2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
17 #ifndef OFPROTO_OFPROTO_PROVIDER_H
18 #define OFPROTO_OFPROTO_PROVIDER_H 1
20 /* Definitions for use within ofproto.
26 * Lots of ofproto data structures are only accessed from a single thread.
27 * Those data structures are generally not thread-safe.
29 * The ofproto-dpif ofproto implementation accesses the flow table from
30 * multiple threads, including modifying the flow table from multiple threads
31 * via the "learn" action, so the flow table and various structures that index
32 * it have been made thread-safe. Refer to comments on individual data
33 * structures for details.
37 #include "classifier.h"
38 #include "guarded-list.h"
42 #include "ofp-errors.h"
44 #include "ofproto/ofproto.h"
45 #include "ovs-atomic.h"
46 #include "ovs-thread.h"
53 struct ofputil_flow_mod;
57 extern struct ovs_mutex ofproto_mutex;
59 /* An OpenFlow switch.
61 * With few exceptions, ofproto implementations may look at these fields but
62 * should not modify them. */
64 struct hmap_node hmap_node; /* In global 'all_ofprotos' hmap. */
65 const struct ofproto_class *ofproto_class;
66 char *type; /* Datapath type. */
67 char *name; /* Datapath name. */
70 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
71 uint64_t datapath_id; /* Datapath ID. */
72 bool forward_bpdu; /* Option to allow forwarding of BPDU frames
73 * when NORMAL action is invoked. */
74 char *mfr_desc; /* Manufacturer (NULL for default)b. */
75 char *hw_desc; /* Hardware (NULL for default). */
76 char *sw_desc; /* Software version (NULL for default). */
77 char *serial_desc; /* Serial number (NULL for default). */
78 char *dp_desc; /* Datapath description (NULL for default). */
79 enum ofp_config_flags frag_handling; /* One of OFPC_*. */
82 struct hmap ports; /* Contains "struct ofport"s. */
83 struct shash port_by_name;
84 struct simap ofp_requests; /* OpenFlow port number requests. */
85 uint16_t alloc_port_no; /* Last allocated OpenFlow port number. */
86 uint16_t max_ports; /* Max possible OpenFlow port num, plus one. */
87 struct hmap ofport_usage; /* Map ofport to last used time. */
88 uint64_t change_seq; /* Change sequence for netdev status. */
91 long long int eviction_group_timer; /* For rate limited reheapification. */
92 struct oftable *tables;
95 /* Rules indexed on their cookie values, in all flow tables. */
96 struct hindex cookies OVS_GUARDED_BY(ofproto_mutex);
98 /* List of expirable flows, in all flow tables. */
99 struct list expirable OVS_GUARDED_BY(ofproto_mutex);
102 * OpenFlow meters start at 1. To avoid confusion we leave the first
103 * pointer in the array un-used, and index directly with the OpenFlow
105 struct ofputil_meter_features meter_features;
106 struct meter **meters; /* 'meter_features.max_meter' + 1 pointers. */
108 /* OpenFlow connections. */
109 struct connmgr *connmgr;
111 /* Flow table operation tracking.
113 * 'state' is meaningful only within ofproto.c, one of the enum
114 * ofproto_state constants defined there.
116 * 'pending' is the list of "struct ofopgroup"s currently pending.
118 * 'n_pending' is the number of elements in 'pending'.
120 * 'deletions' contains pending ofoperations of type OFOPERATION_DELETE,
121 * indexed on its rule's flow.*/
123 struct list pending OVS_GUARDED_BY(ofproto_mutex);
124 unsigned int n_pending OVS_GUARDED_BY(ofproto_mutex);
125 struct hmap deletions OVS_GUARDED_BY(ofproto_mutex);
127 /* Delayed rule executions.
129 * We delay calls to ->ofproto_class->rule_execute() past releasing
130 * ofproto_mutex during a flow_mod, because otherwise a "learn" action
131 * triggered by the executing the packet would try to recursively modify
132 * the flow table and reacquire the global lock. */
133 struct guarded_list rule_executes; /* Contains "struct rule_execute"s. */
135 /* Flow table operation logging. */
136 int n_add, n_delete, n_modify; /* Number of unreported ops of each kind. */
137 long long int first_op, last_op; /* Range of times for unreported ops. */
138 long long int next_op_report; /* Time to report ops, or LLONG_MAX. */
139 long long int op_backoff; /* Earliest time to report ops again. */
141 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
143 * This is deprecated. It is only for compatibility with broken device
144 * drivers in old versions of Linux that do not properly support VLANs when
145 * VLAN devices are not used. When broken device drivers are no longer in
146 * widespread use, we will delete these interfaces. */
147 unsigned long int *vlan_bitmap; /* 4096-bit bitmap of in-use VLANs. */
148 bool vlans_changed; /* True if new VLANs are in use. */
149 int min_mtu; /* Current MTU of non-internal ports. */
152 struct ovs_rwlock groups_rwlock;
153 struct hmap groups OVS_GUARDED; /* Contains "struct ofgroup"s. */
154 uint32_t n_groups[4] OVS_GUARDED; /* # of existing groups of each type. */
155 struct ofputil_group_features ogf;
158 void ofproto_init_tables(struct ofproto *, int n_tables);
159 void ofproto_init_max_ports(struct ofproto *, uint16_t max_ports);
161 struct ofproto *ofproto_lookup(const char *name);
162 struct ofport *ofproto_get_port(const struct ofproto *, ofp_port_t ofp_port);
164 /* An OpenFlow port within a "struct ofproto".
166 * The port's name is netdev_get_name(port->netdev).
168 * With few exceptions, ofproto implementations may look at these fields but
169 * should not modify them. */
171 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
172 struct ofproto *ofproto; /* The ofproto that contains this port. */
173 struct netdev *netdev;
174 struct ofputil_phy_port pp;
175 ofp_port_t ofp_port; /* OpenFlow port number. */
176 long long int created; /* Time created, in msec. */
180 void ofproto_port_set_state(struct ofport *, enum ofputil_port_state);
182 /* OpenFlow table flags:
184 * - "Hidden" tables are not included in OpenFlow operations that operate on
185 * "all tables". For example, a request for flow stats on all tables will
186 * omit flows in hidden tables, table stats requests will omit the table
187 * entirely, and the switch features reply will not count the hidden table.
189 * However, operations that specifically name the particular table still
190 * operate on it. For example, flow_mods and flow stats requests on a
193 * To avoid gaps in table IDs (which have unclear validity in OpenFlow),
194 * hidden tables must be the highest-numbered tables that a provider
197 * - "Read-only" tables can't be changed through OpenFlow operations. (At
198 * the moment all flow table operations go effectively through OpenFlow, so
199 * this means that read-only tables can't be changed at all after the
200 * read-only flag is set.)
202 * The generic ofproto layer never sets these flags. An ofproto provider can
203 * set them if it is appropriate.
206 OFTABLE_HIDDEN = 1 << 0, /* Hide from most OpenFlow operations. */
207 OFTABLE_READONLY = 1 << 1 /* Don't allow OpenFlow to change this table. */
210 /* A flow table within a "struct ofproto".
216 * A cls->rwlock read-lock holder prevents rules from being added or deleted.
218 * Adding or removing rules requires holding ofproto_mutex AND the cls->rwlock
221 * cls->rwlock should be held only briefly. For extended access to a rule,
222 * increment its ref_count with ofproto_rule_ref(). A rule will not be freed
223 * until its ref_count reaches zero.
225 * Modifying a rule requires the rule's own mutex. Holding cls->rwlock (for
226 * read or write) does not allow the holder to modify the rule.
228 * Freeing a rule requires ofproto_mutex and the cls->rwlock write-lock. After
229 * removing the rule from the classifier, release a ref_count from the rule
230 * ('cls''s reference to the rule).
232 * Refer to the thread-safety notes on struct rule for more information.*/
234 enum oftable_flags flags;
235 struct classifier cls; /* Contains "struct rule"s. */
236 char *name; /* Table name exposed via OpenFlow, or NULL. */
238 /* Maximum number of flows or UINT_MAX if there is no limit besides any
239 * limit imposed by resource limitations. */
240 unsigned int max_flows;
242 /* These members determine the handling of an attempt to add a flow that
243 * would cause the table to have more than 'max_flows' flows.
245 * If 'eviction_fields' is NULL, overflows will be rejected with an error.
247 * If 'eviction_fields' is nonnull (regardless of whether n_eviction_fields
248 * is nonzero), an overflow will cause a flow to be removed. The flow to
249 * be removed is chosen to give fairness among groups distinguished by
250 * different values for the subfields within 'groups'. */
251 struct mf_subfield *eviction_fields;
252 size_t n_eviction_fields;
256 * When a flow is added that would cause the table to have more than
257 * 'max_flows' flows, and 'eviction_fields' is nonnull, these groups are
258 * used to decide which rule to evict: the rule is chosen from the eviction
259 * group that contains the greatest number of rules.*/
260 uint32_t eviction_group_id_basis;
261 struct hmap eviction_groups_by_id;
262 struct heap eviction_groups_by_size;
264 /* Table config: contains enum ofp_table_config; accessed atomically. */
268 /* Assigns TABLE to each oftable, in turn, in OFPROTO.
270 * All parameters are evaluated multiple times. */
271 #define OFPROTO_FOR_EACH_TABLE(TABLE, OFPROTO) \
272 for ((TABLE) = (OFPROTO)->tables; \
273 (TABLE) < &(OFPROTO)->tables[(OFPROTO)->n_tables]; \
276 /* An OpenFlow flow within a "struct ofproto".
278 * With few exceptions, ofproto implementations may look at these fields but
279 * should not modify them.
285 * Except near the beginning or ending of its lifespan, rule 'rule' belongs to
286 * the classifier rule->ofproto->tables[rule->table_id].cls. The text below
287 * calls this classifier 'cls'.
292 * The thread safety rules described here for "struct rule" are motivated by
295 * - Prevent threads that read members of "struct rule" from reading bad
296 * data due to changes by some thread concurrently modifying those
299 * - Prevent two threads making changes to members of a given "struct rule"
300 * from interfering with each other.
306 * A rule 'rule' may be accessed without a risk of being freed by code that
307 * holds a read-lock or write-lock on 'cls->rwlock' or that owns a reference to
308 * 'rule->ref_count' (or both). Code that needs to hold onto a rule for a
309 * while should take 'cls->rwlock', find the rule it needs, increment
310 * 'rule->ref_count' with ofproto_rule_ref(), and drop 'cls->rwlock'.
312 * 'rule->ref_count' protects 'rule' from being freed. It doesn't protect the
313 * rule from being deleted from 'cls' (that's 'cls->rwlock') and it doesn't
314 * protect members of 'rule' from modification (that's 'rule->mutex').
316 * 'rule->mutex' protects the members of 'rule' from modification. It doesn't
317 * protect the rule from being deleted from 'cls' (that's 'cls->rwlock') and it
318 * doesn't prevent the rule from being freed (that's 'rule->ref_count').
320 * Regarding thread safety, the members of a rule fall into the following
323 * - Immutable. These members are marked 'const'.
325 * - Members that may be safely read or written only by code holding
326 * ofproto_mutex. These are marked OVS_GUARDED_BY(ofproto_mutex).
328 * - Members that may be safely read only by code holding ofproto_mutex or
329 * 'rule->mutex', and safely written only by coding holding ofproto_mutex
330 * AND 'rule->mutex'. These are marked OVS_GUARDED.
333 /* Where this rule resides in an OpenFlow switch.
335 * These are immutable once the rule is constructed, hence 'const'. */
336 struct ofproto *const ofproto; /* The ofproto that contains this rule. */
337 const struct cls_rule cr; /* In owning ofproto's classifier. */
338 const uint8_t table_id; /* Index in ofproto's 'tables' array. */
340 /* Protects members marked OVS_GUARDED.
341 * Readers only need to hold this mutex.
342 * Writers must hold both this mutex AND ofproto_mutex.
343 * By implication writers can read *without* taking this mutex while they
344 * hold ofproto_mutex. */
345 struct ovs_mutex mutex OVS_ACQ_AFTER(ofproto_mutex);
347 /* Number of references.
348 * The classifier owns one reference.
349 * Any thread trying to keep a rule from being freed should hold its own
351 struct ovs_refcount ref_count;
353 /* Operation now in progress, if nonnull. */
354 struct ofoperation *pending OVS_GUARDED_BY(ofproto_mutex);
356 /* A "flow cookie" is the OpenFlow name for a 64-bit value associated with
358 ovs_be64 flow_cookie OVS_GUARDED;
359 struct hindex_node cookie_node OVS_GUARDED_BY(ofproto_mutex);
361 enum ofputil_flow_mod_flags flags OVS_GUARDED;
364 uint16_t hard_timeout OVS_GUARDED; /* In seconds from ->modified. */
365 uint16_t idle_timeout OVS_GUARDED; /* In seconds from ->used. */
367 /* Eviction groups (see comment on struct eviction_group for explanation) .
369 * 'eviction_group' is this rule's eviction group, or NULL if it is not in
370 * any eviction group. When 'eviction_group' is nonnull, 'evg_node' is in
371 * the ->eviction_group->rules hmap. */
372 struct eviction_group *eviction_group OVS_GUARDED_BY(ofproto_mutex);
373 struct heap_node evg_node OVS_GUARDED_BY(ofproto_mutex);
375 /* OpenFlow actions. See struct rule_actions for more thread-safety
377 struct rule_actions *actions OVS_GUARDED;
379 /* In owning meter's 'rules' list. An empty list if there is no meter. */
380 struct list meter_list_node OVS_GUARDED_BY(ofproto_mutex);
382 /* Flow monitors (e.g. for NXST_FLOW_MONITOR, related to struct ofmonitor).
384 * 'add_seqno' is the sequence number when this rule was created.
385 * 'modify_seqno' is the sequence number when this rule was last modified.
386 * See 'monitor_seqno' in connmgr.c for more information. */
387 enum nx_flow_monitor_flags monitor_flags OVS_GUARDED_BY(ofproto_mutex);
388 uint64_t add_seqno OVS_GUARDED_BY(ofproto_mutex);
389 uint64_t modify_seqno OVS_GUARDED_BY(ofproto_mutex);
391 /* Optimisation for flow expiry. In ofproto's 'expirable' list if this
392 * rule is expirable, otherwise empty. */
393 struct list expirable OVS_GUARDED_BY(ofproto_mutex);
395 /* Times. Last so that they are more likely close to the stats managed
396 * by the provider. */
397 long long int created OVS_GUARDED; /* Creation time. */
399 /* Must hold 'mutex' for both read/write, 'ofproto_mutex' not needed. */
400 long long int modified OVS_GUARDED; /* Time of last modification. */
403 void ofproto_rule_ref(struct rule *);
404 void ofproto_rule_unref(struct rule *);
406 struct rule_actions *rule_get_actions(const struct rule *rule)
407 OVS_EXCLUDED(rule->mutex);
408 struct rule_actions *rule_get_actions__(const struct rule *rule)
409 OVS_REQUIRES(rule->mutex);
411 /* Returns true if 'rule' is an OpenFlow 1.3 "table-miss" rule, false
414 * ("Table-miss" rules are special because a packet_in generated through one
415 * uses OFPR_NO_MATCH as its reason, whereas packet_ins generated by any other
416 * rule use OFPR_ACTION.) */
418 rule_is_table_miss(const struct rule *rule)
420 return rule->cr.priority == 0 && cls_rule_is_catchall(&rule->cr);
423 /* A set of actions within a "struct rule".
429 * A struct rule_actions 'actions' may be accessed without a risk of being
430 * freed by code that holds a read-lock or write-lock on 'rule->mutex' (where
431 * 'rule' is the rule for which 'rule->actions == actions') or that owns a
432 * reference to 'actions->ref_count' (or both). */
433 struct rule_actions {
434 struct ovs_refcount ref_count;
436 /* These members are immutable: they do not change during the struct's
438 struct ofpact *ofpacts; /* Sequence of "struct ofpacts". */
439 unsigned int ofpacts_len; /* Size of 'ofpacts', in bytes. */
440 uint32_t provider_meter_id; /* Datapath meter_id, or UINT32_MAX. */
443 struct rule_actions *rule_actions_create(const struct ofproto *,
444 const struct ofpact *, size_t);
445 void rule_actions_ref(struct rule_actions *);
446 void rule_actions_unref(struct rule_actions *);
448 /* A set of rules to which an OpenFlow operation applies. */
449 struct rule_collection {
450 struct rule **rules; /* The rules. */
451 size_t n; /* Number of rules collected. */
453 size_t capacity; /* Number of rules that will fit in 'rules'. */
454 struct rule *stub[64]; /* Preallocated rules to avoid malloc(). */
457 void rule_collection_init(struct rule_collection *);
458 void rule_collection_add(struct rule_collection *, struct rule *);
459 void rule_collection_ref(struct rule_collection *) OVS_REQUIRES(ofproto_mutex);
460 void rule_collection_unref(struct rule_collection *);
461 void rule_collection_destroy(struct rule_collection *);
463 /* Limits the number of flows allowed in the datapath. Only affects the
464 * ofproto-dpif implementation. */
465 extern unsigned ofproto_flow_limit;
467 /* Number of upcall handler and revalidator threads. Only affects the
468 * ofproto-dpif implementation. */
469 extern size_t n_handlers, n_revalidators;
471 static inline struct rule *
472 rule_from_cls_rule(const struct cls_rule *cls_rule)
474 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
477 void ofproto_rule_expire(struct rule *rule, uint8_t reason)
478 OVS_REQUIRES(ofproto_mutex);
479 void ofproto_rule_delete(struct ofproto *, struct rule *)
480 OVS_EXCLUDED(ofproto_mutex);
481 void ofproto_rule_reduce_timeouts(struct rule *rule, uint16_t idle_timeout,
482 uint16_t hard_timeout)
483 OVS_EXCLUDED(ofproto_mutex);
485 void ofoperation_complete(struct ofoperation *, enum ofperr);
487 bool ofoperation_has_out_port(const struct ofoperation *, ofp_port_t out_port)
488 OVS_REQUIRES(ofproto_mutex);
490 /* A group within a "struct ofproto".
492 * With few exceptions, ofproto implementations may look at these fields but
493 * should not modify them. */
495 /* The rwlock is used to prevent groups from being deleted while child
496 * threads are using them to xlate flows. A read lock means the
497 * group is currently being used. A write lock means the group is
498 * in the process of being deleted or updated. Note that since
499 * a read lock on the groups container is held while searching, and
500 * a group is ever write locked only while holding a write lock
501 * on the container, the user's of groups will never face a group
502 * in the write locked state. */
503 struct ovs_rwlock rwlock OVS_ACQ_AFTER(ofproto_mutex);
504 struct hmap_node hmap_node; /* In struct ofproto's "groups" hmap. */
505 struct ofproto *ofproto; /* The ofproto that contains this group. */
507 enum ofp11_group_type type; /* One of OFPGT_*. */
509 long long int created; /* Creation time. */
510 long long int modified; /* Time of last modification. */
512 struct list buckets; /* Contains "struct ofputil_bucket"s. */
516 bool ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
517 struct ofgroup **group)
518 OVS_TRY_RDLOCK(true, (*group)->rwlock);
520 void ofproto_group_release(struct ofgroup *group)
521 OVS_RELEASES(group->rwlock);
523 /* ofproto class structure, to be defined by each ofproto implementation.
529 * These functions work primarily with four different kinds of data
532 * - "struct ofproto", which represents an OpenFlow switch.
534 * - "struct ofport", which represents a port within an ofproto.
536 * - "struct rule", which represents an OpenFlow flow within an ofproto.
538 * - "struct ofgroup", which represents an OpenFlow 1.1+ group within an
541 * Each of these data structures contains all of the implementation-independent
542 * generic state for the respective concept, called the "base" state. None of
543 * them contains any extra space for ofproto implementations to use. Instead,
544 * each implementation is expected to declare its own data structure that
545 * contains an instance of the generic data structure plus additional
546 * implementation-specific members, called the "derived" state. The
547 * implementation can use casts or (preferably) the CONTAINER_OF macro to
548 * obtain access to derived state given only a pointer to the embedded generic
555 * Four stylized functions accompany each of these data structures:
557 * "alloc" "construct" "destruct" "dealloc"
558 * ------------ ---------------- --------------- --------------
559 * ofproto ->alloc ->construct ->destruct ->dealloc
560 * ofport ->port_alloc ->port_construct ->port_destruct ->port_dealloc
561 * rule ->rule_alloc ->rule_construct ->rule_destruct ->rule_dealloc
562 * group ->group_alloc ->group_construct ->group_destruct ->group_dealloc
564 * "ofproto", "ofport", and "group" have this exact life cycle. The "rule"
565 * data structure also follow this life cycle with some additional elaborations
566 * described under "Rule Life Cycle" below.
568 * Any instance of a given data structure goes through the following life
571 * 1. The client calls the "alloc" function to obtain raw memory. If "alloc"
572 * fails, skip all the other steps.
574 * 2. The client initializes all of the data structure's base state. If this
575 * fails, skip to step 7.
577 * 3. The client calls the "construct" function. The implementation
578 * initializes derived state. It may refer to the already-initialized
579 * base state. If "construct" fails, skip to step 6.
581 * 4. The data structure is now initialized and in use.
583 * 5. When the data structure is no longer needed, the client calls the
584 * "destruct" function. The implementation uninitializes derived state.
585 * The base state has not been uninitialized yet, so the implementation
586 * may still refer to it.
588 * 6. The client uninitializes all of the data structure's base state.
590 * 7. The client calls the "dealloc" to free the raw memory. The
591 * implementation must not refer to base or derived state in the data
592 * structure, because it has already been uninitialized.
594 * Each "alloc" function allocates and returns a new instance of the respective
595 * data structure. The "alloc" function is not given any information about the
596 * use of the new data structure, so it cannot perform much initialization.
597 * Its purpose is just to ensure that the new data structure has enough room
598 * for base and derived state. It may return a null pointer if memory is not
599 * available, in which case none of the other functions is called.
601 * Each "construct" function initializes derived state in its respective data
602 * structure. When "construct" is called, all of the base state has already
603 * been initialized, so the "construct" function may refer to it. The
604 * "construct" function is allowed to fail, in which case the client calls the
605 * "dealloc" function (but not the "destruct" function).
607 * Each "destruct" function uninitializes and frees derived state in its
608 * respective data structure. When "destruct" is called, the base state has
609 * not yet been uninitialized, so the "destruct" function may refer to it. The
610 * "destruct" function is not allowed to fail.
612 * Each "dealloc" function frees raw memory that was allocated by the the
613 * "alloc" function. The memory's base and derived members might not have ever
614 * been initialized (but if "construct" returned successfully, then it has been
615 * "destruct"ed already). The "dealloc" function is not allowed to fail.
621 * Most of these functions return 0 if they are successful or a positive error
622 * code on failure. Depending on the function, valid error codes are either
623 * errno values or OFPERR_* OpenFlow error codes.
625 * Most of these functions are expected to execute synchronously, that is, to
626 * block as necessary to obtain a result. Thus, these functions may return
627 * EAGAIN (or EWOULDBLOCK or EINPROGRESS) only where the function descriptions
628 * explicitly say those errors are a possibility. We may relax this
629 * requirement in the future if and when we encounter performance problems. */
630 struct ofproto_class {
631 /* ## ----------------- ## */
632 /* ## Factory Functions ## */
633 /* ## ----------------- ## */
635 /* Initializes provider. The caller may pass in 'iface_hints',
636 * which contains an shash of "struct iface_hint" elements indexed
637 * by the interface's name. The provider may use these hints to
638 * describe the startup configuration in order to reinitialize its
639 * state. The caller owns the provided data, so a provider must
640 * make copies of anything required. An ofproto provider must
641 * remove any existing state that is not described by the hint, and
642 * may choose to remove it all. */
643 void (*init)(const struct shash *iface_hints);
645 /* Enumerates the types of all support ofproto types into 'types'. The
646 * caller has already initialized 'types' and other ofproto classes might
647 * already have added names to it. */
648 void (*enumerate_types)(struct sset *types);
650 /* Enumerates the names of all existing datapath of the specified 'type'
651 * into 'names' 'all_dps'. The caller has already initialized 'names' as
654 * 'type' is one of the types enumerated by ->enumerate_types().
656 * Returns 0 if successful, otherwise a positive errno value.
658 int (*enumerate_names)(const char *type, struct sset *names);
660 /* Deletes the datapath with the specified 'type' and 'name'. The caller
661 * should have closed any open ofproto with this 'type' and 'name'; this
662 * function is allowed to fail if that is not the case.
664 * 'type' is one of the types enumerated by ->enumerate_types().
665 * 'name' is one of the names enumerated by ->enumerate_names() for 'type'.
667 * Returns 0 if successful, otherwise a positive errno value.
669 int (*del)(const char *type, const char *name);
671 /* Returns the type to pass to netdev_open() when a datapath of type
672 * 'datapath_type' has a port of type 'port_type', for a few special
673 * cases when a netdev type differs from a port type. For example,
674 * when using the userspace datapath, a port of type "internal"
675 * needs to be opened as "tap".
677 * Returns either 'type' itself or a string literal, which must not
679 const char *(*port_open_type)(const char *datapath_type,
680 const char *port_type);
682 /* ## ------------------------ ## */
683 /* ## Top-Level type Functions ## */
684 /* ## ------------------------ ## */
686 /* Performs any periodic activity required on ofprotos of type
689 * An ofproto provider may implement it or not, depending on whether
690 * it needs type-level maintenance.
692 * Returns 0 if successful, otherwise a positive errno value. */
693 int (*type_run)(const char *type);
695 /* Causes the poll loop to wake up when a type 'type''s 'run'
696 * function needs to be called, e.g. by calling the timer or fd
697 * waiting functions in poll-loop.h.
699 * An ofproto provider may implement it or not, depending on whether
700 * it needs type-level maintenance. */
701 void (*type_wait)(const char *type);
703 /* ## --------------------------- ## */
704 /* ## Top-Level ofproto Functions ## */
705 /* ## --------------------------- ## */
707 /* Life-cycle functions for an "ofproto" (see "Life Cycle" above).
713 * ->construct() should not modify any base members of the ofproto. The
714 * client will initialize the ofproto's 'ports' and 'tables' members after
715 * construction is complete.
717 * When ->construct() is called, the client does not yet know how many flow
718 * tables the datapath supports, so ofproto->n_tables will be 0 and
719 * ofproto->tables will be NULL. ->construct() should call
720 * ofproto_init_tables() to allocate and initialize ofproto->n_tables and
721 * ofproto->tables. Each flow table will be initially empty, so
722 * ->construct() should delete flows from the underlying datapath, if
723 * necessary, rather than populating the tables.
725 * If the ofproto knows the maximum port number that the datapath can have,
726 * then it can call ofproto_init_max_ports(). If it does so, then the
727 * client will ensure that the actions it allows to be used through
728 * OpenFlow do not refer to ports above that maximum number.
730 * Only one ofproto instance needs to be supported for any given datapath.
731 * If a datapath is already open as part of one "ofproto", then another
732 * attempt to "construct" the same datapath as part of another ofproto is
733 * allowed to fail with an error.
735 * ->construct() returns 0 if successful, otherwise a positive errno
742 * If 'ofproto' has any pending asynchronous operations, ->destruct()
743 * must complete all of them by calling ofoperation_complete().
745 * ->destruct() must also destroy all remaining rules in the ofproto's
746 * tables, by passing each remaining rule to ofproto_rule_delete(), and
747 * then complete each of those deletions in turn by calling
748 * ofoperation_complete().
750 * (Thus, there is a multi-step process for any rule currently being
751 * inserted or modified at the beginning of destruction: first
752 * ofoperation_complete() that operation, then ofproto_rule_delete() the
753 * rule, then ofoperation_complete() the deletion operation.)
755 * The client will destroy the flow tables themselves after ->destruct()
758 struct ofproto *(*alloc)(void);
759 int (*construct)(struct ofproto *ofproto);
760 void (*destruct)(struct ofproto *ofproto);
761 void (*dealloc)(struct ofproto *ofproto);
763 /* Performs any periodic activity required by 'ofproto'. It should:
765 * - Call connmgr_send_packet_in() for each received packet that missed
766 * in the OpenFlow flow table or that had a OFPP_CONTROLLER output
769 * - Call ofproto_rule_expire() for each OpenFlow flow that has reached
770 * its hard_timeout or idle_timeout, to expire the flow.
772 * (But rules that are part of a pending operation, e.g. rules for
773 * which ->pending is true, may not expire.)
775 * Returns 0 if successful, otherwise a positive errno value. */
776 int (*run)(struct ofproto *ofproto);
778 /* Causes the poll loop to wake up when 'ofproto''s 'run' function needs to
779 * be called, e.g. by calling the timer or fd waiting functions in
781 void (*wait)(struct ofproto *ofproto);
783 /* Adds some memory usage statistics for the implementation of 'ofproto'
784 * into 'usage', for use with memory_report().
786 * This function is optional. */
787 void (*get_memory_usage)(const struct ofproto *ofproto,
788 struct simap *usage);
790 /* Adds some memory usage statistics for the implementation of 'type'
791 * into 'usage', for use with memory_report().
793 * This function is optional. */
794 void (*type_get_memory_usage)(const char *type, struct simap *usage);
796 /* Every "struct rule" in 'ofproto' is about to be deleted, one by one.
797 * This function may prepare for that, for example by clearing state in
798 * advance. It should *not* actually delete any "struct rule"s from
799 * 'ofproto', only prepare for it.
801 * This function is optional; it's really just for optimization in case
802 * it's cheaper to delete all the flows from your hardware in a single pass
803 * than to do it one by one. */
804 void (*flush)(struct ofproto *ofproto);
806 /* Helper for the OpenFlow OFPT_FEATURES_REQUEST request.
808 * The implementation should store true in '*arp_match_ip' if the switch
809 * supports matching IP addresses inside ARP requests and replies, false
812 * The implementation should store in '*actions' a bitmap of the supported
813 * OpenFlow actions. Vendor actions are not included in '*actions'. */
814 void (*get_features)(struct ofproto *ofproto,
816 enum ofputil_action_bitmap *actions);
818 /* Helper for the OpenFlow OFPST_TABLE statistics request.
820 * The 'ots' array contains 'ofproto->n_tables' elements. Each element is
823 * - 'table_id' to the array index.
825 * - 'name' to "table#" where # is the table ID.
827 * - 'match' and 'wildcards' to OFPXMT12_MASK.
829 * - 'write_actions' and 'apply_actions' to OFPAT12_OUTPUT.
831 * - 'write_setfields' and 'apply_setfields' to OFPXMT12_MASK.
833 * - 'metadata_match' and 'metadata_write' to UINT64_MAX.
835 * - 'instructions' to OFPIT11_ALL.
837 * - 'config' to OFPTC11_TABLE_MISS_MASK.
839 * - 'max_entries' to 1,000,000.
841 * - 'active_count' to the classifier_count() for the table.
843 * - 'lookup_count' and 'matched_count' to 0.
845 * The implementation should update any members in each element for which
846 * it has better values:
848 * - 'name' to a more meaningful name.
850 * - 'wildcards' to the set of wildcards actually supported by the table
851 * (if it doesn't support all OpenFlow wildcards).
853 * - 'instructions' to set the instructions actually supported by
856 * - 'write_actions' to set the write actions actually supported by
857 * the table (if it doesn't support all OpenFlow actions).
859 * - 'apply_actions' to set the apply actions actually supported by
860 * the table (if it doesn't support all OpenFlow actions).
862 * - 'write_setfields' to set the write setfields actually supported by
865 * - 'apply_setfields' to set the apply setfields actually supported by
868 * - 'max_entries' to the maximum number of flows actually supported by
871 * - 'lookup_count' to the number of packets looked up in this flow table
874 * - 'matched_count' to the number of packets looked up in this flow
875 * table so far that matched one of the flow entries.
877 * All of the members of struct ofp12_table_stats are in network byte
880 void (*get_tables)(struct ofproto *ofproto, struct ofp12_table_stats *ots);
882 /* ## ---------------- ## */
883 /* ## ofport Functions ## */
884 /* ## ---------------- ## */
886 /* Life-cycle functions for a "struct ofport" (see "Life Cycle" above).
888 * ->port_construct() should not modify any base members of the ofport.
889 * An ofproto implementation should use the 'ofp_port' member of
890 * "struct ofport" as the OpenFlow port number.
892 * ofports are managed by the base ofproto code. The ofproto
893 * implementation should only create and destroy them in response to calls
894 * to these functions. The base ofproto code will create and destroy
895 * ofports in the following situations:
897 * - Just after the ->construct() function is called, the base ofproto
898 * iterates over all of the implementation's ports, using
899 * ->port_dump_start() and related functions, and constructs an ofport
900 * for each dumped port.
902 * - If ->port_poll() reports that a specific port has changed, then the
903 * base ofproto will query that port with ->port_query_by_name() and
904 * construct or destruct ofports as necessary to reflect the updated
907 * - If ->port_poll() returns ENOBUFS to report an unspecified port set
908 * change, then the base ofproto will iterate over all of the
909 * implementation's ports, in the same way as at ofproto
910 * initialization, and construct and destruct ofports to reflect all of
913 * ->port_construct() returns 0 if successful, otherwise a positive errno
916 struct ofport *(*port_alloc)(void);
917 int (*port_construct)(struct ofport *ofport);
918 void (*port_destruct)(struct ofport *ofport);
919 void (*port_dealloc)(struct ofport *ofport);
921 /* Called after 'ofport->netdev' is replaced by a new netdev object. If
922 * the ofproto implementation uses the ofport's netdev internally, then it
923 * should switch to using the new one. The old one has been closed.
925 * An ofproto implementation that doesn't need to do anything in this
926 * function may use a null pointer. */
927 void (*port_modified)(struct ofport *ofport);
929 /* Called after an OpenFlow request changes a port's configuration.
930 * 'ofport->pp.config' contains the new configuration. 'old_config'
931 * contains the previous configuration.
933 * The caller implements OFPUTIL_PC_PORT_DOWN using netdev functions to
934 * turn NETDEV_UP on and off, so this function doesn't have to do anything
935 * for that bit (and it won't be called if that is the only bit that
937 void (*port_reconfigured)(struct ofport *ofport,
938 enum ofputil_port_config old_config);
940 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
941 * initializes '*port' appropriately. Otherwise, returns a positive errno
944 * The caller owns the data in 'port' and must free it with
945 * ofproto_port_destroy() when it is no longer needed. */
946 int (*port_query_by_name)(const struct ofproto *ofproto,
947 const char *devname, struct ofproto_port *port);
949 /* Attempts to add 'netdev' as a port on 'ofproto'. Returns 0 if
950 * successful, otherwise a positive errno value. The caller should
951 * inform the implementation of the OpenFlow port through the
952 * ->port_construct() method.
954 * It doesn't matter whether the new port will be returned by a later call
955 * to ->port_poll(); the implementation may do whatever is more
957 int (*port_add)(struct ofproto *ofproto, struct netdev *netdev);
959 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'. Returns
960 * 0 if successful, otherwise a positive errno value.
962 * It doesn't matter whether the new port will be returned by a later call
963 * to ->port_poll(); the implementation may do whatever is more
965 int (*port_del)(struct ofproto *ofproto, ofp_port_t ofp_port);
968 int (*port_get_stats)(const struct ofport *port,
969 struct netdev_stats *stats);
971 /* Port iteration functions.
973 * The client might not be entirely in control of the ports within an
974 * ofproto. Some hardware implementations, for example, might have a fixed
975 * set of ports in a datapath. For this reason, the client needs a way to
976 * iterate through all the ports that are actually in a datapath. These
977 * functions provide that functionality.
979 * The 'state' pointer provides the implementation a place to
980 * keep track of its position. Its format is opaque to the caller.
982 * The ofproto provider retains ownership of the data that it stores into
983 * ->port_dump_next()'s 'port' argument. The data must remain valid until
984 * at least the next call to ->port_dump_next() or ->port_dump_done() for
985 * 'state'. The caller will not modify or free it.
990 * ->port_dump_start() attempts to begin dumping the ports in 'ofproto'.
991 * On success, it should return 0 and initialize '*statep' with any data
992 * needed for iteration. On failure, returns a positive errno value, and
993 * the client will not call ->port_dump_next() or ->port_dump_done().
995 * ->port_dump_next() attempts to retrieve another port from 'ofproto' for
996 * 'state'. If there is another port, it should store the port's
997 * information into 'port' and return 0. It should return EOF if all ports
998 * have already been iterated. Otherwise, on error, it should return a
999 * positive errno value. This function will not be called again once it
1000 * returns nonzero once for a given iteration (but the 'port_dump_done'
1001 * function will be called afterward).
1003 * ->port_dump_done() allows the implementation to release resources used
1004 * for iteration. The caller might decide to stop iteration in the middle
1005 * by calling this function before ->port_dump_next() returns nonzero.
1013 * error = ofproto->ofproto_class->port_dump_start(ofproto, &state);
1016 * struct ofproto_port port;
1018 * error = ofproto->ofproto_class->port_dump_next(
1019 * ofproto, state, &port);
1023 * // Do something with 'port' here (without modifying or freeing
1024 * // any of its data).
1026 * ofproto->ofproto_class->port_dump_done(ofproto, state);
1028 * // 'error' is now EOF (success) or a positive errno value (failure).
1030 int (*port_dump_start)(const struct ofproto *ofproto, void **statep);
1031 int (*port_dump_next)(const struct ofproto *ofproto, void *state,
1032 struct ofproto_port *port);
1033 int (*port_dump_done)(const struct ofproto *ofproto, void *state);
1035 /* Polls for changes in the set of ports in 'ofproto'. If the set of ports
1036 * in 'ofproto' has changed, then this function should do one of the
1039 * - Preferably: store the name of the device that was added to or deleted
1040 * from 'ofproto' in '*devnamep' and return 0. The caller is responsible
1041 * for freeing '*devnamep' (with free()) when it no longer needs it.
1043 * - Alternatively: return ENOBUFS, without indicating the device that was
1046 * Occasional 'false positives', in which the function returns 0 while
1047 * indicating a device that was not actually added or deleted or returns
1048 * ENOBUFS without any change, are acceptable.
1050 * The purpose of 'port_poll' is to let 'ofproto' know about changes made
1051 * externally to the 'ofproto' object, e.g. by a system administrator via
1052 * ovs-dpctl. Therefore, it's OK, and even preferable, for port_poll() to
1053 * not report changes made through calls to 'port_add' or 'port_del' on the
1054 * same 'ofproto' object. (But it's OK for it to report them too, just
1055 * slightly less efficient.)
1057 * If the set of ports in 'ofproto' has not changed, returns EAGAIN. May
1058 * also return other positive errno values to indicate that something has
1061 * If the set of ports in a datapath is fixed, or if the only way that the
1062 * set of ports in a datapath can change is through ->port_add() and
1063 * ->port_del(), then this function may be a null pointer.
1065 int (*port_poll)(const struct ofproto *ofproto, char **devnamep);
1067 /* Arranges for the poll loop to wake up when ->port_poll() will return a
1068 * value other than EAGAIN.
1070 * If the set of ports in a datapath is fixed, or if the only way that the
1071 * set of ports in a datapath can change is through ->port_add() and
1072 * ->port_del(), or if the poll loop will always wake up anyway when
1073 * ->port_poll() will return a value other than EAGAIN, then this function
1074 * may be a null pointer.
1076 void (*port_poll_wait)(const struct ofproto *ofproto);
1078 /* Checks the status of LACP negotiation for 'port'. Returns 1 if LACP
1079 * partner information for 'port' is up-to-date, 0 if LACP partner
1080 * information is not current (generally indicating a connectivity
1081 * problem), or -1 if LACP is not enabled on 'port'.
1083 * This function may be a null pointer if the ofproto implementation does
1084 * not support LACP. */
1085 int (*port_is_lacp_current)(const struct ofport *port);
1087 /* ## ----------------------- ## */
1088 /* ## OpenFlow Rule Functions ## */
1089 /* ## ----------------------- ## */
1091 /* Chooses an appropriate table for 'match' within 'ofproto'. On
1092 * success, stores the table ID into '*table_idp' and returns 0. On
1093 * failure, returns an OpenFlow error code.
1095 * The choice of table should be a function of 'match' and 'ofproto''s
1096 * datapath capabilities. It should not depend on the flows already in
1097 * 'ofproto''s flow tables. Failure implies that an OpenFlow rule with
1098 * 'match' as its matching condition can never be inserted into 'ofproto',
1099 * even starting from an empty flow table.
1101 * If multiple tables are candidates for inserting the flow, the function
1102 * should choose one arbitrarily (but deterministically).
1104 * If this function is NULL then table 0 is always chosen. */
1105 enum ofperr (*rule_choose_table)(const struct ofproto *ofproto,
1106 const struct match *match,
1107 uint8_t *table_idp);
1109 /* Life-cycle functions for a "struct rule".
1115 * The life cycle of a struct rule is an elaboration of the basic life
1116 * cycle described above under "Life Cycle".
1118 * After a rule is successfully constructed, it is then inserted. If
1119 * insertion completes successfully, then before it is later destructed, it
1122 * You can think of a rule as having the following extra steps inserted
1123 * between "Life Cycle" steps 4 and 5:
1125 * 4.1. The client inserts the rule into the flow table, making it
1126 * visible in flow table lookups.
1128 * 4.2. The client calls "rule_insert". Immediately or eventually, the
1129 * implementation calls ofoperation_complete() to indicate that the
1130 * insertion completed. If the operation failed, skip to step 5.
1132 * 4.3. The rule is now installed in the flow table. Eventually it will
1135 * 4.4. The client removes the rule from the flow table. It is no longer
1136 * visible in flow table lookups.
1138 * 4.5. The client calls "rule_delete". Immediately or eventually, the
1139 * implementation calls ofoperation_complete() to indicate that the
1140 * deletion completed. Deletion is not allowed to fail, so it must
1144 * Asynchronous Operation Support
1145 * ==============================
1147 * The "insert" and "delete" life-cycle operations on rules can operate
1148 * asynchronously, meaning that ->rule_insert() and ->rule_delete() only
1149 * need to initiate their respective operations and do not need to wait for
1150 * them to complete before they return. ->rule_modify_actions() also
1151 * operates asynchronously.
1153 * An ofproto implementation reports the success or failure of an
1154 * asynchronous operation on a rule using the rule's 'pending' member,
1155 * which points to a opaque "struct ofoperation" that represents the
1156 * ongoing operation. When the operation completes, the ofproto
1157 * implementation calls ofoperation_complete(), passing the ofoperation and
1158 * an error indication.
1160 * Only the following contexts may call ofoperation_complete():
1162 * - The function called to initiate the operation, e.g. ->rule_insert()
1163 * or ->rule_delete(). This is the best choice if the operation
1164 * completes quickly.
1166 * - The implementation's ->run() function.
1168 * - The implementation's ->destruct() function.
1170 * The ofproto base code updates the flow table optimistically, assuming
1171 * that the operation will probably succeed:
1173 * - ofproto adds the rule in the flow table before calling
1176 * - ofproto updates the rule's actions and other properties before
1177 * calling ->rule_modify_actions().
1179 * - ofproto removes the rule before calling ->rule_delete().
1181 * With one exception, when an asynchronous operation completes with an
1182 * error, ofoperation_complete() backs out the already applied changes:
1184 * - If adding a rule in the flow table fails, ofproto removes the new
1187 * - If modifying a rule fails, ofproto restores the original actions
1188 * (and other properties).
1190 * - Removing a rule is not allowed to fail. It must always succeed.
1192 * The ofproto base code serializes operations: if any operation is in
1193 * progress on a given rule, ofproto postpones initiating any new operation
1194 * on that rule until the pending operation completes. Therefore, every
1195 * operation must eventually complete through a call to
1196 * ofoperation_complete() to avoid delaying new operations indefinitely
1197 * (including any OpenFlow request that affects the rule in question, even
1198 * just to query its statistics).
1204 * When ->rule_construct() is called, 'rule' is a new rule that is not yet
1205 * inserted into a flow table. ->rule_construct() should initialize enough
1206 * of the rule's derived state for 'rule' to be suitable for inserting into
1207 * a flow table. ->rule_construct() should not modify any base members of
1210 * If ->rule_construct() fails (as indicated by returning a nonzero
1211 * OpenFlow error code), the ofproto base code will uninitialize and
1212 * deallocate 'rule'. See "Rule Life Cycle" above for more details.
1214 * ->rule_construct() may also:
1216 * - Validate that the datapath supports the matching rule in 'rule->cr'
1217 * datapath. For example, if the rule's table does not support
1218 * registers, then it is an error if 'rule->cr' does not wildcard all
1221 * - Validate that the datapath can correctly implement 'rule->ofpacts'.
1223 * Some implementations might need to defer these tasks to ->rule_insert(),
1224 * which is also acceptable.
1230 * Following successful construction, the ofproto base case inserts 'rule'
1231 * into its flow table, then it calls ->rule_insert(). ->rule_insert()
1232 * should set in motion adding the new rule to the datapath flow table. It
1233 * must act as follows:
1235 * - If it completes insertion, either by succeeding or failing, it must
1236 * call ofoperation_complete()
1238 * - If insertion is only partially complete, then it must return without
1239 * calling ofoperation_complete(). Later, when the insertion is
1240 * complete, the ->run() or ->destruct() function must call
1241 * ofoperation_complete() to report success or failure.
1243 * If ->rule_insert() fails, the ofproto base code will remove 'rule' from
1244 * the flow table, destruct, uninitialize, and deallocate 'rule'. See
1245 * "Rule Life Cycle" above for more details.
1251 * The ofproto base code removes 'rule' from its flow table before it calls
1252 * ->rule_delete(). ->rule_delete() should set in motion removing 'rule'
1253 * from the datapath flow table. It must act as follows:
1255 * - If it completes deletion, it must call ofoperation_complete().
1257 * - If deletion is only partially complete, then it must return without
1258 * calling ofoperation_complete(). Later, when the deletion is
1259 * complete, the ->run() or ->destruct() function must call
1260 * ofoperation_complete().
1262 * Rule deletion must not fail.
1268 * ->rule_destruct() must uninitialize derived state.
1270 * Rule destruction must not fail. */
1271 struct rule *(*rule_alloc)(void);
1272 enum ofperr (*rule_construct)(struct rule *rule)
1273 /* OVS_REQUIRES(ofproto_mutex) */;
1274 void (*rule_insert)(struct rule *rule) /* OVS_REQUIRES(ofproto_mutex) */;
1275 void (*rule_delete)(struct rule *rule) /* OVS_REQUIRES(ofproto_mutex) */;
1276 void (*rule_destruct)(struct rule *rule);
1277 void (*rule_dealloc)(struct rule *rule);
1279 /* Obtains statistics for 'rule', storing the number of packets that have
1280 * matched it in '*packet_count' and the number of bytes in those packets
1281 * in '*byte_count'. UINT64_MAX indicates that the packet count or byte
1282 * count is unknown. */
1283 void (*rule_get_stats)(struct rule *rule, uint64_t *packet_count,
1284 uint64_t *byte_count, long long int *used)
1285 /* OVS_EXCLUDED(ofproto_mutex) */;
1287 /* Applies the actions in 'rule' to 'packet'. (This implements sending
1288 * buffered packets for OpenFlow OFPT_FLOW_MOD commands.)
1290 * Takes ownership of 'packet' (so it should eventually free it, with
1293 * 'flow' reflects the flow information for 'packet'. All of the
1294 * information in 'flow' is extracted from 'packet', except for
1295 * flow->tunnel and flow->in_port, which are assigned the correct values
1296 * for the incoming packet. The register values are zeroed. 'packet''s
1297 * header pointers (e.g. packet->l3) are appropriately initialized.
1298 * packet->l3 is aligned on a 32-bit boundary.
1300 * The implementation should add the statistics for 'packet' into 'rule'.
1302 * Returns 0 if successful, otherwise an OpenFlow error code. */
1303 enum ofperr (*rule_execute)(struct rule *rule, const struct flow *flow,
1304 struct ofpbuf *packet);
1306 /* When ->rule_modify_actions() is called, the caller has already replaced
1307 * the OpenFlow actions in 'rule' by a new set. (The original actions are
1308 * in rule->pending->actions.)
1310 * ->rule_modify_actions() should set the following in motion:
1312 * - Validate that the datapath can correctly implement the actions now
1315 * - Update the datapath flow table with the new actions.
1317 * - Only if 'reset_counters' is true, reset any packet or byte counters
1318 * associated with the rule to zero, so that rule_get_stats() will not
1319 * longer count those packets or bytes.
1321 * If the operation synchronously completes, ->rule_modify_actions() may
1322 * call ofoperation_complete() before it returns. Otherwise, ->run()
1323 * should call ofoperation_complete() later, after the operation does
1326 * If the operation fails, then the base ofproto code will restore the
1327 * original 'actions' and 'n_actions' of 'rule'.
1329 * ->rule_modify_actions() should not modify any base members of struct
1331 void (*rule_modify_actions)(struct rule *rule, bool reset_counters)
1332 /* OVS_REQUIRES(ofproto_mutex) */;
1334 /* Changes the OpenFlow IP fragment handling policy to 'frag_handling',
1335 * which takes one of the following values, with the corresponding
1338 * - OFPC_FRAG_NORMAL: The switch should treat IP fragments the same way
1339 * as other packets, omitting TCP and UDP port numbers (always setting
1342 * - OFPC_FRAG_DROP: The switch should drop all IP fragments without
1343 * passing them through the flow table.
1345 * - OFPC_FRAG_REASM: The switch should reassemble IP fragments before
1346 * passing packets through the flow table.
1348 * - OFPC_FRAG_NX_MATCH (a Nicira extension): Similar to OFPC_FRAG_NORMAL,
1349 * except that TCP and UDP port numbers should be included in fragments
1352 * Implementations are not required to support every mode.
1353 * OFPC_FRAG_NORMAL is the default mode when an ofproto is created.
1355 * At the time of the call to ->set_frag_handling(), the current mode is
1356 * available in 'ofproto->frag_handling'. ->set_frag_handling() returns
1357 * true if the requested mode was set, false if it is not supported.
1359 * Upon successful return, the caller changes 'ofproto->frag_handling' to
1360 * reflect the new mode.
1362 bool (*set_frag_handling)(struct ofproto *ofproto,
1363 enum ofp_config_flags frag_handling);
1365 /* Implements the OpenFlow OFPT_PACKET_OUT command. The datapath should
1366 * execute the 'ofpacts_len' bytes of "struct ofpacts" in 'ofpacts'.
1368 * The caller retains ownership of 'packet' and of 'ofpacts', so
1369 * ->packet_out() should not modify or free them.
1371 * This function must validate that it can correctly implement 'ofpacts'.
1372 * If not, then it should return an OpenFlow error code.
1374 * 'flow' reflects the flow information for 'packet'. All of the
1375 * information in 'flow' is extracted from 'packet', except for
1376 * flow->in_port (see below). flow->tunnel and its register values are
1379 * flow->in_port comes from the OpenFlow OFPT_PACKET_OUT message. The
1380 * implementation should reject invalid flow->in_port values by returning
1381 * OFPERR_OFPBRC_BAD_PORT. (If the implementation called
1382 * ofproto_init_max_ports(), then the client will reject these ports
1383 * itself.) For consistency, the implementation should consider valid for
1384 * flow->in_port any value that could possibly be seen in a packet that it
1385 * passes to connmgr_send_packet_in(). Ideally, even an implementation
1386 * that never generates packet-ins (e.g. due to hardware limitations)
1387 * should still allow flow->in_port values for every possible physical port
1388 * and OFPP_LOCAL. The only virtual ports (those above OFPP_MAX) that the
1389 * caller will ever pass in as flow->in_port, other than OFPP_LOCAL, are
1390 * OFPP_NONE and OFPP_CONTROLLER. The implementation should allow both of
1391 * these, treating each of them as packets generated by the controller as
1392 * opposed to packets originating from some switch port.
1394 * (Ordinarily the only effect of flow->in_port is on output actions that
1395 * involve the input port, such as actions that output to OFPP_IN_PORT,
1396 * OFPP_FLOOD, or OFPP_ALL. flow->in_port can also affect Nicira extension
1397 * "resubmit" actions.)
1399 * 'packet' is not matched against the OpenFlow flow table, so its
1400 * statistics should not be included in OpenFlow flow statistics.
1402 * Returns 0 if successful, otherwise an OpenFlow error code. */
1403 enum ofperr (*packet_out)(struct ofproto *ofproto, struct ofpbuf *packet,
1404 const struct flow *flow,
1405 const struct ofpact *ofpacts,
1406 size_t ofpacts_len);
1408 /* ## ------------------------- ## */
1409 /* ## OFPP_NORMAL configuration ## */
1410 /* ## ------------------------- ## */
1412 /* Configures NetFlow on 'ofproto' according to the options in
1413 * 'netflow_options', or turns off NetFlow if 'netflow_options' is NULL.
1415 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
1416 * NetFlow, as does a null pointer. */
1417 int (*set_netflow)(struct ofproto *ofproto,
1418 const struct netflow_options *netflow_options);
1420 void (*get_netflow_ids)(const struct ofproto *ofproto,
1421 uint8_t *engine_type, uint8_t *engine_id);
1423 /* Configures sFlow on 'ofproto' according to the options in
1424 * 'sflow_options', or turns off sFlow if 'sflow_options' is NULL.
1426 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
1427 * sFlow, as does a null pointer. */
1428 int (*set_sflow)(struct ofproto *ofproto,
1429 const struct ofproto_sflow_options *sflow_options);
1431 /* Configures IPFIX on 'ofproto' according to the options in
1432 * 'bridge_exporter_options' and the 'flow_exporters_options'
1433 * array, or turns off IPFIX if 'bridge_exporter_options' and
1434 * 'flow_exporters_options' is NULL.
1436 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
1437 * IPFIX, as does a null pointer. */
1439 struct ofproto *ofproto,
1440 const struct ofproto_ipfix_bridge_exporter_options
1441 *bridge_exporter_options,
1442 const struct ofproto_ipfix_flow_exporter_options
1443 *flow_exporters_options, size_t n_flow_exporters_options);
1445 /* Configures connectivity fault management on 'ofport'.
1447 * If 'cfm_settings' is nonnull, configures CFM according to its members.
1449 * If 'cfm_settings' is null, removes any connectivity fault management
1450 * configuration from 'ofport'.
1452 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1453 * support CFM, as does a null pointer. */
1454 int (*set_cfm)(struct ofport *ofport, const struct cfm_settings *s);
1456 /* Checks the status of CFM configured on 'ofport'. Returns true if the
1457 * port's CFM status was successfully stored into '*status'. Returns false
1458 * if the port did not have CFM configured, in which case '*status' is
1461 * The caller must provide and owns '*status', but it does not own and must
1462 * not modify or free the array returned in 'status->rmps'. */
1463 bool (*get_cfm_status)(const struct ofport *ofport,
1464 struct ofproto_cfm_status *status);
1466 /* Configures BFD on 'ofport'.
1468 * If 'cfg' is NULL, or 'cfg' does not contain the key value pair
1469 * "enable=true", removes BFD from 'ofport'. Otherwise, configures BFD
1470 * according to 'cfg'.
1472 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1473 * support BFD, as does a null pointer. */
1474 int (*set_bfd)(struct ofport *ofport, const struct smap *cfg);
1476 /* Populates 'smap' with the status of BFD on 'ofport'. Returns 0 on
1477 * success, or a positive errno. EOPNOTSUPP as a return value indicates
1478 * that this ofproto_class does not support BFD, as does a null pointer. */
1479 int (*get_bfd_status)(struct ofport *ofport, struct smap *smap);
1481 /* Configures spanning tree protocol (STP) on 'ofproto' using the
1482 * settings defined in 's'.
1484 * If 's' is nonnull, configures STP according to its members.
1486 * If 's' is null, removes any STP configuration from 'ofproto'.
1488 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1489 * support STP, as does a null pointer. */
1490 int (*set_stp)(struct ofproto *ofproto,
1491 const struct ofproto_stp_settings *s);
1493 /* Retrieves state of spanning tree protocol (STP) on 'ofproto'.
1495 * Stores STP state for 'ofproto' in 's'. If the 'enabled' member
1496 * is false, the other member values are not meaningful.
1498 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1499 * support STP, as does a null pointer. */
1500 int (*get_stp_status)(struct ofproto *ofproto,
1501 struct ofproto_stp_status *s);
1503 /* Configures spanning tree protocol (STP) on 'ofport' using the
1504 * settings defined in 's'.
1506 * If 's' is nonnull, configures STP according to its members. The
1507 * caller is responsible for assigning STP port numbers (using the
1508 * 'port_num' member in the range of 1 through 255, inclusive) and
1509 * ensuring there are no duplicates.
1511 * If 's' is null, removes any STP configuration from 'ofport'.
1513 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1514 * support STP, as does a null pointer. */
1515 int (*set_stp_port)(struct ofport *ofport,
1516 const struct ofproto_port_stp_settings *s);
1518 /* Retrieves spanning tree protocol (STP) port status of 'ofport'.
1520 * Stores STP state for 'ofport' in 's'. If the 'enabled' member is
1521 * false, the other member values are not meaningful.
1523 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1524 * support STP, as does a null pointer. */
1525 int (*get_stp_port_status)(struct ofport *ofport,
1526 struct ofproto_port_stp_status *s);
1528 /* Retrieves spanning tree protocol (STP) port statistics of 'ofport'.
1530 * Stores STP state for 'ofport' in 's'. If the 'enabled' member is
1531 * false, the other member values are not meaningful.
1533 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1534 * support STP, as does a null pointer. */
1535 int (*get_stp_port_stats)(struct ofport *ofport,
1536 struct ofproto_port_stp_stats *s);
1538 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
1539 * 'queues' attached to 'ofport'. This data is not intended to be
1540 * sufficient to implement QoS. Instead, providers may use this
1541 * information to implement features which require knowledge of what queues
1542 * exist on a port, and some basic information about them.
1544 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1545 * support QoS, as does a null pointer. */
1546 int (*set_queues)(struct ofport *ofport,
1547 const struct ofproto_port_queue *queues, size_t n_qdscp);
1549 /* If 's' is nonnull, this function registers a "bundle" associated with
1550 * client data pointer 'aux' in 'ofproto'. A bundle is the same concept as
1551 * a Port in OVSDB, that is, it consists of one or more "slave" devices
1552 * (Interfaces, in OVSDB) along with VLAN and LACP configuration and, if
1553 * there is more than one slave, a bonding configuration. If 'aux' is
1554 * already registered then this function updates its configuration to 's'.
1555 * Otherwise, this function registers a new bundle.
1557 * If 's' is NULL, this function unregisters the bundle registered on
1558 * 'ofproto' associated with client data pointer 'aux'. If no such bundle
1559 * has been registered, this has no effect.
1561 * This function affects only the behavior of the NXAST_AUTOPATH action and
1562 * output to the OFPP_NORMAL port. An implementation that does not support
1563 * it at all may set it to NULL or return EOPNOTSUPP. An implementation
1564 * that supports only a subset of the functionality should implement what
1565 * it can and return 0. */
1566 int (*bundle_set)(struct ofproto *ofproto, void *aux,
1567 const struct ofproto_bundle_settings *s);
1569 /* If 'port' is part of any bundle, removes it from that bundle. If the
1570 * bundle now has no ports, deletes the bundle. If the bundle now has only
1571 * one port, deconfigures the bundle's bonding configuration. */
1572 void (*bundle_remove)(struct ofport *ofport);
1574 /* If 's' is nonnull, this function registers a mirror associated with
1575 * client data pointer 'aux' in 'ofproto'. A mirror is the same concept as
1576 * a Mirror in OVSDB. If 'aux' is already registered then this function
1577 * updates its configuration to 's'. Otherwise, this function registers a
1580 * If 's' is NULL, this function unregisters the mirror registered on
1581 * 'ofproto' associated with client data pointer 'aux'. If no such mirror
1582 * has been registered, this has no effect.
1584 * An implementation that does not support mirroring at all may set
1585 * it to NULL or return EOPNOTSUPP. An implementation that supports
1586 * only a subset of the functionality should implement what it can
1588 int (*mirror_set)(struct ofproto *ofproto, void *aux,
1589 const struct ofproto_mirror_settings *s);
1591 /* Retrieves statistics from mirror associated with client data
1592 * pointer 'aux' in 'ofproto'. Stores packet and byte counts in
1593 * 'packets' and 'bytes', respectively. If a particular counter is
1594 * not supported, the appropriate argument is set to UINT64_MAX.
1596 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1597 * support retrieving mirror statistics. */
1598 int (*mirror_get_stats)(struct ofproto *ofproto, void *aux,
1599 uint64_t *packets, uint64_t *bytes);
1601 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs
1602 * on which all packets are flooded, instead of using MAC learning. If
1603 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1605 * This function affects only the behavior of the OFPP_NORMAL action. An
1606 * implementation that does not support it may set it to NULL or return
1608 int (*set_flood_vlans)(struct ofproto *ofproto,
1609 unsigned long *flood_vlans);
1611 /* Returns true if 'aux' is a registered bundle that is currently in use as
1612 * the output for a mirror. */
1613 bool (*is_mirror_output_bundle)(const struct ofproto *ofproto, void *aux);
1615 /* When the configuration option of forward_bpdu changes, this function
1616 * will be invoked. */
1617 void (*forward_bpdu_changed)(struct ofproto *ofproto);
1619 /* Sets the MAC aging timeout for the OFPP_NORMAL action to 'idle_time', in
1620 * seconds, and the maximum number of MAC table entries to
1623 * An implementation that doesn't support configuring these features may
1624 * set this function to NULL or implement it as a no-op. */
1625 void (*set_mac_table_config)(struct ofproto *ofproto,
1626 unsigned int idle_time, size_t max_entries);
1628 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
1630 * This is deprecated. It is only for compatibility with broken device drivers
1631 * in old versions of Linux that do not properly support VLANs when VLAN
1632 * devices are not used. When broken device drivers are no longer in
1633 * widespread use, we will delete these interfaces. */
1635 /* If 'realdev_ofp_port' is nonzero, then this function configures 'ofport'
1636 * as a VLAN splinter port for VLAN 'vid', associated with the real device
1637 * that has OpenFlow port number 'realdev_ofp_port'.
1639 * If 'realdev_ofp_port' is zero, then this function deconfigures 'ofport'
1640 * as a VLAN splinter port.
1642 * This function should be NULL if an implementation does not support it.
1644 int (*set_realdev)(struct ofport *ofport,
1645 ofp_port_t realdev_ofp_port, int vid);
1647 /* ## ------------------------ ## */
1648 /* ## OpenFlow meter functions ## */
1649 /* ## ------------------------ ## */
1651 /* These functions should be NULL if an implementation does not support
1652 * them. They must be all null or all non-null.. */
1654 /* Initializes 'features' to describe the metering features supported by
1656 void (*meter_get_features)(const struct ofproto *ofproto,
1657 struct ofputil_meter_features *features);
1659 /* If '*id' is UINT32_MAX, adds a new meter with the given 'config'. On
1660 * success the function must store a provider meter ID other than
1661 * UINT32_MAX in '*id'. All further references to the meter will be made
1662 * with the returned provider meter id rather than the OpenFlow meter id.
1663 * The caller does not try to interpret the provider meter id, giving the
1664 * implementation the freedom to either use the OpenFlow meter_id value
1665 * provided in the meter configuration, or any other value suitable for the
1668 * If '*id' is a value other than UINT32_MAX, modifies the existing meter
1669 * with that meter provider ID to have configuration 'config', while
1670 * leaving '*id' unchanged. On failure, the existing meter configuration
1671 * is left intact. */
1672 enum ofperr (*meter_set)(struct ofproto *ofproto, ofproto_meter_id *id,
1673 const struct ofputil_meter_config *config);
1675 /* Gets the meter and meter band packet and byte counts for maximum of
1676 * 'stats->n_bands' bands for the meter with provider ID 'id' within
1677 * 'ofproto'. The caller fills in the other stats values. The band stats
1678 * are copied to memory at 'stats->bands' provided by the caller. The
1679 * number of returned band stats is returned in 'stats->n_bands'. */
1680 enum ofperr (*meter_get)(const struct ofproto *ofproto,
1681 ofproto_meter_id id,
1682 struct ofputil_meter_stats *stats);
1684 /* Deletes a meter, making the 'ofproto_meter_id' invalid for any
1686 void (*meter_del)(struct ofproto *, ofproto_meter_id);
1689 /* ## -------------------- ## */
1690 /* ## OpenFlow 1.1+ groups ## */
1691 /* ## -------------------- ## */
1693 struct ofgroup *(*group_alloc)(void);
1694 enum ofperr (*group_construct)(struct ofgroup *);
1695 void (*group_destruct)(struct ofgroup *);
1696 void (*group_dealloc)(struct ofgroup *);
1698 enum ofperr (*group_modify)(struct ofgroup *, struct ofgroup *victim);
1700 enum ofperr (*group_get_stats)(const struct ofgroup *,
1701 struct ofputil_group_stats *);
1704 extern const struct ofproto_class ofproto_dpif_class;
1706 int ofproto_class_register(const struct ofproto_class *);
1707 int ofproto_class_unregister(const struct ofproto_class *);
1709 /* ofproto_flow_mod() returns this value if the flow_mod could not be processed
1710 * because it overlaps with an ongoing flow table operation that has not yet
1711 * completed. The caller should retry the operation later.
1713 * ofproto.c also uses this value internally for additional (similar) purposes.
1715 * This particular value is a good choice because it is large, so that it does
1716 * not collide with any errno value, but not large enough to collide with an
1717 * OFPERR_* value. */
1718 enum { OFPROTO_POSTPONE = 1 << 16 };
1719 BUILD_ASSERT_DECL(OFPROTO_POSTPONE < OFPERR_OFS);
1721 int ofproto_flow_mod(struct ofproto *, struct ofputil_flow_mod *)
1722 OVS_EXCLUDED(ofproto_mutex);
1723 void ofproto_add_flow(struct ofproto *, const struct match *,
1724 unsigned int priority,
1725 const struct ofpact *ofpacts, size_t ofpacts_len)
1726 OVS_EXCLUDED(ofproto_mutex);
1727 bool ofproto_delete_flow(struct ofproto *,
1728 const struct match *, unsigned int priority)
1729 OVS_EXCLUDED(ofproto_mutex);
1730 void ofproto_flush_flows(struct ofproto *);
1732 #endif /* ofproto/ofproto-provider.h */