2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 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-actions.h"
43 #include "ofp-errors.h"
45 #include "ofproto/ofproto.h"
46 #include "ovs-atomic.h"
48 #include "ovs-thread.h"
54 struct ofputil_flow_mod;
59 extern struct ovs_mutex ofproto_mutex;
61 /* An OpenFlow switch.
63 * With few exceptions, ofproto implementations may look at these fields but
64 * should not modify them. */
66 struct hmap_node hmap_node; /* In global 'all_ofprotos' hmap. */
67 const struct ofproto_class *ofproto_class;
68 char *type; /* Datapath type. */
69 char *name; /* Datapath name. */
72 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
73 uint64_t datapath_id; /* Datapath ID. */
74 bool forward_bpdu; /* Option to allow forwarding of BPDU frames
75 * when NORMAL action is invoked. */
76 char *mfr_desc; /* Manufacturer (NULL for default). */
77 char *hw_desc; /* Hardware (NULL for default). */
78 char *sw_desc; /* Software version (NULL for default). */
79 char *serial_desc; /* Serial number (NULL for default). */
80 char *dp_desc; /* Datapath description (NULL for default). */
81 enum ofp_config_flags frag_handling; /* One of OFPC_*. */
84 struct hmap ports; /* Contains "struct ofport"s. */
85 struct shash port_by_name;
86 struct simap ofp_requests; /* OpenFlow port number requests. */
87 uint16_t alloc_port_no; /* Last allocated OpenFlow port number. */
88 uint16_t max_ports; /* Max possible OpenFlow port num, plus one. */
89 struct hmap ofport_usage; /* Map ofport to last used time. */
90 uint64_t change_seq; /* Change sequence for netdev status. */
93 long long int eviction_group_timer; /* For rate limited reheapification. */
94 struct oftable *tables;
97 /* Rules indexed on their cookie values, in all flow tables. */
98 struct hindex cookies OVS_GUARDED_BY(ofproto_mutex);
100 /* List of expirable flows, in all flow tables. */
101 struct list expirable OVS_GUARDED_BY(ofproto_mutex);
104 * OpenFlow meters start at 1. To avoid confusion we leave the first
105 * pointer in the array un-used, and index directly with the OpenFlow
107 struct ofputil_meter_features meter_features;
108 struct meter **meters; /* 'meter_features.max_meter' + 1 pointers. */
110 /* OpenFlow connections. */
111 struct connmgr *connmgr;
113 /* 'meaningful only within ofproto.c, one of the enum ofproto_state
114 * constants defined there. */
117 /* Delayed rule executions.
119 * We delay calls to ->ofproto_class->rule_execute() past releasing
120 * ofproto_mutex during a flow_mod, because otherwise a "learn" action
121 * triggered by the executing the packet would try to recursively modify
122 * the flow table and reacquire the global lock. */
123 struct guarded_list rule_executes; /* Contains "struct rule_execute"s. */
125 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
127 * This is deprecated. It is only for compatibility with broken device
128 * drivers in old versions of Linux that do not properly support VLANs when
129 * VLAN devices are not used. When broken device drivers are no longer in
130 * widespread use, we will delete these interfaces. */
131 unsigned long int *vlan_bitmap; /* 4096-bit bitmap of in-use VLANs. */
132 bool vlans_changed; /* True if new VLANs are in use. */
133 int min_mtu; /* Current MTU of non-internal ports. */
136 struct ovs_rwlock groups_rwlock;
137 struct hmap groups OVS_GUARDED; /* Contains "struct ofgroup"s. */
138 uint32_t n_groups[4] OVS_GUARDED; /* # of existing groups of each type. */
139 struct ofputil_group_features ogf;
142 void ofproto_init_tables(struct ofproto *, int n_tables);
143 void ofproto_init_max_ports(struct ofproto *, uint16_t max_ports);
145 struct ofproto *ofproto_lookup(const char *name);
146 struct ofport *ofproto_get_port(const struct ofproto *, ofp_port_t ofp_port);
148 /* An OpenFlow port within a "struct ofproto".
150 * The port's name is netdev_get_name(port->netdev).
152 * With few exceptions, ofproto implementations may look at these fields but
153 * should not modify them. */
155 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
156 struct ofproto *ofproto; /* The ofproto that contains this port. */
157 struct netdev *netdev;
158 struct ofputil_phy_port pp;
159 ofp_port_t ofp_port; /* OpenFlow port number. */
161 long long int created; /* Time created, in msec. */
165 void ofproto_port_set_state(struct ofport *, enum ofputil_port_state);
167 /* OpenFlow table flags:
169 * - "Hidden" tables are not included in OpenFlow operations that operate on
170 * "all tables". For example, a request for flow stats on all tables will
171 * omit flows in hidden tables, table stats requests will omit the table
172 * entirely, and the switch features reply will not count the hidden table.
174 * However, operations that specifically name the particular table still
175 * operate on it. For example, flow_mods and flow stats requests on a
178 * To avoid gaps in table IDs (which have unclear validity in OpenFlow),
179 * hidden tables must be the highest-numbered tables that a provider
182 * - "Read-only" tables can't be changed through OpenFlow operations. (At
183 * the moment all flow table operations go effectively through OpenFlow, so
184 * this means that read-only tables can't be changed at all after the
185 * read-only flag is set.)
187 * The generic ofproto layer never sets these flags. An ofproto provider can
188 * set them if it is appropriate.
191 OFTABLE_HIDDEN = 1 << 0, /* Hide from most OpenFlow operations. */
192 OFTABLE_READONLY = 1 << 1 /* Don't allow OpenFlow controller to change
196 /* A flow table within a "struct ofproto".
202 * A cls->rwlock read-lock holder prevents rules from being added or deleted.
204 * Adding or removing rules requires holding ofproto_mutex AND the cls->rwlock
207 * cls->rwlock should be held only briefly. For extended access to a rule,
208 * increment its ref_count with ofproto_rule_ref(). A rule will not be freed
209 * until its ref_count reaches zero.
211 * Modifying a rule requires the rule's own mutex. Holding cls->rwlock (for
212 * read or write) does not allow the holder to modify the rule.
214 * Freeing a rule requires ofproto_mutex and the cls->rwlock write-lock. After
215 * removing the rule from the classifier, release a ref_count from the rule
216 * ('cls''s reference to the rule).
218 * Refer to the thread-safety notes on struct rule for more information.*/
220 enum oftable_flags flags;
221 struct classifier cls; /* Contains "struct rule"s. */
222 char *name; /* Table name exposed via OpenFlow, or NULL. */
224 /* Maximum number of flows or UINT_MAX if there is no limit besides any
225 * limit imposed by resource limitations. */
226 unsigned int max_flows;
228 /* These members determine the handling of an attempt to add a flow that
229 * would cause the table to have more than 'max_flows' flows.
231 * If 'eviction_fields' is NULL, overflows will be rejected with an error.
233 * If 'eviction_fields' is nonnull (regardless of whether n_eviction_fields
234 * is nonzero), an overflow will cause a flow to be removed. The flow to
235 * be removed is chosen to give fairness among groups distinguished by
236 * different values for the subfields within 'groups'. */
237 struct mf_subfield *eviction_fields;
238 size_t n_eviction_fields;
242 * When a flow is added that would cause the table to have more than
243 * 'max_flows' flows, and 'eviction_fields' is nonnull, these groups are
244 * used to decide which rule to evict: the rule is chosen from the eviction
245 * group that contains the greatest number of rules.*/
246 uint32_t eviction_group_id_basis;
247 struct hmap eviction_groups_by_id;
248 struct heap eviction_groups_by_size;
250 /* Table config: contains enum ofproto_table_config; accessed atomically. */
253 atomic_ulong n_matched;
254 atomic_ulong n_missed;
257 /* Assigns TABLE to each oftable, in turn, in OFPROTO.
259 * All parameters are evaluated multiple times. */
260 #define OFPROTO_FOR_EACH_TABLE(TABLE, OFPROTO) \
261 for ((TABLE) = (OFPROTO)->tables; \
262 (TABLE) < &(OFPROTO)->tables[(OFPROTO)->n_tables]; \
265 /* An OpenFlow flow within a "struct ofproto".
267 * With few exceptions, ofproto implementations may look at these fields but
268 * should not modify them.
274 * Except near the beginning or ending of its lifespan, rule 'rule' belongs to
275 * the classifier rule->ofproto->tables[rule->table_id].cls. The text below
276 * calls this classifier 'cls'.
281 * The thread safety rules described here for "struct rule" are motivated by
284 * - Prevent threads that read members of "struct rule" from reading bad
285 * data due to changes by some thread concurrently modifying those
288 * - Prevent two threads making changes to members of a given "struct rule"
289 * from interfering with each other.
295 * A rule 'rule' may be accessed without a risk of being freed by code that
296 * holds a read-lock or write-lock on 'cls->rwlock' or that owns a reference to
297 * 'rule->ref_count' (or both). Code that needs to hold onto a rule for a
298 * while should take 'cls->rwlock', find the rule it needs, increment
299 * 'rule->ref_count' with ofproto_rule_ref(), and drop 'cls->rwlock'.
301 * 'rule->ref_count' protects 'rule' from being freed. It doesn't protect the
302 * rule from being deleted from 'cls' (that's 'cls->rwlock') and it doesn't
303 * protect members of 'rule' from modification (that's 'rule->mutex').
305 * 'rule->mutex' protects the members of 'rule' from modification. It doesn't
306 * protect the rule from being deleted from 'cls' (that's 'cls->rwlock') and it
307 * doesn't prevent the rule from being freed (that's 'rule->ref_count').
309 * Regarding thread safety, the members of a rule fall into the following
312 * - Immutable. These members are marked 'const'.
314 * - Members that may be safely read or written only by code holding
315 * ofproto_mutex. These are marked OVS_GUARDED_BY(ofproto_mutex).
317 * - Members that may be safely read only by code holding ofproto_mutex or
318 * 'rule->mutex', and safely written only by coding holding ofproto_mutex
319 * AND 'rule->mutex'. These are marked OVS_GUARDED.
322 /* Where this rule resides in an OpenFlow switch.
324 * These are immutable once the rule is constructed, hence 'const'. */
325 struct ofproto *const ofproto; /* The ofproto that contains this rule. */
326 const struct cls_rule cr; /* In owning ofproto's classifier. */
327 const uint8_t table_id; /* Index in ofproto's 'tables' array. */
329 /* Protects members marked OVS_GUARDED.
330 * Readers only need to hold this mutex.
331 * Writers must hold both this mutex AND ofproto_mutex.
332 * By implication writers can read *without* taking this mutex while they
333 * hold ofproto_mutex. */
334 struct ovs_mutex mutex OVS_ACQ_AFTER(ofproto_mutex);
336 /* Number of references.
337 * The classifier owns one reference.
338 * Any thread trying to keep a rule from being freed should hold its own
340 struct ovs_refcount ref_count;
342 /* A "flow cookie" is the OpenFlow name for a 64-bit value associated with
344 ovs_be64 flow_cookie OVS_GUARDED;
345 struct hindex_node cookie_node OVS_GUARDED_BY(ofproto_mutex);
347 enum ofputil_flow_mod_flags flags OVS_GUARDED;
350 uint16_t hard_timeout OVS_GUARDED; /* In seconds from ->modified. */
351 uint16_t idle_timeout OVS_GUARDED; /* In seconds from ->used. */
353 /* Eviction groups (see comment on struct eviction_group for explanation) .
355 * 'eviction_group' is this rule's eviction group, or NULL if it is not in
356 * any eviction group. When 'eviction_group' is nonnull, 'evg_node' is in
357 * the ->eviction_group->rules hmap. */
358 struct eviction_group *eviction_group OVS_GUARDED_BY(ofproto_mutex);
359 struct heap_node evg_node OVS_GUARDED_BY(ofproto_mutex);
361 /* OpenFlow actions. See struct rule_actions for more thread-safety
363 OVSRCU_TYPE(const struct rule_actions *) actions;
365 /* In owning meter's 'rules' list. An empty list if there is no meter. */
366 struct list meter_list_node OVS_GUARDED_BY(ofproto_mutex);
368 /* Flow monitors (e.g. for NXST_FLOW_MONITOR, related to struct ofmonitor).
370 * 'add_seqno' is the sequence number when this rule was created.
371 * 'modify_seqno' is the sequence number when this rule was last modified.
372 * See 'monitor_seqno' in connmgr.c for more information. */
373 enum nx_flow_monitor_flags monitor_flags OVS_GUARDED_BY(ofproto_mutex);
374 uint64_t add_seqno OVS_GUARDED_BY(ofproto_mutex);
375 uint64_t modify_seqno OVS_GUARDED_BY(ofproto_mutex);
377 /* Optimisation for flow expiry. In ofproto's 'expirable' list if this
378 * rule is expirable, otherwise empty. */
379 struct list expirable OVS_GUARDED_BY(ofproto_mutex);
381 /* Times. Last so that they are more likely close to the stats managed
382 * by the provider. */
383 long long int created OVS_GUARDED; /* Creation time. */
385 /* Must hold 'mutex' for both read/write, 'ofproto_mutex' not needed. */
386 long long int modified OVS_GUARDED; /* Time of last modification. */
389 void ofproto_rule_ref(struct rule *);
390 void ofproto_rule_unref(struct rule *);
392 static inline const struct rule_actions * rule_get_actions(const struct rule *);
393 static inline bool rule_is_table_miss(const struct rule *);
394 static inline bool rule_is_hidden(const struct rule *);
396 /* A set of actions within a "struct rule".
402 * A struct rule_actions may be accessed without a risk of being
403 * freed by code that holds a read-lock or write-lock on 'rule->mutex' (where
404 * 'rule' is the rule for which 'rule->actions == actions') or during the RCU
406 struct rule_actions {
407 /* These members are immutable: they do not change during the struct's
409 uint32_t ofpacts_len; /* Size of 'ofpacts', in bytes. */
410 uint32_t provider_meter_id; /* Datapath meter_id, or UINT32_MAX. */
411 struct ofpact ofpacts[]; /* Sequence of "struct ofpacts". */
413 BUILD_ASSERT_DECL(offsetof(struct rule_actions, ofpacts) % OFPACT_ALIGNTO == 0);
415 const struct rule_actions *rule_actions_create(const struct ofproto *,
416 const struct ofpact *, size_t);
417 void rule_actions_destroy(const struct rule_actions *);
419 /* A set of rules to which an OpenFlow operation applies. */
420 struct rule_collection {
421 struct rule **rules; /* The rules. */
422 size_t n; /* Number of rules collected. */
424 size_t capacity; /* Number of rules that will fit in 'rules'. */
425 struct rule *stub[64]; /* Preallocated rules to avoid malloc(). */
428 void rule_collection_init(struct rule_collection *);
429 void rule_collection_add(struct rule_collection *, struct rule *);
430 void rule_collection_ref(struct rule_collection *) OVS_REQUIRES(ofproto_mutex);
431 void rule_collection_unref(struct rule_collection *);
432 void rule_collection_destroy(struct rule_collection *);
434 /* Limits the number of flows allowed in the datapath. Only affects the
435 * ofproto-dpif implementation. */
436 extern unsigned ofproto_flow_limit;
438 /* Maximum idle time (in ms) for flows to be cached in the datapath.
439 * Revalidators may expire flows more quickly than the configured value based
440 * on system load and other factors. This variable is subject to change. */
441 extern unsigned ofproto_max_idle;
443 /* Number of upcall handler and revalidator threads. Only affects the
444 * ofproto-dpif implementation. */
445 extern size_t n_handlers, n_revalidators;
447 static inline struct rule *rule_from_cls_rule(const struct cls_rule *);
449 void ofproto_rule_expire(struct rule *rule, uint8_t reason)
450 OVS_REQUIRES(ofproto_mutex);
451 void ofproto_rule_delete(struct ofproto *, struct rule *)
452 OVS_EXCLUDED(ofproto_mutex);
453 void ofproto_rule_reduce_timeouts(struct rule *rule, uint16_t idle_timeout,
454 uint16_t hard_timeout)
455 OVS_EXCLUDED(ofproto_mutex);
457 void ofoperation_complete(struct ofoperation *, enum ofperr);
459 /* A group within a "struct ofproto".
461 * With few exceptions, ofproto implementations may look at these fields but
462 * should not modify them. */
464 struct hmap_node hmap_node OVS_GUARDED; /* In ofproto's "groups" hmap. */
466 /* Number of references.
468 * This is needed to keep track of references to the group in the xlate
471 * If the main thread removes the group from an ofproto, we need to
472 * guarantee that the group remains accessible to users of
473 * xlate_group_actions and the xlate_cache, as the xlate_cache will not be
474 * cleaned up until the corresponding datapath flows are revalidated. */
475 struct ovs_refcount ref_count;
477 /* No lock is needed to protect the fields below since they are not
478 * modified after construction. */
479 const struct ofproto *ofproto; /* The ofproto that contains this group. */
480 const uint32_t group_id;
481 const enum ofp11_group_type type; /* One of OFPGT_*. */
483 const long long int created; /* Creation time. */
484 const long long int modified; /* Time of last modification. */
486 struct list buckets; /* Contains "struct ofputil_bucket"s. */
487 const uint32_t n_buckets;
490 bool ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
491 struct ofgroup **group);
493 void ofproto_group_ref(struct ofgroup *);
494 void ofproto_group_unref(struct ofgroup *);
496 /* ofproto class structure, to be defined by each ofproto implementation.
502 * These functions work primarily with four different kinds of data
505 * - "struct ofproto", which represents an OpenFlow switch.
507 * - "struct ofport", which represents a port within an ofproto.
509 * - "struct rule", which represents an OpenFlow flow within an ofproto.
511 * - "struct ofgroup", which represents an OpenFlow 1.1+ group within an
514 * Each of these data structures contains all of the implementation-independent
515 * generic state for the respective concept, called the "base" state. None of
516 * them contains any extra space for ofproto implementations to use. Instead,
517 * each implementation is expected to declare its own data structure that
518 * contains an instance of the generic data structure plus additional
519 * implementation-specific members, called the "derived" state. The
520 * implementation can use casts or (preferably) the CONTAINER_OF macro to
521 * obtain access to derived state given only a pointer to the embedded generic
528 * Four stylized functions accompany each of these data structures:
530 * "alloc" "construct" "destruct" "dealloc"
531 * ------------ ---------------- --------------- --------------
532 * ofproto ->alloc ->construct ->destruct ->dealloc
533 * ofport ->port_alloc ->port_construct ->port_destruct ->port_dealloc
534 * rule ->rule_alloc ->rule_construct ->rule_destruct ->rule_dealloc
535 * group ->group_alloc ->group_construct ->group_destruct ->group_dealloc
537 * "ofproto", "ofport", and "group" have this exact life cycle. The "rule"
538 * data structure also follow this life cycle with some additional elaborations
539 * described under "Rule Life Cycle" below.
541 * Any instance of a given data structure goes through the following life
544 * 1. The client calls the "alloc" function to obtain raw memory. If "alloc"
545 * fails, skip all the other steps.
547 * 2. The client initializes all of the data structure's base state. If this
548 * fails, skip to step 7.
550 * 3. The client calls the "construct" function. The implementation
551 * initializes derived state. It may refer to the already-initialized
552 * base state. If "construct" fails, skip to step 6.
554 * 4. The data structure is now initialized and in use.
556 * 5. When the data structure is no longer needed, the client calls the
557 * "destruct" function. The implementation uninitializes derived state.
558 * The base state has not been uninitialized yet, so the implementation
559 * may still refer to it.
561 * 6. The client uninitializes all of the data structure's base state.
563 * 7. The client calls the "dealloc" to free the raw memory. The
564 * implementation must not refer to base or derived state in the data
565 * structure, because it has already been uninitialized.
567 * Each "alloc" function allocates and returns a new instance of the respective
568 * data structure. The "alloc" function is not given any information about the
569 * use of the new data structure, so it cannot perform much initialization.
570 * Its purpose is just to ensure that the new data structure has enough room
571 * for base and derived state. It may return a null pointer if memory is not
572 * available, in which case none of the other functions is called.
574 * Each "construct" function initializes derived state in its respective data
575 * structure. When "construct" is called, all of the base state has already
576 * been initialized, so the "construct" function may refer to it. The
577 * "construct" function is allowed to fail, in which case the client calls the
578 * "dealloc" function (but not the "destruct" function).
580 * Each "destruct" function uninitializes and frees derived state in its
581 * respective data structure. When "destruct" is called, the base state has
582 * not yet been uninitialized, so the "destruct" function may refer to it. The
583 * "destruct" function is not allowed to fail.
585 * Each "dealloc" function frees raw memory that was allocated by the the
586 * "alloc" function. The memory's base and derived members might not have ever
587 * been initialized (but if "construct" returned successfully, then it has been
588 * "destruct"ed already). The "dealloc" function is not allowed to fail.
594 * Most of these functions return 0 if they are successful or a positive error
595 * code on failure. Depending on the function, valid error codes are either
596 * errno values or OFPERR_* OpenFlow error codes.
598 * Most of these functions are expected to execute synchronously, that is, to
599 * block as necessary to obtain a result. Thus, these functions may return
600 * EAGAIN (or EWOULDBLOCK or EINPROGRESS) only where the function descriptions
601 * explicitly say those errors are a possibility. We may relax this
602 * requirement in the future if and when we encounter performance problems. */
603 struct ofproto_class {
604 /* ## ----------------- ## */
605 /* ## Factory Functions ## */
606 /* ## ----------------- ## */
608 /* Initializes provider. The caller may pass in 'iface_hints',
609 * which contains an shash of "struct iface_hint" elements indexed
610 * by the interface's name. The provider may use these hints to
611 * describe the startup configuration in order to reinitialize its
612 * state. The caller owns the provided data, so a provider must
613 * make copies of anything required. An ofproto provider must
614 * remove any existing state that is not described by the hint, and
615 * may choose to remove it all. */
616 void (*init)(const struct shash *iface_hints);
618 /* Enumerates the types of all support ofproto types into 'types'. The
619 * caller has already initialized 'types' and other ofproto classes might
620 * already have added names to it. */
621 void (*enumerate_types)(struct sset *types);
623 /* Enumerates the names of all existing datapath of the specified 'type'
624 * into 'names' 'all_dps'. The caller has already initialized 'names' as
627 * 'type' is one of the types enumerated by ->enumerate_types().
629 * Returns 0 if successful, otherwise a positive errno value.
631 int (*enumerate_names)(const char *type, struct sset *names);
633 /* Deletes the datapath with the specified 'type' and 'name'. The caller
634 * should have closed any open ofproto with this 'type' and 'name'; this
635 * function is allowed to fail if that is not the case.
637 * 'type' is one of the types enumerated by ->enumerate_types().
638 * 'name' is one of the names enumerated by ->enumerate_names() for 'type'.
640 * Returns 0 if successful, otherwise a positive errno value.
642 int (*del)(const char *type, const char *name);
644 /* Returns the type to pass to netdev_open() when a datapath of type
645 * 'datapath_type' has a port of type 'port_type', for a few special
646 * cases when a netdev type differs from a port type. For example,
647 * when using the userspace datapath, a port of type "internal"
648 * needs to be opened as "tap".
650 * Returns either 'type' itself or a string literal, which must not
652 const char *(*port_open_type)(const char *datapath_type,
653 const char *port_type);
655 /* ## ------------------------ ## */
656 /* ## Top-Level type Functions ## */
657 /* ## ------------------------ ## */
659 /* Performs any periodic activity required on ofprotos of type
662 * An ofproto provider may implement it or not, depending on whether
663 * it needs type-level maintenance.
665 * Returns 0 if successful, otherwise a positive errno value. */
666 int (*type_run)(const char *type);
668 /* Causes the poll loop to wake up when a type 'type''s 'run'
669 * function needs to be called, e.g. by calling the timer or fd
670 * waiting functions in poll-loop.h.
672 * An ofproto provider may implement it or not, depending on whether
673 * it needs type-level maintenance. */
674 void (*type_wait)(const char *type);
676 /* ## --------------------------- ## */
677 /* ## Top-Level ofproto Functions ## */
678 /* ## --------------------------- ## */
680 /* Life-cycle functions for an "ofproto" (see "Life Cycle" above).
686 * ->construct() should not modify any base members of the ofproto. The
687 * client will initialize the ofproto's 'ports' and 'tables' members after
688 * construction is complete.
690 * When ->construct() is called, the client does not yet know how many flow
691 * tables the datapath supports, so ofproto->n_tables will be 0 and
692 * ofproto->tables will be NULL. ->construct() should call
693 * ofproto_init_tables() to allocate and initialize ofproto->n_tables and
694 * ofproto->tables. Each flow table will be initially empty, so
695 * ->construct() should delete flows from the underlying datapath, if
696 * necessary, rather than populating the tables.
698 * If the ofproto knows the maximum port number that the datapath can have,
699 * then it can call ofproto_init_max_ports(). If it does so, then the
700 * client will ensure that the actions it allows to be used through
701 * OpenFlow do not refer to ports above that maximum number.
703 * Only one ofproto instance needs to be supported for any given datapath.
704 * If a datapath is already open as part of one "ofproto", then another
705 * attempt to "construct" the same datapath as part of another ofproto is
706 * allowed to fail with an error.
708 * ->construct() returns 0 if successful, otherwise a positive errno
715 * If 'ofproto' has any pending asynchronous operations, ->destruct()
716 * must complete all of them by calling ofoperation_complete().
718 * ->destruct() must also destroy all remaining rules in the ofproto's
719 * tables, by passing each remaining rule to ofproto_rule_delete(), and
720 * then complete each of those deletions in turn by calling
721 * ofoperation_complete().
723 * (Thus, there is a multi-step process for any rule currently being
724 * inserted or modified at the beginning of destruction: first
725 * ofoperation_complete() that operation, then ofproto_rule_delete() the
726 * rule, then ofoperation_complete() the deletion operation.)
728 * The client will destroy the flow tables themselves after ->destruct()
731 struct ofproto *(*alloc)(void);
732 int (*construct)(struct ofproto *ofproto);
733 void (*destruct)(struct ofproto *ofproto);
734 void (*dealloc)(struct ofproto *ofproto);
736 /* Performs any periodic activity required by 'ofproto'. It should:
738 * - Call connmgr_send_packet_in() for each received packet that missed
739 * in the OpenFlow flow table or that had a OFPP_CONTROLLER output
742 * - Call ofproto_rule_expire() for each OpenFlow flow that has reached
743 * its hard_timeout or idle_timeout, to expire the flow.
745 * (But rules that are part of a pending operation, e.g. rules for
746 * which ->pending is true, may not expire.)
748 * Returns 0 if successful, otherwise a positive errno value. */
749 int (*run)(struct ofproto *ofproto);
751 /* Causes the poll loop to wake up when 'ofproto''s 'run' function needs to
752 * be called, e.g. by calling the timer or fd waiting functions in
754 void (*wait)(struct ofproto *ofproto);
756 /* Adds some memory usage statistics for the implementation of 'ofproto'
757 * into 'usage', for use with memory_report().
759 * This function is optional. */
760 void (*get_memory_usage)(const struct ofproto *ofproto,
761 struct simap *usage);
763 /* Adds some memory usage statistics for the implementation of 'type'
764 * into 'usage', for use with memory_report().
766 * This function is optional. */
767 void (*type_get_memory_usage)(const char *type, struct simap *usage);
769 /* Every "struct rule" in 'ofproto' is about to be deleted, one by one.
770 * This function may prepare for that, for example by clearing state in
771 * advance. It should *not* actually delete any "struct rule"s from
772 * 'ofproto', only prepare for it.
774 * This function is optional; it's really just for optimization in case
775 * it's cheaper to delete all the flows from your hardware in a single pass
776 * than to do it one by one. */
777 void (*flush)(struct ofproto *ofproto);
779 /* Helper for the OpenFlow OFPT_FEATURES_REQUEST request.
781 * The implementation should store true in '*arp_match_ip' if the switch
782 * supports matching IP addresses inside ARP requests and replies, false
785 * The implementation should store in '*actions' a bitmap of the supported
786 * OpenFlow actions. Vendor actions are not included in '*actions'. */
787 void (*get_features)(struct ofproto *ofproto,
789 enum ofputil_action_bitmap *actions);
791 /* Helper for the OpenFlow OFPST_TABLE statistics request.
793 * The 'ots' array contains 'ofproto->n_tables' elements. Each element is
796 * - 'table_id' to the array index.
798 * - 'name' to "table#" where # is the table ID.
800 * - 'match' and 'wildcards' to OFPXMT12_MASK.
802 * - 'write_actions' and 'apply_actions' to OFPAT12_OUTPUT.
804 * - 'write_setfields' and 'apply_setfields' to OFPXMT12_MASK.
806 * - 'metadata_match' and 'metadata_write' to OVS_BE64_MAX.
808 * - 'instructions' to OFPIT11_ALL.
810 * - 'config' to OFPTC11_TABLE_MISS_MASK.
812 * - 'max_entries' to 1,000,000.
814 * - 'active_count' to the classifier_count() for the table.
816 * - 'lookup_count' and 'matched_count' to 0.
818 * The implementation should update any members in each element for which
819 * it has better values:
821 * - 'name' to a more meaningful name.
823 * - 'wildcards' to the set of wildcards actually supported by the table
824 * (if it doesn't support all OpenFlow wildcards).
826 * - 'instructions' to set the instructions actually supported by
829 * - 'write_actions' to set the write actions actually supported by
830 * the table (if it doesn't support all OpenFlow actions).
832 * - 'apply_actions' to set the apply actions actually supported by
833 * the table (if it doesn't support all OpenFlow actions).
835 * - 'write_setfields' to set the write setfields actually supported by
838 * - 'apply_setfields' to set the apply setfields actually supported by
841 * - 'max_entries' to the maximum number of flows actually supported by
844 * - 'lookup_count' to the number of packets looked up in this flow table
847 * - 'matched_count' to the number of packets looked up in this flow
848 * table so far that matched one of the flow entries.
850 * All of the members of struct ofp12_table_stats are in network byte
853 void (*get_tables)(struct ofproto *ofproto, struct ofp12_table_stats *ots);
855 /* ## ---------------- ## */
856 /* ## ofport Functions ## */
857 /* ## ---------------- ## */
859 /* Life-cycle functions for a "struct ofport" (see "Life Cycle" above).
861 * ->port_construct() should not modify any base members of the ofport.
862 * An ofproto implementation should use the 'ofp_port' member of
863 * "struct ofport" as the OpenFlow port number.
865 * ofports are managed by the base ofproto code. The ofproto
866 * implementation should only create and destroy them in response to calls
867 * to these functions. The base ofproto code will create and destroy
868 * ofports in the following situations:
870 * - Just after the ->construct() function is called, the base ofproto
871 * iterates over all of the implementation's ports, using
872 * ->port_dump_start() and related functions, and constructs an ofport
873 * for each dumped port.
875 * - If ->port_poll() reports that a specific port has changed, then the
876 * base ofproto will query that port with ->port_query_by_name() and
877 * construct or destruct ofports as necessary to reflect the updated
880 * - If ->port_poll() returns ENOBUFS to report an unspecified port set
881 * change, then the base ofproto will iterate over all of the
882 * implementation's ports, in the same way as at ofproto
883 * initialization, and construct and destruct ofports to reflect all of
886 * ->port_construct() returns 0 if successful, otherwise a positive errno
889 struct ofport *(*port_alloc)(void);
890 int (*port_construct)(struct ofport *ofport);
891 void (*port_destruct)(struct ofport *ofport);
892 void (*port_dealloc)(struct ofport *ofport);
894 /* Called after 'ofport->netdev' is replaced by a new netdev object. If
895 * the ofproto implementation uses the ofport's netdev internally, then it
896 * should switch to using the new one. The old one has been closed.
898 * An ofproto implementation that doesn't need to do anything in this
899 * function may use a null pointer. */
900 void (*port_modified)(struct ofport *ofport);
902 /* Called after an OpenFlow request changes a port's configuration.
903 * 'ofport->pp.config' contains the new configuration. 'old_config'
904 * contains the previous configuration.
906 * The caller implements OFPUTIL_PC_PORT_DOWN using netdev functions to
907 * turn NETDEV_UP on and off, so this function doesn't have to do anything
908 * for that bit (and it won't be called if that is the only bit that
910 void (*port_reconfigured)(struct ofport *ofport,
911 enum ofputil_port_config old_config);
913 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
914 * initializes '*port' appropriately. Otherwise, returns a positive errno
917 * The caller owns the data in 'port' and must free it with
918 * ofproto_port_destroy() when it is no longer needed. */
919 int (*port_query_by_name)(const struct ofproto *ofproto,
920 const char *devname, struct ofproto_port *port);
922 /* Attempts to add 'netdev' as a port on 'ofproto'. Returns 0 if
923 * successful, otherwise a positive errno value. The caller should
924 * inform the implementation of the OpenFlow port through the
925 * ->port_construct() method.
927 * It doesn't matter whether the new port will be returned by a later call
928 * to ->port_poll(); the implementation may do whatever is more
930 int (*port_add)(struct ofproto *ofproto, struct netdev *netdev);
932 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'. Returns
933 * 0 if successful, otherwise a positive errno value.
935 * It doesn't matter whether the new port will be returned by a later call
936 * to ->port_poll(); the implementation may do whatever is more
938 int (*port_del)(struct ofproto *ofproto, ofp_port_t ofp_port);
941 int (*port_get_stats)(const struct ofport *port,
942 struct netdev_stats *stats);
944 /* Port iteration functions.
946 * The client might not be entirely in control of the ports within an
947 * ofproto. Some hardware implementations, for example, might have a fixed
948 * set of ports in a datapath. For this reason, the client needs a way to
949 * iterate through all the ports that are actually in a datapath. These
950 * functions provide that functionality.
952 * The 'state' pointer provides the implementation a place to
953 * keep track of its position. Its format is opaque to the caller.
955 * The ofproto provider retains ownership of the data that it stores into
956 * ->port_dump_next()'s 'port' argument. The data must remain valid until
957 * at least the next call to ->port_dump_next() or ->port_dump_done() for
958 * 'state'. The caller will not modify or free it.
963 * ->port_dump_start() attempts to begin dumping the ports in 'ofproto'.
964 * On success, it should return 0 and initialize '*statep' with any data
965 * needed for iteration. On failure, returns a positive errno value, and
966 * the client will not call ->port_dump_next() or ->port_dump_done().
968 * ->port_dump_next() attempts to retrieve another port from 'ofproto' for
969 * 'state'. If there is another port, it should store the port's
970 * information into 'port' and return 0. It should return EOF if all ports
971 * have already been iterated. Otherwise, on error, it should return a
972 * positive errno value. This function will not be called again once it
973 * returns nonzero once for a given iteration (but the 'port_dump_done'
974 * function will be called afterward).
976 * ->port_dump_done() allows the implementation to release resources used
977 * for iteration. The caller might decide to stop iteration in the middle
978 * by calling this function before ->port_dump_next() returns nonzero.
986 * error = ofproto->ofproto_class->port_dump_start(ofproto, &state);
989 * struct ofproto_port port;
991 * error = ofproto->ofproto_class->port_dump_next(
992 * ofproto, state, &port);
996 * // Do something with 'port' here (without modifying or freeing
997 * // any of its data).
999 * ofproto->ofproto_class->port_dump_done(ofproto, state);
1001 * // 'error' is now EOF (success) or a positive errno value (failure).
1003 int (*port_dump_start)(const struct ofproto *ofproto, void **statep);
1004 int (*port_dump_next)(const struct ofproto *ofproto, void *state,
1005 struct ofproto_port *port);
1006 int (*port_dump_done)(const struct ofproto *ofproto, void *state);
1008 /* Polls for changes in the set of ports in 'ofproto'. If the set of ports
1009 * in 'ofproto' has changed, then this function should do one of the
1012 * - Preferably: store the name of the device that was added to or deleted
1013 * from 'ofproto' in '*devnamep' and return 0. The caller is responsible
1014 * for freeing '*devnamep' (with free()) when it no longer needs it.
1016 * - Alternatively: return ENOBUFS, without indicating the device that was
1019 * Occasional 'false positives', in which the function returns 0 while
1020 * indicating a device that was not actually added or deleted or returns
1021 * ENOBUFS without any change, are acceptable.
1023 * The purpose of 'port_poll' is to let 'ofproto' know about changes made
1024 * externally to the 'ofproto' object, e.g. by a system administrator via
1025 * ovs-dpctl. Therefore, it's OK, and even preferable, for port_poll() to
1026 * not report changes made through calls to 'port_add' or 'port_del' on the
1027 * same 'ofproto' object. (But it's OK for it to report them too, just
1028 * slightly less efficient.)
1030 * If the set of ports in 'ofproto' has not changed, returns EAGAIN. May
1031 * also return other positive errno values to indicate that something has
1034 * If the set of ports in a datapath is fixed, or if the only way that the
1035 * set of ports in a datapath can change is through ->port_add() and
1036 * ->port_del(), then this function may be a null pointer.
1038 int (*port_poll)(const struct ofproto *ofproto, char **devnamep);
1040 /* Arranges for the poll loop to wake up when ->port_poll() will return a
1041 * value other than EAGAIN.
1043 * If the set of ports in a datapath is fixed, or if the only way that the
1044 * set of ports in a datapath can change is through ->port_add() and
1045 * ->port_del(), or if the poll loop will always wake up anyway when
1046 * ->port_poll() will return a value other than EAGAIN, then this function
1047 * may be a null pointer.
1049 void (*port_poll_wait)(const struct ofproto *ofproto);
1051 /* Checks the status of LACP negotiation for 'port'. Returns 1 if LACP
1052 * partner information for 'port' is up-to-date, 0 if LACP partner
1053 * information is not current (generally indicating a connectivity
1054 * problem), or -1 if LACP is not enabled on 'port'.
1056 * This function may be a null pointer if the ofproto implementation does
1057 * not support LACP. */
1058 int (*port_is_lacp_current)(const struct ofport *port);
1060 /* ## ----------------------- ## */
1061 /* ## OpenFlow Rule Functions ## */
1062 /* ## ----------------------- ## */
1064 /* Chooses an appropriate table for 'match' within 'ofproto'. On
1065 * success, stores the table ID into '*table_idp' and returns 0. On
1066 * failure, returns an OpenFlow error code.
1068 * The choice of table should be a function of 'match' and 'ofproto''s
1069 * datapath capabilities. It should not depend on the flows already in
1070 * 'ofproto''s flow tables. Failure implies that an OpenFlow rule with
1071 * 'match' as its matching condition can never be inserted into 'ofproto',
1072 * even starting from an empty flow table.
1074 * If multiple tables are candidates for inserting the flow, the function
1075 * should choose one arbitrarily (but deterministically).
1077 * If this function is NULL then table 0 is always chosen. */
1078 enum ofperr (*rule_choose_table)(const struct ofproto *ofproto,
1079 const struct match *match,
1080 uint8_t *table_idp);
1082 /* Life-cycle functions for a "struct rule".
1088 * The life cycle of a struct rule is an elaboration of the basic life
1089 * cycle described above under "Life Cycle".
1091 * After a rule is successfully constructed, it is then inserted. If
1092 * insertion completes successfully, then before it is later destructed, it
1095 * You can think of a rule as having the following extra steps inserted
1096 * between "Life Cycle" steps 4 and 5:
1098 * 4.1. The client inserts the rule into the flow table, making it
1099 * visible in flow table lookups.
1101 * 4.2. The client calls "rule_insert". Immediately or eventually, the
1102 * implementation calls ofoperation_complete() to indicate that the
1103 * insertion completed. If the operation failed, skip to step 5.
1105 * 4.3. The rule is now installed in the flow table. Eventually it will
1108 * 4.4. The client removes the rule from the flow table. It is no longer
1109 * visible in flow table lookups.
1111 * 4.5. The client calls "rule_delete". Immediately or eventually, the
1112 * implementation calls ofoperation_complete() to indicate that the
1113 * deletion completed. Deletion is not allowed to fail, so it must
1117 * Asynchronous Operation Support
1118 * ==============================
1120 * The "insert" and "delete" life-cycle operations on rules can operate
1121 * asynchronously, meaning that ->rule_insert() and ->rule_delete() only
1122 * need to initiate their respective operations and do not need to wait for
1123 * them to complete before they return. ->rule_modify_actions() also
1124 * operates asynchronously.
1126 * An ofproto implementation reports the success or failure of an
1127 * asynchronous operation on a rule using the rule's 'pending' member,
1128 * which points to a opaque "struct ofoperation" that represents the
1129 * ongoing operation. When the operation completes, the ofproto
1130 * implementation calls ofoperation_complete(), passing the ofoperation and
1131 * an error indication.
1133 * Only the following contexts may call ofoperation_complete():
1135 * - The function called to initiate the operation, e.g. ->rule_insert()
1136 * or ->rule_delete(). This is the best choice if the operation
1137 * completes quickly.
1139 * - The implementation's ->run() function.
1141 * - The implementation's ->destruct() function.
1143 * The ofproto base code updates the flow table optimistically, assuming
1144 * that the operation will probably succeed:
1146 * - ofproto adds the rule in the flow table before calling
1147 * ->rule_insert(). If adding a rule in the flow table fails, ofproto
1148 * removes the new rule in the call to ofoperation_complete().
1150 * - ofproto updates the rule's actions and other properties before
1151 * calling ->rule_modify_actions(). Modifying a rule is not allowed to
1152 * fail (but ->rule_premodify_actions() can prevent the modification
1153 * attempt in advance).
1155 * - ofproto removes the rule before calling ->rule_delete(). Removing a
1156 * rule is not allowed to fail. It must always succeed.
1158 * The ofproto base code serializes operations: if any operation is in
1159 * progress on a given rule, ofproto postpones initiating any new operation
1160 * on that rule until the pending operation completes. Therefore, every
1161 * operation must eventually complete through a call to
1162 * ofoperation_complete() to avoid delaying new operations indefinitely
1163 * (including any OpenFlow request that affects the rule in question, even
1164 * just to query its statistics).
1170 * When ->rule_construct() is called, 'rule' is a new rule that is not yet
1171 * inserted into a flow table. ->rule_construct() should initialize enough
1172 * of the rule's derived state for 'rule' to be suitable for inserting into
1173 * a flow table. ->rule_construct() should not modify any base members of
1176 * If ->rule_construct() fails (as indicated by returning a nonzero
1177 * OpenFlow error code), the ofproto base code will uninitialize and
1178 * deallocate 'rule'. See "Rule Life Cycle" above for more details.
1180 * ->rule_construct() may also:
1182 * - Validate that the datapath supports the matching rule in 'rule->cr'
1183 * datapath. For example, if the rule's table does not support
1184 * registers, then it is an error if 'rule->cr' does not wildcard all
1187 * - Validate that the datapath can correctly implement 'rule->ofpacts'.
1189 * Some implementations might need to defer these tasks to ->rule_insert(),
1190 * which is also acceptable.
1196 * Following successful construction, the ofproto base case inserts 'rule'
1197 * into its flow table, then it calls ->rule_insert(). ->rule_insert()
1198 * should set in motion adding the new rule to the datapath flow table. It
1199 * must act as follows:
1201 * - If it completes insertion, either by succeeding or failing, it must
1202 * call ofoperation_complete()
1204 * - If insertion is only partially complete, then it must return without
1205 * calling ofoperation_complete(). Later, when the insertion is
1206 * complete, the ->run() or ->destruct() function must call
1207 * ofoperation_complete() to report success or failure.
1209 * If ->rule_insert() fails, the ofproto base code will remove 'rule' from
1210 * the flow table, destruct, uninitialize, and deallocate 'rule'. See
1211 * "Rule Life Cycle" above for more details.
1217 * The ofproto base code removes 'rule' from its flow table before it calls
1218 * ->rule_delete(). ->rule_delete() should set in motion removing 'rule'
1219 * from the datapath flow table. It must act as follows:
1221 * - If it completes deletion, it must call ofoperation_complete().
1223 * - If deletion is only partially complete, then it must return without
1224 * calling ofoperation_complete(). Later, when the deletion is
1225 * complete, the ->run() or ->destruct() function must call
1226 * ofoperation_complete().
1228 * Rule deletion must not fail.
1234 * ->rule_destruct() must uninitialize derived state.
1236 * Rule destruction must not fail. */
1237 struct rule *(*rule_alloc)(void);
1238 enum ofperr (*rule_construct)(struct rule *rule)
1239 /* OVS_REQUIRES(ofproto_mutex) */;
1240 enum ofperr (*rule_insert)(struct rule *rule)
1241 /* OVS_REQUIRES(ofproto_mutex) */;
1242 void (*rule_delete)(struct rule *rule) /* OVS_REQUIRES(ofproto_mutex) */;
1243 void (*rule_destruct)(struct rule *rule);
1244 void (*rule_dealloc)(struct rule *rule);
1246 /* Obtains statistics for 'rule', storing the number of packets that have
1247 * matched it in '*packet_count' and the number of bytes in those packets
1248 * in '*byte_count'. UINT64_MAX indicates that the packet count or byte
1249 * count is unknown. */
1250 void (*rule_get_stats)(struct rule *rule, uint64_t *packet_count,
1251 uint64_t *byte_count, long long int *used)
1252 /* OVS_EXCLUDED(ofproto_mutex) */;
1254 /* Applies the actions in 'rule' to 'packet'. (This implements sending
1255 * buffered packets for OpenFlow OFPT_FLOW_MOD commands.)
1257 * Takes ownership of 'packet' (so it should eventually free it, with
1260 * 'flow' reflects the flow information for 'packet'. All of the
1261 * information in 'flow' is extracted from 'packet', except for
1262 * flow->tunnel and flow->in_port, which are assigned the correct values
1263 * for the incoming packet. The register values are zeroed. 'packet''s
1264 * header pointers and offsets (e.g. packet->l3) are appropriately
1265 * initialized. packet->l3 is aligned on a 32-bit boundary.
1267 * The implementation should add the statistics for 'packet' into 'rule'.
1269 * Returns 0 if successful, otherwise an OpenFlow error code. */
1270 enum ofperr (*rule_execute)(struct rule *rule, const struct flow *flow,
1271 struct ofpbuf *packet);
1273 /* If the datapath can properly implement changing 'rule''s actions to the
1274 * 'ofpacts_len' bytes in 'ofpacts', returns 0. Otherwise, returns an enum
1275 * ofperr indicating why the new actions wouldn't work.
1277 * May be a null pointer if any set of actions is acceptable. */
1278 enum ofperr (*rule_premodify_actions)(const struct rule *rule,
1279 const struct ofpact *ofpacts,
1281 /* OVS_REQUIRES(ofproto_mutex) */;
1283 /* When ->rule_modify_actions() is called, the caller has already replaced
1284 * the OpenFlow actions in 'rule' by a new set. (The original actions are
1285 * in rule->pending->actions.)
1287 * ->rule_modify_actions() should set the following in motion:
1289 * - Update the datapath flow table with the new actions.
1291 * - Only if 'reset_counters' is true, reset any packet or byte counters
1292 * associated with the rule to zero, so that rule_get_stats() will not
1293 * longer count those packets or bytes.
1295 * If the operation synchronously completes, ->rule_modify_actions() may
1296 * call ofoperation_complete() before it returns. Otherwise, ->run()
1297 * should call ofoperation_complete() later, after the operation does
1300 * Rule modification must not fail.
1302 * ->rule_modify_actions() should not modify any base members of struct
1304 void (*rule_modify_actions)(struct rule *rule, bool reset_counters)
1305 /* OVS_REQUIRES(ofproto_mutex) */;
1307 /* Changes the OpenFlow IP fragment handling policy to 'frag_handling',
1308 * which takes one of the following values, with the corresponding
1311 * - OFPC_FRAG_NORMAL: The switch should treat IP fragments the same way
1312 * as other packets, omitting TCP and UDP port numbers (always setting
1315 * - OFPC_FRAG_DROP: The switch should drop all IP fragments without
1316 * passing them through the flow table.
1318 * - OFPC_FRAG_REASM: The switch should reassemble IP fragments before
1319 * passing packets through the flow table.
1321 * - OFPC_FRAG_NX_MATCH (a Nicira extension): Similar to OFPC_FRAG_NORMAL,
1322 * except that TCP and UDP port numbers should be included in fragments
1325 * Implementations are not required to support every mode.
1326 * OFPC_FRAG_NORMAL is the default mode when an ofproto is created.
1328 * At the time of the call to ->set_frag_handling(), the current mode is
1329 * available in 'ofproto->frag_handling'. ->set_frag_handling() returns
1330 * true if the requested mode was set, false if it is not supported.
1332 * Upon successful return, the caller changes 'ofproto->frag_handling' to
1333 * reflect the new mode.
1335 bool (*set_frag_handling)(struct ofproto *ofproto,
1336 enum ofp_config_flags frag_handling);
1338 /* Implements the OpenFlow OFPT_PACKET_OUT command. The datapath should
1339 * execute the 'ofpacts_len' bytes of "struct ofpacts" in 'ofpacts'.
1341 * The caller retains ownership of 'packet' and of 'ofpacts', so
1342 * ->packet_out() should not modify or free them.
1344 * This function must validate that it can correctly implement 'ofpacts'.
1345 * If not, then it should return an OpenFlow error code.
1347 * 'flow' reflects the flow information for 'packet'. All of the
1348 * information in 'flow' is extracted from 'packet', except for
1349 * flow->in_port (see below). flow->tunnel and its register values are
1352 * flow->in_port comes from the OpenFlow OFPT_PACKET_OUT message. The
1353 * implementation should reject invalid flow->in_port values by returning
1354 * OFPERR_OFPBRC_BAD_PORT. (If the implementation called
1355 * ofproto_init_max_ports(), then the client will reject these ports
1356 * itself.) For consistency, the implementation should consider valid for
1357 * flow->in_port any value that could possibly be seen in a packet that it
1358 * passes to connmgr_send_packet_in(). Ideally, even an implementation
1359 * that never generates packet-ins (e.g. due to hardware limitations)
1360 * should still allow flow->in_port values for every possible physical port
1361 * and OFPP_LOCAL. The only virtual ports (those above OFPP_MAX) that the
1362 * caller will ever pass in as flow->in_port, other than OFPP_LOCAL, are
1363 * OFPP_NONE and OFPP_CONTROLLER. The implementation should allow both of
1364 * these, treating each of them as packets generated by the controller as
1365 * opposed to packets originating from some switch port.
1367 * (Ordinarily the only effect of flow->in_port is on output actions that
1368 * involve the input port, such as actions that output to OFPP_IN_PORT,
1369 * OFPP_FLOOD, or OFPP_ALL. flow->in_port can also affect Nicira extension
1370 * "resubmit" actions.)
1372 * 'packet' is not matched against the OpenFlow flow table, so its
1373 * statistics should not be included in OpenFlow flow statistics.
1375 * Returns 0 if successful, otherwise an OpenFlow error code. */
1376 enum ofperr (*packet_out)(struct ofproto *ofproto, struct ofpbuf *packet,
1377 const struct flow *flow,
1378 const struct ofpact *ofpacts,
1379 size_t ofpacts_len);
1381 /* ## ------------------------- ## */
1382 /* ## OFPP_NORMAL configuration ## */
1383 /* ## ------------------------- ## */
1385 /* Configures NetFlow on 'ofproto' according to the options in
1386 * 'netflow_options', or turns off NetFlow if 'netflow_options' is NULL.
1388 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
1389 * NetFlow, as does a null pointer. */
1390 int (*set_netflow)(struct ofproto *ofproto,
1391 const struct netflow_options *netflow_options);
1393 void (*get_netflow_ids)(const struct ofproto *ofproto,
1394 uint8_t *engine_type, uint8_t *engine_id);
1396 /* Configures sFlow on 'ofproto' according to the options in
1397 * 'sflow_options', or turns off sFlow if 'sflow_options' is NULL.
1399 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
1400 * sFlow, as does a null pointer. */
1401 int (*set_sflow)(struct ofproto *ofproto,
1402 const struct ofproto_sflow_options *sflow_options);
1404 /* Configures IPFIX on 'ofproto' according to the options in
1405 * 'bridge_exporter_options' and the 'flow_exporters_options'
1406 * array, or turns off IPFIX if 'bridge_exporter_options' and
1407 * 'flow_exporters_options' is NULL.
1409 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
1410 * IPFIX, as does a null pointer. */
1412 struct ofproto *ofproto,
1413 const struct ofproto_ipfix_bridge_exporter_options
1414 *bridge_exporter_options,
1415 const struct ofproto_ipfix_flow_exporter_options
1416 *flow_exporters_options, size_t n_flow_exporters_options);
1418 /* Configures connectivity fault management on 'ofport'.
1420 * If 'cfm_settings' is nonnull, configures CFM according to its members.
1422 * If 'cfm_settings' is null, removes any connectivity fault management
1423 * configuration from 'ofport'.
1425 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1426 * support CFM, as does a null pointer. */
1427 int (*set_cfm)(struct ofport *ofport, const struct cfm_settings *s);
1429 /* Checks the status of CFM configured on 'ofport'. Returns 0 if the
1430 * port's CFM status was successfully stored into '*status'. Returns
1431 * negative number if there is no status change since last update.
1432 * Returns positive errno otherwise.
1434 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1435 * support CFM, as does a null pointer.
1437 * The caller must provide and own '*status', and it must free the array
1438 * returned in 'status->rmps'. '*status' is indeterminate if the return
1439 * value is non-zero. */
1440 int (*get_cfm_status)(const struct ofport *ofport,
1441 struct ofproto_cfm_status *status);
1443 /* Configures BFD on 'ofport'.
1445 * If 'cfg' is NULL, or 'cfg' does not contain the key value pair
1446 * "enable=true", removes BFD from 'ofport'. Otherwise, configures BFD
1447 * according to 'cfg'.
1449 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1450 * support BFD, as does a null pointer. */
1451 int (*set_bfd)(struct ofport *ofport, const struct smap *cfg);
1453 /* Populates 'smap' with the status of BFD on 'ofport'. Returns 0 on
1454 * success. Returns a negative number if there is no status change since
1455 * last update. Returns a positive errno otherwise.
1457 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1458 * support BFD, as does a null pointer. */
1459 int (*get_bfd_status)(struct ofport *ofport, struct smap *smap);
1461 /* Configures spanning tree protocol (STP) on 'ofproto' using the
1462 * settings defined in 's'.
1464 * If 's' is nonnull, configures STP according to its members.
1466 * If 's' is null, removes any STP configuration from 'ofproto'.
1468 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1469 * support STP, as does a null pointer. */
1470 int (*set_stp)(struct ofproto *ofproto,
1471 const struct ofproto_stp_settings *s);
1473 /* Retrieves state of spanning tree protocol (STP) on 'ofproto'.
1475 * Stores STP state for 'ofproto' in 's'. If the 'enabled' member
1476 * is false, the other member values are not meaningful.
1478 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1479 * support STP, as does a null pointer. */
1480 int (*get_stp_status)(struct ofproto *ofproto,
1481 struct ofproto_stp_status *s);
1483 /* Configures spanning tree protocol (STP) on 'ofport' using the
1484 * settings defined in 's'.
1486 * If 's' is nonnull, configures STP according to its members. The
1487 * caller is responsible for assigning STP port numbers (using the
1488 * 'port_num' member in the range of 1 through 255, inclusive) and
1489 * ensuring there are no duplicates.
1491 * If 's' is null, removes any STP configuration from 'ofport'.
1493 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1494 * support STP, as does a null pointer. */
1495 int (*set_stp_port)(struct ofport *ofport,
1496 const struct ofproto_port_stp_settings *s);
1498 /* Retrieves spanning tree protocol (STP) port status of 'ofport'.
1500 * Stores STP state for 'ofport' in 's'. If the 'enabled' member is
1501 * false, the other member values are not meaningful.
1503 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1504 * support STP, as does a null pointer. */
1505 int (*get_stp_port_status)(struct ofport *ofport,
1506 struct ofproto_port_stp_status *s);
1508 /* Retrieves spanning tree protocol (STP) port statistics of 'ofport'.
1510 * Stores STP state for 'ofport' in 's'. If the 'enabled' member is
1511 * false, the other member values are not meaningful.
1513 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1514 * support STP, as does a null pointer. */
1515 int (*get_stp_port_stats)(struct ofport *ofport,
1516 struct ofproto_port_stp_stats *s);
1518 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
1519 * 'queues' attached to 'ofport'. This data is not intended to be
1520 * sufficient to implement QoS. Instead, providers may use this
1521 * information to implement features which require knowledge of what queues
1522 * exist on a port, and some basic information about them.
1524 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1525 * support QoS, as does a null pointer. */
1526 int (*set_queues)(struct ofport *ofport,
1527 const struct ofproto_port_queue *queues, size_t n_qdscp);
1529 /* If 's' is nonnull, this function registers a "bundle" associated with
1530 * client data pointer 'aux' in 'ofproto'. A bundle is the same concept as
1531 * a Port in OVSDB, that is, it consists of one or more "slave" devices
1532 * (Interfaces, in OVSDB) along with VLAN and LACP configuration and, if
1533 * there is more than one slave, a bonding configuration. If 'aux' is
1534 * already registered then this function updates its configuration to 's'.
1535 * Otherwise, this function registers a new bundle.
1537 * If 's' is NULL, this function unregisters the bundle registered on
1538 * 'ofproto' associated with client data pointer 'aux'. If no such bundle
1539 * has been registered, this has no effect.
1541 * This function affects only the behavior of the NXAST_AUTOPATH action and
1542 * output to the OFPP_NORMAL port. An implementation that does not support
1543 * it at all may set it to NULL or return EOPNOTSUPP. An implementation
1544 * that supports only a subset of the functionality should implement what
1545 * it can and return 0. */
1546 int (*bundle_set)(struct ofproto *ofproto, void *aux,
1547 const struct ofproto_bundle_settings *s);
1549 /* If 'port' is part of any bundle, removes it from that bundle. If the
1550 * bundle now has no ports, deletes the bundle. If the bundle now has only
1551 * one port, deconfigures the bundle's bonding configuration. */
1552 void (*bundle_remove)(struct ofport *ofport);
1554 /* If 's' is nonnull, this function registers a mirror associated with
1555 * client data pointer 'aux' in 'ofproto'. A mirror is the same concept as
1556 * a Mirror in OVSDB. If 'aux' is already registered then this function
1557 * updates its configuration to 's'. Otherwise, this function registers a
1560 * If 's' is NULL, this function unregisters the mirror registered on
1561 * 'ofproto' associated with client data pointer 'aux'. If no such mirror
1562 * has been registered, this has no effect.
1564 * An implementation that does not support mirroring at all may set
1565 * it to NULL or return EOPNOTSUPP. An implementation that supports
1566 * only a subset of the functionality should implement what it can
1568 int (*mirror_set)(struct ofproto *ofproto, void *aux,
1569 const struct ofproto_mirror_settings *s);
1571 /* Retrieves statistics from mirror associated with client data
1572 * pointer 'aux' in 'ofproto'. Stores packet and byte counts in
1573 * 'packets' and 'bytes', respectively. If a particular counter is
1574 * not supported, the appropriate argument is set to UINT64_MAX.
1576 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1577 * support retrieving mirror statistics. */
1578 int (*mirror_get_stats)(struct ofproto *ofproto, void *aux,
1579 uint64_t *packets, uint64_t *bytes);
1581 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs
1582 * on which all packets are flooded, instead of using MAC learning. If
1583 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1585 * This function affects only the behavior of the OFPP_NORMAL action. An
1586 * implementation that does not support it may set it to NULL or return
1588 int (*set_flood_vlans)(struct ofproto *ofproto,
1589 unsigned long *flood_vlans);
1591 /* Returns true if 'aux' is a registered bundle that is currently in use as
1592 * the output for a mirror. */
1593 bool (*is_mirror_output_bundle)(const struct ofproto *ofproto, void *aux);
1595 /* When the configuration option of forward_bpdu changes, this function
1596 * will be invoked. */
1597 void (*forward_bpdu_changed)(struct ofproto *ofproto);
1599 /* Sets the MAC aging timeout for the OFPP_NORMAL action to 'idle_time', in
1600 * seconds, and the maximum number of MAC table entries to
1603 * An implementation that doesn't support configuring these features may
1604 * set this function to NULL or implement it as a no-op. */
1605 void (*set_mac_table_config)(struct ofproto *ofproto,
1606 unsigned int idle_time, size_t max_entries);
1608 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
1610 * This is deprecated. It is only for compatibility with broken device drivers
1611 * in old versions of Linux that do not properly support VLANs when VLAN
1612 * devices are not used. When broken device drivers are no longer in
1613 * widespread use, we will delete these interfaces. */
1615 /* If 'realdev_ofp_port' is nonzero, then this function configures 'ofport'
1616 * as a VLAN splinter port for VLAN 'vid', associated with the real device
1617 * that has OpenFlow port number 'realdev_ofp_port'.
1619 * If 'realdev_ofp_port' is zero, then this function deconfigures 'ofport'
1620 * as a VLAN splinter port.
1622 * This function should be NULL if an implementation does not support it.
1624 int (*set_realdev)(struct ofport *ofport,
1625 ofp_port_t realdev_ofp_port, int vid);
1627 /* ## ------------------------ ## */
1628 /* ## OpenFlow meter functions ## */
1629 /* ## ------------------------ ## */
1631 /* These functions should be NULL if an implementation does not support
1632 * them. They must be all null or all non-null.. */
1634 /* Initializes 'features' to describe the metering features supported by
1636 void (*meter_get_features)(const struct ofproto *ofproto,
1637 struct ofputil_meter_features *features);
1639 /* If '*id' is UINT32_MAX, adds a new meter with the given 'config'. On
1640 * success the function must store a provider meter ID other than
1641 * UINT32_MAX in '*id'. All further references to the meter will be made
1642 * with the returned provider meter id rather than the OpenFlow meter id.
1643 * The caller does not try to interpret the provider meter id, giving the
1644 * implementation the freedom to either use the OpenFlow meter_id value
1645 * provided in the meter configuration, or any other value suitable for the
1648 * If '*id' is a value other than UINT32_MAX, modifies the existing meter
1649 * with that meter provider ID to have configuration 'config', while
1650 * leaving '*id' unchanged. On failure, the existing meter configuration
1651 * is left intact. */
1652 enum ofperr (*meter_set)(struct ofproto *ofproto, ofproto_meter_id *id,
1653 const struct ofputil_meter_config *config);
1655 /* Gets the meter and meter band packet and byte counts for maximum of
1656 * 'stats->n_bands' bands for the meter with provider ID 'id' within
1657 * 'ofproto'. The caller fills in the other stats values. The band stats
1658 * are copied to memory at 'stats->bands' provided by the caller. The
1659 * number of returned band stats is returned in 'stats->n_bands'. */
1660 enum ofperr (*meter_get)(const struct ofproto *ofproto,
1661 ofproto_meter_id id,
1662 struct ofputil_meter_stats *stats);
1664 /* Deletes a meter, making the 'ofproto_meter_id' invalid for any
1666 void (*meter_del)(struct ofproto *, ofproto_meter_id);
1669 /* ## -------------------- ## */
1670 /* ## OpenFlow 1.1+ groups ## */
1671 /* ## -------------------- ## */
1673 struct ofgroup *(*group_alloc)(void);
1674 enum ofperr (*group_construct)(struct ofgroup *);
1675 void (*group_destruct)(struct ofgroup *);
1676 void (*group_dealloc)(struct ofgroup *);
1678 enum ofperr (*group_modify)(struct ofgroup *);
1680 enum ofperr (*group_get_stats)(const struct ofgroup *,
1681 struct ofputil_group_stats *);
1684 extern const struct ofproto_class ofproto_dpif_class;
1686 int ofproto_class_register(const struct ofproto_class *);
1687 int ofproto_class_unregister(const struct ofproto_class *);
1689 int ofproto_flow_mod(struct ofproto *, struct ofputil_flow_mod *)
1690 OVS_EXCLUDED(ofproto_mutex);
1691 void ofproto_add_flow(struct ofproto *, const struct match *,
1692 unsigned int priority,
1693 const struct ofpact *ofpacts, size_t ofpacts_len)
1694 OVS_EXCLUDED(ofproto_mutex);
1695 void ofproto_delete_flow(struct ofproto *,
1696 const struct match *, unsigned int priority)
1697 OVS_EXCLUDED(ofproto_mutex);
1698 void ofproto_flush_flows(struct ofproto *);
1701 static inline const struct rule_actions *
1702 rule_get_actions(const struct rule *rule)
1704 return ovsrcu_get(const struct rule_actions *, &rule->actions);
1707 /* Returns true if 'rule' is an OpenFlow 1.3 "table-miss" rule, false
1710 * ("Table-miss" rules are special because a packet_in generated through one
1711 * uses OFPR_NO_MATCH as its reason, whereas packet_ins generated by any other
1712 * rule use OFPR_ACTION.) */
1714 rule_is_table_miss(const struct rule *rule)
1716 return rule->cr.priority == 0 && cls_rule_is_catchall(&rule->cr);
1719 /* Returns true if 'rule' should be hidden from the controller.
1721 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1722 * (e.g. by in-band control) and are intentionally hidden from the
1725 rule_is_hidden(const struct rule *rule)
1727 return rule->cr.priority > UINT16_MAX;
1730 static inline struct rule *
1731 rule_from_cls_rule(const struct cls_rule *cls_rule)
1733 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1736 #endif /* ofproto/ofproto-provider.h */