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;
64 uint16_t max_ports; /* Max possible OpenFlow port num, plus one. */
67 struct oftable *tables;
70 /* OpenFlow connections. */
71 struct connmgr *connmgr;
73 /* Flow table operation tracking. */
74 int state; /* Internal state. */
75 struct list pending; /* List of "struct ofopgroup"s. */
76 unsigned int n_pending; /* list_size(&pending). */
77 struct hmap deletions; /* All OFOPERATION_DELETE "ofoperation"s. */
79 /* Flow table operation logging. */
80 int n_add, n_delete, n_modify; /* Number of unreported ops of each kind. */
81 long long int first_op, last_op; /* Range of times for unreported ops. */
82 long long int next_op_report; /* Time to report ops, or LLONG_MAX. */
83 long long int op_backoff; /* Earliest time to report ops again. */
85 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
87 * This is deprecated. It is only for compatibility with broken device
88 * drivers in old versions of Linux that do not properly support VLANs when
89 * VLAN devices are not used. When broken device drivers are no longer in
90 * widespread use, we will delete these interfaces. */
91 unsigned long int *vlan_bitmap; /* 4096-bit bitmap of in-use VLANs. */
92 bool vlans_changed; /* True if new VLANs are in use. */
93 int min_mtu; /* Current MTU of non-internal ports. */
96 void ofproto_init_tables(struct ofproto *, int n_tables);
97 void ofproto_init_max_ports(struct ofproto *, uint16_t max_ports);
99 struct ofproto *ofproto_lookup(const char *name);
100 struct ofport *ofproto_get_port(const struct ofproto *, uint16_t ofp_port);
102 /* An OpenFlow port within a "struct ofproto".
104 * With few exceptions, ofproto implementations may look at these fields but
105 * should not modify them. */
107 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
108 struct ofproto *ofproto; /* The ofproto that contains this port. */
109 struct netdev *netdev;
110 struct ofputil_phy_port pp;
111 uint16_t ofp_port; /* OpenFlow port number. */
112 unsigned int change_seq;
116 void ofproto_port_set_state(struct ofport *, enum ofputil_port_state);
119 OFTABLE_HIDDEN = 1 << 0, /* Hide from most OpenFlow operations. */
120 OFTABLE_READONLY = 1 << 1 /* Don't allow OpenFlow to change this table. */
123 /* A flow table within a "struct ofproto". */
125 enum oftable_flags flags;
126 struct classifier cls; /* Contains "struct rule"s. */
127 char *name; /* Table name exposed via OpenFlow, or NULL. */
129 /* Maximum number of flows or UINT_MAX if there is no limit besides any
130 * limit imposed by resource limitations. */
131 unsigned int max_flows;
133 /* These members determine the handling of an attempt to add a flow that
134 * would cause the table to have more than 'max_flows' flows.
136 * If 'eviction_fields' is NULL, overflows will be rejected with an error.
138 * If 'eviction_fields' is nonnull (regardless of whether n_eviction_fields
139 * is nonzero), an overflow will cause a flow to be removed. The flow to
140 * be removed is chosen to give fairness among groups distinguished by
141 * different values for the subfields within 'groups'. */
142 struct mf_subfield *eviction_fields;
143 size_t n_eviction_fields;
147 * When a flow is added that would cause the table to have more than
148 * 'max_flows' flows, and 'eviction_fields' is nonnull, these groups are
149 * used to decide which rule to evict: the rule is chosen from the eviction
150 * group that contains the greatest number of rules.*/
151 uint32_t eviction_group_id_basis;
152 struct hmap eviction_groups_by_id;
153 struct heap eviction_groups_by_size;
156 /* Assigns TABLE to each oftable, in turn, in OFPROTO.
158 * All parameters are evaluated multiple times. */
159 #define OFPROTO_FOR_EACH_TABLE(TABLE, OFPROTO) \
160 for ((TABLE) = (OFPROTO)->tables; \
161 (TABLE) < &(OFPROTO)->tables[(OFPROTO)->n_tables]; \
164 /* An OpenFlow flow within a "struct ofproto".
166 * With few exceptions, ofproto implementations may look at these fields but
167 * should not modify them. */
169 struct list ofproto_node; /* Owned by ofproto base code. */
170 struct ofproto *ofproto; /* The ofproto that contains this rule. */
171 struct cls_rule cr; /* In owning ofproto's classifier. */
173 struct ofoperation *pending; /* Operation now in progress, if nonnull. */
175 ovs_be64 flow_cookie; /* Controller-issued identifier. */
177 long long int created; /* Creation time. */
178 long long int modified; /* Time of last modification. */
179 long long int used; /* Last use; time created if never used. */
180 uint16_t hard_timeout; /* In seconds from ->modified. */
181 uint16_t idle_timeout; /* In seconds from ->used. */
182 uint8_t table_id; /* Index in ofproto's 'tables' array. */
183 bool send_flow_removed; /* Send a flow removed message? */
185 /* Eviction groups. */
186 bool evictable; /* If false, prevents eviction. */
187 struct heap_node evg_node; /* In eviction_group's "rules" heap. */
188 struct eviction_group *eviction_group; /* NULL if not in any group. */
190 struct ofpact *ofpacts; /* Sequence of "struct ofpacts". */
191 unsigned int ofpacts_len; /* Size of 'ofpacts', in bytes. */
194 enum nx_flow_monitor_flags monitor_flags;
195 uint64_t add_seqno; /* Sequence number when added. */
196 uint64_t modify_seqno; /* Sequence number when changed. */
199 static inline struct rule *
200 rule_from_cls_rule(const struct cls_rule *cls_rule)
202 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
205 void ofproto_rule_update_used(struct rule *, long long int used);
206 void ofproto_rule_expire(struct rule *, uint8_t reason);
207 void ofproto_rule_destroy(struct rule *);
209 bool ofproto_rule_has_out_port(const struct rule *, uint16_t out_port);
211 void ofoperation_complete(struct ofoperation *, enum ofperr);
212 struct rule *ofoperation_get_victim(struct ofoperation *);
214 bool ofoperation_has_out_port(const struct ofoperation *, uint16_t out_port);
216 bool ofproto_rule_is_hidden(const struct rule *);
218 /* ofproto class structure, to be defined by each ofproto implementation.
224 * These functions work primarily with three different kinds of data
227 * - "struct ofproto", which represents an OpenFlow switch.
229 * - "struct ofport", which represents a port within an ofproto.
231 * - "struct rule", which represents an OpenFlow flow within an ofproto.
233 * Each of these data structures contains all of the implementation-independent
234 * generic state for the respective concept, called the "base" state. None of
235 * them contains any extra space for ofproto implementations to use. Instead,
236 * each implementation is expected to declare its own data structure that
237 * contains an instance of the generic data structure plus additional
238 * implementation-specific members, called the "derived" state. The
239 * implementation can use casts or (preferably) the CONTAINER_OF macro to
240 * obtain access to derived state given only a pointer to the embedded generic
247 * Four stylized functions accompany each of these data structures:
249 * "alloc" "construct" "destruct" "dealloc"
250 * ------------ ---------------- --------------- --------------
251 * ofproto ->alloc ->construct ->destruct ->dealloc
252 * ofport ->port_alloc ->port_construct ->port_destruct ->port_dealloc
253 * rule ->rule_alloc ->rule_construct ->rule_destruct ->rule_dealloc
255 * Any instance of a given data structure goes through the following life
258 * 1. The client calls the "alloc" function to obtain raw memory. If "alloc"
259 * fails, skip all the other steps.
261 * 2. The client initializes all of the data structure's base state. If this
262 * fails, skip to step 7.
264 * 3. The client calls the "construct" function. The implementation
265 * initializes derived state. It may refer to the already-initialized
266 * base state. If "construct" fails, skip to step 6.
268 * 4. The data structure is now initialized and in use.
270 * 5. When the data structure is no longer needed, the client calls the
271 * "destruct" function. The implementation uninitializes derived state.
272 * The base state has not been uninitialized yet, so the implementation
273 * may still refer to it.
275 * 6. The client uninitializes all of the data structure's base state.
277 * 7. The client calls the "dealloc" to free the raw memory. The
278 * implementation must not refer to base or derived state in the data
279 * structure, because it has already been uninitialized.
281 * Each "alloc" function allocates and returns a new instance of the respective
282 * data structure. The "alloc" function is not given any information about the
283 * use of the new data structure, so it cannot perform much initialization.
284 * Its purpose is just to ensure that the new data structure has enough room
285 * for base and derived state. It may return a null pointer if memory is not
286 * available, in which case none of the other functions is called.
288 * Each "construct" function initializes derived state in its respective data
289 * structure. When "construct" is called, all of the base state has already
290 * been initialized, so the "construct" function may refer to it. The
291 * "construct" function is allowed to fail, in which case the client calls the
292 * "dealloc" function (but not the "destruct" function).
294 * Each "destruct" function uninitializes and frees derived state in its
295 * respective data structure. When "destruct" is called, the base state has
296 * not yet been uninitialized, so the "destruct" function may refer to it. The
297 * "destruct" function is not allowed to fail.
299 * Each "dealloc" function frees raw memory that was allocated by the the
300 * "alloc" function. The memory's base and derived members might not have ever
301 * been initialized (but if "construct" returned successfully, then it has been
302 * "destruct"ed already). The "dealloc" function is not allowed to fail.
308 * Most of these functions return 0 if they are successful or a positive error
309 * code on failure. Depending on the function, valid error codes are either
310 * errno values or OFPERR_* OpenFlow error codes.
312 * Most of these functions are expected to execute synchronously, that is, to
313 * block as necessary to obtain a result. Thus, these functions may return
314 * EAGAIN (or EWOULDBLOCK or EINPROGRESS) only where the function descriptions
315 * explicitly say those errors are a possibility. We may relax this
316 * requirement in the future if and when we encounter performance problems. */
317 struct ofproto_class {
318 /* ## ----------------- ## */
319 /* ## Factory Functions ## */
320 /* ## ----------------- ## */
322 /* Enumerates the types of all support ofproto types into 'types'. The
323 * caller has already initialized 'types' and other ofproto classes might
324 * already have added names to it. */
325 void (*enumerate_types)(struct sset *types);
327 /* Enumerates the names of all existing datapath of the specified 'type'
328 * into 'names' 'all_dps'. The caller has already initialized 'names' as
331 * 'type' is one of the types enumerated by ->enumerate_types().
333 * Returns 0 if successful, otherwise a positive errno value.
335 int (*enumerate_names)(const char *type, struct sset *names);
337 /* Deletes the datapath with the specified 'type' and 'name'. The caller
338 * should have closed any open ofproto with this 'type' and 'name'; this
339 * function is allowed to fail if that is not the case.
341 * 'type' is one of the types enumerated by ->enumerate_types().
342 * 'name' is one of the names enumerated by ->enumerate_names() for 'type'.
344 * Returns 0 if successful, otherwise a positive errno value.
346 int (*del)(const char *type, const char *name);
348 /* ## --------------------------- ## */
349 /* ## Top-Level ofproto Functions ## */
350 /* ## --------------------------- ## */
352 /* Life-cycle functions for an "ofproto" (see "Life Cycle" above).
358 * ->construct() should not modify any base members of the ofproto. The
359 * client will initialize the ofproto's 'ports' and 'tables' members after
360 * construction is complete.
362 * When ->construct() is called, the client does not yet know how many flow
363 * tables the datapath supports, so ofproto->n_tables will be 0 and
364 * ofproto->tables will be NULL. ->construct() should call
365 * ofproto_init_tables() to allocate and initialize ofproto->n_tables and
366 * ofproto->tables. Each flow table will be initially empty, so
367 * ->construct() should delete flows from the underlying datapath, if
368 * necessary, rather than populating the tables.
370 * If the ofproto knows the maximum port number that the datapath can have,
371 * then it can call ofproto_init_max_ports(). If it does so, then the
372 * client will ensure that the actions it allows to be used through
373 * OpenFlow do not refer to ports above that maximum number.
375 * Only one ofproto instance needs to be supported for any given datapath.
376 * If a datapath is already open as part of one "ofproto", then another
377 * attempt to "construct" the same datapath as part of another ofproto is
378 * allowed to fail with an error.
380 * ->construct() returns 0 if successful, otherwise a positive errno
387 * If 'ofproto' has any pending asynchronous operations, ->destruct()
388 * must complete all of them by calling ofoperation_complete().
390 * ->destruct() must also destroy all remaining rules in the ofproto's
391 * tables, by passing each remaining rule to ofproto_rule_destroy(). The
392 * client will destroy the flow tables themselves after ->destruct()
395 struct ofproto *(*alloc)(void);
396 int (*construct)(struct ofproto *ofproto);
397 void (*destruct)(struct ofproto *ofproto);
398 void (*dealloc)(struct ofproto *ofproto);
400 /* Performs any periodic activity required by 'ofproto'. It should:
402 * - Call connmgr_send_packet_in() for each received packet that missed
403 * in the OpenFlow flow table or that had a OFPP_CONTROLLER output
406 * - Call ofproto_rule_expire() for each OpenFlow flow that has reached
407 * its hard_timeout or idle_timeout, to expire the flow.
409 * (But rules that are part of a pending operation, e.g. rules for
410 * which ->pending is true, may not expire.)
412 * Returns 0 if successful, otherwise a positive errno value. */
413 int (*run)(struct ofproto *ofproto);
415 /* Performs periodic activity required by 'ofproto' that needs to be done
416 * with the least possible latency.
418 * This is run multiple times per main loop. An ofproto provider may
419 * implement it or not, according to whether it provides a performance
420 * boost for that ofproto implementation. */
421 int (*run_fast)(struct ofproto *ofproto);
423 /* Causes the poll loop to wake up when 'ofproto''s 'run' function needs to
424 * be called, e.g. by calling the timer or fd waiting functions in
426 void (*wait)(struct ofproto *ofproto);
428 /* Adds some memory usage statistics for the implementation of 'ofproto'
429 * into 'usage', for use with memory_report().
431 * This function is optional. */
432 void (*get_memory_usage)(const struct ofproto *ofproto,
433 struct simap *usage);
435 /* Every "struct rule" in 'ofproto' is about to be deleted, one by one.
436 * This function may prepare for that, for example by clearing state in
437 * advance. It should *not* actually delete any "struct rule"s from
438 * 'ofproto', only prepare for it.
440 * This function is optional; it's really just for optimization in case
441 * it's cheaper to delete all the flows from your hardware in a single pass
442 * than to do it one by one. */
443 void (*flush)(struct ofproto *ofproto);
445 /* Helper for the OpenFlow OFPT_FEATURES_REQUEST request.
447 * The implementation should store true in '*arp_match_ip' if the switch
448 * supports matching IP addresses inside ARP requests and replies, false
451 * The implementation should store in '*actions' a bitmap of the supported
452 * OpenFlow actions. Vendor actions are not included in '*actions'. */
453 void (*get_features)(struct ofproto *ofproto,
455 enum ofputil_action_bitmap *actions);
457 /* Helper for the OpenFlow OFPST_TABLE statistics request.
459 * The 'ots' array contains 'ofproto->n_tables' elements. Each element is
462 * - 'table_id' to the array index.
464 * - 'name' to "table#" where # is the table ID.
466 * - 'wildcards' to OFPFW10_ALL.
468 * - 'max_entries' to 1,000,000.
470 * - 'active_count' to the classifier_count() for the table.
472 * - 'lookup_count' and 'matched_count' to 0.
474 * The implementation should update any members in each element for which
475 * it has better values:
477 * - 'name' to a more meaningful name.
479 * - 'wildcards' to the set of wildcards actually supported by the table
480 * (if it doesn't support all OpenFlow wildcards).
482 * - 'max_entries' to the maximum number of flows actually supported by
485 * - 'lookup_count' to the number of packets looked up in this flow table
488 * - 'matched_count' to the number of packets looked up in this flow
489 * table so far that matched one of the flow entries.
491 * Keep in mind that all of the members of struct ofp10_table_stats are in
492 * network byte order.
494 void (*get_tables)(struct ofproto *ofproto, struct ofp10_table_stats *ots);
496 /* ## ---------------- ## */
497 /* ## ofport Functions ## */
498 /* ## ---------------- ## */
500 /* Life-cycle functions for a "struct ofport" (see "Life Cycle" above).
502 * ->port_construct() should not modify any base members of the ofport.
504 * ofports are managed by the base ofproto code. The ofproto
505 * implementation should only create and destroy them in response to calls
506 * to these functions. The base ofproto code will create and destroy
507 * ofports in the following situations:
509 * - Just after the ->construct() function is called, the base ofproto
510 * iterates over all of the implementation's ports, using
511 * ->port_dump_start() and related functions, and constructs an ofport
512 * for each dumped port.
514 * - If ->port_poll() reports that a specific port has changed, then the
515 * base ofproto will query that port with ->port_query_by_name() and
516 * construct or destruct ofports as necessary to reflect the updated
519 * - If ->port_poll() returns ENOBUFS to report an unspecified port set
520 * change, then the base ofproto will iterate over all of the
521 * implementation's ports, in the same way as at ofproto
522 * initialization, and construct and destruct ofports to reflect all of
525 * ->port_construct() returns 0 if successful, otherwise a positive errno
528 struct ofport *(*port_alloc)(void);
529 int (*port_construct)(struct ofport *ofport);
530 void (*port_destruct)(struct ofport *ofport);
531 void (*port_dealloc)(struct ofport *ofport);
533 /* Called after 'ofport->netdev' is replaced by a new netdev object. If
534 * the ofproto implementation uses the ofport's netdev internally, then it
535 * should switch to using the new one. The old one has been closed.
537 * An ofproto implementation that doesn't need to do anything in this
538 * function may use a null pointer. */
539 void (*port_modified)(struct ofport *ofport);
541 /* Called after an OpenFlow request changes a port's configuration.
542 * 'ofport->pp.config' contains the new configuration. 'old_config'
543 * contains the previous configuration.
545 * The caller implements OFPUTIL_PC_PORT_DOWN using netdev functions to
546 * turn NETDEV_UP on and off, so this function doesn't have to do anything
547 * for that bit (and it won't be called if that is the only bit that
549 void (*port_reconfigured)(struct ofport *ofport,
550 enum ofputil_port_config old_config);
552 /* Looks up a port named 'devname' in 'ofproto'. On success, initializes
553 * '*port' appropriately.
555 * The caller owns the data in 'port' and must free it with
556 * ofproto_port_destroy() when it is no longer needed. */
557 int (*port_query_by_name)(const struct ofproto *ofproto,
558 const char *devname, struct ofproto_port *port);
560 /* Attempts to add 'netdev' as a port on 'ofproto'. Returns 0 if
561 * successful, otherwise a positive errno value. If successful, sets
562 * '*ofp_portp' to the new port's port number.
564 * It doesn't matter whether the new port will be returned by a later call
565 * to ->port_poll(); the implementation may do whatever is more
567 int (*port_add)(struct ofproto *ofproto, struct netdev *netdev,
568 uint16_t *ofp_portp);
570 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'. Returns
571 * 0 if successful, otherwise a positive errno value.
573 * It doesn't matter whether the new port will be returned by a later call
574 * to ->port_poll(); the implementation may do whatever is more
576 int (*port_del)(struct ofproto *ofproto, uint16_t ofp_port);
579 int (*port_get_stats)(const struct ofport *port,
580 struct netdev_stats *stats);
582 /* Port iteration functions.
584 * The client might not be entirely in control of the ports within an
585 * ofproto. Some hardware implementations, for example, might have a fixed
586 * set of ports in a datapath, and the Linux datapath allows the system
587 * administrator to externally add and remove ports with ovs-dpctl. For
588 * this reason, the client needs a way to iterate through all the ports
589 * that are actually in a datapath. These functions provide that
592 * The 'state' pointer provides the implementation a place to
593 * keep track of its position. Its format is opaque to the caller.
595 * The ofproto provider retains ownership of the data that it stores into
596 * ->port_dump_next()'s 'port' argument. The data must remain valid until
597 * at least the next call to ->port_dump_next() or ->port_dump_done() for
598 * 'state'. The caller will not modify or free it.
603 * ->port_dump_start() attempts to begin dumping the ports in 'ofproto'.
604 * On success, it should return 0 and initialize '*statep' with any data
605 * needed for iteration. On failure, returns a positive errno value, and
606 * the client will not call ->port_dump_next() or ->port_dump_done().
608 * ->port_dump_next() attempts to retrieve another port from 'ofproto' for
609 * 'state'. If there is another port, it should store the port's
610 * information into 'port' and return 0. It should return EOF if all ports
611 * have already been iterated. Otherwise, on error, it should return a
612 * positive errno value. This function will not be called again once it
613 * returns nonzero once for a given iteration (but the 'port_dump_done'
614 * function will be called afterward).
616 * ->port_dump_done() allows the implementation to release resources used
617 * for iteration. The caller might decide to stop iteration in the middle
618 * by calling this function before ->port_dump_next() returns nonzero.
626 * error = ofproto->ofproto_class->port_dump_start(ofproto, &state);
629 * struct ofproto_port port;
631 * error = ofproto->ofproto_class->port_dump_next(
632 * ofproto, state, &port);
636 * // Do something with 'port' here (without modifying or freeing
637 * // any of its data).
639 * ofproto->ofproto_class->port_dump_done(ofproto, state);
641 * // 'error' is now EOF (success) or a positive errno value (failure).
643 int (*port_dump_start)(const struct ofproto *ofproto, void **statep);
644 int (*port_dump_next)(const struct ofproto *ofproto, void *state,
645 struct ofproto_port *port);
646 int (*port_dump_done)(const struct ofproto *ofproto, void *state);
648 /* Polls for changes in the set of ports in 'ofproto'. If the set of ports
649 * in 'ofproto' has changed, then this function should do one of the
652 * - Preferably: store the name of the device that was added to or deleted
653 * from 'ofproto' in '*devnamep' and return 0. The caller is responsible
654 * for freeing '*devnamep' (with free()) when it no longer needs it.
656 * - Alternatively: return ENOBUFS, without indicating the device that was
659 * Occasional 'false positives', in which the function returns 0 while
660 * indicating a device that was not actually added or deleted or returns
661 * ENOBUFS without any change, are acceptable.
663 * The purpose of 'port_poll' is to let 'ofproto' know about changes made
664 * externally to the 'ofproto' object, e.g. by a system administrator via
665 * ovs-dpctl. Therefore, it's OK, and even preferable, for port_poll() to
666 * not report changes made through calls to 'port_add' or 'port_del' on the
667 * same 'ofproto' object. (But it's OK for it to report them too, just
668 * slightly less efficient.)
670 * If the set of ports in 'ofproto' has not changed, returns EAGAIN. May
671 * also return other positive errno values to indicate that something has
674 * If the set of ports in a datapath is fixed, or if the only way that the
675 * set of ports in a datapath can change is through ->port_add() and
676 * ->port_del(), then this function may be a null pointer.
678 int (*port_poll)(const struct ofproto *ofproto, char **devnamep);
680 /* Arranges for the poll loop to wake up when ->port_poll() will return a
681 * value other than EAGAIN.
683 * If the set of ports in a datapath is fixed, or if the only way that the
684 * set of ports in a datapath can change is through ->port_add() and
685 * ->port_del(), or if the poll loop will always wake up anyway when
686 * ->port_poll() will return a value other than EAGAIN, then this function
687 * may be a null pointer.
689 void (*port_poll_wait)(const struct ofproto *ofproto);
691 /* Checks the status of LACP negotiation for 'port'. Returns 1 if LACP
692 * partner information for 'port' is up-to-date, 0 if LACP partner
693 * information is not current (generally indicating a connectivity
694 * problem), or -1 if LACP is not enabled on 'port'.
696 * This function may be a null pointer if the ofproto implementation does
697 * not support LACP. */
698 int (*port_is_lacp_current)(const struct ofport *port);
700 /* ## ----------------------- ## */
701 /* ## OpenFlow Rule Functions ## */
702 /* ## ----------------------- ## */
706 /* Chooses an appropriate table for 'cls_rule' within 'ofproto'. On
707 * success, stores the table ID into '*table_idp' and returns 0. On
708 * failure, returns an OpenFlow error code.
710 * The choice of table should be a function of 'cls_rule' and 'ofproto''s
711 * datapath capabilities. It should not depend on the flows already in
712 * 'ofproto''s flow tables. Failure implies that an OpenFlow rule with
713 * 'cls_rule' as its matching condition can never be inserted into
714 * 'ofproto', even starting from an empty flow table.
716 * If multiple tables are candidates for inserting the flow, the function
717 * should choose one arbitrarily (but deterministically).
719 * If this function is NULL then table 0 is always chosen. */
720 enum ofperr (*rule_choose_table)(const struct ofproto *ofproto,
721 const struct cls_rule *cls_rule,
724 /* Life-cycle functions for a "struct rule" (see "Life Cycle" above).
727 * Asynchronous Operation Support
728 * ==============================
730 * The life-cycle operations on rules can operate asynchronously, meaning
731 * that ->rule_construct() and ->rule_destruct() only need to initiate
732 * their respective operations and do not need to wait for them to complete
733 * before they return. ->rule_modify_actions() also operates
736 * An ofproto implementation reports the success or failure of an
737 * asynchronous operation on a rule using the rule's 'pending' member,
738 * which points to a opaque "struct ofoperation" that represents the
739 * ongoing opreation. When the operation completes, the ofproto
740 * implementation calls ofoperation_complete(), passing the ofoperation and
741 * an error indication.
743 * Only the following contexts may call ofoperation_complete():
745 * - The function called to initiate the operation,
746 * e.g. ->rule_construct() or ->rule_destruct(). This is the best
747 * choice if the operation completes quickly.
749 * - The implementation's ->run() function.
751 * - The implementation's ->destruct() function.
753 * The ofproto base code updates the flow table optimistically, assuming
754 * that the operation will probably succeed:
756 * - ofproto adds or replaces the rule in the flow table before calling
757 * ->rule_construct().
759 * - ofproto updates the rule's actions before calling
760 * ->rule_modify_actions().
762 * - ofproto removes the rule before calling ->rule_destruct().
764 * With one exception, when an asynchronous operation completes with an
765 * error, ofoperation_complete() backs out the already applied changes:
767 * - If adding or replacing a rule in the flow table fails, ofproto
768 * removes the new rule or restores the original rule.
770 * - If modifying a rule's actions fails, ofproto restores the original
773 * - Removing a rule is not allowed to fail. It must always succeed.
775 * The ofproto base code serializes operations: if any operation is in
776 * progress on a given rule, ofproto postpones initiating any new operation
777 * on that rule until the pending operation completes. Therefore, every
778 * operation must eventually complete through a call to
779 * ofoperation_complete() to avoid delaying new operations indefinitely
780 * (including any OpenFlow request that affects the rule in question, even
781 * just to query its statistics).
787 * When ->rule_construct() is called, the caller has already inserted
788 * 'rule' into 'rule->ofproto''s flow table numbered 'rule->table_id'.
789 * There are two cases:
791 * - 'rule' is a new rule in its flow table. In this case,
792 * ofoperation_get_victim(rule) returns NULL.
794 * - 'rule' is replacing an existing rule in its flow table that had the
795 * same matching criteria and priority. In this case,
796 * ofoperation_get_victim(rule) returns the rule being replaced (the
799 * ->rule_construct() should set the following in motion:
801 * - Validate that the matching rule in 'rule->cr' is supported by the
802 * datapath. For example, if the rule's table does not support
803 * registers, then it is an error if 'rule->cr' does not wildcard all
806 * - Validate that the datapath can correctly implement 'rule->ofpacts'.
808 * - If the rule is valid, update the datapath flow table, adding the new
809 * rule or replacing the existing one.
811 * - If 'rule' is replacing an existing rule, uninitialize any derived
812 * state for the victim rule, as in step 5 in the "Life Cycle"
815 * (On failure, the ofproto code will roll back the insertion from the flow
816 * table, either removing 'rule' or replacing it by the victim rule if
819 * ->rule_construct() must act in one of the following ways:
821 * - If it succeeds, it must call ofoperation_complete() and return 0.
823 * - If it fails, it must act in one of the following ways:
825 * * Call ofoperation_complete() and return 0.
827 * * Return an OpenFlow error code. (Do not call
828 * ofoperation_complete() in this case.)
830 * Either way, ->rule_destruct() will not be called for 'rule', but
831 * ->rule_dealloc() will be.
833 * - If the operation is only partially complete, then it must return 0.
834 * Later, when the operation is complete, the ->run() or ->destruct()
835 * function must call ofoperation_complete() to report success or
838 * ->rule_construct() should not modify any base members of struct rule.
844 * When ->rule_destruct() is called, the caller has already removed 'rule'
845 * from 'rule->ofproto''s flow table. ->rule_destruct() should set in
846 * motion removing 'rule' from the datapath flow table. If removal
847 * completes synchronously, it should call ofoperation_complete().
848 * Otherwise, the ->run() or ->destruct() function must later call
849 * ofoperation_complete() after the operation completes.
851 * Rule destruction must not fail. */
852 struct rule *(*rule_alloc)(void);
853 enum ofperr (*rule_construct)(struct rule *rule);
854 void (*rule_destruct)(struct rule *rule);
855 void (*rule_dealloc)(struct rule *rule);
857 /* Obtains statistics for 'rule', storing the number of packets that have
858 * matched it in '*packet_count' and the number of bytes in those packets
859 * in '*byte_count'. UINT64_MAX indicates that the packet count or byte
860 * count is unknown. */
861 void (*rule_get_stats)(struct rule *rule, uint64_t *packet_count,
862 uint64_t *byte_count);
864 /* Applies the actions in 'rule' to 'packet'. (This implements sending
865 * buffered packets for OpenFlow OFPT_FLOW_MOD commands.)
867 * Takes ownership of 'packet' (so it should eventually free it, with
870 * 'flow' reflects the flow information for 'packet'. All of the
871 * information in 'flow' is extracted from 'packet', except for
872 * flow->tun_id and flow->in_port, which are assigned the correct values
873 * for the incoming packet. The register values are zeroed. 'packet''s
874 * header pointers (e.g. packet->l3) are appropriately initialized.
876 * The implementation should add the statistics for 'packet' into 'rule'.
878 * Returns 0 if successful, otherwise an OpenFlow error code. */
879 enum ofperr (*rule_execute)(struct rule *rule, const struct flow *flow,
880 struct ofpbuf *packet);
882 /* When ->rule_modify_actions() is called, the caller has already replaced
883 * the OpenFlow actions in 'rule' by a new set. (The original actions are
884 * in rule->pending->actions.)
886 * ->rule_modify_actions() should set the following in motion:
888 * - Validate that the datapath can correctly implement the actions now
891 * - Update the datapath flow table with the new actions.
893 * If the operation synchronously completes, ->rule_modify_actions() may
894 * call ofoperation_complete() before it returns. Otherwise, ->run()
895 * should call ofoperation_complete() later, after the operation does
898 * If the operation fails, then the base ofproto code will restore the
899 * original 'actions' and 'n_actions' of 'rule'.
901 * ->rule_modify_actions() should not modify any base members of struct
903 void (*rule_modify_actions)(struct rule *rule);
905 /* Changes the OpenFlow IP fragment handling policy to 'frag_handling',
906 * which takes one of the following values, with the corresponding
909 * - OFPC_FRAG_NORMAL: The switch should treat IP fragments the same way
910 * as other packets, omitting TCP and UDP port numbers (always setting
913 * - OFPC_FRAG_DROP: The switch should drop all IP fragments without
914 * passing them through the flow table.
916 * - OFPC_FRAG_REASM: The switch should reassemble IP fragments before
917 * passing packets through the flow table.
919 * - OFPC_FRAG_NX_MATCH (a Nicira extension): Similar to OFPC_FRAG_NORMAL,
920 * except that TCP and UDP port numbers should be included in fragments
923 * Implementations are not required to support every mode.
924 * OFPC_FRAG_NORMAL is the default mode when an ofproto is created.
926 * At the time of the call to ->set_frag_handling(), the current mode is
927 * available in 'ofproto->frag_handling'. ->set_frag_handling() returns
928 * true if the requested mode was set, false if it is not supported.
930 * Upon successful return, the caller changes 'ofproto->frag_handling' to
931 * reflect the new mode.
933 bool (*set_frag_handling)(struct ofproto *ofproto,
934 enum ofp_config_flags frag_handling);
936 /* Implements the OpenFlow OFPT_PACKET_OUT command. The datapath should
937 * execute the 'ofpacts_len' bytes of "struct ofpacts" in 'ofpacts'.
939 * The caller retains ownership of 'packet' and of 'ofpacts', so
940 * ->packet_out() should not modify or free them.
942 * This function must validate that it can correctly implement 'ofpacts'.
943 * If not, then it should return an OpenFlow error code.
945 * 'flow' reflects the flow information for 'packet'. All of the
946 * information in 'flow' is extracted from 'packet', except for
947 * flow->in_port (see below). flow->tun_id and its register values are
950 * flow->in_port comes from the OpenFlow OFPT_PACKET_OUT message. The
951 * implementation should reject invalid flow->in_port values by returning
952 * OFPERR_NXBRC_BAD_IN_PORT. (If the implementation called
953 * ofproto_init_max_ports(), then the client will reject these ports
954 * itself.) For consistency, the implementation should consider valid for
955 * flow->in_port any value that could possibly be seen in a packet that it
956 * passes to connmgr_send_packet_in(). Ideally, even an implementation
957 * that never generates packet-ins (e.g. due to hardware limitations)
958 * should still allow flow->in_port values for every possible physical port
959 * and OFPP_LOCAL. The only virtual ports (those above OFPP_MAX) that the
960 * caller will ever pass in as flow->in_port, other than OFPP_LOCAL, are
961 * OFPP_NONE and OFPP_CONTROLLER. The implementation should allow both of
962 * these, treating each of them as packets generated by the controller as
963 * opposed to packets originating from some switch port.
965 * (Ordinarily the only effect of flow->in_port is on output actions that
966 * involve the input port, such as actions that output to OFPP_IN_PORT,
967 * OFPP_FLOOD, or OFPP_ALL. flow->in_port can also affect Nicira extension
968 * "resubmit" actions.)
970 * 'packet' is not matched against the OpenFlow flow table, so its
971 * statistics should not be included in OpenFlow flow statistics.
973 * Returns 0 if successful, otherwise an OpenFlow error code. */
974 enum ofperr (*packet_out)(struct ofproto *ofproto, struct ofpbuf *packet,
975 const struct flow *flow,
976 const struct ofpact *ofpacts,
979 /* ## ------------------------- ## */
980 /* ## OFPP_NORMAL configuration ## */
981 /* ## ------------------------- ## */
983 /* Configures NetFlow on 'ofproto' according to the options in
984 * 'netflow_options', or turns off NetFlow if 'netflow_options' is NULL.
986 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
987 * NetFlow, as does a null pointer. */
988 int (*set_netflow)(struct ofproto *ofproto,
989 const struct netflow_options *netflow_options);
991 void (*get_netflow_ids)(const struct ofproto *ofproto,
992 uint8_t *engine_type, uint8_t *engine_id);
994 /* Configures sFlow on 'ofproto' according to the options in
995 * 'sflow_options', or turns off sFlow if 'sflow_options' is NULL.
997 * EOPNOTSUPP as a return value indicates that 'ofproto' does not support
998 * sFlow, as does a null pointer. */
999 int (*set_sflow)(struct ofproto *ofproto,
1000 const struct ofproto_sflow_options *sflow_options);
1002 /* Configures connectivity fault management on 'ofport'.
1004 * If 'cfm_settings' is nonnull, configures CFM according to its members.
1006 * If 'cfm_settings' is null, removes any connectivity fault management
1007 * configuration from 'ofport'.
1009 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1010 * support CFM, as does a null pointer. */
1011 int (*set_cfm)(struct ofport *ofport, const struct cfm_settings *s);
1013 /* Checks the fault status of CFM configured on 'ofport'. Returns a
1014 * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
1015 * indicating a connectivity problem). Returns zero if CFM is not faulted,
1016 * and -1 if CFM is not enabled on 'port'.
1018 * This function may be a null pointer if the ofproto implementation does
1019 * not support CFM. */
1020 int (*get_cfm_fault)(const struct ofport *ofport);
1022 /* Check the operational status reported by the remote CFM endpoint of
1023 * 'ofp_port' Returns 1 if operationally up, 0 if operationally down, and
1024 * -1 if CFM is not enabled on 'ofp_port' or does not support operational
1027 * This function may be a null pointer if the ofproto implementation does
1028 * not support CFM. */
1029 int (*get_cfm_opup)(const struct ofport *ofport);
1031 /* Gets the MPIDs of the remote maintenance points broadcasting to
1032 * 'ofport'. Populates 'rmps' with a provider owned array of MPIDs, and
1033 * 'n_rmps' with the number of MPIDs in 'rmps'. Returns a number less than
1034 * 0 if CFM is not enabled of 'ofport'.
1036 * This function may be a null pointer if the ofproto implementation does
1037 * not support CFM. */
1038 int (*get_cfm_remote_mpids)(const struct ofport *ofport,
1039 const uint64_t **rmps, size_t *n_rmps);
1041 /* Checks the health of CFM configured on 'ofport'. Returns an integer
1042 * to indicate the health percentage of the 'ofport' which is an average of
1043 * the health of all the remote_mps. Returns an integer between 0 and 100
1044 * where 0 means that the 'ofport' is very unhealthy and 100 means the
1045 * 'ofport' is perfectly healthy. Returns -1 if CFM is not enabled on
1046 * 'port' or if the number of remote_mpids is > 1.
1048 * This function may be a null pointer if the ofproto implementation does
1049 * not support CFM. */
1050 int (*get_cfm_health)(const struct ofport *ofport);
1052 /* Configures spanning tree protocol (STP) on 'ofproto' using the
1053 * settings defined in 's'.
1055 * If 's' is nonnull, configures STP according to its members.
1057 * If 's' is null, removes any STP configuration from 'ofproto'.
1059 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1060 * support STP, as does a null pointer. */
1061 int (*set_stp)(struct ofproto *ofproto,
1062 const struct ofproto_stp_settings *s);
1064 /* Retrieves state of spanning tree protocol (STP) on 'ofproto'.
1066 * Stores STP state for 'ofproto' in 's'. If the 'enabled' member
1067 * is false, the other member values are not meaningful.
1069 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1070 * support STP, as does a null pointer. */
1071 int (*get_stp_status)(struct ofproto *ofproto,
1072 struct ofproto_stp_status *s);
1074 /* Configures spanning tree protocol (STP) on 'ofport' using the
1075 * settings defined in 's'.
1077 * If 's' is nonnull, configures STP according to its members. The
1078 * caller is responsible for assigning STP port numbers (using the
1079 * 'port_num' member in the range of 1 through 255, inclusive) and
1080 * ensuring there are no duplicates.
1082 * If 's' is null, removes any STP configuration from 'ofport'.
1084 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1085 * support STP, as does a null pointer. */
1086 int (*set_stp_port)(struct ofport *ofport,
1087 const struct ofproto_port_stp_settings *s);
1089 /* Retrieves spanning tree protocol (STP) port status of 'ofport'.
1091 * Stores STP state for 'ofport' in 's'. If the 'enabled' member is
1092 * false, the other member values are not meaningful.
1094 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1095 * support STP, as does a null pointer. */
1096 int (*get_stp_port_status)(struct ofport *ofport,
1097 struct ofproto_port_stp_status *s);
1099 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
1100 * 'queues' attached to 'ofport'. This data is not intended to be
1101 * sufficient to implement QoS. Instead, providers may use this
1102 * information to implement features which require knowledge of what queues
1103 * exist on a port, and some basic information about them.
1105 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1106 * support QoS, as does a null pointer. */
1107 int (*set_queues)(struct ofport *ofport,
1108 const struct ofproto_port_queue *queues, size_t n_qdscp);
1110 /* If 's' is nonnull, this function registers a "bundle" associated with
1111 * client data pointer 'aux' in 'ofproto'. A bundle is the same concept as
1112 * a Port in OVSDB, that is, it consists of one or more "slave" devices
1113 * (Interfaces, in OVSDB) along with VLAN and LACP configuration and, if
1114 * there is more than one slave, a bonding configuration. If 'aux' is
1115 * already registered then this function updates its configuration to 's'.
1116 * Otherwise, this function registers a new bundle.
1118 * If 's' is NULL, this function unregisters the bundle registered on
1119 * 'ofproto' associated with client data pointer 'aux'. If no such bundle
1120 * has been registered, this has no effect.
1122 * This function affects only the behavior of the NXAST_AUTOPATH action and
1123 * output to the OFPP_NORMAL port. An implementation that does not support
1124 * it at all may set it to NULL or return EOPNOTSUPP. An implementation
1125 * that supports only a subset of the functionality should implement what
1126 * it can and return 0. */
1127 int (*bundle_set)(struct ofproto *ofproto, void *aux,
1128 const struct ofproto_bundle_settings *s);
1130 /* If 'port' is part of any bundle, removes it from that bundle. If the
1131 * bundle now has no ports, deletes the bundle. If the bundle now has only
1132 * one port, deconfigures the bundle's bonding configuration. */
1133 void (*bundle_remove)(struct ofport *ofport);
1135 /* If 's' is nonnull, this function registers a mirror associated with
1136 * client data pointer 'aux' in 'ofproto'. A mirror is the same concept as
1137 * a Mirror in OVSDB. If 'aux' is already registered then this function
1138 * updates its configuration to 's'. Otherwise, this function registers a
1141 * If 's' is NULL, this function unregisters the mirror registered on
1142 * 'ofproto' associated with client data pointer 'aux'. If no such mirror
1143 * has been registered, this has no effect.
1145 * An implementation that does not support mirroring at all may set
1146 * it to NULL or return EOPNOTSUPP. An implementation that supports
1147 * only a subset of the functionality should implement what it can
1149 int (*mirror_set)(struct ofproto *ofproto, void *aux,
1150 const struct ofproto_mirror_settings *s);
1152 /* Retrieves statistics from mirror associated with client data
1153 * pointer 'aux' in 'ofproto'. Stores packet and byte counts in
1154 * 'packets' and 'bytes', respectively. If a particular counter is
1155 * not supported, the appropriate argument is set to UINT64_MAX.
1157 * EOPNOTSUPP as a return value indicates that this ofproto_class does not
1158 * support retrieving mirror statistics. */
1159 int (*mirror_get_stats)(struct ofproto *ofproto, void *aux,
1160 uint64_t *packets, uint64_t *bytes);
1162 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs
1163 * on which all packets are flooded, instead of using MAC learning. If
1164 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1166 * This function affects only the behavior of the OFPP_NORMAL action. An
1167 * implementation that does not support it may set it to NULL or return
1169 int (*set_flood_vlans)(struct ofproto *ofproto,
1170 unsigned long *flood_vlans);
1172 /* Returns true if 'aux' is a registered bundle that is currently in use as
1173 * the output for a mirror. */
1174 bool (*is_mirror_output_bundle)(const struct ofproto *ofproto, void *aux);
1176 /* When the configuration option of forward_bpdu changes, this function
1177 * will be invoked. */
1178 void (*forward_bpdu_changed)(struct ofproto *ofproto);
1180 /* Sets the MAC aging timeout for the OFPP_NORMAL action to 'idle_time',
1182 void (*set_mac_idle_time)(struct ofproto *ofproto, unsigned int idle_time);
1184 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
1186 * This is deprecated. It is only for compatibility with broken device drivers
1187 * in old versions of Linux that do not properly support VLANs when VLAN
1188 * devices are not used. When broken device drivers are no longer in
1189 * widespread use, we will delete these interfaces. */
1191 /* If 'realdev_ofp_port' is nonzero, then this function configures 'ofport'
1192 * as a VLAN splinter port for VLAN 'vid', associated with the real device
1193 * that has OpenFlow port number 'realdev_ofp_port'.
1195 * If 'realdev_ofp_port' is zero, then this function deconfigures 'ofport'
1196 * as a VLAN splinter port.
1198 * This function should be NULL if a an implementation does not support
1200 int (*set_realdev)(struct ofport *ofport,
1201 uint16_t realdev_ofp_port, int vid);
1204 extern const struct ofproto_class ofproto_dpif_class;
1206 int ofproto_class_register(const struct ofproto_class *);
1207 int ofproto_class_unregister(const struct ofproto_class *);
1209 /* ofproto_flow_mod() returns this value if the flow_mod could not be processed
1210 * because it overlaps with an ongoing flow table operation that has not yet
1211 * completed. The caller should retry the operation later.
1213 * ofproto.c also uses this value internally for additional (similar) purposes.
1215 * This particular value is a good choice because it is large, so that it does
1216 * not collide with any errno value, but not large enough to collide with an
1217 * OFPERR_* value. */
1218 enum { OFPROTO_POSTPONE = 1 << 16 };
1219 BUILD_ASSERT_DECL(OFPROTO_POSTPONE < OFPERR_OFS);
1221 int ofproto_flow_mod(struct ofproto *, const struct ofputil_flow_mod *);
1222 void ofproto_add_flow(struct ofproto *, const struct cls_rule *,
1223 const struct ofpact *ofpacts, size_t ofpacts_len);
1224 bool ofproto_delete_flow(struct ofproto *, const struct cls_rule *);
1225 void ofproto_flush_flows(struct ofproto *);
1227 #endif /* ofproto/ofproto-provider.h */