2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
17 #ifndef OFPROTO_OFPROTO_PROVIDER_H
18 #define OFPROTO_OFPROTO_PROVIDER_H 1
20 /* Definitions for use within ofproto. */
22 #include "ofproto/ofproto.h"
24 #include "classifier.h"
27 #include "ofp-errors.h"
31 struct ofputil_flow_mod;
33 /* An OpenFlow switch.
35 * With few exceptions, ofproto implementations may look at these fields but
36 * should not modify them. */
38 const struct ofproto_class *ofproto_class;
39 char *type; /* Datapath type. */
40 char *name; /* Datapath name. */
41 struct hmap_node hmap_node; /* In global 'all_ofprotos' hmap. */
44 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
45 uint64_t datapath_id; /* Datapath ID. */
46 unsigned flow_eviction_threshold; /* Threshold at which to begin flow
47 * table eviction. Only affects the
48 * ofproto-dpif implementation */
49 bool forward_bpdu; /* Option to allow forwarding of BPDU frames
50 * when NORMAL action is invoked. */
51 char *mfr_desc; /* Manufacturer. */
52 char *hw_desc; /* Hardware. */
53 char *sw_desc; /* Software version. */
54 char *serial_desc; /* Serial number. */
55 char *dp_desc; /* Datapath description. */
56 enum ofp_config_flags frag_handling; /* One of OFPC_*. */
59 struct hmap ports; /* Contains "struct ofport"s. */
60 struct shash port_by_name;
63 struct oftable *tables;
66 /* OpenFlow connections. */
67 struct connmgr *connmgr;
69 /* Flow table operation tracking. */
70 int state; /* Internal state. */
71 struct list pending; /* List of "struct ofopgroup"s. */
72 unsigned int n_pending; /* list_size(&pending). */
73 struct hmap deletions; /* All OFOPERATION_DELETE "ofoperation"s. */
75 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
77 * This is deprecated. It is only for compatibility with broken device
78 * drivers in old versions of Linux that do not properly support VLANs when
79 * VLAN devices are not used. When broken device drivers are no longer in
80 * widespread use, we will delete these interfaces. */
81 unsigned long int *vlan_bitmap; /* 4096-bit bitmap of in-use VLANs. */
82 bool vlans_changed; /* True if new VLANs are in use. */
85 void ofproto_init_tables(struct ofproto *, int n_tables);
87 struct ofproto *ofproto_lookup(const char *name);
88 struct ofport *ofproto_get_port(const struct ofproto *, uint16_t ofp_port);
90 /* An OpenFlow port within a "struct ofproto".
92 * With few exceptions, ofproto implementations may look at these fields but
93 * should not modify them. */
95 struct ofproto *ofproto; /* The ofproto that contains this port. */
96 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
97 struct netdev *netdev;
98 struct ofp_phy_port opp;
99 uint16_t ofp_port; /* OpenFlow port number. */
100 unsigned int change_seq;
104 void ofproto_port_set_state(struct ofport *, ovs_be32 state);
107 OFTABLE_HIDDEN = 1 << 0, /* Hide from most OpenFlow operations. */
108 OFTABLE_READONLY = 1 << 1 /* Don't allow OpenFlow to change this table. */
111 /* A flow table within a "struct ofproto". */
113 enum oftable_flags flags;
114 struct classifier cls; /* Contains "struct rule"s. */
115 char *name; /* Table name exposed via OpenFlow, or NULL. */
117 /* Maximum number of flows or UINT_MAX if there is no limit besides any
118 * limit imposed by resource limitations. */
119 unsigned int max_flows;
121 /* These members determine the handling of an attempt to add a flow that
122 * would cause the table to have more than 'max_flows' flows.
124 * If 'eviction_fields' is NULL, overflows will be rejected with an error.
126 * If 'eviction_fields' is nonnull (regardless of whether n_eviction_fields
127 * is nonzero), an overflow will cause a flow to be removed. The flow to
128 * be removed is chosen to give fairness among groups distinguished by
129 * different values for the subfields within 'groups'. */
130 struct mf_subfield *eviction_fields;
131 size_t n_eviction_fields;
135 * When a flow is added that would cause the table to have more than
136 * 'max_flows' flows, and 'eviction_fields' is nonnull, these groups are
137 * used to decide which rule to evict: the rule is chosen from the eviction
138 * group that contains the greatest number of rules.*/
139 uint32_t eviction_group_id_basis;
140 struct hmap eviction_groups_by_id;
141 struct heap eviction_groups_by_size;
144 /* Assigns TABLE to each oftable, in turn, in OFPROTO.
146 * All parameters are evaluated multiple times. */
147 #define OFPROTO_FOR_EACH_TABLE(TABLE, OFPROTO) \
148 for ((TABLE) = (OFPROTO)->tables; \
149 (TABLE) < &(OFPROTO)->tables[(OFPROTO)->n_tables]; \
152 /* An OpenFlow flow within a "struct ofproto".
154 * With few exceptions, ofproto implementations may look at these fields but
155 * should not modify them. */
157 struct ofproto *ofproto; /* The ofproto that contains this rule. */
158 struct list ofproto_node; /* Owned by ofproto base code. */
159 struct cls_rule cr; /* In owning ofproto's classifier. */
161 struct ofoperation *pending; /* Operation now in progress, if nonnull. */
163 ovs_be64 flow_cookie; /* Controller-issued identifier. */
165 long long int created; /* Creation time. */
166 long long int modified; /* Time of last modification. */
167 long long int used; /* Last use; time created if never used. */
168 uint16_t hard_timeout; /* In seconds from ->modified. */
169 uint16_t idle_timeout; /* In seconds from ->used. */
170 uint8_t table_id; /* Index in ofproto's 'tables' array. */
171 bool send_flow_removed; /* Send a flow removed message? */
173 /* Eviction groups. */
174 bool evictable; /* If false, prevents eviction. */
175 struct heap_node evg_node; /* In eviction_group's "rules" heap. */
176 struct eviction_group *eviction_group; /* NULL if not in any group. */
178 union ofp_action *actions; /* OpenFlow actions. */
179 int n_actions; /* Number of elements in actions[]. */
182 static inline struct rule *
183 rule_from_cls_rule(const struct cls_rule *cls_rule)
185 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
188 void ofproto_rule_update_used(struct rule *, long long int used);
189 void ofproto_rule_expire(struct rule *, uint8_t reason);
190 void ofproto_rule_destroy(struct rule *);
192 void ofoperation_complete(struct ofoperation *, enum ofperr);
193 struct rule *ofoperation_get_victim(struct ofoperation *);
195 /* ofproto class structure, to be defined by each ofproto implementation.
201 * These functions work primarily with three different kinds of data
204 * - "struct ofproto", which represents an OpenFlow switch.
206 * - "struct ofport", which represents a port within an ofproto.
208 * - "struct rule", which represents an OpenFlow flow within an ofproto.
210 * Each of these data structures contains all of the implementation-independent
211 * generic state for the respective concept, called the "base" state. None of
212 * them contains any extra space for ofproto implementations to use. Instead,
213 * each implementation is expected to declare its own data structure that
214 * contains an instance of the generic data structure plus additional
215 * implementation-specific members, called the "derived" state. The
216 * implementation can use casts or (preferably) the CONTAINER_OF macro to
217 * obtain access to derived state given only a pointer to the embedded generic
224 * Four stylized functions accompany each of these data structures:
226 * "alloc" "construct" "destruct" "dealloc"
227 * ------------ ---------------- --------------- --------------
228 * ofproto ->alloc ->construct ->destruct ->dealloc
229 * ofport ->port_alloc ->port_construct ->port_destruct ->port_dealloc
230 * rule ->rule_alloc ->rule_construct ->rule_destruct ->rule_dealloc
232 * Any instance of a given data structure goes through the following life
235 * 1. The client calls the "alloc" function to obtain raw memory. If "alloc"
236 * fails, skip all the other steps.
238 * 2. The client initializes all of the data structure's base state. If this
239 * fails, skip to step 7.
241 * 3. The client calls the "construct" function. The implementation
242 * initializes derived state. It may refer to the already-initialized
243 * base state. If "construct" fails, skip to step 6.
245 * 4. The data structure is now initialized and in use.
247 * 5. When the data structure is no longer needed, the client calls the
248 * "destruct" function. The implementation uninitializes derived state.
249 * The base state has not been uninitialized yet, so the implementation
250 * may still refer to it.
252 * 6. The client uninitializes all of the data structure's base state.
254 * 7. The client calls the "dealloc" to free the raw memory. The
255 * implementation must not refer to base or derived state in the data
256 * structure, because it has already been uninitialized.
258 * Each "alloc" function allocates and returns a new instance of the respective
259 * data structure. The "alloc" function is not given any information about the
260 * use of the new data structure, so it cannot perform much initialization.
261 * Its purpose is just to ensure that the new data structure has enough room
262 * for base and derived state. It may return a null pointer if memory is not
263 * available, in which case none of the other functions is called.
265 * Each "construct" function initializes derived state in its respective data
266 * structure. When "construct" is called, all of the base state has already
267 * been initialized, so the "construct" function may refer to it. The
268 * "construct" function is allowed to fail, in which case the client calls the
269 * "dealloc" function (but not the "destruct" function).
271 * Each "destruct" function uninitializes and frees derived state in its
272 * respective data structure. When "destruct" is called, the base state has
273 * not yet been uninitialized, so the "destruct" function may refer to it. The
274 * "destruct" function is not allowed to fail.
276 * Each "dealloc" function frees raw memory that was allocated by the the
277 * "alloc" function. The memory's base and derived members might not have ever
278 * been initialized (but if "construct" returned successfully, then it has been
279 * "destruct"ed already). The "dealloc" function is not allowed to fail.
285 * Most of these functions return 0 if they are successful or a positive error
286 * code on failure. Depending on the function, valid error codes are either
287 * errno values or OFPERR_* OpenFlow error codes.
289 * Most of these functions are expected to execute synchronously, that is, to
290 * block as necessary to obtain a result. Thus, these functions may return
291 * EAGAIN (or EWOULDBLOCK or EINPROGRESS) only where the function descriptions
292 * explicitly say those errors are a possibility. We may relax this
293 * requirement in the future if and when we encounter performance problems. */
294 struct ofproto_class {
295 /* ## ----------------- ## */
296 /* ## Factory Functions ## */
297 /* ## ----------------- ## */
299 /* Enumerates the types of all support ofproto types into 'types'. The
300 * caller has already initialized 'types' and other ofproto classes might
301 * already have added names to it. */
302 void (*enumerate_types)(struct sset *types);
304 /* Enumerates the names of all existing datapath of the specified 'type'
305 * into 'names' 'all_dps'. The caller has already initialized 'names' as
308 * 'type' is one of the types enumerated by ->enumerate_types().
310 * Returns 0 if successful, otherwise a positive errno value.
312 int (*enumerate_names)(const char *type, struct sset *names);
314 /* Deletes the datapath with the specified 'type' and 'name'. The caller
315 * should have closed any open ofproto with this 'type' and 'name'; this
316 * function is allowed to fail if that is not the case.
318 * 'type' is one of the types enumerated by ->enumerate_types().
319 * 'name' is one of the names enumerated by ->enumerate_names() for 'type'.
321 * Returns 0 if successful, otherwise a positive errno value.
323 int (*del)(const char *type, const char *name);
325 /* ## --------------------------- ## */
326 /* ## Top-Level ofproto Functions ## */
327 /* ## --------------------------- ## */
329 /* Life-cycle functions for an "ofproto" (see "Life Cycle" above).
335 * ->construct() should not modify any base members of the ofproto. The
336 * client will initialize the ofproto's 'ports' and 'tables' members after
337 * construction is complete.
339 * When ->construct() is called, the client does not yet know how many flow
340 * tables the datapath supports, so ofproto->n_tables will be 0 and
341 * ofproto->tables will be NULL. ->construct() should call
342 * ofproto_init_tables() to allocate and initialize ofproto->n_tables and
343 * ofproto->tables. Each flow table will be initially empty, so
344 * ->construct() should delete flows from the underlying datapath, if
345 * necessary, rather than populating the tables.
347 * Only one ofproto instance needs to be supported for any given datapath.
348 * If a datapath is already open as part of one "ofproto", then another
349 * attempt to "construct" the same datapath as part of another ofproto is
350 * allowed to fail with an error.
352 * ->construct() returns 0 if successful, otherwise a positive errno
359 * If 'ofproto' has any pending asynchronous operations, ->destruct()
360 * must complete all of them by calling ofoperation_complete().
362 * ->destruct() must also destroy all remaining rules in the ofproto's
363 * tables, by passing each remaining rule to ofproto_rule_destroy(). The
364 * client will destroy the flow tables themselves after ->destruct()
367 struct ofproto *(*alloc)(void);
368 int (*construct)(struct ofproto *ofproto);
369 void (*destruct)(struct ofproto *ofproto);
370 void (*dealloc)(struct ofproto *ofproto);
372 /* Performs any periodic activity required by 'ofproto'. It should:
374 * - Call connmgr_send_packet_in() for each received packet that missed
375 * in the OpenFlow flow table or that had a OFPP_CONTROLLER output
378 * - Call ofproto_rule_expire() for each OpenFlow flow that has reached
379 * its hard_timeout or idle_timeout, to expire the flow.
381 * Returns 0 if successful, otherwise a positive errno value. */
382 int (*run)(struct ofproto *ofproto);
384 /* Performs periodic activity required by 'ofproto' that needs to be done
385 * with the least possible latency.
387 * This is run multiple times per main loop. An ofproto provider may
388 * implement it or not, according to whether it provides a performance
389 * boost for that ofproto implementation. */
390 int (*run_fast)(struct ofproto *ofproto);
392 /* Causes the poll loop to wake up when 'ofproto''s 'run' function needs to
393 * be called, e.g. by calling the timer or fd waiting functions in
395 void (*wait)(struct ofproto *ofproto);
397 /* Every "struct rule" in 'ofproto' is about to be deleted, one by one.
398 * This function may prepare for that, for example by clearing state in
399 * advance. It should *not* actually delete any "struct rule"s from
400 * 'ofproto', only prepare for it.
402 * This function is optional; it's really just for optimization in case
403 * it's cheaper to delete all the flows from your hardware in a single pass
404 * than to do it one by one. */
405 void (*flush)(struct ofproto *ofproto);
407 /* Helper for the OpenFlow OFPT_FEATURES_REQUEST request.
409 * The implementation should store true in '*arp_match_ip' if the switch
410 * supports matching IP addresses inside ARP requests and replies, false
413 * The implementation should store in '*actions' a bitmap of the supported
414 * OpenFlow actions: the bit with value (1 << n) should be set to 1 if the
415 * implementation supports the action with value 'n', and to 0 otherwise.
416 * For example, if the implementation supports the OFPAT_OUTPUT and
417 * OFPAT_ENQUEUE actions, but no others, it would set '*actions' to (1 <<
418 * OFPAT_OUTPUT) | (1 << OFPAT_ENQUEUE). Vendor actions are not included
420 void (*get_features)(struct ofproto *ofproto,
421 bool *arp_match_ip, uint32_t *actions);
423 /* Helper for the OpenFlow OFPST_TABLE statistics request.
425 * The 'ots' array contains 'ofproto->n_tables' elements. Each element is
428 * - 'table_id' to the array index.
430 * - 'name' to "table#" where # is the table ID.
432 * - 'wildcards' to OFPFW_ALL.
434 * - 'max_entries' to 1,000,000.
436 * - 'active_count' to the classifier_count() for the table.
438 * - 'lookup_count' and 'matched_count' to 0.
440 * The implementation should update any members in each element for which
441 * it has better values:
443 * - 'name' to a more meaningful name.
445 * - 'wildcards' to the set of wildcards actually supported by the table
446 * (if it doesn't support all OpenFlow wildcards).
448 * - 'max_entries' to the maximum number of flows actually supported by
451 * - 'lookup_count' to the number of packets looked up in this flow table
454 * - 'matched_count' to the number of packets looked up in this flow
455 * table so far that matched one of the flow entries.
457 * Keep in mind that all of the members of struct ofp_table_stats are in
458 * network byte order.
460 void (*get_tables)(struct ofproto *ofproto, struct ofp_table_stats *ots);
462 /* ## ---------------- ## */
463 /* ## ofport Functions ## */
464 /* ## ---------------- ## */
466 /* Life-cycle functions for a "struct ofport" (see "Life Cycle" above).
468 * ->port_construct() should not modify any base members of the ofport.
470 * ofports are managed by the base ofproto code. The ofproto
471 * implementation should only create and destroy them in response to calls
472 * to these functions. The base ofproto code will create and destroy
473 * ofports in the following situations:
475 * - Just after the ->construct() function is called, the base ofproto
476 * iterates over all of the implementation's ports, using
477 * ->port_dump_start() and related functions, and constructs an ofport
478 * for each dumped port.
480 * - If ->port_poll() reports that a specific port has changed, then the
481 * base ofproto will query that port with ->port_query_by_name() and
482 * construct or destruct ofports as necessary to reflect the updated
485 * - If ->port_poll() returns ENOBUFS to report an unspecified port set
486 * change, then the base ofproto will iterate over all of the
487 * implementation's ports, in the same way as at ofproto
488 * initialization, and construct and destruct ofports to reflect all of
491 * ->port_construct() returns 0 if successful, otherwise a positive errno
494 struct ofport *(*port_alloc)(void);
495 int (*port_construct)(struct ofport *ofport);
496 void (*port_destruct)(struct ofport *ofport);
497 void (*port_dealloc)(struct ofport *ofport);
499 /* Called after 'ofport->netdev' is replaced by a new netdev object. If
500 * the ofproto implementation uses the ofport's netdev internally, then it
501 * should switch to using the new one. The old one has been closed.
503 * An ofproto implementation that doesn't need to do anything in this
504 * function may use a null pointer. */
505 void (*port_modified)(struct ofport *ofport);
507 /* Called after an OpenFlow OFPT_PORT_MOD request changes a port's
508 * configuration. 'ofport->opp.config' contains the new configuration.
509 * 'old_config' contains the previous configuration.
511 * The caller implements OFPPC_PORT_DOWN using netdev functions to turn
512 * NETDEV_UP on and off, so this function doesn't have to do anything for
513 * that bit (and it won't be called if that is the only bit that
515 void (*port_reconfigured)(struct ofport *ofport, ovs_be32 old_config);
517 /* Looks up a port named 'devname' in 'ofproto'. On success, initializes
518 * '*port' appropriately.
520 * The caller owns the data in 'port' and must free it with
521 * ofproto_port_destroy() when it is no longer needed. */
522 int (*port_query_by_name)(const struct ofproto *ofproto,
523 const char *devname, struct ofproto_port *port);
525 /* Attempts to add 'netdev' as a port on 'ofproto'. Returns 0 if
526 * successful, otherwise a positive errno value. If successful, sets
527 * '*ofp_portp' to the new port's port number.
529 * It doesn't matter whether the new port will be returned by a later call
530 * to ->port_poll(); the implementation may do whatever is more
532 int (*port_add)(struct ofproto *ofproto, struct netdev *netdev,
533 uint16_t *ofp_portp);
535 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'. Returns
536 * 0 if successful, otherwise a positive errno value.
538 * It doesn't matter whether the new port will be returned by a later call
539 * to ->port_poll(); the implementation may do whatever is more
541 int (*port_del)(struct ofproto *ofproto, uint16_t ofp_port);
544 int (*port_get_stats)(const struct ofport *port,
545 struct netdev_stats *stats);
547 /* Port iteration functions.
549 * The client might not be entirely in control of the ports within an
550 * ofproto. Some hardware implementations, for example, might have a fixed
551 * set of ports in a datapath, and the Linux datapath allows the system
552 * administrator to externally add and remove ports with ovs-dpctl. For
553 * this reason, the client needs a way to iterate through all the ports
554 * that are actually in a datapath. These functions provide that
557 * The 'state' pointer provides the implementation a place to
558 * keep track of its position. Its format is opaque to the caller.
560 * The ofproto provider retains ownership of the data that it stores into
561 * ->port_dump_next()'s 'port' argument. The data must remain valid until
562 * at least the next call to ->port_dump_next() or ->port_dump_done() for
563 * 'state'. The caller will not modify or free it.
568 * ->port_dump_start() attempts to begin dumping the ports in 'ofproto'.
569 * On success, it should return 0 and initialize '*statep' with any data
570 * needed for iteration. On failure, returns a positive errno value, and
571 * the client will not call ->port_dump_next() or ->port_dump_done().
573 * ->port_dump_next() attempts to retrieve another port from 'ofproto' for
574 * 'state'. If there is another port, it should store the port's
575 * information into 'port' and return 0. It should return EOF if all ports
576 * have already been iterated. Otherwise, on error, it should return a
577 * positive errno value. This function will not be called again once it
578 * returns nonzero once for a given iteration (but the 'port_dump_done'
579 * function will be called afterward).
581 * ->port_dump_done() allows the implementation to release resources used
582 * for iteration. The caller might decide to stop iteration in the middle
583 * by calling this function before ->port_dump_next() returns nonzero.
591 * error = ofproto->ofproto_class->port_dump_start(ofproto, &state);
594 * struct ofproto_port port;
596 * error = ofproto->ofproto_class->port_dump_next(
597 * ofproto, state, &port);
601 * // Do something with 'port' here (without modifying or freeing
602 * // any of its data).
604 * ofproto->ofproto_class->port_dump_done(ofproto, state);
606 * // 'error' is now EOF (success) or a positive errno value (failure).
608 int (*port_dump_start)(const struct ofproto *ofproto, void **statep);
609 int (*port_dump_next)(const struct ofproto *ofproto, void *state,
610 struct ofproto_port *port);
611 int (*port_dump_done)(const struct ofproto *ofproto, void *state);
613 /* Polls for changes in the set of ports in 'ofproto'. If the set of ports
614 * in 'ofproto' has changed, then this function should do one of the
617 * - Preferably: store the name of the device that was added to or deleted
618 * from 'ofproto' in '*devnamep' and return 0. The caller is responsible
619 * for freeing '*devnamep' (with free()) when it no longer needs it.
621 * - Alternatively: return ENOBUFS, without indicating the device that was
624 * Occasional 'false positives', in which the function returns 0 while
625 * indicating a device that was not actually added or deleted or returns
626 * ENOBUFS without any change, are acceptable.
628 * The purpose of 'port_poll' is to let 'ofproto' know about changes made
629 * externally to the 'ofproto' object, e.g. by a system administrator via
630 * ovs-dpctl. Therefore, it's OK, and even preferable, for port_poll() to
631 * not report changes made through calls to 'port_add' or 'port_del' on the
632 * same 'ofproto' object. (But it's OK for it to report them too, just
633 * slightly less efficient.)
635 * If the set of ports in 'ofproto' has not changed, returns EAGAIN. May
636 * also return other positive errno values to indicate that something has
639 * If the set of ports in a datapath is fixed, or if the only way that the
640 * set of ports in a datapath can change is through ->port_add() and
641 * ->port_del(), then this function may be a null pointer.
643 int (*port_poll)(const struct ofproto *ofproto, char **devnamep);
645 /* Arranges for the poll loop to wake up when ->port_poll() will return a
646 * value other than EAGAIN.
648 * If the set of ports in a datapath is fixed, or if the only way that the
649 * set of ports in a datapath can change is through ->port_add() and
650 * ->port_del(), or if the poll loop will always wake up anyway when
651 * ->port_poll() will return a value other than EAGAIN, then this function
652 * may be a null pointer.
654 void (*port_poll_wait)(const struct ofproto *ofproto);
656 /* Checks the status of LACP negotiation for 'port'. Returns 1 if LACP
657 * partner information for 'port' is up-to-date, 0 if LACP partner
658 * information is not current (generally indicating a connectivity
659 * problem), or -1 if LACP is not enabled on 'port'.
661 * This function may be a null pointer if the ofproto implementation does
662 * not support LACP. */
663 int (*port_is_lacp_current)(const struct ofport *port);
665 /* ## ----------------------- ## */
666 /* ## OpenFlow Rule Functions ## */
667 /* ## ----------------------- ## */
671 /* Chooses an appropriate table for 'cls_rule' within 'ofproto'. On
672 * success, stores the table ID into '*table_idp' and returns 0. On
673 * failure, returns an OpenFlow error code.
675 * The choice of table should be a function of 'cls_rule' and 'ofproto''s
676 * datapath capabilities. It should not depend on the flows already in
677 * 'ofproto''s flow tables. Failure implies that an OpenFlow rule with
678 * 'cls_rule' as its matching condition can never be inserted into
679 * 'ofproto', even starting from an empty flow table.
681 * If multiple tables are candidates for inserting the flow, the function
682 * should choose one arbitrarily (but deterministically).
684 * If this function is NULL then table 0 is always chosen. */
685 enum ofperr (*rule_choose_table)(const struct ofproto *ofproto,
686 const struct cls_rule *cls_rule,
689 /* Life-cycle functions for a "struct rule" (see "Life Cycle" above).
692 * Asynchronous Operation Support
693 * ==============================
695 * The life-cycle operations on rules can operate asynchronously, meaning
696 * that ->rule_construct() and ->rule_destruct() only need to initiate
697 * their respective operations and do not need to wait for them to complete
698 * before they return. ->rule_modify_actions() also operates
701 * An ofproto implementation reports the success or failure of an
702 * asynchronous operation on a rule using the rule's 'pending' member,
703 * which points to a opaque "struct ofoperation" that represents the
704 * ongoing opreation. When the operation completes, the ofproto
705 * implementation calls ofoperation_complete(), passing the ofoperation and
706 * an error indication.
708 * Only the following contexts may call ofoperation_complete():
710 * - The function called to initiate the operation,
711 * e.g. ->rule_construct() or ->rule_destruct(). This is the best
712 * choice if the operation completes quickly.
714 * - The implementation's ->run() function.
716 * - The implementation's ->destruct() function.
718 * The ofproto base code updates the flow table optimistically, assuming
719 * that the operation will probably succeed:
721 * - ofproto adds or replaces the rule in the flow table before calling
722 * ->rule_construct().
724 * - ofproto updates the rule's actions before calling
725 * ->rule_modify_actions().
727 * - ofproto removes the rule before calling ->rule_destruct().
729 * With one exception, when an asynchronous operation completes with an
730 * error, ofoperation_complete() backs out the already applied changes:
732 * - If adding or replacing a rule in the flow table fails, ofproto
733 * removes the new rule or restores the original rule.
735 * - If modifying a rule's actions fails, ofproto restores the original
738 * - Removing a rule is not allowed to fail. It must always succeed.
740 * The ofproto base code serializes operations: if any operation is in
741 * progress on a given rule, ofproto postpones initiating any new operation
742 * on that rule until the pending operation completes. Therefore, every
743 * operation must eventually complete through a call to
744 * ofoperation_complete() to avoid delaying new operations indefinitely
745 * (including any OpenFlow request that affects the rule in question, even
746 * just to query its statistics).
752 * When ->rule_construct() is called, the caller has already inserted
753 * 'rule' into 'rule->ofproto''s flow table numbered 'rule->table_id'.
754 * There are two cases:
756 * - 'rule' is a new rule in its flow table. In this case,
757 * ofoperation_get_victim(rule) returns NULL.
759 * - 'rule' is replacing an existing rule in its flow table that had the
760 * same matching criteria and priority. In this case,
761 * ofoperation_get_victim(rule) returns the rule being replaced (the
764 * ->rule_construct() should set the following in motion:
766 * - Validate that the matching rule in 'rule->cr' is supported by the
767 * datapath. For example, if the rule's table does not support
768 * registers, then it is an error if 'rule->cr' does not wildcard all
771 * - Validate that 'rule->actions' and 'rule->n_actions' are well-formed
772 * OpenFlow actions that the datapath can correctly implement. The
773 * validate_actions() function (in ofp-util.c) can be useful as a model
774 * for action validation, but it accepts all of the OpenFlow actions
775 * that OVS understands. If your ofproto implementation only
776 * implements a subset of those, then you should implement your own
779 * - If the rule is valid, update the datapath flow table, adding the new
780 * rule or replacing the existing one.
782 * - If 'rule' is replacing an existing rule, uninitialize any derived
783 * state for the victim rule, as in step 5 in the "Life Cycle"
786 * (On failure, the ofproto code will roll back the insertion from the flow
787 * table, either removing 'rule' or replacing it by the victim rule if
790 * ->rule_construct() must act in one of the following ways:
792 * - If it succeeds, it must call ofoperation_complete() and return 0.
794 * - If it fails, it must act in one of the following ways:
796 * * Call ofoperation_complete() and return 0.
798 * * Return an OpenFlow error code. (Do not call
799 * ofoperation_complete() in this case.)
801 * Either way, ->rule_destruct() will not be called for 'rule', but
802 * ->rule_dealloc() will be.
804 * - If the operation is only partially complete, then it must return 0.
805 * Later, when the operation is complete, the ->run() or ->destruct()
806 * function must call ofoperation_complete() to report success or
809 * ->rule_construct() should not modify any base members of struct rule.
815 * When ->rule_destruct() is called, the caller has already removed 'rule'
816 * from 'rule->ofproto''s flow table. ->rule_destruct() should set in
817 * motion removing 'rule' from the datapath flow table. If removal
818 * completes synchronously, it should call ofoperation_complete().
819 * Otherwise, the ->run() or ->destruct() function must later call
820 * ofoperation_complete() after the operation completes.
822 * Rule destruction must not fail. */
823 struct rule *(*rule_alloc)(void);
824 enum ofperr (*rule_construct)(struct rule *rule);
825 void (*rule_destruct)(struct rule *rule);
826 void (*rule_dealloc)(struct rule *rule);
828 /* Obtains statistics for 'rule', storing the number of packets that have
829 * matched it in '*packet_count' and the number of bytes in those packets
830 * in '*byte_count'. UINT64_MAX indicates that the packet count or byte
831 * count is unknown. */
832 void (*rule_get_stats)(struct rule *rule, uint64_t *packet_count,
833 uint64_t *byte_count);
835 /* Applies the actions in 'rule' to 'packet'. (This implements sending
836 * buffered packets for OpenFlow OFPT_FLOW_MOD commands.)
838 * Takes ownership of 'packet' (so it should eventually free it, with
841 * 'flow' reflects the flow information for 'packet'. All of the
842 * information in 'flow' is extracted from 'packet', except for
843 * flow->tun_id and flow->in_port, which are assigned the correct values
844 * for the incoming packet. The register values are zeroed.
846 * The statistics for 'packet' should be included in 'rule'.
848 * Returns 0 if successful, otherwise an OpenFlow error code. */
849 enum ofperr (*rule_execute)(struct rule *rule, const struct flow *flow,
850 struct ofpbuf *packet);
852 /* When ->rule_modify_actions() is called, the caller has already replaced
853 * the OpenFlow actions in 'rule' by a new set. (The original actions are
854 * in rule->pending->actions.)
856 * ->rule_modify_actions() should set the following in motion:
858 * - Validate that the actions now in 'rule' are well-formed OpenFlow
859 * actions that the datapath can correctly implement.
861 * - Update the datapath flow table with the new actions.
863 * If the operation synchronously completes, ->rule_modify_actions() may
864 * call ofoperation_complete() before it returns. Otherwise, ->run()
865 * should call ofoperation_complete() later, after the operation does
868 * If the operation fails, then the base ofproto code will restore the
869 * original 'actions' and 'n_actions' of 'rule'.
871 * ->rule_modify_actions() should not modify any base members of struct
873 void (*rule_modify_actions)(struct rule *rule);
875 /* Changes the OpenFlow IP fragment handling policy to 'frag_handling',
876 * which takes one of the following values, with the corresponding
879 * - OFPC_FRAG_NORMAL: The switch should treat IP fragments the same way
880 * as other packets, omitting TCP and UDP port numbers (always setting
883 * - OFPC_FRAG_DROP: The switch should drop all IP fragments without
884 * passing them through the flow table.
886 * - OFPC_FRAG_REASM: The switch should reassemble IP fragments before
887 * passing packets through the flow table.
889 * - OFPC_FRAG_NX_MATCH (a Nicira extension): Similar to OFPC_FRAG_NORMAL,
890 * except that TCP and UDP port numbers should be included in fragments
893 * Implementations are not required to support every mode.
894 * OFPC_FRAG_NORMAL is the default mode when an ofproto is created.
896 * At the time of the call to ->set_frag_handling(), the current mode is
897 * available in 'ofproto->frag_handling'. ->set_frag_handling() returns
898 * true if the requested mode was set, false if it is not supported.
900 * Upon successful return, the caller changes 'ofproto->frag_handling' to
901 * reflect the new mode.
903 bool (*set_frag_handling)(struct ofproto *ofproto,
904 enum ofp_config_flags frag_handling);
906 /* Implements the OpenFlow OFPT_PACKET_OUT command. The datapath should
907 * execute the 'n_actions' in the 'actions' array on 'packet'.
909 * The caller retains ownership of 'packet', so ->packet_out() should not
912 * This function must validate that the 'n_actions' elements in 'actions'
913 * are well-formed OpenFlow actions that can be correctly implemented by
914 * the datapath. If not, then it should return an OpenFlow error code.
916 * 'flow' reflects the flow information for 'packet'. All of the
917 * information in 'flow' is extracted from 'packet', except for
918 * flow->in_port, which is taken from the OFPT_PACKET_OUT message.
919 * flow->tun_id and its register values are zeroed.
921 * 'packet' is not matched against the OpenFlow flow table, so its
922 * statistics should not be included in OpenFlow flow statistics.
924 * Returns 0 if successful, otherwise an OpenFlow error code. */
925 enum ofperr (*packet_out)(struct ofproto *ofproto, struct ofpbuf *packet,
926 const struct flow *flow,
927 const union ofp_action *actions,
930 /* ## ------------------------- ## */
931 /* ## OFPP_NORMAL configuration ## */
932 /* ## ------------------------- ## */
934 /* Configures NetFlow on 'ofproto' according to the options in
935 * 'netflow_options', or turns off NetFlow if 'netflow_options' is NULL.
937 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
938 * NetFlow, as does a null pointer. */
939 int (*set_netflow)(struct ofproto *ofproto,
940 const struct netflow_options *netflow_options);
942 void (*get_netflow_ids)(const struct ofproto *ofproto,
943 uint8_t *engine_type, uint8_t *engine_id);
945 /* Configures sFlow on 'ofproto' according to the options in
946 * 'sflow_options', or turns off sFlow if 'sflow_options' is NULL.
948 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
949 * sFlow, as does a null pointer. */
950 int (*set_sflow)(struct ofproto *ofproto,
951 const struct ofproto_sflow_options *sflow_options);
953 /* Configures connectivity fault management on 'ofport'.
955 * If 'cfm_settings' is nonnull, configures CFM according to its members.
957 * If 'cfm_settings' is null, removes any connectivity fault management
958 * configuration from 'ofport'.
960 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
961 * support CFM, as does a null pointer. */
962 int (*set_cfm)(struct ofport *ofport, const struct cfm_settings *s);
964 /* Checks the fault status of CFM configured on 'ofport'. Returns a
965 * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
966 * indicating a connectivity problem). Returns zero if CFM is not faulted,
967 * and -1 if CFM is not enabled on 'port'.
969 * This function may be a null pointer if the ofproto implementation does
970 * not support CFM. */
971 int (*get_cfm_fault)(const struct ofport *ofport);
973 /* Gets the MPIDs of the remote maintenance points broadcasting to
974 * 'ofport'. Populates 'rmps' with a provider owned array of MPIDs, and
975 * 'n_rmps' with the number of MPIDs in 'rmps'. Returns a number less than
976 * 0 if CFM is not enabled of 'ofport'.
978 * This function may be a null pointer if the ofproto implementation does
979 * not support CFM. */
980 int (*get_cfm_remote_mpids)(const struct ofport *ofport,
981 const uint64_t **rmps, size_t *n_rmps);
983 /* Configures spanning tree protocol (STP) on 'ofproto' using the
984 * settings defined in 's'.
986 * If 's' is nonnull, configures STP according to its members.
988 * If 's' is null, removes any STP configuration from 'ofproto'.
990 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
991 * support STP, as does a null pointer. */
992 int (*set_stp)(struct ofproto *ofproto,
993 const struct ofproto_stp_settings *s);
995 /* Retrieves state of spanning tree protocol (STP) on 'ofproto'.
997 * Stores STP state for 'ofproto' in 's'. If the 'enabled' member
998 * is false, the other member values are not meaningful.
1000 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1001 * support STP, as does a null pointer. */
1002 int (*get_stp_status)(struct ofproto *ofproto,
1003 struct ofproto_stp_status *s);
1005 /* Configures spanning tree protocol (STP) on 'ofport' using the
1006 * settings defined in 's'.
1008 * If 's' is nonnull, configures STP according to its members. The
1009 * caller is responsible for assigning STP port numbers (using the
1010 * 'port_num' member in the range of 1 through 255, inclusive) and
1011 * ensuring there are no duplicates.
1013 * If 's' is null, removes any STP configuration from 'ofport'.
1015 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1016 * support STP, as does a null pointer. */
1017 int (*set_stp_port)(struct ofport *ofport,
1018 const struct ofproto_port_stp_settings *s);
1020 /* Retrieves spanning tree protocol (STP) port status of 'ofport'.
1022 * Stores STP state for 'ofport' in 's'. If the 'enabled' member is
1023 * false, the other member values are not meaningful.
1025 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1026 * support STP, as does a null pointer. */
1027 int (*get_stp_port_status)(struct ofport *ofport,
1028 struct ofproto_port_stp_status *s);
1030 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
1031 * 'queues' attached to 'ofport'. This data is not intended to be
1032 * sufficient to implement QoS. Instead, providers may use this
1033 * information to implement features which require knowledge of what queues
1034 * exist on a port, and some basic information about them.
1036 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1037 * support QoS, as does a null pointer. */
1038 int (*set_queues)(struct ofport *ofport,
1039 const struct ofproto_port_queue *queues, size_t n_qdscp);
1041 /* If 's' is nonnull, this function registers a "bundle" associated with
1042 * client data pointer 'aux' in 'ofproto'. A bundle is the same concept as
1043 * a Port in OVSDB, that is, it consists of one or more "slave" devices
1044 * (Interfaces, in OVSDB) along with VLAN and LACP configuration and, if
1045 * there is more than one slave, a bonding configuration. If 'aux' is
1046 * already registered then this function updates its configuration to 's'.
1047 * Otherwise, this function registers a new bundle.
1049 * If 's' is NULL, this function unregisters the bundle registered on
1050 * 'ofproto' associated with client data pointer 'aux'. If no such bundle
1051 * has been registered, this has no effect.
1053 * This function affects only the behavior of the NXAST_AUTOPATH action and
1054 * output to the OFPP_NORMAL port. An implementation that does not support
1055 * it at all may set it to NULL or return EOPNOTSUPP. An implementation
1056 * that supports only a subset of the functionality should implement what
1057 * it can and return 0. */
1058 int (*bundle_set)(struct ofproto *ofproto, void *aux,
1059 const struct ofproto_bundle_settings *s);
1061 /* If 'port' is part of any bundle, removes it from that bundle. If the
1062 * bundle now has no ports, deletes the bundle. If the bundle now has only
1063 * one port, deconfigures the bundle's bonding configuration. */
1064 void (*bundle_remove)(struct ofport *ofport);
1066 /* If 's' is nonnull, this function registers a mirror associated with
1067 * client data pointer 'aux' in 'ofproto'. A mirror is the same concept as
1068 * a Mirror in OVSDB. If 'aux' is already registered then this function
1069 * updates its configuration to 's'. Otherwise, this function registers a
1072 * If 's' is NULL, this function unregisters the mirror registered on
1073 * 'ofproto' associated with client data pointer 'aux'. If no such mirror
1074 * has been registered, this has no effect.
1076 * An implementation that does not support mirroring at all may set
1077 * it to NULL or return EOPNOTSUPP. An implementation that supports
1078 * only a subset of the functionality should implement what it can
1080 int (*mirror_set)(struct ofproto *ofproto, void *aux,
1081 const struct ofproto_mirror_settings *s);
1083 /* Retrieves statistics from mirror associated with client data
1084 * pointer 'aux' in 'ofproto'. Stores packet and byte counts in
1085 * 'packets' and 'bytes', respectively. If a particular counter is
1086 * not supported, the appropriate argument is set to UINT64_MAX.
1088 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1089 * support retrieving mirror statistics. */
1090 int (*mirror_get_stats)(struct ofproto *ofproto, void *aux,
1091 uint64_t *packets, uint64_t *bytes);
1093 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs
1094 * on which all packets are flooded, instead of using MAC learning. If
1095 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1097 * This function affects only the behavior of the OFPP_NORMAL action. An
1098 * implementation that does not support it may set it to NULL or return
1100 int (*set_flood_vlans)(struct ofproto *ofproto,
1101 unsigned long *flood_vlans);
1103 /* Returns true if 'aux' is a registered bundle that is currently in use as
1104 * the output for a mirror. */
1105 bool (*is_mirror_output_bundle)(const struct ofproto *ofproto, void *aux);
1107 /* When the configuration option of forward_bpdu changes, this function
1108 * will be invoked. */
1109 void (*forward_bpdu_changed)(struct ofproto *ofproto);
1111 /* Sets the MAC aging timeout for the OFPP_NORMAL action to 'idle_time',
1113 void (*set_mac_idle_time)(struct ofproto *ofproto, unsigned int idle_time);
1115 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
1117 * This is deprecated. It is only for compatibility with broken device drivers
1118 * in old versions of Linux that do not properly support VLANs when VLAN
1119 * devices are not used. When broken device drivers are no longer in
1120 * widespread use, we will delete these interfaces. */
1122 /* If 'realdev_ofp_port' is nonzero, then this function configures 'ofport'
1123 * as a VLAN splinter port for VLAN 'vid', associated with the real device
1124 * that has OpenFlow port number 'realdev_ofp_port'.
1126 * If 'realdev_ofp_port' is zero, then this function deconfigures 'ofport'
1127 * as a VLAN splinter port.
1129 * This function should be NULL if a an implementation does not support
1131 int (*set_realdev)(struct ofport *ofport,
1132 uint16_t realdev_ofp_port, int vid);
1135 extern const struct ofproto_class ofproto_dpif_class;
1137 int ofproto_class_register(const struct ofproto_class *);
1138 int ofproto_class_unregister(const struct ofproto_class *);
1140 /* ofproto_flow_mod() returns this value if the flow_mod could not be processed
1141 * because it overlaps with an ongoing flow table operation that has not yet
1142 * completed. The caller should retry the operation later.
1144 * ofproto.c also uses this value internally for additional (similar) purposes.
1146 * This particular value is a good choice because it is large, so that it does
1147 * not collide with any errno value, but not large enough to collide with an
1148 * OFPERR_* value. */
1149 enum { OFPROTO_POSTPONE = 1 << 16 };
1150 BUILD_ASSERT_DECL(OFPROTO_POSTPONE < OFPERR_OFS);
1152 int ofproto_flow_mod(struct ofproto *, const struct ofputil_flow_mod *);
1153 void ofproto_add_flow(struct ofproto *, const struct cls_rule *,
1154 const union ofp_action *, size_t n_actions);
1155 bool ofproto_delete_flow(struct ofproto *, const struct cls_rule *);
1156 void ofproto_flush_flows(struct ofproto *);
1158 #endif /* ofproto/ofproto-provider.h */