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 struct hmap_node hmap_node; /* In global 'all_ofprotos' hmap. */
40 const struct ofproto_class *ofproto_class;
41 char *type; /* Datapath type. */
42 char *name; /* Datapath name. */
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. */
84 int min_mtu; /* Current MTU of non-internal ports. */
87 void ofproto_init_tables(struct ofproto *, int n_tables);
89 struct ofproto *ofproto_lookup(const char *name);
90 struct ofport *ofproto_get_port(const struct ofproto *, uint16_t ofp_port);
92 /* An OpenFlow port within a "struct ofproto".
94 * With few exceptions, ofproto implementations may look at these fields but
95 * should not modify them. */
97 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
98 struct ofproto *ofproto; /* The ofproto that contains this port. */
99 struct netdev *netdev;
100 struct ofputil_phy_port pp;
101 uint16_t ofp_port; /* OpenFlow port number. */
102 unsigned int change_seq;
106 void ofproto_port_set_state(struct ofport *, enum ofputil_port_state);
109 OFTABLE_HIDDEN = 1 << 0, /* Hide from most OpenFlow operations. */
110 OFTABLE_READONLY = 1 << 1 /* Don't allow OpenFlow to change this table. */
113 /* A flow table within a "struct ofproto". */
115 enum oftable_flags flags;
116 struct classifier cls; /* Contains "struct rule"s. */
117 char *name; /* Table name exposed via OpenFlow, or NULL. */
119 /* Maximum number of flows or UINT_MAX if there is no limit besides any
120 * limit imposed by resource limitations. */
121 unsigned int max_flows;
123 /* These members determine the handling of an attempt to add a flow that
124 * would cause the table to have more than 'max_flows' flows.
126 * If 'eviction_fields' is NULL, overflows will be rejected with an error.
128 * If 'eviction_fields' is nonnull (regardless of whether n_eviction_fields
129 * is nonzero), an overflow will cause a flow to be removed. The flow to
130 * be removed is chosen to give fairness among groups distinguished by
131 * different values for the subfields within 'groups'. */
132 struct mf_subfield *eviction_fields;
133 size_t n_eviction_fields;
137 * When a flow is added that would cause the table to have more than
138 * 'max_flows' flows, and 'eviction_fields' is nonnull, these groups are
139 * used to decide which rule to evict: the rule is chosen from the eviction
140 * group that contains the greatest number of rules.*/
141 uint32_t eviction_group_id_basis;
142 struct hmap eviction_groups_by_id;
143 struct heap eviction_groups_by_size;
146 /* Assigns TABLE to each oftable, in turn, in OFPROTO.
148 * All parameters are evaluated multiple times. */
149 #define OFPROTO_FOR_EACH_TABLE(TABLE, OFPROTO) \
150 for ((TABLE) = (OFPROTO)->tables; \
151 (TABLE) < &(OFPROTO)->tables[(OFPROTO)->n_tables]; \
154 /* An OpenFlow flow within a "struct ofproto".
156 * With few exceptions, ofproto implementations may look at these fields but
157 * should not modify them. */
159 struct list ofproto_node; /* Owned by ofproto base code. */
160 struct ofproto *ofproto; /* The ofproto that contains this rule. */
161 struct cls_rule cr; /* In owning ofproto's classifier. */
163 struct ofoperation *pending; /* Operation now in progress, if nonnull. */
165 ovs_be64 flow_cookie; /* Controller-issued identifier. */
167 long long int created; /* Creation time. */
168 long long int modified; /* Time of last modification. */
169 long long int used; /* Last use; time created if never used. */
170 uint16_t hard_timeout; /* In seconds from ->modified. */
171 uint16_t idle_timeout; /* In seconds from ->used. */
172 uint8_t table_id; /* Index in ofproto's 'tables' array. */
173 bool send_flow_removed; /* Send a flow removed message? */
175 /* Eviction groups. */
176 bool evictable; /* If false, prevents eviction. */
177 struct heap_node evg_node; /* In eviction_group's "rules" heap. */
178 struct eviction_group *eviction_group; /* NULL if not in any group. */
180 union ofp_action *actions; /* OpenFlow actions. */
181 int n_actions; /* Number of elements in actions[]. */
184 static inline struct rule *
185 rule_from_cls_rule(const struct cls_rule *cls_rule)
187 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
190 void ofproto_rule_update_used(struct rule *, long long int used);
191 void ofproto_rule_expire(struct rule *, uint8_t reason);
192 void ofproto_rule_destroy(struct rule *);
194 void ofoperation_complete(struct ofoperation *, enum ofperr);
195 struct rule *ofoperation_get_victim(struct ofoperation *);
197 /* ofproto class structure, to be defined by each ofproto implementation.
203 * These functions work primarily with three different kinds of data
206 * - "struct ofproto", which represents an OpenFlow switch.
208 * - "struct ofport", which represents a port within an ofproto.
210 * - "struct rule", which represents an OpenFlow flow within an ofproto.
212 * Each of these data structures contains all of the implementation-independent
213 * generic state for the respective concept, called the "base" state. None of
214 * them contains any extra space for ofproto implementations to use. Instead,
215 * each implementation is expected to declare its own data structure that
216 * contains an instance of the generic data structure plus additional
217 * implementation-specific members, called the "derived" state. The
218 * implementation can use casts or (preferably) the CONTAINER_OF macro to
219 * obtain access to derived state given only a pointer to the embedded generic
226 * Four stylized functions accompany each of these data structures:
228 * "alloc" "construct" "destruct" "dealloc"
229 * ------------ ---------------- --------------- --------------
230 * ofproto ->alloc ->construct ->destruct ->dealloc
231 * ofport ->port_alloc ->port_construct ->port_destruct ->port_dealloc
232 * rule ->rule_alloc ->rule_construct ->rule_destruct ->rule_dealloc
234 * Any instance of a given data structure goes through the following life
237 * 1. The client calls the "alloc" function to obtain raw memory. If "alloc"
238 * fails, skip all the other steps.
240 * 2. The client initializes all of the data structure's base state. If this
241 * fails, skip to step 7.
243 * 3. The client calls the "construct" function. The implementation
244 * initializes derived state. It may refer to the already-initialized
245 * base state. If "construct" fails, skip to step 6.
247 * 4. The data structure is now initialized and in use.
249 * 5. When the data structure is no longer needed, the client calls the
250 * "destruct" function. The implementation uninitializes derived state.
251 * The base state has not been uninitialized yet, so the implementation
252 * may still refer to it.
254 * 6. The client uninitializes all of the data structure's base state.
256 * 7. The client calls the "dealloc" to free the raw memory. The
257 * implementation must not refer to base or derived state in the data
258 * structure, because it has already been uninitialized.
260 * Each "alloc" function allocates and returns a new instance of the respective
261 * data structure. The "alloc" function is not given any information about the
262 * use of the new data structure, so it cannot perform much initialization.
263 * Its purpose is just to ensure that the new data structure has enough room
264 * for base and derived state. It may return a null pointer if memory is not
265 * available, in which case none of the other functions is called.
267 * Each "construct" function initializes derived state in its respective data
268 * structure. When "construct" is called, all of the base state has already
269 * been initialized, so the "construct" function may refer to it. The
270 * "construct" function is allowed to fail, in which case the client calls the
271 * "dealloc" function (but not the "destruct" function).
273 * Each "destruct" function uninitializes and frees derived state in its
274 * respective data structure. When "destruct" is called, the base state has
275 * not yet been uninitialized, so the "destruct" function may refer to it. The
276 * "destruct" function is not allowed to fail.
278 * Each "dealloc" function frees raw memory that was allocated by the the
279 * "alloc" function. The memory's base and derived members might not have ever
280 * been initialized (but if "construct" returned successfully, then it has been
281 * "destruct"ed already). The "dealloc" function is not allowed to fail.
287 * Most of these functions return 0 if they are successful or a positive error
288 * code on failure. Depending on the function, valid error codes are either
289 * errno values or OFPERR_* OpenFlow error codes.
291 * Most of these functions are expected to execute synchronously, that is, to
292 * block as necessary to obtain a result. Thus, these functions may return
293 * EAGAIN (or EWOULDBLOCK or EINPROGRESS) only where the function descriptions
294 * explicitly say those errors are a possibility. We may relax this
295 * requirement in the future if and when we encounter performance problems. */
296 struct ofproto_class {
297 /* ## ----------------- ## */
298 /* ## Factory Functions ## */
299 /* ## ----------------- ## */
301 /* Enumerates the types of all support ofproto types into 'types'. The
302 * caller has already initialized 'types' and other ofproto classes might
303 * already have added names to it. */
304 void (*enumerate_types)(struct sset *types);
306 /* Enumerates the names of all existing datapath of the specified 'type'
307 * into 'names' 'all_dps'. The caller has already initialized 'names' as
310 * 'type' is one of the types enumerated by ->enumerate_types().
312 * Returns 0 if successful, otherwise a positive errno value.
314 int (*enumerate_names)(const char *type, struct sset *names);
316 /* Deletes the datapath with the specified 'type' and 'name'. The caller
317 * should have closed any open ofproto with this 'type' and 'name'; this
318 * function is allowed to fail if that is not the case.
320 * 'type' is one of the types enumerated by ->enumerate_types().
321 * 'name' is one of the names enumerated by ->enumerate_names() for 'type'.
323 * Returns 0 if successful, otherwise a positive errno value.
325 int (*del)(const char *type, const char *name);
327 /* ## --------------------------- ## */
328 /* ## Top-Level ofproto Functions ## */
329 /* ## --------------------------- ## */
331 /* Life-cycle functions for an "ofproto" (see "Life Cycle" above).
337 * ->construct() should not modify any base members of the ofproto. The
338 * client will initialize the ofproto's 'ports' and 'tables' members after
339 * construction is complete.
341 * When ->construct() is called, the client does not yet know how many flow
342 * tables the datapath supports, so ofproto->n_tables will be 0 and
343 * ofproto->tables will be NULL. ->construct() should call
344 * ofproto_init_tables() to allocate and initialize ofproto->n_tables and
345 * ofproto->tables. Each flow table will be initially empty, so
346 * ->construct() should delete flows from the underlying datapath, if
347 * necessary, rather than populating the tables.
349 * Only one ofproto instance needs to be supported for any given datapath.
350 * If a datapath is already open as part of one "ofproto", then another
351 * attempt to "construct" the same datapath as part of another ofproto is
352 * allowed to fail with an error.
354 * ->construct() returns 0 if successful, otherwise a positive errno
361 * If 'ofproto' has any pending asynchronous operations, ->destruct()
362 * must complete all of them by calling ofoperation_complete().
364 * ->destruct() must also destroy all remaining rules in the ofproto's
365 * tables, by passing each remaining rule to ofproto_rule_destroy(). The
366 * client will destroy the flow tables themselves after ->destruct()
369 struct ofproto *(*alloc)(void);
370 int (*construct)(struct ofproto *ofproto);
371 void (*destruct)(struct ofproto *ofproto);
372 void (*dealloc)(struct ofproto *ofproto);
374 /* Performs any periodic activity required by 'ofproto'. It should:
376 * - Call connmgr_send_packet_in() for each received packet that missed
377 * in the OpenFlow flow table or that had a OFPP_CONTROLLER output
380 * - Call ofproto_rule_expire() for each OpenFlow flow that has reached
381 * its hard_timeout or idle_timeout, to expire the flow.
383 * Returns 0 if successful, otherwise a positive errno value. */
384 int (*run)(struct ofproto *ofproto);
386 /* Performs periodic activity required by 'ofproto' that needs to be done
387 * with the least possible latency.
389 * This is run multiple times per main loop. An ofproto provider may
390 * implement it or not, according to whether it provides a performance
391 * boost for that ofproto implementation. */
392 int (*run_fast)(struct ofproto *ofproto);
394 /* Causes the poll loop to wake up when 'ofproto''s 'run' function needs to
395 * be called, e.g. by calling the timer or fd waiting functions in
397 void (*wait)(struct ofproto *ofproto);
399 /* Every "struct rule" in 'ofproto' is about to be deleted, one by one.
400 * This function may prepare for that, for example by clearing state in
401 * advance. It should *not* actually delete any "struct rule"s from
402 * 'ofproto', only prepare for it.
404 * This function is optional; it's really just for optimization in case
405 * it's cheaper to delete all the flows from your hardware in a single pass
406 * than to do it one by one. */
407 void (*flush)(struct ofproto *ofproto);
409 /* Helper for the OpenFlow OFPT_FEATURES_REQUEST request.
411 * The implementation should store true in '*arp_match_ip' if the switch
412 * supports matching IP addresses inside ARP requests and replies, false
415 * The implementation should store in '*actions' a bitmap of the supported
416 * OpenFlow actions. Vendor actions are not included in '*actions'. */
417 void (*get_features)(struct ofproto *ofproto,
419 enum ofputil_action_bitmap *actions);
421 /* Helper for the OpenFlow OFPST_TABLE statistics request.
423 * The 'ots' array contains 'ofproto->n_tables' elements. Each element is
426 * - 'table_id' to the array index.
428 * - 'name' to "table#" where # is the table ID.
430 * - 'wildcards' to OFPFW_ALL.
432 * - 'max_entries' to 1,000,000.
434 * - 'active_count' to the classifier_count() for the table.
436 * - 'lookup_count' and 'matched_count' to 0.
438 * The implementation should update any members in each element for which
439 * it has better values:
441 * - 'name' to a more meaningful name.
443 * - 'wildcards' to the set of wildcards actually supported by the table
444 * (if it doesn't support all OpenFlow wildcards).
446 * - 'max_entries' to the maximum number of flows actually supported by
449 * - 'lookup_count' to the number of packets looked up in this flow table
452 * - 'matched_count' to the number of packets looked up in this flow
453 * table so far that matched one of the flow entries.
455 * Keep in mind that all of the members of struct ofp_table_stats are in
456 * network byte order.
458 void (*get_tables)(struct ofproto *ofproto, struct ofp_table_stats *ots);
460 /* ## ---------------- ## */
461 /* ## ofport Functions ## */
462 /* ## ---------------- ## */
464 /* Life-cycle functions for a "struct ofport" (see "Life Cycle" above).
466 * ->port_construct() should not modify any base members of the ofport.
468 * ofports are managed by the base ofproto code. The ofproto
469 * implementation should only create and destroy them in response to calls
470 * to these functions. The base ofproto code will create and destroy
471 * ofports in the following situations:
473 * - Just after the ->construct() function is called, the base ofproto
474 * iterates over all of the implementation's ports, using
475 * ->port_dump_start() and related functions, and constructs an ofport
476 * for each dumped port.
478 * - If ->port_poll() reports that a specific port has changed, then the
479 * base ofproto will query that port with ->port_query_by_name() and
480 * construct or destruct ofports as necessary to reflect the updated
483 * - If ->port_poll() returns ENOBUFS to report an unspecified port set
484 * change, then the base ofproto will iterate over all of the
485 * implementation's ports, in the same way as at ofproto
486 * initialization, and construct and destruct ofports to reflect all of
489 * ->port_construct() returns 0 if successful, otherwise a positive errno
492 struct ofport *(*port_alloc)(void);
493 int (*port_construct)(struct ofport *ofport);
494 void (*port_destruct)(struct ofport *ofport);
495 void (*port_dealloc)(struct ofport *ofport);
497 /* Called after 'ofport->netdev' is replaced by a new netdev object. If
498 * the ofproto implementation uses the ofport's netdev internally, then it
499 * should switch to using the new one. The old one has been closed.
501 * An ofproto implementation that doesn't need to do anything in this
502 * function may use a null pointer. */
503 void (*port_modified)(struct ofport *ofport);
505 /* Called after an OpenFlow request changes a port's configuration.
506 * 'ofport->pp.config' contains the new configuration. 'old_config'
507 * contains the previous configuration.
509 * The caller implements OFPUTIL_PC_PORT_DOWN using netdev functions to
510 * turn NETDEV_UP on and off, so this function doesn't have to do anything
511 * for that bit (and it won't be called if that is the only bit that
513 void (*port_reconfigured)(struct ofport *ofport,
514 enum ofputil_port_config old_config);
516 /* Looks up a port named 'devname' in 'ofproto'. On success, initializes
517 * '*port' appropriately.
519 * The caller owns the data in 'port' and must free it with
520 * ofproto_port_destroy() when it is no longer needed. */
521 int (*port_query_by_name)(const struct ofproto *ofproto,
522 const char *devname, struct ofproto_port *port);
524 /* Attempts to add 'netdev' as a port on 'ofproto'. Returns 0 if
525 * successful, otherwise a positive errno value. If successful, sets
526 * '*ofp_portp' to the new port's port number.
528 * It doesn't matter whether the new port will be returned by a later call
529 * to ->port_poll(); the implementation may do whatever is more
531 int (*port_add)(struct ofproto *ofproto, struct netdev *netdev,
532 uint16_t *ofp_portp);
534 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'. Returns
535 * 0 if successful, otherwise a positive errno value.
537 * It doesn't matter whether the new port will be returned by a later call
538 * to ->port_poll(); the implementation may do whatever is more
540 int (*port_del)(struct ofproto *ofproto, uint16_t ofp_port);
543 int (*port_get_stats)(const struct ofport *port,
544 struct netdev_stats *stats);
546 /* Port iteration functions.
548 * The client might not be entirely in control of the ports within an
549 * ofproto. Some hardware implementations, for example, might have a fixed
550 * set of ports in a datapath, and the Linux datapath allows the system
551 * administrator to externally add and remove ports with ovs-dpctl. For
552 * this reason, the client needs a way to iterate through all the ports
553 * that are actually in a datapath. These functions provide that
556 * The 'state' pointer provides the implementation a place to
557 * keep track of its position. Its format is opaque to the caller.
559 * The ofproto provider retains ownership of the data that it stores into
560 * ->port_dump_next()'s 'port' argument. The data must remain valid until
561 * at least the next call to ->port_dump_next() or ->port_dump_done() for
562 * 'state'. The caller will not modify or free it.
567 * ->port_dump_start() attempts to begin dumping the ports in 'ofproto'.
568 * On success, it should return 0 and initialize '*statep' with any data
569 * needed for iteration. On failure, returns a positive errno value, and
570 * the client will not call ->port_dump_next() or ->port_dump_done().
572 * ->port_dump_next() attempts to retrieve another port from 'ofproto' for
573 * 'state'. If there is another port, it should store the port's
574 * information into 'port' and return 0. It should return EOF if all ports
575 * have already been iterated. Otherwise, on error, it should return a
576 * positive errno value. This function will not be called again once it
577 * returns nonzero once for a given iteration (but the 'port_dump_done'
578 * function will be called afterward).
580 * ->port_dump_done() allows the implementation to release resources used
581 * for iteration. The caller might decide to stop iteration in the middle
582 * by calling this function before ->port_dump_next() returns nonzero.
590 * error = ofproto->ofproto_class->port_dump_start(ofproto, &state);
593 * struct ofproto_port port;
595 * error = ofproto->ofproto_class->port_dump_next(
596 * ofproto, state, &port);
600 * // Do something with 'port' here (without modifying or freeing
601 * // any of its data).
603 * ofproto->ofproto_class->port_dump_done(ofproto, state);
605 * // 'error' is now EOF (success) or a positive errno value (failure).
607 int (*port_dump_start)(const struct ofproto *ofproto, void **statep);
608 int (*port_dump_next)(const struct ofproto *ofproto, void *state,
609 struct ofproto_port *port);
610 int (*port_dump_done)(const struct ofproto *ofproto, void *state);
612 /* Polls for changes in the set of ports in 'ofproto'. If the set of ports
613 * in 'ofproto' has changed, then this function should do one of the
616 * - Preferably: store the name of the device that was added to or deleted
617 * from 'ofproto' in '*devnamep' and return 0. The caller is responsible
618 * for freeing '*devnamep' (with free()) when it no longer needs it.
620 * - Alternatively: return ENOBUFS, without indicating the device that was
623 * Occasional 'false positives', in which the function returns 0 while
624 * indicating a device that was not actually added or deleted or returns
625 * ENOBUFS without any change, are acceptable.
627 * The purpose of 'port_poll' is to let 'ofproto' know about changes made
628 * externally to the 'ofproto' object, e.g. by a system administrator via
629 * ovs-dpctl. Therefore, it's OK, and even preferable, for port_poll() to
630 * not report changes made through calls to 'port_add' or 'port_del' on the
631 * same 'ofproto' object. (But it's OK for it to report them too, just
632 * slightly less efficient.)
634 * If the set of ports in 'ofproto' has not changed, returns EAGAIN. May
635 * also return other positive errno values to indicate that something has
638 * If the set of ports in a datapath is fixed, or if the only way that the
639 * set of ports in a datapath can change is through ->port_add() and
640 * ->port_del(), then this function may be a null pointer.
642 int (*port_poll)(const struct ofproto *ofproto, char **devnamep);
644 /* Arranges for the poll loop to wake up when ->port_poll() will return a
645 * value other than EAGAIN.
647 * If the set of ports in a datapath is fixed, or if the only way that the
648 * set of ports in a datapath can change is through ->port_add() and
649 * ->port_del(), or if the poll loop will always wake up anyway when
650 * ->port_poll() will return a value other than EAGAIN, then this function
651 * may be a null pointer.
653 void (*port_poll_wait)(const struct ofproto *ofproto);
655 /* Checks the status of LACP negotiation for 'port'. Returns 1 if LACP
656 * partner information for 'port' is up-to-date, 0 if LACP partner
657 * information is not current (generally indicating a connectivity
658 * problem), or -1 if LACP is not enabled on 'port'.
660 * This function may be a null pointer if the ofproto implementation does
661 * not support LACP. */
662 int (*port_is_lacp_current)(const struct ofport *port);
664 /* ## ----------------------- ## */
665 /* ## OpenFlow Rule Functions ## */
666 /* ## ----------------------- ## */
670 /* Chooses an appropriate table for 'cls_rule' within 'ofproto'. On
671 * success, stores the table ID into '*table_idp' and returns 0. On
672 * failure, returns an OpenFlow error code.
674 * The choice of table should be a function of 'cls_rule' and 'ofproto''s
675 * datapath capabilities. It should not depend on the flows already in
676 * 'ofproto''s flow tables. Failure implies that an OpenFlow rule with
677 * 'cls_rule' as its matching condition can never be inserted into
678 * 'ofproto', even starting from an empty flow table.
680 * If multiple tables are candidates for inserting the flow, the function
681 * should choose one arbitrarily (but deterministically).
683 * If this function is NULL then table 0 is always chosen. */
684 enum ofperr (*rule_choose_table)(const struct ofproto *ofproto,
685 const struct cls_rule *cls_rule,
688 /* Life-cycle functions for a "struct rule" (see "Life Cycle" above).
691 * Asynchronous Operation Support
692 * ==============================
694 * The life-cycle operations on rules can operate asynchronously, meaning
695 * that ->rule_construct() and ->rule_destruct() only need to initiate
696 * their respective operations and do not need to wait for them to complete
697 * before they return. ->rule_modify_actions() also operates
700 * An ofproto implementation reports the success or failure of an
701 * asynchronous operation on a rule using the rule's 'pending' member,
702 * which points to a opaque "struct ofoperation" that represents the
703 * ongoing opreation. When the operation completes, the ofproto
704 * implementation calls ofoperation_complete(), passing the ofoperation and
705 * an error indication.
707 * Only the following contexts may call ofoperation_complete():
709 * - The function called to initiate the operation,
710 * e.g. ->rule_construct() or ->rule_destruct(). This is the best
711 * choice if the operation completes quickly.
713 * - The implementation's ->run() function.
715 * - The implementation's ->destruct() function.
717 * The ofproto base code updates the flow table optimistically, assuming
718 * that the operation will probably succeed:
720 * - ofproto adds or replaces the rule in the flow table before calling
721 * ->rule_construct().
723 * - ofproto updates the rule's actions before calling
724 * ->rule_modify_actions().
726 * - ofproto removes the rule before calling ->rule_destruct().
728 * With one exception, when an asynchronous operation completes with an
729 * error, ofoperation_complete() backs out the already applied changes:
731 * - If adding or replacing a rule in the flow table fails, ofproto
732 * removes the new rule or restores the original rule.
734 * - If modifying a rule's actions fails, ofproto restores the original
737 * - Removing a rule is not allowed to fail. It must always succeed.
739 * The ofproto base code serializes operations: if any operation is in
740 * progress on a given rule, ofproto postpones initiating any new operation
741 * on that rule until the pending operation completes. Therefore, every
742 * operation must eventually complete through a call to
743 * ofoperation_complete() to avoid delaying new operations indefinitely
744 * (including any OpenFlow request that affects the rule in question, even
745 * just to query its statistics).
751 * When ->rule_construct() is called, the caller has already inserted
752 * 'rule' into 'rule->ofproto''s flow table numbered 'rule->table_id'.
753 * There are two cases:
755 * - 'rule' is a new rule in its flow table. In this case,
756 * ofoperation_get_victim(rule) returns NULL.
758 * - 'rule' is replacing an existing rule in its flow table that had the
759 * same matching criteria and priority. In this case,
760 * ofoperation_get_victim(rule) returns the rule being replaced (the
763 * ->rule_construct() should set the following in motion:
765 * - Validate that the matching rule in 'rule->cr' is supported by the
766 * datapath. For example, if the rule's table does not support
767 * registers, then it is an error if 'rule->cr' does not wildcard all
770 * - Validate that 'rule->actions' and 'rule->n_actions' are well-formed
771 * OpenFlow actions that the datapath can correctly implement. The
772 * validate_actions() function (in ofp-util.c) can be useful as a model
773 * for action validation, but it accepts all of the OpenFlow actions
774 * that OVS understands. If your ofproto implementation only
775 * implements a subset of those, then you should implement your own
778 * - If the rule is valid, update the datapath flow table, adding the new
779 * rule or replacing the existing one.
781 * - If 'rule' is replacing an existing rule, uninitialize any derived
782 * state for the victim rule, as in step 5 in the "Life Cycle"
785 * (On failure, the ofproto code will roll back the insertion from the flow
786 * table, either removing 'rule' or replacing it by the victim rule if
789 * ->rule_construct() must act in one of the following ways:
791 * - If it succeeds, it must call ofoperation_complete() and return 0.
793 * - If it fails, it must act in one of the following ways:
795 * * Call ofoperation_complete() and return 0.
797 * * Return an OpenFlow error code. (Do not call
798 * ofoperation_complete() in this case.)
800 * Either way, ->rule_destruct() will not be called for 'rule', but
801 * ->rule_dealloc() will be.
803 * - If the operation is only partially complete, then it must return 0.
804 * Later, when the operation is complete, the ->run() or ->destruct()
805 * function must call ofoperation_complete() to report success or
808 * ->rule_construct() should not modify any base members of struct rule.
814 * When ->rule_destruct() is called, the caller has already removed 'rule'
815 * from 'rule->ofproto''s flow table. ->rule_destruct() should set in
816 * motion removing 'rule' from the datapath flow table. If removal
817 * completes synchronously, it should call ofoperation_complete().
818 * Otherwise, the ->run() or ->destruct() function must later call
819 * ofoperation_complete() after the operation completes.
821 * Rule destruction must not fail. */
822 struct rule *(*rule_alloc)(void);
823 enum ofperr (*rule_construct)(struct rule *rule);
824 void (*rule_destruct)(struct rule *rule);
825 void (*rule_dealloc)(struct rule *rule);
827 /* Obtains statistics for 'rule', storing the number of packets that have
828 * matched it in '*packet_count' and the number of bytes in those packets
829 * in '*byte_count'. UINT64_MAX indicates that the packet count or byte
830 * count is unknown. */
831 void (*rule_get_stats)(struct rule *rule, uint64_t *packet_count,
832 uint64_t *byte_count);
834 /* Applies the actions in 'rule' to 'packet'. (This implements sending
835 * buffered packets for OpenFlow OFPT_FLOW_MOD commands.)
837 * Takes ownership of 'packet' (so it should eventually free it, with
840 * 'flow' reflects the flow information for 'packet'. All of the
841 * information in 'flow' is extracted from 'packet', except for
842 * flow->tun_id and flow->in_port, which are assigned the correct values
843 * for the incoming packet. The register values are zeroed. 'packet''s
844 * header pointers (e.g. packet->l3) are appropriately initialized.
846 * The implementation should add the statistics for 'packet' into 'rule'.
848 * Returns 0 if successful, otherwise an OpenFlow error code. */
849 enum ofperr (*rule_execute)(struct rule *rule, const struct flow *flow,
850 struct ofpbuf *packet);
852 /* When ->rule_modify_actions() is called, the caller has already replaced
853 * the OpenFlow actions in 'rule' by a new set. (The original actions are
854 * in rule->pending->actions.)
856 * ->rule_modify_actions() should set the following in motion:
858 * - Validate that the actions now in 'rule' are well-formed OpenFlow
859 * actions that the datapath can correctly implement.
861 * - Update the datapath flow table with the new actions.
863 * If the operation synchronously completes, ->rule_modify_actions() may
864 * call ofoperation_complete() before it returns. Otherwise, ->run()
865 * should call ofoperation_complete() later, after the operation does
868 * If the operation fails, then the base ofproto code will restore the
869 * original 'actions' and 'n_actions' of 'rule'.
871 * ->rule_modify_actions() should not modify any base members of struct
873 void (*rule_modify_actions)(struct rule *rule);
875 /* Changes the OpenFlow IP fragment handling policy to 'frag_handling',
876 * which takes one of the following values, with the corresponding
879 * - OFPC_FRAG_NORMAL: The switch should treat IP fragments the same way
880 * as other packets, omitting TCP and UDP port numbers (always setting
883 * - OFPC_FRAG_DROP: The switch should drop all IP fragments without
884 * passing them through the flow table.
886 * - OFPC_FRAG_REASM: The switch should reassemble IP fragments before
887 * passing packets through the flow table.
889 * - OFPC_FRAG_NX_MATCH (a Nicira extension): Similar to OFPC_FRAG_NORMAL,
890 * except that TCP and UDP port numbers should be included in fragments
893 * Implementations are not required to support every mode.
894 * OFPC_FRAG_NORMAL is the default mode when an ofproto is created.
896 * At the time of the call to ->set_frag_handling(), the current mode is
897 * available in 'ofproto->frag_handling'. ->set_frag_handling() returns
898 * true if the requested mode was set, false if it is not supported.
900 * Upon successful return, the caller changes 'ofproto->frag_handling' to
901 * reflect the new mode.
903 bool (*set_frag_handling)(struct ofproto *ofproto,
904 enum ofp_config_flags frag_handling);
906 /* Implements the OpenFlow OFPT_PACKET_OUT command. The datapath should
907 * execute the 'n_actions' in the 'actions' array on 'packet'.
909 * The caller retains ownership of 'packet', so ->packet_out() should not
912 * This function must validate that the 'n_actions' elements in 'actions'
913 * are well-formed OpenFlow actions that can be correctly implemented by
914 * the datapath. If not, then it should return an OpenFlow error code.
916 * 'flow' reflects the flow information for 'packet'. All of the
917 * information in 'flow' is extracted from 'packet', except for
918 * flow->in_port, which is taken from the OFPT_PACKET_OUT message.
919 * flow->tun_id and its register values are zeroed.
921 * 'packet' is not matched against the OpenFlow flow table, so its
922 * statistics should not be included in OpenFlow flow statistics.
924 * Returns 0 if successful, otherwise an OpenFlow error code. */
925 enum ofperr (*packet_out)(struct ofproto *ofproto, struct ofpbuf *packet,
926 const struct flow *flow,
927 const union ofp_action *actions,
930 /* ## ------------------------- ## */
931 /* ## OFPP_NORMAL configuration ## */
932 /* ## ------------------------- ## */
934 /* Configures NetFlow on 'ofproto' according to the options in
935 * 'netflow_options', or turns off NetFlow if 'netflow_options' is NULL.
937 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
938 * NetFlow, as does a null pointer. */
939 int (*set_netflow)(struct ofproto *ofproto,
940 const struct netflow_options *netflow_options);
942 void (*get_netflow_ids)(const struct ofproto *ofproto,
943 uint8_t *engine_type, uint8_t *engine_id);
945 /* Configures sFlow on 'ofproto' according to the options in
946 * 'sflow_options', or turns off sFlow if 'sflow_options' is NULL.
948 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
949 * sFlow, as does a null pointer. */
950 int (*set_sflow)(struct ofproto *ofproto,
951 const struct ofproto_sflow_options *sflow_options);
953 /* Configures connectivity fault management on 'ofport'.
955 * If 'cfm_settings' is nonnull, configures CFM according to its members.
957 * If 'cfm_settings' is null, removes any connectivity fault management
958 * configuration from 'ofport'.
960 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
961 * support CFM, as does a null pointer. */
962 int (*set_cfm)(struct ofport *ofport, const struct cfm_settings *s);
964 /* Checks the fault status of CFM configured on 'ofport'. Returns a
965 * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
966 * indicating a connectivity problem). Returns zero if CFM is not faulted,
967 * and -1 if CFM is not enabled on 'port'.
969 * This function may be a null pointer if the ofproto implementation does
970 * not support CFM. */
971 int (*get_cfm_fault)(const struct ofport *ofport);
973 /* Gets the MPIDs of the remote maintenance points broadcasting to
974 * 'ofport'. Populates 'rmps' with a provider owned array of MPIDs, and
975 * 'n_rmps' with the number of MPIDs in 'rmps'. Returns a number less than
976 * 0 if CFM is not enabled of 'ofport'.
978 * This function may be a null pointer if the ofproto implementation does
979 * not support CFM. */
980 int (*get_cfm_remote_mpids)(const struct ofport *ofport,
981 const uint64_t **rmps, size_t *n_rmps);
983 /* Checks the health of CFM configured on 'ofport'. Returns an integer
984 * to indicate the health percentage of the 'ofport' which is an average of
985 * the health of all the remote_mps. Returns an integer between 0 and 100
986 * where 0 means that the 'ofport' is very unhealthy and 100 means the
987 * 'ofport' is perfectly healthy. Returns -1 if CFM is not enabled on
988 * 'port' or if the number of remote_mpids is > 1.
990 * This function may be a null pointer if the ofproto implementation does
991 * not support CFM. */
992 int (*get_cfm_health)(const struct ofport *ofport);
994 /* Configures spanning tree protocol (STP) on 'ofproto' using the
995 * settings defined in 's'.
997 * If 's' is nonnull, configures STP according to its members.
999 * If 's' is null, removes any STP configuration from 'ofproto'.
1001 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1002 * support STP, as does a null pointer. */
1003 int (*set_stp)(struct ofproto *ofproto,
1004 const struct ofproto_stp_settings *s);
1006 /* Retrieves state of spanning tree protocol (STP) on 'ofproto'.
1008 * Stores STP state for 'ofproto' in 's'. If the 'enabled' member
1009 * is false, the other member values are not meaningful.
1011 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1012 * support STP, as does a null pointer. */
1013 int (*get_stp_status)(struct ofproto *ofproto,
1014 struct ofproto_stp_status *s);
1016 /* Configures spanning tree protocol (STP) on 'ofport' using the
1017 * settings defined in 's'.
1019 * If 's' is nonnull, configures STP according to its members. The
1020 * caller is responsible for assigning STP port numbers (using the
1021 * 'port_num' member in the range of 1 through 255, inclusive) and
1022 * ensuring there are no duplicates.
1024 * If 's' is null, removes any STP configuration from 'ofport'.
1026 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1027 * support STP, as does a null pointer. */
1028 int (*set_stp_port)(struct ofport *ofport,
1029 const struct ofproto_port_stp_settings *s);
1031 /* Retrieves spanning tree protocol (STP) port status of 'ofport'.
1033 * Stores STP state for 'ofport' in 's'. If the 'enabled' member is
1034 * false, the other member values are not meaningful.
1036 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1037 * support STP, as does a null pointer. */
1038 int (*get_stp_port_status)(struct ofport *ofport,
1039 struct ofproto_port_stp_status *s);
1041 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
1042 * 'queues' attached to 'ofport'. This data is not intended to be
1043 * sufficient to implement QoS. Instead, providers may use this
1044 * information to implement features which require knowledge of what queues
1045 * exist on a port, and some basic information about them.
1047 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1048 * support QoS, as does a null pointer. */
1049 int (*set_queues)(struct ofport *ofport,
1050 const struct ofproto_port_queue *queues, size_t n_qdscp);
1052 /* If 's' is nonnull, this function registers a "bundle" associated with
1053 * client data pointer 'aux' in 'ofproto'. A bundle is the same concept as
1054 * a Port in OVSDB, that is, it consists of one or more "slave" devices
1055 * (Interfaces, in OVSDB) along with VLAN and LACP configuration and, if
1056 * there is more than one slave, a bonding configuration. If 'aux' is
1057 * already registered then this function updates its configuration to 's'.
1058 * Otherwise, this function registers a new bundle.
1060 * If 's' is NULL, this function unregisters the bundle registered on
1061 * 'ofproto' associated with client data pointer 'aux'. If no such bundle
1062 * has been registered, this has no effect.
1064 * This function affects only the behavior of the NXAST_AUTOPATH action and
1065 * output to the OFPP_NORMAL port. An implementation that does not support
1066 * it at all may set it to NULL or return EOPNOTSUPP. An implementation
1067 * that supports only a subset of the functionality should implement what
1068 * it can and return 0. */
1069 int (*bundle_set)(struct ofproto *ofproto, void *aux,
1070 const struct ofproto_bundle_settings *s);
1072 /* If 'port' is part of any bundle, removes it from that bundle. If the
1073 * bundle now has no ports, deletes the bundle. If the bundle now has only
1074 * one port, deconfigures the bundle's bonding configuration. */
1075 void (*bundle_remove)(struct ofport *ofport);
1077 /* If 's' is nonnull, this function registers a mirror associated with
1078 * client data pointer 'aux' in 'ofproto'. A mirror is the same concept as
1079 * a Mirror in OVSDB. If 'aux' is already registered then this function
1080 * updates its configuration to 's'. Otherwise, this function registers a
1083 * If 's' is NULL, this function unregisters the mirror registered on
1084 * 'ofproto' associated with client data pointer 'aux'. If no such mirror
1085 * has been registered, this has no effect.
1087 * An implementation that does not support mirroring at all may set
1088 * it to NULL or return EOPNOTSUPP. An implementation that supports
1089 * only a subset of the functionality should implement what it can
1091 int (*mirror_set)(struct ofproto *ofproto, void *aux,
1092 const struct ofproto_mirror_settings *s);
1094 /* Retrieves statistics from mirror associated with client data
1095 * pointer 'aux' in 'ofproto'. Stores packet and byte counts in
1096 * 'packets' and 'bytes', respectively. If a particular counter is
1097 * not supported, the appropriate argument is set to UINT64_MAX.
1099 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1100 * support retrieving mirror statistics. */
1101 int (*mirror_get_stats)(struct ofproto *ofproto, void *aux,
1102 uint64_t *packets, uint64_t *bytes);
1104 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs
1105 * on which all packets are flooded, instead of using MAC learning. If
1106 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1108 * This function affects only the behavior of the OFPP_NORMAL action. An
1109 * implementation that does not support it may set it to NULL or return
1111 int (*set_flood_vlans)(struct ofproto *ofproto,
1112 unsigned long *flood_vlans);
1114 /* Returns true if 'aux' is a registered bundle that is currently in use as
1115 * the output for a mirror. */
1116 bool (*is_mirror_output_bundle)(const struct ofproto *ofproto, void *aux);
1118 /* When the configuration option of forward_bpdu changes, this function
1119 * will be invoked. */
1120 void (*forward_bpdu_changed)(struct ofproto *ofproto);
1122 /* Sets the MAC aging timeout for the OFPP_NORMAL action to 'idle_time',
1124 void (*set_mac_idle_time)(struct ofproto *ofproto, unsigned int idle_time);
1126 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
1128 * This is deprecated. It is only for compatibility with broken device drivers
1129 * in old versions of Linux that do not properly support VLANs when VLAN
1130 * devices are not used. When broken device drivers are no longer in
1131 * widespread use, we will delete these interfaces. */
1133 /* If 'realdev_ofp_port' is nonzero, then this function configures 'ofport'
1134 * as a VLAN splinter port for VLAN 'vid', associated with the real device
1135 * that has OpenFlow port number 'realdev_ofp_port'.
1137 * If 'realdev_ofp_port' is zero, then this function deconfigures 'ofport'
1138 * as a VLAN splinter port.
1140 * This function should be NULL if a an implementation does not support
1142 int (*set_realdev)(struct ofport *ofport,
1143 uint16_t realdev_ofp_port, int vid);
1146 extern const struct ofproto_class ofproto_dpif_class;
1148 int ofproto_class_register(const struct ofproto_class *);
1149 int ofproto_class_unregister(const struct ofproto_class *);
1151 /* ofproto_flow_mod() returns this value if the flow_mod could not be processed
1152 * because it overlaps with an ongoing flow table operation that has not yet
1153 * completed. The caller should retry the operation later.
1155 * ofproto.c also uses this value internally for additional (similar) purposes.
1157 * This particular value is a good choice because it is large, so that it does
1158 * not collide with any errno value, but not large enough to collide with an
1159 * OFPERR_* value. */
1160 enum { OFPROTO_POSTPONE = 1 << 16 };
1161 BUILD_ASSERT_DECL(OFPROTO_POSTPONE < OFPERR_OFS);
1163 int ofproto_flow_mod(struct ofproto *, const struct ofputil_flow_mod *);
1164 void ofproto_add_flow(struct ofproto *, const struct cls_rule *,
1165 const union ofp_action *, size_t n_actions);
1166 bool ofproto_delete_flow(struct ofproto *, const struct cls_rule *);
1167 void ofproto_flush_flows(struct ofproto *);
1169 #endif /* ofproto/ofproto-provider.h */