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 update_mtu(struct ofproto
*, struct ofport
*);
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;
390 ofproto
->min_mtu
= INT_MAX
;
392 error
= ofproto
->ofproto_class
->construct(ofproto
);
394 VLOG_ERR("failed to open datapath %s: %s",
395 datapath_name
, strerror(error
));
396 ofproto_destroy__(ofproto
);
400 assert(ofproto
->n_tables
);
402 ofproto
->datapath_id
= pick_datapath_id(ofproto
);
403 VLOG_INFO("using datapath ID %016"PRIx64
, ofproto
->datapath_id
);
411 ofproto_init_tables(struct ofproto
*ofproto
, int n_tables
)
413 struct oftable
*table
;
415 assert(!ofproto
->n_tables
);
416 assert(n_tables
>= 1 && n_tables
<= 255);
418 ofproto
->n_tables
= n_tables
;
419 ofproto
->tables
= xmalloc(n_tables
* sizeof *ofproto
->tables
);
420 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
426 ofproto_set_datapath_id(struct ofproto
*p
, uint64_t datapath_id
)
428 uint64_t old_dpid
= p
->datapath_id
;
429 p
->datapath_id
= datapath_id
? datapath_id
: pick_datapath_id(p
);
430 if (p
->datapath_id
!= old_dpid
) {
431 VLOG_INFO("datapath ID changed to %016"PRIx64
, p
->datapath_id
);
433 /* Force all active connections to reconnect, since there is no way to
434 * notify a controller that the datapath ID has changed. */
435 ofproto_reconnect_controllers(p
);
440 ofproto_set_controllers(struct ofproto
*p
,
441 const struct ofproto_controller
*controllers
,
442 size_t n_controllers
)
444 connmgr_set_controllers(p
->connmgr
, controllers
, n_controllers
);
448 ofproto_set_fail_mode(struct ofproto
*p
, enum ofproto_fail_mode fail_mode
)
450 connmgr_set_fail_mode(p
->connmgr
, fail_mode
);
453 /* Drops the connections between 'ofproto' and all of its controllers, forcing
454 * them to reconnect. */
456 ofproto_reconnect_controllers(struct ofproto
*ofproto
)
458 connmgr_reconnect(ofproto
->connmgr
);
461 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
462 * in-band control should guarantee access, in the same way that in-band
463 * control guarantees access to OpenFlow controllers. */
465 ofproto_set_extra_in_band_remotes(struct ofproto
*ofproto
,
466 const struct sockaddr_in
*extras
, size_t n
)
468 connmgr_set_extra_in_band_remotes(ofproto
->connmgr
, extras
, n
);
471 /* Sets the OpenFlow queue used by flows set up by in-band control on
472 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
473 * flows will use the default queue. */
475 ofproto_set_in_band_queue(struct ofproto
*ofproto
, int queue_id
)
477 connmgr_set_in_band_queue(ofproto
->connmgr
, queue_id
);
480 /* Sets the number of flows at which eviction from the kernel flow table
483 ofproto_set_flow_eviction_threshold(struct ofproto
*ofproto
, unsigned threshold
)
485 if (threshold
< OFPROTO_FLOW_EVICTION_THRESHOLD_MIN
) {
486 ofproto
->flow_eviction_threshold
= OFPROTO_FLOW_EVICTION_THRESHOLD_MIN
;
488 ofproto
->flow_eviction_threshold
= threshold
;
492 /* If forward_bpdu is true, the NORMAL action will forward frames with
493 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
494 * the NORMAL action will drop these frames. */
496 ofproto_set_forward_bpdu(struct ofproto
*ofproto
, bool forward_bpdu
)
498 bool old_val
= ofproto
->forward_bpdu
;
499 ofproto
->forward_bpdu
= forward_bpdu
;
500 if (old_val
!= ofproto
->forward_bpdu
) {
501 if (ofproto
->ofproto_class
->forward_bpdu_changed
) {
502 ofproto
->ofproto_class
->forward_bpdu_changed(ofproto
);
507 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
508 * 'idle_time', in seconds. */
510 ofproto_set_mac_idle_time(struct ofproto
*ofproto
, unsigned idle_time
)
512 if (ofproto
->ofproto_class
->set_mac_idle_time
) {
513 ofproto
->ofproto_class
->set_mac_idle_time(ofproto
, idle_time
);
518 ofproto_set_desc(struct ofproto
*p
,
519 const char *mfr_desc
, const char *hw_desc
,
520 const char *sw_desc
, const char *serial_desc
,
523 struct ofp_desc_stats
*ods
;
526 if (strlen(mfr_desc
) >= sizeof ods
->mfr_desc
) {
527 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
528 sizeof ods
->mfr_desc
);
531 p
->mfr_desc
= xstrdup(mfr_desc
);
534 if (strlen(hw_desc
) >= sizeof ods
->hw_desc
) {
535 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
536 sizeof ods
->hw_desc
);
539 p
->hw_desc
= xstrdup(hw_desc
);
542 if (strlen(sw_desc
) >= sizeof ods
->sw_desc
) {
543 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
544 sizeof ods
->sw_desc
);
547 p
->sw_desc
= xstrdup(sw_desc
);
550 if (strlen(serial_desc
) >= sizeof ods
->serial_num
) {
551 VLOG_WARN("truncating serial_desc, must be less than %zu "
553 sizeof ods
->serial_num
);
555 free(p
->serial_desc
);
556 p
->serial_desc
= xstrdup(serial_desc
);
559 if (strlen(dp_desc
) >= sizeof ods
->dp_desc
) {
560 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
561 sizeof ods
->dp_desc
);
564 p
->dp_desc
= xstrdup(dp_desc
);
569 ofproto_set_snoops(struct ofproto
*ofproto
, const struct sset
*snoops
)
571 return connmgr_set_snoops(ofproto
->connmgr
, snoops
);
575 ofproto_set_netflow(struct ofproto
*ofproto
,
576 const struct netflow_options
*nf_options
)
578 if (nf_options
&& sset_is_empty(&nf_options
->collectors
)) {
582 if (ofproto
->ofproto_class
->set_netflow
) {
583 return ofproto
->ofproto_class
->set_netflow(ofproto
, nf_options
);
585 return nf_options
? EOPNOTSUPP
: 0;
590 ofproto_set_sflow(struct ofproto
*ofproto
,
591 const struct ofproto_sflow_options
*oso
)
593 if (oso
&& sset_is_empty(&oso
->targets
)) {
597 if (ofproto
->ofproto_class
->set_sflow
) {
598 return ofproto
->ofproto_class
->set_sflow(ofproto
, oso
);
600 return oso
? EOPNOTSUPP
: 0;
604 /* Spanning Tree Protocol (STP) configuration. */
606 /* Configures STP on 'ofproto' using the settings defined in 's'. If
607 * 's' is NULL, disables STP.
609 * Returns 0 if successful, otherwise a positive errno value. */
611 ofproto_set_stp(struct ofproto
*ofproto
,
612 const struct ofproto_stp_settings
*s
)
614 return (ofproto
->ofproto_class
->set_stp
615 ? ofproto
->ofproto_class
->set_stp(ofproto
, s
)
619 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
620 * 'enabled' member of 's' is false, then the other members are not
623 * Returns 0 if successful, otherwise a positive errno value. */
625 ofproto_get_stp_status(struct ofproto
*ofproto
,
626 struct ofproto_stp_status
*s
)
628 return (ofproto
->ofproto_class
->get_stp_status
629 ? ofproto
->ofproto_class
->get_stp_status(ofproto
, s
)
633 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
634 * in 's'. The caller is responsible for assigning STP port numbers
635 * (using the 'port_num' member in the range of 1 through 255, inclusive)
636 * and ensuring there are no duplicates. If the 's' is NULL, then STP
637 * is disabled on the port.
639 * Returns 0 if successful, otherwise a positive errno value.*/
641 ofproto_port_set_stp(struct ofproto
*ofproto
, uint16_t ofp_port
,
642 const struct ofproto_port_stp_settings
*s
)
644 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
646 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16
,
647 ofproto
->name
, ofp_port
);
651 return (ofproto
->ofproto_class
->set_stp_port
652 ? ofproto
->ofproto_class
->set_stp_port(ofport
, s
)
656 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
657 * 's'. If the 'enabled' member in 's' is false, then the other members
658 * are not meaningful.
660 * Returns 0 if successful, otherwise a positive errno value.*/
662 ofproto_port_get_stp_status(struct ofproto
*ofproto
, uint16_t ofp_port
,
663 struct ofproto_port_stp_status
*s
)
665 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
667 VLOG_WARN("%s: cannot get STP status on nonexistent port %"PRIu16
,
668 ofproto
->name
, ofp_port
);
672 return (ofproto
->ofproto_class
->get_stp_port_status
673 ? ofproto
->ofproto_class
->get_stp_port_status(ofport
, s
)
677 /* Queue DSCP configuration. */
679 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
680 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
681 * to implement QoS. Instead, it is used to implement features which require
682 * knowledge of what queues exist on a port, and some basic information about
685 * Returns 0 if successful, otherwise a positive errno value. */
687 ofproto_port_set_queues(struct ofproto
*ofproto
, uint16_t ofp_port
,
688 const struct ofproto_port_queue
*queues
,
691 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
694 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16
,
695 ofproto
->name
, ofp_port
);
699 return (ofproto
->ofproto_class
->set_queues
700 ? ofproto
->ofproto_class
->set_queues(ofport
, queues
, n_queues
)
704 /* Connectivity Fault Management configuration. */
706 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
708 ofproto_port_clear_cfm(struct ofproto
*ofproto
, uint16_t ofp_port
)
710 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
711 if (ofport
&& ofproto
->ofproto_class
->set_cfm
) {
712 ofproto
->ofproto_class
->set_cfm(ofport
, NULL
);
716 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
717 * basic configuration from the configuration members in 'cfm', and the remote
718 * maintenance point ID from remote_mpid. Ignores the statistics members of
721 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
723 ofproto_port_set_cfm(struct ofproto
*ofproto
, uint16_t ofp_port
,
724 const struct cfm_settings
*s
)
726 struct ofport
*ofport
;
729 ofport
= ofproto_get_port(ofproto
, ofp_port
);
731 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16
,
732 ofproto
->name
, ofp_port
);
736 /* XXX: For configuration simplicity, we only support one remote_mpid
737 * outside of the CFM module. It's not clear if this is the correct long
738 * term solution or not. */
739 error
= (ofproto
->ofproto_class
->set_cfm
740 ? ofproto
->ofproto_class
->set_cfm(ofport
, s
)
743 VLOG_WARN("%s: CFM configuration on port %"PRIu16
" (%s) failed (%s)",
744 ofproto
->name
, ofp_port
, netdev_get_name(ofport
->netdev
),
749 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
750 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
751 * 0 if LACP partner information is not current (generally indicating a
752 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
754 ofproto_port_is_lacp_current(struct ofproto
*ofproto
, uint16_t ofp_port
)
756 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
757 return (ofport
&& ofproto
->ofproto_class
->port_is_lacp_current
758 ? ofproto
->ofproto_class
->port_is_lacp_current(ofport
)
764 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
765 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
766 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
767 * configuration plus, if there is more than one slave, a bonding
770 * If 'aux' is already registered then this function updates its configuration
771 * to 's'. Otherwise, this function registers a new bundle.
773 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
776 ofproto_bundle_register(struct ofproto
*ofproto
, void *aux
,
777 const struct ofproto_bundle_settings
*s
)
779 return (ofproto
->ofproto_class
->bundle_set
780 ? ofproto
->ofproto_class
->bundle_set(ofproto
, aux
, s
)
784 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
785 * If no such bundle has been registered, this has no effect. */
787 ofproto_bundle_unregister(struct ofproto
*ofproto
, void *aux
)
789 return ofproto_bundle_register(ofproto
, aux
, NULL
);
793 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
794 * If 'aux' is already registered then this function updates its configuration
795 * to 's'. Otherwise, this function registers a new mirror. */
797 ofproto_mirror_register(struct ofproto
*ofproto
, void *aux
,
798 const struct ofproto_mirror_settings
*s
)
800 return (ofproto
->ofproto_class
->mirror_set
801 ? ofproto
->ofproto_class
->mirror_set(ofproto
, aux
, s
)
805 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
806 * If no mirror has been registered, this has no effect. */
808 ofproto_mirror_unregister(struct ofproto
*ofproto
, void *aux
)
810 return ofproto_mirror_register(ofproto
, aux
, NULL
);
813 /* Retrieves statistics from mirror associated with client data pointer
814 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
815 * 'bytes', respectively. If a particular counters is not supported,
816 * the appropriate argument is set to UINT64_MAX. */
818 ofproto_mirror_get_stats(struct ofproto
*ofproto
, void *aux
,
819 uint64_t *packets
, uint64_t *bytes
)
821 if (!ofproto
->ofproto_class
->mirror_get_stats
) {
822 *packets
= *bytes
= UINT64_MAX
;
826 return ofproto
->ofproto_class
->mirror_get_stats(ofproto
, aux
,
830 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
831 * which all packets are flooded, instead of using MAC learning. If
832 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
834 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
837 ofproto_set_flood_vlans(struct ofproto
*ofproto
, unsigned long *flood_vlans
)
839 return (ofproto
->ofproto_class
->set_flood_vlans
840 ? ofproto
->ofproto_class
->set_flood_vlans(ofproto
, flood_vlans
)
844 /* Returns true if 'aux' is a registered bundle that is currently in use as the
845 * output for a mirror. */
847 ofproto_is_mirror_output_bundle(const struct ofproto
*ofproto
, void *aux
)
849 return (ofproto
->ofproto_class
->is_mirror_output_bundle
850 ? ofproto
->ofproto_class
->is_mirror_output_bundle(ofproto
, aux
)
854 /* Configuration of OpenFlow tables. */
856 /* Returns the number of OpenFlow tables in 'ofproto'. */
858 ofproto_get_n_tables(const struct ofproto
*ofproto
)
860 return ofproto
->n_tables
;
863 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
864 * settings from 's'. 'table_id' must be in the range 0 through the number of
865 * OpenFlow tables in 'ofproto' minus 1, inclusive.
867 * For read-only tables, only the name may be configured. */
869 ofproto_configure_table(struct ofproto
*ofproto
, int table_id
,
870 const struct ofproto_table_settings
*s
)
872 struct oftable
*table
;
874 assert(table_id
>= 0 && table_id
< ofproto
->n_tables
);
875 table
= &ofproto
->tables
[table_id
];
877 oftable_set_name(table
, s
->name
);
879 if (table
->flags
& OFTABLE_READONLY
) {
884 oftable_enable_eviction(table
, s
->groups
, s
->n_groups
);
886 oftable_disable_eviction(table
);
889 table
->max_flows
= s
->max_flows
;
890 if (classifier_count(&table
->cls
) > table
->max_flows
891 && table
->eviction_fields
) {
892 /* 'table' contains more flows than allowed. We might not be able to
893 * evict them right away because of the asynchronous nature of flow
894 * table changes. Schedule eviction for later. */
895 switch (ofproto
->state
) {
897 ofproto
->state
= S_EVICT
;
901 /* We're already deleting flows, nothing more to do. */
908 ofproto_has_snoops(const struct ofproto
*ofproto
)
910 return connmgr_has_snoops(ofproto
->connmgr
);
914 ofproto_get_snoops(const struct ofproto
*ofproto
, struct sset
*snoops
)
916 connmgr_get_snoops(ofproto
->connmgr
, snoops
);
920 ofproto_flush__(struct ofproto
*ofproto
)
922 struct ofopgroup
*group
;
923 struct oftable
*table
;
925 if (ofproto
->ofproto_class
->flush
) {
926 ofproto
->ofproto_class
->flush(ofproto
);
929 group
= ofopgroup_create_unattached(ofproto
);
930 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
931 struct rule
*rule
, *next_rule
;
932 struct cls_cursor cursor
;
934 if (table
->flags
& OFTABLE_HIDDEN
) {
938 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
939 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
940 if (!rule
->pending
) {
941 ofoperation_create(group
, rule
, OFOPERATION_DELETE
);
942 oftable_remove_rule(rule
);
943 ofproto
->ofproto_class
->rule_destruct(rule
);
947 ofopgroup_submit(group
);
951 ofproto_destroy__(struct ofproto
*ofproto
)
953 struct oftable
*table
;
955 assert(list_is_empty(&ofproto
->pending
));
956 assert(!ofproto
->n_pending
);
958 connmgr_destroy(ofproto
->connmgr
);
960 hmap_remove(&all_ofprotos
, &ofproto
->hmap_node
);
963 free(ofproto
->mfr_desc
);
964 free(ofproto
->hw_desc
);
965 free(ofproto
->sw_desc
);
966 free(ofproto
->serial_desc
);
967 free(ofproto
->dp_desc
);
968 hmap_destroy(&ofproto
->ports
);
969 shash_destroy(&ofproto
->port_by_name
);
971 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
972 oftable_destroy(table
);
974 free(ofproto
->tables
);
976 hmap_destroy(&ofproto
->deletions
);
978 free(ofproto
->vlan_bitmap
);
980 ofproto
->ofproto_class
->dealloc(ofproto
);
984 ofproto_destroy(struct ofproto
*p
)
986 struct ofport
*ofport
, *next_ofport
;
993 HMAP_FOR_EACH_SAFE (ofport
, next_ofport
, hmap_node
, &p
->ports
) {
994 ofport_destroy(ofport
);
997 p
->ofproto_class
->destruct(p
);
998 ofproto_destroy__(p
);
1001 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1002 * kernel datapath, for example, this destroys the datapath in the kernel, and
1003 * with the netdev-based datapath, it tears down the data structures that
1004 * represent the datapath.
1006 * The datapath should not be currently open as an ofproto. */
1008 ofproto_delete(const char *name
, const char *type
)
1010 const struct ofproto_class
*class = ofproto_class_find__(type
);
1011 return (!class ? EAFNOSUPPORT
1012 : !class->del
? EACCES
1013 : class->del(type
, name
));
1017 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
1019 if (error
== ENOBUFS
) {
1020 reinit_ports(ofproto
);
1021 } else if (!error
) {
1022 update_port(ofproto
, devname
);
1028 ofproto_run(struct ofproto
*p
)
1030 struct ofport
*ofport
;
1034 error
= p
->ofproto_class
->run(p
);
1035 if (error
&& error
!= EAGAIN
) {
1036 VLOG_ERR_RL(&rl
, "%s: run failed (%s)", p
->name
, strerror(error
));
1039 if (p
->ofproto_class
->port_poll
) {
1040 while ((error
= p
->ofproto_class
->port_poll(p
, &devname
)) != EAGAIN
) {
1041 process_port_change(p
, error
, devname
);
1045 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1046 unsigned int change_seq
= netdev_change_seq(ofport
->netdev
);
1047 if (ofport
->change_seq
!= change_seq
) {
1048 ofport
->change_seq
= change_seq
;
1049 update_port(p
, netdev_get_name(ofport
->netdev
));
1055 connmgr_run(p
->connmgr
, handle_openflow
);
1059 connmgr_run(p
->connmgr
, NULL
);
1061 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1062 p
->state
= S_OPENFLOW
;
1067 connmgr_run(p
->connmgr
, NULL
);
1069 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1070 connmgr_flushed(p
->connmgr
);
1071 p
->state
= S_OPENFLOW
;
1082 /* Performs periodic activity required by 'ofproto' that needs to be done
1083 * with the least possible latency.
1085 * It makes sense to call this function a couple of times per poll loop, to
1086 * provide a significant performance boost on some benchmarks with the
1087 * ofproto-dpif implementation. */
1089 ofproto_run_fast(struct ofproto
*p
)
1093 error
= p
->ofproto_class
->run_fast
? p
->ofproto_class
->run_fast(p
) : 0;
1094 if (error
&& error
!= EAGAIN
) {
1095 VLOG_ERR_RL(&rl
, "%s: fastpath run failed (%s)",
1096 p
->name
, strerror(error
));
1102 ofproto_wait(struct ofproto
*p
)
1104 struct ofport
*ofport
;
1106 p
->ofproto_class
->wait(p
);
1107 if (p
->ofproto_class
->port_poll_wait
) {
1108 p
->ofproto_class
->port_poll_wait(p
);
1111 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1112 if (ofport
->change_seq
!= netdev_change_seq(ofport
->netdev
)) {
1113 poll_immediate_wake();
1119 connmgr_wait(p
->connmgr
, true);
1124 connmgr_wait(p
->connmgr
, false);
1125 if (list_is_empty(&p
->pending
) && hmap_is_empty(&p
->deletions
)) {
1126 poll_immediate_wake();
1133 ofproto_is_alive(const struct ofproto
*p
)
1135 return connmgr_has_controllers(p
->connmgr
);
1139 ofproto_get_ofproto_controller_info(const struct ofproto
*ofproto
,
1142 connmgr_get_controller_info(ofproto
->connmgr
, info
);
1146 ofproto_free_ofproto_controller_info(struct shash
*info
)
1148 connmgr_free_controller_info(info
);
1151 /* Makes a deep copy of 'old' into 'port'. */
1153 ofproto_port_clone(struct ofproto_port
*port
, const struct ofproto_port
*old
)
1155 port
->name
= xstrdup(old
->name
);
1156 port
->type
= xstrdup(old
->type
);
1157 port
->ofp_port
= old
->ofp_port
;
1160 /* Frees memory allocated to members of 'ofproto_port'.
1162 * Do not call this function on an ofproto_port obtained from
1163 * ofproto_port_dump_next(): that function retains ownership of the data in the
1166 ofproto_port_destroy(struct ofproto_port
*ofproto_port
)
1168 free(ofproto_port
->name
);
1169 free(ofproto_port
->type
);
1172 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1174 * This function provides no status indication. An error status for the entire
1175 * dump operation is provided when it is completed by calling
1176 * ofproto_port_dump_done().
1179 ofproto_port_dump_start(struct ofproto_port_dump
*dump
,
1180 const struct ofproto
*ofproto
)
1182 dump
->ofproto
= ofproto
;
1183 dump
->error
= ofproto
->ofproto_class
->port_dump_start(ofproto
,
1187 /* Attempts to retrieve another port from 'dump', which must have been created
1188 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1189 * 'port' and returns true. On failure, returns false.
1191 * Failure might indicate an actual error or merely that the last port has been
1192 * dumped. An error status for the entire dump operation is provided when it
1193 * is completed by calling ofproto_port_dump_done().
1195 * The ofproto owns the data stored in 'port'. It will remain valid until at
1196 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1197 * ofproto_port_dump_done(). */
1199 ofproto_port_dump_next(struct ofproto_port_dump
*dump
,
1200 struct ofproto_port
*port
)
1202 const struct ofproto
*ofproto
= dump
->ofproto
;
1208 dump
->error
= ofproto
->ofproto_class
->port_dump_next(ofproto
, dump
->state
,
1211 ofproto
->ofproto_class
->port_dump_done(ofproto
, dump
->state
);
1217 /* Completes port table dump operation 'dump', which must have been created
1218 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1219 * error-free, otherwise a positive errno value describing the problem. */
1221 ofproto_port_dump_done(struct ofproto_port_dump
*dump
)
1223 const struct ofproto
*ofproto
= dump
->ofproto
;
1225 dump
->error
= ofproto
->ofproto_class
->port_dump_done(ofproto
,
1228 return dump
->error
== EOF
? 0 : dump
->error
;
1231 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
1232 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
1233 * is non-null). On failure, returns a positive errno value and sets
1234 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
1236 ofproto_port_add(struct ofproto
*ofproto
, struct netdev
*netdev
,
1237 uint16_t *ofp_portp
)
1242 error
= ofproto
->ofproto_class
->port_add(ofproto
, netdev
, &ofp_port
);
1244 update_port(ofproto
, netdev_get_name(netdev
));
1247 *ofp_portp
= error
? OFPP_NONE
: ofp_port
;
1252 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1253 * initializes '*port' appropriately; on failure, returns a positive errno
1256 * The caller owns the data in 'ofproto_port' and must free it with
1257 * ofproto_port_destroy() when it is no longer needed. */
1259 ofproto_port_query_by_name(const struct ofproto
*ofproto
, const char *devname
,
1260 struct ofproto_port
*port
)
1264 error
= ofproto
->ofproto_class
->port_query_by_name(ofproto
, devname
, port
);
1266 memset(port
, 0, sizeof *port
);
1271 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1272 * Returns 0 if successful, otherwise a positive errno. */
1274 ofproto_port_del(struct ofproto
*ofproto
, uint16_t ofp_port
)
1276 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
1277 const char *name
= ofport
? netdev_get_name(ofport
->netdev
) : "<unknown>";
1280 error
= ofproto
->ofproto_class
->port_del(ofproto
, ofp_port
);
1281 if (!error
&& ofport
) {
1282 /* 'name' is the netdev's name and update_port() is going to close the
1283 * netdev. Just in case update_port() refers to 'name' after it
1284 * destroys 'ofport', make a copy of it around the update_port()
1286 char *devname
= xstrdup(name
);
1287 update_port(ofproto
, devname
);
1293 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1294 * performs the 'n_actions' actions in 'actions'. The new flow will not
1297 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1298 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1299 * controllers; otherwise, it will be hidden.
1301 * The caller retains ownership of 'cls_rule' and 'actions'.
1303 * This is a helper function for in-band control and fail-open. */
1305 ofproto_add_flow(struct ofproto
*ofproto
, const struct cls_rule
*cls_rule
,
1306 const union ofp_action
*actions
, size_t n_actions
)
1308 const struct rule
*rule
;
1310 rule
= rule_from_cls_rule(classifier_find_rule_exactly(
1311 &ofproto
->tables
[0].cls
, cls_rule
));
1312 if (!rule
|| !ofputil_actions_equal(rule
->actions
, rule
->n_actions
,
1313 actions
, n_actions
)) {
1314 struct ofputil_flow_mod fm
;
1316 memset(&fm
, 0, sizeof fm
);
1318 fm
.buffer_id
= UINT32_MAX
;
1319 fm
.actions
= (union ofp_action
*) actions
;
1320 fm
.n_actions
= n_actions
;
1321 add_flow(ofproto
, NULL
, &fm
, NULL
);
1325 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1326 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1327 * operation cannot be initiated now but may be retried later.
1329 * This is a helper function for in-band control and fail-open. */
1331 ofproto_flow_mod(struct ofproto
*ofproto
, const struct ofputil_flow_mod
*fm
)
1333 return handle_flow_mod__(ofproto
, NULL
, fm
, NULL
);
1336 /* Searches for a rule with matching criteria exactly equal to 'target' in
1337 * ofproto's table 0 and, if it finds one, deletes it.
1339 * This is a helper function for in-band control and fail-open. */
1341 ofproto_delete_flow(struct ofproto
*ofproto
, const struct cls_rule
*target
)
1345 rule
= rule_from_cls_rule(classifier_find_rule_exactly(
1346 &ofproto
->tables
[0].cls
, target
));
1348 /* No such rule -> success. */
1350 } else if (rule
->pending
) {
1351 /* An operation on the rule is already pending -> failure.
1352 * Caller must retry later if it's important. */
1355 /* Initiate deletion -> success. */
1356 struct ofopgroup
*group
= ofopgroup_create_unattached(ofproto
);
1357 ofoperation_create(group
, rule
, OFOPERATION_DELETE
);
1358 oftable_remove_rule(rule
);
1359 ofproto
->ofproto_class
->rule_destruct(rule
);
1360 ofopgroup_submit(group
);
1366 /* Starts the process of deleting all of the flows from all of ofproto's flow
1367 * tables and then reintroducing the flows required by in-band control and
1368 * fail-open. The process will complete in a later call to ofproto_run(). */
1370 ofproto_flush_flows(struct ofproto
*ofproto
)
1372 COVERAGE_INC(ofproto_flush
);
1373 ofproto
->state
= S_FLUSH
;
1377 reinit_ports(struct ofproto
*p
)
1379 struct ofproto_port_dump dump
;
1380 struct sset devnames
;
1381 struct ofport
*ofport
;
1382 struct ofproto_port ofproto_port
;
1383 const char *devname
;
1385 COVERAGE_INC(ofproto_reinit_ports
);
1387 sset_init(&devnames
);
1388 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1389 sset_add(&devnames
, netdev_get_name(ofport
->netdev
));
1391 OFPROTO_PORT_FOR_EACH (&ofproto_port
, &dump
, p
) {
1392 sset_add(&devnames
, ofproto_port
.name
);
1395 SSET_FOR_EACH (devname
, &devnames
) {
1396 update_port(p
, devname
);
1398 sset_destroy(&devnames
);
1401 /* Opens and returns a netdev for 'ofproto_port', or a null pointer if the
1402 * netdev cannot be opened. On success, also fills in 'opp'. */
1403 static struct netdev
*
1404 ofport_open(const struct ofproto_port
*ofproto_port
,
1405 struct ofputil_phy_port
*pp
)
1407 enum netdev_flags flags
;
1408 struct netdev
*netdev
;
1411 error
= netdev_open(ofproto_port
->name
, ofproto_port
->type
, &netdev
);
1413 VLOG_WARN_RL(&rl
, "ignoring port %s (%"PRIu16
") because netdev %s "
1414 "cannot be opened (%s)",
1415 ofproto_port
->name
, ofproto_port
->ofp_port
,
1416 ofproto_port
->name
, strerror(error
));
1420 pp
->port_no
= ofproto_port
->ofp_port
;
1421 netdev_get_etheraddr(netdev
, pp
->hw_addr
);
1422 ovs_strlcpy(pp
->name
, ofproto_port
->name
, sizeof pp
->name
);
1423 netdev_get_flags(netdev
, &flags
);
1424 pp
->config
= flags
& NETDEV_UP
? 0 : OFPUTIL_PC_PORT_DOWN
;
1425 pp
->state
= netdev_get_carrier(netdev
) ? 0 : OFPUTIL_PS_LINK_DOWN
;
1426 netdev_get_features(netdev
, &pp
->curr
, &pp
->advertised
,
1427 &pp
->supported
, &pp
->peer
);
1428 pp
->curr_speed
= netdev_features_to_bps(pp
->curr
);
1429 pp
->max_speed
= netdev_features_to_bps(pp
->supported
);
1434 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1435 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1438 ofport_equal(const struct ofputil_phy_port
*a
,
1439 const struct ofputil_phy_port
*b
)
1441 return (eth_addr_equals(a
->hw_addr
, b
->hw_addr
)
1442 && a
->state
== b
->state
1443 && !((a
->config
^ b
->config
) & OFPUTIL_PC_PORT_DOWN
)
1444 && a
->curr
== b
->curr
1445 && a
->advertised
== b
->advertised
1446 && a
->supported
== b
->supported
1447 && a
->peer
== b
->peer
1448 && a
->curr_speed
== b
->curr_speed
1449 && a
->max_speed
== b
->max_speed
);
1452 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1453 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1454 * one with the same name or port number). */
1456 ofport_install(struct ofproto
*p
,
1457 struct netdev
*netdev
, const struct ofputil_phy_port
*pp
)
1459 const char *netdev_name
= netdev_get_name(netdev
);
1460 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
= pp
->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 update_mtu(p
, ofport
);
1481 /* Let the ofproto_class initialize its private data. */
1482 error
= p
->ofproto_class
->port_construct(ofport
);
1486 connmgr_send_port_status(p
->connmgr
, pp
, OFPPR_ADD
);
1490 VLOG_WARN_RL(&rl
, "%s: could not add port %s (%s)",
1491 p
->name
, netdev_name
, strerror(error
));
1493 ofport_destroy__(ofport
);
1495 netdev_close(netdev
);
1499 /* Removes 'ofport' from 'p' and destroys it. */
1501 ofport_remove(struct ofport
*ofport
)
1503 connmgr_send_port_status(ofport
->ofproto
->connmgr
, &ofport
->pp
,
1505 ofport_destroy(ofport
);
1508 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1511 ofport_remove_with_name(struct ofproto
*ofproto
, const char *name
)
1513 struct ofport
*port
= shash_find_data(&ofproto
->port_by_name
, name
);
1515 ofport_remove(port
);
1519 /* Updates 'port' with new 'pp' description.
1521 * Does not handle a name or port number change. The caller must implement
1522 * such a change as a delete followed by an add. */
1524 ofport_modified(struct ofport
*port
, struct ofputil_phy_port
*pp
)
1526 memcpy(port
->pp
.hw_addr
, pp
->hw_addr
, ETH_ADDR_LEN
);
1527 port
->pp
.config
= ((port
->pp
.config
& ~OFPUTIL_PC_PORT_DOWN
)
1528 | (pp
->config
& OFPUTIL_PC_PORT_DOWN
));
1529 port
->pp
.state
= pp
->state
;
1530 port
->pp
.curr
= pp
->curr
;
1531 port
->pp
.advertised
= pp
->advertised
;
1532 port
->pp
.supported
= pp
->supported
;
1533 port
->pp
.peer
= pp
->peer
;
1534 port
->pp
.curr_speed
= pp
->curr_speed
;
1535 port
->pp
.max_speed
= pp
->max_speed
;
1537 connmgr_send_port_status(port
->ofproto
->connmgr
, &port
->pp
, OFPPR_MODIFY
);
1540 /* Update OpenFlow 'state' in 'port' and notify controller. */
1542 ofproto_port_set_state(struct ofport
*port
, enum ofputil_port_state state
)
1544 if (port
->pp
.state
!= state
) {
1545 port
->pp
.state
= state
;
1546 connmgr_send_port_status(port
->ofproto
->connmgr
, &port
->pp
,
1552 ofproto_port_unregister(struct ofproto
*ofproto
, uint16_t ofp_port
)
1554 struct ofport
*port
= ofproto_get_port(ofproto
, ofp_port
);
1556 if (port
->ofproto
->ofproto_class
->set_realdev
) {
1557 port
->ofproto
->ofproto_class
->set_realdev(port
, 0, 0);
1559 if (port
->ofproto
->ofproto_class
->set_stp_port
) {
1560 port
->ofproto
->ofproto_class
->set_stp_port(port
, NULL
);
1562 if (port
->ofproto
->ofproto_class
->set_cfm
) {
1563 port
->ofproto
->ofproto_class
->set_cfm(port
, NULL
);
1565 if (port
->ofproto
->ofproto_class
->bundle_remove
) {
1566 port
->ofproto
->ofproto_class
->bundle_remove(port
);
1572 ofport_destroy__(struct ofport
*port
)
1574 struct ofproto
*ofproto
= port
->ofproto
;
1575 const char *name
= netdev_get_name(port
->netdev
);
1577 hmap_remove(&ofproto
->ports
, &port
->hmap_node
);
1578 shash_delete(&ofproto
->port_by_name
,
1579 shash_find(&ofproto
->port_by_name
, name
));
1581 netdev_close(port
->netdev
);
1582 ofproto
->ofproto_class
->port_dealloc(port
);
1586 ofport_destroy(struct ofport
*port
)
1589 port
->ofproto
->ofproto_class
->port_destruct(port
);
1590 ofport_destroy__(port
);
1595 ofproto_get_port(const struct ofproto
*ofproto
, uint16_t ofp_port
)
1597 struct ofport
*port
;
1599 HMAP_FOR_EACH_IN_BUCKET (port
, hmap_node
,
1600 hash_int(ofp_port
, 0), &ofproto
->ports
) {
1601 if (port
->ofp_port
== ofp_port
) {
1609 ofproto_port_get_stats(const struct ofport
*port
, struct netdev_stats
*stats
)
1611 struct ofproto
*ofproto
= port
->ofproto
;
1614 if (ofproto
->ofproto_class
->port_get_stats
) {
1615 error
= ofproto
->ofproto_class
->port_get_stats(port
, stats
);
1624 update_port(struct ofproto
*ofproto
, const char *name
)
1626 struct ofproto_port ofproto_port
;
1627 struct ofputil_phy_port pp
;
1628 struct netdev
*netdev
;
1629 struct ofport
*port
;
1631 COVERAGE_INC(ofproto_update_port
);
1633 /* Fetch 'name''s location and properties from the datapath. */
1634 netdev
= (!ofproto_port_query_by_name(ofproto
, name
, &ofproto_port
)
1635 ? ofport_open(&ofproto_port
, &pp
)
1638 port
= ofproto_get_port(ofproto
, ofproto_port
.ofp_port
);
1639 if (port
&& !strcmp(netdev_get_name(port
->netdev
), name
)) {
1640 struct netdev
*old_netdev
= port
->netdev
;
1642 /* 'name' hasn't changed location. Any properties changed? */
1643 if (!ofport_equal(&port
->pp
, &pp
)) {
1644 ofport_modified(port
, &pp
);
1647 update_mtu(ofproto
, port
);
1649 /* Install the newly opened netdev in case it has changed.
1650 * Don't close the old netdev yet in case port_modified has to
1651 * remove a retained reference to it.*/
1652 port
->netdev
= netdev
;
1653 port
->change_seq
= netdev_change_seq(netdev
);
1655 if (port
->ofproto
->ofproto_class
->port_modified
) {
1656 port
->ofproto
->ofproto_class
->port_modified(port
);
1659 netdev_close(old_netdev
);
1661 /* If 'port' is nonnull then its name differs from 'name' and thus
1662 * we should delete it. If we think there's a port named 'name'
1663 * then its port number must be wrong now so delete it too. */
1665 ofport_remove(port
);
1667 ofport_remove_with_name(ofproto
, name
);
1668 ofport_install(ofproto
, netdev
, &pp
);
1671 /* Any port named 'name' is gone now. */
1672 ofport_remove_with_name(ofproto
, name
);
1674 ofproto_port_destroy(&ofproto_port
);
1678 init_ports(struct ofproto
*p
)
1680 struct ofproto_port_dump dump
;
1681 struct ofproto_port ofproto_port
;
1683 OFPROTO_PORT_FOR_EACH (&ofproto_port
, &dump
, p
) {
1684 uint16_t ofp_port
= ofproto_port
.ofp_port
;
1685 if (ofproto_get_port(p
, ofp_port
)) {
1686 VLOG_WARN_RL(&rl
, "ignoring duplicate port %"PRIu16
" in datapath",
1688 } else if (shash_find(&p
->port_by_name
, ofproto_port
.name
)) {
1689 VLOG_WARN_RL(&rl
, "ignoring duplicate device %s in datapath",
1692 struct ofputil_phy_port pp
;
1693 struct netdev
*netdev
;
1695 netdev
= ofport_open(&ofproto_port
, &pp
);
1697 ofport_install(p
, netdev
, &pp
);
1705 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
1706 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
1708 find_min_mtu(struct ofproto
*p
)
1710 struct ofport
*ofport
;
1713 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1714 struct netdev
*netdev
= ofport
->netdev
;
1717 /* Skip any internal ports, since that's what we're trying to
1719 if (!strcmp(netdev_get_type(netdev
), "internal")) {
1723 if (netdev_get_mtu(netdev
, &dev_mtu
)) {
1726 if (!mtu
|| dev_mtu
< mtu
) {
1731 return mtu
? mtu
: ETH_PAYLOAD_MAX
;
1734 /* Update MTU of all datapath devices on 'p' to the minimum of the
1735 * non-datapath ports in event of 'port' added or changed. */
1737 update_mtu(struct ofproto
*p
, struct ofport
*port
)
1739 struct ofport
*ofport
;
1740 struct netdev
*netdev
= port
->netdev
;
1741 int dev_mtu
, old_min
;
1743 if (netdev_get_mtu(netdev
, &dev_mtu
)) {
1747 if (!strcmp(netdev_get_type(port
->netdev
), "internal")) {
1748 if (dev_mtu
> p
->min_mtu
) {
1749 if (!netdev_set_mtu(port
->netdev
, p
->min_mtu
)) {
1750 dev_mtu
= p
->min_mtu
;
1753 port
->mtu
= dev_mtu
;
1757 /* For non-internal port find new min mtu. */
1758 old_min
= p
->min_mtu
;
1759 port
->mtu
= dev_mtu
;
1760 p
->min_mtu
= find_min_mtu(p
);
1761 if (p
->min_mtu
== old_min
) {
1765 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1766 struct netdev
*netdev
= ofport
->netdev
;
1768 if (!strcmp(netdev_get_type(netdev
), "internal")) {
1769 if (!netdev_set_mtu(netdev
, p
->min_mtu
)) {
1770 ofport
->mtu
= p
->min_mtu
;
1777 ofproto_rule_destroy__(struct rule
*rule
)
1780 free(rule
->actions
);
1781 rule
->ofproto
->ofproto_class
->rule_dealloc(rule
);
1785 /* This function allows an ofproto implementation to destroy any rules that
1786 * remain when its ->destruct() function is called. The caller must have
1787 * already uninitialized any derived members of 'rule' (step 5 described in the
1788 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
1789 * This function implements steps 6 and 7.
1791 * This function should only be called from an ofproto implementation's
1792 * ->destruct() function. It is not suitable elsewhere. */
1794 ofproto_rule_destroy(struct rule
*rule
)
1796 assert(!rule
->pending
);
1797 oftable_remove_rule(rule
);
1798 ofproto_rule_destroy__(rule
);
1801 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1802 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1805 rule_has_out_port(const struct rule
*rule
, uint16_t out_port
)
1807 const union ofp_action
*oa
;
1810 if (out_port
== OFPP_NONE
) {
1813 OFPUTIL_ACTION_FOR_EACH_UNSAFE (oa
, left
, rule
->actions
, rule
->n_actions
) {
1814 if (action_outputs_to_port(oa
, htons(out_port
))) {
1821 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1822 * statistics appropriately. 'packet' must have at least sizeof(struct
1823 * ofp_packet_in) bytes of headroom.
1825 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1826 * with statistics for 'packet' either way.
1828 * Takes ownership of 'packet'. */
1830 rule_execute(struct rule
*rule
, uint16_t in_port
, struct ofpbuf
*packet
)
1834 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
1836 flow_extract(packet
, 0, 0, in_port
, &flow
);
1837 return rule
->ofproto
->ofproto_class
->rule_execute(rule
, &flow
, packet
);
1840 /* Returns true if 'rule' should be hidden from the controller.
1842 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1843 * (e.g. by in-band control) and are intentionally hidden from the
1846 rule_is_hidden(const struct rule
*rule
)
1848 return rule
->cr
.priority
> UINT16_MAX
;
1851 static enum oftable_flags
1852 rule_get_flags(const struct rule
*rule
)
1854 return rule
->ofproto
->tables
[rule
->table_id
].flags
;
1858 rule_is_modifiable(const struct rule
*rule
)
1860 return !(rule_get_flags(rule
) & OFTABLE_READONLY
);
1864 handle_echo_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
1866 ofconn_send_reply(ofconn
, make_echo_reply(oh
));
1871 handle_features_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
1873 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
1874 struct ofputil_switch_features features
;
1875 struct ofport
*port
;
1879 ofproto
->ofproto_class
->get_features(ofproto
, &arp_match_ip
,
1881 assert(features
.actions
& OFPUTIL_A_OUTPUT
); /* sanity check */
1883 features
.datapath_id
= ofproto
->datapath_id
;
1884 features
.n_buffers
= pktbuf_capacity();
1885 features
.n_tables
= ofproto
->n_tables
;
1886 features
.capabilities
= (OFPUTIL_C_FLOW_STATS
| OFPUTIL_C_TABLE_STATS
|
1887 OFPUTIL_C_PORT_STATS
| OFPUTIL_C_QUEUE_STATS
);
1889 features
.capabilities
|= OFPUTIL_C_ARP_MATCH_IP
;
1892 b
= ofputil_encode_switch_features(&features
, ofconn_get_protocol(ofconn
),
1894 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
1895 ofputil_put_switch_features_port(&port
->pp
, b
);
1898 ofconn_send_reply(ofconn
, b
);
1903 handle_get_config_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
1905 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
1906 struct ofp_switch_config
*osc
;
1907 enum ofp_config_flags flags
;
1911 osc
= make_openflow_xid(sizeof *osc
, OFPT_GET_CONFIG_REPLY
, oh
->xid
, &buf
);
1912 flags
= ofproto
->frag_handling
;
1913 if (ofconn_get_invalid_ttl_to_controller(ofconn
)) {
1914 flags
|= OFPC_INVALID_TTL_TO_CONTROLLER
;
1916 osc
->flags
= htons(flags
);
1917 osc
->miss_send_len
= htons(ofconn_get_miss_send_len(ofconn
));
1918 ofconn_send_reply(ofconn
, buf
);
1924 handle_set_config(struct ofconn
*ofconn
, const struct ofp_switch_config
*osc
)
1926 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
1927 uint16_t flags
= ntohs(osc
->flags
);
1929 if (ofconn_get_type(ofconn
) != OFCONN_PRIMARY
1930 || ofconn_get_role(ofconn
) != NX_ROLE_SLAVE
) {
1931 enum ofp_config_flags cur
= ofproto
->frag_handling
;
1932 enum ofp_config_flags next
= flags
& OFPC_FRAG_MASK
;
1934 assert((cur
& OFPC_FRAG_MASK
) == cur
);
1936 if (ofproto
->ofproto_class
->set_frag_handling(ofproto
, next
)) {
1937 ofproto
->frag_handling
= next
;
1939 VLOG_WARN_RL(&rl
, "%s: unsupported fragment handling mode %s",
1941 ofputil_frag_handling_to_string(next
));
1945 ofconn_set_invalid_ttl_to_controller(ofconn
,
1946 (flags
& OFPC_INVALID_TTL_TO_CONTROLLER
));
1948 ofconn_set_miss_send_len(ofconn
, ntohs(osc
->miss_send_len
));
1953 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
1954 * error message code for the caller to propagate upward. Otherwise, returns
1957 * The log message mentions 'msg_type'. */
1959 reject_slave_controller(struct ofconn
*ofconn
)
1961 if (ofconn_get_type(ofconn
) == OFCONN_PRIMARY
1962 && ofconn_get_role(ofconn
) == NX_ROLE_SLAVE
) {
1963 return OFPERR_OFPBRC_EPERM
;
1970 handle_packet_out(struct ofconn
*ofconn
, const struct ofp_packet_out
*opo
)
1972 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
1973 struct ofputil_packet_out po
;
1974 struct ofpbuf
*payload
;
1978 COVERAGE_INC(ofproto_packet_out
);
1980 error
= reject_slave_controller(ofconn
);
1985 /* Decode message. */
1986 error
= ofputil_decode_packet_out(&po
, opo
);
1992 if (po
.buffer_id
!= UINT32_MAX
) {
1993 error
= ofconn_pktbuf_retrieve(ofconn
, po
.buffer_id
, &payload
, NULL
);
1994 if (error
|| !payload
) {
1998 payload
= xmalloc(sizeof *payload
);
1999 ofpbuf_use_const(payload
, po
.packet
, po
.packet_len
);
2002 /* Send out packet. */
2003 flow_extract(payload
, 0, 0, po
.in_port
, &flow
);
2004 error
= p
->ofproto_class
->packet_out(p
, payload
, &flow
,
2005 po
.actions
, po
.n_actions
);
2006 ofpbuf_delete(payload
);
2012 update_port_config(struct ofport
*port
,
2013 enum ofputil_port_config config
,
2014 enum ofputil_port_config mask
)
2016 enum ofputil_port_config old_config
= port
->pp
.config
;
2017 enum ofputil_port_config toggle
;
2019 toggle
= (config
^ port
->pp
.config
) & mask
;
2020 if (toggle
& OFPUTIL_PC_PORT_DOWN
) {
2021 if (config
& OFPUTIL_PC_PORT_DOWN
) {
2022 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, true);
2024 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, true);
2026 toggle
&= ~OFPUTIL_PC_PORT_DOWN
;
2029 port
->pp
.config
^= toggle
;
2030 if (port
->pp
.config
!= old_config
) {
2031 port
->ofproto
->ofproto_class
->port_reconfigured(port
, old_config
);
2036 handle_port_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2038 struct ofproto
*p
= ofconn_get_ofproto(ofconn
);
2039 struct ofputil_port_mod pm
;
2040 struct ofport
*port
;
2043 error
= reject_slave_controller(ofconn
);
2048 error
= ofputil_decode_port_mod(oh
, &pm
);
2053 port
= ofproto_get_port(p
, pm
.port_no
);
2055 return OFPERR_OFPPMFC_BAD_PORT
;
2056 } else if (!eth_addr_equals(port
->pp
.hw_addr
, pm
.hw_addr
)) {
2057 return OFPERR_OFPPMFC_BAD_HW_ADDR
;
2059 update_port_config(port
, pm
.config
, pm
.mask
);
2061 netdev_set_advertisements(port
->netdev
, pm
.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
= htons(port
->pp
.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 a
2453 * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
2454 * indicating a connectivity problem). Returns zero if CFM is not faulted,
2455 * and -1 if CFM is not enabled on 'port'. */
2457 ofproto_port_get_cfm_fault(const struct ofproto
*ofproto
, uint16_t ofp_port
)
2459 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
2460 return (ofport
&& ofproto
->ofproto_class
->get_cfm_fault
2461 ? ofproto
->ofproto_class
->get_cfm_fault(ofport
)
2465 /* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
2466 * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by
2467 * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a
2468 * number less than 0 if CFM is not enabled on 'ofp_port'. */
2470 ofproto_port_get_cfm_remote_mpids(const struct ofproto
*ofproto
,
2471 uint16_t ofp_port
, const uint64_t **rmps
,
2474 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
2478 return (ofport
&& ofproto
->ofproto_class
->get_cfm_remote_mpids
2479 ? ofproto
->ofproto_class
->get_cfm_remote_mpids(ofport
, rmps
,
2484 /* Checks the health of the CFM for 'ofp_port' within 'ofproto'. Returns an
2485 * integer value between 0 and 100 to indicate the health of the port as a
2486 * percentage which is the average of cfm health of all the remote_mpids or
2487 * returns -1 if CFM is not enabled on 'ofport'. */
2489 ofproto_port_get_cfm_health(const struct ofproto
*ofproto
, uint16_t ofp_port
)
2491 struct ofport
*ofport
= ofproto_get_port(ofproto
, ofp_port
);
2492 return (ofport
&& ofproto
->ofproto_class
->get_cfm_health
2493 ? ofproto
->ofproto_class
->get_cfm_health(ofport
)
2498 handle_aggregate_stats_request(struct ofconn
*ofconn
,
2499 const struct ofp_stats_msg
*osm
)
2501 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2502 struct ofputil_flow_stats_request request
;
2503 struct ofputil_aggregate_stats stats
;
2504 bool unknown_packets
, unknown_bytes
;
2505 struct ofpbuf
*reply
;
2510 error
= ofputil_decode_flow_stats_request(&request
, &osm
->header
);
2515 error
= collect_rules_loose(ofproto
, request
.table_id
, &request
.match
,
2516 request
.cookie
, request
.cookie_mask
,
2517 request
.out_port
, &rules
);
2522 memset(&stats
, 0, sizeof stats
);
2523 unknown_packets
= unknown_bytes
= false;
2524 LIST_FOR_EACH (rule
, ofproto_node
, &rules
) {
2525 uint64_t packet_count
;
2526 uint64_t byte_count
;
2528 ofproto
->ofproto_class
->rule_get_stats(rule
, &packet_count
,
2531 if (packet_count
== UINT64_MAX
) {
2532 unknown_packets
= true;
2534 stats
.packet_count
+= packet_count
;
2537 if (byte_count
== UINT64_MAX
) {
2538 unknown_bytes
= true;
2540 stats
.byte_count
+= byte_count
;
2545 if (unknown_packets
) {
2546 stats
.packet_count
= UINT64_MAX
;
2548 if (unknown_bytes
) {
2549 stats
.byte_count
= UINT64_MAX
;
2552 reply
= ofputil_encode_aggregate_stats_reply(&stats
, osm
);
2553 ofconn_send_reply(ofconn
, reply
);
2558 struct queue_stats_cbdata
{
2559 struct ofport
*ofport
;
2560 struct list replies
;
2564 put_queue_stats(struct queue_stats_cbdata
*cbdata
, uint32_t queue_id
,
2565 const struct netdev_queue_stats
*stats
)
2567 struct ofp_queue_stats
*reply
;
2569 reply
= ofputil_append_stats_reply(sizeof *reply
, &cbdata
->replies
);
2570 reply
->port_no
= htons(cbdata
->ofport
->pp
.port_no
);
2571 memset(reply
->pad
, 0, sizeof reply
->pad
);
2572 reply
->queue_id
= htonl(queue_id
);
2573 put_32aligned_be64(&reply
->tx_bytes
, htonll(stats
->tx_bytes
));
2574 put_32aligned_be64(&reply
->tx_packets
, htonll(stats
->tx_packets
));
2575 put_32aligned_be64(&reply
->tx_errors
, htonll(stats
->tx_errors
));
2579 handle_queue_stats_dump_cb(uint32_t queue_id
,
2580 struct netdev_queue_stats
*stats
,
2583 struct queue_stats_cbdata
*cbdata
= cbdata_
;
2585 put_queue_stats(cbdata
, queue_id
, stats
);
2589 handle_queue_stats_for_port(struct ofport
*port
, uint32_t queue_id
,
2590 struct queue_stats_cbdata
*cbdata
)
2592 cbdata
->ofport
= port
;
2593 if (queue_id
== OFPQ_ALL
) {
2594 netdev_dump_queue_stats(port
->netdev
,
2595 handle_queue_stats_dump_cb
, cbdata
);
2597 struct netdev_queue_stats stats
;
2599 if (!netdev_get_queue_stats(port
->netdev
, queue_id
, &stats
)) {
2600 put_queue_stats(cbdata
, queue_id
, &stats
);
2606 handle_queue_stats_request(struct ofconn
*ofconn
,
2607 const struct ofp_queue_stats_request
*qsr
)
2609 struct ofproto
*ofproto
= ofconn_get_ofproto(ofconn
);
2610 struct queue_stats_cbdata cbdata
;
2611 struct ofport
*port
;
2612 unsigned int port_no
;
2615 COVERAGE_INC(ofproto_queue_req
);
2617 ofputil_start_stats_reply(&qsr
->osm
, &cbdata
.replies
);
2619 port_no
= ntohs(qsr
->port_no
);
2620 queue_id
= ntohl(qsr
->queue_id
);
2621 if (port_no
== OFPP_ALL
) {
2622 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
2623 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
2625 } else if (port_no
< OFPP_MAX
) {
2626 port
= ofproto_get_port(ofproto
, port_no
);
2628 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
2631 ofpbuf_list_delete(&cbdata
.replies
);
2632 return OFPERR_OFPQOFC_BAD_PORT
;
2634 ofconn_send_replies(ofconn
, &cbdata
.replies
);
2640 is_flow_deletion_pending(const struct ofproto
*ofproto
,
2641 const struct cls_rule
*cls_rule
,
2644 if (!hmap_is_empty(&ofproto
->deletions
)) {
2645 struct ofoperation
*op
;
2647 HMAP_FOR_EACH_WITH_HASH (op
, hmap_node
,
2648 cls_rule_hash(cls_rule
, table_id
),
2649 &ofproto
->deletions
) {
2650 if (cls_rule_equal(cls_rule
, &op
->rule
->cr
)) {
2659 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2660 * in which no matching flow already exists in the flow table.
2662 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2663 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
2664 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
2665 * initiated now but may be retried later.
2667 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2670 add_flow(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
2671 const struct ofputil_flow_mod
*fm
, const struct ofp_header
*request
)
2673 struct oftable
*table
;
2674 struct ofopgroup
*group
;
2675 struct rule
*victim
;
2679 error
= check_table_id(ofproto
, fm
->table_id
);
2685 if (fm
->table_id
== 0xff) {
2687 if (ofproto
->ofproto_class
->rule_choose_table
) {
2688 error
= ofproto
->ofproto_class
->rule_choose_table(ofproto
, &fm
->cr
,
2693 assert(table_id
< ofproto
->n_tables
);
2694 table
= &ofproto
->tables
[table_id
];
2696 table
= &ofproto
->tables
[0];
2698 } else if (fm
->table_id
< ofproto
->n_tables
) {
2699 table
= &ofproto
->tables
[fm
->table_id
];
2701 return OFPERR_NXFMFC_BAD_TABLE_ID
;
2704 if (table
->flags
& OFTABLE_READONLY
) {
2705 return OFPERR_OFPBRC_EPERM
;
2708 /* Check for overlap, if requested. */
2709 if (fm
->flags
& OFPFF_CHECK_OVERLAP
2710 && classifier_rule_overlaps(&table
->cls
, &fm
->cr
)) {
2711 return OFPERR_OFPFMFC_OVERLAP
;
2714 /* Serialize against pending deletion. */
2715 if (is_flow_deletion_pending(ofproto
, &fm
->cr
, table
- ofproto
->tables
)) {
2716 return OFPROTO_POSTPONE
;
2719 /* Allocate new rule. */
2720 rule
= ofproto
->ofproto_class
->rule_alloc();
2722 VLOG_WARN_RL(&rl
, "%s: failed to create rule (%s)",
2723 ofproto
->name
, strerror(error
));
2726 rule
->ofproto
= ofproto
;
2728 rule
->pending
= NULL
;
2729 rule
->flow_cookie
= fm
->cookie
;
2730 rule
->created
= rule
->modified
= rule
->used
= time_msec();
2731 rule
->idle_timeout
= fm
->idle_timeout
;
2732 rule
->hard_timeout
= fm
->hard_timeout
;
2733 rule
->table_id
= table
- ofproto
->tables
;
2734 rule
->send_flow_removed
= (fm
->flags
& OFPFF_SEND_FLOW_REM
) != 0;
2735 rule
->actions
= ofputil_actions_clone(fm
->actions
, fm
->n_actions
);
2736 rule
->n_actions
= fm
->n_actions
;
2737 rule
->evictable
= true;
2738 rule
->eviction_group
= NULL
;
2740 /* Insert new rule. */
2741 victim
= oftable_replace_rule(rule
);
2742 if (victim
&& !rule_is_modifiable(victim
)) {
2743 error
= OFPERR_OFPBRC_EPERM
;
2744 } else if (victim
&& victim
->pending
) {
2745 error
= OFPROTO_POSTPONE
;
2749 if (classifier_count(&table
->cls
) > table
->max_flows
) {
2752 was_evictable
= rule
->evictable
;
2753 rule
->evictable
= false;
2754 evict
= choose_rule_to_evict(table
);
2755 rule
->evictable
= was_evictable
;
2758 error
= OFPERR_OFPFMFC_ALL_TABLES_FULL
;
2760 } else if (evict
->pending
) {
2761 error
= OFPROTO_POSTPONE
;
2768 group
= ofopgroup_create(ofproto
, ofconn
, request
, fm
->buffer_id
);
2769 ofoperation_create(group
, rule
, OFOPERATION_ADD
);
2770 rule
->pending
->victim
= victim
;
2772 error
= ofproto
->ofproto_class
->rule_construct(rule
);
2774 ofoperation_destroy(rule
->pending
);
2776 delete_flow__(evict
, group
);
2778 ofopgroup_submit(group
);
2782 /* Back out if an error occurred. */
2784 oftable_substitute_rule(rule
, victim
);
2785 ofproto_rule_destroy__(rule
);
2790 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2792 /* Modifies the rules listed in 'rules', changing their actions to match those
2795 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2798 * Returns 0 on success, otherwise an OpenFlow error code. */
2800 modify_flows__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
2801 const struct ofputil_flow_mod
*fm
,
2802 const struct ofp_header
*request
, struct list
*rules
)
2804 struct ofopgroup
*group
;
2808 group
= ofopgroup_create(ofproto
, ofconn
, request
, fm
->buffer_id
);
2809 error
= OFPERR_OFPBRC_EPERM
;
2810 LIST_FOR_EACH (rule
, ofproto_node
, rules
) {
2811 if (rule_is_modifiable(rule
)) {
2812 /* At least one rule is modifiable, don't report EPERM error. */
2818 if (!ofputil_actions_equal(fm
->actions
, fm
->n_actions
,
2819 rule
->actions
, rule
->n_actions
)) {
2820 ofoperation_create(group
, rule
, OFOPERATION_MODIFY
);
2821 rule
->pending
->actions
= rule
->actions
;
2822 rule
->pending
->n_actions
= rule
->n_actions
;
2823 rule
->actions
= ofputil_actions_clone(fm
->actions
, fm
->n_actions
);
2824 rule
->n_actions
= fm
->n_actions
;
2825 ofproto
->ofproto_class
->rule_modify_actions(rule
);
2827 rule
->modified
= time_msec();
2829 rule
->flow_cookie
= fm
->cookie
;
2831 ofopgroup_submit(group
);
2836 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
2839 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2842 modify_flows_loose(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
2843 const struct ofputil_flow_mod
*fm
,
2844 const struct ofp_header
*request
)
2849 error
= collect_rules_loose(ofproto
, fm
->table_id
, &fm
->cr
,
2850 fm
->cookie
, fm
->cookie_mask
,
2852 return (error
? error
2853 : list_is_empty(&rules
) ? add_flow(ofproto
, ofconn
, fm
, request
)
2854 : modify_flows__(ofproto
, ofconn
, fm
, request
, &rules
));
2857 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
2860 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2863 modify_flow_strict(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
2864 const struct ofputil_flow_mod
*fm
,
2865 const struct ofp_header
*request
)
2870 error
= collect_rules_strict(ofproto
, fm
->table_id
, &fm
->cr
,
2871 fm
->cookie
, fm
->cookie_mask
,
2873 return (error
? error
2874 : list_is_empty(&rules
) ? add_flow(ofproto
, ofconn
, fm
, request
)
2875 : list_is_singleton(&rules
) ? modify_flows__(ofproto
, ofconn
,
2876 fm
, request
, &rules
)
2880 /* OFPFC_DELETE implementation. */
2883 delete_flow__(struct rule
*rule
, struct ofopgroup
*group
)
2885 struct ofproto
*ofproto
= rule
->ofproto
;
2887 ofproto_rule_send_removed(rule
, OFPRR_DELETE
);
2889 ofoperation_create(group
, rule
, OFOPERATION_DELETE
);
2890 oftable_remove_rule(rule
);
2891 ofproto
->ofproto_class
->rule_destruct(rule
);
2894 /* Deletes the rules listed in 'rules'.
2896 * Returns 0 on success, otherwise an OpenFlow error code. */
2898 delete_flows__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
2899 const struct ofp_header
*request
, struct list
*rules
)
2901 struct rule
*rule
, *next
;
2902 struct ofopgroup
*group
;
2904 group
= ofopgroup_create(ofproto
, ofconn
, request
, UINT32_MAX
);
2905 LIST_FOR_EACH_SAFE (rule
, next
, ofproto_node
, rules
) {
2906 delete_flow__(rule
, group
);
2908 ofopgroup_submit(group
);
2913 /* Implements OFPFC_DELETE. */
2915 delete_flows_loose(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
2916 const struct ofputil_flow_mod
*fm
,
2917 const struct ofp_header
*request
)
2922 error
= collect_rules_loose(ofproto
, fm
->table_id
, &fm
->cr
,
2923 fm
->cookie
, fm
->cookie_mask
,
2924 fm
->out_port
, &rules
);
2925 return (error
? error
2926 : !list_is_empty(&rules
) ? delete_flows__(ofproto
, ofconn
, request
,
2931 /* Implements OFPFC_DELETE_STRICT. */
2933 delete_flow_strict(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
2934 const struct ofputil_flow_mod
*fm
,
2935 const struct ofp_header
*request
)
2940 error
= collect_rules_strict(ofproto
, fm
->table_id
, &fm
->cr
,
2941 fm
->cookie
, fm
->cookie_mask
,
2942 fm
->out_port
, &rules
);
2943 return (error
? error
2944 : list_is_singleton(&rules
) ? delete_flows__(ofproto
, ofconn
,
2950 ofproto_rule_send_removed(struct rule
*rule
, uint8_t reason
)
2952 struct ofputil_flow_removed fr
;
2954 if (rule_is_hidden(rule
) || !rule
->send_flow_removed
) {
2959 fr
.cookie
= rule
->flow_cookie
;
2961 calc_flow_duration__(rule
->created
, time_msec(),
2962 &fr
.duration_sec
, &fr
.duration_nsec
);
2963 fr
.idle_timeout
= rule
->idle_timeout
;
2964 rule
->ofproto
->ofproto_class
->rule_get_stats(rule
, &fr
.packet_count
,
2967 connmgr_send_flow_removed(rule
->ofproto
->connmgr
, &fr
);
2971 ofproto_rule_update_used(struct rule
*rule
, long long int used
)
2973 if (used
> rule
->used
) {
2974 struct eviction_group
*evg
= rule
->eviction_group
;
2978 heap_change(&evg
->rules
, &rule
->evg_node
,
2979 rule_eviction_priority(rule
));
2984 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
2985 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
2988 * ofproto implementation ->run() functions should use this function to expire
2989 * OpenFlow flows. */
2991 ofproto_rule_expire(struct rule
*rule
, uint8_t reason
)
2993 struct ofproto
*ofproto
= rule
->ofproto
;
2994 struct ofopgroup
*group
;
2996 assert(reason
== OFPRR_HARD_TIMEOUT
|| reason
== OFPRR_IDLE_TIMEOUT
);
2998 ofproto_rule_send_removed(rule
, reason
);
3000 group
= ofopgroup_create_unattached(ofproto
);
3001 ofoperation_create(group
, rule
, OFOPERATION_DELETE
);
3002 oftable_remove_rule(rule
);
3003 ofproto
->ofproto_class
->rule_destruct(rule
);
3004 ofopgroup_submit(group
);
3008 handle_flow_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3010 struct ofputil_flow_mod fm
;
3013 error
= reject_slave_controller(ofconn
);
3018 error
= ofputil_decode_flow_mod(&fm
, oh
, ofconn_get_protocol(ofconn
));
3023 /* We do not support the emergency flow cache. It will hopefully get
3024 * dropped from OpenFlow in the near future. */
3025 if (fm
.flags
& OFPFF_EMERG
) {
3026 /* There isn't a good fit for an error code, so just state that the
3027 * flow table is full. */
3028 return OFPERR_OFPFMFC_ALL_TABLES_FULL
;
3031 return handle_flow_mod__(ofconn_get_ofproto(ofconn
), ofconn
, &fm
, oh
);
3035 handle_flow_mod__(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3036 const struct ofputil_flow_mod
*fm
,
3037 const struct ofp_header
*oh
)
3039 if (ofproto
->n_pending
>= 50) {
3040 assert(!list_is_empty(&ofproto
->pending
));
3041 return OFPROTO_POSTPONE
;
3044 switch (fm
->command
) {
3046 return add_flow(ofproto
, ofconn
, fm
, oh
);
3049 return modify_flows_loose(ofproto
, ofconn
, fm
, oh
);
3051 case OFPFC_MODIFY_STRICT
:
3052 return modify_flow_strict(ofproto
, ofconn
, fm
, oh
);
3055 return delete_flows_loose(ofproto
, ofconn
, fm
, oh
);
3057 case OFPFC_DELETE_STRICT
:
3058 return delete_flow_strict(ofproto
, ofconn
, fm
, oh
);
3061 if (fm
->command
> 0xff) {
3062 VLOG_WARN_RL(&rl
, "flow_mod has explicit table_id but "
3063 "flow_mod_table_id extension is not enabled");
3065 return OFPERR_OFPFMFC_BAD_COMMAND
;
3070 handle_role_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3072 struct nx_role_request
*nrr
= (struct nx_role_request
*) oh
;
3073 struct nx_role_request
*reply
;
3077 role
= ntohl(nrr
->role
);
3078 if (role
!= NX_ROLE_OTHER
&& role
!= NX_ROLE_MASTER
3079 && role
!= NX_ROLE_SLAVE
) {
3080 return OFPERR_OFPRRFC_BAD_ROLE
;
3083 if (ofconn_get_role(ofconn
) != role
3084 && ofconn_has_pending_opgroups(ofconn
)) {
3085 return OFPROTO_POSTPONE
;
3088 ofconn_set_role(ofconn
, role
);
3090 reply
= make_nxmsg_xid(sizeof *reply
, NXT_ROLE_REPLY
, oh
->xid
, &buf
);
3091 reply
->role
= htonl(role
);
3092 ofconn_send_reply(ofconn
, buf
);
3098 handle_nxt_flow_mod_table_id(struct ofconn
*ofconn
,
3099 const struct ofp_header
*oh
)
3101 const struct nx_flow_mod_table_id
*msg
3102 = (const struct nx_flow_mod_table_id
*) oh
;
3103 enum ofputil_protocol cur
, next
;
3105 cur
= ofconn_get_protocol(ofconn
);
3106 next
= ofputil_protocol_set_tid(cur
, msg
->set
!= 0);
3107 ofconn_set_protocol(ofconn
, next
);
3113 handle_nxt_set_flow_format(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3115 const struct nx_set_flow_format
*msg
3116 = (const struct nx_set_flow_format
*) oh
;
3117 enum ofputil_protocol cur
, next
;
3118 enum ofputil_protocol next_base
;
3120 next_base
= ofputil_nx_flow_format_to_protocol(ntohl(msg
->format
));
3122 return OFPERR_OFPBRC_EPERM
;
3125 cur
= ofconn_get_protocol(ofconn
);
3126 next
= ofputil_protocol_set_base(cur
, next_base
);
3127 if (cur
!= next
&& ofconn_has_pending_opgroups(ofconn
)) {
3128 /* Avoid sending async messages in surprising protocol. */
3129 return OFPROTO_POSTPONE
;
3132 ofconn_set_protocol(ofconn
, next
);
3137 handle_nxt_set_packet_in_format(struct ofconn
*ofconn
,
3138 const struct ofp_header
*oh
)
3140 const struct nx_set_packet_in_format
*msg
;
3143 msg
= (const struct nx_set_packet_in_format
*) oh
;
3144 format
= ntohl(msg
->format
);
3145 if (format
!= NXPIF_OPENFLOW10
&& format
!= NXPIF_NXM
) {
3146 return OFPERR_OFPBRC_EPERM
;
3149 if (format
!= ofconn_get_packet_in_format(ofconn
)
3150 && ofconn_has_pending_opgroups(ofconn
)) {
3151 /* Avoid sending async message in surprsing packet in format. */
3152 return OFPROTO_POSTPONE
;
3155 ofconn_set_packet_in_format(ofconn
, format
);
3160 handle_nxt_set_async_config(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3162 const struct nx_async_config
*msg
= (const struct nx_async_config
*) oh
;
3163 uint32_t master
[OAM_N_TYPES
];
3164 uint32_t slave
[OAM_N_TYPES
];
3166 master
[OAM_PACKET_IN
] = ntohl(msg
->packet_in_mask
[0]);
3167 master
[OAM_PORT_STATUS
] = ntohl(msg
->port_status_mask
[0]);
3168 master
[OAM_FLOW_REMOVED
] = ntohl(msg
->flow_removed_mask
[0]);
3170 slave
[OAM_PACKET_IN
] = ntohl(msg
->packet_in_mask
[1]);
3171 slave
[OAM_PORT_STATUS
] = ntohl(msg
->port_status_mask
[1]);
3172 slave
[OAM_FLOW_REMOVED
] = ntohl(msg
->flow_removed_mask
[1]);
3174 ofconn_set_async_config(ofconn
, master
, slave
);
3180 handle_nxt_set_controller_id(struct ofconn
*ofconn
,
3181 const struct ofp_header
*oh
)
3183 const struct nx_controller_id
*nci
;
3185 nci
= (const struct nx_controller_id
*) oh
;
3186 if (!is_all_zeros(nci
->zero
, sizeof nci
->zero
)) {
3187 return OFPERR_NXBRC_MUST_BE_ZERO
;
3190 ofconn_set_controller_id(ofconn
, ntohs(nci
->controller_id
));
3195 handle_barrier_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3197 struct ofp_header
*ob
;
3200 if (ofconn_has_pending_opgroups(ofconn
)) {
3201 return OFPROTO_POSTPONE
;
3204 ob
= make_openflow_xid(sizeof *ob
, OFPT10_BARRIER_REPLY
, oh
->xid
, &buf
);
3205 ofconn_send_reply(ofconn
, buf
);
3210 handle_openflow__(struct ofconn
*ofconn
, const struct ofpbuf
*msg
)
3212 const struct ofp_header
*oh
= msg
->data
;
3213 const struct ofputil_msg_type
*type
;
3216 error
= ofputil_decode_msg_type(oh
, &type
);
3221 switch (ofputil_msg_type_code(type
)) {
3222 /* OpenFlow requests. */
3223 case OFPUTIL_OFPT_ECHO_REQUEST
:
3224 return handle_echo_request(ofconn
, oh
);
3226 case OFPUTIL_OFPT_FEATURES_REQUEST
:
3227 return handle_features_request(ofconn
, oh
);
3229 case OFPUTIL_OFPT_GET_CONFIG_REQUEST
:
3230 return handle_get_config_request(ofconn
, oh
);
3232 case OFPUTIL_OFPT_SET_CONFIG
:
3233 return handle_set_config(ofconn
, msg
->data
);
3235 case OFPUTIL_OFPT_PACKET_OUT
:
3236 return handle_packet_out(ofconn
, msg
->data
);
3238 case OFPUTIL_OFPT_PORT_MOD
:
3239 return handle_port_mod(ofconn
, oh
);
3241 case OFPUTIL_OFPT_FLOW_MOD
:
3242 return handle_flow_mod(ofconn
, oh
);
3244 case OFPUTIL_OFPT_BARRIER_REQUEST
:
3245 return handle_barrier_request(ofconn
, oh
);
3247 /* OpenFlow replies. */
3248 case OFPUTIL_OFPT_ECHO_REPLY
:
3251 /* Nicira extension requests. */
3252 case OFPUTIL_NXT_ROLE_REQUEST
:
3253 return handle_role_request(ofconn
, oh
);
3255 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID
:
3256 return handle_nxt_flow_mod_table_id(ofconn
, oh
);
3258 case OFPUTIL_NXT_SET_FLOW_FORMAT
:
3259 return handle_nxt_set_flow_format(ofconn
, oh
);
3261 case OFPUTIL_NXT_SET_PACKET_IN_FORMAT
:
3262 return handle_nxt_set_packet_in_format(ofconn
, oh
);
3264 case OFPUTIL_NXT_SET_CONTROLLER_ID
:
3265 return handle_nxt_set_controller_id(ofconn
, oh
);
3267 case OFPUTIL_NXT_FLOW_MOD
:
3268 return handle_flow_mod(ofconn
, oh
);
3270 case OFPUTIL_NXT_FLOW_AGE
:
3271 /* Nothing to do. */
3274 case OFPUTIL_NXT_SET_ASYNC_CONFIG
:
3275 return handle_nxt_set_async_config(ofconn
, oh
);
3277 /* Statistics requests. */
3278 case OFPUTIL_OFPST_DESC_REQUEST
:
3279 return handle_desc_stats_request(ofconn
, msg
->data
);
3281 case OFPUTIL_OFPST_FLOW_REQUEST
:
3282 case OFPUTIL_NXST_FLOW_REQUEST
:
3283 return handle_flow_stats_request(ofconn
, msg
->data
);
3285 case OFPUTIL_OFPST_AGGREGATE_REQUEST
:
3286 case OFPUTIL_NXST_AGGREGATE_REQUEST
:
3287 return handle_aggregate_stats_request(ofconn
, msg
->data
);
3289 case OFPUTIL_OFPST_TABLE_REQUEST
:
3290 return handle_table_stats_request(ofconn
, msg
->data
);
3292 case OFPUTIL_OFPST_PORT_REQUEST
:
3293 return handle_port_stats_request(ofconn
, msg
->data
);
3295 case OFPUTIL_OFPST_QUEUE_REQUEST
:
3296 return handle_queue_stats_request(ofconn
, msg
->data
);
3298 case OFPUTIL_MSG_INVALID
:
3299 case OFPUTIL_OFPT_HELLO
:
3300 case OFPUTIL_OFPT_ERROR
:
3301 case OFPUTIL_OFPT_FEATURES_REPLY
:
3302 case OFPUTIL_OFPT_GET_CONFIG_REPLY
:
3303 case OFPUTIL_OFPT_PACKET_IN
:
3304 case OFPUTIL_OFPT_FLOW_REMOVED
:
3305 case OFPUTIL_OFPT_PORT_STATUS
:
3306 case OFPUTIL_OFPT_BARRIER_REPLY
:
3307 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST
:
3308 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY
:
3309 case OFPUTIL_OFPST_DESC_REPLY
:
3310 case OFPUTIL_OFPST_FLOW_REPLY
:
3311 case OFPUTIL_OFPST_QUEUE_REPLY
:
3312 case OFPUTIL_OFPST_PORT_REPLY
:
3313 case OFPUTIL_OFPST_TABLE_REPLY
:
3314 case OFPUTIL_OFPST_AGGREGATE_REPLY
:
3315 case OFPUTIL_NXT_ROLE_REPLY
:
3316 case OFPUTIL_NXT_FLOW_REMOVED
:
3317 case OFPUTIL_NXT_PACKET_IN
:
3318 case OFPUTIL_NXST_FLOW_REPLY
:
3319 case OFPUTIL_NXST_AGGREGATE_REPLY
:
3321 return (oh
->type
== OFPT10_STATS_REQUEST
||
3322 oh
->type
== OFPT10_STATS_REPLY
3323 ? OFPERR_OFPBRC_BAD_STAT
3324 : OFPERR_OFPBRC_BAD_TYPE
);
3329 handle_openflow(struct ofconn
*ofconn
, struct ofpbuf
*ofp_msg
)
3331 int error
= handle_openflow__(ofconn
, ofp_msg
);
3332 if (error
&& error
!= OFPROTO_POSTPONE
) {
3333 ofconn_send_error(ofconn
, ofp_msg
->data
, error
);
3335 COVERAGE_INC(ofproto_recv_openflow
);
3336 return error
!= OFPROTO_POSTPONE
;
3339 /* Asynchronous operations. */
3341 /* Creates and returns a new ofopgroup that is not associated with any
3342 * OpenFlow connection.
3344 * The caller should add operations to the returned group with
3345 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3346 static struct ofopgroup
*
3347 ofopgroup_create_unattached(struct ofproto
*ofproto
)
3349 struct ofopgroup
*group
= xzalloc(sizeof *group
);
3350 group
->ofproto
= ofproto
;
3351 list_init(&group
->ofproto_node
);
3352 list_init(&group
->ops
);
3353 list_init(&group
->ofconn_node
);
3357 /* Creates and returns a new ofopgroup for 'ofproto'.
3359 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
3360 * connection. The 'request' and 'buffer_id' arguments are ignored.
3362 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
3363 * If the ofopgroup eventually fails, then the error reply will include
3364 * 'request'. If the ofopgroup eventually succeeds, then the packet with
3365 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
3367 * The caller should add operations to the returned group with
3368 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3369 static struct ofopgroup
*
3370 ofopgroup_create(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3371 const struct ofp_header
*request
, uint32_t buffer_id
)
3373 struct ofopgroup
*group
= ofopgroup_create_unattached(ofproto
);
3375 size_t request_len
= ntohs(request
->length
);
3377 assert(ofconn_get_ofproto(ofconn
) == ofproto
);
3379 ofconn_add_opgroup(ofconn
, &group
->ofconn_node
);
3380 group
->ofconn
= ofconn
;
3381 group
->request
= xmemdup(request
, MIN(request_len
, 64));
3382 group
->buffer_id
= buffer_id
;
3387 /* Submits 'group' for processing.
3389 * If 'group' contains no operations (e.g. none were ever added, or all of the
3390 * ones that were added completed synchronously), then it is destroyed
3391 * immediately. Otherwise it is added to the ofproto's list of pending
3394 ofopgroup_submit(struct ofopgroup
*group
)
3396 if (list_is_empty(&group
->ops
)) {
3397 ofopgroup_destroy(group
);
3399 list_push_back(&group
->ofproto
->pending
, &group
->ofproto_node
);
3400 group
->ofproto
->n_pending
++;
3405 ofopgroup_destroy(struct ofopgroup
*group
)
3407 assert(list_is_empty(&group
->ops
));
3408 if (!list_is_empty(&group
->ofproto_node
)) {
3409 assert(group
->ofproto
->n_pending
> 0);
3410 group
->ofproto
->n_pending
--;
3411 list_remove(&group
->ofproto_node
);
3413 if (!list_is_empty(&group
->ofconn_node
)) {
3414 list_remove(&group
->ofconn_node
);
3416 ofconn_send_error(group
->ofconn
, group
->request
, group
->error
);
3418 connmgr_retry(group
->ofproto
->connmgr
);
3420 free(group
->request
);
3424 /* Initiates a new operation on 'rule', of the specified 'type', within
3425 * 'group'. Prior to calling, 'rule' must not have any pending operation. */
3427 ofoperation_create(struct ofopgroup
*group
, struct rule
*rule
,
3428 enum ofoperation_type type
)
3430 struct ofoperation
*op
;
3432 assert(!rule
->pending
);
3434 op
= rule
->pending
= xzalloc(sizeof *op
);
3436 list_push_back(&group
->ops
, &op
->group_node
);
3440 op
->flow_cookie
= rule
->flow_cookie
;
3442 if (type
== OFOPERATION_DELETE
) {
3443 hmap_insert(&op
->group
->ofproto
->deletions
, &op
->hmap_node
,
3444 cls_rule_hash(&rule
->cr
, rule
->table_id
));
3449 ofoperation_destroy(struct ofoperation
*op
)
3451 struct ofopgroup
*group
= op
->group
;
3454 op
->rule
->pending
= NULL
;
3456 if (op
->type
== OFOPERATION_DELETE
) {
3457 hmap_remove(&group
->ofproto
->deletions
, &op
->hmap_node
);
3459 list_remove(&op
->group_node
);
3463 if (list_is_empty(&group
->ops
) && !list_is_empty(&group
->ofproto_node
)) {
3464 ofopgroup_destroy(group
);
3468 /* Indicates that 'op' completed with status 'error', which is either 0 to
3469 * indicate success or an OpenFlow error code on failure.
3471 * If 'error' is 0, indicating success, the operation will be committed
3472 * permanently to the flow table. There is one interesting subcase:
3474 * - If 'op' is an "add flow" operation that is replacing an existing rule in
3475 * the flow table (the "victim" rule) by a new one, then the caller must
3476 * have uninitialized any derived state in the victim rule, as in step 5 in
3477 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
3478 * performs steps 6 and 7 for the victim rule, most notably by calling its
3479 * ->rule_dealloc() function.
3481 * If 'error' is nonzero, then generally the operation will be rolled back:
3483 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
3484 * restores the original rule. The caller must have uninitialized any
3485 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
3486 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
3487 * and 7 for the new rule, calling its ->rule_dealloc() function.
3489 * - If 'op' is a "modify flow" operation, ofproto restores the original
3492 * - 'op' must not be a "delete flow" operation. Removing a rule is not
3493 * allowed to fail. It must always succeed.
3495 * Please see the large comment in ofproto/ofproto-provider.h titled
3496 * "Asynchronous Operation Support" for more information. */
3498 ofoperation_complete(struct ofoperation
*op
, enum ofperr error
)
3500 struct ofopgroup
*group
= op
->group
;
3501 struct rule
*rule
= op
->rule
;
3502 struct ofproto
*ofproto
= rule
->ofproto
;
3504 assert(rule
->pending
== op
);
3505 assert(op
->status
< 0);
3509 && op
->type
!= OFOPERATION_DELETE
3511 && group
->buffer_id
!= UINT32_MAX
3512 && list_is_singleton(&op
->group_node
)) {
3513 struct ofpbuf
*packet
;
3516 error
= ofconn_pktbuf_retrieve(group
->ofconn
, group
->buffer_id
,
3520 error
= rule_execute(rule
, in_port
, packet
);
3523 if (!group
->error
) {
3524 group
->error
= error
;
3528 case OFOPERATION_ADD
:
3530 ofproto_rule_destroy__(op
->victim
);
3531 if ((rule
->cr
.wc
.vlan_tci_mask
& htons(VLAN_VID_MASK
))
3532 == htons(VLAN_VID_MASK
)) {
3533 if (ofproto
->vlan_bitmap
) {
3534 uint16_t vid
= vlan_tci_to_vid(rule
->cr
.flow
.vlan_tci
);
3536 if (!bitmap_is_set(ofproto
->vlan_bitmap
, vid
)) {
3537 bitmap_set1(ofproto
->vlan_bitmap
, vid
);
3538 ofproto
->vlans_changed
= true;
3541 ofproto
->vlans_changed
= true;
3545 oftable_substitute_rule(rule
, op
->victim
);
3546 ofproto_rule_destroy__(rule
);
3550 case OFOPERATION_DELETE
:
3552 ofproto_rule_destroy__(rule
);
3556 case OFOPERATION_MODIFY
:
3558 rule
->modified
= time_msec();
3560 free(rule
->actions
);
3561 rule
->actions
= op
->actions
;
3562 rule
->n_actions
= op
->n_actions
;
3570 ofoperation_destroy(op
);
3574 ofoperation_get_victim(struct ofoperation
*op
)
3576 assert(op
->type
== OFOPERATION_ADD
);
3581 pick_datapath_id(const struct ofproto
*ofproto
)
3583 const struct ofport
*port
;
3585 port
= ofproto_get_port(ofproto
, OFPP_LOCAL
);
3587 uint8_t ea
[ETH_ADDR_LEN
];
3590 error
= netdev_get_etheraddr(port
->netdev
, ea
);
3592 return eth_addr_to_uint64(ea
);
3594 VLOG_WARN("could not get MAC address for %s (%s)",
3595 netdev_get_name(port
->netdev
), strerror(error
));
3597 return ofproto
->fallback_dpid
;
3601 pick_fallback_dpid(void)
3603 uint8_t ea
[ETH_ADDR_LEN
];
3604 eth_addr_nicira_random(ea
);
3605 return eth_addr_to_uint64(ea
);
3608 /* Table overflow policy. */
3610 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
3611 * is not configured to evict rules or if the table contains no evictable
3612 * rules. (Rules with 'evictable' set to false or with no timeouts are not
3614 static struct rule
*
3615 choose_rule_to_evict(struct oftable
*table
)
3617 struct eviction_group
*evg
;
3619 if (!table
->eviction_fields
) {
3623 /* In the common case, the outer and inner loops here will each be entered
3626 * - The inner loop normally "return"s in its first iteration. If the
3627 * eviction group has any evictable rules, then it always returns in
3630 * - The outer loop only iterates more than once if the largest eviction
3631 * group has no evictable rules.
3633 * - The outer loop can exit only if table's 'max_flows' is all filled up
3634 * by unevictable rules'. */
3635 HEAP_FOR_EACH (evg
, size_node
, &table
->eviction_groups_by_size
) {
3638 HEAP_FOR_EACH (rule
, evg_node
, &evg
->rules
) {
3639 if (rule
->evictable
) {
3648 /* Searches 'ofproto' for tables that have more flows than their configured
3649 * maximum and that have flow eviction enabled, and evicts as many flows as
3650 * necessary and currently feasible from them.
3652 * This triggers only when an OpenFlow table has N flows in it and then the
3653 * client configures a maximum number of flows less than N. */
3655 ofproto_evict(struct ofproto
*ofproto
)
3657 struct ofopgroup
*group
;
3658 struct oftable
*table
;
3660 group
= ofopgroup_create_unattached(ofproto
);
3661 OFPROTO_FOR_EACH_TABLE (table
, ofproto
) {
3662 while (classifier_count(&table
->cls
) > table
->max_flows
3663 && table
->eviction_fields
) {
3666 rule
= choose_rule_to_evict(table
);
3667 if (!rule
|| rule
->pending
) {
3671 ofoperation_create(group
, rule
, OFOPERATION_DELETE
);
3672 oftable_remove_rule(rule
);
3673 ofproto
->ofproto_class
->rule_destruct(rule
);
3676 ofopgroup_submit(group
);
3679 /* Eviction groups. */
3681 /* Returns the priority to use for an eviction_group that contains 'n_rules'
3682 * rules. The priority contains low-order random bits to ensure that eviction
3683 * groups with the same number of rules are prioritized randomly. */
3685 eviction_group_priority(size_t n_rules
)
3687 uint16_t size
= MIN(UINT16_MAX
, n_rules
);
3688 return (size
<< 16) | random_uint16();
3691 /* Updates 'evg', an eviction_group within 'table', following a change that
3692 * adds or removes rules in 'evg'. */
3694 eviction_group_resized(struct oftable
*table
, struct eviction_group
*evg
)
3696 heap_change(&table
->eviction_groups_by_size
, &evg
->size_node
,
3697 eviction_group_priority(heap_count(&evg
->rules
)));
3700 /* Destroys 'evg', an eviction_group within 'table':
3702 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
3703 * rules themselves, just removes them from the eviction group.)
3705 * - Removes 'evg' from 'table'.
3709 eviction_group_destroy(struct oftable
*table
, struct eviction_group
*evg
)
3711 while (!heap_is_empty(&evg
->rules
)) {
3714 rule
= CONTAINER_OF(heap_pop(&evg
->rules
), struct rule
, evg_node
);
3715 rule
->eviction_group
= NULL
;
3717 hmap_remove(&table
->eviction_groups_by_id
, &evg
->id_node
);
3718 heap_remove(&table
->eviction_groups_by_size
, &evg
->size_node
);
3719 heap_destroy(&evg
->rules
);
3723 /* Removes 'rule' from its eviction group, if any. */
3725 eviction_group_remove_rule(struct rule
*rule
)
3727 if (rule
->eviction_group
) {
3728 struct oftable
*table
= &rule
->ofproto
->tables
[rule
->table_id
];
3729 struct eviction_group
*evg
= rule
->eviction_group
;
3731 rule
->eviction_group
= NULL
;
3732 heap_remove(&evg
->rules
, &rule
->evg_node
);
3733 if (heap_is_empty(&evg
->rules
)) {
3734 eviction_group_destroy(table
, evg
);
3736 eviction_group_resized(table
, evg
);
3741 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
3742 * returns the hash value. */
3744 eviction_group_hash_rule(struct rule
*rule
)
3746 struct oftable
*table
= &rule
->ofproto
->tables
[rule
->table_id
];
3747 const struct mf_subfield
*sf
;
3750 hash
= table
->eviction_group_id_basis
;
3751 for (sf
= table
->eviction_fields
;
3752 sf
< &table
->eviction_fields
[table
->n_eviction_fields
];
3755 if (mf_are_prereqs_ok(sf
->field
, &rule
->cr
.flow
)) {
3756 union mf_value value
;
3758 mf_get_value(sf
->field
, &rule
->cr
.flow
, &value
);
3760 bitwise_zero(&value
, sf
->field
->n_bytes
, 0, sf
->ofs
);
3762 if (sf
->ofs
+ sf
->n_bits
< sf
->field
->n_bytes
* 8) {
3763 unsigned int start
= sf
->ofs
+ sf
->n_bits
;
3764 bitwise_zero(&value
, sf
->field
->n_bytes
, start
,
3765 sf
->field
->n_bytes
* 8 - start
);
3767 hash
= hash_bytes(&value
, sf
->field
->n_bytes
, hash
);
3769 hash
= hash_int(hash
, 0);
3776 /* Returns an eviction group within 'table' with the given 'id', creating one
3778 static struct eviction_group
*
3779 eviction_group_find(struct oftable
*table
, uint32_t id
)
3781 struct eviction_group
*evg
;
3783 HMAP_FOR_EACH_WITH_HASH (evg
, id_node
, id
, &table
->eviction_groups_by_id
) {
3787 evg
= xmalloc(sizeof *evg
);
3788 hmap_insert(&table
->eviction_groups_by_id
, &evg
->id_node
, id
);
3789 heap_insert(&table
->eviction_groups_by_size
, &evg
->size_node
,
3790 eviction_group_priority(0));
3791 heap_init(&evg
->rules
);
3796 /* Returns an eviction priority for 'rule'. The return value should be
3797 * interpreted so that higher priorities make a rule more attractive candidates
3800 rule_eviction_priority(struct rule
*rule
)
3802 long long int hard_expiration
;
3803 long long int idle_expiration
;
3804 long long int expiration
;
3805 uint32_t expiration_offset
;
3807 /* Calculate time of expiration. */
3808 hard_expiration
= (rule
->hard_timeout
3809 ? rule
->modified
+ rule
->hard_timeout
* 1000
3811 idle_expiration
= (rule
->idle_timeout
3812 ? rule
->used
+ rule
->idle_timeout
* 1000
3814 expiration
= MIN(hard_expiration
, idle_expiration
);
3815 if (expiration
== LLONG_MAX
) {
3819 /* Calculate the time of expiration as a number of (approximate) seconds
3820 * after program startup.
3822 * This should work OK for program runs that last UINT32_MAX seconds or
3823 * less. Therefore, please restart OVS at least once every 136 years. */
3824 expiration_offset
= (expiration
>> 10) - (time_boot_msec() >> 10);
3826 /* Invert the expiration offset because we're using a max-heap. */
3827 return UINT32_MAX
- expiration_offset
;
3830 /* Adds 'rule' to an appropriate eviction group for its oftable's
3831 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
3832 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
3835 * The caller must ensure that 'rule' is not already in an eviction group. */
3837 eviction_group_add_rule(struct rule
*rule
)
3839 struct ofproto
*ofproto
= rule
->ofproto
;
3840 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
3842 if (table
->eviction_fields
3843 && (rule
->hard_timeout
|| rule
->idle_timeout
)) {
3844 struct eviction_group
*evg
;
3846 evg
= eviction_group_find(table
, eviction_group_hash_rule(rule
));
3848 rule
->eviction_group
= evg
;
3849 heap_insert(&evg
->rules
, &rule
->evg_node
,
3850 rule_eviction_priority(rule
));
3851 eviction_group_resized(table
, evg
);
3857 /* Initializes 'table'. */
3859 oftable_init(struct oftable
*table
)
3861 memset(table
, 0, sizeof *table
);
3862 classifier_init(&table
->cls
);
3865 /* Destroys 'table', including its classifier and eviction groups.
3867 * The caller is responsible for freeing 'table' itself. */
3869 oftable_destroy(struct oftable
*table
)
3871 assert(classifier_is_empty(&table
->cls
));
3872 oftable_disable_eviction(table
);
3873 classifier_destroy(&table
->cls
);
3877 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
3878 * string, then 'table' will use its default name.
3880 * This only affects the name exposed for a table exposed through the OpenFlow
3881 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
3883 oftable_set_name(struct oftable
*table
, const char *name
)
3885 if (name
&& name
[0]) {
3886 int len
= strnlen(name
, OFP_MAX_TABLE_NAME_LEN
);
3887 if (!table
->name
|| strncmp(name
, table
->name
, len
)) {
3889 table
->name
= xmemdup0(name
, len
);
3897 /* oftables support a choice of two policies when adding a rule would cause the
3898 * number of flows in the table to exceed the configured maximum number: either
3899 * they can refuse to add the new flow or they can evict some existing flow.
3900 * This function configures the former policy on 'table'. */
3902 oftable_disable_eviction(struct oftable
*table
)
3904 if (table
->eviction_fields
) {
3905 struct eviction_group
*evg
, *next
;
3907 HMAP_FOR_EACH_SAFE (evg
, next
, id_node
,
3908 &table
->eviction_groups_by_id
) {
3909 eviction_group_destroy(table
, evg
);
3911 hmap_destroy(&table
->eviction_groups_by_id
);
3912 heap_destroy(&table
->eviction_groups_by_size
);
3914 free(table
->eviction_fields
);
3915 table
->eviction_fields
= NULL
;
3916 table
->n_eviction_fields
= 0;
3920 /* oftables support a choice of two policies when adding a rule would cause the
3921 * number of flows in the table to exceed the configured maximum number: either
3922 * they can refuse to add the new flow or they can evict some existing flow.
3923 * This function configures the latter policy on 'table', with fairness based
3924 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
3925 * 'n_fields' as 0 disables fairness.) */
3927 oftable_enable_eviction(struct oftable
*table
,
3928 const struct mf_subfield
*fields
, size_t n_fields
)
3930 struct cls_cursor cursor
;
3933 if (table
->eviction_fields
3934 && n_fields
== table
->n_eviction_fields
3936 || !memcmp(fields
, table
->eviction_fields
,
3937 n_fields
* sizeof *fields
))) {
3942 oftable_disable_eviction(table
);
3944 table
->n_eviction_fields
= n_fields
;
3945 table
->eviction_fields
= xmemdup(fields
, n_fields
* sizeof *fields
);
3947 table
->eviction_group_id_basis
= random_uint32();
3948 hmap_init(&table
->eviction_groups_by_id
);
3949 heap_init(&table
->eviction_groups_by_size
);
3951 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
3952 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3953 eviction_group_add_rule(rule
);
3957 /* Removes 'rule' from the oftable that contains it. */
3959 oftable_remove_rule(struct rule
*rule
)
3961 struct ofproto
*ofproto
= rule
->ofproto
;
3962 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
3964 classifier_remove(&table
->cls
, &rule
->cr
);
3965 eviction_group_remove_rule(rule
);
3968 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
3969 * oftable that has an identical cls_rule. Returns the rule that was removed,
3970 * if any, and otherwise NULL. */
3971 static struct rule
*
3972 oftable_replace_rule(struct rule
*rule
)
3974 struct ofproto
*ofproto
= rule
->ofproto
;
3975 struct oftable
*table
= &ofproto
->tables
[rule
->table_id
];
3976 struct rule
*victim
;
3978 victim
= rule_from_cls_rule(classifier_replace(&table
->cls
, &rule
->cr
));
3980 eviction_group_remove_rule(victim
);
3982 eviction_group_add_rule(rule
);
3986 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
3988 oftable_substitute_rule(struct rule
*old
, struct rule
*new)
3991 oftable_replace_rule(new);
3993 oftable_remove_rule(old
);
3997 /* unixctl commands. */
4000 ofproto_lookup(const char *name
)
4002 struct ofproto
*ofproto
;
4004 HMAP_FOR_EACH_WITH_HASH (ofproto
, hmap_node
, hash_string(name
, 0),
4006 if (!strcmp(ofproto
->name
, name
)) {
4014 ofproto_unixctl_list(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
4015 const char *argv
[] OVS_UNUSED
, void *aux OVS_UNUSED
)
4017 struct ofproto
*ofproto
;
4021 HMAP_FOR_EACH (ofproto
, hmap_node
, &all_ofprotos
) {
4022 ds_put_format(&results
, "%s\n", ofproto
->name
);
4024 unixctl_command_reply(conn
, ds_cstr(&results
));
4025 ds_destroy(&results
);
4029 ofproto_unixctl_init(void)
4031 static bool registered
;
4037 unixctl_command_register("ofproto/list", "", 0, 0,
4038 ofproto_unixctl_list
, NULL
);
4041 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4043 * This is deprecated. It is only for compatibility with broken device drivers
4044 * in old versions of Linux that do not properly support VLANs when VLAN
4045 * devices are not used. When broken device drivers are no longer in
4046 * widespread use, we will delete these interfaces. */
4048 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4049 * (exactly) by an OpenFlow rule in 'ofproto'. */
4051 ofproto_get_vlan_usage(struct ofproto
*ofproto
, unsigned long int *vlan_bitmap
)
4053 const struct oftable
*oftable
;
4055 free(ofproto
->vlan_bitmap
);
4056 ofproto
->vlan_bitmap
= bitmap_allocate(4096);
4057 ofproto
->vlans_changed
= false;
4059 OFPROTO_FOR_EACH_TABLE (oftable
, ofproto
) {
4060 const struct cls_table
*table
;
4062 HMAP_FOR_EACH (table
, hmap_node
, &oftable
->cls
.tables
) {
4063 if ((table
->wc
.vlan_tci_mask
& htons(VLAN_VID_MASK
))
4064 == htons(VLAN_VID_MASK
)) {
4065 const struct cls_rule
*rule
;
4067 HMAP_FOR_EACH (rule
, hmap_node
, &table
->rules
) {
4068 uint16_t vid
= vlan_tci_to_vid(rule
->flow
.vlan_tci
);
4069 bitmap_set1(vlan_bitmap
, vid
);
4070 bitmap_set1(ofproto
->vlan_bitmap
, vid
);
4077 /* Returns true if new VLANs have come into use by the flow table since the
4078 * last call to ofproto_get_vlan_usage().
4080 * We don't track when old VLANs stop being used. */
4082 ofproto_has_vlan_usage_changed(const struct ofproto
*ofproto
)
4084 return ofproto
->vlans_changed
;
4087 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4088 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4089 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4090 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4091 * then the VLAN device is un-enslaved. */
4093 ofproto_port_set_realdev(struct ofproto
*ofproto
, uint16_t vlandev_ofp_port
,
4094 uint16_t realdev_ofp_port
, int vid
)
4096 struct ofport
*ofport
;
4099 assert(vlandev_ofp_port
!= realdev_ofp_port
);
4101 ofport
= ofproto_get_port(ofproto
, vlandev_ofp_port
);
4103 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16
,
4104 ofproto
->name
, vlandev_ofp_port
);
4108 if (!ofproto
->ofproto_class
->set_realdev
) {
4109 if (!vlandev_ofp_port
) {
4112 VLOG_WARN("%s: vlan splinters not supported", ofproto
->name
);
4116 error
= ofproto
->ofproto_class
->set_realdev(ofport
, realdev_ofp_port
, vid
);
4118 VLOG_WARN("%s: setting realdev on port %"PRIu16
" (%s) failed (%s)",
4119 ofproto
->name
, vlandev_ofp_port
,
4120 netdev_get_name(ofport
->netdev
), strerror(error
));