction_group_add_rule(rule
);
4597 /* Removes 'rule' from the oftable that contains it. */
4599 oftable_remove_rule(struct rule
*rule
)
4601 struct ofproto
*ofproto
= rule
->ofproto
;
4602 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
4604 classifier_remove(&table
->cls
, &rule
->cr
);
4605 eviction_group_remove_rule(rule
);
4608 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4609 * oftable that has an identical cls_rule. Returns the rule that was removed,
4610 * if any, and otherwise NULL. */
4611 static struct rule
*
4612 oftable_replace_rule(struct rule
*rule
)
4614 struct ofproto
*ofproto
= rule
->ofproto
;
4615 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
4616 struct rule
*victim
;
4618 victim
= rule_from_cls_rule(classifier_replace(&table
->cls
, &rule
->cr
));
4620 eviction_group_remove_rule(victim
);
4622 eviction_group_add_rule(rule
);
4626 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4628 oftable_substitute_rule(struct rule
*old
, struct rule
*new)
4631 oftable_replace_rule(new);
4633 oftable_remove_rule(old
);
4637 /* unixctl commands. */
4640 ofproto_lookup(const char *name
)
4642 struct ofproto
*ofproto
;
4644 HMAP_FOR_EACH_WITH_HASH (ofproto
, hmap_node
, hash_string(name
, 0),
4646 if (!strcmp(ofproto
->name
, name
)) {
4654 ofproto_unixctl_list(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
4655 const char *argv
[] OVS_UNUSED
, void *aux OVS_UNUSED
)
4657 struct ofproto
*ofproto
;
4661 HMAP_FOR_EACH (ofproto
, hmap_node
, &all_ofprotos
) {
4662 ds_put_format(&results
, "%s\n", ofproto
->name
);
4664 unixctl_command_reply(conn
, ds_cstr(&results
));
4665 ds_destroy(&results
);
4669 ofproto_unixctl_init(void)
4671 static bool registered
;
4677 unixctl_command_register("ofproto/list", "", 0, 0,
4678 ofproto_unixctl_list
, NULL
);
4681 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4683 * This is deprecated. It is only for compatibility with broken device drivers
4684 * in old versions of Linux that do not properly support VLANs when VLAN
4685 * devices are not used. When broken device drivers are no longer in
4686 * widespread use, we will delete these interfaces. */
4688 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4689 * (exactly) by an OpenFlow rule in 'ofproto'. */
4691 ofproto_get_vlan_usage(struct ofproto
*ofproto
, unsigned long int *vlan_bitmap
)
4693 const struct oftable
*oftable
;
4695 free(ofproto
->vlan_bitmap
);
4696 ofproto
->vlan_bitmap
= bitmap_allocate(4096);
4697 ofproto
->vlans_changed
= false;
4699 OFPROTO_FOR_EACH_TABLE (oftable
, ofproto
) {
4700 const struct cls_table
*table
;
4702 HMAP_FOR_EACH (table
, hmap_node
, &oftable
->cls
.tables
) {
4703 if (minimask_get_vid_mask(&table
->mask
) == VLAN_VID_MASK
) {
4704 const struct cls_rule
*rule
;
4706 HMAP_FOR_EACH (rule
, hmap_node
, &table
->rules
) {
4707 uint16_t vid
= miniflow_get_vid(&rule
->match
.flow
);
4708 bitmap_set1(vlan_bitmap
, vid
);
4709 bitmap_set1(ofproto
->vlan_bitmap
, vid
);
4716 /* Returns true if new VLANs have come into use by the flow table since the
4717 * last call to ofproto_get_vlan_usage().
4719 * We don't track when old VLANs stop being used. */
4721 ofproto_has_vlan_usage_changed(const struct ofproto
*ofproto
)
4723 return ofproto
->vlans_changed
;
4726 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4727 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4728 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4729 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4730 * then the VLAN device is un-enslaved. */
4732 ofproto_port_set_realdev(struct ofproto
*ofproto
, uint16_t vlandev_ofp_port
,
4733 uint16_t realdev_ofp_port
, int vid
)
4735 struct ofport
*ofport
;
4738 assert(vlandev_ofp_port
!= realdev_ofp_port
);
4740 ofport
= ofproto_get_port(ofproto
, vlandev_ofp_port
);
4742 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16
,
4743 ofproto
->name
, vlandev_ofp_port
);
4747 if (!ofproto
->ofproto_class
->set_realdev
) {
4748 if (!vlandev_ofp_port
) {
4751 VLOG_WARN("%s: vlan splinters not supported", ofproto
->name
);
4755 error
= ofproto
->ofproto_class
->set_realdev(ofport
, realdev_ofp_port
, vid
);
4757 VLOG_WARN("%s: setting realdev on port %"PRIu16
" (%s) failed (%s)",
4758 ofproto
->name
, vlandev_ofp_port
,
4759 netdev_get_name(ofport
->netdev
), strerror(error
));