2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
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. */
22 #include "ofproto/ofproto.h"
24 #include "classifier.h"
27 #include "ofp-errors.h"
32 struct ofputil_flow_mod;
34 /* An OpenFlow switch.
36 * With few exceptions, ofproto implementations may look at these fields but
37 * should not modify them. */
39 const struct ofproto_class *ofproto_class;
40 char *type; /* Datapath type. */
41 char *name; /* Datapath name. */
42 struct hmap_node hmap_node; /* In global 'all_ofprotos' hmap. */
45 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
46 uint64_t datapath_id; /* Datapath ID. */
47 unsigned flow_eviction_threshold; /* Threshold at which to begin flow
48 * table eviction. Only affects the
49 * ofproto-dpif implementation */
50 bool forward_bpdu; /* Option to allow forwarding of BPDU frames
51 * when NORMAL action is invoked. */
52 char *mfr_desc; /* Manufacturer. */
53 char *hw_desc; /* Hardware. */
54 char *sw_desc; /* Software version. */
55 char *serial_desc; /* Serial number. */
56 char *dp_desc; /* Datapath description. */
57 enum ofp_config_flags frag_handling; /* One of OFPC_*. */
60 struct hmap ports; /* Contains "struct ofport"s. */
61 struct shash port_by_name;
64 struct oftable *tables;
67 /* OpenFlow connections. */
68 struct connmgr *connmgr;
70 /* Flow table operation tracking. */
71 int state; /* Internal state. */
72 struct list pending; /* List of "struct ofopgroup"s. */
73 unsigned int n_pending; /* list_size(&pending). */
74 struct hmap deletions; /* All OFOPERATION_DELETE "ofoperation"s. */
76 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
78 * This is deprecated. It is only for compatibility with broken device
79 * drivers in old versions of Linux that do not properly support VLANs when
80 * VLAN devices are not used. When broken device drivers are no longer in
81 * widespread use, we will delete these interfaces. */
82 unsigned long int *vlan_bitmap; /* 4096-bit bitmap of in-use VLANs. */
83 bool vlans_changed; /* True if new VLANs are in use. */
86 void ofproto_init_tables(struct ofproto *, int n_tables);
88 struct ofproto *ofproto_lookup(const char *name);
89 struct ofport *ofproto_get_port(const struct ofproto *, uint16_t ofp_port);
91 /* An OpenFlow port within a "struct ofproto".
93 * With few exceptions, ofproto implementations may look at these fields but
94 * should not modify them. */
96 struct ofproto *ofproto; /* The ofproto that contains this port. */
97 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
98 struct netdev *netdev;
99 struct ofputil_phy_port pp;
100 uint16_t ofp_port; /* OpenFlow port number. */
101 unsigned int change_seq;
105 void ofproto_port_set_state(struct ofport *, enum ofputil_port_state);
108 OFTABLE_HIDDEN = 1 << 0, /* Hide from most OpenFlow operations. */
109 OFTABLE_READONLY = 1 << 1 /* Don't allow OpenFlow to change this table. */
112 /* A flow table within a "struct ofproto". */
114 enum oftable_flags flags;
115 struct classifier cls; /* Contains "struct rule"s. */
116 char *name; /* Table name exposed via OpenFlow, or NULL. */
118 /* Maximum number of flows or UINT_MAX if there is no limit besides any
119 * limit imposed by resource limitations. */
120 unsigned int max_flows;
122 /* These members determine the handling of an attempt to add a flow that
123 * would cause the table to have more than 'max_flows' flows.
125 * If 'eviction_fields' is NULL, overflows will be rejected with an error.
127 * If 'eviction_fields' is nonnull (regardless of whether n_eviction_fields
128 * is nonzero), an overflow will cause a flow to be removed. The flow to
129 * be removed is chosen to give fairness among groups distinguished by
130 * different values for the subfields within 'groups'. */
131 struct mf_subfield *eviction_fields;
132 size_t n_eviction_fields;
136 * When a flow is added that would cause the table to have more than
137 * 'max_flows' flows, and 'eviction_fields' is nonnull, these groups are
138 * used to decide which rule to evict: the rule is chosen from the eviction
139 * group that contains the greatest number of rules.*/
140 uint32_t eviction_group_id_basis;
141 struct hmap eviction_groups_by_id;
142 struct heap eviction_groups_by_size;
145 /* Assigns TABLE to each oftable, in turn, in OFPROTO.
147 * All parameters are evaluated multiple times. */
148 #define OFPROTO_FOR_EACH_TABLE(TABLE, OFPROTO) \
149 for ((TABLE) = (OFPROTO)->tables; \
150 (TABLE) < &(OFPROTO)->tables[(OFPROTO)->n_tables]; \
153 /* An OpenFlow flow within a "struct ofproto".
155 * With few exceptions, ofproto implementations may look at these fields but
156 * should not modify them. */
158 struct ofproto *ofproto; /* The ofproto that contains this rule. */
159 struct list ofproto_node; /* Owned by ofproto base code. */
160 struct cls_rule cr; /* In owning ofproto's classifier. */
162 struct ofoperation *pending; /* Operation now in progress, if nonnull. */
164 ovs_be64 flow_cookie; /* Controller-issued identifier. */
166 long long int created; /* Creation time. */
167 long long int modified; /* Time of last modification. */
168 long long int used; /* Last use; time created if never used. */
169 uint16_t hard_timeout; /* In seconds from ->modified. */
170 uint16_t idle_timeout; /* In seconds from ->used. */
171 uint8_t table_id; /* Index in ofproto's 'tables' array. */
172 bool send_flow_removed; /* Send a flow removed message? */
174 /* Eviction groups. */
175 bool evictable; /* If false, prevents eviction. */
176 struct heap_node evg_node; /* In eviction_group's "rules" heap. */
177 struct eviction_group *eviction_group; /* NULL if not in any group. */
179 union ofp_action *actions; /* OpenFlow actions. */
180 int n_actions; /* Number of elements in actions[]. */
183 static inline struct rule *
184 rule_from_cls_rule(const struct cls_rule *cls_rule)
186 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
189 void ofproto_rule_update_used(struct rule *, long long int used);
190 void ofproto_rule_expire(struct rule *, uint8_t reason);
191 void ofproto_rule_destroy(struct rule *);
193 void ofoperation_complete(struct ofoperation *, enum ofperr);
194 struct rule *ofoperation_get_victim(struct ofoperation *);
196 /* ofproto class structure, to be defined by each ofproto implementation.
202 * These functions work primarily with three different kinds of data
205 * - "struct ofproto", which represents an OpenFlow switch.
207 * - "struct ofport", which represents a port within an ofproto.
209 * - "struct rule", which represents an OpenFlow flow within an ofproto.
211 * Each of these data structures contains all of the implementation-independent
212 * generic state for the respective concept, called the "base" state. None of
213 * them contains any extra space for ofproto implementations to use. Instead,
214 * each implementation is expected to declare its own data structure that
215 * contains an instance of the generic data structure plus additional
216 * implementation-specific members, called the "derived" state. The
217 * implementation can use casts or (preferably) the CONTAINER_OF macro to
218 * obtain access to derived state given only a pointer to the embedded generic
225 * Four stylized functions accompany each of these data structures:
227 * "alloc" "construct" "destruct" "dealloc"
228 * ------------ ---------------- --------------- --------------
229 * ofproto ->alloc ->construct ->destruct ->dealloc
230 * ofport ->port_alloc ->port_construct ->port_destruct ->port_dealloc
231 * rule ->rule_alloc ->rule_construct ->rule_destruct ->rule_dealloc
233 * Any instance of a given data structure goes through the following life
236 * 1. The client calls the "alloc" function to obtain raw memory. If "alloc"
237 * fails, skip all the other steps.
239 * 2. The client initializes all of the data structure's base state. If this
240 * fails, skip to step 7.
242 * 3. The client calls the "construct" function. The implementation
243 * initializes derived state. It may refer to the already-initialized
244 * base state. If "construct" fails, skip to step 6.
246 * 4. The data structure is now initialized and in use.
248 * 5. When the data structure is no longer needed, the client calls the
249 * "destruct" function. The implementation uninitializes derived state.
250 * The base state has not been uninitialized yet, so the implementation
251 * may still refer to it.
253 * 6. The client uninitializes all of the data structure's base state.
255 * 7. The client calls the "dealloc" to free the raw memory. The
256 * implementation must not refer to base or derived state in the data
257 * structure, because it has already been uninitialized.
259 * Each "alloc" function allocates and returns a new instance of the respective
260 * data structure. The "alloc" function is not given any information about the
261 * use of the new data structure, so it cannot perform much initialization.
262 * Its purpose is just to ensure that the new data structure has enough room
263 * for base and derived state. It may return a null pointer if memory is not
264 * available, in which case none of the other functions is called.
266 * Each "construct" function initializes derived state in its respective data
267 * structure. When "construct" is called, all of the base state has already
268 * been initialized, so the "construct" function may refer to it. The
269 * "construct" function is allowed to fail, in which case the client calls the
270 * "dealloc" function (but not the "destruct" function).
272 * Each "destruct" function uninitializes and frees derived state in its
273 * respective data structure. When "destruct" is called, the base state has
274 * not yet been uninitialized, so the "destruct" function may refer to it. The
275 * "destruct" function is not allowed to fail.
277 * Each "dealloc" function frees raw memory that was allocated by the the
278 * "alloc" function. The memory's base and derived members might not have ever
279 * been initialized (but if "construct" returned successfully, then it has been
280 * "destruct"ed already). The "dealloc" function is not allowed to fail.
286 * Most of these functions return 0 if they are successful or a positive error
287 * code on failure. Depending on the function, valid error codes are either
288 * errno values or OFPERR_* OpenFlow error codes.
290 * Most of these functions are expected to execute synchronously, that is, to
291 * block as necessary to obtain a result. Thus, these functions may return
292 * EAGAIN (or EWOULDBLOCK or EINPROGRESS) only where the function descriptions
293 * explicitly say those errors are a possibility. We may relax this
294 * requirement in the future if and when we encounter performance problems. */
295 struct ofproto_class {
296 /* ## ----------------- ## */
297 /* ## Factory Functions ## */
298 /* ## ----------------- ## */
300 /* Enumerates the types of all support ofproto types into 'types'. The
301 * caller has already initialized 'types' and other ofproto classes might
302 * already have added names to it. */
303 void (*enumerate_types)(struct sset *types);
305 /* Enumerates the names of all existing datapath of the specified 'type'
306 * into 'names' 'all_dps'. The caller has already initialized 'names' as
309 * 'type' is one of the types enumerated by ->enumerate_types().
311 * Returns 0 if successful, otherwise a positive errno value.
313 int (*enumerate_names)(const char *type, struct sset *names);
315 /* Deletes the datapath with the specified 'type' and 'name'. The caller
316 * should have closed any open ofproto with this 'type' and 'name'; this
317 * function is allowed to fail if that is not the case.
319 * 'type' is one of the types enumerated by ->enumerate_types().
320 * 'name' is one of the names enumerated by ->enumerate_names() for 'type'.
322 * Returns 0 if successful, otherwise a positive errno value.
324 int (*del)(const char *type, const char *name);
326 /* ## --------------------------- ## */
327 /* ## Top-Level ofproto Functions ## */
328 /* ## --------------------------- ## */
330 /* Life-cycle functions for an "ofproto" (see "Life Cycle" above).
336 * ->construct() should not modify any base members of the ofproto. The
337 * client will initialize the ofproto's 'ports' and 'tables' members after
338 * construction is complete.
340 * When ->construct() is called, the client does not yet know how many flow
341 * tables the datapath supports, so ofproto->n_tables will be 0 and
342 * ofproto->tables will be NULL. ->construct() should call
343 * ofproto_init_tables() to allocate and initialize ofproto->n_tables and
344 * ofproto->tables. Each flow table will be initially empty, so
345 * ->construct() should delete flows from the underlying datapath, if
346 * necessary, rather than populating the tables.
348 * Only one ofproto instance needs to be supported for any given datapath.
349 * If a datapath is already open as part of one "ofproto", then another
350 * attempt to "construct" the same datapath as part of another ofproto is
351 * allowed to fail with an error.
353 * ->construct() returns 0 if successful, otherwise a positive errno
360 * If 'ofproto' has any pending asynchronous operations, ->destruct()
361 * must complete all of them by calling ofoperation_complete().
363 * ->destruct() must also destroy all remaining rules in the ofproto's
364 * tables, by passing each remaining rule to ofproto_rule_destroy(). The
365 * client will destroy the flow tables themselves after ->destruct()
368 struct ofproto *(*alloc)(void);
369 int (*construct)(struct ofproto *ofproto);
370 void (*destruct)(struct ofproto *ofproto);
371 void (*dealloc)(struct ofproto *ofproto);
373 /* Performs any periodic activity required by 'ofproto'. It should:
375 * - Call connmgr_send_packet_in() for each received packet that missed
376 * in the OpenFlow flow table or that had a OFPP_CONTROLLER output
379 * - Call ofproto_rule_expire() for each OpenFlow flow that has reached
380 * its hard_timeout or idle_timeout, to expire the flow.
382 * Returns 0 if successful, otherwise a positive errno value. */
383 int (*run)(struct ofproto *ofproto);
385 /* Performs periodic activity required by 'ofproto' that needs to be done
386 * with the least possible latency.
388 * This is run multiple times per main loop. An ofproto provider may
389 * implement it or not, according to whether it provides a performance
390 * boost for that ofproto implementation. */
391 int (*run_fast)(struct ofproto *ofproto);
393 /* Causes the poll loop to wake up when 'ofproto''s 'run' function needs to
394 * be called, e.g. by calling the timer or fd waiting functions in
396 void (*wait)(struct ofproto *ofproto);
398 /* Every "struct rule" in 'ofproto' is about to be deleted, one by one.
399 * This function may prepare for that, for example by clearing state in
400 * advance. It should *not* actually delete any "struct rule"s from
401 * 'ofproto', only prepare for it.
403 * This function is optional; it's really just for optimization in case
404 * it's cheaper to delete all the flows from your hardware in a single pass
405 * than to do it one by one. */
406 void (*flush)(struct ofproto *ofproto);
408 /* Helper for the OpenFlow OFPT_FEATURES_REQUEST request.
410 * The implementation should store true in '*arp_match_ip' if the switch
411 * supports matching IP addresses inside ARP requests and replies, false
414 * The implementation should store in '*actions' a bitmap of the supported
415 * OpenFlow actions. Vendor actions are not included in '*actions'. */
416 void (*get_features)(struct ofproto *ofproto,
418 enum ofputil_action_bitmap *actions);
420 /* Helper for the OpenFlow OFPST_TABLE statistics request.
422 * The 'ots' array contains 'ofproto->n_tables' elements. Each element is
425 * - 'table_id' to the array index.
427 * - 'name' to "table#" where # is the table ID.
429 * - 'wildcards' to OFPFW_ALL.
431 * - 'max_entries' to 1,000,000.
433 * - 'active_count' to the classifier_count() for the table.
435 * - 'lookup_count' and 'matched_count' to 0.
437 * The implementation should update any members in each element for which
438 * it has better values:
440 * - 'name' to a more meaningful name.
442 * - 'wildcards' to the set of wildcards actually supported by the table
443 * (if it doesn't support all OpenFlow wildcards).
445 * - 'max_entries' to the maximum number of flows actually supported by
448 * - 'lookup_count' to the number of packets looked up in this flow table
451 * - 'matched_count' to the number of packets looked up in this flow
452 * table so far that matched one of the flow entries.
454 * Keep in mind that all of the members of struct ofp_table_stats are in
455 * network byte order.
457 void (*get_tables)(struct ofproto *ofproto, struct ofp_table_stats *ots);
459 /* ## ---------------- ## */
460 /* ## ofport Functions ## */
461 /* ## ---------------- ## */
463 /* Life-cycle functions for a "struct ofport" (see "Life Cycle" above).
465 * ->port_construct() should not modify any base members of the ofport.
467 * ofports are managed by the base ofproto code. The ofproto
468 * implementation should only create and destroy them in response to calls
469 * to these functions. The base ofproto code will create and destroy
470 * ofports in the following situations:
472 * - Just after the ->construct() function is called, the base ofproto
473 * iterates over all of the implementation's ports, using
474 * ->port_dump_start() and related functions, and constructs an ofport
475 * for each dumped port.
477 * - If ->port_poll() reports that a specific port has changed, then the
478 * base ofproto will query that port with ->port_query_by_name() and
479 * construct or destruct ofports as necessary to reflect the updated
482 * - If ->port_poll() returns ENOBUFS to report an unspecified port set
483 * change, then the base ofproto will iterate over all of the
484 * implementation's ports, in the same way as at ofproto
485 * initialization, and construct and destruct ofports to reflect all of
488 * ->port_construct() returns 0 if successful, otherwise a positive errno
491 struct ofport *(*port_alloc)(void);
492 int (*port_construct)(struct ofport *ofport);
493 void (*port_destruct)(struct ofport *ofport);
494 void (*port_dealloc)(struct ofport *ofport);
496 /* Called after 'ofport->netdev' is replaced by a new netdev object. If
497 * the ofproto implementation uses the ofport's netdev internally, then it
498 * should switch to using the new one. The old one has been closed.
500 * An ofproto implementation that doesn't need to do anything in this
501 * function may use a null pointer. */
502 void (*port_modified)(struct ofport *ofport);
504 /* Called after an OpenFlow request changes a port's configuration.
505 * 'ofport->pp.config' contains the new configuration. 'old_config'
506 * contains the previous configuration.
508 * The caller implements OFPUTIL_PC_PORT_DOWN using netdev functions to
509 * turn NETDEV_UP on and off, so this function doesn't have to do anything
510 * for that bit (and it won't be called if that is the only bit that
512 void (*port_reconfigured)(struct ofport *ofport,
513 enum ofputil_port_config old_config);
515 /* Looks up a port named 'devname' in 'ofproto'. On success, initializes
516 * '*port' appropriately.
518 * The caller owns the data in 'port' and must free it with
519 * ofproto_port_destroy() when it is no longer needed. */
520 int (*port_query_by_name)(const struct ofproto *ofproto,
521 const char *devname, struct ofproto_port *port);
523 /* Attempts to add 'netdev' as a port on 'ofproto'. Returns 0 if
524 * successful, otherwise a positive errno value. If successful, sets
525 * '*ofp_portp' to the new port's port number.
527 * It doesn't matter whether the new port will be returned by a later call
528 * to ->port_poll(); the implementation may do whatever is more
530 int (*port_add)(struct ofproto *ofproto, struct netdev *netdev,
531 uint16_t *ofp_portp);
533 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'. Returns
534 * 0 if successful, otherwise a positive errno value.
536 * It doesn't matter whether the new port will be returned by a later call
537 * to ->port_poll(); the implementation may do whatever is more
539 int (*port_del)(struct ofproto *ofproto, uint16_t ofp_port);
542 int (*port_get_stats)(const struct ofport *port,
543 struct netdev_stats *stats);
545 /* Port iteration functions.
547 * The client might not be entirely in control of the ports within an
548 * ofproto. Some hardware implementations, for example, might have a fixed
549 * set of ports in a datapath, and the Linux datapath allows the system
550 * administrator to externally add and remove ports with ovs-dpctl. For
551 * this reason, the client needs a way to iterate through all the ports
552 * that are actually in a datapath. These functions provide that
555 * The 'state' pointer provides the implementation a place to
556 * keep track of its position. Its format is opaque to the caller.
558 * The ofproto provider retains ownership of the data that it stores into
559 * ->port_dump_next()'s 'port' argument. The data must remain valid until
560 * at least the next call to ->port_dump_next() or ->port_dump_done() for
561 * 'state'. The caller will not modify or free it.
566 * ->port_dump_start() attempts to begin dumping the ports in 'ofproto'.
567 * On success, it should return 0 and initialize '*statep' with any data
568 * needed for iteration. On failure, returns a positive errno value, and
569 * the client will not call ->port_dump_next() or ->port_dump_done().
571 * ->port_dump_next() attempts to retrieve another port from 'ofproto' for
572 * 'state'. If there is another port, it should store the port's
573 * information into 'port' and return 0. It should return EOF if all ports
574 * have already been iterated. Otherwise, on error, it should return a
575 * positive errno value. This function will not be called again once it
576 * returns nonzero once for a given iteration (but the 'port_dump_done'
577 * function will be called afterward).
579 * ->port_dump_done() allows the implementation to release resources used
580 * for iteration. The caller might decide to stop iteration in the middle
581 * by calling this function before ->port_dump_next() returns nonzero.
589 * error = ofproto->ofproto_class->port_dump_start(ofproto, &state);
592 * struct ofproto_port port;
594 * error = ofproto->ofproto_class->port_dump_next(
595 * ofproto, state, &port);
599 * // Do something with 'port' here (without modifying or freeing
600 * // any of its data).
602 * ofproto->ofproto_class->port_dump_done(ofproto, state);
604 * // 'error' is now EOF (success) or a positive errno value (failure).
606 int (*port_dump_start)(const struct ofproto *ofproto, void **statep);
607 int (*port_dump_next)(const struct ofproto *ofproto, void *state,
608 struct ofproto_port *port);
609 int (*port_dump_done)(const struct ofproto *ofproto, void *state);
611 /* Polls for changes in the set of ports in 'ofproto'. If the set of ports
612 * in 'ofproto' has changed, then this function should do one of the
615 * - Preferably: store the name of the device that was added to or deleted
616 * from 'ofproto' in '*devnamep' and return 0. The caller is responsible
617 * for freeing '*devnamep' (with free()) when it no longer needs it.
619 * - Alternatively: return ENOBUFS, without indicating the device that was
622 * Occasional 'false positives', in which the function returns 0 while
623 * indicating a device that was not actually added or deleted or returns
624 * ENOBUFS without any change, are acceptable.
626 * The purpose of 'port_poll' is to let 'ofproto' know about changes made
627 * externally to the 'ofproto' object, e.g. by a system administrator via
628 * ovs-dpctl. Therefore, it's OK, and even preferable, for port_poll() to
629 * not report changes made through calls to 'port_add' or 'port_del' on the
630 * same 'ofproto' object. (But it's OK for it to report them too, just
631 * slightly less efficient.)
633 * If the set of ports in 'ofproto' has not changed, returns EAGAIN. May
634 * also return other positive errno values to indicate that something has
637 * If the set of ports in a datapath is fixed, or if the only way that the
638 * set of ports in a datapath can change is through ->port_add() and
639 * ->port_del(), then this function may be a null pointer.
641 int (*port_poll)(const struct ofproto *ofproto, char **devnamep);
643 /* Arranges for the poll loop to wake up when ->port_poll() will return a
644 * value other than EAGAIN.
646 * If the set of ports in a datapath is fixed, or if the only way that the
647 * set of ports in a datapath can change is through ->port_add() and
648 * ->port_del(), or if the poll loop will always wake up anyway when
649 * ->port_poll() will return a value other than EAGAIN, then this function
650 * may be a null pointer.
652 void (*port_poll_wait)(const struct ofproto *ofproto);
654 /* Checks the status of LACP negotiation for 'port'. Returns 1 if LACP
655 * partner information for 'port' is up-to-date, 0 if LACP partner
656 * information is not current (generally indicating a connectivity
657 * problem), or -1 if LACP is not enabled on 'port'.
659 * This function may be a null pointer if the ofproto implementation does
660 * not support LACP. */
661 int (*port_is_lacp_current)(const struct ofport *port);
663 /* ## ----------------------- ## */
664 /* ## OpenFlow Rule Functions ## */
665 /* ## ----------------------- ## */
669 /* Chooses an appropriate table for 'cls_rule' within 'ofproto'. On
670 * success, stores the table ID into '*table_idp' and returns 0. On
671 * failure, returns an OpenFlow error code.
673 * The choice of table should be a function of 'cls_rule' and 'ofproto''s
674 * datapath capabilities. It should not depend on the flows already in
675 * 'ofproto''s flow tables. Failure implies that an OpenFlow rule with
676 * 'cls_rule' as its matching condition can never be inserted into
677 * 'ofproto', even starting from an empty flow table.
679 * If multiple tables are candidates for inserting the flow, the function
680 * should choose one arbitrarily (but deterministically).
682 * If this function is NULL then table 0 is always chosen. */
683 enum ofperr (*rule_choose_table)(const struct ofproto *ofproto,
684 const struct cls_rule *cls_rule,
687 /* Life-cycle functions for a "struct rule" (see "Life Cycle" above).
690 * Asynchronous Operation Support
691 * ==============================
693 * The life-cycle operations on rules can operate asynchronously, meaning
694 * that ->rule_construct() and ->rule_destruct() only need to initiate
695 * their respective operations and do not need to wait for them to complete
696 * before they return. ->rule_modify_actions() also operates
699 * An ofproto implementation reports the success or failure of an
700 * asynchronous operation on a rule using the rule's 'pending' member,
701 * which points to a opaque "struct ofoperation" that represents the
702 * ongoing opreation. When the operation completes, the ofproto
703 * implementation calls ofoperation_complete(), passing the ofoperation and
704 * an error indication.
706 * Only the following contexts may call ofoperation_complete():
708 * - The function called to initiate the operation,
709 * e.g. ->rule_construct() or ->rule_destruct(). This is the best
710 * choice if the operation completes quickly.
712 * - The implementation's ->run() function.
714 * - The implementation's ->destruct() function.
716 * The ofproto base code updates the flow table optimistically, assuming
717 * that the operation will probably succeed:
719 * - ofproto adds or replaces the rule in the flow table before calling
720 * ->rule_construct().
722 * - ofproto updates the rule's actions before calling
723 * ->rule_modify_actions().
725 * - ofproto removes the rule before calling ->rule_destruct().
727 * With one exception, when an asynchronous operation completes with an
728 * error, ofoperation_complete() backs out the already applied changes:
730 * - If adding or replacing a rule in the flow table fails, ofproto
731 * removes the new rule or restores the original rule.
733 * - If modifying a rule's actions fails, ofproto restores the original
736 * - Removing a rule is not allowed to fail. It must always succeed.
738 * The ofproto base code serializes operations: if any operation is in
739 * progress on a given rule, ofproto postpones initiating any new operation
740 * on that rule until the pending operation completes. Therefore, every
741 * operation must eventually complete through a call to
742 * ofoperation_complete() to avoid delaying new operations indefinitely
743 * (including any OpenFlow request that affects the rule in question, even
744 * just to query its statistics).
750 * When ->rule_construct() is called, the caller has already inserted
751 * 'rule' into 'rule->ofproto''s flow table numbered 'rule->table_id'.
752 * There are two cases:
754 * - 'rule' is a new rule in its flow table. In this case,
755 * ofoperation_get_victim(rule) returns NULL.
757 * - 'rule' is replacing an existing rule in its flow table that had the
758 * same matching criteria and priority. In this case,
759 * ofoperation_get_victim(rule) returns the rule being replaced (the
762 * ->rule_construct() should set the following in motion:
764 * - Validate that the matching rule in 'rule->cr' is supported by the
765 * datapath. For example, if the rule's table does not support
766 * registers, then it is an error if 'rule->cr' does not wildcard all
769 * - Validate that 'rule->actions' and 'rule->n_actions' are well-formed
770 * OpenFlow actions that the datapath can correctly implement. The
771 * validate_actions() function (in ofp-util.c) can be useful as a model
772 * for action validation, but it accepts all of the OpenFlow actions
773 * that OVS understands. If your ofproto implementation only
774 * implements a subset of those, then you should implement your own
777 * - If the rule is valid, update the datapath flow table, adding the new
778 * rule or replacing the existing one.
780 * - If 'rule' is replacing an existing rule, uninitialize any derived
781 * state for the victim rule, as in step 5 in the "Life Cycle"
784 * (On failure, the ofproto code will roll back the insertion from the flow
785 * table, either removing 'rule' or replacing it by the victim rule if
788 * ->rule_construct() must act in one of the following ways:
790 * - If it succeeds, it must call ofoperation_complete() and return 0.
792 * - If it fails, it must act in one of the following ways:
794 * * Call ofoperation_complete() and return 0.
796 * * Return an OpenFlow error code. (Do not call
797 * ofoperation_complete() in this case.)
799 * Either way, ->rule_destruct() will not be called for 'rule', but
800 * ->rule_dealloc() will be.
802 * - If the operation is only partially complete, then it must return 0.
803 * Later, when the operation is complete, the ->run() or ->destruct()
804 * function must call ofoperation_complete() to report success or
807 * ->rule_construct() should not modify any base members of struct rule.
813 * When ->rule_destruct() is called, the caller has already removed 'rule'
814 * from 'rule->ofproto''s flow table. ->rule_destruct() should set in
815 * motion removing 'rule' from the datapath flow table. If removal
816 * completes synchronously, it should call ofoperation_complete().
817 * Otherwise, the ->run() or ->destruct() function must later call
818 * ofoperation_complete() after the operation completes.
820 * Rule destruction must not fail. */
821 struct rule *(*rule_alloc)(void);
822 enum ofperr (*rule_construct)(struct rule *rule);
823 void (*rule_destruct)(struct rule *rule);
824 void (*rule_dealloc)(struct rule *rule);
826 /* Obtains statistics for 'rule', storing the number of packets that have
827 * matched it in '*packet_count' and the number of bytes in those packets
828 * in '*byte_count'. UINT64_MAX indicates that the packet count or byte
829 * count is unknown. */
830 void (*rule_get_stats)(struct rule *rule, uint64_t *packet_count,
831 uint64_t *byte_count);
833 /* Applies the actions in 'rule' to 'packet'. (This implements sending
834 * buffered packets for OpenFlow OFPT_FLOW_MOD commands.)
836 * Takes ownership of 'packet' (so it should eventually free it, with
839 * 'flow' reflects the flow information for 'packet'. All of the
840 * information in 'flow' is extracted from 'packet', except for
841 * flow->tun_id and flow->in_port, which are assigned the correct values
842 * for the incoming packet. The register values are zeroed. 'packet''s
843 * header pointers (e.g. packet->l3) are appropriately initialized.
845 * The implementation should add the statistics for 'packet' into 'rule'.
847 * Returns 0 if successful, otherwise an OpenFlow error code. */
848 enum ofperr (*rule_execute)(struct rule *rule, const struct flow *flow,
849 struct ofpbuf *packet);
851 /* When ->rule_modify_actions() is called, the caller has already replaced
852 * the OpenFlow actions in 'rule' by a new set. (The original actions are
853 * in rule->pending->actions.)
855 * ->rule_modify_actions() should set the following in motion:
857 * - Validate that the actions now in 'rule' are well-formed OpenFlow
858 * actions that the datapath can correctly implement.
860 * - Update the datapath flow table with the new actions.
862 * If the operation synchronously completes, ->rule_modify_actions() may
863 * call ofoperation_complete() before it returns. Otherwise, ->run()
864 * should call ofoperation_complete() later, after the operation does
867 * If the operation fails, then the base ofproto code will restore the
868 * original 'actions' and 'n_actions' of 'rule'.
870 * ->rule_modify_actions() should not modify any base members of struct
872 void (*rule_modify_actions)(struct rule *rule);
874 /* Changes the OpenFlow IP fragment handling policy to 'frag_handling',
875 * which takes one of the following values, with the corresponding
878 * - OFPC_FRAG_NORMAL: The switch should treat IP fragments the same way
879 * as other packets, omitting TCP and UDP port numbers (always setting
882 * - OFPC_FRAG_DROP: The switch should drop all IP fragments without
883 * passing them through the flow table.
885 * - OFPC_FRAG_REASM: The switch should reassemble IP fragments before
886 * passing packets through the flow table.
888 * - OFPC_FRAG_NX_MATCH (a Nicira extension): Similar to OFPC_FRAG_NORMAL,
889 * except that TCP and UDP port numbers should be included in fragments
892 * Implementations are not required to support every mode.
893 * OFPC_FRAG_NORMAL is the default mode when an ofproto is created.
895 * At the time of the call to ->set_frag_handling(), the current mode is
896 * available in 'ofproto->frag_handling'. ->set_frag_handling() returns
897 * true if the requested mode was set, false if it is not supported.
899 * Upon successful return, the caller changes 'ofproto->frag_handling' to
900 * reflect the new mode.
902 bool (*set_frag_handling)(struct ofproto *ofproto,
903 enum ofp_config_flags frag_handling);
905 /* Implements the OpenFlow OFPT_PACKET_OUT command. The datapath should
906 * execute the 'n_actions' in the 'actions' array on 'packet'.
908 * The caller retains ownership of 'packet', so ->packet_out() should not
911 * This function must validate that the 'n_actions' elements in 'actions'
912 * are well-formed OpenFlow actions that can be correctly implemented by
913 * the datapath. If not, then it should return an OpenFlow error code.
915 * 'flow' reflects the flow information for 'packet'. All of the
916 * information in 'flow' is extracted from 'packet', except for
917 * flow->in_port, which is taken from the OFPT_PACKET_OUT message.
918 * flow->tun_id and its register values are zeroed.
920 * 'packet' is not matched against the OpenFlow flow table, so its
921 * statistics should not be included in OpenFlow flow statistics.
923 * Returns 0 if successful, otherwise an OpenFlow error code. */
924 enum ofperr (*packet_out)(struct ofproto *ofproto, struct ofpbuf *packet,
925 const struct flow *flow,
926 const union ofp_action *actions,
929 /* ## ------------------------- ## */
930 /* ## OFPP_NORMAL configuration ## */
931 /* ## ------------------------- ## */
933 /* Configures NetFlow on 'ofproto' according to the options in
934 * 'netflow_options', or turns off NetFlow if 'netflow_options' is NULL.
936 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
937 * NetFlow, as does a null pointer. */
938 int (*set_netflow)(struct ofproto *ofproto,
939 const struct netflow_options *netflow_options);
941 void (*get_netflow_ids)(const struct ofproto *ofproto,
942 uint8_t *engine_type, uint8_t *engine_id);
944 /* Configures sFlow on 'ofproto' according to the options in
945 * 'sflow_options', or turns off sFlow if 'sflow_options' is NULL.
947 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
948 * sFlow, as does a null pointer. */
949 int (*set_sflow)(struct ofproto *ofproto,
950 const struct ofproto_sflow_options *sflow_options);
952 /* Configures connectivity fault management on 'ofport'.
954 * If 'cfm_settings' is nonnull, configures CFM according to its members.
956 * If 'cfm_settings' is null, removes any connectivity fault management
957 * configuration from 'ofport'.
959 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
960 * support CFM, as does a null pointer. */
961 int (*set_cfm)(struct ofport *ofport, const struct cfm_settings *s);
963 /* Checks the fault status of CFM configured on 'ofport'. Returns a
964 * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
965 * indicating a connectivity problem). Returns zero if CFM is not faulted,
966 * and -1 if CFM is not enabled on 'port'.
968 * This function may be a null pointer if the ofproto implementation does
969 * not support CFM. */
970 int (*get_cfm_fault)(const struct ofport *ofport);
972 /* Gets the MPIDs of the remote maintenance points broadcasting to
973 * 'ofport'. Populates 'rmps' with a provider owned array of MPIDs, and
974 * 'n_rmps' with the number of MPIDs in 'rmps'. Returns a number less than
975 * 0 if CFM is not enabled of 'ofport'.
977 * This function may be a null pointer if the ofproto implementation does
978 * not support CFM. */
979 int (*get_cfm_remote_mpids)(const struct ofport *ofport,
980 const uint64_t **rmps, size_t *n_rmps);
982 /* Configures spanning tree protocol (STP) on 'ofproto' using the
983 * settings defined in 's'.
985 * If 's' is nonnull, configures STP according to its members.
987 * If 's' is null, removes any STP configuration from 'ofproto'.
989 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
990 * support STP, as does a null pointer. */
991 int (*set_stp)(struct ofproto *ofproto,
992 const struct ofproto_stp_settings *s);
994 /* Retrieves state of spanning tree protocol (STP) on 'ofproto'.
996 * Stores STP state for 'ofproto' in 's'. If the 'enabled' member
997 * is false, the other member values are not meaningful.
999 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1000 * support STP, as does a null pointer. */
1001 int (*get_stp_status)(struct ofproto *ofproto,
1002 struct ofproto_stp_status *s);
1004 /* Configures spanning tree protocol (STP) on 'ofport' using the
1005 * settings defined in 's'.
1007 * If 's' is nonnull, configures STP according to its members. The
1008 * caller is responsible for assigning STP port numbers (using the
1009 * 'port_num' member in the range of 1 through 255, inclusive) and
1010 * ensuring there are no duplicates.
1012 * If 's' is null, removes any STP configuration from 'ofport'.
1014 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1015 * support STP, as does a null pointer. */
1016 int (*set_stp_port)(struct ofport *ofport,
1017 const struct ofproto_port_stp_settings *s);
1019 /* Retrieves spanning tree protocol (STP) port status of 'ofport'.
1021 * Stores STP state for 'ofport' in 's'. If the 'enabled' member is
1022 * false, the other member values are not meaningful.
1024 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1025 * support STP, as does a null pointer. */
1026 int (*get_stp_port_status)(struct ofport *ofport,
1027 struct ofproto_port_stp_status *s);
1029 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
1030 * 'queues' attached to 'ofport'. This data is not intended to be
1031 * sufficient to implement QoS. Instead, providers may use this
1032 * information to implement features which require knowledge of what queues
1033 * exist on a port, and some basic information about them.
1035 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1036 * support QoS, as does a null pointer. */
1037 int (*set_queues)(struct ofport *ofport,
1038 const struct ofproto_port_queue *queues, size_t n_qdscp);
1040 /* If 's' is nonnull, this function registers a "bundle" associated with
1041 * client data pointer 'aux' in 'ofproto'. A bundle is the same concept as
1042 * a Port in OVSDB, that is, it consists of one or more "slave" devices
1043 * (Interfaces, in OVSDB) along with VLAN and LACP configuration and, if
1044 * there is more than one slave, a bonding configuration. If 'aux' is
1045 * already registered then this function updates its configuration to 's'.
1046 * Otherwise, this function registers a new bundle.
1048 * If 's' is NULL, this function unregisters the bundle registered on
1049 * 'ofproto' associated with client data pointer 'aux'. If no such bundle
1050 * has been registered, this has no effect.
1052 * This function affects only the behavior of the NXAST_AUTOPATH action and
1053 * output to the OFPP_NORMAL port. An implementation that does not support
1054 * it at all may set it to NULL or return EOPNOTSUPP. An implementation
1055 * that supports only a subset of the functionality should implement what
1056 * it can and return 0. */
1057 int (*bundle_set)(struct ofproto *ofproto, void *aux,
1058 const struct ofproto_bundle_settings *s);
1060 /* If 'port' is part of any bundle, removes it from that bundle. If the
1061 * bundle now has no ports, deletes the bundle. If the bundle now has only
1062 * one port, deconfigures the bundle's bonding configuration. */
1063 void (*bundle_remove)(struct ofport *ofport);
1065 /* If 's' is nonnull, this function registers a mirror associated with
1066 * client data pointer 'aux' in 'ofproto'. A mirror is the same concept as
1067 * a Mirror in OVSDB. If 'aux' is already registered then this function
1068 * updates its configuration to 's'. Otherwise, this function registers a
1071 * If 's' is NULL, this function unregisters the mirror registered on
1072 * 'ofproto' associated with client data pointer 'aux'. If no such mirror
1073 * has been registered, this has no effect.
1075 * An implementation that does not support mirroring at all may set
1076 * it to NULL or return EOPNOTSUPP. An implementation that supports
1077 * only a subset of the functionality should implement what it can
1079 int (*mirror_set)(struct ofproto *ofproto, void *aux,
1080 const struct ofproto_mirror_settings *s);
1082 /* Retrieves statistics from mirror associated with client data
1083 * pointer 'aux' in 'ofproto'. Stores packet and byte counts in
1084 * 'packets' and 'bytes', respectively. If a particular counter is
1085 * not supported, the appropriate argument is set to UINT64_MAX.
1087 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1088 * support retrieving mirror statistics. */
1089 int (*mirror_get_stats)(struct ofproto *ofproto, void *aux,
1090 uint64_t *packets, uint64_t *bytes);
1092 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs
1093 * on which all packets are flooded, instead of using MAC learning. If
1094 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1096 * This function affects only the behavior of the OFPP_NORMAL action. An
1097 * implementation that does not support it may set it to NULL or return
1099 int (*set_flood_vlans)(struct ofproto *ofproto,
1100 unsigned long *flood_vlans);
1102 /* Returns true if 'aux' is a registered bundle that is currently in use as
1103 * the output for a mirror. */
1104 bool (*is_mirror_output_bundle)(const struct ofproto *ofproto, void *aux);
1106 /* When the configuration option of forward_bpdu changes, this function
1107 * will be invoked. */
1108 void (*forward_bpdu_changed)(struct ofproto *ofproto);
1110 /* Sets the MAC aging timeout for the OFPP_NORMAL action to 'idle_time',
1112 void (*set_mac_idle_time)(struct ofproto *ofproto, unsigned int idle_time);
1114 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
1116 * This is deprecated. It is only for compatibility with broken device drivers
1117 * in old versions of Linux that do not properly support VLANs when VLAN
1118 * devices are not used. When broken device drivers are no longer in
1119 * widespread use, we will delete these interfaces. */
1121 /* If 'realdev_ofp_port' is nonzero, then this function configures 'ofport'
1122 * as a VLAN splinter port for VLAN 'vid', associated with the real device
1123 * that has OpenFlow port number 'realdev_ofp_port'.
1125 * If 'realdev_ofp_port' is zero, then this function deconfigures 'ofport'
1126 * as a VLAN splinter port.
1128 * This function should be NULL if a an implementation does not support
1130 int (*set_realdev)(struct ofport *ofport,
1131 uint16_t realdev_ofp_port, int vid);
1134 extern const struct ofproto_class ofproto_dpif_class;
1136 int ofproto_class_register(const struct ofproto_class *);
1137 int ofproto_class_unregister(const struct ofproto_class *);
1139 /* ofproto_flow_mod() returns this value if the flow_mod could not be processed
1140 * because it overlaps with an ongoing flow table operation that has not yet
1141 * completed. The caller should retry the operation later.
1143 * ofproto.c also uses this value internally for additional (similar) purposes.
1145 * This particular value is a good choice because it is large, so that it does
1146 * not collide with any errno value, but not large enough to collide with an
1147 * OFPERR_* value. */
1148 enum { OFPROTO_POSTPONE = 1 << 16 };
1149 BUILD_ASSERT_DECL(OFPROTO_POSTPONE < OFPERR_OFS);
1151 int ofproto_flow_mod(struct ofproto *, const struct ofputil_flow_mod *);
1152 void ofproto_add_flow(struct ofproto *, const struct cls_rule *,
1153 const union ofp_action *, size_t n_actions);
1154 bool ofproto_delete_flow(struct ofproto *, const struct cls_rule *);
1155 void ofproto_flush_flows(struct ofproto *);
1157 #endif /* ofproto/ofproto-provider.h */