2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 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 ofputil_frag_handling frag_handling;
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;
96 cls_version_t tables_version; /* Controls which rules are visible to
99 /* Rules indexed on their cookie values, in all flow tables. */
100 struct hindex cookies OVS_GUARDED_BY(ofproto_mutex);
101 struct hmap learned_cookies OVS_GUARDED_BY(ofproto_mutex);
103 /* List of expirable flows, in all flow tables. */
104 struct ovs_list expirable OVS_GUARDED_BY(ofproto_mutex);
107 * OpenFlow meters start at 1. To avoid confusion we leave the first
108 * pointer in the array un-used, and index directly with the OpenFlow
110 struct ofputil_meter_features meter_features;
111 struct meter **meters; /* 'meter_features.max_meter' + 1 pointers. */
113 /* OpenFlow connections. */
114 struct connmgr *connmgr;
116 /* Delayed rule executions.
118 * We delay calls to ->ofproto_class->rule_execute() past releasing
119 * ofproto_mutex during a flow_mod, because otherwise a "learn" action
120 * triggered by the executing the packet would try to recursively modify
121 * the flow table and reacquire the global lock. */
122 struct guarded_list rule_executes; /* Contains "struct rule_execute"s. */
124 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
126 * This is deprecated. It is only for compatibility with broken device
127 * drivers in old versions of Linux that do not properly support VLANs when
128 * VLAN devices are not used. When broken device drivers are no longer in
129 * widespread use, we will delete these interfaces. */
130 unsigned long int *vlan_bitmap; /* 4096-bit bitmap of in-use VLANs. */
131 bool vlans_changed; /* True if new VLANs are in use. */
132 int min_mtu; /* Current MTU of non-internal ports. */
135 struct ovs_rwlock groups_rwlock;
136 struct hmap groups OVS_GUARDED; /* Contains "struct ofgroup"s. */
137 uint32_t n_groups[4] OVS_GUARDED; /* # of existing groups of each type. */
138 struct ofputil_group_features ogf;
141 void ofproto_init_tables(struct ofproto *, int n_tables);
142 void ofproto_init_max_ports(struct ofproto *, uint16_t max_ports);
144 struct ofproto *ofproto_lookup(const char *name);
145 struct ofport *ofproto_get_port(const struct ofproto *, ofp_port_t ofp_port);
147 /* An OpenFlow port within a "struct ofproto".
149 * The port's name is netdev_get_name(port->netdev).
151 * With few exceptions, ofproto implementations may look at these fields but
152 * should not modify them. */
154 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
155 struct ofproto *ofproto; /* The ofproto that contains this port. */
156 struct netdev *netdev;
157 struct ofputil_phy_port pp;
158 ofp_port_t ofp_port; /* OpenFlow port number. */
160 long long int created; /* Time created, in msec. */
164 void ofproto_port_set_state(struct ofport *, enum ofputil_port_state);
166 /* OpenFlow table flags:
168 * - "Hidden" tables are not included in OpenFlow operations that operate on
169 * "all tables". For example, a request for flow stats on all tables will
170 * omit flows in hidden tables, table stats requests will omit the table
171 * entirely, and the switch features reply will not count the hidden table.
173 * However, operations that specifically name the particular table still
174 * operate on it. For example, flow_mods and flow stats requests on a
177 * To avoid gaps in table IDs (which have unclear validity in OpenFlow),
178 * hidden tables must be the highest-numbered tables that a provider
181 * - "Read-only" tables can't be changed through OpenFlow operations. (At
182 * the moment all flow table operations go effectively through OpenFlow, so
183 * this means that read-only tables can't be changed at all after the
184 * read-only flag is set.)
186 * The generic ofproto layer never sets these flags. An ofproto provider can
187 * set them if it is appropriate.
190 OFTABLE_HIDDEN = 1 << 0, /* Hide from most OpenFlow operations. */
191 OFTABLE_READONLY = 1 << 1 /* Don't allow OpenFlow controller to change
195 /* A flow table within a "struct ofproto".
201 * Adding or removing rules requires holding ofproto_mutex.
203 * Rules in 'cls' are RCU protected. For extended access to a rule, try
204 * incrementing its ref_count with ofproto_rule_try_ref(), or
205 * ofproto_rule_ref(), if the rule is still known to be in 'cls'. A rule
206 * will be freed using ovsrcu_postpone() once its 'ref_count' reaches zero.
208 * Modifying a rule requires the rule's own mutex.
210 * Freeing a rule requires ofproto_mutex. After removing the rule from the
211 * classifier, release a ref_count from the rule ('cls''s reference to the
214 * Refer to the thread-safety notes on struct rule for more information.*/
216 enum oftable_flags flags;
217 struct classifier cls; /* Contains "struct rule"s. */
218 char *name; /* Table name exposed via OpenFlow, or NULL. */
220 /* Maximum number of flows or UINT_MAX if there is no limit besides any
221 * limit imposed by resource limitations. */
222 unsigned int max_flows;
223 /* Current number of flows, not counting temporary duplicates nor deferred
225 unsigned int n_flows;
227 /* These members determine the handling of an attempt to add a flow that
228 * would cause the table to have more than 'max_flows' flows.
230 * If 'eviction_fields' is NULL, overflows will be rejected with an error.
232 * If 'eviction_fields' is nonnull (regardless of whether n_eviction_fields
233 * is nonzero), an overflow will cause a flow to be removed. The flow to
234 * be removed is chosen to give fairness among groups distinguished by
235 * different values for the subfields within 'groups'. */
236 struct mf_subfield *eviction_fields;
237 size_t n_eviction_fields;
241 * When a flow is added that would cause the table to have more than
242 * 'max_flows' flows, and 'eviction_fields' is nonnull, these groups are
243 * used to decide which rule to evict: the rule is chosen from the eviction
244 * group that contains the greatest number of rules.*/
245 uint32_t eviction_group_id_basis;
246 struct hmap eviction_groups_by_id;
247 struct heap eviction_groups_by_size;
249 /* Flow table miss handling configuration. */
250 ATOMIC(enum ofputil_table_miss) miss_config;
252 /* Eviction is enabled if either the client (vswitchd) enables it or an
253 * OpenFlow controller enables it; thus, a nonzero value indicates that
254 * eviction is enabled. */
255 #define EVICTION_CLIENT (1 << 0) /* Set to 1 if client enables eviction. */
256 #define EVICTION_OPENFLOW (1 << 1) /* Set to 1 if OpenFlow enables eviction. */
257 unsigned int eviction;
259 /* If true, vacancy events are enabled; otherwise they are disabled. */
260 bool vacancy_enabled;
262 /* Non-zero values for vacancy_up and vacancy_down indicates that vacancy
263 * is enabled by table-mod, else these values are set to zero when
264 * vacancy is disabled */
265 uint8_t vacancy_down; /* Vacancy threshold when space decreases (%). */
266 uint8_t vacancy_up; /* Vacancy threshold when space increases (%). */
268 atomic_ulong n_matched;
269 atomic_ulong n_missed;
272 /* Assigns TABLE to each oftable, in turn, in OFPROTO.
274 * All parameters are evaluated multiple times. */
275 #define OFPROTO_FOR_EACH_TABLE(TABLE, OFPROTO) \
276 for ((TABLE) = (OFPROTO)->tables; \
277 (TABLE) < &(OFPROTO)->tables[(OFPROTO)->n_tables]; \
280 /* An OpenFlow flow within a "struct ofproto".
282 * With few exceptions, ofproto implementations may look at these fields but
283 * should not modify them.
289 * Except near the beginning or ending of its lifespan, rule 'rule' belongs to
290 * the classifier rule->ofproto->tables[rule->table_id].cls. The text below
291 * calls this classifier 'cls'.
296 * The thread safety rules described here for "struct rule" are motivated by
299 * - Prevent threads that read members of "struct rule" from reading bad
300 * data due to changes by some thread concurrently modifying those
303 * - Prevent two threads making changes to members of a given "struct rule"
304 * from interfering with each other.
310 * A rule 'rule' may be accessed without a risk of being freed by a thread
311 * until the thread quiesces (i.e., rules are RCU protected and destructed
312 * using ovsrcu_postpone()). Code that needs to hold onto a rule for a
313 * while should increment 'rule->ref_count' either with ofproto_rule_ref()
314 * (if 'ofproto_mutex' is held), or with ofproto_rule_try_ref() (when some
315 * other thread might remove the rule from 'cls'). ofproto_rule_try_ref()
316 * will fail if the rule has already been scheduled for destruction.
318 * 'rule->ref_count' protects 'rule' from being freed. It doesn't protect the
319 * rule from being deleted from 'cls' (that's 'ofproto_mutex') and it doesn't
320 * protect members of 'rule' from modification (that's 'rule->mutex').
322 * 'rule->mutex' protects the members of 'rule' from modification. It doesn't
323 * protect the rule from being deleted from 'cls' (that's 'ofproto_mutex') and
324 * it doesn't prevent the rule from being freed (that's 'rule->ref_count').
326 * Regarding thread safety, the members of a rule fall into the following
329 * - Immutable. These members are marked 'const'.
331 * - Members that may be safely read or written only by code holding
332 * ofproto_mutex. These are marked OVS_GUARDED_BY(ofproto_mutex).
334 * - Members that may be safely read only by code holding ofproto_mutex or
335 * 'rule->mutex', and safely written only by coding holding ofproto_mutex
336 * AND 'rule->mutex'. These are marked OVS_GUARDED.
339 /* Where this rule resides in an OpenFlow switch.
341 * These are immutable once the rule is constructed, hence 'const'. */
342 struct ofproto *const ofproto; /* The ofproto that contains this rule. */
343 const struct cls_rule cr; /* In owning ofproto's classifier. */
344 const uint8_t table_id; /* Index in ofproto's 'tables' array. */
345 bool removed; /* Rule has been removed from the ofproto
346 * data structures. */
348 /* Protects members marked OVS_GUARDED.
349 * Readers only need to hold this mutex.
350 * Writers must hold both this mutex AND ofproto_mutex.
351 * By implication writers can read *without* taking this mutex while they
352 * hold ofproto_mutex. */
353 struct ovs_mutex mutex OVS_ACQ_AFTER(ofproto_mutex);
355 /* Number of references.
356 * The classifier owns one reference.
357 * Any thread trying to keep a rule from being freed should hold its own
359 struct ovs_refcount ref_count;
361 /* A "flow cookie" is the OpenFlow name for a 64-bit value associated with
363 ovs_be64 flow_cookie OVS_GUARDED;
364 struct hindex_node cookie_node OVS_GUARDED_BY(ofproto_mutex);
366 enum ofputil_flow_mod_flags flags OVS_GUARDED;
369 uint16_t hard_timeout OVS_GUARDED; /* In seconds from ->modified. */
370 uint16_t idle_timeout OVS_GUARDED; /* In seconds from ->used. */
372 /* Eviction precedence. */
373 const uint16_t importance;
375 /* Removal reason for sending flow removed message.
376 * Used only if 'flags' has OFPUTIL_FF_SEND_FLOW_REM set and if the
377 * value is not OVS_OFPRR_NONE. */
378 uint8_t removed_reason;
380 /* Eviction groups (see comment on struct eviction_group for explanation) .
382 * 'eviction_group' is this rule's eviction group, or NULL if it is not in
383 * any eviction group. When 'eviction_group' is nonnull, 'evg_node' is in
384 * the ->eviction_group->rules hmap. */
385 struct eviction_group *eviction_group OVS_GUARDED_BY(ofproto_mutex);
386 struct heap_node evg_node OVS_GUARDED_BY(ofproto_mutex);
388 /* OpenFlow actions. See struct rule_actions for more thread-safety
390 const struct rule_actions * const actions;
392 /* In owning meter's 'rules' list. An empty list if there is no meter. */
393 struct ovs_list meter_list_node OVS_GUARDED_BY(ofproto_mutex);
395 /* Flow monitors (e.g. for NXST_FLOW_MONITOR, related to struct ofmonitor).
397 * 'add_seqno' is the sequence number when this rule was created.
398 * 'modify_seqno' is the sequence number when this rule was last modified.
399 * See 'monitor_seqno' in connmgr.c for more information. */
400 enum nx_flow_monitor_flags monitor_flags OVS_GUARDED_BY(ofproto_mutex);
401 uint64_t add_seqno OVS_GUARDED_BY(ofproto_mutex);
402 uint64_t modify_seqno OVS_GUARDED_BY(ofproto_mutex);
404 /* Optimisation for flow expiry. In ofproto's 'expirable' list if this
405 * rule is expirable, otherwise empty. */
406 struct ovs_list expirable OVS_GUARDED_BY(ofproto_mutex);
408 /* Times. Last so that they are more likely close to the stats managed
409 * by the provider. */
410 long long int created OVS_GUARDED; /* Creation time. */
412 /* Must hold 'mutex' for both read/write, 'ofproto_mutex' not needed. */
413 long long int modified OVS_GUARDED; /* Time of last modification. */
416 void ofproto_rule_ref(struct rule *);
417 bool ofproto_rule_try_ref(struct rule *);
418 void ofproto_rule_unref(struct rule *);
420 static inline const struct rule_actions * rule_get_actions(const struct rule *);
421 static inline bool rule_is_table_miss(const struct rule *);
422 static inline bool rule_is_hidden(const struct rule *);
424 /* A set of actions within a "struct rule".
430 * A struct rule_actions may be accessed without a risk of being freed by
431 * code that holds 'rule->mutex' (where 'rule' is the rule for which
432 * 'rule->actions == actions') or during the RCU active period.
434 * All members are immutable: they do not change during the rule's
436 struct rule_actions {
439 * 'has_meter' is true if 'ofpacts' contains an OFPACT_METER action.
441 * 'has_learn_with_delete' is true if 'ofpacts' contains an OFPACT_LEARN
442 * action whose flags include NX_LEARN_F_DELETE_LEARNED. */
444 bool has_learn_with_delete;
447 uint32_t ofpacts_len; /* Size of 'ofpacts', in bytes. */
448 struct ofpact ofpacts[]; /* Sequence of "struct ofpacts". */
450 BUILD_ASSERT_DECL(offsetof(struct rule_actions, ofpacts) % OFPACT_ALIGNTO == 0);
452 const struct rule_actions *rule_actions_create(const struct ofpact *, size_t);
453 void rule_actions_destroy(const struct rule_actions *);
454 bool ofproto_rule_has_out_port(const struct rule *, ofp_port_t port)
455 OVS_REQUIRES(ofproto_mutex);
457 /* A set of rules to which an OpenFlow operation applies. */
458 struct rule_collection {
459 struct rule **rules; /* The rules. */
460 size_t n; /* Number of rules collected. */
462 size_t capacity; /* Number of rules that will fit in 'rules'. */
463 struct rule *stub[64]; /* Preallocated rules to avoid malloc(). */
466 void rule_collection_init(struct rule_collection *);
467 void rule_collection_add(struct rule_collection *, struct rule *);
468 void rule_collection_ref(struct rule_collection *) OVS_REQUIRES(ofproto_mutex);
469 void rule_collection_unref(struct rule_collection *);
470 void rule_collection_destroy(struct rule_collection *);
472 /* Limits the number of flows allowed in the datapath. Only affects the
473 * ofproto-dpif implementation. */
474 extern unsigned ofproto_flow_limit;
476 /* Maximum idle time (in ms) for flows to be cached in the datapath.
477 * Revalidators may expire flows more quickly than the configured value based
478 * on system load and other factors. This variable is subject to change. */
479 extern unsigned ofproto_max_idle;
481 /* Number of upcall handler and revalidator threads. Only affects the
482 * ofproto-dpif implementation. */
483 extern size_t n_handlers, n_revalidators;
485 /* Cpu mask for pmd threads. */
486 extern char *pmd_cpu_mask;
488 static inline struct rule *rule_from_cls_rule(const struct cls_rule *);
490 void ofproto_rule_expire(struct rule *rule, uint8_t reason)
491 OVS_REQUIRES(ofproto_mutex);
492 void ofproto_rule_delete(struct ofproto *, struct rule *)
493 OVS_EXCLUDED(ofproto_mutex);
494 void ofproto_rule_reduce_timeouts(struct rule *rule, uint16_t idle_timeout,
495 uint16_t hard_timeout)
496 OVS_EXCLUDED(ofproto_mutex);
498 /* A group within a "struct ofproto".
500 * With few exceptions, ofproto implementations may look at these fields but
501 * should not modify them. */
503 struct hmap_node hmap_node OVS_GUARDED; /* In ofproto's "groups" hmap. */
505 /* Number of references.
507 * This is needed to keep track of references to the group in the xlate
510 * If the main thread removes the group from an ofproto, we need to
511 * guarantee that the group remains accessible to users of
512 * xlate_group_actions and the xlate_cache, as the xlate_cache will not be
513 * cleaned up until the corresponding datapath flows are revalidated. */
514 struct ovs_refcount ref_count;
516 /* No lock is needed to protect the fields below since they are not
517 * modified after construction. */
518 const struct ofproto *ofproto; /* The ofproto that contains this group. */
519 const uint32_t group_id;
520 const enum ofp11_group_type type; /* One of OFPGT_*. */
522 const long long int created; /* Creation time. */
523 const long long int modified; /* Time of last modification. */
525 struct ovs_list buckets; /* Contains "struct ofputil_bucket"s. */
526 const uint32_t n_buckets;
528 const struct ofputil_group_props props;
531 bool ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
532 struct ofgroup **group);
534 void ofproto_group_ref(struct ofgroup *);
535 void ofproto_group_unref(struct ofgroup *);
537 void ofproto_group_delete_all(struct ofproto *);
539 /* ofproto class structure, to be defined by each ofproto implementation.
545 * These functions work primarily with four different kinds of data
548 * - "struct ofproto", which represents an OpenFlow switch.
550 * - "struct ofport", which represents a port within an ofproto.
552 * - "struct rule", which represents an OpenFlow flow within an ofproto.
554 * - "struct ofgroup", which represents an OpenFlow 1.1+ group within an
557 * Each of these data structures contains all of the implementation-independent
558 * generic state for the respective concept, called the "base" state. None of
559 * them contains any extra space for ofproto implementations to use. Instead,
560 * each implementation is expected to declare its own data structure that
561 * contains an instance of the generic data structure plus additional
562 * implementation-specific members, called the "derived" state. The
563 * implementation can use casts or (preferably) the CONTAINER_OF macro to
564 * obtain access to derived state given only a pointer to the embedded generic
571 * Four stylized functions accompany each of these data structures:
573 * "alloc" "construct" "destruct" "dealloc"
574 * ------------ ---------------- --------------- --------------
575 * ofproto ->alloc ->construct ->destruct ->dealloc
576 * ofport ->port_alloc ->port_construct ->port_destruct ->port_dealloc
577 * rule ->rule_alloc ->rule_construct ->rule_destruct ->rule_dealloc
578 * group ->group_alloc ->group_construct ->group_destruct ->group_dealloc
580 * "ofproto", "ofport", and "group" have this exact life cycle. The "rule"
581 * data structure also follow this life cycle with some additional elaborations
582 * described under "Rule Life Cycle" below.
584 * Any instance of a given data structure goes through the following life
587 * 1. The client calls the "alloc" function to obtain raw memory. If "alloc"
588 * fails, skip all the other steps.
590 * 2. The client initializes all of the data structure's base state. If this
591 * fails, skip to step 7.
593 * 3. The client calls the "construct" function. The implementation
594 * initializes derived state. It may refer to the already-initialized
595 * base state. If "construct" fails, skip to step 6.
597 * 4. The data structure is now initialized and in use.
599 * 5. When the data structure is no longer needed, the client calls the
600 * "destruct" function. The implementation uninitializes derived state.
601 * The base state has not been uninitialized yet, so the implementation
602 * may still refer to it.
604 * 6. The client uninitializes all of the data structure's base state.
606 * 7. The client calls the "dealloc" to free the raw memory. The
607 * implementation must not refer to base or derived state in the data
608 * structure, because it has already been uninitialized.
610 * Each "alloc" function allocates and returns a new instance of the respective
611 * data structure. The "alloc" function is not given any information about the
612 * use of the new data structure, so it cannot perform much initialization.
613 * Its purpose is just to ensure that the new data structure has enough room
614 * for base and derived state. It may return a null pointer if memory is not
615 * available, in which case none of the other functions is called.
617 * Each "construct" function initializes derived state in its respective data
618 * structure. When "construct" is called, all of the base state has already
619 * been initialized, so the "construct" function may refer to it. The
620 * "construct" function is allowed to fail, in which case the client calls the
621 * "dealloc" function (but not the "destruct" function).
623 * Each "destruct" function uninitializes and frees derived state in its
624 * respective data structure. When "destruct" is called, the base state has
625 * not yet been uninitialized, so the "destruct" function may refer to it. The
626 * "destruct" function is not allowed to fail.
628 * Each "dealloc" function frees raw memory that was allocated by the
629 * "alloc" function. The memory's base and derived members might not have ever
630 * been initialized (but if "construct" returned successfully, then it has been
631 * "destruct"ed already). The "dealloc" function is not allowed to fail.
637 * Most of these functions return 0 if they are successful or a positive error
638 * code on failure. Depending on the function, valid error codes are either
639 * errno values or OFPERR_* OpenFlow error codes.
641 * Most of these functions are expected to execute synchronously, that is, to
642 * block as necessary to obtain a result. Thus, these functions may return
643 * EAGAIN (or EWOULDBLOCK or EINPROGRESS) only where the function descriptions
644 * explicitly say those errors are a possibility. We may relax this
645 * requirement in the future if and when we encounter performance problems. */
646 struct ofproto_class {
647 /* ## ----------------- ## */
648 /* ## Factory Functions ## */
649 /* ## ----------------- ## */
651 /* Initializes provider. The caller may pass in 'iface_hints',
652 * which contains an shash of "struct iface_hint" elements indexed
653 * by the interface's name. The provider may use these hints to
654 * describe the startup configuration in order to reinitialize its
655 * state. The caller owns the provided data, so a provider must
656 * make copies of anything required. An ofproto provider must
657 * remove any existing state that is not described by the hint, and
658 * may choose to remove it all. */
659 void (*init)(const struct shash *iface_hints);
661 /* Enumerates the types of all supported ofproto types into 'types'. The
662 * caller has already initialized 'types'. The implementation should add
663 * its own types to 'types' but not remove any existing ones, because other
664 * ofproto classes might already have added names to it. */
665 void (*enumerate_types)(struct sset *types);
667 /* Enumerates the names of all existing datapath of the specified 'type'
668 * into 'names' 'all_dps'. The caller has already initialized 'names' as
671 * 'type' is one of the types enumerated by ->enumerate_types().
673 * Returns 0 if successful, otherwise a positive errno value.
675 int (*enumerate_names)(const char *type, struct sset *names);
677 /* Deletes the datapath with the specified 'type' and 'name'. The caller
678 * should have closed any open ofproto with this 'type' and 'name'; this
679 * function is allowed to fail if that is not the case.
681 * 'type' is one of the types enumerated by ->enumerate_types().
682 * 'name' is one of the names enumerated by ->enumerate_names() for 'type'.
684 * Returns 0 if successful, otherwise a positive errno value.
686 int (*del)(const char *type, const char *name);
688 /* Returns the type to pass to netdev_open() when a datapath of type
689 * 'datapath_type' has a port of type 'port_type', for a few special
690 * cases when a netdev type differs from a port type. For example,
691 * when using the userspace datapath, a port of type "internal"
692 * needs to be opened as "tap".
694 * Returns either 'type' itself or a string literal, which must not
696 const char *(*port_open_type)(const char *datapath_type,
697 const char *port_type);
699 /* ## ------------------------ ## */
700 /* ## Top-Level type Functions ## */
701 /* ## ------------------------ ## */
703 /* Performs any periodic activity required on ofprotos of type
706 * An ofproto provider may implement it or not, depending on whether
707 * it needs type-level maintenance.
709 * Returns 0 if successful, otherwise a positive errno value. */
710 int (*type_run)(const char *type);
712 /* Causes the poll loop to wake up when a type 'type''s 'run'
713 * function needs to be called, e.g. by calling the timer or fd
714 * waiting functions in poll-loop.h.
716 * An ofproto provider may implement it or not, depending on whether
717 * it needs type-level maintenance. */
718 void (*type_wait)(const char *type);
720 /* ## --------------------------- ## */
721 /* ## Top-Level ofproto Functions ## */
722 /* ## --------------------------- ## */
724 /* Life-cycle functions for an "ofproto" (see "Life Cycle" above).
730 * ->construct() should not modify any base members of the ofproto. The
731 * client will initialize the ofproto's 'ports' and 'tables' members after
732 * construction is complete.
734 * When ->construct() is called, the client does not yet know how many flow
735 * tables the datapath supports, so ofproto->n_tables will be 0 and
736 * ofproto->tables will be NULL. ->construct() should call
737 * ofproto_init_tables() to allocate and initialize ofproto->n_tables and
738 * ofproto->tables. Each flow table will be initially empty, so
739 * ->construct() should delete flows from the underlying datapath, if
740 * necessary, rather than populating the tables.
742 * If the ofproto knows the maximum port number that the datapath can have,
743 * then it can call ofproto_init_max_ports(). If it does so, then the
744 * client will ensure that the actions it allows to be used through
745 * OpenFlow do not refer to ports above that maximum number.
747 * Only one ofproto instance needs to be supported for any given datapath.
748 * If a datapath is already open as part of one "ofproto", then another
749 * attempt to "construct" the same datapath as part of another ofproto is
750 * allowed to fail with an error.
752 * ->construct() returns 0 if successful, otherwise a positive errno
759 * ->destruct() must also destroy all remaining rules in the ofproto's
760 * tables, by passing each remaining rule to ofproto_rule_delete(), then
761 * destroy all remaining groups by calling ofproto_group_delete_all().
763 * The client will destroy the flow tables themselves after ->destruct()
766 struct ofproto *(*alloc)(void);
767 int (*construct)(struct ofproto *ofproto);
768 void (*destruct)(struct ofproto *ofproto);
769 void (*dealloc)(struct ofproto *ofproto);
771 /* Performs any periodic activity required by 'ofproto'. It should:
773 * - Call connmgr_send_packet_in() for each received packet that missed
774 * in the OpenFlow flow table or that had a OFPP_CONTROLLER output
777 * - Call ofproto_rule_expire() for each OpenFlow flow that has reached
778 * its hard_timeout or idle_timeout, to expire the flow.
780 * Returns 0 if successful, otherwise a positive errno value. */
781 int (*run)(struct ofproto *ofproto);
783 /* Causes the poll loop to wake up when 'ofproto''s 'run' function needs to
784 * be called, e.g. by calling the timer or fd waiting functions in
786 void (*wait)(struct ofproto *ofproto);
788 /* Adds some memory usage statistics for the implementation of 'ofproto'
789 * into 'usage', for use with memory_report().
791 * This function is optional. */
792 void (*get_memory_usage)(const struct ofproto *ofproto,
793 struct simap *usage);
795 /* Adds some memory usage statistics for the implementation of 'type'
796 * into 'usage', for use with memory_report().
798 * This function is optional. */
799 void (*type_get_memory_usage)(const char *type, struct simap *usage);
801 /* Every "struct rule" in 'ofproto' is about to be deleted, one by one.
802 * This function may prepare for that, for example by clearing state in
803 * advance. It should *not* actually delete any "struct rule"s from
804 * 'ofproto', only prepare for it.
806 * This function is optional; it's really just for optimization in case
807 * it's cheaper to delete all the flows from your hardware in a single pass
808 * than to do it one by one. */
809 void (*flush)(struct ofproto *ofproto);
811 /* Helper for the OpenFlow OFPT_TABLE_FEATURES request.
813 * The 'features' array contains 'ofproto->n_tables' elements. Each
814 * element is initialized as:
816 * - 'table_id' to the array index.
818 * - 'name' to "table#" where # is the table ID.
820 * - 'metadata_match' and 'metadata_write' to OVS_BE64_MAX.
822 * - 'config' to the table miss configuration.
824 * - 'max_entries' to 1,000,000.
826 * - Both 'nonmiss' and 'miss' to:
828 * * 'next' to all 1-bits for all later tables.
830 * * 'instructions' to all instructions.
832 * * 'write' and 'apply' both to:
834 * - 'ofpacts': All actions.
836 * - 'set_fields': All fields.
838 * - 'match', 'mask', and 'wildcard' to all fields.
840 * If 'stats' is nonnull, it also contains 'ofproto->n_tables' elements.
841 * Each element is initialized as:
843 * - 'table_id' to the array index.
845 * - 'active_count' to the 'n_flows' of struct ofproto for the table.
847 * - 'lookup_count' and 'matched_count' to 0.
849 * The implementation should update any members in each element for which
850 * it has better values:
852 * - Any member of 'features' to better describe the implementation's
855 * - 'lookup_count' to the number of packets looked up in this flow table
858 * - 'matched_count' to the number of packets looked up in this flow
859 * table so far that matched one of the flow entries.
861 void (*query_tables)(struct ofproto *ofproto,
862 struct ofputil_table_features *features,
863 struct ofputil_table_stats *stats);
865 /* Sets the current tables version the provider should use for classifier
867 void (*set_tables_version)(struct ofproto *ofproto, cls_version_t version);
868 /* ## ---------------- ## */
869 /* ## ofport Functions ## */
870 /* ## ---------------- ## */
872 /* Life-cycle functions for a "struct ofport" (see "Life Cycle" above).
874 * ->port_construct() should not modify any base members of the ofport.
875 * An ofproto implementation should use the 'ofp_port' member of
876 * "struct ofport" as the OpenFlow port number.
878 * ofports are managed by the base ofproto code. The ofproto
879 * implementation should only create and destroy them in response to calls
880 * to these functions. The base ofproto code will create and destroy
881 * ofports in the following situations:
883 * - Just after the ->construct() function is called, the base ofproto
884 * iterates over all of the implementation's ports, using
885 * ->port_dump_start() and related functions, and constructs an ofport
886 * for each dumped port.
888 * - If ->port_poll() reports that a specific port has changed, then the
889 * base ofproto will query that port with ->port_query_by_name() and
890 * construct or destruct ofports as necessary to reflect the updated
893 * - If ->port_poll() returns ENOBUFS to report an unspecified port set
894 * change, then the base ofproto will iterate over all of the
895 * implementation's ports, in the same way as at ofproto
896 * initialization, and construct and destruct ofports to reflect all of
899 * - On graceful shutdown, the base ofproto code will destruct all
902 * ->port_construct() returns 0 if successful, otherwise a positive errno
909 * ->port_destruct() takes a 'del' parameter. If the provider implements
910 * the datapath itself (e.g. dpif-netdev, it can ignore 'del'. On the
911 * other hand, if the provider is an interface to an external datapath
912 * (e.g. to a kernel module that implement the datapath) then 'del' should
913 * influence destruction behavior in the following way:
915 * - If 'del' is true, it should remove the port from the underlying
916 * implementation. This is the common case.
918 * - If 'del' is false, it should leave the port in the underlying
919 * implementation. This is used when Open vSwitch is performing a
920 * graceful shutdown and ensures that, across Open vSwitch restarts,
921 * the underlying ports are not removed and recreated. That makes an
922 * important difference for, e.g., "internal" ports that have
923 * configured IP addresses; without this distinction, the IP address
924 * and other configured state for the port is lost.
926 struct ofport *(*port_alloc)(void);
927 int (*port_construct)(struct ofport *ofport);
928 void (*port_destruct)(struct ofport *ofport, bool del);
929 void (*port_dealloc)(struct ofport *ofport);
931 /* Called after 'ofport->netdev' is replaced by a new netdev object. If
932 * the ofproto implementation uses the ofport's netdev internally, then it
933 * should switch to using the new one. The old one has been closed.
935 * An ofproto implementation that doesn't need to do anything in this
936 * function may use a null pointer. */
937 void (*port_modified)(struct ofport *ofport);
939 /* Called after an OpenFlow request changes a port's configuration.
940 * 'ofport->pp.config' contains the new configuration. 'old_config'
941 * contains the previous configuration.
943 * The caller implements OFPUTIL_PC_PORT_DOWN using netdev functions to
944 * turn NETDEV_UP on and off, so this function doesn't have to do anything
945 * for that bit (and it won't be called if that is the only bit that
947 void (*port_reconfigured)(struct ofport *ofport,
948 enum ofputil_port_config old_config);
950 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
951 * initializes '*port' appropriately. Otherwise, returns a positive errno
954 * The caller owns the data in 'port' and must free it with
955 * ofproto_port_destroy() when it is no longer needed. */
956 int (*port_query_by_name)(const struct ofproto *ofproto,
957 const char *devname, struct ofproto_port *port);
959 /* Attempts to add 'netdev' as a port on 'ofproto'. Returns 0 if
960 * successful, otherwise a positive errno value. The caller should
961 * inform the implementation of the OpenFlow port through the
962 * ->port_construct() method.
964 * It doesn't matter whether the new port will be returned by a later call
965 * to ->port_poll(); the implementation may do whatever is more
967 int (*port_add)(struct ofproto *ofproto, struct netdev *netdev);
969 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'. Returns
970 * 0 if successful, otherwise a positive errno value.
972 * It doesn't matter whether the new port will be returned by a later call
973 * to ->port_poll(); the implementation may do whatever is more
975 int (*port_del)(struct ofproto *ofproto, ofp_port_t ofp_port);
978 int (*port_get_stats)(const struct ofport *port,
979 struct netdev_stats *stats);
981 /* Port iteration functions.
983 * The client might not be entirely in control of the ports within an
984 * ofproto. Some hardware implementations, for example, might have a fixed
985 * set of ports in a datapath. For this reason, the client needs a way to
986 * iterate through all the ports that are actually in a datapath. These
987 * functions provide that functionality.
989 * The 'state' pointer provides the implementation a place to
990 * keep track of its position. Its format is opaque to the caller.
992 * The ofproto provider retains ownership of the data that it stores into
993 * ->port_dump_next()'s 'port' argument. The data must remain valid until
994 * at least the next call to ->port_dump_next() or ->port_dump_done() for
995 * 'state'. The caller will not modify or free it.
1000 * ->port_dump_start() attempts to begin dumping the ports in 'ofproto'.
1001 * On success, it should return 0 and initialize '*statep' with any data
1002 * needed for iteration. On failure, returns a positive errno value, and
1003 * the client will not call ->port_dump_next() or ->port_dump_done().
1005 * ->port_dump_next() attempts to retrieve another port from 'ofproto' for
1006 * 'state'. If there is another port, it should store the port's
1007 * information into 'port' and return 0. It should return EOF if all ports
1008 * have already been iterated. Otherwise, on error, it should return a
1009 * positive errno value. This function will not be called again once it
1010 * returns nonzero once for a given iteration (but the 'port_dump_done'
1011 * function will be called afterward).
1013 * ->port_dump_done() allows the implementation to release resources used
1014 * for iteration. The caller might decide to stop iteration in the middle
1015 * by calling this function before ->port_dump_next() returns nonzero.
1023 * error = ofproto->ofproto_class->port_dump_start(ofproto, &state);
1026 * struct ofproto_port port;
1028 * error = ofproto->ofproto_class->port_dump_next(
1029 * ofproto, state, &port);
1033 * // Do something with 'port' here (without modifying or freeing
1034 * // any of its data).
1036 * ofproto->ofproto_class->port_dump_done(ofproto, state);
1038 * // 'error' is now EOF (success) or a positive errno value (failure).
1040 int (*port_dump_start)(const struct ofproto *ofproto, void **statep);
1041 int (*port_dump_next)(const struct ofproto *ofproto, void *state,
1042 struct ofproto_port *port);
1043 int (*port_dump_done)(const struct ofproto *ofproto, void *state);
1045 /* Polls for changes in the set of ports in 'ofproto'. If the set of ports
1046 * in 'ofproto' has changed, then this function should do one of the
1049 * - Preferably: store the name of the device that was added to or deleted
1050 * from 'ofproto' in '*devnamep' and return 0. The caller is responsible
1051 * for freeing '*devnamep' (with free()) when it no longer needs it.
1053 * - Alternatively: return ENOBUFS, without indicating the device that was
1056 * Occasional 'false positives', in which the function returns 0 while
1057 * indicating a device that was not actually added or deleted or returns
1058 * ENOBUFS without any change, are acceptable.
1060 * The purpose of 'port_poll' is to let 'ofproto' know about changes made
1061 * externally to the 'ofproto' object, e.g. by a system administrator via
1062 * ovs-dpctl. Therefore, it's OK, and even preferable, for port_poll() to
1063 * not report changes made through calls to 'port_add' or 'port_del' on the
1064 * same 'ofproto' object. (But it's OK for it to report them too, just
1065 * slightly less efficient.)
1067 * If the set of ports in 'ofproto' has not changed, returns EAGAIN. May
1068 * also return other positive errno values to indicate that something has
1071 * If the set of ports in a datapath is fixed, or if the only way that the
1072 * set of ports in a datapath can change is through ->port_add() and
1073 * ->port_del(), then this function may be a null pointer.
1075 int (*port_poll)(const struct ofproto *ofproto, char **devnamep);
1077 /* Arranges for the poll loop to wake up when ->port_poll() will return a
1078 * value other than EAGAIN.
1080 * If the set of ports in a datapath is fixed, or if the only way that the
1081 * set of ports in a datapath can change is through ->port_add() and
1082 * ->port_del(), or if the poll loop will always wake up anyway when
1083 * ->port_poll() will return a value other than EAGAIN, then this function
1084 * may be a null pointer.
1086 void (*port_poll_wait)(const struct ofproto *ofproto);
1088 /* Checks the status of LACP negotiation for 'port'. Returns 1 if LACP
1089 * partner information for 'port' is up-to-date, 0 if LACP partner
1090 * information is not current (generally indicating a connectivity
1091 * problem), or -1 if LACP is not enabled on 'port'.
1093 * This function may be a null pointer if the ofproto implementation does
1096 int (*port_is_lacp_current)(const struct ofport *port);
1098 /* Get LACP port stats. Returns -1 if LACP is not enabled on 'port'.
1100 * This function may be a null pointer if the ofproto implementation does
1103 int (*port_get_lacp_stats)(const struct ofport *port,
1104 struct lacp_slave_stats *stats);
1106 /* ## ----------------------- ## */
1107 /* ## OpenFlow Rule Functions ## */
1108 /* ## ----------------------- ## */
1110 /* Chooses an appropriate table for 'match' within 'ofproto'. On
1111 * success, stores the table ID into '*table_idp' and returns 0. On
1112 * failure, returns an OpenFlow error code.
1114 * The choice of table should be a function of 'match' and 'ofproto''s
1115 * datapath capabilities. It should not depend on the flows already in
1116 * 'ofproto''s flow tables. Failure implies that an OpenFlow rule with
1117 * 'match' as its matching condition can never be inserted into 'ofproto',
1118 * even starting from an empty flow table.
1120 * If multiple tables are candidates for inserting the flow, the function
1121 * should choose one arbitrarily (but deterministically).
1123 * If this function is NULL then table 0 is always chosen. */
1124 enum ofperr (*rule_choose_table)(const struct ofproto *ofproto,
1125 const struct match *match,
1126 uint8_t *table_idp);
1128 /* Life-cycle functions for a "struct rule".
1134 * The life cycle of a struct rule is an elaboration of the basic life
1135 * cycle described above under "Life Cycle".
1137 * After a rule is successfully constructed, it is then inserted. If
1138 * insertion is successful, then before it is later destructed, it is
1141 * You can think of a rule as having the following extra steps inserted
1142 * between "Life Cycle" steps 4 and 5:
1144 * 4.1. The client inserts the rule into the flow table, making it
1145 * visible in flow table lookups.
1147 * 4.2. The client calls "rule_insert" to insert the flow. The
1148 * implementation attempts to install the flow in the underlying
1149 * hardware and returns an error code indicate success or failure.
1150 * On failure, go to step 5.
1152 * 4.3. The rule is now installed in the flow table. Eventually it will
1155 * 4.4. The client removes the rule from the flow table. It is no longer
1156 * visible in flow table lookups.
1158 * 4.5. The client calls "rule_delete". The implementation uninstalls
1159 * the flow from the underlying hardware. Deletion is not allowed
1166 * When ->rule_construct() is called, 'rule' is a new rule that is not yet
1167 * inserted into a flow table. ->rule_construct() should initialize enough
1168 * of the rule's derived state for 'rule' to be suitable for inserting into
1169 * a flow table. ->rule_construct() should not modify any base members of
1172 * If ->rule_construct() fails (as indicated by returning a nonzero
1173 * OpenFlow error code), the ofproto base code will uninitialize and
1174 * deallocate 'rule'. See "Rule Life Cycle" above for more details.
1176 * ->rule_construct() must also:
1178 * - Validate that the datapath supports the matching rule in 'rule->cr'
1179 * datapath. For example, if the rule's table does not support
1180 * registers, then it is an error if 'rule->cr' does not wildcard all
1183 * - Validate that the datapath can correctly implement 'rule->ofpacts'.
1185 * After a successful construction the rest of the rule life cycle calls
1186 * may not fail, so ->rule_construct() must also make sure that the rule
1187 * can be inserted in to the datapath.
1193 * Following successful construction, the ofproto base case inserts 'rule'
1194 * into its flow table, then it calls ->rule_insert(). ->rule_insert()
1195 * must add the new rule to the datapath flow table and return only after
1196 * this is complete. The 'new_rule' may be a duplicate of an 'old_rule'.
1197 * In this case the 'old_rule' is non-null, and the implementation should
1198 * forward rule statistics from the 'old_rule' to the 'new_rule' if
1199 * 'forward_stats' is 'true'. This may not fail.
1205 * The ofproto base code removes 'rule' from its flow table before it calls
1206 * ->rule_delete(). ->rule_delete() must remove 'rule' from the datapath
1207 * flow table and return only after this has completed successfully.
1209 * Rule deletion must not fail.
1215 * ->rule_destruct() must uninitialize derived state.
1217 * Rule destruction must not fail. */
1218 struct rule *(*rule_alloc)(void);
1219 enum ofperr (*rule_construct)(struct rule *rule)
1220 /* OVS_REQUIRES(ofproto_mutex) */;
1221 void (*rule_insert)(struct rule *rule, struct rule *old_rule,
1223 /* OVS_REQUIRES(ofproto_mutex) */;
1224 void (*rule_delete)(struct rule *rule) /* OVS_REQUIRES(ofproto_mutex) */;
1225 void (*rule_destruct)(struct rule *rule);
1226 void (*rule_dealloc)(struct rule *rule);
1228 /* Obtains statistics for 'rule', storing the number of packets that have
1229 * matched it in '*packet_count' and the number of bytes in those packets
1230 * in '*byte_count'. UINT64_MAX indicates that the packet count or byte
1231 * count is unknown. */
1232 void (*rule_get_stats)(struct rule *rule, uint64_t *packet_count,
1233 uint64_t *byte_count, long long int *used)
1234 /* OVS_EXCLUDED(ofproto_mutex) */;
1236 /* Applies the actions in 'rule' to 'packet'. (This implements sending
1237 * buffered packets for OpenFlow OFPT_FLOW_MOD commands.)
1239 * Takes ownership of 'packet' (so it should eventually free it, with
1242 * 'flow' reflects the flow information for 'packet'. All of the
1243 * information in 'flow' is extracted from 'packet', except for
1244 * flow->tunnel and flow->in_port, which are assigned the correct values
1245 * for the incoming packet. The register values are zeroed. 'packet''s
1246 * header pointers and offsets (e.g. packet->l3) are appropriately
1247 * initialized. packet->l3 is aligned on a 32-bit boundary.
1249 * The implementation should add the statistics for 'packet' into 'rule'.
1251 * Returns 0 if successful, otherwise an OpenFlow error code. */
1252 enum ofperr (*rule_execute)(struct rule *rule, const struct flow *flow,
1253 struct dp_packet *packet);
1255 /* Changes the OpenFlow IP fragment handling policy to 'frag_handling',
1256 * which takes one of the following values, with the corresponding
1259 * - OFPUTIL_FRAG_NORMAL: The switch should treat IP fragments the same
1260 * way as other packets, omitting TCP and UDP port numbers (always
1261 * setting them to 0).
1263 * - OFPUTIL_FRAG_DROP: The switch should drop all IP fragments without
1264 * passing them through the flow table.
1266 * - OFPUTIL_FRAG_REASM: The switch should reassemble IP fragments before
1267 * passing packets through the flow table.
1269 * - OFPUTIL_FRAG_NX_MATCH (a Nicira extension): Similar to
1270 * OFPUTIL_FRAG_NORMAL, except that TCP and UDP port numbers should be
1271 * included in fragments with offset 0.
1273 * Implementations are not required to support every mode.
1274 * OFPUTIL_FRAG_NORMAL is the default mode when an ofproto is created.
1276 * At the time of the call to ->set_frag_handling(), the current mode is
1277 * available in 'ofproto->frag_handling'. ->set_frag_handling() returns
1278 * true if the requested mode was set, false if it is not supported.
1280 * Upon successful return, the caller changes 'ofproto->frag_handling' to
1281 * reflect the new mode.
1283 bool (*set_frag_handling)(struct ofproto *ofproto,
1284 enum ofputil_frag_handling frag_handling);
1286 /* Implements the OpenFlow OFPT_PACKET_OUT command. The datapath should
1287 * execute the 'ofpacts_len' bytes of "struct ofpacts" in 'ofpacts'.
1289 * The caller retains ownership of 'packet' and of 'ofpacts', so
1290 * ->packet_out() should not modify or free them.
1292 * This function must validate that it can correctly implement 'ofpacts'.
1293 * If not, then it should return an OpenFlow error code.
1295 * 'flow' reflects the flow information for 'packet'. All of the
1296 * information in 'flow' is extracted from 'packet', except for
1297 * flow->in_port (see below). flow->tunnel and its register values are
1300 * flow->in_port comes from the OpenFlow OFPT_PACKET_OUT message. The
1301 * implementation should reject invalid flow->in_port values by returning
1302 * OFPERR_OFPBRC_BAD_PORT. (If the implementation called
1303 * ofproto_init_max_ports(), then the client will reject these ports
1304 * itself.) For consistency, the implementation should consider valid for
1305 * flow->in_port any value that could possibly be seen in a packet that it
1306 * passes to connmgr_send_packet_in(). Ideally, even an implementation
1307 * that never generates packet-ins (e.g. due to hardware limitations)
1308 * should still allow flow->in_port values for every possible physical port
1309 * and OFPP_LOCAL. The only virtual ports (those above OFPP_MAX) that the
1310 * caller will ever pass in as flow->in_port, other than OFPP_LOCAL, are
1311 * OFPP_NONE and OFPP_CONTROLLER. The implementation should allow both of
1312 * these, treating each of them as packets generated by the controller as
1313 * opposed to packets originating from some switch port.
1315 * (Ordinarily the only effect of flow->in_port is on output actions that
1316 * involve the input port, such as actions that output to OFPP_IN_PORT,
1317 * OFPP_FLOOD, or OFPP_ALL. flow->in_port can also affect Nicira extension
1318 * "resubmit" actions.)
1320 * 'packet' is not matched against the OpenFlow flow table, so its
1321 * statistics should not be included in OpenFlow flow statistics.
1323 * Returns 0 if successful, otherwise an OpenFlow error code. */
1324 enum ofperr (*packet_out)(struct ofproto *ofproto, struct dp_packet *packet,
1325 const struct flow *flow,
1326 const struct ofpact *ofpacts,
1327 size_t ofpacts_len);
1329 /* ## ------------------------- ## */
1330 /* ## OFPP_NORMAL configuration ## */
1331 /* ## ------------------------- ## */
1333 /* Configures NetFlow on 'ofproto' according to the options in
1334 * 'netflow_options', or turns off NetFlow if 'netflow_options' is NULL.
1336 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
1337 * NetFlow, as does a null pointer. */
1338 int (*set_netflow)(struct ofproto *ofproto,
1339 const struct netflow_options *netflow_options);
1341 void (*get_netflow_ids)(const struct ofproto *ofproto,
1342 uint8_t *engine_type, uint8_t *engine_id);
1344 /* Configures sFlow on 'ofproto' according to the options in
1345 * 'sflow_options', or turns off sFlow if 'sflow_options' is NULL.
1347 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
1348 * sFlow, as does a null pointer. */
1349 int (*set_sflow)(struct ofproto *ofproto,
1350 const struct ofproto_sflow_options *sflow_options);
1352 /* Configures IPFIX on 'ofproto' according to the options in
1353 * 'bridge_exporter_options' and the 'flow_exporters_options'
1354 * array, or turns off IPFIX if 'bridge_exporter_options' and
1355 * 'flow_exporters_options' is NULL.
1357 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
1358 * IPFIX, as does a null pointer. */
1360 struct ofproto *ofproto,
1361 const struct ofproto_ipfix_bridge_exporter_options
1362 *bridge_exporter_options,
1363 const struct ofproto_ipfix_flow_exporter_options
1364 *flow_exporters_options, size_t n_flow_exporters_options);
1366 /* Configures connectivity fault management on 'ofport'.
1368 * If 'cfm_settings' is nonnull, configures CFM according to its members.
1370 * If 'cfm_settings' is null, removes any connectivity fault management
1371 * configuration from 'ofport'.
1373 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1374 * support CFM, as does a null pointer. */
1375 int (*set_cfm)(struct ofport *ofport, const struct cfm_settings *s);
1377 /* Checks the status change of CFM on 'ofport'. Returns true if
1378 * there is status change since last call or if CFM is not specified. */
1379 bool (*cfm_status_changed)(struct ofport *ofport);
1381 /* Populates 'smap' with the status of CFM on 'ofport'. Returns 0 on
1382 * success, or a positive errno. EOPNOTSUPP as a return value indicates
1383 * that this ofproto_class does not support CFM, as does a null pointer.
1385 * The caller must provide and own '*status', and it must free the array
1386 * returned in 'status->rmps'. '*status' is indeterminate if the return
1387 * value is non-zero. */
1388 int (*get_cfm_status)(const struct ofport *ofport,
1389 struct cfm_status *status);
1391 /* Configures LLDP on 'ofport'.
1393 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1394 * support LLDP, as does a null pointer. */
1395 int (*set_lldp)(struct ofport *ofport, const struct smap *cfg);
1397 /* Checks the status of LLDP configured on 'ofport'. Returns true if the
1398 * port's LLDP status was successfully stored into '*status'. Returns
1399 * false if the port did not have LLDP configured, in which case '*status'
1402 * The caller must provide and own '*status'. '*status' is indeterminate
1403 * if the return value is non-zero. */
1404 bool (*get_lldp_status)(const struct ofport *ofport,
1405 struct lldp_status *status);
1407 /* Configures Auto Attach.
1409 * If 's' is nonnull, configures Auto Attach according to its members.
1411 * If 's' is null, removes any Auto Attach configuration.
1413 int (*set_aa)(struct ofproto *ofproto,
1414 const struct aa_settings *s);
1416 /* If 's' is nonnull, this function registers a mapping associated with
1417 * client data pointer 'aux' in 'ofproto'. If 'aux' is already registered
1418 * then this function updates its configuration to 's'. Otherwise, this
1419 * function registers a new mapping.
1421 * An implementation that does not support mapping at all may set
1422 * it to NULL or return EOPNOTSUPP. An implementation that supports
1423 * only a subset of the functionality should implement what it can
1426 int (*aa_mapping_set)(struct ofproto *ofproto, void *aux,
1427 const struct aa_mapping_settings *s);
1429 /* If 's' is nonnull, this function unregisters a mapping associated with
1430 * client data pointer 'aux' in 'ofproto'. If 'aux' is already registered
1431 * then this function updates its configuration to 's'. Otherwise, this
1432 * function unregisters a new mapping.
1434 * An implementation that does not support mapping at all may set
1435 * it to NULL or return EOPNOTSUPP. An implementation that supports
1436 * only a subset of the functionality should implement what it can
1439 int (*aa_mapping_unset)(struct ofproto *ofproto, void *aux);
1442 * Returns the a list of AutoAttach VLAN operations. When Auto Attach is
1443 * enabled, the VLAN associated with an I-SID/VLAN mapping is first
1444 * negotiated with an Auto Attach Server. Once an I-SID VLAN mapping
1445 * becomes active, the corresponding VLAN needs to be communicated to the
1446 * bridge in order to add the VLAN to the trunk port linking the Auto
1447 * Attach Client (in this case openvswitch) and the Auto Attach Server.
1449 * The list entries are of type "struct bridge_aa_vlan". Each entry
1450 * specifies the operation (add or remove), the interface on which to
1451 * execute the operation and the VLAN.
1453 int (*aa_vlan_get_queued)(struct ofproto *ofproto, struct ovs_list *list);
1456 * Returns the current number of entries in the list of VLAN operations
1457 * in the Auto Attach Client (see previous function description
1458 * aa_vlan_get_queued). Returns 0 if Auto Attach is disabled.
1460 unsigned int (*aa_vlan_get_queue_size)(struct ofproto *ofproto);
1462 /* Configures BFD on 'ofport'.
1464 * If 'cfg' is NULL, or 'cfg' does not contain the key value pair
1465 * "enable=true", removes BFD from 'ofport'. Otherwise, configures BFD
1466 * according to 'cfg'.
1468 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1469 * support BFD, as does a null pointer. */
1470 int (*set_bfd)(struct ofport *ofport, const struct smap *cfg);
1472 /* Checks the status change of BFD on 'ofport'. Returns true if there
1473 * is status change since last call or if BFD is not specified. */
1474 bool (*bfd_status_changed)(struct ofport *ofport);
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 /* Configures Rapid Spanning Tree Protocol (RSTP) on 'ofproto' using the
1539 * settings defined in 's'.
1541 * If 's' is nonnull, configures RSTP according to its members.
1543 * If 's' is null, removes any RSTP configuration from 'ofproto'.
1545 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1546 * support RSTP, as does a null pointer. */
1547 void (*set_rstp)(struct ofproto *ofproto,
1548 const struct ofproto_rstp_settings *s);
1550 /* Retrieves state of Rapid Spanning Tree Protocol (RSTP) on 'ofproto'.
1552 * Stores RSTP state for 'ofproto' in 's'. If the 'enabled' member
1553 * is false, the other member values are not meaningful.
1555 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1556 * support RSTP, as does a null pointer. */
1557 void (*get_rstp_status)(struct ofproto *ofproto,
1558 struct ofproto_rstp_status *s);
1560 /* Configures Rapid Spanning Tree Protocol (RSTP) on 'ofport' using the
1561 * settings defined in 's'.
1563 * If 's' is nonnull, configures RSTP according to its members. The
1564 * caller is responsible for assigning RSTP port numbers (using the
1565 * 'port_num' member in the range of 1 through 255, inclusive) and
1566 * ensuring there are no duplicates.
1568 * If 's' is null, removes any RSTP configuration from 'ofport'.
1570 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1571 * support STP, as does a null pointer. */
1572 void (*set_rstp_port)(struct ofport *ofport,
1573 const struct ofproto_port_rstp_settings *s);
1575 /* Retrieves Rapid Spanning Tree Protocol (RSTP) port status of 'ofport'.
1577 * Stores RSTP state for 'ofport' in 's'. If the 'enabled' member is
1578 * false, the other member values are not meaningful.
1580 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1581 * support RSTP, as does a null pointer. */
1582 void (*get_rstp_port_status)(struct ofport *ofport,
1583 struct ofproto_port_rstp_status *s);
1585 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
1586 * 'queues' attached to 'ofport'. This data is not intended to be
1587 * sufficient to implement QoS. Instead, providers may use this
1588 * information to implement features which require knowledge of what queues
1589 * exist on a port, and some basic information about them.
1591 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1592 * support QoS, as does a null pointer. */
1593 int (*set_queues)(struct ofport *ofport,
1594 const struct ofproto_port_queue *queues, size_t n_qdscp);
1596 /* If 's' is nonnull, this function registers a "bundle" associated with
1597 * client data pointer 'aux' in 'ofproto'. A bundle is the same concept as
1598 * a Port in OVSDB, that is, it consists of one or more "slave" devices
1599 * (Interfaces, in OVSDB) along with VLAN and LACP configuration and, if
1600 * there is more than one slave, a bonding configuration. If 'aux' is
1601 * already registered then this function updates its configuration to 's'.
1602 * Otherwise, this function registers a new bundle.
1604 * If 's' is NULL, this function unregisters the bundle registered on
1605 * 'ofproto' associated with client data pointer 'aux'. If no such bundle
1606 * has been registered, this has no effect.
1608 * This function affects only the behavior of the NXAST_AUTOPATH action and
1609 * output to the OFPP_NORMAL port. An implementation that does not support
1610 * it at all may set it to NULL or return EOPNOTSUPP. An implementation
1611 * that supports only a subset of the functionality should implement what
1612 * it can and return 0. */
1613 int (*bundle_set)(struct ofproto *ofproto, void *aux,
1614 const struct ofproto_bundle_settings *s);
1616 /* If 'port' is part of any bundle, removes it from that bundle. If the
1617 * bundle now has no ports, deletes the bundle. If the bundle now has only
1618 * one port, deconfigures the bundle's bonding configuration. */
1619 void (*bundle_remove)(struct ofport *ofport);
1621 /* If 's' is nonnull, this function registers a mirror associated with
1622 * client data pointer 'aux' in 'ofproto'. A mirror is the same concept as
1623 * a Mirror in OVSDB. If 'aux' is already registered then this function
1624 * updates its configuration to 's'. Otherwise, this function registers a
1627 * If 's' is NULL, this function unregisters the mirror registered on
1628 * 'ofproto' associated with client data pointer 'aux'. If no such mirror
1629 * has been registered, this has no effect.
1631 * An implementation that does not support mirroring at all may set
1632 * it to NULL or return EOPNOTSUPP. An implementation that supports
1633 * only a subset of the functionality should implement what it can
1635 int (*mirror_set)(struct ofproto *ofproto, void *aux,
1636 const struct ofproto_mirror_settings *s);
1638 /* Retrieves statistics from mirror associated with client data
1639 * pointer 'aux' in 'ofproto'. Stores packet and byte counts in
1640 * 'packets' and 'bytes', respectively. If a particular counter is
1641 * not supported, the appropriate argument is set to UINT64_MAX.
1643 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1644 * support retrieving mirror statistics. */
1645 int (*mirror_get_stats)(struct ofproto *ofproto, void *aux,
1646 uint64_t *packets, uint64_t *bytes);
1648 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs
1649 * on which all packets are flooded, instead of using MAC learning. If
1650 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1652 * This function affects only the behavior of the OFPP_NORMAL action. An
1653 * implementation that does not support it may set it to NULL or return
1655 int (*set_flood_vlans)(struct ofproto *ofproto,
1656 unsigned long *flood_vlans);
1658 /* Returns true if 'aux' is a registered bundle that is currently in use as
1659 * the output for a mirror. */
1660 bool (*is_mirror_output_bundle)(const struct ofproto *ofproto, void *aux);
1662 /* When the configuration option of forward_bpdu changes, this function
1663 * will be invoked. */
1664 void (*forward_bpdu_changed)(struct ofproto *ofproto);
1666 /* Sets the MAC aging timeout for the OFPP_NORMAL action to 'idle_time', in
1667 * seconds, and the maximum number of MAC table entries to
1670 * An implementation that doesn't support configuring these features may
1671 * set this function to NULL or implement it as a no-op. */
1672 void (*set_mac_table_config)(struct ofproto *ofproto,
1673 unsigned int idle_time, size_t max_entries);
1675 /* Configures multicast snooping on 'ofport' using the settings
1678 * If 's' is nonnull, this function updates multicast snooping
1679 * configuration to 's' in 'ofproto'.
1681 * If 's' is NULL, this function disables multicast snooping
1684 * An implementation that does not support multicast snooping may set
1685 * it to NULL or return EOPNOTSUPP. */
1686 int (*set_mcast_snooping)(struct ofproto *ofproto,
1687 const struct ofproto_mcast_snooping_settings *s);
1689 /* Configures multicast snooping port's flood setting on 'ofproto'.
1691 * If 's' is nonnull, this function updates multicast snooping
1692 * configuration to 's' in 'ofproto'.
1694 * If 's' is NULL, this function doesn't change anything.
1696 * An implementation that does not support multicast snooping may set
1697 * it to NULL or return EOPNOTSUPP. */
1698 int (*set_mcast_snooping_port)(struct ofproto *ofproto_, void *aux,
1699 const struct ofproto_mcast_snooping_port_settings *s);
1701 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
1703 * This is deprecated. It is only for compatibility with broken device drivers
1704 * in old versions of Linux that do not properly support VLANs when VLAN
1705 * devices are not used. When broken device drivers are no longer in
1706 * widespread use, we will delete these interfaces. */
1708 /* If 'realdev_ofp_port' is nonzero, then this function configures 'ofport'
1709 * as a VLAN splinter port for VLAN 'vid', associated with the real device
1710 * that has OpenFlow port number 'realdev_ofp_port'.
1712 * If 'realdev_ofp_port' is zero, then this function deconfigures 'ofport'
1713 * as a VLAN splinter port.
1715 * This function should be NULL if an implementation does not support it.
1717 int (*set_realdev)(struct ofport *ofport,
1718 ofp_port_t realdev_ofp_port, int vid);
1720 /* ## ------------------------ ## */
1721 /* ## OpenFlow meter functions ## */
1722 /* ## ------------------------ ## */
1724 /* These functions should be NULL if an implementation does not support
1725 * them. They must be all null or all non-null.. */
1727 /* Initializes 'features' to describe the metering features supported by
1729 void (*meter_get_features)(const struct ofproto *ofproto,
1730 struct ofputil_meter_features *features);
1732 /* If '*id' is UINT32_MAX, adds a new meter with the given 'config'. On
1733 * success the function must store a provider meter ID other than
1734 * UINT32_MAX in '*id'. All further references to the meter will be made
1735 * with the returned provider meter id rather than the OpenFlow meter id.
1736 * The caller does not try to interpret the provider meter id, giving the
1737 * implementation the freedom to either use the OpenFlow meter_id value
1738 * provided in the meter configuration, or any other value suitable for the
1741 * If '*id' is a value other than UINT32_MAX, modifies the existing meter
1742 * with that meter provider ID to have configuration 'config', while
1743 * leaving '*id' unchanged. On failure, the existing meter configuration
1744 * is left intact. */
1745 enum ofperr (*meter_set)(struct ofproto *ofproto, ofproto_meter_id *id,
1746 const struct ofputil_meter_config *config);
1748 /* Gets the meter and meter band packet and byte counts for maximum of
1749 * 'stats->n_bands' bands for the meter with provider ID 'id' within
1750 * 'ofproto'. The caller fills in the other stats values. The band stats
1751 * are copied to memory at 'stats->bands' provided by the caller. The
1752 * number of returned band stats is returned in 'stats->n_bands'. */
1753 enum ofperr (*meter_get)(const struct ofproto *ofproto,
1754 ofproto_meter_id id,
1755 struct ofputil_meter_stats *stats);
1757 /* Deletes a meter, making the 'ofproto_meter_id' invalid for any
1759 void (*meter_del)(struct ofproto *, ofproto_meter_id);
1762 /* ## -------------------- ## */
1763 /* ## OpenFlow 1.1+ groups ## */
1764 /* ## -------------------- ## */
1766 struct ofgroup *(*group_alloc)(void);
1767 enum ofperr (*group_construct)(struct ofgroup *);
1768 void (*group_destruct)(struct ofgroup *);
1769 void (*group_dealloc)(struct ofgroup *);
1771 enum ofperr (*group_modify)(struct ofgroup *);
1773 enum ofperr (*group_get_stats)(const struct ofgroup *,
1774 struct ofputil_group_stats *);
1776 /* ## --------------------- ## */
1777 /* ## Datapath information ## */
1778 /* ## --------------------- ## */
1779 /* Retrieve the version string of the datapath. The version
1780 * string can be NULL if it can not be determined.
1782 * The version retuned is read only. The caller should not
1785 * This function should be NULL if an implementation does not support it.
1787 const char *(*get_datapath_version)(const struct ofproto *);
1790 extern const struct ofproto_class ofproto_dpif_class;
1792 int ofproto_class_register(const struct ofproto_class *);
1793 int ofproto_class_unregister(const struct ofproto_class *);
1795 /* flow_mod with execution context. */
1796 struct ofproto_flow_mod {
1797 struct ofputil_flow_mod fm;
1799 cls_version_t version; /* Version in which changes take
1801 struct rule_collection old_rules; /* Affected rules. */
1802 struct rule_collection new_rules; /* Replacement rules. */
1805 /* port_mod with execution context. */
1806 struct ofproto_port_mod {
1807 struct ofputil_port_mod pm;
1808 struct ofport *port; /* Affected port. */
1811 enum ofperr ofproto_flow_mod(struct ofproto *, struct ofproto_flow_mod *)
1812 OVS_EXCLUDED(ofproto_mutex);
1813 void ofproto_add_flow(struct ofproto *, const struct match *, int priority,
1814 const struct ofpact *ofpacts, size_t ofpacts_len)
1815 OVS_EXCLUDED(ofproto_mutex);
1816 void ofproto_delete_flow(struct ofproto *, const struct match *, int priority)
1817 OVS_EXCLUDED(ofproto_mutex);
1818 void ofproto_flush_flows(struct ofproto *);
1820 enum ofperr ofproto_check_ofpacts(struct ofproto *,
1821 const struct ofpact ofpacts[],
1822 size_t ofpacts_len);
1824 static inline const struct rule_actions *
1825 rule_get_actions(const struct rule *rule)
1827 return rule->actions;
1830 /* Returns true if 'rule' is an OpenFlow 1.3 "table-miss" rule, false
1833 * ("Table-miss" rules are special because a packet_in generated through one
1834 * uses OFPR_NO_MATCH as its reason, whereas packet_ins generated by any other
1835 * rule use OFPR_ACTION.) */
1837 rule_is_table_miss(const struct rule *rule)
1839 return rule->cr.priority == 0 && cls_rule_is_catchall(&rule->cr);
1842 /* Returns true if 'rule' should be hidden from the controller.
1844 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1845 * (e.g. by in-band control) and are intentionally hidden from the
1848 rule_is_hidden(const struct rule *rule)
1850 return rule->cr.priority > UINT16_MAX;
1853 static inline struct rule *
1854 rule_from_cls_rule(const struct cls_rule *cls_rule)
1856 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1859 #endif /* ofproto/ofproto-provider.h */