2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
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-errors.h"
36 #include "ofp-print.h"
39 #include "ofproto-provider.h"
40 #include "openflow/nicira-ext.h"
41 #include "openflow/openflow.h"
45 #include "poll-loop.h"
50 #include "unaligned.h"
54 VLOG_DEFINE_THIS_MODULE(ofproto
);
56 COVERAGE_DEFINE(ofproto_error
);
57 COVERAGE_DEFINE(ofproto_flush
);
58 COVERAGE_DEFINE(ofproto_no_packet_in
);
59 COVERAGE_DEFINE(ofproto_packet_out
);
60 COVERAGE_DEFINE(ofproto_queue_req
);
61 COVERAGE_DEFINE(ofproto_recv_openflow
);
62 COVERAGE_DEFINE(ofproto_reinit_ports
);
63 COVERAGE_DEFINE(ofproto_uninstallable
);
64 COVERAGE_DEFINE(ofproto_update_port
);
67 S_OPENFLOW
, /* Processing OpenFlow commands. */
68 S_EVICT
, /* Evicting flows from over-limit tables. */
69 S_FLUSH
, /* Deleting all flow table rules. */
72 enum ofoperation_type
{
78 /* A single OpenFlow request can execute any number of operations. The
79 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
80 * ofconn to which an error reply should be sent if necessary.
82 * ofproto initiates some operations internally. These operations are still
83 * assigned to groups but will not have an associated ofconn. */
85 struct ofproto
*ofproto
; /* Owning ofproto. */
86 struct list ofproto_node
; /* In ofproto's "pending" list. */
87 struct list ops
; /* List of "struct ofoperation"s. */
89 /* Data needed to send OpenFlow reply on failure or to send a buffered
92 * If list_is_empty(ofconn_node) then this ofopgroup never had an
93 * associated ofconn or its ofconn's connection dropped after it initiated
94 * the operation. In the latter case 'ofconn' is a wild pointer that
95 * refers to freed memory, so the 'ofconn' member must be used only if
96 * !list_is_empty(ofconn_node).
98 struct list ofconn_node
; /* In ofconn's list of pending opgroups. */
99 struct ofconn
*ofconn
; /* ofconn for reply (but see note above). */
100 struct ofp_header
*request
; /* Original request (truncated at 64 bytes). */
101 uint32_t buffer_id
; /* Buffer id from original request. */
102 int error
; /* 0 if no error yet, otherwise error code. */
105 static struct ofopgroup
*ofopgroup_create_unattached(struct ofproto
*);
106 static struct ofopgroup
*ofopgroup_create(struct ofproto
*, struct ofconn
*,
107 const struct ofp_header
*,
109 static void ofopgroup_submit(struct ofopgroup
*);
110 static void ofopgroup_destroy(struct ofopgroup
*);
112 /* A single flow table operation. */
114 struct ofopgroup
*group
; /* Owning group. */
115 struct list group_node
; /* In ofopgroup's "ops" list. */
116 struct hmap_node hmap_node
; /* In ofproto's "deletions" hmap. */
117 struct rule
*rule
; /* Rule being operated upon. */
118 enum ofoperation_type type
; /* Type of operation. */
119 int status
; /* -1 if pending, otherwise 0 or error code. */
120 struct rule
*victim
; /* OFOPERATION_ADDING: Replaced rule. */
121 union ofp_action
*actions
; /* OFOPERATION_MODIFYING: Replaced actions. */
122 int n_actions
; /* OFOPERATION_MODIFYING: # of old actions. */
123 ovs_be64 flow_cookie
; /* Rule's old flow cookie. */
126 static void ofoperation_create(struct ofopgroup
*, struct rule
*,
127 enum ofoperation_type
);
128 static void ofoperation_destroy(struct ofoperation
*);
131 static void oftable_init(struct oftable
*);
132 static void oftable_destroy(struct oftable
*);
134 static void oftable_set_name(struct oftable
*, const char *name
);
136 static void oftable_disable_eviction(struct oftable
*);
137 static void oftable_enable_eviction(struct oftable
*,
138 const struct mf_subfield
*fields
,
141 static void oftable_remove_rule(struct rule
*);
142 static struct rule
*oftable_replace_rule(struct rule
*);
143 static void oftable_substitute_rule(struct rule
*old
, struct rule
*new);
145 /* A set of rules within a single OpenFlow table (oftable) that have the same
146 * values for the oftable's eviction_fields. A rule to be evicted, when one is
147 * needed, is taken from the eviction group that contains the greatest number
150 * An oftable owns any number of eviction groups, each of which contains any
153 * Membership in an eviction group is imprecise, based on the hash of the
154 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
155 * That is, if two rules have different eviction_fields, but those
156 * eviction_fields hash to the same value, then they will belong to the same
157 * eviction_group anyway.
159 * (When eviction is not enabled on an oftable, we don't track any eviction
160 * groups, to save time and space.) */
161 struct eviction_group
{
162 struct hmap_node id_node
; /* In oftable's "eviction_groups_by_id". */
163 struct heap_node size_node
; /* In oftable's "eviction_groups_by_size". */
164 struct heap rules
; /* Contains "struct rule"s. */
167 static struct rule
*choose_rule_to_evict(struct oftable
*);
168 static void ofproto_evict(struct ofproto
*);
169 static uint32_t rule_eviction_priority(struct rule
*);
172 static void ofport_destroy__(struct ofport
*);
173 static void ofport_destroy(struct ofport
*);
175 static void update_port(struct ofproto
*, const char *devname
);
176 static int init_ports(struct ofproto
*);
177 static void reinit_ports(struct ofproto
*);
180 static void ofproto_rule_destroy__(struct rule
*);
181 static void ofproto_rule_send_removed(struct rule
*, uint8_t reason
);
182 static bool rule_is_modifiable(const struct rule
*);
183 static bool rule_is_hidden(const struct rule
*);
186 static enum ofperr
add_flow(struct ofproto
*, struct ofconn
*,
187 const struct ofputil_flow_mod
*,
188 const struct ofp_header
*);
189 static void delete_flow__(struct rule
*, struct ofopgroup
*);
190 static bool handle_openflow(struct ofconn
*, struct ofpbuf
*);
191 static enum ofperr
handle_flow_mod__(struct ofproto
*, struct ofconn
*,
192 const struct ofputil_flow_mod
*,
193 const struct ofp_header
*);
196 static uint64_t pick_datapath_id(const struct ofproto
*);
197 static uint64_t pick_fallback_dpid(void);
198 static void ofproto_destroy__(struct ofproto
*);
199 static void set_internal_devs_mtu(struct ofproto
*);
202 static void ofproto_unixctl_init(void);
204 /* All registered ofproto classes, in probe order. */
205 static const struct ofproto_class
**ofproto_classes
;
206 static size_t n_ofproto_classes
;
207 static size_t allocated_ofproto_classes
;
209 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
210 static struct hmap all_ofprotos
= HMAP_INITIALIZER(&all_ofprotos
);
212 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
215 ofproto_initialize(void)
221 ofproto_class_register(&ofproto_dpif_class
);
225 /* 'type' should be a normalized datapath type, as returned by
226 * ofproto_normalize_type(). Returns the corresponding ofproto_class
227 * structure, or a null pointer if there is none registered for 'type'. */
228 static const struct ofproto_class
*
229 ofproto_class_find__(const char *type
)
233 ofproto_initialize();
234 for (i
= 0; i
< n_ofproto_classes
; i
++) {
235 const struct ofproto_class
*class = ofproto_classes
[i
];
240 class->enumerate_types(&types
);
241 found
= sset_contains(&types
, type
);
242 sset_destroy(&types
);
248 VLOG_WARN("unknown datapath type %s", type
);
252 /* Registers a new ofproto class. After successful registration, new ofprotos
253 * of that type can be created using ofproto_create(). */
255 ofproto_class_register(const struct ofproto_class
*new_class
)
259 for (i
= 0; i
< n_ofproto_classes
; i
++) {
260 if (ofproto_classes
[i
] == new_class
) {
265 if (n_ofproto_classes
>= allocated_ofproto_classes
) {
266 ofproto_classes
= x2nrealloc(ofproto_classes
,
267 &allocated_ofproto_classes
,
268 sizeof *ofproto_classes
);
270 ofproto_classes
[n_ofproto_classes
++] = new_class
;
274 /* Unregisters a datapath provider. 'type' must have been previously
275 * registered and not currently be in use by any ofprotos. After
276 * unregistration new datapaths of that type cannot be opened using
277 * ofproto_create(). */
279 ofproto_class_unregister(const struct ofproto_class
*class)
283 for (i
= 0; i
< n_ofproto_classes
; i
++) {
284 if (ofproto_classes
[i
] == class) {
285 for (i
++; i
< n_ofproto_classes
; i
++) {
286 ofproto_classes
[i
- 1] = ofproto_classes
[i
];
292 VLOG_WARN("attempted to unregister an ofproto class that is not "
297 /* Clears 'types' and enumerates all registered ofproto types into it. The
298 * caller must first initialize the sset. */
300 ofproto_enumerate_types(struct sset
*types
)
304 ofproto_initialize();
305 for (i
= 0; i
< n_ofproto_classes
; i
++) {
306 ofproto_classes
[i
]->enumerate_types(types
);
310 /* Returns the fully spelled out name for the given ofproto 'type'.
312 * Normalized type string can be compared with strcmp(). Unnormalized type
313 * string might be the same even if they have different spellings. */
315 ofproto_normalize_type(const char *type
)
317 return type
&& type
[0] ? type
: "system";
320 /* Clears 'names' and enumerates the names of all known created ofprotos with
321 * the given 'type'. The caller must first initialize the sset. Returns 0 if
322 * successful, otherwise a positive errno value.
324 * Some kinds of datapaths might not be practically enumerable. This is not
325 * considered an error. */
327 ofproto_enumerate_names(const char *type
, struct sset
*names
)
329 const struct ofproto_class
*class = ofproto_class_find__(type
);
330 return class ? class->enumerate_names(type
, names
) : EAFNOSUPPORT
;
334 ofproto_create(const char *datapath_name
, const char *datapath_type
,
335 struct ofproto
**ofprotop
)
337 const struct ofproto_class
*class;
338 struct ofproto
*ofproto
;
343 ofproto_initialize();
344 ofproto_unixctl_init();
346 datapath_type
= ofproto_normalize_type(datapath_type
);
347 class = ofproto_class_find__(datapath_type
);
349 VLOG_WARN("could not create datapath %s of unknown type %s",
350 datapath_name
, datapath_type
);
354 ofproto
= class->alloc();
356 VLOG_ERR("failed to allocate datapath %s of type %s",
357 datapath_name
, datapath_type
);
362 memset(ofproto
, 0, sizeof *ofproto
);
363 ofproto
->ofproto_class
= class;
364 ofproto
->name
= xstrdup(datapath_name
);
365 ofproto
->type
= xstrdup(datapath_type
);
366 hmap_insert(&all_ofprotos
, &ofproto
->hmap_node
,
367 hash_string(ofproto
->name
, 0));
368 ofproto
->datapath_id
= 0;
369 ofproto_set_flow_eviction_threshold(ofproto
,
370 OFPROTO_FLOW_EVICTON_THRESHOLD_DEFAULT
);
371 ofproto
->forward_bpdu
= false;
372 ofproto
->fallback_dpid
= pick_fallback_dpid();
373 ofproto
->mfr_desc
= xstrdup(DEFAULT_MFR_DESC
);
374 ofproto
->hw_desc
= xstrdup(DEFAULT_HW_DESC
);
375 ofproto
->sw_desc
= xstrdup(DEFAULT_SW_DESC
);
376 ofproto
->serial_desc
= xstrdup(DEFAULT_SERIAL_DESC
);
377 ofproto
->dp_desc
= xstrdup(DEFAULT_DP_DESC
);
378 ofproto
->frag_handling
= OFPC_FRAG_NORMAL
;
379 hmap_init(&ofproto
->ports
);
380 shash_init(&ofproto
->port_by_name
);
381 ofproto
->tables
= NULL
;
382 ofproto
->n_tables
= 0;
383 ofproto
->connmgr
= connmgr_create(ofproto
, datapath_name
, datapath_name
);
384 ofproto
->state
= S_OPENFLOW
;
385 list_init(&ofproto
->pending
);
386 ofproto
->n_pending
= 0;
387 hmap_init(&ofproto
->deletions
);
388 ofproto
->vlan_bitmap
= NULL
;
389 ofproto
->vlans_changed
= false;
391 error
= ofproto
->ofproto_class
->construct(ofproto
);
393 VLOG_ERR("failed to open datapath %s: %s",
394 datapath_name
, strerror(error
));
395 ofproto_destroy__(ofproto
);
399 assert(ofproto
->n_tables
);
401 ofproto
->datapath_id
= pick_datapath_id(ofproto
);
402 VLOG_INFO("using datapath ID %016"PRIx64
, ofproto
->datapath_id
);
410 ofproto_init_tables(struct ofproto
*ofproto
, int n_tables
)
412 struct oftable
*table
;
414 assert(!ofproto
->n_tables
);
415 assert(n_tables
>= 1 && n_tables
<= 255);
417 ofproto
->n_tables
= n_tables
;
418 ofproto
->tables
= xmalloc(n_tables
* sizeof *ofproto
->tables
);
419 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
425 ofproto_set_datapath_id(struct ofproto
*p
, uint64_t datapath_id
)
427 uint64_t old_dpid
= p
->datapath_id
;
428 p
->datapath_id
= datapath_id
? datapath_id
: pick_datapath_id(p
);
429 if (p
->datapath_id
!= old_dpid
) {
430 VLOG_INFO("datapath ID changed to %016"PRIx64
, p
->datapath_id
);
432 /* Force all active connections to reconnect, since there is no way to
433 * notify a controller that the datapath ID has changed. */
434 ofproto_reconnect_controllers(p
);
439 ofproto_set_controllers(struct ofproto
*p
,
440 const struct ofproto_controller
*controllers
,
441 size_t n_controllers
)
443 connmgr_set_controllers(p
->connmgr
, controllers
, n_controllers
);
447 ofproto_set_fail_mode(struct ofproto
*p
, enum ofproto_fail_mode fail_mode
)
449 connmgr_set_fail_mode(p
->connmgr
, fail_mode
);
452 /* Drops the connections between 'ofproto' and all of its controllers, forcing
453 * them to reconnect. */
455 ofproto_reconnect_controllers(struct ofproto
*ofproto
)
457 connmgr_reconnect(ofproto
->connmgr
);
460 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
461 * in-band control should guarantee access, in the same way that in-band
462 * control guarantees access to OpenFlow controllers. */
464 ofproto_set_extra_in_band_remotes(struct ofproto
*ofproto
,
465 const struct sockaddr_in
*extras
, size_t n
)
467 connmgr_set_extra_in_band_remotes(ofproto
->connmgr
, extras
, n
);
470 /* Sets the OpenFlow queue used by flows set up by in-band control on
471 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
472 * flows will use the default queue. */
474 ofproto_set_in_band_queue(struct ofproto
*ofproto
, int queue_id
)
476 connmgr_set_in_band_queue(ofproto
->connmgr
, queue_id
);
479 /* Sets the number of flows at which eviction from the kernel flow table
482 ofproto_set_flow_eviction_threshold(struct ofproto
*ofproto
, unsigned threshold
)
484 if (threshold
< OFPROTO_FLOW_EVICTION_THRESHOLD_MIN
) {
485 ofproto
->flow_eviction_threshold
= OFPROTO_FLOW_EVICTION_THRESHOLD_MIN
;
487 ofproto
->flow_eviction_threshold
= threshold
;
491 /* If forward_bpdu is true, the NORMAL action will forward frames with
492 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
493 * the NORMAL action will drop these frames. */
495 ofproto_set_forward_bpdu(struct ofproto
*ofproto
, bool forward_bpdu
)
497 bool old_val
= ofproto
->forward_bpdu
;
498 ofproto
->forward_bpdu
= forward_bpdu
;
499 if (old_val
!= ofproto
->forward_bpdu
) {
500 if (ofproto
->ofproto_class
->forward_bpdu_changed
) {
501 ofproto
->ofproto_class
->forward_bpdu_changed(ofproto
);
506 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
507 * 'idle_time', in seconds. */
509 ofproto_set_mac_idle_time(struct ofproto
*ofproto
, unsigned idle_time
)
511 if (ofproto
->ofproto_class
->set_mac_idle_time
) {
512 ofproto
->ofproto_class
->set_mac_idle_time(ofproto
, idle_time
);
517 ofproto_set_desc(struct ofproto
*p
,
518 const char *mfr_desc
, const char *hw_desc
,
519 const char *sw_desc
, const char *serial_desc
,
522 struct ofp_desc_stats
*ods
;
525 if (strlen(mfr_desc
) >= sizeof ods
->mfr_desc
) {
526 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
527 sizeof ods
->mfr_desc
);
530 p
->mfr_desc
= xstrdup(mfr_desc
);
533 if (strlen(hw_desc
) >= sizeof ods
->hw_desc
) {
534 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
535 sizeof ods
->hw_desc
);
538 p
->hw_desc
= xstrdup(hw_desc
);
541 if (strlen(sw_desc
) >= sizeof ods
->sw_desc
) {
542 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
543 sizeof ods
->sw_desc
);
546 p
->sw_desc
= xstrdup(sw_desc
);
549 if (strlen(serial_desc
) >= sizeof ods
->serial_num
) {
550 VLOG_WARN("truncating serial_desc, must be less than %zu "
552 sizeof ods
->serial_num
);
554 free(p
->serial_desc
);
555 p
->serial_desc
= xstrdup(serial_desc
);
558 if (strlen(dp_desc
) >= sizeof ods
->dp_desc
) {
559 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
560 sizeof ods
->dp_desc
);
563 p
->dp_desc
= xstrdup(dp_desc
);
568 ofproto_set_snoops(struct ofproto
*ofproto
, const struct sset
*snoops
)
570 return connmgr_set_snoops(ofproto
->connmgr
, snoops
);
574 ofproto_set_netflow(struct ofproto
*ofproto
,
575 const struct netflow_options
*nf_options
)
577 if (nf_options
&& sset_is_empty(&nf_options
->collectors
)) {
581 if (ofproto
->ofproto_class
->set_netflow
) {
582 return ofproto
->ofproto_class
->set_netflow(ofproto
, nf_options
);
584 return nf_options
? EOPNOTSUPP
: 0;
589 ofproto_set_sflow(struct ofproto
*ofproto
,
590 const struct ofproto_sflow_options
*oso
)
592 if (oso
&& sset_is_empty(&oso
->targets
)) {
596 if (ofproto
->ofproto_class
->set_sflow
) {
597 return ofproto
->ofproto_class
->set_sflow(ofproto
, oso
);
599 return oso
? EOPNOTSUPP
: 0;
603 /* Spanning Tree Protocol (STP) configuration. */
605 /* Configures STP on 'ofproto' using the settings defined in 's'. If
606 * 's' is NULL, disables STP.
608 * Returns 0 if successful, otherwise a positive errno value. */
610 ofproto_set_stp(struct ofproto
*ofproto
,
611 const struct ofproto_stp_settings
*s
)
613 return (ofproto
->ofproto_class
->set_stp
614 ? ofproto
->ofproto_class
->set_stp(ofproto
, s
)
618 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
619 * 'enabled' member of 's' is false, then the other members are not
622 * Returns 0 if successful, otherwise a positive errno value. */
624 ofproto_get_stp_status(struct ofproto
*ofproto
,
625 struct ofproto_stp_status
*s
)
627 return (ofproto
->ofproto_class
->get_stp_status
628 ? ofproto
->ofproto_class
->get_stp_status(ofproto
, s
)
632 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
633 * in 's'. The caller is responsible for assigning STP port numbers
634 * (using the 'port_num' member in the range of 1 through 255, inclusive)
635 * and ensuring there are no duplicates. If the 's' is NULL, then STP
636 * is disabled on the port.
638 * Returns 0 if successful, otherwise a positive errno value.*/
640 ofproto_port_set_stp(struct ofproto
*ofproto
, uint16_t ofp_port
,
641 const struct ofproto_port_stp_settings
*s
)
643 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
645 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16
,
646 ofproto
->name
, ofp_port
);
650 return (ofproto
->ofproto_class
->set_stp_port
651 ? ofproto
->ofproto_class
->set_stp_port(ofport
, s
)
655 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
656 * 's'. If the 'enabled' member in 's' is false, then the other members
657 * are not meaningful.
659 * Returns 0 if successful, otherwise a positive errno value.*/
661 ofproto_port_get_stp_status(struct ofproto
*ofproto
, uint16_t ofp_port
,
662 struct ofproto_port_stp_status
*s
)
664 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
666 VLOG_WARN("%s: cannot get STP status on nonexistent port %"PRIu16
,
667 ofproto
->name
, ofp_port
);
671 return (ofproto
->ofproto_class
->get_stp_port_status
672 ? ofproto
->ofproto_class
->get_stp_port_status(ofport
, s
)
676 /* Queue DSCP configuration. */
678 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
679 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
680 * to implement QoS. Instead, it is used to implement features which require
681 * knowledge of what queues exist on a port, and some basic information about
684 * Returns 0 if successful, otherwise a positive errno value. */
686 ofproto_port_set_queues(struct ofproto
*ofproto
, uint16_t ofp_port
,
687 const struct ofproto_port_queue
*queues
,
690 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
693 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16
,
694 ofproto
->name
, ofp_port
);
698 return (ofproto
->ofproto_class
->set_queues
699 ? ofproto
->ofproto_class
->set_queues(ofport
, queues
, n_queues
)
703 /* Connectivity Fault Management configuration. */
705 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
707 ofproto_port_clear_cfm(struct ofproto
*ofproto
, uint16_t ofp_port
)
709 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
710 if (ofport
&& ofproto
->ofproto_class
->set_cfm
) {
711 ofproto
->ofproto_class
->set_cfm(ofport
, NULL
);
715 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
716 * basic configuration from the configuration members in 'cfm', and the remote
717 * maintenance point ID from remote_mpid. Ignores the statistics members of
720 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
722 ofproto_port_set_cfm(struct ofproto
*ofproto
, uint16_t ofp_port
,
723 const struct cfm_settings
*s
)
725 struct ofport
*ofport
;
728 ofport
= ofproto_get_port(ofproto
, ofp_port
);
730 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16
,
731 ofproto
->name
, ofp_port
);
735 /* XXX: For configuration simplicity, we only support one remote_mpid
736 * outside of the CFM module. It's not clear if this is the correct long
737 * term solution or not. */
738 error
= (ofproto
->ofproto_class
->set_cfm
739 ? ofproto
->ofproto_class
->set_cfm(ofport
, s
)
742 VLOG_WARN("%s: CFM configuration on port %"PRIu16
" (%s) failed (%s)",
743 ofproto
->name
, ofp_port
, netdev_get_name(ofport
->netdev
),
748 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
749 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
750 * 0 if LACP partner information is not current (generally indicating a
751 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
753 ofproto_port_is_lacp_current(struct ofproto
*ofproto
, uint16_t ofp_port
)
755 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
756 return (ofport
&& ofproto
->ofproto_class
->port_is_lacp_current
757 ? ofproto
->ofproto_class
->port_is_lacp_current(ofport
)
763 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
764 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
765 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
766 * configuration plus, if there is more than one slave, a bonding
769 * If 'aux' is already registered then this function updates its configuration
770 * to 's'. Otherwise, this function registers a new bundle.
772 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
775 ofproto_bundle_register(struct ofproto
*ofproto
, void *aux
,
776 const struct ofproto_bundle_settings
*s
)
778 return (ofproto
->ofproto_class
->bundle_set
779 ? ofproto
->ofproto_class
->bundle_set(ofproto
, aux
, s
)
783 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
784 * If no such bundle has been registered, this has no effect. */
786 ofproto_bundle_unregister(struct ofproto
*ofproto
, void *aux
)
788 return ofproto_bundle_register(ofproto
, aux
, NULL
);
792 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
793 * If 'aux' is already registered then this function updates its configuration
794 * to 's'. Otherwise, this function registers a new mirror. */
796 ofproto_mirror_register(struct ofproto
*ofproto
, void *aux
,
797 const struct ofproto_mirror_settings
*s
)
799 return (ofproto
->ofproto_class
->mirror_set
800 ? ofproto
->ofproto_class
->mirror_set(ofproto
, aux
, s
)
804 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
805 * If no mirror has been registered, this has no effect. */
807 ofproto_mirror_unregister(struct ofproto
*ofproto
, void *aux
)
809 return ofproto_mirror_register(ofproto
, aux
, NULL
);
812 /* Retrieves statistics from mirror associated with client data pointer
813 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
814 * 'bytes', respectively. If a particular counters is not supported,
815 * the appropriate argument is set to UINT64_MAX. */
817 ofproto_mirror_get_stats(struct ofproto
*ofproto
, void *aux
,
818 uint64_t *packets
, uint64_t *bytes
)
820 if (!ofproto
->ofproto_class
->mirror_get_stats
) {
821 *packets
= *bytes
= UINT64_MAX
;
825 return ofproto
->ofproto_class
->mirror_get_stats(ofproto
, aux
,
829 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
830 * which all packets are flooded, instead of using MAC learning. If
831 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
833 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
836 ofproto_set_flood_vlans(struct ofproto
*ofproto
, unsigned long *flood_vlans
)
838 return (ofproto
->ofproto_class
->set_flood_vlans
839 ? ofproto
->ofproto_class
->set_flood_vlans(ofproto
, flood_vlans
)
843 /* Returns true if 'aux' is a registered bundle that is currently in use as the
844 * output for a mirror. */
846 ofproto_is_mirror_output_bundle(const struct ofproto
*ofproto
, void *aux
)
848 return (ofproto
->ofproto_class
->is_mirror_output_bundle
849 ? ofproto
->ofproto_class
->is_mirror_output_bundle(ofproto
, aux
)
853 /* Configuration of OpenFlow tables. */
855 /* Returns the number of OpenFlow tables in 'ofproto'. */
857 ofproto_get_n_tables(const struct ofproto
*ofproto
)
859 return ofproto
->n_tables
;
862 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
863 * settings from 's'. 'table_id' must be in the range 0 through the number of
864 * OpenFlow tables in 'ofproto' minus 1, inclusive.
866 * For read-only tables, only the name may be configured. */
868 ofproto_configure_table(struct ofproto
*ofproto
, int table_id
,
869 const struct ofproto_table_settings
*s
)
871 struct oftable
*table
;
873 assert(table_id
>= 0 && table_id
< ofproto
->n_tables
);
874 table
= &ofproto
->tables
[table_id
];
876 oftable_set_name(table
, s
->name
);
878 if (table
->flags
& OFTABLE_READONLY
) {
883 oftable_enable_eviction(table
, s
->groups
, s
->n_groups
);
885 oftable_disable_eviction(table
);
888 table
->max_flows
= s
->max_flows
;
889 if (classifier_count(&table
->cls
) > table
->max_flows
890 && table
->eviction_fields
) {
891 /* 'table' contains more flows than allowed. We might not be able to
892 * evict them right away because of the asynchronous nature of flow
893 * table changes. Schedule eviction for later. */
894 switch (ofproto
->state
) {
896 ofproto
->state
= S_EVICT
;
900 /* We're already deleting flows, nothing more to do. */
907 ofproto_has_snoops(const struct ofproto
*ofproto
)
909 return connmgr_has_snoops(ofproto
->connmgr
);
913 ofproto_get_snoops(const struct ofproto
*ofproto
, struct sset
*snoops
)
915 connmgr_get_snoops(ofproto
->connmgr
, snoops
);
919 ofproto_flush__(struct ofproto
*ofproto
)
921 struct ofopgroup
*group
;
922 struct oftable
*table
;
924 if (ofproto
->ofproto_class
->flush
) {
925 ofproto
->ofproto_class
->flush(ofproto
);
928 group
= ofopgroup_create_unattached(ofproto
);
929 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
930 struct rule
*rule
, *next_rule
;
931 struct cls_cursor cursor
;
933 if (table
->flags
& OFTABLE_HIDDEN
) {
937 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
938 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
939 if (!rule
->pending
) {
940 ofoperation_create(group
, rule
, OFOPERATION_DELETE
);
941 oftable_remove_rule(rule
);
942 ofproto
->ofproto_class
->rule_destruct(rule
);
946 ofopgroup_submit(group
);
950 ofproto_destroy__(struct ofproto
*ofproto
)
952 struct oftable
*table
;
954 assert(list_is_empty(&ofproto
->pending
));
955 assert(!ofproto
->n_pending
);
957 connmgr_destroy(ofproto
->connmgr
);
959 hmap_remove(&all_ofprotos
, &ofproto
->hmap_node
);
962 free(ofproto
->mfr_desc
);
963 free(ofproto
->hw_desc
);
964 free(ofproto
->sw_desc
);
965 free(ofproto
->serial_desc
);
966 free(ofproto
->dp_desc
);
967 hmap_destroy(&ofproto
->ports
);
968 shash_destroy(&ofproto
->port_by_name
);
970 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
971 oftable_destroy(table
);
973 free(ofproto
->tables
);
975 hmap_destroy(&ofproto
->deletions
);
977 free(ofproto
->vlan_bitmap
);
979 ofproto
->ofproto_class
->dealloc(ofproto
);
983 ofproto_destroy(struct ofproto
*p
)
985 struct ofport
*ofport
, *next_ofport
;
992 HMAP_FOR_EACH_SAFE (ofport
, next_ofport
, hmap_node
, &p
->ports
) {
993 ofport_destroy(ofport
);
996 p
->ofproto_class
->destruct(p
);
997 ofproto_destroy__(p
);
1000 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1001 * kernel datapath, for example, this destroys the datapath in the kernel, and
1002 * with the netdev-based datapath, it tears down the data structures that
1003 * represent the datapath.
1005 * The datapath should not be currently open as an ofproto. */
1007 ofproto_delete(const char *name
, const char *type
)
1009 const struct ofproto_class
*class = ofproto_class_find__(type
);
1010 return (!class ? EAFNOSUPPORT
1011 : !class->del
? EACCES
1012 : class->del(type
, name
));
1016 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
1018 if (error
== ENOBUFS
) {
1019 reinit_ports(ofproto
);
1020 } else if (!error
) {
1021 update_port(ofproto
, devname
);
1027 ofproto_run(struct ofproto
*p
)
1029 struct ofport
*ofport
;
1033 error
= p
->ofproto_class
->run(p
);
1034 if (error
&& error
!= EAGAIN
) {
1035 VLOG_ERR_RL(&rl
, "%s: run failed (%s)", p
->name
, strerror(error
));
1038 if (p
->ofproto_class
->port_poll
) {
1039 while ((error
= p
->ofproto_class
->port_poll(p
, &devname
)) != EAGAIN
) {
1040 process_port_change(p
, error
, devname
);
1044 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1045 unsigned int change_seq
= netdev_change_seq(ofport
->netdev
);
1046 if (ofport
->change_seq
!= change_seq
) {
1047 ofport
->change_seq
= change_seq
;
1048 update_port(p
, netdev_get_name(ofport
->netdev
));
1054 connmgr_run(p
->connmgr
, handle_openflow
);
1058 connmgr_run(p
->connmgr
, NULL
);
1060 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1061 p
->state
= S_OPENFLOW
;
1066 connmgr_run(p
->connmgr
, NULL
);
1068 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1069 connmgr_flushed(p
->connmgr
);
1070 p
->state
= S_OPENFLOW
;
1081 /* Performs periodic activity required by 'ofproto' that needs to be done
1082 * with the least possible latency.
1084 * It makes sense to call this function a couple of times per poll loop, to
1085 * provide a significant performance boost on some benchmarks with the
1086 * ofproto-dpif implementation. */
1088 ofproto_run_fast(struct ofproto
*p
)
1092 error
= p
->ofproto_class
->run_fast
? p
->ofproto_class
->run_fast(p
) : 0;
1093 if (error
&& error
!= EAGAIN
) {
1094 VLOG_ERR_RL(&rl
, "%s: fastpath run failed (%s)",
1095 p
->name
, strerror(error
));
1101 ofproto_wait(struct ofproto
*p
)
1103 struct ofport
*ofport
;
1105 p
->ofproto_class
->wait(p
);
1106 if (p
->ofproto_class
->port_poll_wait
) {
1107 p
->ofproto_class
->port_poll_wait(p
);
1110 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1111 if (ofport
->change_seq
!= netdev_change_seq(ofport
->netdev
)) {
1112 poll_immediate_wake();
1118 connmgr_wait(p
->connmgr
, true);
1123 connmgr_wait(p
->connmgr
, false);
1124 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1125 poll_immediate_wake();
1132 ofproto_is_alive(const struct ofproto
*p
)
1134 return connmgr_has_controllers(p
->connmgr
);
1138 ofproto_get_ofproto_controller_info(const struct ofproto
*ofproto
,
1141 connmgr_get_controller_info(ofproto
->connmgr
, info
);
1145 ofproto_free_ofproto_controller_info(struct shash
*info
)
1147 connmgr_free_controller_info(info
);
1150 /* Makes a deep copy of 'old' into 'port'. */
1152 ofproto_port_clone(struct ofproto_port
*port
, const struct ofproto_port
*old
)
1154 port
->name
= xstrdup(old
->name
);
1155 port
->type
= xstrdup(old
->type
);
1156 port
->ofp_port
= old
->ofp_port
;
1159 /* Frees memory allocated to members of 'ofproto_port'.
1161 * Do not call this function on an ofproto_port obtained from
1162 * ofproto_port_dump_next(): that function retains ownership of the data in the
1165 ofproto_port_destroy(struct ofproto_port
*ofproto_port
)
1167 free(ofproto_port
->name
);
1168 free(ofproto_port
->type
);
1171 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1173 * This function provides no status indication. An error status for the entire
1174 * dump operation is provided when it is completed by calling
1175 * ofproto_port_dump_done().
1178 ofproto_port_dump_start(struct ofproto_port_dump
*dump
,
1179 const struct ofproto
*ofproto
)
1181 dump
->ofproto
= ofproto
;
1182 dump
->error
= ofproto
->ofproto_class
->port_dump_start(ofproto
,
1186 /* Attempts to retrieve another port from 'dump', which must have been created
1187 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1188 * 'port' and returns true. On failure, returns false.
1190 * Failure might indicate an actual error or merely that the last port has been
1191 * dumped. An error status for the entire dump operation is provided when it
1192 * is completed by calling ofproto_port_dump_done().
1194 * The ofproto owns the data stored in 'port'. It will remain valid until at
1195 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1196 * ofproto_port_dump_done(). */
1198 ofproto_port_dump_next(struct ofproto_port_dump
*dump
,
1199 struct ofproto_port
*port
)
1201 const struct ofproto
*ofproto
= dump
->ofproto
;
1207 dump
->error
= ofproto
->ofproto_class
->port_dump_next(ofproto
, dump
->state
,
1210 ofproto
->ofproto_class
->port_dump_done(ofproto
, dump
->state
);
1216 /* Completes port table dump operation 'dump', which must have been created
1217 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1218 * error-free, otherwise a positive errno value describing the problem. */
1220 ofproto_port_dump_done(struct ofproto_port_dump
*dump
)
1222 const struct ofproto
*ofproto
= dump
->ofproto
;
1224 dump
->error
= ofproto
->ofproto_class
->port_dump_done(ofproto
,
1227 return dump
->error
== EOF
? 0 : dump
->error
;
1230 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
1231 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
1232 * is non-null). On failure, returns a positive errno value and sets
1233 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
1235 ofproto_port_add(struct ofproto
*ofproto
, struct netdev
*netdev
,
1236 uint16_t *ofp_portp
)
1241 error
= ofproto
->ofproto_class
->port_add(ofproto
, netdev
, &ofp_port
);
1243 update_port(ofproto
, netdev_get_name(netdev
));
1246 *ofp_portp
= error
? OFPP_NONE
: ofp_port
;
1251 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1252 * initializes '*port' appropriately; on failure, returns a positive errno
1255 * The caller owns the data in 'ofproto_port' and must free it with
1256 * ofproto_port_destroy() when it is no longer needed. */
1258 ofproto_port_query_by_name(const struct ofproto
*ofproto
, const char *devname
,
1259 struct ofproto_port
*port
)
1263 error
= ofproto
->ofproto_class
->port_query_by_name(ofproto
, devname
, port
);
1265 memset(port
, 0, sizeof *port
);
1270 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1271 * Returns 0 if successful, otherwise a positive errno. */
1273 ofproto_port_del(struct ofproto
*ofproto
, uint16_t ofp_port
)
1275 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
1276 const char *name
= ofport
? netdev_get_name(ofport
->netdev
) : "<unknown>";
1279 error
= ofproto
->ofproto_class
->port_del(ofproto
, ofp_port
);
1280 if (!error
&& ofport
) {
1281 /* 'name' is the netdev's name and update_port() is going to close the
1282 * netdev. Just in case update_port() refers to 'name' after it
1283 * destroys 'ofport', make a copy of it around the update_port()
1285 char *devname
= xstrdup(name
);
1286 update_port(ofproto
, devname
);
1292 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1293 * performs the 'n_actions' actions in 'actions'. The new flow will not
1296 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1297 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1298 * controllers; otherwise, it will be hidden.
1300 * The caller retains ownership of 'cls_rule' and 'actions'.
1302 * This is a helper function for in-band control and fail-open. */
1304 ofproto_add_flow(struct ofproto
*ofproto
, const struct cls_rule
*cls_rule
,
1305 const union ofp_action
*actions
, size_t n_actions
)
1307 const struct rule
*rule
;
1309 rule
= rule_from_cls_rule(classifier_find_rule_exactly(
1310 &ofproto
->tables
[0].cls
, cls_rule
));
1311 if (!rule
|| !ofputil_actions_equal(rule
->actions
, rule
->n_actions
,
1312 actions
, n_actions
)) {
1313 struct ofputil_flow_mod fm
;
1315 memset(&fm
, 0, sizeof fm
);
1317 fm
.buffer_id
= UINT32_MAX
;
1318 fm
.actions
= (union ofp_action
*) actions
;
1319 fm
.n_actions
= n_actions
;
1320 add_flow(ofproto
, NULL
, &fm
, NULL
);
1324 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1325 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1326 * operation cannot be initiated now but may be retried later.
1328 * This is a helper function for in-band control and fail-open. */
1330 ofproto_flow_mod(struct ofproto
*ofproto
, const struct ofputil_flow_mod
*fm
)
1332 return handle_flow_mod__(ofproto
, NULL
, fm
, NULL
);
1335 /* Searches for a rule with matching criteria exactly equal to 'target' in
1336 * ofproto's table 0 and, if it finds one, deletes it.
1338 * This is a helper function for in-band control and fail-open. */
1340 ofproto_delete_flow(struct ofproto
*ofproto
, const struct cls_rule
*target
)
1344 rule
= rule_from_cls_rule(classifier_find_rule_exactly(
1345 &ofproto
->tables
[0].cls
, target
));
1347 /* No such rule -> success. */
1349 } else if (rule
->pending
) {
1350 /* An operation on the rule is already pending -> failure.
1351 * Caller must retry later if it's important. */
1354 /* Initiate deletion -> success. */
1355 struct ofopgroup
*group
= ofopgroup_create_unattached(ofproto
);
1356 ofoperation_create(group
, rule
, OFOPERATION_DELETE
);
1357 oftable_remove_rule(rule
);
1358 ofproto
->ofproto_class
->rule_destruct(rule
);
1359 ofopgroup_submit(group
);
1365 /* Starts the process of deleting all of the flows from all of ofproto's flow
1366 * tables and then reintroducing the flows required by in-band control and
1367 * fail-open. The process will complete in a later call to ofproto_run(). */
1369 ofproto_flush_flows(struct ofproto
*ofproto
)
1371 COVERAGE_INC(ofproto_flush
);
1372 ofproto
->state
= S_FLUSH
;
1376 reinit_ports(struct ofproto
*p
)
1378 struct ofproto_port_dump dump
;
1379 struct sset devnames
;
1380 struct ofport
*ofport
;
1381 struct ofproto_port ofproto_port
;
1382 const char *devname
;
1384 COVERAGE_INC(ofproto_reinit_ports
);
1386 sset_init(&devnames
);
1387 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1388 sset_add(&devnames
, netdev_get_name(ofport
->netdev
));
1390 OFPROTO_PORT_FOR_EACH (&ofproto_port
, &dump
, p
) {
1391 sset_add(&devnames
, ofproto_port
.name
);
1394 SSET_FOR_EACH (devname
, &devnames
) {
1395 update_port(p
, devname
);
1397 sset_destroy(&devnames
);
1400 /* Opens and returns a netdev for 'ofproto_port', or a null pointer if the
1401 * netdev cannot be opened. On success, also fills in 'opp'. */
1402 static struct netdev
*
1403 ofport_open(const struct ofproto_port
*ofproto_port
, struct ofp_phy_port
*opp
)
1405 uint32_t curr
, advertised
, supported
, peer
;
1406 enum netdev_flags flags
;
1407 struct netdev
*netdev
;
1410 error
= netdev_open(ofproto_port
->name
, ofproto_port
->type
, &netdev
);
1412 VLOG_WARN_RL(&rl
, "ignoring port %s (%"PRIu16
") because netdev %s "
1413 "cannot be opened (%s)",
1414 ofproto_port
->name
, ofproto_port
->ofp_port
,
1415 ofproto_port
->name
, strerror(error
));
1419 netdev_get_flags(netdev
, &flags
);
1420 netdev_get_features(netdev
, &curr
, &advertised
, &supported
, &peer
);
1422 opp
->port_no
= htons(ofproto_port
->ofp_port
);
1423 netdev_get_etheraddr(netdev
, opp
->hw_addr
);
1424 ovs_strzcpy(opp
->name
, ofproto_port
->name
, sizeof opp
->name
);
1425 opp
->config
= flags
& NETDEV_UP
? 0 : htonl(OFPPC_PORT_DOWN
);
1426 opp
->state
= netdev_get_carrier(netdev
) ? 0 : htonl(OFPPS_LINK_DOWN
);
1427 opp
->curr
= htonl(curr
);
1428 opp
->advertised
= htonl(advertised
);
1429 opp
->supported
= htonl(supported
);
1430 opp
->peer
= htonl(peer
);
1435 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1436 * port number, and 'config' bits other than OFPPC_PORT_DOWN are
1439 ofport_equal(const struct ofp_phy_port
*a
, const struct ofp_phy_port
*b
)
1441 BUILD_ASSERT_DECL(sizeof *a
== 48); /* Detect ofp_phy_port changes. */
1442 return (!memcmp(a
->hw_addr
, b
->hw_addr
, sizeof a
->hw_addr
)
1443 && a
->state
== b
->state
1444 && !((a
->config
^ b
->config
) & htonl(OFPPC_PORT_DOWN
))
1445 && a
->curr
== b
->curr
1446 && a
->advertised
== b
->advertised
1447 && a
->supported
== b
->supported
1448 && a
->peer
== b
->peer
);
1451 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1452 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1453 * one with the same name or port number). */
1455 ofport_install(struct ofproto
*p
,
1456 struct netdev
*netdev
, const struct ofp_phy_port
*opp
)
1458 const char *netdev_name
= netdev_get_name(netdev
);
1459 struct ofport
*ofport
;
1463 /* Create ofport. */
1464 ofport
= p
->ofproto_class
->port_alloc();
1469 ofport
->ofproto
= p
;
1470 ofport
->netdev
= netdev
;
1471 ofport
->change_seq
= netdev_change_seq(netdev
);
1473 ofport
->ofp_port
= ntohs(opp
->port_no
);
1475 /* Add port to 'p'. */
1476 hmap_insert(&p
->ports
, &ofport
->hmap_node
, hash_int(ofport
->ofp_port
, 0));
1477 shash_add(&p
->port_by_name
, netdev_name
, ofport
);
1479 if (!netdev_get_mtu(netdev
, &dev_mtu
)) {
1480 set_internal_devs_mtu(p
);
1481 ofport
->mtu
= dev_mtu
;
1486 /* Let the ofproto_class initialize its private data. */
1487 error
= p
->ofproto_class
->port_construct(ofport
);
1491 connmgr_send_port_status(p
->connmgr
, opp
, OFPPR_ADD
);
1495 VLOG_WARN_RL(&rl
, "%s: could not add port %s (%s)",
1496 p
->name
, netdev_name
, strerror(error
));
1498 ofport_destroy__(ofport
);
1500 netdev_close(netdev
);
1504 /* Removes 'ofport' from 'p' and destroys it. */
1506 ofport_remove(struct ofport
*ofport
)
1508 connmgr_send_port_status(ofport
->ofproto
->connmgr
, &ofport
->opp
,
1510 ofport_destroy(ofport
);
1513 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1516 ofport_remove_with_name(struct ofproto
*ofproto
, const char *name
)
1518 struct ofport
*port
= shash_find_data(&ofproto
->port_by_name
, name
);
1520 ofport_remove(port
);
1524 /* Updates 'port' with new 'opp' description.
1526 * Does not handle a name or port number change. The caller must implement
1527 * such a change as a delete followed by an add. */
1529 ofport_modified(struct ofport
*port
, struct ofp_phy_port
*opp
)
1531 memcpy(port
->opp
.hw_addr
, opp
->hw_addr
, ETH_ADDR_LEN
);
1532 port
->opp
.config
= ((port
->opp
.config
& ~htonl(OFPPC_PORT_DOWN
))
1533 | (opp
->config
& htonl(OFPPC_PORT_DOWN
)));
1534 port
->opp
.state
= opp
->state
;
1535 port
->opp
.curr
= opp
->curr
;
1536 port
->opp
.advertised
= opp
->advertised
;
1537 port
->opp
.supported
= opp
->supported
;
1538 port
->opp
.peer
= opp
->peer
;
1540 connmgr_send_port_status(port
->ofproto
->connmgr
, &port
->opp
, OFPPR_MODIFY
);
1543 /* Update OpenFlow 'state' in 'port' and notify controller. */
1545 ofproto_port_set_state(struct ofport
*port
, ovs_be32 state
)
1547 if (port
->opp
.state
!= state
) {
1548 port
->opp
.state
= state
;
1549 connmgr_send_port_status(port
->ofproto
->connmgr
, &port
->opp
,
1555 ofproto_port_unregister(struct ofproto
*ofproto
, uint16_t ofp_port
)
1557 struct ofport
*port
= ofproto_get_port(ofproto
, ofp_port
);
1559 if (port
->ofproto
->ofproto_class
->set_realdev
) {
1560 port
->ofproto
->ofproto_class
->set_realdev(port
, 0, 0);
1562 if (port
->ofproto
->ofproto_class
->set_stp_port
) {
1563 port
->ofproto
->ofproto_class
->set_stp_port(port
, NULL
);
1565 if (port
->ofproto
->ofproto_class
->set_cfm
) {
1566 port
->ofproto
->ofproto_class
->set_cfm(port
, NULL
);
1568 if (port
->ofproto
->ofproto_class
->bundle_remove
) {
1569 port
->ofproto
->ofproto_class
->bundle_remove(port
);
1575 ofport_destroy__(struct ofport
*port
)
1577 struct ofproto
*ofproto
= port
->ofproto
;
1578 const char *name
= netdev_get_name(port
->netdev
);
1580 hmap_remove(&ofproto
->ports
, &port
->hmap_node
);
1581 shash_delete(&ofproto
->port_by_name
,
1582 shash_find(&ofproto
->port_by_name
, name
));
1584 netdev_close(port
->netdev
);
1585 ofproto
->ofproto_class
->port_dealloc(port
);
1589 ofport_destroy(struct ofport
*port
)
1592 port
->ofproto
->ofproto_class
->port_destruct(port
);
1593 ofport_destroy__(port
);
1598 ofproto_get_port(const struct ofproto
*ofproto
, uint16_t ofp_port
)
1600 struct ofport
*port
;
1602 HMAP_FOR_EACH_IN_BUCKET (port
, hmap_node
,
1603 hash_int(ofp_port
, 0), &ofproto
->ports
) {
1604 if (port
->ofp_port
== ofp_port
) {
1612 ofproto_port_get_stats(const struct ofport
*port
, struct netdev_stats
*stats
)
1614 struct ofproto
*ofproto
= port
->ofproto
;
1617 if (ofproto
->ofproto_class
->port_get_stats
) {
1618 error
= ofproto
->ofproto_class
->port_get_stats(port
, stats
);
1627 update_port(struct ofproto
*ofproto
, const char *name
)
1629 struct ofproto_port ofproto_port
;
1630 struct ofp_phy_port opp
;
1631 struct netdev
*netdev
;
1632 struct ofport
*port
;
1634 COVERAGE_INC(ofproto_update_port
);
1636 /* Fetch 'name''s location and properties from the datapath. */
1637 netdev
= (!ofproto_port_query_by_name(ofproto
, name
, &ofproto_port
)
1638 ? ofport_open(&ofproto_port
, &opp
)
1641 port
= ofproto_get_port(ofproto
, ofproto_port
.ofp_port
);
1642 if (port
&& !strcmp(netdev_get_name(port
->netdev
), name
)) {
1643 struct netdev
*old_netdev
= port
->netdev
;
1646 /* 'name' hasn't changed location. Any properties changed? */
1647 if (!ofport_equal(&port
->opp
, &opp
)) {
1648 ofport_modified(port
, &opp
);
1651 /* If this is a non-internal port and the MTU changed, check
1652 * if the datapath's MTU needs to be updated. */
1653 if (strcmp(netdev_get_type(netdev
), "internal")
1654 && !netdev_get_mtu(netdev
, &dev_mtu
)
1655 && port
->mtu
!= dev_mtu
) {
1656 set_internal_devs_mtu(ofproto
);
1657 port
->mtu
= dev_mtu
;
1660 /* Install the newly opened netdev in case it has changed.
1661 * Don't close the old netdev yet in case port_modified has to
1662 * remove a retained reference to it.*/
1663 port
->netdev
= netdev
;
1664 port
->change_seq
= netdev_change_seq(netdev
);
1666 if (port
->ofproto
->ofproto_class
->port_modified
) {
1667 port
->ofproto
->ofproto_class
->port_modified(port
);
1670 netdev_close(old_netdev
);
1672 /* If 'port' is nonnull then its name differs from 'name' and thus
1673 * we should delete it. If we think there's a port named 'name'
1674 * then its port number must be wrong now so delete it too. */
1676 ofport_remove(port
);
1678 ofport_remove_with_name(ofproto
, name
);
1679 ofport_install(ofproto
, netdev
, &opp
);
1682 /* Any port named 'name' is gone now. */
1683 ofport_remove_with_name(ofproto
, name
);
1685 ofproto_port_destroy(&ofproto_port
);
1689 init_ports(struct ofproto
*p
)
1691 struct ofproto_port_dump dump
;
1692 struct ofproto_port ofproto_port
;
1694 OFPROTO_PORT_FOR_EACH (&ofproto_port
, &dump
, p
) {
1695 uint16_t ofp_port
= ofproto_port
.ofp_port
;
1696 if (ofproto_get_port(p
, ofp_port
)) {
1697 VLOG_WARN_RL(&rl
, "ignoring duplicate port %"PRIu16
" in datapath",
1699 } else if (shash_find(&p
->port_by_name
, ofproto_port
.name
)) {
1700 VLOG_WARN_RL(&rl
, "ignoring duplicate device %s in datapath",
1703 struct ofp_phy_port opp
;
1704 struct netdev
*netdev
;
1706 netdev
= ofport_open(&ofproto_port
, &opp
);
1708 ofport_install(p
, netdev
, &opp
);
1716 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
1717 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
1719 find_min_mtu(struct ofproto
*p
)
1721 struct ofport
*ofport
;
1724 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1725 struct netdev
*netdev
= ofport
->netdev
;
1728 /* Skip any internal ports, since that's what we're trying to
1730 if (!strcmp(netdev_get_type(netdev
), "internal")) {
1734 if (netdev_get_mtu(netdev
, &dev_mtu
)) {
1737 if (!mtu
|| dev_mtu
< mtu
) {
1742 return mtu
? mtu
: ETH_PAYLOAD_MAX
;
1745 /* Set the MTU of all datapath devices on 'p' to the minimum of the
1746 * non-datapath ports. */
1748 set_internal_devs_mtu(struct ofproto
*p
)
1750 struct ofport
*ofport
;
1751 int mtu
= find_min_mtu(p
);
1753 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1754 struct netdev
*netdev
= ofport
->netdev
;
1756 if (!strcmp(netdev_get_type(netdev
), "internal")) {
1757 netdev_set_mtu(netdev
, mtu
);
1763 ofproto_rule_destroy__(struct rule
*rule
)
1766 free(rule
->actions
);
1767 rule
->ofproto
->ofproto_class
->rule_dealloc(rule
);
1771 /* This function allows an ofproto implementation to destroy any rules that
1772 * remain when its ->destruct() function is called. The caller must have
1773 * already uninitialized any derived members of 'rule' (step 5 described in the
1774 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
1775 * This function implements steps 6 and 7.
1777 * This function should only be called from an ofproto implementation's
1778 * ->destruct() function. It is not suitable elsewhere. */
1780 ofproto_rule_destroy(struct rule
*rule
)
1782 assert(!rule
->pending
);
1783 oftable_remove_rule(rule
);
1784 ofproto_rule_destroy__(rule
);
1787 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1788 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1791 rule_has_out_port(const struct rule
*rule
, uint16_t out_port
)
1793 const union ofp_action
*oa
;
1796 if (out_port
== OFPP_NONE
) {
1799 OFPUTIL_ACTION_FOR_EACH_UNSAFE (oa
, left
, rule
->actions
, rule
->n_actions
) {
1800 if (action_outputs_to_port(oa
, htons(out_port
))) {
1807 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1808 * statistics appropriately. 'packet' must have at least sizeof(struct
1809 * ofp_packet_in) bytes of headroom.
1811 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1812 * with statistics for 'packet' either way.
1814 * Takes ownership of 'packet'. */
1816 rule_execute(struct rule
*rule
, uint16_t in_port
, struct ofpbuf
*packet
)
1820 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
1822 flow_extract(packet
, 0, 0, in_port
, &flow
);
1823 return rule
->ofproto
->ofproto_class
->rule_execute(rule
, &flow
, packet
);
1826 /* Returns true if 'rule' should be hidden from the controller.
1828 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1829 * (e.g. by in-band control) and are intentionally hidden from the
1832 rule_is_hidden(const struct rule
*rule
)
1834 return rule
->cr
.priority
> UINT16_MAX
;
1837 static enum oftable_flags
1838 rule_get_flags(const struct rule
*rule
)
1840 return rule
->ofproto
->tables
[rule
->table_id
].flags
;
1844 rule_is_modifiable(const struct rule
*rule
)
1846 return !(rule_get_flags(rule
) & OFTABLE_READONLY
);
1850 handle_echo_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
1852 ofconn_send_reply(ofconn
, make_echo_reply(oh
));
1857 handle_features_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
1859 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
1860 struct ofp_switch_features
*osf
;
1862 struct ofport
*port
;
1866 ofproto
->ofproto_class
->get_features(ofproto
, &arp_match_ip
, &actions
);
1867 assert(actions
& (1 << OFPAT_OUTPUT
)); /* sanity check */
1869 osf
= make_openflow_xid(sizeof *osf
, OFPT_FEATURES_REPLY
, oh
->xid
, &buf
);
1870 osf
->datapath_id
= htonll(ofproto
->datapath_id
);
1871 osf
->n_buffers
= htonl(pktbuf_capacity());
1872 osf
->n_tables
= ofproto
->n_tables
;
1873 osf
->capabilities
= htonl(OFPC_FLOW_STATS
| OFPC_TABLE_STATS
|
1874 OFPC_PORT_STATS
| OFPC_QUEUE_STATS
);
1876 osf
->capabilities
|= htonl(OFPC_ARP_MATCH_IP
);
1878 osf
->actions
= htonl(actions
);
1880 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
1881 ofpbuf_put(buf
, &port
->opp
, sizeof port
->opp
);
1884 ofconn_send_reply(ofconn
, buf
);
1889 handle_get_config_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
1891 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
1892 struct ofp_switch_config
*osc
;
1893 enum ofp_config_flags flags
;
1897 osc
= make_openflow_xid(sizeof *osc
, OFPT_GET_CONFIG_REPLY
, oh
->xid
, &buf
);
1898 flags
= ofproto
->frag_handling
;
1899 if (ofconn_get_invalid_ttl_to_controller(ofconn
)) {
1900 flags
|= OFPC_INVALID_TTL_TO_CONTROLLER
;
1902 osc
->flags
= htons(flags
);
1903 osc
->miss_send_len
= htons(ofconn_get_miss_send_len(ofconn
));
1904 ofconn_send_reply(ofconn
, buf
);
1910 handle_set_config(struct ofconn
*ofconn
, const struct ofp_switch_config
*osc
)
1912 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
1913 uint16_t flags
= ntohs(osc
->flags
);
1915 if (ofconn_get_type(ofconn
) != OFCONN_PRIMARY
1916 || ofconn_get_role(ofconn
) != NX_ROLE_SLAVE
) {
1917 enum ofp_config_flags cur
= ofproto
->frag_handling
;
1918 enum ofp_config_flags next
= flags
& OFPC_FRAG_MASK
;
1920 assert((cur
& OFPC_FRAG_MASK
) == cur
);
1922 if (ofproto
->ofproto_class
->set_frag_handling(ofproto
, next
)) {
1923 ofproto
->frag_handling
= next
;
1925 VLOG_WARN_RL(&rl
, "%s: unsupported fragment handling mode %s",
1927 ofputil_frag_handling_to_string(next
));
1931 ofconn_set_invalid_ttl_to_controller(ofconn
,
1932 (flags
& OFPC_INVALID_TTL_TO_CONTROLLER
));
1934 ofconn_set_miss_send_len(ofconn
, ntohs(osc
->miss_send_len
));
1939 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
1940 * error message code for the caller to propagate upward. Otherwise, returns
1943 * The log message mentions 'msg_type'. */
1945 reject_slave_controller(struct ofconn
*ofconn
)
1947 if (ofconn_get_type(ofconn
) == OFCONN_PRIMARY
1948 && ofconn_get_role(ofconn
) == NX_ROLE_SLAVE
) {
1949 return OFPERR_OFPBRC_EPERM
;
1956 handle_packet_out(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
1958 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
1959 struct ofp_packet_out
*opo
;
1960 struct ofpbuf payload
, *buffer
;
1961 union ofp_action
*ofp_actions
;
1962 struct ofpbuf request
;
1964 size_t n_ofp_actions
;
1968 COVERAGE_INC(ofproto_packet_out
);
1970 error
= reject_slave_controller(ofconn
);
1975 /* Get ofp_packet_out. */
1976 ofpbuf_use_const(&request
, oh
, ntohs(oh
->length
));
1977 opo
= ofpbuf_pull(&request
, offsetof(struct ofp_packet_out
, actions
));
1980 error
= ofputil_pull_actions(&request
, ntohs(opo
->actions_len
),
1981 &ofp_actions
, &n_ofp_actions
);
1987 if (opo
->buffer_id
!= htonl(UINT32_MAX
)) {
1988 error
= ofconn_pktbuf_retrieve(ofconn
, ntohl(opo
->buffer_id
),
1990 if (error
|| !buffer
) {
1999 /* Get in_port and partially validate it.
2001 * We don't know what range of ports the ofproto actually implements, but
2002 * we do know that only certain reserved ports (numbered OFPP_MAX and
2003 * above) are valid. */
2004 in_port
= ntohs(opo
->in_port
);
2005 if (in_port
>= OFPP_MAX
&& in_port
!= OFPP_LOCAL
&& in_port
!= OFPP_NONE
) {
2006 return OFPERR_NXBRC_BAD_IN_PORT
;
2009 /* Send out packet. */
2010 flow_extract(&payload
, 0, 0, in_port
, &flow
);
2011 error
= p
->ofproto_class
->packet_out(p
, &payload
, &flow
,
2012 ofp_actions
, n_ofp_actions
);
2013 ofpbuf_delete(buffer
);
2019 update_port_config(struct ofport
*port
, ovs_be32 config
, ovs_be32 mask
)
2021 ovs_be32 old_config
= port
->opp
.config
;
2023 mask
&= config
^ port
->opp
.config
;
2024 if (mask
& htonl(OFPPC_PORT_DOWN
)) {
2025 if (config
& htonl(OFPPC_PORT_DOWN
)) {
2026 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, true);
2028 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, true);
2032 port
->opp
.config
^= mask
& (htonl(OFPPC_NO_RECV
| OFPPC_NO_RECV_STP
|
2033 OFPPC_NO_FLOOD
| OFPPC_NO_FWD
|
2034 OFPPC_NO_PACKET_IN
));
2035 if (port
->opp
.config
!= old_config
) {
2036 port
->ofproto
->ofproto_class
->port_reconfigured(port
, old_config
);
2041 handle_port_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2043 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2044 const struct ofp_port_mod
*opm
= (const struct ofp_port_mod
*) oh
;
2045 struct ofport
*port
;
2048 error
= reject_slave_controller(ofconn
);
2053 port
= ofproto_get_port(p
, ntohs(opm
->port_no
));
2055 return OFPERR_OFPPMFC_BAD_PORT
;
2056 } else if (memcmp(port
->opp
.hw_addr
, opm
->hw_addr
, OFP_ETH_ALEN
)) {
2057 return OFPERR_OFPPMFC_BAD_HW_ADDR
;
2059 update_port_config(port
, opm
->config
, opm
->mask
);
2060 if (opm
->advertise
) {
2061 netdev_set_advertisements(port
->netdev
, ntohl(opm
->advertise
));
2068 handle_desc_stats_request(struct ofconn
*ofconn
,
2069 const struct ofp_stats_msg
*request
)
2071 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2072 struct ofp_desc_stats
*ods
;
2075 ods
= ofputil_make_stats_reply(sizeof *ods
, request
, &msg
);
2076 ovs_strlcpy(ods
->mfr_desc
, p
->mfr_desc
, sizeof ods
->mfr_desc
);
2077 ovs_strlcpy(ods
->hw_desc
, p
->hw_desc
, sizeof ods
->hw_desc
);
2078 ovs_strlcpy(ods
->sw_desc
, p
->sw_desc
, sizeof ods
->sw_desc
);
2079 ovs_strlcpy(ods
->serial_num
, p
->serial_desc
, sizeof ods
->serial_num
);
2080 ovs_strlcpy(ods
->dp_desc
, p
->dp_desc
, sizeof ods
->dp_desc
);
2081 ofconn_send_reply(ofconn
, msg
);
2087 handle_table_stats_request(struct ofconn
*ofconn
,
2088 const struct ofp_stats_msg
*request
)
2090 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2091 struct ofp_table_stats
*ots
;
2095 ofputil_make_stats_reply(sizeof(struct ofp_stats_msg
), request
, &msg
);
2097 ots
= ofpbuf_put_zeros(msg
, sizeof *ots
* p
->n_tables
);
2098 for (i
= 0; i
< p
->n_tables
; i
++) {
2099 ots
[i
].table_id
= i
;
2100 sprintf(ots
[i
].name
, "table%zu", i
);
2101 ots
[i
].wildcards
= htonl(OFPFW_ALL
);
2102 ots
[i
].max_entries
= htonl(1000000); /* An arbitrary big number. */
2103 ots
[i
].active_count
= htonl(classifier_count(&p
->tables
[i
].cls
));
2106 p
->ofproto_class
->get_tables(p
, ots
);
2108 for (i
= 0; i
< p
->n_tables
; i
++) {
2109 const struct oftable
*table
= &p
->tables
[i
];
2112 ovs_strzcpy(ots
[i
].name
, table
->name
, sizeof ots
[i
].name
);
2115 if (table
->max_flows
< ntohl(ots
[i
].max_entries
)) {
2116 ots
[i
].max_entries
= htonl(table
->max_flows
);
2120 ofconn_send_reply(ofconn
, msg
);
2125 append_port_stat(struct ofport
*port
, struct list
*replies
)
2127 struct netdev_stats stats
;
2128 struct ofp_port_stats
*ops
;
2130 /* Intentionally ignore return value, since errors will set
2131 * 'stats' to all-1s, which is correct for OpenFlow, and
2132 * netdev_get_stats() will log errors. */
2133 ofproto_port_get_stats(port
, &stats
);
2135 ops
= ofputil_append_stats_reply(sizeof *ops
, replies
);
2136 ops
->port_no
= port
->opp
.port_no
;
2137 memset(ops
->pad
, 0, sizeof ops
->pad
);
2138 put_32aligned_be64(&ops
->rx_packets
, htonll(stats
.rx_packets
));
2139 put_32aligned_be64(&ops
->tx_packets
, htonll(stats
.tx_packets
));
2140 put_32aligned_be64(&ops
->rx_bytes
, htonll(stats
.rx_bytes
));
2141 put_32aligned_be64(&ops
->tx_bytes
, htonll(stats
.tx_bytes
));
2142 put_32aligned_be64(&ops
->rx_dropped
, htonll(stats
.rx_dropped
));
2143 put_32aligned_be64(&ops
->tx_dropped
, htonll(stats
.tx_dropped
));
2144 put_32aligned_be64(&ops
->rx_errors
, htonll(stats
.rx_errors
));
2145 put_32aligned_be64(&ops
->tx_errors
, htonll(stats
.tx_errors
));
2146 put_32aligned_be64(&ops
->rx_frame_err
, htonll(stats
.rx_frame_errors
));
2147 put_32aligned_be64(&ops
->rx_over_err
, htonll(stats
.rx_over_errors
));
2148 put_32aligned_be64(&ops
->rx_crc_err
, htonll(stats
.rx_crc_errors
));
2149 put_32aligned_be64(&ops
->collisions
, htonll(stats
.collisions
));
2153 handle_port_stats_request(struct ofconn
*ofconn
,
2154 const struct ofp_port_stats_request
*psr
)
2156 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2157 struct ofport
*port
;
2158 struct list replies
;
2160 ofputil_start_stats_reply(&psr
->osm
, &replies
);
2161 if (psr
->port_no
!= htons(OFPP_NONE
)) {
2162 port
= ofproto_get_port(p
, ntohs(psr
->port_no
));
2164 append_port_stat(port
, &replies
);
2167 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
2168 append_port_stat(port
, &replies
);
2172 ofconn_send_replies(ofconn
, &replies
);
2177 calc_flow_duration__(long long int start
, long long int now
,
2178 uint32_t *sec
, uint32_t *nsec
)
2180 long long int msecs
= now
- start
;
2181 *sec
= msecs
/ 1000;
2182 *nsec
= (msecs
% 1000) * (1000 * 1000);
2185 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2186 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2188 check_table_id(const struct ofproto
*ofproto
, uint8_t table_id
)
2190 return (table_id
== 0xff || table_id
< ofproto
->n_tables
2192 : OFPERR_NXBRC_BAD_TABLE_ID
);
2196 static struct oftable
*
2197 next_visible_table(struct ofproto
*ofproto
, uint8_t table_id
)
2199 struct oftable
*table
;
2201 for (table
= &ofproto
->tables
[table_id
];
2202 table
< &ofproto
->tables
[ofproto
->n_tables
];
2204 if (!(table
->flags
& OFTABLE_HIDDEN
)) {
2212 static struct oftable
*
2213 first_matching_table(struct ofproto
*ofproto
, uint8_t table_id
)
2215 if (table_id
== 0xff) {
2216 return next_visible_table(ofproto
, 0);
2217 } else if (table_id
< ofproto
->n_tables
) {
2218 return &ofproto
->tables
[table_id
];
2224 static struct oftable
*
2225 next_matching_table(struct ofproto
*ofproto
,
2226 struct oftable
*table
, uint8_t table_id
)
2228 return (table_id
== 0xff
2229 ? next_visible_table(ofproto
, (table
- ofproto
->tables
) + 1)
2233 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2235 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2236 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2238 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2239 * only once, for that table. (This can be used to access tables marked
2242 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2243 * entered at all. (Perhaps you should have validated TABLE_ID with
2244 * check_table_id().)
2246 * All parameters are evaluated multiple times.
2248 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2249 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2251 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2253 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2254 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2255 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2258 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2259 * to 'out_port' are included.
2261 * Hidden rules are always omitted.
2263 * Returns 0 on success, otherwise an OpenFlow error code. */
2265 collect_rules_loose(struct ofproto
*ofproto
, uint8_t table_id
,
2266 const struct cls_rule
*match
,
2267 ovs_be64 cookie
, ovs_be64 cookie_mask
,
2268 uint16_t out_port
, struct list
*rules
)
2270 struct oftable
*table
;
2273 error
= check_table_id(ofproto
, table_id
);
2279 FOR_EACH_MATCHING_TABLE (table
, table_id
, ofproto
) {
2280 struct cls_cursor cursor
;
2283 cls_cursor_init(&cursor
, &table
->cls
, match
);
2284 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
2285 if (rule
->pending
) {
2286 return OFPROTO_POSTPONE
;
2288 if (!rule_is_hidden(rule
) && rule_has_out_port(rule
, out_port
)
2289 && !((rule
->flow_cookie
^ cookie
) & cookie_mask
)) {
2290 list_push_back(rules
, &rule
->ofproto_node
);
2297 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2298 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2299 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2302 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2303 * to 'out_port' are included.
2305 * Hidden rules are always omitted.
2307 * Returns 0 on success, otherwise an OpenFlow error code. */
2309 collect_rules_strict(struct ofproto
*ofproto
, uint8_t table_id
,
2310 const struct cls_rule
*match
,
2311 ovs_be64 cookie
, ovs_be64 cookie_mask
,
2312 uint16_t out_port
, struct list
*rules
)
2314 struct oftable
*table
;
2317 error
= check_table_id(ofproto
, table_id
);
2323 FOR_EACH_MATCHING_TABLE (table
, table_id
, ofproto
) {
2326 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&table
->cls
,
2329 if (rule
->pending
) {
2330 return OFPROTO_POSTPONE
;
2332 if (!rule_is_hidden(rule
) && rule_has_out_port(rule
, out_port
)
2333 && !((rule
->flow_cookie
^ cookie
) & cookie_mask
)) {
2334 list_push_back(rules
, &rule
->ofproto_node
);
2341 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2342 * forced into the range of a uint16_t. */
2344 age_secs(long long int age_ms
)
2346 return (age_ms
< 0 ? 0
2347 : age_ms
>= UINT16_MAX
* 1000 ? UINT16_MAX
2348 : (unsigned int) age_ms
/ 1000);
2352 handle_flow_stats_request(struct ofconn
*ofconn
,
2353 const struct ofp_stats_msg
*osm
)
2355 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2356 struct ofputil_flow_stats_request fsr
;
2357 struct list replies
;
2362 error
= ofputil_decode_flow_stats_request(&fsr
, &osm
->header
);
2367 error
= collect_rules_loose(ofproto
, fsr
.table_id
, &fsr
.match
,
2368 fsr
.cookie
, fsr
.cookie_mask
,
2369 fsr
.out_port
, &rules
);
2374 ofputil_start_stats_reply(osm
, &replies
);
2375 LIST_FOR_EACH (rule
, ofproto_node
, &rules
) {
2376 long long int now
= time_msec();
2377 struct ofputil_flow_stats fs
;
2380 fs
.cookie
= rule
->flow_cookie
;
2381 fs
.table_id
= rule
->table_id
;
2382 calc_flow_duration__(rule
->created
, now
, &fs
.duration_sec
,
2384 fs
.idle_timeout
= rule
->idle_timeout
;
2385 fs
.hard_timeout
= rule
->hard_timeout
;
2386 fs
.idle_age
= age_secs(now
- rule
->used
);
2387 fs
.hard_age
= age_secs(now
- rule
->modified
);
2388 ofproto
->ofproto_class
->rule_get_stats(rule
, &fs
.packet_count
,
2390 fs
.actions
= rule
->actions
;
2391 fs
.n_actions
= rule
->n_actions
;
2392 ofputil_append_flow_stats_reply(&fs
, &replies
);
2394 ofconn_send_replies(ofconn
, &replies
);
2400 flow_stats_ds(struct rule
*rule
, struct ds
*results
)
2402 uint64_t packet_count
, byte_count
;
2404 rule
->ofproto
->ofproto_class
->rule_get_stats(rule
,
2405 &packet_count
, &byte_count
);
2407 if (rule
->table_id
!= 0) {
2408 ds_put_format(results
, "table_id=%"PRIu8
", ", rule
->table_id
);
2410 ds_put_format(results
, "duration=%llds, ",
2411 (time_msec() - rule
->created
) / 1000);
2412 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
2413 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
2414 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
2415 cls_rule_format(&rule
->cr
, results
);
2416 ds_put_char(results
, ',');
2417 if (rule
->n_actions
> 0) {
2418 ofp_print_actions(results
, rule
->actions
, rule
->n_actions
);
2420 ds_put_cstr(results
, "drop");
2422 ds_put_cstr(results
, "\n");
2425 /* Adds a pretty-printed description of all flows to 'results', including
2426 * hidden flows (e.g., set up by in-band control). */
2428 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
2430 struct oftable
*table
;
2432 OFPROTO_FOR_EACH_TABLE (table
, p
) {
2433 struct cls_cursor cursor
;
2436 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
2437 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
2438 flow_stats_ds(rule
, results
);
2443 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2444 * '*engine_type' and '*engine_id', respectively. */
2446 ofproto_get_netflow_ids(const struct ofproto
*ofproto
,
2447 uint8_t *engine_type
, uint8_t *engine_id
)
2449 ofproto
->ofproto_class
->get_netflow_ids(ofproto
, engine_type
, engine_id
);
2452 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns 1
2453 * if CFM is faulted (generally indiciating a connectivity problem), 0 if CFM
2454 * is not faulted, and -1 if CFM is not enabled on 'ofp_port'. */
2456 ofproto_port_get_cfm_fault(const struct ofproto
*ofproto
, uint16_t ofp_port
)
2458 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
2459 return (ofport
&& ofproto
->ofproto_class
->get_cfm_fault
2460 ? ofproto
->ofproto_class
->get_cfm_fault(ofport
)
2464 /* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
2465 * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by
2466 * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a
2467 * number less than 0 if CFM is not enabled on 'ofp_port'. */
2469 ofproto_port_get_cfm_remote_mpids(const struct ofproto
*ofproto
,
2470 uint16_t ofp_port
, const uint64_t **rmps
,
2473 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
2477 return (ofport
&& ofproto
->ofproto_class
->get_cfm_remote_mpids
2478 ? ofproto
->ofproto_class
->get_cfm_remote_mpids(ofport
, rmps
,
2484 handle_aggregate_stats_request(struct ofconn
*ofconn
,
2485 const struct ofp_stats_msg
*osm
)
2487 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2488 struct ofputil_flow_stats_request request
;
2489 struct ofputil_aggregate_stats stats
;
2490 bool unknown_packets
, unknown_bytes
;
2491 struct ofpbuf
*reply
;
2496 error
= ofputil_decode_flow_stats_request(&request
, &osm
->header
);
2501 error
= collect_rules_loose(ofproto
, request
.table_id
, &request
.match
,
2502 request
.cookie
, request
.cookie_mask
,
2503 request
.out_port
, &rules
);
2508 memset(&stats
, 0, sizeof stats
);
2509 unknown_packets
= unknown_bytes
= false;
2510 LIST_FOR_EACH (rule
, ofproto_node
, &rules
) {
2511 uint64_t packet_count
;
2512 uint64_t byte_count
;
2514 ofproto
->ofproto_class
->rule_get_stats(rule
, &packet_count
,
2517 if (packet_count
== UINT64_MAX
) {
2518 unknown_packets
= true;
2520 stats
.packet_count
+= packet_count
;
2523 if (byte_count
== UINT64_MAX
) {
2524 unknown_bytes
= true;
2526 stats
.byte_count
+= byte_count
;
2531 if (unknown_packets
) {
2532 stats
.packet_count
= UINT64_MAX
;
2534 if (unknown_bytes
) {
2535 stats
.byte_count
= UINT64_MAX
;
2538 reply
= ofputil_encode_aggregate_stats_reply(&stats
, osm
);
2539 ofconn_send_reply(ofconn
, reply
);
2544 struct queue_stats_cbdata
{
2545 struct ofport
*ofport
;
2546 struct list replies
;
2550 put_queue_stats(struct queue_stats_cbdata
*cbdata
, uint32_t queue_id
,
2551 const struct netdev_queue_stats
*stats
)
2553 struct ofp_queue_stats
*reply
;
2555 reply
= ofputil_append_stats_reply(sizeof *reply
, &cbdata
->replies
);
2556 reply
->port_no
= cbdata
->ofport
->opp
.port_no
;
2557 memset(reply
->pad
, 0, sizeof reply
->pad
);
2558 reply
->queue_id
= htonl(queue_id
);
2559 put_32aligned_be64(&reply
->tx_bytes
, htonll(stats
->tx_bytes
));
2560 put_32aligned_be64(&reply
->tx_packets
, htonll(stats
->tx_packets
));
2561 put_32aligned_be64(&reply
->tx_errors
, htonll(stats
->tx_errors
));
2565 handle_queue_stats_dump_cb(uint32_t queue_id
,
2566 struct netdev_queue_stats
*stats
,
2569 struct queue_stats_cbdata
*cbdata
= cbdata_
;
2571 put_queue_stats(cbdata
, queue_id
, stats
);
2575 handle_queue_stats_for_port(struct ofport
*port
, uint32_t queue_id
,
2576 struct queue_stats_cbdata
*cbdata
)
2578 cbdata
->ofport
= port
;
2579 if (queue_id
== OFPQ_ALL
) {
2580 netdev_dump_queue_stats(port
->netdev
,
2581 handle_queue_stats_dump_cb
, cbdata
);
2583 struct netdev_queue_stats stats
;
2585 if (!netdev_get_queue_stats(port
->netdev
, queue_id
, &stats
)) {
2586 put_queue_stats(cbdata
, queue_id
, &stats
);
2592 handle_queue_stats_request(struct ofconn
*ofconn
,
2593 const struct ofp_queue_stats_request
*qsr
)
2595 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2596 struct queue_stats_cbdata cbdata
;
2597 struct ofport
*port
;
2598 unsigned int port_no
;
2601 COVERAGE_INC(ofproto_queue_req
);
2603 ofputil_start_stats_reply(&qsr
->osm
, &cbdata
.replies
);
2605 port_no
= ntohs(qsr
->port_no
);
2606 queue_id
= ntohl(qsr
->queue_id
);
2607 if (port_no
== OFPP_ALL
) {
2608 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
2609 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
2611 } else if (port_no
< OFPP_MAX
) {
2612 port
= ofproto_get_port(ofproto
, port_no
);
2614 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
2617 ofpbuf_list_delete(&cbdata
.replies
);
2618 return OFPERR_OFPQOFC_BAD_PORT
;
2620 ofconn_send_replies(ofconn
, &cbdata
.replies
);
2626 is_flow_deletion_pending(const struct ofproto
*ofproto
,
2627 const struct cls_rule
*cls_rule
,
2630 if (!hmap_is_empty(&ofproto
->deletions
)) {
2631 struct ofoperation
*op
;
2633 HMAP_FOR_EACH_WITH_HASH (op
, hmap_node
,
2634 cls_rule_hash(cls_rule
, table_id
),
2635 &ofproto
->deletions
) {
2636 if (cls_rule_equal(cls_rule
, &op
->rule
->cr
)) {
2645 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2646 * in which no matching flow already exists in the flow table.
2648 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2649 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
2650 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
2651 * initiated now but may be retried later.
2653 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2656 add_flow(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
2657 const struct ofputil_flow_mod
*fm
, const struct ofp_header
*request
)
2659 struct oftable
*table
;
2660 struct ofopgroup
*group
;
2661 struct rule
*victim
;
2665 error
= check_table_id(ofproto
, fm
->table_id
);
2671 if (fm
->table_id
== 0xff) {
2673 if (ofproto
->ofproto_class
->rule_choose_table
) {
2674 error
= ofproto
->ofproto_class
->rule_choose_table(ofproto
, &fm
->cr
,
2679 assert(table_id
< ofproto
->n_tables
);
2680 table
= &ofproto
->tables
[table_id
];
2682 table
= &ofproto
->tables
[0];
2684 } else if (fm
->table_id
< ofproto
->n_tables
) {
2685 table
= &ofproto
->tables
[fm
->table_id
];
2687 return OFPERR_NXFMFC_BAD_TABLE_ID
;
2690 if (table
->flags
& OFTABLE_READONLY
) {
2691 return OFPERR_OFPBRC_EPERM
;
2694 /* Check for overlap, if requested. */
2695 if (fm
->flags
& OFPFF_CHECK_OVERLAP
2696 && classifier_rule_overlaps(&table
->cls
, &fm
->cr
)) {
2697 return OFPERR_OFPFMFC_OVERLAP
;
2700 /* Serialize against pending deletion. */
2701 if (is_flow_deletion_pending(ofproto
, &fm
->cr
, table
- ofproto
->tables
)) {
2702 return OFPROTO_POSTPONE
;
2705 /* Allocate new rule. */
2706 rule
= ofproto
->ofproto_class
->rule_alloc();
2708 VLOG_WARN_RL(&rl
, "%s: failed to create rule (%s)",
2709 ofproto
->name
, strerror(error
));
2712 rule
->ofproto
= ofproto
;
2714 rule
->pending
= NULL
;
2715 rule
->flow_cookie
= fm
->cookie
;
2716 rule
->created
= rule
->modified
= rule
->used
= time_msec();
2717 rule
->idle_timeout
= fm
->idle_timeout
;
2718 rule
->hard_timeout
= fm
->hard_timeout
;
2719 rule
->table_id
= table
- ofproto
->tables
;
2720 rule
->send_flow_removed
= (fm
->flags
& OFPFF_SEND_FLOW_REM
) != 0;
2721 rule
->actions
= ofputil_actions_clone(fm
->actions
, fm
->n_actions
);
2722 rule
->n_actions
= fm
->n_actions
;
2723 rule
->evictable
= true;
2724 rule
->eviction_group
= NULL
;
2726 /* Insert new rule. */
2727 victim
= oftable_replace_rule(rule
);
2728 if (victim
&& !rule_is_modifiable(victim
)) {
2729 error
= OFPERR_OFPBRC_EPERM
;
2730 } else if (victim
&& victim
->pending
) {
2731 error
= OFPROTO_POSTPONE
;
2735 if (classifier_count(&table
->cls
) > table
->max_flows
) {
2738 was_evictable
= rule
->evictable
;
2739 rule
->evictable
= false;
2740 evict
= choose_rule_to_evict(table
);
2741 rule
->evictable
= was_evictable
;
2744 error
= OFPERR_OFPFMFC_ALL_TABLES_FULL
;
2746 } else if (evict
->pending
) {
2747 error
= OFPROTO_POSTPONE
;
2754 group
= ofopgroup_create(ofproto
, ofconn
, request
, fm
->buffer_id
);
2755 ofoperation_create(group
, rule
, OFOPERATION_ADD
);
2756 rule
->pending
->victim
= victim
;
2758 error
= ofproto
->ofproto_class
->rule_construct(rule
);
2760 ofoperation_destroy(rule
->pending
);
2762 delete_flow__(evict
, group
);
2764 ofopgroup_submit(group
);
2768 /* Back out if an error occurred. */
2770 oftable_substitute_rule(rule
, victim
);
2771 ofproto_rule_destroy__(rule
);
2776 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2778 /* Modifies the rules listed in 'rules', changing their actions to match those
2781 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2784 * Returns 0 on success, otherwise an OpenFlow error code. */
2786 modify_flows__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
2787 const struct ofputil_flow_mod
*fm
,
2788 const struct ofp_header
*request
, struct list
*rules
)
2790 struct ofopgroup
*group
;
2794 group
= ofopgroup_create(ofproto
, ofconn
, request
, fm
->buffer_id
);
2795 error
= OFPERR_OFPBRC_EPERM
;
2796 LIST_FOR_EACH (rule
, ofproto_node
, rules
) {
2797 if (rule_is_modifiable(rule
)) {
2798 /* At least one rule is modifiable, don't report EPERM error. */
2804 if (!ofputil_actions_equal(fm
->actions
, fm
->n_actions
,
2805 rule
->actions
, rule
->n_actions
)) {
2806 ofoperation_create(group
, rule
, OFOPERATION_MODIFY
);
2807 rule
->pending
->actions
= rule
->actions
;
2808 rule
->pending
->n_actions
= rule
->n_actions
;
2809 rule
->actions
= ofputil_actions_clone(fm
->actions
, fm
->n_actions
);
2810 rule
->n_actions
= fm
->n_actions
;
2811 ofproto
->ofproto_class
->rule_modify_actions(rule
);
2813 rule
->modified
= time_msec();
2815 rule
->flow_cookie
= fm
->cookie
;
2817 ofopgroup_submit(group
);
2822 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
2825 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2828 modify_flows_loose(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
2829 const struct ofputil_flow_mod
*fm
,
2830 const struct ofp_header
*request
)
2835 error
= collect_rules_loose(ofproto
, fm
->table_id
, &fm
->cr
,
2836 fm
->cookie
, fm
->cookie_mask
,
2838 return (error
? error
2839 : list_is_empty(&rules
) ? add_flow(ofproto
, ofconn
, fm
, request
)
2840 : modify_flows__(ofproto
, ofconn
, fm
, request
, &rules
));
2843 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
2846 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2849 modify_flow_strict(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
2850 const struct ofputil_flow_mod
*fm
,
2851 const struct ofp_header
*request
)
2856 error
= collect_rules_strict(ofproto
, fm
->table_id
, &fm
->cr
,
2857 fm
->cookie
, fm
->cookie_mask
,
2859 return (error
? error
2860 : list_is_empty(&rules
) ? add_flow(ofproto
, ofconn
, fm
, request
)
2861 : list_is_singleton(&rules
) ? modify_flows__(ofproto
, ofconn
,
2862 fm
, request
, &rules
)
2866 /* OFPFC_DELETE implementation. */
2869 delete_flow__(struct rule
*rule
, struct ofopgroup
*group
)
2871 struct ofproto
*ofproto
= rule
->ofproto
;
2873 ofproto_rule_send_removed(rule
, OFPRR_DELETE
);
2875 ofoperation_create(group
, rule
, OFOPERATION_DELETE
);
2876 oftable_remove_rule(rule
);
2877 ofproto
->ofproto_class
->rule_destruct(rule
);
2880 /* Deletes the rules listed in 'rules'.
2882 * Returns 0 on success, otherwise an OpenFlow error code. */
2884 delete_flows__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
2885 const struct ofp_header
*request
, struct list
*rules
)
2887 struct rule
*rule
, *next
;
2888 struct ofopgroup
*group
;
2890 group
= ofopgroup_create(ofproto
, ofconn
, request
, UINT32_MAX
);
2891 LIST_FOR_EACH_SAFE (rule
, next
, ofproto_node
, rules
) {
2892 delete_flow__(rule
, group
);
2894 ofopgroup_submit(group
);
2899 /* Implements OFPFC_DELETE. */
2901 delete_flows_loose(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
2902 const struct ofputil_flow_mod
*fm
,
2903 const struct ofp_header
*request
)
2908 error
= collect_rules_loose(ofproto
, fm
->table_id
, &fm
->cr
,
2909 fm
->cookie
, fm
->cookie_mask
,
2910 fm
->out_port
, &rules
);
2911 return (error
? error
2912 : !list_is_empty(&rules
) ? delete_flows__(ofproto
, ofconn
, request
,
2917 /* Implements OFPFC_DELETE_STRICT. */
2919 delete_flow_strict(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
2920 const struct ofputil_flow_mod
*fm
,
2921 const struct ofp_header
*request
)
2926 error
= collect_rules_strict(ofproto
, fm
->table_id
, &fm
->cr
,
2927 fm
->cookie
, fm
->cookie_mask
,
2928 fm
->out_port
, &rules
);
2929 return (error
? error
2930 : list_is_singleton(&rules
) ? delete_flows__(ofproto
, ofconn
,
2936 ofproto_rule_send_removed(struct rule
*rule
, uint8_t reason
)
2938 struct ofputil_flow_removed fr
;
2940 if (rule_is_hidden(rule
) || !rule
->send_flow_removed
) {
2945 fr
.cookie
= rule
->flow_cookie
;
2947 calc_flow_duration__(rule
->created
, time_msec(),
2948 &fr
.duration_sec
, &fr
.duration_nsec
);
2949 fr
.idle_timeout
= rule
->idle_timeout
;
2950 rule
->ofproto
->ofproto_class
->rule_get_stats(rule
, &fr
.packet_count
,
2953 connmgr_send_flow_removed(rule
->ofproto
->connmgr
, &fr
);
2957 ofproto_rule_update_used(struct rule
*rule
, long long int used
)
2959 if (used
> rule
->used
) {
2960 struct eviction_group
*evg
= rule
->eviction_group
;
2964 heap_change(&evg
->rules
, &rule
->evg_node
,
2965 rule_eviction_priority(rule
));
2970 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
2971 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
2974 * ofproto implementation ->run() functions should use this function to expire
2975 * OpenFlow flows. */
2977 ofproto_rule_expire(struct rule
*rule
, uint8_t reason
)
2979 struct ofproto
*ofproto
= rule
->ofproto
;
2980 struct ofopgroup
*group
;
2982 assert(reason
== OFPRR_HARD_TIMEOUT
|| reason
== OFPRR_IDLE_TIMEOUT
);
2984 ofproto_rule_send_removed(rule
, reason
);
2986 group
= ofopgroup_create_unattached(ofproto
);
2987 ofoperation_create(group
, rule
, OFOPERATION_DELETE
);
2988 oftable_remove_rule(rule
);
2989 ofproto
->ofproto_class
->rule_destruct(rule
);
2990 ofopgroup_submit(group
);
2994 handle_flow_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2996 struct ofputil_flow_mod fm
;
2999 error
= reject_slave_controller(ofconn
);
3004 error
= ofputil_decode_flow_mod(&fm
, oh
,
3005 ofconn_get_flow_mod_table_id(ofconn
));
3010 /* We do not support the emergency flow cache. It will hopefully get
3011 * dropped from OpenFlow in the near future. */
3012 if (fm
.flags
& OFPFF_EMERG
) {
3013 /* There isn't a good fit for an error code, so just state that the
3014 * flow table is full. */
3015 return OFPERR_OFPFMFC_ALL_TABLES_FULL
;
3018 return handle_flow_mod__(ofconn_get_ofproto(ofconn
), ofconn
, &fm
, oh
);
3022 handle_flow_mod__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3023 const struct ofputil_flow_mod
*fm
,
3024 const struct ofp_header
*oh
)
3026 if (ofproto
->n_pending
>= 50) {
3027 assert(!list_is_empty(&ofproto
->pending
));
3028 return OFPROTO_POSTPONE
;
3031 switch (fm
->command
) {
3033 return add_flow(ofproto
, ofconn
, fm
, oh
);
3036 return modify_flows_loose(ofproto
, ofconn
, fm
, oh
);
3038 case OFPFC_MODIFY_STRICT
:
3039 return modify_flow_strict(ofproto
, ofconn
, fm
, oh
);
3042 return delete_flows_loose(ofproto
, ofconn
, fm
, oh
);
3044 case OFPFC_DELETE_STRICT
:
3045 return delete_flow_strict(ofproto
, ofconn
, fm
, oh
);
3048 if (fm
->command
> 0xff) {
3049 VLOG_WARN_RL(&rl
, "flow_mod has explicit table_id but "
3050 "flow_mod_table_id extension is not enabled");
3052 return OFPERR_OFPFMFC_BAD_COMMAND
;
3057 handle_role_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3059 struct nx_role_request
*nrr
= (struct nx_role_request
*) oh
;
3060 struct nx_role_request
*reply
;
3064 if (ofconn_get_type(ofconn
) != OFCONN_PRIMARY
) {
3065 return OFPERR_OFPBRC_EPERM
;
3068 role
= ntohl(nrr
->role
);
3069 if (role
!= NX_ROLE_OTHER
&& role
!= NX_ROLE_MASTER
3070 && role
!= NX_ROLE_SLAVE
) {
3071 return OFPERR_NXBRC_BAD_ROLE
;
3074 if (ofconn_get_role(ofconn
) != role
3075 && ofconn_has_pending_opgroups(ofconn
)) {
3076 return OFPROTO_POSTPONE
;
3079 ofconn_set_role(ofconn
, role
);
3081 reply
= make_nxmsg_xid(sizeof *reply
, NXT_ROLE_REPLY
, oh
->xid
, &buf
);
3082 reply
->role
= htonl(role
);
3083 ofconn_send_reply(ofconn
, buf
);
3089 handle_nxt_flow_mod_table_id(struct ofconn
*ofconn
,
3090 const struct ofp_header
*oh
)
3092 const struct nx_flow_mod_table_id
*msg
3093 = (const struct nx_flow_mod_table_id
*) oh
;
3095 ofconn_set_flow_mod_table_id(ofconn
, msg
->set
!= 0);
3100 handle_nxt_set_flow_format(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3102 const struct nx_set_flow_format
*msg
3103 = (const struct nx_set_flow_format
*) oh
;
3106 format
= ntohl(msg
->format
);
3107 if (format
!= NXFF_OPENFLOW10
&& format
!= NXFF_NXM
) {
3108 return OFPERR_OFPBRC_EPERM
;
3111 if (format
!= ofconn_get_flow_format(ofconn
)
3112 && ofconn_has_pending_opgroups(ofconn
)) {
3113 /* Avoid sending async messages in surprising flow format. */
3114 return OFPROTO_POSTPONE
;
3117 ofconn_set_flow_format(ofconn
, format
);
3122 handle_nxt_set_packet_in_format(struct ofconn
*ofconn
,
3123 const struct ofp_header
*oh
)
3125 const struct nx_set_packet_in_format
*msg
;
3128 msg
= (const struct nx_set_packet_in_format
*) oh
;
3129 format
= ntohl(msg
->format
);
3130 if (format
!= NXPIF_OPENFLOW10
&& format
!= NXPIF_NXM
) {
3131 return OFPERR_OFPBRC_EPERM
;
3134 if (format
!= ofconn_get_packet_in_format(ofconn
)
3135 && ofconn_has_pending_opgroups(ofconn
)) {
3136 /* Avoid sending async message in surprsing packet in format. */
3137 return OFPROTO_POSTPONE
;
3140 ofconn_set_packet_in_format(ofconn
, format
);
3145 handle_barrier_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3147 struct ofp_header
*ob
;
3150 if (ofconn_has_pending_opgroups(ofconn
)) {
3151 return OFPROTO_POSTPONE
;
3154 ob
= make_openflow_xid(sizeof *ob
, OFPT_BARRIER_REPLY
, oh
->xid
, &buf
);
3155 ofconn_send_reply(ofconn
, buf
);
3160 handle_openflow__(struct ofconn
*ofconn
, const struct ofpbuf
*msg
)
3162 const struct ofp_header
*oh
= msg
->data
;
3163 const struct ofputil_msg_type
*type
;
3166 error
= ofputil_decode_msg_type(oh
, &type
);
3171 switch (ofputil_msg_type_code(type
)) {
3172 /* OpenFlow requests. */
3173 case OFPUTIL_OFPT_ECHO_REQUEST
:
3174 return handle_echo_request(ofconn
, oh
);
3176 case OFPUTIL_OFPT_FEATURES_REQUEST
:
3177 return handle_features_request(ofconn
, oh
);
3179 case OFPUTIL_OFPT_GET_CONFIG_REQUEST
:
3180 return handle_get_config_request(ofconn
, oh
);
3182 case OFPUTIL_OFPT_SET_CONFIG
:
3183 return handle_set_config(ofconn
, msg
->data
);
3185 case OFPUTIL_OFPT_PACKET_OUT
:
3186 return handle_packet_out(ofconn
, oh
);
3188 case OFPUTIL_OFPT_PORT_MOD
:
3189 return handle_port_mod(ofconn
, oh
);
3191 case OFPUTIL_OFPT_FLOW_MOD
:
3192 return handle_flow_mod(ofconn
, oh
);
3194 case OFPUTIL_OFPT_BARRIER_REQUEST
:
3195 return handle_barrier_request(ofconn
, oh
);
3197 /* OpenFlow replies. */
3198 case OFPUTIL_OFPT_ECHO_REPLY
:
3201 /* Nicira extension requests. */
3202 case OFPUTIL_NXT_ROLE_REQUEST
:
3203 return handle_role_request(ofconn
, oh
);
3205 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID
:
3206 return handle_nxt_flow_mod_table_id(ofconn
, oh
);
3208 case OFPUTIL_NXT_SET_FLOW_FORMAT
:
3209 return handle_nxt_set_flow_format(ofconn
, oh
);
3211 case OFPUTIL_NXT_SET_PACKET_IN_FORMAT
:
3212 return handle_nxt_set_packet_in_format(ofconn
, oh
);
3214 case OFPUTIL_NXT_FLOW_MOD
:
3215 return handle_flow_mod(ofconn
, oh
);
3217 case OFPUTIL_NXT_FLOW_AGE
:
3218 /* Nothing to do. */
3221 /* Statistics requests. */
3222 case OFPUTIL_OFPST_DESC_REQUEST
:
3223 return handle_desc_stats_request(ofconn
, msg
->data
);
3225 case OFPUTIL_OFPST_FLOW_REQUEST
:
3226 case OFPUTIL_NXST_FLOW_REQUEST
:
3227 return handle_flow_stats_request(ofconn
, msg
->data
);
3229 case OFPUTIL_OFPST_AGGREGATE_REQUEST
:
3230 case OFPUTIL_NXST_AGGREGATE_REQUEST
:
3231 return handle_aggregate_stats_request(ofconn
, msg
->data
);
3233 case OFPUTIL_OFPST_TABLE_REQUEST
:
3234 return handle_table_stats_request(ofconn
, msg
->data
);
3236 case OFPUTIL_OFPST_PORT_REQUEST
:
3237 return handle_port_stats_request(ofconn
, msg
->data
);
3239 case OFPUTIL_OFPST_QUEUE_REQUEST
:
3240 return handle_queue_stats_request(ofconn
, msg
->data
);
3242 case OFPUTIL_MSG_INVALID
:
3243 case OFPUTIL_OFPT_HELLO
:
3244 case OFPUTIL_OFPT_ERROR
:
3245 case OFPUTIL_OFPT_FEATURES_REPLY
:
3246 case OFPUTIL_OFPT_GET_CONFIG_REPLY
:
3247 case OFPUTIL_OFPT_PACKET_IN
:
3248 case OFPUTIL_OFPT_FLOW_REMOVED
:
3249 case OFPUTIL_OFPT_PORT_STATUS
:
3250 case OFPUTIL_OFPT_BARRIER_REPLY
:
3251 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST
:
3252 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY
:
3253 case OFPUTIL_OFPST_DESC_REPLY
:
3254 case OFPUTIL_OFPST_FLOW_REPLY
:
3255 case OFPUTIL_OFPST_QUEUE_REPLY
:
3256 case OFPUTIL_OFPST_PORT_REPLY
:
3257 case OFPUTIL_OFPST_TABLE_REPLY
:
3258 case OFPUTIL_OFPST_AGGREGATE_REPLY
:
3259 case OFPUTIL_NXT_ROLE_REPLY
:
3260 case OFPUTIL_NXT_FLOW_REMOVED
:
3261 case OFPUTIL_NXT_PACKET_IN
:
3262 case OFPUTIL_NXST_FLOW_REPLY
:
3263 case OFPUTIL_NXST_AGGREGATE_REPLY
:
3265 if (oh
->type
== OFPT_STATS_REQUEST
|| oh
->type
== OFPT_STATS_REPLY
) {
3266 return OFPERR_OFPBRC_BAD_STAT
;
3268 return OFPERR_OFPBRC_BAD_TYPE
;
3274 handle_openflow(struct ofconn
*ofconn
, struct ofpbuf
*ofp_msg
)
3276 int error
= handle_openflow__(ofconn
, ofp_msg
);
3277 if (error
&& error
!= OFPROTO_POSTPONE
) {
3278 ofconn_send_error(ofconn
, ofp_msg
->data
, error
);
3280 COVERAGE_INC(ofproto_recv_openflow
);
3281 return error
!= OFPROTO_POSTPONE
;
3284 /* Asynchronous operations. */
3286 /* Creates and returns a new ofopgroup that is not associated with any
3287 * OpenFlow connection.
3289 * The caller should add operations to the returned group with
3290 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3291 static struct ofopgroup
*
3292 ofopgroup_create_unattached(struct ofproto
*ofproto
)
3294 struct ofopgroup
*group
= xzalloc(sizeof *group
);
3295 group
->ofproto
= ofproto
;
3296 list_init(&group
->ofproto_node
);
3297 list_init(&group
->ops
);
3298 list_init(&group
->ofconn_node
);
3302 /* Creates and returns a new ofopgroup for 'ofproto'.
3304 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
3305 * connection. The 'request' and 'buffer_id' arguments are ignored.
3307 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
3308 * If the ofopgroup eventually fails, then the error reply will include
3309 * 'request'. If the ofopgroup eventually succeeds, then the packet with
3310 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
3312 * The caller should add operations to the returned group with
3313 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3314 static struct ofopgroup
*
3315 ofopgroup_create(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3316 const struct ofp_header
*request
, uint32_t buffer_id
)
3318 struct ofopgroup
*group
= ofopgroup_create_unattached(ofproto
);
3320 size_t request_len
= ntohs(request
->length
);
3322 assert(ofconn_get_ofproto(ofconn
) == ofproto
);
3324 ofconn_add_opgroup(ofconn
, &group
->ofconn_node
);
3325 group
->ofconn
= ofconn
;
3326 group
->request
= xmemdup(request
, MIN(request_len
, 64));
3327 group
->buffer_id
= buffer_id
;
3332 /* Submits 'group' for processing.
3334 * If 'group' contains no operations (e.g. none were ever added, or all of the
3335 * ones that were added completed synchronously), then it is destroyed
3336 * immediately. Otherwise it is added to the ofproto's list of pending
3339 ofopgroup_submit(struct ofopgroup
*group
)
3341 if (list_is_empty(&group
->ops
)) {
3342 ofopgroup_destroy(group
);
3344 list_push_back(&group
->ofproto
->pending
, &group
->ofproto_node
);
3345 group
->ofproto
->n_pending
++;
3350 ofopgroup_destroy(struct ofopgroup
*group
)
3352 assert(list_is_empty(&group
->ops
));
3353 if (!list_is_empty(&group
->ofproto_node
)) {
3354 assert(group
->ofproto
->n_pending
> 0);
3355 group
->ofproto
->n_pending
--;
3356 list_remove(&group
->ofproto_node
);
3358 if (!list_is_empty(&group
->ofconn_node
)) {
3359 list_remove(&group
->ofconn_node
);
3361 ofconn_send_error(group
->ofconn
, group
->request
, group
->error
);
3363 connmgr_retry(group
->ofproto
->connmgr
);
3365 free(group
->request
);
3369 /* Initiates a new operation on 'rule', of the specified 'type', within
3370 * 'group'. Prior to calling, 'rule' must not have any pending operation. */
3372 ofoperation_create(struct ofopgroup
*group
, struct rule
*rule
,
3373 enum ofoperation_type type
)
3375 struct ofoperation
*op
;
3377 assert(!rule
->pending
);
3379 op
= rule
->pending
= xzalloc(sizeof *op
);
3381 list_push_back(&group
->ops
, &op
->group_node
);
3385 op
->flow_cookie
= rule
->flow_cookie
;
3387 if (type
== OFOPERATION_DELETE
) {
3388 hmap_insert(&op
->group
->ofproto
->deletions
, &op
->hmap_node
,
3389 cls_rule_hash(&rule
->cr
, rule
->table_id
));
3394 ofoperation_destroy(struct ofoperation
*op
)
3396 struct ofopgroup
*group
= op
->group
;
3399 op
->rule
->pending
= NULL
;
3401 if (op
->type
== OFOPERATION_DELETE
) {
3402 hmap_remove(&group
->ofproto
->deletions
, &op
->hmap_node
);
3404 list_remove(&op
->group_node
);
3408 if (list_is_empty(&group
->ops
) && !list_is_empty(&group
->ofproto_node
)) {
3409 ofopgroup_destroy(group
);
3413 /* Indicates that 'op' completed with status 'error', which is either 0 to
3414 * indicate success or an OpenFlow error code on failure.
3416 * If 'error' is 0, indicating success, the operation will be committed
3417 * permanently to the flow table. There is one interesting subcase:
3419 * - If 'op' is an "add flow" operation that is replacing an existing rule in
3420 * the flow table (the "victim" rule) by a new one, then the caller must
3421 * have uninitialized any derived state in the victim rule, as in step 5 in
3422 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
3423 * performs steps 6 and 7 for the victim rule, most notably by calling its
3424 * ->rule_dealloc() function.
3426 * If 'error' is nonzero, then generally the operation will be rolled back:
3428 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
3429 * restores the original rule. The caller must have uninitialized any
3430 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
3431 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
3432 * and 7 for the new rule, calling its ->rule_dealloc() function.
3434 * - If 'op' is a "modify flow" operation, ofproto restores the original
3437 * - 'op' must not be a "delete flow" operation. Removing a rule is not
3438 * allowed to fail. It must always succeed.
3440 * Please see the large comment in ofproto/ofproto-provider.h titled
3441 * "Asynchronous Operation Support" for more information. */
3443 ofoperation_complete(struct ofoperation
*op
, enum ofperr error
)
3445 struct ofopgroup
*group
= op
->group
;
3446 struct rule
*rule
= op
->rule
;
3447 struct ofproto
*ofproto
= rule
->ofproto
;
3449 assert(rule
->pending
== op
);
3450 assert(op
->status
< 0);
3454 && op
->type
!= OFOPERATION_DELETE
3456 && group
->buffer_id
!= UINT32_MAX
3457 && list_is_singleton(&op
->group_node
)) {
3458 struct ofpbuf
*packet
;
3461 error
= ofconn_pktbuf_retrieve(group
->ofconn
, group
->buffer_id
,
3465 error
= rule_execute(rule
, in_port
, packet
);
3468 if (!group
->error
) {
3469 group
->error
= error
;
3473 case OFOPERATION_ADD
:
3475 ofproto_rule_destroy__(op
->victim
);
3476 if ((rule
->cr
.wc
.vlan_tci_mask
& htons(VLAN_VID_MASK
))
3477 == htons(VLAN_VID_MASK
)) {
3478 if (ofproto
->vlan_bitmap
) {
3479 uint16_t vid
= vlan_tci_to_vid(rule
->cr
.flow
.vlan_tci
);
3481 if (!bitmap_is_set(ofproto
->vlan_bitmap
, vid
)) {
3482 bitmap_set1(ofproto
->vlan_bitmap
, vid
);
3483 ofproto
->vlans_changed
= true;
3486 ofproto
->vlans_changed
= true;
3490 oftable_substitute_rule(rule
, op
->victim
);
3491 ofproto_rule_destroy__(rule
);
3495 case OFOPERATION_DELETE
:
3497 ofproto_rule_destroy__(rule
);
3501 case OFOPERATION_MODIFY
:
3503 rule
->modified
= time_msec();
3505 free(rule
->actions
);
3506 rule
->actions
= op
->actions
;
3507 rule
->n_actions
= op
->n_actions
;
3515 ofoperation_destroy(op
);
3519 ofoperation_get_victim(struct ofoperation
*op
)
3521 assert(op
->type
== OFOPERATION_ADD
);
3526 pick_datapath_id(const struct ofproto
*ofproto
)
3528 const struct ofport
*port
;
3530 port
= ofproto_get_port(ofproto
, OFPP_LOCAL
);
3532 uint8_t ea
[ETH_ADDR_LEN
];
3535 error
= netdev_get_etheraddr(port
->netdev
, ea
);
3537 return eth_addr_to_uint64(ea
);
3539 VLOG_WARN("could not get MAC address for %s (%s)",
3540 netdev_get_name(port
->netdev
), strerror(error
));
3542 return ofproto
->fallback_dpid
;
3546 pick_fallback_dpid(void)
3548 uint8_t ea
[ETH_ADDR_LEN
];
3549 eth_addr_nicira_random(ea
);
3550 return eth_addr_to_uint64(ea
);
3553 /* Table overflow policy. */
3555 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
3556 * is not configured to evict rules or if the table contains no evictable
3557 * rules. (Rules with 'evictable' set to false or with no timeouts are not
3559 static struct rule
*
3560 choose_rule_to_evict(struct oftable
*table
)
3562 struct eviction_group
*evg
;
3564 if (!table
->eviction_fields
) {
3568 /* In the common case, the outer and inner loops here will each be entered
3571 * - The inner loop normally "return"s in its first iteration. If the
3572 * eviction group has any evictable rules, then it always returns in
3575 * - The outer loop only iterates more than once if the largest eviction
3576 * group has no evictable rules.
3578 * - The outer loop can exit only if table's 'max_flows' is all filled up
3579 * by unevictable rules'. */
3580 HEAP_FOR_EACH (evg
, size_node
, &table
->eviction_groups_by_size
) {
3583 HEAP_FOR_EACH (rule
, evg_node
, &evg
->rules
) {
3584 if (rule
->evictable
) {
3593 /* Searches 'ofproto' for tables that have more flows than their configured
3594 * maximum and that have flow eviction enabled, and evicts as many flows as
3595 * necessary and currently feasible from them.
3597 * This triggers only when an OpenFlow table has N flows in it and then the
3598 * client configures a maximum number of flows less than N. */
3600 ofproto_evict(struct ofproto
*ofproto
)
3602 struct ofopgroup
*group
;
3603 struct oftable
*table
;
3605 group
= ofopgroup_create_unattached(ofproto
);
3606 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
3607 while (classifier_count(&table
->cls
) > table
->max_flows
3608 && table
->eviction_fields
) {
3611 rule
= choose_rule_to_evict(table
);
3612 if (!rule
|| rule
->pending
) {
3616 ofoperation_create(group
, rule
, OFOPERATION_DELETE
);
3617 oftable_remove_rule(rule
);
3618 ofproto
->ofproto_class
->rule_destruct(rule
);
3621 ofopgroup_submit(group
);
3624 /* Eviction groups. */
3626 /* Returns the priority to use for an eviction_group that contains 'n_rules'
3627 * rules. The priority contains low-order random bits to ensure that eviction
3628 * groups with the same number of rules are prioritized randomly. */
3630 eviction_group_priority(size_t n_rules
)
3632 uint16_t size
= MIN(UINT16_MAX
, n_rules
);
3633 return (size
<< 16) | random_uint16();
3636 /* Updates 'evg', an eviction_group within 'table', following a change that
3637 * adds or removes rules in 'evg'. */
3639 eviction_group_resized(struct oftable
*table
, struct eviction_group
*evg
)
3641 heap_change(&table
->eviction_groups_by_size
, &evg
->size_node
,
3642 eviction_group_priority(heap_count(&evg
->rules
)));
3645 /* Destroys 'evg', an eviction_group within 'table':
3647 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
3648 * rules themselves, just removes them from the eviction group.)
3650 * - Removes 'evg' from 'table'.
3654 eviction_group_destroy(struct oftable
*table
, struct eviction_group
*evg
)
3656 while (!heap_is_empty(&evg
->rules
)) {
3659 rule
= CONTAINER_OF(heap_pop(&evg
->rules
), struct rule
, evg_node
);
3660 rule
->eviction_group
= NULL
;
3662 hmap_remove(&table
->eviction_groups_by_id
, &evg
->id_node
);
3663 heap_remove(&table
->eviction_groups_by_size
, &evg
->size_node
);
3664 heap_destroy(&evg
->rules
);
3668 /* Removes 'rule' from its eviction group, if any. */
3670 eviction_group_remove_rule(struct rule
*rule
)
3672 if (rule
->eviction_group
) {
3673 struct oftable
*table
= &rule
->ofproto
->tables
[rule
->table_id
];
3674 struct eviction_group
*evg
= rule
->eviction_group
;
3676 rule
->eviction_group
= NULL
;
3677 heap_remove(&evg
->rules
, &rule
->evg_node
);
3678 if (heap_is_empty(&evg
->rules
)) {
3679 eviction_group_destroy(table
, evg
);
3681 eviction_group_resized(table
, evg
);
3686 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
3687 * returns the hash value. */
3689 eviction_group_hash_rule(struct rule
*rule
)
3691 struct oftable
*table
= &rule
->ofproto
->tables
[rule
->table_id
];
3692 const struct mf_subfield
*sf
;
3695 hash
= table
->eviction_group_id_basis
;
3696 for (sf
= table
->eviction_fields
;
3697 sf
< &table
->eviction_fields
[table
->n_eviction_fields
];
3700 if (mf_are_prereqs_ok(sf
->field
, &rule
->cr
.flow
)) {
3701 union mf_value value
;
3703 mf_get_value(sf
->field
, &rule
->cr
.flow
, &value
);
3705 bitwise_zero(&value
, sf
->field
->n_bytes
, 0, sf
->ofs
);
3707 if (sf
->ofs
+ sf
->n_bits
< sf
->field
->n_bytes
* 8) {
3708 unsigned int start
= sf
->ofs
+ sf
->n_bits
;
3709 bitwise_zero(&value
, sf
->field
->n_bytes
, start
,
3710 sf
->field
->n_bytes
* 8 - start
);
3712 hash
= hash_bytes(&value
, sf
->field
->n_bytes
, hash
);
3714 hash
= hash_int(hash
, 0);
3721 /* Returns an eviction group within 'table' with the given 'id', creating one
3723 static struct eviction_group
*
3724 eviction_group_find(struct oftable
*table
, uint32_t id
)
3726 struct eviction_group
*evg
;
3728 HMAP_FOR_EACH_WITH_HASH (evg
, id_node
, id
, &table
->eviction_groups_by_id
) {
3732 evg
= xmalloc(sizeof *evg
);
3733 hmap_insert(&table
->eviction_groups_by_id
, &evg
->id_node
, id
);
3734 heap_insert(&table
->eviction_groups_by_size
, &evg
->size_node
,
3735 eviction_group_priority(0));
3736 heap_init(&evg
->rules
);
3741 /* Returns an eviction priority for 'rule'. The return value should be
3742 * interpreted so that higher priorities make a rule more attractive candidates
3745 rule_eviction_priority(struct rule
*rule
)
3747 long long int hard_expiration
;
3748 long long int idle_expiration
;
3749 long long int expiration
;
3750 uint32_t expiration_offset
;
3752 /* Calculate time of expiration. */
3753 hard_expiration
= (rule
->hard_timeout
3754 ? rule
->modified
+ rule
->hard_timeout
* 1000
3756 idle_expiration
= (rule
->idle_timeout
3757 ? rule
->used
+ rule
->idle_timeout
* 1000
3759 expiration
= MIN(hard_expiration
, idle_expiration
);
3760 if (expiration
== LLONG_MAX
) {
3764 /* Calculate the time of expiration as a number of (approximate) seconds
3765 * after program startup.
3767 * This should work OK for program runs that last UINT32_MAX seconds or
3768 * less. Therefore, please restart OVS at least once every 136 years. */
3769 expiration_offset
= (expiration
>> 10) - (time_boot_msec() >> 10);
3771 /* Invert the expiration offset because we're using a max-heap. */
3772 return UINT32_MAX
- expiration_offset
;
3775 /* Adds 'rule' to an appropriate eviction group for its oftable's
3776 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
3777 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
3780 * The caller must ensure that 'rule' is not already in an eviction group. */
3782 eviction_group_add_rule(struct rule
*rule
)
3784 struct ofproto
*ofproto
= rule
->ofproto
;
3785 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
3787 if (table
->eviction_fields
3788 && (rule
->hard_timeout
|| rule
->idle_timeout
)) {
3789 struct eviction_group
*evg
;
3791 evg
= eviction_group_find(table
, eviction_group_hash_rule(rule
));
3793 rule
->eviction_group
= evg
;
3794 heap_insert(&evg
->rules
, &rule
->evg_node
,
3795 rule_eviction_priority(rule
));
3796 eviction_group_resized(table
, evg
);
3802 /* Initializes 'table'. */
3804 oftable_init(struct oftable
*table
)
3806 memset(table
, 0, sizeof *table
);
3807 classifier_init(&table
->cls
);
3810 /* Destroys 'table', including its classifier and eviction groups.
3812 * The caller is responsible for freeing 'table' itself. */
3814 oftable_destroy(struct oftable
*table
)
3816 assert(classifier_is_empty(&table
->cls
));
3817 oftable_disable_eviction(table
);
3818 classifier_destroy(&table
->cls
);
3822 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
3823 * string, then 'table' will use its default name.
3825 * This only affects the name exposed for a table exposed through the OpenFlow
3826 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
3828 oftable_set_name(struct oftable
*table
, const char *name
)
3830 if (name
&& name
[0]) {
3831 int len
= strnlen(name
, OFP_MAX_TABLE_NAME_LEN
);
3832 if (!table
->name
|| strncmp(name
, table
->name
, len
)) {
3834 table
->name
= xmemdup0(name
, len
);
3842 /* oftables support a choice of two policies when adding a rule would cause the
3843 * number of flows in the table to exceed the configured maximum number: either
3844 * they can refuse to add the new flow or they can evict some existing flow.
3845 * This function configures the former policy on 'table'. */
3847 oftable_disable_eviction(struct oftable
*table
)
3849 if (table
->eviction_fields
) {
3850 struct eviction_group
*evg
, *next
;
3852 HMAP_FOR_EACH_SAFE (evg
, next
, id_node
,
3853 &table
->eviction_groups_by_id
) {
3854 eviction_group_destroy(table
, evg
);
3856 hmap_destroy(&table
->eviction_groups_by_id
);
3857 heap_destroy(&table
->eviction_groups_by_size
);
3859 free(table
->eviction_fields
);
3860 table
->eviction_fields
= NULL
;
3861 table
->n_eviction_fields
= 0;
3865 /* oftables support a choice of two policies when adding a rule would cause the
3866 * number of flows in the table to exceed the configured maximum number: either
3867 * they can refuse to add the new flow or they can evict some existing flow.
3868 * This function configures the latter policy on 'table', with fairness based
3869 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
3870 * 'n_fields' as 0 disables fairness.) */
3872 oftable_enable_eviction(struct oftable
*table
,
3873 const struct mf_subfield
*fields
, size_t n_fields
)
3875 struct cls_cursor cursor
;
3878 if (table
->eviction_fields
3879 && n_fields
== table
->n_eviction_fields
3881 || !memcmp(fields
, table
->eviction_fields
,
3882 n_fields
* sizeof *fields
))) {
3887 oftable_disable_eviction(table
);
3889 table
->n_eviction_fields
= n_fields
;
3890 table
->eviction_fields
= xmemdup(fields
, n_fields
* sizeof *fields
);
3892 table
->eviction_group_id_basis
= random_uint32();
3893 hmap_init(&table
->eviction_groups_by_id
);
3894 heap_init(&table
->eviction_groups_by_size
);
3896 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
3897 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3898 eviction_group_add_rule(rule
);
3902 /* Removes 'rule' from the oftable that contains it. */
3904 oftable_remove_rule(struct rule
*rule
)
3906 struct ofproto
*ofproto
= rule
->ofproto
;
3907 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
3909 classifier_remove(&table
->cls
, &rule
->cr
);
3910 eviction_group_remove_rule(rule
);
3913 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
3914 * oftable that has an identical cls_rule. Returns the rule that was removed,
3915 * if any, and otherwise NULL. */
3916 static struct rule
*
3917 oftable_replace_rule(struct rule
*rule
)
3919 struct ofproto
*ofproto
= rule
->ofproto
;
3920 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
3921 struct rule
*victim
;
3923 victim
= rule_from_cls_rule(classifier_replace(&table
->cls
, &rule
->cr
));
3925 eviction_group_remove_rule(victim
);
3927 eviction_group_add_rule(rule
);
3931 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
3933 oftable_substitute_rule(struct rule
*old
, struct rule
*new)
3936 oftable_replace_rule(new);
3938 oftable_remove_rule(old
);
3942 /* unixctl commands. */
3945 ofproto_lookup(const char *name
)
3947 struct ofproto
*ofproto
;
3949 HMAP_FOR_EACH_WITH_HASH (ofproto
, hmap_node
, hash_string(name
, 0),
3951 if (!strcmp(ofproto
->name
, name
)) {
3959 ofproto_unixctl_list(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
3960 const char *argv
[] OVS_UNUSED
, void *aux OVS_UNUSED
)
3962 struct ofproto
*ofproto
;
3966 HMAP_FOR_EACH (ofproto
, hmap_node
, &all_ofprotos
) {
3967 ds_put_format(&results
, "%s\n", ofproto
->name
);
3969 unixctl_command_reply(conn
, 200, ds_cstr(&results
));
3970 ds_destroy(&results
);
3974 ofproto_unixctl_init(void)
3976 static bool registered
;
3982 unixctl_command_register("ofproto/list", "", 0, 0,
3983 ofproto_unixctl_list
, NULL
);
3986 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
3988 * This is deprecated. It is only for compatibility with broken device drivers
3989 * in old versions of Linux that do not properly support VLANs when VLAN
3990 * devices are not used. When broken device drivers are no longer in
3991 * widespread use, we will delete these interfaces. */
3993 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
3994 * (exactly) by an OpenFlow rule in 'ofproto'. */
3996 ofproto_get_vlan_usage(struct ofproto
*ofproto
, unsigned long int *vlan_bitmap
)
3998 const struct oftable
*oftable
;
4000 free(ofproto
->vlan_bitmap
);
4001 ofproto
->vlan_bitmap
= bitmap_allocate(4096);
4002 ofproto
->vlans_changed
= false;
4004 OFPROTO_FOR_EACH_TABLE (oftable
, ofproto
) {
4005 const struct cls_table
*table
;
4007 HMAP_FOR_EACH (table
, hmap_node
, &oftable
->cls
.tables
) {
4008 if ((table
->wc
.vlan_tci_mask
& htons(VLAN_VID_MASK
))
4009 == htons(VLAN_VID_MASK
)) {
4010 const struct cls_rule
*rule
;
4012 HMAP_FOR_EACH (rule
, hmap_node
, &table
->rules
) {
4013 uint16_t vid
= vlan_tci_to_vid(rule
->flow
.vlan_tci
);
4014 bitmap_set1(vlan_bitmap
, vid
);
4015 bitmap_set1(ofproto
->vlan_bitmap
, vid
);
4022 /* Returns true if new VLANs have come into use by the flow table since the
4023 * last call to ofproto_get_vlan_usage().
4025 * We don't track when old VLANs stop being used. */
4027 ofproto_has_vlan_usage_changed(const struct ofproto
*ofproto
)
4029 return ofproto
->vlans_changed
;
4032 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4033 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4034 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4035 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4036 * then the VLAN device is un-enslaved. */
4038 ofproto_port_set_realdev(struct ofproto
*ofproto
, uint16_t vlandev_ofp_port
,
4039 uint16_t realdev_ofp_port
, int vid
)
4041 struct ofport
*ofport
;
4044 assert(vlandev_ofp_port
!= realdev_ofp_port
);
4046 ofport
= ofproto_get_port(ofproto
, vlandev_ofp_port
);
4048 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16
,
4049 ofproto
->name
, vlandev_ofp_port
);
4053 if (!ofproto
->ofproto_class
->set_realdev
) {
4054 if (!vlandev_ofp_port
) {
4057 VLOG_WARN("%s: vlan splinters not supported", ofproto
->name
);
4061 error
= ofproto
->ofproto_class
->set_realdev(ofport
, realdev_ofp_port
, vid
);
4063 VLOG_WARN("%s: setting realdev on port %"PRIu16
" (%s) failed (%s)",
4064 ofproto
->name
, vlandev_ofp_port
,
4065 netdev_get_name(ofport
->netdev
), strerror(error
));