2 * Copyright (c) 2009, 2010, 2011 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"
29 struct ofputil_flow_mod;
31 /* An OpenFlow switch.
33 * With few exceptions, ofproto implementations may look at these fields but
34 * should not modify them. */
36 const struct ofproto_class *ofproto_class;
37 char *type; /* Datapath type. */
38 char *name; /* Datapath name. */
39 struct hmap_node hmap_node; /* In global 'all_ofprotos' hmap. */
42 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
43 uint64_t datapath_id; /* Datapath ID. */
44 unsigned flow_eviction_threshold; /* Threshold at which to begin flow
45 * table eviction. Only affects the
46 * ofproto-dpif implementation */
47 bool forward_bpdu; /* Option to allow forwarding of BPDU frames
48 * when NORMAL action is invoked. */
49 char *mfr_desc; /* Manufacturer. */
50 char *hw_desc; /* Hardware. */
51 char *sw_desc; /* Software version. */
52 char *serial_desc; /* Serial number. */
53 char *dp_desc; /* Datapath description. */
54 enum ofp_config_flags frag_handling; /* One of OFPC_*. */
57 struct hmap ports; /* Contains "struct ofport"s. */
58 struct shash port_by_name;
61 struct classifier *tables; /* Each classifier contains "struct rule"s. */
64 /* OpenFlow connections. */
65 struct connmgr *connmgr;
67 /* Flow table operation tracking. */
68 int state; /* Internal state. */
69 struct list pending; /* List of "struct ofopgroup"s. */
70 unsigned int n_pending; /* list_size(&pending). */
71 struct hmap deletions; /* All OFOPERATION_DELETE "ofoperation"s. */
73 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
75 * This is deprecated. It is only for compatibility with broken device
76 * drivers in old versions of Linux that do not properly support VLANs when
77 * VLAN devices are not used. When broken device drivers are no longer in
78 * widespread use, we will delete these interfaces. */
79 unsigned long int *vlan_bitmap; /* 4096-bit bitmap of in-use VLANs. */
80 bool vlans_changed; /* True if new VLANs are in use. */
83 struct ofproto *ofproto_lookup(const char *name);
84 struct ofport *ofproto_get_port(const struct ofproto *, uint16_t ofp_port);
86 /* Assigns CLS to each classifier table, in turn, in OFPROTO.
88 * All parameters are evaluated multiple times. */
89 #define OFPROTO_FOR_EACH_TABLE(CLS, OFPROTO) \
90 for ((CLS) = (OFPROTO)->tables; \
91 (CLS) < &(OFPROTO)->tables[(OFPROTO)->n_tables]; \
94 /* An OpenFlow port within a "struct ofproto".
96 * With few exceptions, ofproto implementations may look at these fields but
97 * should not modify them. */
99 struct ofproto *ofproto; /* The ofproto that contains this port. */
100 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
101 struct netdev *netdev;
102 struct ofp_phy_port opp;
103 uint16_t ofp_port; /* OpenFlow port number. */
104 unsigned int change_seq;
108 void ofproto_port_set_state(struct ofport *, ovs_be32 state);
110 /* An OpenFlow flow within a "struct ofproto".
112 * With few exceptions, ofproto implementations may look at these fields but
113 * should not modify them. */
115 struct ofproto *ofproto; /* The ofproto that contains this rule. */
116 struct list ofproto_node; /* Owned by ofproto base code. */
117 struct cls_rule cr; /* In owning ofproto's classifier. */
119 struct ofoperation *pending; /* Operation now in progress, if nonnull. */
121 ovs_be64 flow_cookie; /* Controller-issued identifier. */
123 long long int created; /* Creation time. */
124 long long int modified; /* Time of last modification. */
125 uint16_t idle_timeout; /* In seconds from time of last use. */
126 uint16_t hard_timeout; /* In seconds from last modification. */
127 uint8_t table_id; /* Index in ofproto's 'tables' array. */
128 bool send_flow_removed; /* Send a flow removed message? */
130 union ofp_action *actions; /* OpenFlow actions. */
131 int n_actions; /* Number of elements in actions[]. */
134 static inline struct rule *
135 rule_from_cls_rule(const struct cls_rule *cls_rule)
137 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
140 void ofproto_rule_expire(struct rule *, uint8_t reason);
141 void ofproto_rule_destroy(struct rule *);
143 void ofoperation_complete(struct ofoperation *, int status);
144 struct rule *ofoperation_get_victim(struct ofoperation *);
146 /* ofproto class structure, to be defined by each ofproto implementation.
152 * These functions work primarily with three different kinds of data
155 * - "struct ofproto", which represents an OpenFlow switch.
157 * - "struct ofport", which represents a port within an ofproto.
159 * - "struct rule", which represents an OpenFlow flow within an ofproto.
161 * Each of these data structures contains all of the implementation-independent
162 * generic state for the respective concept, called the "base" state. None of
163 * them contains any extra space for ofproto implementations to use. Instead,
164 * each implementation is expected to declare its own data structure that
165 * contains an instance of the generic data structure plus additional
166 * implementation-specific members, called the "derived" state. The
167 * implementation can use casts or (preferably) the CONTAINER_OF macro to
168 * obtain access to derived state given only a pointer to the embedded generic
175 * Four stylized functions accompany each of these data structures:
177 * "alloc" "construct" "destruct" "dealloc"
178 * ------------ ---------------- --------------- --------------
179 * ofproto ->alloc ->construct ->destruct ->dealloc
180 * ofport ->port_alloc ->port_construct ->port_destruct ->port_dealloc
181 * rule ->rule_alloc ->rule_construct ->rule_destruct ->rule_dealloc
183 * Any instance of a given data structure goes through the following life
186 * 1. The client calls the "alloc" function to obtain raw memory. If "alloc"
187 * fails, skip all the other steps.
189 * 2. The client initializes all of the data structure's base state. If this
190 * fails, skip to step 7.
192 * 3. The client calls the "construct" function. The implementation
193 * initializes derived state. It may refer to the already-initialized
194 * base state. If "construct" fails, skip to step 6.
196 * 4. The data structure is now initialized and in use.
198 * 5. When the data structure is no longer needed, the client calls the
199 * "destruct" function. The implementation uninitializes derived state.
200 * The base state has not been uninitialized yet, so the implementation
201 * may still refer to it.
203 * 6. The client uninitializes all of the data structure's base state.
205 * 7. The client calls the "dealloc" to free the raw memory. The
206 * implementation must not refer to base or derived state in the data
207 * structure, because it has already been uninitialized.
209 * Each "alloc" function allocates and returns a new instance of the respective
210 * data structure. The "alloc" function is not given any information about the
211 * use of the new data structure, so it cannot perform much initialization.
212 * Its purpose is just to ensure that the new data structure has enough room
213 * for base and derived state. It may return a null pointer if memory is not
214 * available, in which case none of the other functions is called.
216 * Each "construct" function initializes derived state in its respective data
217 * structure. When "construct" is called, all of the base state has already
218 * been initialized, so the "construct" function may refer to it. The
219 * "construct" function is allowed to fail, in which case the client calls the
220 * "dealloc" function (but not the "destruct" function).
222 * Each "destruct" function uninitializes and frees derived state in its
223 * respective data structure. When "destruct" is called, the base state has
224 * not yet been uninitialized, so the "destruct" function may refer to it. The
225 * "destruct" function is not allowed to fail.
227 * Each "dealloc" function frees raw memory that was allocated by the the
228 * "alloc" function. The memory's base and derived members might not have ever
229 * been initialized (but if "construct" returned successfully, then it has been
230 * "destruct"ed already). The "dealloc" function is not allowed to fail.
236 * Most of these functions return 0 if they are successful or a positive error
237 * code on failure. Depending on the function, valid error codes are either
238 * errno values or OpenFlow error codes constructed with ofp_mkerr().
240 * Most of these functions are expected to execute synchronously, that is, to
241 * block as necessary to obtain a result. Thus, these functions may return
242 * EAGAIN (or EWOULDBLOCK or EINPROGRESS) only where the function descriptions
243 * explicitly say those errors are a possibility. We may relax this
244 * requirement in the future if and when we encounter performance problems. */
245 struct ofproto_class {
246 /* ## ----------------- ## */
247 /* ## Factory Functions ## */
248 /* ## ----------------- ## */
250 /* Enumerates the types of all support ofproto types into 'types'. The
251 * caller has already initialized 'types' and other ofproto classes might
252 * already have added names to it. */
253 void (*enumerate_types)(struct sset *types);
255 /* Enumerates the names of all existing datapath of the specified 'type'
256 * into 'names' 'all_dps'. The caller has already initialized 'names' as
259 * 'type' is one of the types enumerated by ->enumerate_types().
261 * Returns 0 if successful, otherwise a positive errno value.
263 int (*enumerate_names)(const char *type, struct sset *names);
265 /* Deletes the datapath with the specified 'type' and 'name'. The caller
266 * should have closed any open ofproto with this 'type' and 'name'; this
267 * function is allowed to fail if that is not the case.
269 * 'type' is one of the types enumerated by ->enumerate_types().
270 * 'name' is one of the names enumerated by ->enumerate_names() for 'type'.
272 * Returns 0 if successful, otherwise a positive errno value.
274 int (*del)(const char *type, const char *name);
276 /* ## --------------------------- ## */
277 /* ## Top-Level ofproto Functions ## */
278 /* ## --------------------------- ## */
280 /* Life-cycle functions for an "ofproto" (see "Life Cycle" above).
286 * ->construct() should not modify any base members of the ofproto. The
287 * client will initialize the ofproto's 'ports' and 'tables' members after
288 * construction is complete.
290 * When ->construct() is called, the client does not yet know how many flow
291 * tables the datapath supports, so ofproto->n_tables will be 0 and
292 * ofproto->tables will be NULL. ->construct() should store the number of
293 * flow tables supported by the datapath (between 1 and 255, inclusive)
294 * into '*n_tables'. After a successful return, the client will initialize
295 * the base 'n_tables' member to '*n_tables' and allocate and initialize
296 * the base 'tables' member as the specified number of empty flow tables.
297 * Each flow table will be initially empty, so ->construct() should delete
298 * flows from the underlying datapath, if necessary, rather than populating
301 * Only one ofproto instance needs to be supported for any given datapath.
302 * If a datapath is already open as part of one "ofproto", then another
303 * attempt to "construct" the same datapath as part of another ofproto is
304 * allowed to fail with an error.
306 * ->construct() returns 0 if successful, otherwise a positive errno
313 * If 'ofproto' has any pending asynchronous operations, ->destruct()
314 * must complete all of them by calling ofoperation_complete().
316 * ->destruct() must also destroy all remaining rules in the ofproto's
317 * tables, by passing each remaining rule to ofproto_rule_destroy(). The
318 * client will destroy the flow tables themselves after ->destruct()
321 struct ofproto *(*alloc)(void);
322 int (*construct)(struct ofproto *ofproto, int *n_tables);
323 void (*destruct)(struct ofproto *ofproto);
324 void (*dealloc)(struct ofproto *ofproto);
326 /* Performs any periodic activity required by 'ofproto'. It should:
328 * - Call connmgr_send_packet_in() for each received packet that missed
329 * in the OpenFlow flow table or that had a OFPP_CONTROLLER output
332 * - Call ofproto_rule_expire() for each OpenFlow flow that has reached
333 * its hard_timeout or idle_timeout, to expire the flow.
335 * Returns 0 if successful, otherwise a positive errno value. */
336 int (*run)(struct ofproto *ofproto);
338 /* Performs periodic activity required by 'ofproto' that needs to be done
339 * with the least possible latency.
341 * This is run multiple times per main loop. An ofproto provider may
342 * implement it or not, according to whether it provides a performance
343 * boost for that ofproto implementation. */
344 int (*run_fast)(struct ofproto *ofproto);
346 /* Causes the poll loop to wake up when 'ofproto''s 'run' function needs to
347 * be called, e.g. by calling the timer or fd waiting functions in
349 void (*wait)(struct ofproto *ofproto);
351 /* Every "struct rule" in 'ofproto' is about to be deleted, one by one.
352 * This function may prepare for that, for example by clearing state in
353 * advance. It should *not* actually delete any "struct rule"s from
354 * 'ofproto', only prepare for it.
356 * This function is optional; it's really just for optimization in case
357 * it's cheaper to delete all the flows from your hardware in a single pass
358 * than to do it one by one. */
359 void (*flush)(struct ofproto *ofproto);
361 /* Helper for the OpenFlow OFPT_FEATURES_REQUEST request.
363 * The implementation should store true in '*arp_match_ip' if the switch
364 * supports matching IP addresses inside ARP requests and replies, false
367 * The implementation should store in '*actions' a bitmap of the supported
368 * OpenFlow actions: the bit with value (1 << n) should be set to 1 if the
369 * implementation supports the action with value 'n', and to 0 otherwise.
370 * For example, if the implementation supports the OFPAT_OUTPUT and
371 * OFPAT_ENQUEUE actions, but no others, it would set '*actions' to (1 <<
372 * OFPAT_OUTPUT) | (1 << OFPAT_ENQUEUE). Vendor actions are not included
374 void (*get_features)(struct ofproto *ofproto,
375 bool *arp_match_ip, uint32_t *actions);
377 /* Helper for the OpenFlow OFPST_TABLE statistics request.
379 * The 'ots' array contains 'ofproto->n_tables' elements. Each element is
382 * - 'table_id' to the array index.
384 * - 'name' to "table#" where # is the table ID.
386 * - 'wildcards' to OFPFW_ALL.
388 * - 'max_entries' to 1,000,000.
390 * - 'active_count' to the classifier_count() for the table.
392 * - 'lookup_count' and 'matched_count' to 0.
394 * The implementation should update any members in each element for which
395 * it has better values:
397 * - 'name' to a more meaningful name.
399 * - 'wildcards' to the set of wildcards actually supported by the table
400 * (if it doesn't support all OpenFlow wildcards).
402 * - 'max_entries' to the maximum number of flows actually supported by
405 * - 'lookup_count' to the number of packets looked up in this flow table
408 * - 'matched_count' to the number of packets looked up in this flow
409 * table so far that matched one of the flow entries.
411 * Keep in mind that all of the members of struct ofp_table_stats are in
412 * network byte order.
414 void (*get_tables)(struct ofproto *ofproto, struct ofp_table_stats *ots);
416 /* ## ---------------- ## */
417 /* ## ofport Functions ## */
418 /* ## ---------------- ## */
420 /* Life-cycle functions for a "struct ofport" (see "Life Cycle" above).
422 * ->port_construct() should not modify any base members of the ofport.
424 * ofports are managed by the base ofproto code. The ofproto
425 * implementation should only create and destroy them in response to calls
426 * to these functions. The base ofproto code will create and destroy
427 * ofports in the following situations:
429 * - Just after the ->construct() function is called, the base ofproto
430 * iterates over all of the implementation's ports, using
431 * ->port_dump_start() and related functions, and constructs an ofport
432 * for each dumped port.
434 * - If ->port_poll() reports that a specific port has changed, then the
435 * base ofproto will query that port with ->port_query_by_name() and
436 * construct or destruct ofports as necessary to reflect the updated
439 * - If ->port_poll() returns ENOBUFS to report an unspecified port set
440 * change, then the base ofproto will iterate over all of the
441 * implementation's ports, in the same way as at ofproto
442 * initialization, and construct and destruct ofports to reflect all of
445 * ->port_construct() returns 0 if successful, otherwise a positive errno
448 struct ofport *(*port_alloc)(void);
449 int (*port_construct)(struct ofport *ofport);
450 void (*port_destruct)(struct ofport *ofport);
451 void (*port_dealloc)(struct ofport *ofport);
453 /* Called after 'ofport->netdev' is replaced by a new netdev object. If
454 * the ofproto implementation uses the ofport's netdev internally, then it
455 * should switch to using the new one. The old one has been closed.
457 * An ofproto implementation that doesn't need to do anything in this
458 * function may use a null pointer. */
459 void (*port_modified)(struct ofport *ofport);
461 /* Called after an OpenFlow OFPT_PORT_MOD request changes a port's
462 * configuration. 'ofport->opp.config' contains the new configuration.
463 * 'old_config' contains the previous configuration.
465 * The caller implements OFPPC_PORT_DOWN using netdev functions to turn
466 * NETDEV_UP on and off, so this function doesn't have to do anything for
467 * that bit (and it won't be called if that is the only bit that
469 void (*port_reconfigured)(struct ofport *ofport, ovs_be32 old_config);
471 /* Looks up a port named 'devname' in 'ofproto'. On success, initializes
472 * '*port' appropriately.
474 * The caller owns the data in 'port' and must free it with
475 * ofproto_port_destroy() when it is no longer needed. */
476 int (*port_query_by_name)(const struct ofproto *ofproto,
477 const char *devname, struct ofproto_port *port);
479 /* Attempts to add 'netdev' as a port on 'ofproto'. Returns 0 if
480 * successful, otherwise a positive errno value. If successful, sets
481 * '*ofp_portp' to the new port's port number.
483 * It doesn't matter whether the new port will be returned by a later call
484 * to ->port_poll(); the implementation may do whatever is more
486 int (*port_add)(struct ofproto *ofproto, struct netdev *netdev,
487 uint16_t *ofp_portp);
489 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'. Returns
490 * 0 if successful, otherwise a positive errno value.
492 * It doesn't matter whether the new port will be returned by a later call
493 * to ->port_poll(); the implementation may do whatever is more
495 int (*port_del)(struct ofproto *ofproto, uint16_t ofp_port);
497 /* Port iteration functions.
499 * The client might not be entirely in control of the ports within an
500 * ofproto. Some hardware implementations, for example, might have a fixed
501 * set of ports in a datapath, and the Linux datapath allows the system
502 * administrator to externally add and remove ports with ovs-dpctl. For
503 * this reason, the client needs a way to iterate through all the ports
504 * that are actually in a datapath. These functions provide that
507 * The 'state' pointer provides the implementation a place to
508 * keep track of its position. Its format is opaque to the caller.
510 * The ofproto provider retains ownership of the data that it stores into
511 * ->port_dump_next()'s 'port' argument. The data must remain valid until
512 * at least the next call to ->port_dump_next() or ->port_dump_done() for
513 * 'state'. The caller will not modify or free it.
518 * ->port_dump_start() attempts to begin dumping the ports in 'ofproto'.
519 * On success, it should return 0 and initialize '*statep' with any data
520 * needed for iteration. On failure, returns a positive errno value, and
521 * the client will not call ->port_dump_next() or ->port_dump_done().
523 * ->port_dump_next() attempts to retrieve another port from 'ofproto' for
524 * 'state'. If there is another port, it should store the port's
525 * information into 'port' and return 0. It should return EOF if all ports
526 * have already been iterated. Otherwise, on error, it should return a
527 * positive errno value. This function will not be called again once it
528 * returns nonzero once for a given iteration (but the 'port_dump_done'
529 * function will be called afterward).
531 * ->port_dump_done() allows the implementation to release resources used
532 * for iteration. The caller might decide to stop iteration in the middle
533 * by calling this function before ->port_dump_next() returns nonzero.
541 * error = ofproto->ofproto_class->port_dump_start(ofproto, &state);
544 * struct ofproto_port port;
546 * error = ofproto->ofproto_class->port_dump_next(
547 * ofproto, state, &port);
551 * // Do something with 'port' here (without modifying or freeing
552 * // any of its data).
554 * ofproto->ofproto_class->port_dump_done(ofproto, state);
556 * // 'error' is now EOF (success) or a positive errno value (failure).
558 int (*port_dump_start)(const struct ofproto *ofproto, void **statep);
559 int (*port_dump_next)(const struct ofproto *ofproto, void *state,
560 struct ofproto_port *port);
561 int (*port_dump_done)(const struct ofproto *ofproto, void *state);
563 /* Polls for changes in the set of ports in 'ofproto'. If the set of ports
564 * in 'ofproto' has changed, then this function should do one of the
567 * - Preferably: store the name of the device that was added to or deleted
568 * from 'ofproto' in '*devnamep' and return 0. The caller is responsible
569 * for freeing '*devnamep' (with free()) when it no longer needs it.
571 * - Alternatively: return ENOBUFS, without indicating the device that was
574 * Occasional 'false positives', in which the function returns 0 while
575 * indicating a device that was not actually added or deleted or returns
576 * ENOBUFS without any change, are acceptable.
578 * The purpose of 'port_poll' is to let 'ofproto' know about changes made
579 * externally to the 'ofproto' object, e.g. by a system administrator via
580 * ovs-dpctl. Therefore, it's OK, and even preferable, for port_poll() to
581 * not report changes made through calls to 'port_add' or 'port_del' on the
582 * same 'ofproto' object. (But it's OK for it to report them too, just
583 * slightly less efficient.)
585 * If the set of ports in 'ofproto' has not changed, returns EAGAIN. May
586 * also return other positive errno values to indicate that something has
589 * If the set of ports in a datapath is fixed, or if the only way that the
590 * set of ports in a datapath can change is through ->port_add() and
591 * ->port_del(), then this function may be a null pointer.
593 int (*port_poll)(const struct ofproto *ofproto, char **devnamep);
595 /* Arranges for the poll loop to wake up when ->port_poll() will return a
596 * value other than EAGAIN.
598 * If the set of ports in a datapath is fixed, or if the only way that the
599 * set of ports in a datapath can change is through ->port_add() and
600 * ->port_del(), or if the poll loop will always wake up anyway when
601 * ->port_poll() will return a value other than EAGAIN, then this function
602 * may be a null pointer.
604 void (*port_poll_wait)(const struct ofproto *ofproto);
606 /* Checks the status of LACP negotiation for 'port'. Returns 1 if LACP
607 * partner information for 'port' is up-to-date, 0 if LACP partner
608 * information is not current (generally indicating a connectivity
609 * problem), or -1 if LACP is not enabled on 'port'.
611 * This function may be a null pointer if the ofproto implementation does
612 * not support LACP. */
613 int (*port_is_lacp_current)(const struct ofport *port);
615 /* ## ----------------------- ## */
616 /* ## OpenFlow Rule Functions ## */
617 /* ## ----------------------- ## */
621 /* Chooses an appropriate table for 'cls_rule' within 'ofproto'. On
622 * success, stores the table ID into '*table_idp' and returns 0. On
623 * failure, returns an OpenFlow error code (as returned by ofp_mkerr()).
625 * The choice of table should be a function of 'cls_rule' and 'ofproto''s
626 * datapath capabilities. It should not depend on the flows already in
627 * 'ofproto''s flow tables. Failure implies that an OpenFlow rule with
628 * 'cls_rule' as its matching condition can never be inserted into
629 * 'ofproto', even starting from an empty flow table.
631 * If multiple tables are candidates for inserting the flow, the function
632 * should choose one arbitrarily (but deterministically).
634 * If this function is NULL then table 0 is always chosen. */
635 int (*rule_choose_table)(const struct ofproto *ofproto,
636 const struct cls_rule *cls_rule,
639 /* Life-cycle functions for a "struct rule" (see "Life Cycle" above).
642 * Asynchronous Operation Support
643 * ==============================
645 * The life-cycle operations on rules can operate asynchronously, meaning
646 * that ->rule_construct() and ->rule_destruct() only need to initiate
647 * their respective operations and do not need to wait for them to complete
648 * before they return. ->rule_modify_actions() also operates
651 * An ofproto implementation reports the success or failure of an
652 * asynchronous operation on a rule using the rule's 'pending' member,
653 * which points to a opaque "struct ofoperation" that represents the
654 * ongoing opreation. When the operation completes, the ofproto
655 * implementation calls ofoperation_complete(), passing the ofoperation and
656 * an error indication.
658 * Only the following contexts may call ofoperation_complete():
660 * - The function called to initiate the operation,
661 * e.g. ->rule_construct() or ->rule_destruct(). This is the best
662 * choice if the operation completes quickly.
664 * - The implementation's ->run() function.
666 * - The implementation's ->destruct() function.
668 * The ofproto base code updates the flow table optimistically, assuming
669 * that the operation will probably succeed:
671 * - ofproto adds or replaces the rule in the flow table before calling
672 * ->rule_construct().
674 * - ofproto updates the rule's actions before calling
675 * ->rule_modify_actions().
677 * - ofproto removes the rule before calling ->rule_destruct().
679 * With one exception, when an asynchronous operation completes with an
680 * error, ofoperation_complete() backs out the already applied changes:
682 * - If adding or replacing a rule in the flow table fails, ofproto
683 * removes the new rule or restores the original rule.
685 * - If modifying a rule's actions fails, ofproto restores the original
688 * - Removing a rule is not allowed to fail. It must always succeed.
690 * The ofproto base code serializes operations: if any operation is in
691 * progress on a given rule, ofproto postpones initiating any new operation
692 * on that rule until the pending operation completes. Therefore, every
693 * operation must eventually complete through a call to
694 * ofoperation_complete() to avoid delaying new operations indefinitely
695 * (including any OpenFlow request that affects the rule in question, even
696 * just to query its statistics).
702 * When ->rule_construct() is called, the caller has already inserted
703 * 'rule' into 'rule->ofproto''s flow table numbered 'rule->table_id'.
704 * There are two cases:
706 * - 'rule' is a new rule in its flow table. In this case,
707 * ofoperation_get_victim(rule) returns NULL.
709 * - 'rule' is replacing an existing rule in its flow table that had the
710 * same matching criteria and priority. In this case,
711 * ofoperation_get_victim(rule) returns the rule being replaced (the
714 * ->rule_construct() should set the following in motion:
716 * - Validate that the matching rule in 'rule->cr' is supported by the
717 * datapath. For example, if the rule's table does not support
718 * registers, then it is an error if 'rule->cr' does not wildcard all
721 * - Validate that 'rule->actions' and 'rule->n_actions' are well-formed
722 * OpenFlow actions that the datapath can correctly implement. The
723 * validate_actions() function (in ofp-util.c) can be useful as a model
724 * for action validation, but it accepts all of the OpenFlow actions
725 * that OVS understands. If your ofproto implementation only
726 * implements a subset of those, then you should implement your own
729 * - If the rule is valid, update the datapath flow table, adding the new
730 * rule or replacing the existing one.
732 * - If 'rule' is replacing an existing rule, uninitialize any derived
733 * state for the victim rule, as in step 5 in the "Life Cycle"
736 * (On failure, the ofproto code will roll back the insertion from the flow
737 * table, either removing 'rule' or replacing it by the victim rule if
740 * ->rule_construct() must act in one of the following ways:
742 * - If it succeeds, it must call ofoperation_complete() and return 0.
744 * - If it fails, it must act in one of the following ways:
746 * * Call ofoperation_complete() and return 0.
748 * * Return an OpenFlow error code (as returned by ofp_mkerr()). (Do
749 * not call ofoperation_complete() in this case.)
751 * Either way, ->rule_destruct() will not be called for 'rule', but
752 * ->rule_dealloc() will be.
754 * - If the operation is only partially complete, then it must return 0.
755 * Later, when the operation is complete, the ->run() or ->destruct()
756 * function must call ofoperation_complete() to report success or
759 * ->rule_construct() should not modify any base members of struct rule.
765 * When ->rule_destruct() is called, the caller has already removed 'rule'
766 * from 'rule->ofproto''s flow table. ->rule_destruct() should set in
767 * motion removing 'rule' from the datapath flow table. If removal
768 * completes synchronously, it should call ofoperation_complete().
769 * Otherwise, the ->run() or ->destruct() function must later call
770 * ofoperation_complete() after the operation completes.
772 * Rule destruction must not fail. */
773 struct rule *(*rule_alloc)(void);
774 int (*rule_construct)(struct rule *rule);
775 void (*rule_destruct)(struct rule *rule);
776 void (*rule_dealloc)(struct rule *rule);
778 /* Obtains statistics for 'rule', storing the number of packets that have
779 * matched it in '*packet_count' and the number of bytes in those packets
780 * in '*byte_count'. UINT64_MAX indicates that the packet count or byte
781 * count is unknown. */
782 void (*rule_get_stats)(struct rule *rule, uint64_t *packet_count,
783 uint64_t *byte_count);
785 /* Applies the actions in 'rule' to 'packet'. (This implements sending
786 * buffered packets for OpenFlow OFPT_FLOW_MOD commands.)
788 * Takes ownership of 'packet' (so it should eventually free it, with
791 * 'flow' reflects the flow information for 'packet'. All of the
792 * information in 'flow' is extracted from 'packet', except for
793 * flow->tun_id and flow->in_port, which are assigned the correct values
794 * for the incoming packet. The register values are zeroed.
796 * The statistics for 'packet' should be included in 'rule'.
798 * Returns 0 if successful, otherwise an OpenFlow error code (as returned
799 * by ofp_mkerr()). */
800 int (*rule_execute)(struct rule *rule, const struct flow *flow,
801 struct ofpbuf *packet);
803 /* When ->rule_modify_actions() is called, the caller has already replaced
804 * the OpenFlow actions in 'rule' by a new set. (The original actions are
805 * in rule->pending->actions.)
807 * ->rule_modify_actions() should set the following in motion:
809 * - Validate that the actions now in 'rule' are well-formed OpenFlow
810 * actions that the datapath can correctly implement.
812 * - Update the datapath flow table with the new actions.
814 * If the operation synchronously completes, ->rule_modify_actions() may
815 * call ofoperation_complete() before it returns. Otherwise, ->run()
816 * should call ofoperation_complete() later, after the operation does
819 * If the operation fails, then the base ofproto code will restore the
820 * original 'actions' and 'n_actions' of 'rule'.
822 * ->rule_modify_actions() should not modify any base members of struct
824 void (*rule_modify_actions)(struct rule *rule);
826 /* Changes the OpenFlow IP fragment handling policy to 'frag_handling',
827 * which takes one of the following values, with the corresponding
830 * - OFPC_FRAG_NORMAL: The switch should treat IP fragments the same way
831 * as other packets, omitting TCP and UDP port numbers (always setting
834 * - OFPC_FRAG_DROP: The switch should drop all IP fragments without
835 * passing them through the flow table.
837 * - OFPC_FRAG_REASM: The switch should reassemble IP fragments before
838 * passing packets through the flow table.
840 * - OFPC_FRAG_NX_MATCH (a Nicira extension): Similar to OFPC_FRAG_NORMAL,
841 * except that TCP and UDP port numbers should be included in fragments
844 * Implementations are not required to support every mode.
845 * OFPC_FRAG_NORMAL is the default mode when an ofproto is created.
847 * At the time of the call to ->set_frag_handling(), the current mode is
848 * available in 'ofproto->frag_handling'. ->set_frag_handling() returns
849 * true if the requested mode was set, false if it is not supported.
851 * Upon successful return, the caller changes 'ofproto->frag_handling' to
852 * reflect the new mode.
854 bool (*set_frag_handling)(struct ofproto *ofproto,
855 enum ofp_config_flags frag_handling);
857 /* Implements the OpenFlow OFPT_PACKET_OUT command. The datapath should
858 * execute the 'n_actions' in the 'actions' array on 'packet'.
860 * The caller retains ownership of 'packet', so ->packet_out() should not
863 * This function must validate that the 'n_actions' elements in 'actions'
864 * are well-formed OpenFlow actions that can be correctly implemented by
865 * the datapath. If not, then it should return an OpenFlow error code (as
866 * returned by ofp_mkerr()).
868 * 'flow' reflects the flow information for 'packet'. All of the
869 * information in 'flow' is extracted from 'packet', except for
870 * flow->in_port, which is taken from the OFPT_PACKET_OUT message.
871 * flow->tun_id and its register values are zeroed.
873 * 'packet' is not matched against the OpenFlow flow table, so its
874 * statistics should not be included in OpenFlow flow statistics.
876 * Returns 0 if successful, otherwise an OpenFlow error code (as returned
877 * by ofp_mkerr()). */
878 int (*packet_out)(struct ofproto *ofproto, struct ofpbuf *packet,
879 const struct flow *flow,
880 const union ofp_action *actions,
883 /* ## ------------------------- ## */
884 /* ## OFPP_NORMAL configuration ## */
885 /* ## ------------------------- ## */
887 /* Configures NetFlow on 'ofproto' according to the options in
888 * 'netflow_options', or turns off NetFlow if 'netflow_options' is NULL.
890 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
891 * NetFlow, as does a null pointer. */
892 int (*set_netflow)(struct ofproto *ofproto,
893 const struct netflow_options *netflow_options);
895 void (*get_netflow_ids)(const struct ofproto *ofproto,
896 uint8_t *engine_type, uint8_t *engine_id);
898 /* Configures sFlow on 'ofproto' according to the options in
899 * 'sflow_options', or turns off sFlow if 'sflow_options' is NULL.
901 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
902 * sFlow, as does a null pointer. */
903 int (*set_sflow)(struct ofproto *ofproto,
904 const struct ofproto_sflow_options *sflow_options);
906 /* Configures connectivity fault management on 'ofport'.
908 * If 'cfm_settings' is nonnull, configures CFM according to its members.
910 * If 'cfm_settings' is null, removes any connectivity fault management
911 * configuration from 'ofport'.
913 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
914 * support CFM, as does a null pointer. */
915 int (*set_cfm)(struct ofport *ofport, const struct cfm_settings *s);
917 /* Checks the fault status of CFM configured on 'ofport'. Returns 1 if CFM
918 * is faulted (generally indicating a connectivity problem), 0 if CFM is
919 * not faulted, or -1 if CFM is not enabled on 'port'
921 * This function may be a null pointer if the ofproto implementation does
922 * not support CFM. */
923 int (*get_cfm_fault)(const struct ofport *ofport);
925 /* Gets the MPIDs of the remote maintenance points broadcasting to
926 * 'ofport'. Populates 'rmps' with a provider owned array of MPIDs, and
927 * 'n_rmps' with the number of MPIDs in 'rmps'. Returns a number less than
928 * 0 if CFM is not enabled of 'ofport'.
930 * This function may be a null pointer if the ofproto implementation does
931 * not support CFM. */
932 int (*get_cfm_remote_mpids)(const struct ofport *ofport,
933 const uint64_t **rmps, size_t *n_rmps);
935 /* Configures spanning tree protocol (STP) on 'ofproto' using the
936 * settings defined in 's'.
938 * If 's' is nonnull, configures STP according to its members.
940 * If 's' is null, removes any STP configuration from 'ofproto'.
942 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
943 * support STP, as does a null pointer. */
944 int (*set_stp)(struct ofproto *ofproto,
945 const struct ofproto_stp_settings *s);
947 /* Retrieves state of spanning tree protocol (STP) on 'ofproto'.
949 * Stores STP state for 'ofproto' in 's'. If the 'enabled' member
950 * is false, the other member values are not meaningful.
952 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
953 * support STP, as does a null pointer. */
954 int (*get_stp_status)(struct ofproto *ofproto,
955 struct ofproto_stp_status *s);
957 /* Configures spanning tree protocol (STP) on 'ofport' using the
958 * settings defined in 's'.
960 * If 's' is nonnull, configures STP according to its members. The
961 * caller is responsible for assigning STP port numbers (using the
962 * 'port_num' member in the range of 1 through 255, inclusive) and
963 * ensuring there are no duplicates.
965 * If 's' is null, removes any STP configuration from 'ofport'.
967 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
968 * support STP, as does a null pointer. */
969 int (*set_stp_port)(struct ofport *ofport,
970 const struct ofproto_port_stp_settings *s);
972 /* Retrieves spanning tree protocol (STP) port status of 'ofport'.
974 * Stores STP state for 'ofport' in 's'. If the 'enabled' member is
975 * false, the other member values are not meaningful.
977 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
978 * support STP, as does a null pointer. */
979 int (*get_stp_port_status)(struct ofport *ofport,
980 struct ofproto_port_stp_status *s);
982 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
983 * 'queues' attached to 'ofport'. This data is not intended to be
984 * sufficient to implement QoS. Instead, providers may use this
985 * information to implement features which require knowledge of what queues
986 * exist on a port, and some basic information about them.
988 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
989 * support QoS, as does a null pointer. */
990 int (*set_queues)(struct ofport *ofport,
991 const struct ofproto_port_queue *queues, size_t n_qdscp);
993 /* If 's' is nonnull, this function registers a "bundle" associated with
994 * client data pointer 'aux' in 'ofproto'. A bundle is the same concept as
995 * a Port in OVSDB, that is, it consists of one or more "slave" devices
996 * (Interfaces, in OVSDB) along with VLAN and LACP configuration and, if
997 * there is more than one slave, a bonding configuration. If 'aux' is
998 * already registered then this function updates its configuration to 's'.
999 * Otherwise, this function registers a new bundle.
1001 * If 's' is NULL, this function unregisters the bundle registered on
1002 * 'ofproto' associated with client data pointer 'aux'. If no such bundle
1003 * has been registered, this has no effect.
1005 * This function affects only the behavior of the NXAST_AUTOPATH action and
1006 * output to the OFPP_NORMAL port. An implementation that does not support
1007 * it at all may set it to NULL or return EOPNOTSUPP. An implementation
1008 * that supports only a subset of the functionality should implement what
1009 * it can and return 0. */
1010 int (*bundle_set)(struct ofproto *ofproto, void *aux,
1011 const struct ofproto_bundle_settings *s);
1013 /* If 'port' is part of any bundle, removes it from that bundle. If the
1014 * bundle now has no ports, deletes the bundle. If the bundle now has only
1015 * one port, deconfigures the bundle's bonding configuration. */
1016 void (*bundle_remove)(struct ofport *ofport);
1018 /* If 's' is nonnull, this function registers a mirror associated with
1019 * client data pointer 'aux' in 'ofproto'. A mirror is the same concept as
1020 * a Mirror in OVSDB. If 'aux' is already registered then this function
1021 * updates its configuration to 's'. Otherwise, this function registers a
1024 * If 's' is NULL, this function unregisters the mirror registered on
1025 * 'ofproto' associated with client data pointer 'aux'. If no such mirror
1026 * has been registered, this has no effect.
1028 * An implementation that does not support mirroring at all may set
1029 * it to NULL or return EOPNOTSUPP. An implementation that supports
1030 * only a subset of the functionality should implement what it can
1032 int (*mirror_set)(struct ofproto *ofproto, void *aux,
1033 const struct ofproto_mirror_settings *s);
1035 /* Retrieves statistics from mirror associated with client data
1036 * pointer 'aux' in 'ofproto'. Stores packet and byte counts in
1037 * 'packets' and 'bytes', respectively. If a particular counter is
1038 * not supported, the appropriate argument is set to UINT64_MAX.
1040 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1041 * support retrieving mirror statistics. */
1042 int (*mirror_get_stats)(struct ofproto *ofproto, void *aux,
1043 uint64_t *packets, uint64_t *bytes);
1045 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs
1046 * on which all packets are flooded, instead of using MAC learning. If
1047 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1049 * This function affects only the behavior of the OFPP_NORMAL action. An
1050 * implementation that does not support it may set it to NULL or return
1052 int (*set_flood_vlans)(struct ofproto *ofproto,
1053 unsigned long *flood_vlans);
1055 /* Returns true if 'aux' is a registered bundle that is currently in use as
1056 * the output for a mirror. */
1057 bool (*is_mirror_output_bundle)(const struct ofproto *ofproto, void *aux);
1059 /* When the configuration option of forward_bpdu changes, this function
1060 * will be invoked. */
1061 void (*forward_bpdu_changed)(struct ofproto *ofproto);
1063 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
1065 * This is deprecated. It is only for compatibility with broken device drivers
1066 * in old versions of Linux that do not properly support VLANs when VLAN
1067 * devices are not used. When broken device drivers are no longer in
1068 * widespread use, we will delete these interfaces. */
1070 /* If 'realdev_ofp_port' is nonzero, then this function configures 'ofport'
1071 * as a VLAN splinter port for VLAN 'vid', associated with the real device
1072 * that has OpenFlow port number 'realdev_ofp_port'.
1074 * If 'realdev_ofp_port' is zero, then this function deconfigures 'ofport'
1075 * as a VLAN splinter port.
1077 * This function should be NULL if a an implementation does not support
1079 int (*set_realdev)(struct ofport *ofport,
1080 uint16_t realdev_ofp_port, int vid);
1083 extern const struct ofproto_class ofproto_dpif_class;
1085 int ofproto_class_register(const struct ofproto_class *);
1086 int ofproto_class_unregister(const struct ofproto_class *);
1088 /* ofproto_flow_mod() returns this value if the flow_mod could not be processed
1089 * because it overlaps with an ongoing flow table operation that has not yet
1090 * completed. The caller should retry the operation later.
1092 * ofproto.c also uses this value internally for additional (similar) purposes.
1094 * This particular value is a good choice because it is negative (so it won't
1095 * collide with any errno value or any value returned by ofp_mkerr()) and large
1096 * (so it won't accidentally collide with EOF or a negative errno value). */
1097 enum { OFPROTO_POSTPONE = -100000 };
1099 int ofproto_flow_mod(struct ofproto *, const struct ofputil_flow_mod *);
1100 void ofproto_add_flow(struct ofproto *, const struct cls_rule *,
1101 const union ofp_action *, size_t n_actions);
1102 bool ofproto_delete_flow(struct ofproto *, const struct cls_rule *);
1103 void ofproto_flush_flows(struct ofproto *);
1105 #endif /* ofproto/ofproto-provider.h */