2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
17 #ifndef OFPROTO_OFPROTO_PROVIDER_H
18 #define OFPROTO_OFPROTO_PROVIDER_H 1
20 /* Definitions for use within ofproto. */
22 #include "ofproto/ofproto.h"
24 #include "classifier.h"
27 #include "ofp-errors.h"
33 struct ofputil_flow_mod;
36 /* An OpenFlow switch.
38 * With few exceptions, ofproto implementations may look at these fields but
39 * should not modify them. */
41 struct hmap_node hmap_node; /* In global 'all_ofprotos' hmap. */
42 const struct ofproto_class *ofproto_class;
43 char *type; /* Datapath type. */
44 char *name; /* Datapath name. */
47 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
48 uint64_t datapath_id; /* Datapath ID. */
49 unsigned flow_eviction_threshold; /* Threshold at which to begin flow
50 * table eviction. Only affects the
51 * ofproto-dpif implementation */
52 bool forward_bpdu; /* Option to allow forwarding of BPDU frames
53 * when NORMAL action is invoked. */
54 char *mfr_desc; /* Manufacturer. */
55 char *hw_desc; /* Hardware. */
56 char *sw_desc; /* Software version. */
57 char *serial_desc; /* Serial number. */
58 char *dp_desc; /* Datapath description. */
59 enum ofp_config_flags frag_handling; /* One of OFPC_*. */
62 struct hmap ports; /* Contains "struct ofport"s. */
63 struct shash port_by_name;
66 struct oftable *tables;
69 /* OpenFlow connections. */
70 struct connmgr *connmgr;
72 /* Flow table operation tracking. */
73 int state; /* Internal state. */
74 struct list pending; /* List of "struct ofopgroup"s. */
75 unsigned int n_pending; /* list_size(&pending). */
76 struct hmap deletions; /* All OFOPERATION_DELETE "ofoperation"s. */
78 /* Flow table operation logging. */
79 int n_add, n_delete, n_modify; /* Number of unreported ops of each kind. */
80 long long int first_op, last_op; /* Range of times for unreported ops. */
81 long long int next_op_report; /* Time to report ops, or LLONG_MAX. */
82 long long int op_backoff; /* Earliest time to report ops again. */
84 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
86 * This is deprecated. It is only for compatibility with broken device
87 * drivers in old versions of Linux that do not properly support VLANs when
88 * VLAN devices are not used. When broken device drivers are no longer in
89 * widespread use, we will delete these interfaces. */
90 unsigned long int *vlan_bitmap; /* 4096-bit bitmap of in-use VLANs. */
91 bool vlans_changed; /* True if new VLANs are in use. */
92 int min_mtu; /* Current MTU of non-internal ports. */
95 void ofproto_init_tables(struct ofproto *, int n_tables);
97 struct ofproto *ofproto_lookup(const char *name);
98 struct ofport *ofproto_get_port(const struct ofproto *, uint16_t ofp_port);
100 /* An OpenFlow port within a "struct ofproto".
102 * With few exceptions, ofproto implementations may look at these fields but
103 * should not modify them. */
105 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
106 struct ofproto *ofproto; /* The ofproto that contains this port. */
107 struct netdev *netdev;
108 struct ofputil_phy_port pp;
109 uint16_t ofp_port; /* OpenFlow port number. */
110 unsigned int change_seq;
114 void ofproto_port_set_state(struct ofport *, enum ofputil_port_state);
117 OFTABLE_HIDDEN = 1 << 0, /* Hide from most OpenFlow operations. */
118 OFTABLE_READONLY = 1 << 1 /* Don't allow OpenFlow to change this table. */
121 /* A flow table within a "struct ofproto". */
123 enum oftable_flags flags;
124 struct classifier cls; /* Contains "struct rule"s. */
125 char *name; /* Table name exposed via OpenFlow, or NULL. */
127 /* Maximum number of flows or UINT_MAX if there is no limit besides any
128 * limit imposed by resource limitations. */
129 unsigned int max_flows;
131 /* These members determine the handling of an attempt to add a flow that
132 * would cause the table to have more than 'max_flows' flows.
134 * If 'eviction_fields' is NULL, overflows will be rejected with an error.
136 * If 'eviction_fields' is nonnull (regardless of whether n_eviction_fields
137 * is nonzero), an overflow will cause a flow to be removed. The flow to
138 * be removed is chosen to give fairness among groups distinguished by
139 * different values for the subfields within 'groups'. */
140 struct mf_subfield *eviction_fields;
141 size_t n_eviction_fields;
145 * When a flow is added that would cause the table to have more than
146 * 'max_flows' flows, and 'eviction_fields' is nonnull, these groups are
147 * used to decide which rule to evict: the rule is chosen from the eviction
148 * group that contains the greatest number of rules.*/
149 uint32_t eviction_group_id_basis;
150 struct hmap eviction_groups_by_id;
151 struct heap eviction_groups_by_size;
154 /* Assigns TABLE to each oftable, in turn, in OFPROTO.
156 * All parameters are evaluated multiple times. */
157 #define OFPROTO_FOR_EACH_TABLE(TABLE, OFPROTO) \
158 for ((TABLE) = (OFPROTO)->tables; \
159 (TABLE) < &(OFPROTO)->tables[(OFPROTO)->n_tables]; \
162 /* An OpenFlow flow within a "struct ofproto".
164 * With few exceptions, ofproto implementations may look at these fields but
165 * should not modify them. */
167 struct list ofproto_node; /* Owned by ofproto base code. */
168 struct ofproto *ofproto; /* The ofproto that contains this rule. */
169 struct cls_rule cr; /* In owning ofproto's classifier. */
171 struct ofoperation *pending; /* Operation now in progress, if nonnull. */
173 ovs_be64 flow_cookie; /* Controller-issued identifier. */
175 long long int created; /* Creation time. */
176 long long int modified; /* Time of last modification. */
177 long long int used; /* Last use; time created if never used. */
178 uint16_t hard_timeout; /* In seconds from ->modified. */
179 uint16_t idle_timeout; /* In seconds from ->used. */
180 uint8_t table_id; /* Index in ofproto's 'tables' array. */
181 bool send_flow_removed; /* Send a flow removed message? */
183 /* Eviction groups. */
184 bool evictable; /* If false, prevents eviction. */
185 struct heap_node evg_node; /* In eviction_group's "rules" heap. */
186 struct eviction_group *eviction_group; /* NULL if not in any group. */
188 struct ofpact *ofpacts; /* Sequence of "struct ofpacts". */
189 unsigned int ofpacts_len; /* Size of 'ofpacts', in bytes. */
192 enum nx_flow_monitor_flags monitor_flags;
193 uint64_t add_seqno; /* Sequence number when added. */
194 uint64_t modify_seqno; /* Sequence number when changed. */
197 static inline struct rule *
198 rule_from_cls_rule(const struct cls_rule *cls_rule)
200 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
203 void ofproto_rule_update_used(struct rule *, long long int used);
204 void ofproto_rule_expire(struct rule *, uint8_t reason);
205 void ofproto_rule_destroy(struct rule *);
207 bool ofproto_rule_has_out_port(const struct rule *, uint16_t out_port);
209 void ofoperation_complete(struct ofoperation *, enum ofperr);
210 struct rule *ofoperation_get_victim(struct ofoperation *);
212 bool ofoperation_has_out_port(const struct ofoperation *, uint16_t out_port);
214 bool ofproto_rule_is_hidden(const struct rule *);
216 /* ofproto class structure, to be defined by each ofproto implementation.
222 * These functions work primarily with three different kinds of data
225 * - "struct ofproto", which represents an OpenFlow switch.
227 * - "struct ofport", which represents a port within an ofproto.
229 * - "struct rule", which represents an OpenFlow flow within an ofproto.
231 * Each of these data structures contains all of the implementation-independent
232 * generic state for the respective concept, called the "base" state. None of
233 * them contains any extra space for ofproto implementations to use. Instead,
234 * each implementation is expected to declare its own data structure that
235 * contains an instance of the generic data structure plus additional
236 * implementation-specific members, called the "derived" state. The
237 * implementation can use casts or (preferably) the CONTAINER_OF macro to
238 * obtain access to derived state given only a pointer to the embedded generic
245 * Four stylized functions accompany each of these data structures:
247 * "alloc" "construct" "destruct" "dealloc"
248 * ------------ ---------------- --------------- --------------
249 * ofproto ->alloc ->construct ->destruct ->dealloc
250 * ofport ->port_alloc ->port_construct ->port_destruct ->port_dealloc
251 * rule ->rule_alloc ->rule_construct ->rule_destruct ->rule_dealloc
253 * Any instance of a given data structure goes through the following life
256 * 1. The client calls the "alloc" function to obtain raw memory. If "alloc"
257 * fails, skip all the other steps.
259 * 2. The client initializes all of the data structure's base state. If this
260 * fails, skip to step 7.
262 * 3. The client calls the "construct" function. The implementation
263 * initializes derived state. It may refer to the already-initialized
264 * base state. If "construct" fails, skip to step 6.
266 * 4. The data structure is now initialized and in use.
268 * 5. When the data structure is no longer needed, the client calls the
269 * "destruct" function. The implementation uninitializes derived state.
270 * The base state has not been uninitialized yet, so the implementation
271 * may still refer to it.
273 * 6. The client uninitializes all of the data structure's base state.
275 * 7. The client calls the "dealloc" to free the raw memory. The
276 * implementation must not refer to base or derived state in the data
277 * structure, because it has already been uninitialized.
279 * Each "alloc" function allocates and returns a new instance of the respective
280 * data structure. The "alloc" function is not given any information about the
281 * use of the new data structure, so it cannot perform much initialization.
282 * Its purpose is just to ensure that the new data structure has enough room
283 * for base and derived state. It may return a null pointer if memory is not
284 * available, in which case none of the other functions is called.
286 * Each "construct" function initializes derived state in its respective data
287 * structure. When "construct" is called, all of the base state has already
288 * been initialized, so the "construct" function may refer to it. The
289 * "construct" function is allowed to fail, in which case the client calls the
290 * "dealloc" function (but not the "destruct" function).
292 * Each "destruct" function uninitializes and frees derived state in its
293 * respective data structure. When "destruct" is called, the base state has
294 * not yet been uninitialized, so the "destruct" function may refer to it. The
295 * "destruct" function is not allowed to fail.
297 * Each "dealloc" function frees raw memory that was allocated by the the
298 * "alloc" function. The memory's base and derived members might not have ever
299 * been initialized (but if "construct" returned successfully, then it has been
300 * "destruct"ed already). The "dealloc" function is not allowed to fail.
306 * Most of these functions return 0 if they are successful or a positive error
307 * code on failure. Depending on the function, valid error codes are either
308 * errno values or OFPERR_* OpenFlow error codes.
310 * Most of these functions are expected to execute synchronously, that is, to
311 * block as necessary to obtain a result. Thus, these functions may return
312 * EAGAIN (or EWOULDBLOCK or EINPROGRESS) only where the function descriptions
313 * explicitly say those errors are a possibility. We may relax this
314 * requirement in the future if and when we encounter performance problems. */
315 struct ofproto_class {
316 /* ## ----------------- ## */
317 /* ## Factory Functions ## */
318 /* ## ----------------- ## */
320 /* Enumerates the types of all support ofproto types into 'types'. The
321 * caller has already initialized 'types' and other ofproto classes might
322 * already have added names to it. */
323 void (*enumerate_types)(struct sset *types);
325 /* Enumerates the names of all existing datapath of the specified 'type'
326 * into 'names' 'all_dps'. The caller has already initialized 'names' as
329 * 'type' is one of the types enumerated by ->enumerate_types().
331 * Returns 0 if successful, otherwise a positive errno value.
333 int (*enumerate_names)(const char *type, struct sset *names);
335 /* Deletes the datapath with the specified 'type' and 'name'. The caller
336 * should have closed any open ofproto with this 'type' and 'name'; this
337 * function is allowed to fail if that is not the case.
339 * 'type' is one of the types enumerated by ->enumerate_types().
340 * 'name' is one of the names enumerated by ->enumerate_names() for 'type'.
342 * Returns 0 if successful, otherwise a positive errno value.
344 int (*del)(const char *type, const char *name);
346 /* ## --------------------------- ## */
347 /* ## Top-Level ofproto Functions ## */
348 /* ## --------------------------- ## */
350 /* Life-cycle functions for an "ofproto" (see "Life Cycle" above).
356 * ->construct() should not modify any base members of the ofproto. The
357 * client will initialize the ofproto's 'ports' and 'tables' members after
358 * construction is complete.
360 * When ->construct() is called, the client does not yet know how many flow
361 * tables the datapath supports, so ofproto->n_tables will be 0 and
362 * ofproto->tables will be NULL. ->construct() should call
363 * ofproto_init_tables() to allocate and initialize ofproto->n_tables and
364 * ofproto->tables. Each flow table will be initially empty, so
365 * ->construct() should delete flows from the underlying datapath, if
366 * necessary, rather than populating the tables.
368 * Only one ofproto instance needs to be supported for any given datapath.
369 * If a datapath is already open as part of one "ofproto", then another
370 * attempt to "construct" the same datapath as part of another ofproto is
371 * allowed to fail with an error.
373 * ->construct() returns 0 if successful, otherwise a positive errno
380 * If 'ofproto' has any pending asynchronous operations, ->destruct()
381 * must complete all of them by calling ofoperation_complete().
383 * ->destruct() must also destroy all remaining rules in the ofproto's
384 * tables, by passing each remaining rule to ofproto_rule_destroy(). The
385 * client will destroy the flow tables themselves after ->destruct()
388 struct ofproto *(*alloc)(void);
389 int (*construct)(struct ofproto *ofproto);
390 void (*destruct)(struct ofproto *ofproto);
391 void (*dealloc)(struct ofproto *ofproto);
393 /* Performs any periodic activity required by 'ofproto'. It should:
395 * - Call connmgr_send_packet_in() for each received packet that missed
396 * in the OpenFlow flow table or that had a OFPP_CONTROLLER output
399 * - Call ofproto_rule_expire() for each OpenFlow flow that has reached
400 * its hard_timeout or idle_timeout, to expire the flow.
402 * (But rules that are part of a pending operation, e.g. rules for
403 * which ->pending is true, may not expire.)
405 * Returns 0 if successful, otherwise a positive errno value. */
406 int (*run)(struct ofproto *ofproto);
408 /* Performs periodic activity required by 'ofproto' that needs to be done
409 * with the least possible latency.
411 * This is run multiple times per main loop. An ofproto provider may
412 * implement it or not, according to whether it provides a performance
413 * boost for that ofproto implementation. */
414 int (*run_fast)(struct ofproto *ofproto);
416 /* Causes the poll loop to wake up when 'ofproto''s 'run' function needs to
417 * be called, e.g. by calling the timer or fd waiting functions in
419 void (*wait)(struct ofproto *ofproto);
421 /* Adds some memory usage statistics for the implementation of 'ofproto'
422 * into 'usage', for use with memory_report().
424 * This function is optional. */
425 void (*get_memory_usage)(const struct ofproto *ofproto,
426 struct simap *usage);
428 /* Every "struct rule" in 'ofproto' is about to be deleted, one by one.
429 * This function may prepare for that, for example by clearing state in
430 * advance. It should *not* actually delete any "struct rule"s from
431 * 'ofproto', only prepare for it.
433 * This function is optional; it's really just for optimization in case
434 * it's cheaper to delete all the flows from your hardware in a single pass
435 * than to do it one by one. */
436 void (*flush)(struct ofproto *ofproto);
438 /* Helper for the OpenFlow OFPT_FEATURES_REQUEST request.
440 * The implementation should store true in '*arp_match_ip' if the switch
441 * supports matching IP addresses inside ARP requests and replies, false
444 * The implementation should store in '*actions' a bitmap of the supported
445 * OpenFlow actions. Vendor actions are not included in '*actions'. */
446 void (*get_features)(struct ofproto *ofproto,
448 enum ofputil_action_bitmap *actions);
450 /* Helper for the OpenFlow OFPST_TABLE statistics request.
452 * The 'ots' array contains 'ofproto->n_tables' elements. Each element is
455 * - 'table_id' to the array index.
457 * - 'name' to "table#" where # is the table ID.
459 * - 'wildcards' to OFPFW10_ALL.
461 * - 'max_entries' to 1,000,000.
463 * - 'active_count' to the classifier_count() for the table.
465 * - 'lookup_count' and 'matched_count' to 0.
467 * The implementation should update any members in each element for which
468 * it has better values:
470 * - 'name' to a more meaningful name.
472 * - 'wildcards' to the set of wildcards actually supported by the table
473 * (if it doesn't support all OpenFlow wildcards).
475 * - 'max_entries' to the maximum number of flows actually supported by
478 * - 'lookup_count' to the number of packets looked up in this flow table
481 * - 'matched_count' to the number of packets looked up in this flow
482 * table so far that matched one of the flow entries.
484 * Keep in mind that all of the members of struct ofp10_table_stats are in
485 * network byte order.
487 void (*get_tables)(struct ofproto *ofproto, struct ofp10_table_stats *ots);
489 /* ## ---------------- ## */
490 /* ## ofport Functions ## */
491 /* ## ---------------- ## */
493 /* Life-cycle functions for a "struct ofport" (see "Life Cycle" above).
495 * ->port_construct() should not modify any base members of the ofport.
497 * ofports are managed by the base ofproto code. The ofproto
498 * implementation should only create and destroy them in response to calls
499 * to these functions. The base ofproto code will create and destroy
500 * ofports in the following situations:
502 * - Just after the ->construct() function is called, the base ofproto
503 * iterates over all of the implementation's ports, using
504 * ->port_dump_start() and related functions, and constructs an ofport
505 * for each dumped port.
507 * - If ->port_poll() reports that a specific port has changed, then the
508 * base ofproto will query that port with ->port_query_by_name() and
509 * construct or destruct ofports as necessary to reflect the updated
512 * - If ->port_poll() returns ENOBUFS to report an unspecified port set
513 * change, then the base ofproto will iterate over all of the
514 * implementation's ports, in the same way as at ofproto
515 * initialization, and construct and destruct ofports to reflect all of
518 * ->port_construct() returns 0 if successful, otherwise a positive errno
521 struct ofport *(*port_alloc)(void);
522 int (*port_construct)(struct ofport *ofport);
523 void (*port_destruct)(struct ofport *ofport);
524 void (*port_dealloc)(struct ofport *ofport);
526 /* Called after 'ofport->netdev' is replaced by a new netdev object. If
527 * the ofproto implementation uses the ofport's netdev internally, then it
528 * should switch to using the new one. The old one has been closed.
530 * An ofproto implementation that doesn't need to do anything in this
531 * function may use a null pointer. */
532 void (*port_modified)(struct ofport *ofport);
534 /* Called after an OpenFlow request changes a port's configuration.
535 * 'ofport->pp.config' contains the new configuration. 'old_config'
536 * contains the previous configuration.
538 * The caller implements OFPUTIL_PC_PORT_DOWN using netdev functions to
539 * turn NETDEV_UP on and off, so this function doesn't have to do anything
540 * for that bit (and it won't be called if that is the only bit that
542 void (*port_reconfigured)(struct ofport *ofport,
543 enum ofputil_port_config old_config);
545 /* Looks up a port named 'devname' in 'ofproto'. On success, initializes
546 * '*port' appropriately.
548 * The caller owns the data in 'port' and must free it with
549 * ofproto_port_destroy() when it is no longer needed. */
550 int (*port_query_by_name)(const struct ofproto *ofproto,
551 const char *devname, struct ofproto_port *port);
553 /* Attempts to add 'netdev' as a port on 'ofproto'. Returns 0 if
554 * successful, otherwise a positive errno value. If successful, sets
555 * '*ofp_portp' to the new port's port number.
557 * It doesn't matter whether the new port will be returned by a later call
558 * to ->port_poll(); the implementation may do whatever is more
560 int (*port_add)(struct ofproto *ofproto, struct netdev *netdev,
561 uint16_t *ofp_portp);
563 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'. Returns
564 * 0 if successful, otherwise a positive errno value.
566 * It doesn't matter whether the new port will be returned by a later call
567 * to ->port_poll(); the implementation may do whatever is more
569 int (*port_del)(struct ofproto *ofproto, uint16_t ofp_port);
572 int (*port_get_stats)(const struct ofport *port,
573 struct netdev_stats *stats);
575 /* Port iteration functions.
577 * The client might not be entirely in control of the ports within an
578 * ofproto. Some hardware implementations, for example, might have a fixed
579 * set of ports in a datapath, and the Linux datapath allows the system
580 * administrator to externally add and remove ports with ovs-dpctl. For
581 * this reason, the client needs a way to iterate through all the ports
582 * that are actually in a datapath. These functions provide that
585 * The 'state' pointer provides the implementation a place to
586 * keep track of its position. Its format is opaque to the caller.
588 * The ofproto provider retains ownership of the data that it stores into
589 * ->port_dump_next()'s 'port' argument. The data must remain valid until
590 * at least the next call to ->port_dump_next() or ->port_dump_done() for
591 * 'state'. The caller will not modify or free it.
596 * ->port_dump_start() attempts to begin dumping the ports in 'ofproto'.
597 * On success, it should return 0 and initialize '*statep' with any data
598 * needed for iteration. On failure, returns a positive errno value, and
599 * the client will not call ->port_dump_next() or ->port_dump_done().
601 * ->port_dump_next() attempts to retrieve another port from 'ofproto' for
602 * 'state'. If there is another port, it should store the port's
603 * information into 'port' and return 0. It should return EOF if all ports
604 * have already been iterated. Otherwise, on error, it should return a
605 * positive errno value. This function will not be called again once it
606 * returns nonzero once for a given iteration (but the 'port_dump_done'
607 * function will be called afterward).
609 * ->port_dump_done() allows the implementation to release resources used
610 * for iteration. The caller might decide to stop iteration in the middle
611 * by calling this function before ->port_dump_next() returns nonzero.
619 * error = ofproto->ofproto_class->port_dump_start(ofproto, &state);
622 * struct ofproto_port port;
624 * error = ofproto->ofproto_class->port_dump_next(
625 * ofproto, state, &port);
629 * // Do something with 'port' here (without modifying or freeing
630 * // any of its data).
632 * ofproto->ofproto_class->port_dump_done(ofproto, state);
634 * // 'error' is now EOF (success) or a positive errno value (failure).
636 int (*port_dump_start)(const struct ofproto *ofproto, void **statep);
637 int (*port_dump_next)(const struct ofproto *ofproto, void *state,
638 struct ofproto_port *port);
639 int (*port_dump_done)(const struct ofproto *ofproto, void *state);
641 /* Polls for changes in the set of ports in 'ofproto'. If the set of ports
642 * in 'ofproto' has changed, then this function should do one of the
645 * - Preferably: store the name of the device that was added to or deleted
646 * from 'ofproto' in '*devnamep' and return 0. The caller is responsible
647 * for freeing '*devnamep' (with free()) when it no longer needs it.
649 * - Alternatively: return ENOBUFS, without indicating the device that was
652 * Occasional 'false positives', in which the function returns 0 while
653 * indicating a device that was not actually added or deleted or returns
654 * ENOBUFS without any change, are acceptable.
656 * The purpose of 'port_poll' is to let 'ofproto' know about changes made
657 * externally to the 'ofproto' object, e.g. by a system administrator via
658 * ovs-dpctl. Therefore, it's OK, and even preferable, for port_poll() to
659 * not report changes made through calls to 'port_add' or 'port_del' on the
660 * same 'ofproto' object. (But it's OK for it to report them too, just
661 * slightly less efficient.)
663 * If the set of ports in 'ofproto' has not changed, returns EAGAIN. May
664 * also return other positive errno values to indicate that something has
667 * If the set of ports in a datapath is fixed, or if the only way that the
668 * set of ports in a datapath can change is through ->port_add() and
669 * ->port_del(), then this function may be a null pointer.
671 int (*port_poll)(const struct ofproto *ofproto, char **devnamep);
673 /* Arranges for the poll loop to wake up when ->port_poll() will return a
674 * value other than EAGAIN.
676 * If the set of ports in a datapath is fixed, or if the only way that the
677 * set of ports in a datapath can change is through ->port_add() and
678 * ->port_del(), or if the poll loop will always wake up anyway when
679 * ->port_poll() will return a value other than EAGAIN, then this function
680 * may be a null pointer.
682 void (*port_poll_wait)(const struct ofproto *ofproto);
684 /* Checks the status of LACP negotiation for 'port'. Returns 1 if LACP
685 * partner information for 'port' is up-to-date, 0 if LACP partner
686 * information is not current (generally indicating a connectivity
687 * problem), or -1 if LACP is not enabled on 'port'.
689 * This function may be a null pointer if the ofproto implementation does
690 * not support LACP. */
691 int (*port_is_lacp_current)(const struct ofport *port);
693 /* ## ----------------------- ## */
694 /* ## OpenFlow Rule Functions ## */
695 /* ## ----------------------- ## */
699 /* Chooses an appropriate table for 'cls_rule' within 'ofproto'. On
700 * success, stores the table ID into '*table_idp' and returns 0. On
701 * failure, returns an OpenFlow error code.
703 * The choice of table should be a function of 'cls_rule' and 'ofproto''s
704 * datapath capabilities. It should not depend on the flows already in
705 * 'ofproto''s flow tables. Failure implies that an OpenFlow rule with
706 * 'cls_rule' as its matching condition can never be inserted into
707 * 'ofproto', even starting from an empty flow table.
709 * If multiple tables are candidates for inserting the flow, the function
710 * should choose one arbitrarily (but deterministically).
712 * If this function is NULL then table 0 is always chosen. */
713 enum ofperr (*rule_choose_table)(const struct ofproto *ofproto,
714 const struct cls_rule *cls_rule,
717 /* Life-cycle functions for a "struct rule" (see "Life Cycle" above).
720 * Asynchronous Operation Support
721 * ==============================
723 * The life-cycle operations on rules can operate asynchronously, meaning
724 * that ->rule_construct() and ->rule_destruct() only need to initiate
725 * their respective operations and do not need to wait for them to complete
726 * before they return. ->rule_modify_actions() also operates
729 * An ofproto implementation reports the success or failure of an
730 * asynchronous operation on a rule using the rule's 'pending' member,
731 * which points to a opaque "struct ofoperation" that represents the
732 * ongoing opreation. When the operation completes, the ofproto
733 * implementation calls ofoperation_complete(), passing the ofoperation and
734 * an error indication.
736 * Only the following contexts may call ofoperation_complete():
738 * - The function called to initiate the operation,
739 * e.g. ->rule_construct() or ->rule_destruct(). This is the best
740 * choice if the operation completes quickly.
742 * - The implementation's ->run() function.
744 * - The implementation's ->destruct() function.
746 * The ofproto base code updates the flow table optimistically, assuming
747 * that the operation will probably succeed:
749 * - ofproto adds or replaces the rule in the flow table before calling
750 * ->rule_construct().
752 * - ofproto updates the rule's actions before calling
753 * ->rule_modify_actions().
755 * - ofproto removes the rule before calling ->rule_destruct().
757 * With one exception, when an asynchronous operation completes with an
758 * error, ofoperation_complete() backs out the already applied changes:
760 * - If adding or replacing a rule in the flow table fails, ofproto
761 * removes the new rule or restores the original rule.
763 * - If modifying a rule's actions fails, ofproto restores the original
766 * - Removing a rule is not allowed to fail. It must always succeed.
768 * The ofproto base code serializes operations: if any operation is in
769 * progress on a given rule, ofproto postpones initiating any new operation
770 * on that rule until the pending operation completes. Therefore, every
771 * operation must eventually complete through a call to
772 * ofoperation_complete() to avoid delaying new operations indefinitely
773 * (including any OpenFlow request that affects the rule in question, even
774 * just to query its statistics).
780 * When ->rule_construct() is called, the caller has already inserted
781 * 'rule' into 'rule->ofproto''s flow table numbered 'rule->table_id'.
782 * There are two cases:
784 * - 'rule' is a new rule in its flow table. In this case,
785 * ofoperation_get_victim(rule) returns NULL.
787 * - 'rule' is replacing an existing rule in its flow table that had the
788 * same matching criteria and priority. In this case,
789 * ofoperation_get_victim(rule) returns the rule being replaced (the
792 * ->rule_construct() should set the following in motion:
794 * - Validate that the matching rule in 'rule->cr' is supported by the
795 * datapath. For example, if the rule's table does not support
796 * registers, then it is an error if 'rule->cr' does not wildcard all
799 * - Validate that 'rule->ofpacts' is a sequence of well-formed actions
800 * that the datapath can correctly implement. If your ofproto
801 * implementation only implements a subset of the actions that Open
802 * vSwitch understands, then you should implement your own action
805 * - If the rule is valid, update the datapath flow table, adding the new
806 * rule or replacing the existing one.
808 * - If 'rule' is replacing an existing rule, uninitialize any derived
809 * state for the victim rule, as in step 5 in the "Life Cycle"
812 * (On failure, the ofproto code will roll back the insertion from the flow
813 * table, either removing 'rule' or replacing it by the victim rule if
816 * ->rule_construct() must act in one of the following ways:
818 * - If it succeeds, it must call ofoperation_complete() and return 0.
820 * - If it fails, it must act in one of the following ways:
822 * * Call ofoperation_complete() and return 0.
824 * * Return an OpenFlow error code. (Do not call
825 * ofoperation_complete() in this case.)
827 * Either way, ->rule_destruct() will not be called for 'rule', but
828 * ->rule_dealloc() will be.
830 * - If the operation is only partially complete, then it must return 0.
831 * Later, when the operation is complete, the ->run() or ->destruct()
832 * function must call ofoperation_complete() to report success or
835 * ->rule_construct() should not modify any base members of struct rule.
841 * When ->rule_destruct() is called, the caller has already removed 'rule'
842 * from 'rule->ofproto''s flow table. ->rule_destruct() should set in
843 * motion removing 'rule' from the datapath flow table. If removal
844 * completes synchronously, it should call ofoperation_complete().
845 * Otherwise, the ->run() or ->destruct() function must later call
846 * ofoperation_complete() after the operation completes.
848 * Rule destruction must not fail. */
849 struct rule *(*rule_alloc)(void);
850 enum ofperr (*rule_construct)(struct rule *rule);
851 void (*rule_destruct)(struct rule *rule);
852 void (*rule_dealloc)(struct rule *rule);
854 /* Obtains statistics for 'rule', storing the number of packets that have
855 * matched it in '*packet_count' and the number of bytes in those packets
856 * in '*byte_count'. UINT64_MAX indicates that the packet count or byte
857 * count is unknown. */
858 void (*rule_get_stats)(struct rule *rule, uint64_t *packet_count,
859 uint64_t *byte_count);
861 /* Applies the actions in 'rule' to 'packet'. (This implements sending
862 * buffered packets for OpenFlow OFPT_FLOW_MOD commands.)
864 * Takes ownership of 'packet' (so it should eventually free it, with
867 * 'flow' reflects the flow information for 'packet'. All of the
868 * information in 'flow' is extracted from 'packet', except for
869 * flow->tun_id and flow->in_port, which are assigned the correct values
870 * for the incoming packet. The register values are zeroed. 'packet''s
871 * header pointers (e.g. packet->l3) are appropriately initialized.
873 * The implementation should add the statistics for 'packet' into 'rule'.
875 * Returns 0 if successful, otherwise an OpenFlow error code. */
876 enum ofperr (*rule_execute)(struct rule *rule, const struct flow *flow,
877 struct ofpbuf *packet);
879 /* When ->rule_modify_actions() is called, the caller has already replaced
880 * the OpenFlow actions in 'rule' by a new set. (The original actions are
881 * in rule->pending->actions.)
883 * ->rule_modify_actions() should set the following in motion:
885 * - Validate that the actions now in 'rule' are well-formed OpenFlow
886 * actions that the datapath can correctly implement.
888 * - Update the datapath flow table with the new actions.
890 * If the operation synchronously completes, ->rule_modify_actions() may
891 * call ofoperation_complete() before it returns. Otherwise, ->run()
892 * should call ofoperation_complete() later, after the operation does
895 * If the operation fails, then the base ofproto code will restore the
896 * original 'actions' and 'n_actions' of 'rule'.
898 * ->rule_modify_actions() should not modify any base members of struct
900 void (*rule_modify_actions)(struct rule *rule);
902 /* Changes the OpenFlow IP fragment handling policy to 'frag_handling',
903 * which takes one of the following values, with the corresponding
906 * - OFPC_FRAG_NORMAL: The switch should treat IP fragments the same way
907 * as other packets, omitting TCP and UDP port numbers (always setting
910 * - OFPC_FRAG_DROP: The switch should drop all IP fragments without
911 * passing them through the flow table.
913 * - OFPC_FRAG_REASM: The switch should reassemble IP fragments before
914 * passing packets through the flow table.
916 * - OFPC_FRAG_NX_MATCH (a Nicira extension): Similar to OFPC_FRAG_NORMAL,
917 * except that TCP and UDP port numbers should be included in fragments
920 * Implementations are not required to support every mode.
921 * OFPC_FRAG_NORMAL is the default mode when an ofproto is created.
923 * At the time of the call to ->set_frag_handling(), the current mode is
924 * available in 'ofproto->frag_handling'. ->set_frag_handling() returns
925 * true if the requested mode was set, false if it is not supported.
927 * Upon successful return, the caller changes 'ofproto->frag_handling' to
928 * reflect the new mode.
930 bool (*set_frag_handling)(struct ofproto *ofproto,
931 enum ofp_config_flags frag_handling);
933 /* Implements the OpenFlow OFPT_PACKET_OUT command. The datapath should
934 * execute the 'ofpacts_len' bytes of "struct ofpacts" in 'ofpacts'.
936 * The caller retains ownership of 'packet' and of 'ofpacts', so
937 * ->packet_out() should not modify or free them.
939 * This function must validate that it can implement 'ofpacts'. If not,
940 * then it should return an OpenFlow error code.
942 * 'flow' reflects the flow information for 'packet'. All of the
943 * information in 'flow' is extracted from 'packet', except for
944 * flow->in_port (see below). flow->tun_id and its register values are
947 * flow->in_port comes from the OpenFlow OFPT_PACKET_OUT message. The
948 * implementation should reject invalid flow->in_port values by returning
949 * OFPERR_NXBRC_BAD_IN_PORT. For consistency, the implementation should
950 * consider valid for flow->in_port any value that could possibly be seen
951 * in a packet that it passes to connmgr_send_packet_in(). Ideally, even
952 * an implementation that never generates packet-ins (e.g. due to hardware
953 * limitations) should still allow flow->in_port values for every possible
954 * physical port and OFPP_LOCAL. The only virtual ports (those above
955 * OFPP_MAX) that the caller will ever pass in as flow->in_port, other than
956 * OFPP_LOCAL, are OFPP_NONE and OFPP_CONTROLLER. The implementation
957 * should allow both of these, treating each of them as packets generated
958 * by the controller as opposed to packets originating from some switch
961 * (Ordinarily the only effect of flow->in_port is on output actions that
962 * involve the input port, such as actions that output to OFPP_IN_PORT,
963 * OFPP_FLOOD, or OFPP_ALL. flow->in_port can also affect Nicira extension
964 * "resubmit" actions.)
966 * 'packet' is not matched against the OpenFlow flow table, so its
967 * statistics should not be included in OpenFlow flow statistics.
969 * Returns 0 if successful, otherwise an OpenFlow error code. */
970 enum ofperr (*packet_out)(struct ofproto *ofproto, struct ofpbuf *packet,
971 const struct flow *flow,
972 const struct ofpact *ofpacts,
975 /* ## ------------------------- ## */
976 /* ## OFPP_NORMAL configuration ## */
977 /* ## ------------------------- ## */
979 /* Configures NetFlow on 'ofproto' according to the options in
980 * 'netflow_options', or turns off NetFlow if 'netflow_options' is NULL.
982 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
983 * NetFlow, as does a null pointer. */
984 int (*set_netflow)(struct ofproto *ofproto,
985 const struct netflow_options *netflow_options);
987 void (*get_netflow_ids)(const struct ofproto *ofproto,
988 uint8_t *engine_type, uint8_t *engine_id);
990 /* Configures sFlow on 'ofproto' according to the options in
991 * 'sflow_options', or turns off sFlow if 'sflow_options' is NULL.
993 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
994 * sFlow, as does a null pointer. */
995 int (*set_sflow)(struct ofproto *ofproto,
996 const struct ofproto_sflow_options *sflow_options);
998 /* Configures connectivity fault management on 'ofport'.
1000 * If 'cfm_settings' is nonnull, configures CFM according to its members.
1002 * If 'cfm_settings' is null, removes any connectivity fault management
1003 * configuration from 'ofport'.
1005 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1006 * support CFM, as does a null pointer. */
1007 int (*set_cfm)(struct ofport *ofport, const struct cfm_settings *s);
1009 /* Checks the fault status of CFM configured on 'ofport'. Returns a
1010 * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
1011 * indicating a connectivity problem). Returns zero if CFM is not faulted,
1012 * and -1 if CFM is not enabled on 'port'.
1014 * This function may be a null pointer if the ofproto implementation does
1015 * not support CFM. */
1016 int (*get_cfm_fault)(const struct ofport *ofport);
1018 /* Gets the MPIDs of the remote maintenance points broadcasting to
1019 * 'ofport'. Populates 'rmps' with a provider owned array of MPIDs, and
1020 * 'n_rmps' with the number of MPIDs in 'rmps'. Returns a number less than
1021 * 0 if CFM is not enabled of 'ofport'.
1023 * This function may be a null pointer if the ofproto implementation does
1024 * not support CFM. */
1025 int (*get_cfm_remote_mpids)(const struct ofport *ofport,
1026 const uint64_t **rmps, size_t *n_rmps);
1028 /* Checks the health of CFM configured on 'ofport'. Returns an integer
1029 * to indicate the health percentage of the 'ofport' which is an average of
1030 * the health of all the remote_mps. Returns an integer between 0 and 100
1031 * where 0 means that the 'ofport' is very unhealthy and 100 means the
1032 * 'ofport' is perfectly healthy. Returns -1 if CFM is not enabled on
1033 * 'port' or if the number of remote_mpids is > 1.
1035 * This function may be a null pointer if the ofproto implementation does
1036 * not support CFM. */
1037 int (*get_cfm_health)(const struct ofport *ofport);
1039 /* Configures spanning tree protocol (STP) on 'ofproto' using the
1040 * settings defined in 's'.
1042 * If 's' is nonnull, configures STP according to its members.
1044 * If 's' is null, removes any STP configuration from 'ofproto'.
1046 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1047 * support STP, as does a null pointer. */
1048 int (*set_stp)(struct ofproto *ofproto,
1049 const struct ofproto_stp_settings *s);
1051 /* Retrieves state of spanning tree protocol (STP) on 'ofproto'.
1053 * Stores STP state for 'ofproto' in 's'. If the 'enabled' member
1054 * is false, the other member values are not meaningful.
1056 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1057 * support STP, as does a null pointer. */
1058 int (*get_stp_status)(struct ofproto *ofproto,
1059 struct ofproto_stp_status *s);
1061 /* Configures spanning tree protocol (STP) on 'ofport' using the
1062 * settings defined in 's'.
1064 * If 's' is nonnull, configures STP according to its members. The
1065 * caller is responsible for assigning STP port numbers (using the
1066 * 'port_num' member in the range of 1 through 255, inclusive) and
1067 * ensuring there are no duplicates.
1069 * If 's' is null, removes any STP configuration from 'ofport'.
1071 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1072 * support STP, as does a null pointer. */
1073 int (*set_stp_port)(struct ofport *ofport,
1074 const struct ofproto_port_stp_settings *s);
1076 /* Retrieves spanning tree protocol (STP) port status of 'ofport'.
1078 * Stores STP state for 'ofport' in 's'. If the 'enabled' member is
1079 * false, the other member values are not meaningful.
1081 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1082 * support STP, as does a null pointer. */
1083 int (*get_stp_port_status)(struct ofport *ofport,
1084 struct ofproto_port_stp_status *s);
1086 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
1087 * 'queues' attached to 'ofport'. This data is not intended to be
1088 * sufficient to implement QoS. Instead, providers may use this
1089 * information to implement features which require knowledge of what queues
1090 * exist on a port, and some basic information about them.
1092 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1093 * support QoS, as does a null pointer. */
1094 int (*set_queues)(struct ofport *ofport,
1095 const struct ofproto_port_queue *queues, size_t n_qdscp);
1097 /* If 's' is nonnull, this function registers a "bundle" associated with
1098 * client data pointer 'aux' in 'ofproto'. A bundle is the same concept as
1099 * a Port in OVSDB, that is, it consists of one or more "slave" devices
1100 * (Interfaces, in OVSDB) along with VLAN and LACP configuration and, if
1101 * there is more than one slave, a bonding configuration. If 'aux' is
1102 * already registered then this function updates its configuration to 's'.
1103 * Otherwise, this function registers a new bundle.
1105 * If 's' is NULL, this function unregisters the bundle registered on
1106 * 'ofproto' associated with client data pointer 'aux'. If no such bundle
1107 * has been registered, this has no effect.
1109 * This function affects only the behavior of the NXAST_AUTOPATH action and
1110 * output to the OFPP_NORMAL port. An implementation that does not support
1111 * it at all may set it to NULL or return EOPNOTSUPP. An implementation
1112 * that supports only a subset of the functionality should implement what
1113 * it can and return 0. */
1114 int (*bundle_set)(struct ofproto *ofproto, void *aux,
1115 const struct ofproto_bundle_settings *s);
1117 /* If 'port' is part of any bundle, removes it from that bundle. If the
1118 * bundle now has no ports, deletes the bundle. If the bundle now has only
1119 * one port, deconfigures the bundle's bonding configuration. */
1120 void (*bundle_remove)(struct ofport *ofport);
1122 /* If 's' is nonnull, this function registers a mirror associated with
1123 * client data pointer 'aux' in 'ofproto'. A mirror is the same concept as
1124 * a Mirror in OVSDB. If 'aux' is already registered then this function
1125 * updates its configuration to 's'. Otherwise, this function registers a
1128 * If 's' is NULL, this function unregisters the mirror registered on
1129 * 'ofproto' associated with client data pointer 'aux'. If no such mirror
1130 * has been registered, this has no effect.
1132 * An implementation that does not support mirroring at all may set
1133 * it to NULL or return EOPNOTSUPP. An implementation that supports
1134 * only a subset of the functionality should implement what it can
1136 int (*mirror_set)(struct ofproto *ofproto, void *aux,
1137 const struct ofproto_mirror_settings *s);
1139 /* Retrieves statistics from mirror associated with client data
1140 * pointer 'aux' in 'ofproto'. Stores packet and byte counts in
1141 * 'packets' and 'bytes', respectively. If a particular counter is
1142 * not supported, the appropriate argument is set to UINT64_MAX.
1144 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1145 * support retrieving mirror statistics. */
1146 int (*mirror_get_stats)(struct ofproto *ofproto, void *aux,
1147 uint64_t *packets, uint64_t *bytes);
1149 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs
1150 * on which all packets are flooded, instead of using MAC learning. If
1151 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1153 * This function affects only the behavior of the OFPP_NORMAL action. An
1154 * implementation that does not support it may set it to NULL or return
1156 int (*set_flood_vlans)(struct ofproto *ofproto,
1157 unsigned long *flood_vlans);
1159 /* Returns true if 'aux' is a registered bundle that is currently in use as
1160 * the output for a mirror. */
1161 bool (*is_mirror_output_bundle)(const struct ofproto *ofproto, void *aux);
1163 /* When the configuration option of forward_bpdu changes, this function
1164 * will be invoked. */
1165 void (*forward_bpdu_changed)(struct ofproto *ofproto);
1167 /* Sets the MAC aging timeout for the OFPP_NORMAL action to 'idle_time',
1169 void (*set_mac_idle_time)(struct ofproto *ofproto, unsigned int idle_time);
1171 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
1173 * This is deprecated. It is only for compatibility with broken device drivers
1174 * in old versions of Linux that do not properly support VLANs when VLAN
1175 * devices are not used. When broken device drivers are no longer in
1176 * widespread use, we will delete these interfaces. */
1178 /* If 'realdev_ofp_port' is nonzero, then this function configures 'ofport'
1179 * as a VLAN splinter port for VLAN 'vid', associated with the real device
1180 * that has OpenFlow port number 'realdev_ofp_port'.
1182 * If 'realdev_ofp_port' is zero, then this function deconfigures 'ofport'
1183 * as a VLAN splinter port.
1185 * This function should be NULL if a an implementation does not support
1187 int (*set_realdev)(struct ofport *ofport,
1188 uint16_t realdev_ofp_port, int vid);
1191 extern const struct ofproto_class ofproto_dpif_class;
1193 int ofproto_class_register(const struct ofproto_class *);
1194 int ofproto_class_unregister(const struct ofproto_class *);
1196 /* ofproto_flow_mod() returns this value if the flow_mod could not be processed
1197 * because it overlaps with an ongoing flow table operation that has not yet
1198 * completed. The caller should retry the operation later.
1200 * ofproto.c also uses this value internally for additional (similar) purposes.
1202 * This particular value is a good choice because it is large, so that it does
1203 * not collide with any errno value, but not large enough to collide with an
1204 * OFPERR_* value. */
1205 enum { OFPROTO_POSTPONE = 1 << 16 };
1206 BUILD_ASSERT_DECL(OFPROTO_POSTPONE < OFPERR_OFS);
1208 int ofproto_flow_mod(struct ofproto *, const struct ofputil_flow_mod *);
1209 void ofproto_add_flow(struct ofproto *, const struct cls_rule *,
1210 const struct ofpact *ofpacts, size_t ofpacts_len);
1211 bool ofproto_delete_flow(struct ofproto *, const struct cls_rule *);
1212 void ofproto_flush_flows(struct ofproto *);
1214 #endif /* ofproto/ofproto-provider.h */