2 * services/mesh.h - deal with mesh of query states and handle events for that.
4 * Copyright (c) 2007, NLnet Labs. All rights reserved.
6 * This software is open source.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * Redistributions of source code must retain the above copyright notice,
13 * this list of conditions and the following disclaimer.
15 * Redistributions in binary form must reproduce the above copyright notice,
16 * this list of conditions and the following disclaimer in the documentation
17 * and/or other materials provided with the distribution.
19 * Neither the name of the NLNET LABS nor the names of its contributors may
20 * be used to endorse or promote products derived from this software without
21 * specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
26 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
27 * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
28 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
29 * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
30 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
39 * This file contains functions to assist in dealing with a mesh of
40 * query states. This mesh is supposed to be thread-specific.
41 * It consists of query states (per qname, qtype, qclass) and connections
42 * between query states and the super and subquery states, and replies to
43 * send back to clients.
46 #ifndef SERVICES_MESH_H
47 #define SERVICES_MESH_H
49 #include "util/rbtree.h"
50 #include "util/netevent.h"
51 #include "util/data/msgparse.h"
52 #include "util/module.h"
53 #include "services/modstack.h"
60 struct outbound_entry;
62 struct respip_client_info;
65 * Maximum number of mesh state activations. Any more is likely an
66 * infinite loop in the module. It is then terminated.
68 #define MESH_MAX_ACTIVATION 3000
71 * Max number of references-to-references-to-references.. search size.
72 * Any more is treated like 'too large', and the creation of a new
73 * dependency is failed (so that no loops can be created).
75 #define MESH_MAX_SUBSUB 1024
78 * Mesh of query states
81 /** active module stack */
82 struct module_stack mods;
83 /** environment for new states */
84 struct module_env* env;
86 /** set of runnable queries (mesh_state.run_node) */
88 /** rbtree of all current queries (mesh_state.node)*/
91 /** count of the total number of mesh_reply entries */
92 size_t num_reply_addrs;
93 /** count of the number of mesh_states that have mesh_replies
94 * Because a state can send results to multiple reply addresses,
95 * this number must be equal or lower than num_reply_addrs. */
96 size_t num_reply_states;
97 /** number of mesh_states that have no mesh_replies, and also
98 * an empty set of super-states, thus are 'toplevel' or detached
99 * internal opportunistic queries */
100 size_t num_detached_states;
101 /** number of reply states in the forever list */
102 size_t num_forever_states;
104 /** max total number of reply states to have */
105 size_t max_reply_states;
106 /** max forever number of reply states to have */
107 size_t max_forever_states;
109 /** stats, cumulative number of reply states jostled out */
110 size_t stats_jostled;
111 /** stats, cumulative number of incoming client msgs dropped */
112 size_t stats_dropped;
113 /** number of replies sent */
115 /** sum of waiting times for the replies */
116 struct timeval replies_sum_wait;
117 /** histogram of time values */
118 struct timehist* histogram;
119 /** (extended stats) secure replies */
121 /** (extended stats) bogus replies */
123 /** (extended stats) rcodes in replies */
124 size_t ans_rcode[16];
125 /** (extended stats) rcode nodata in replies */
128 /** backup of query if other operations recurse and need the
130 struct sldns_buffer* qbuf_bak;
132 /** double linked list of the run-to-completion query states.
133 * These are query states with a reply */
134 struct mesh_state* forever_first;
135 /** last entry in run forever list */
136 struct mesh_state* forever_last;
138 /** double linked list of the query states that can be jostled out
139 * by new queries if too old. These are query states with a reply */
140 struct mesh_state* jostle_first;
141 /** last entry in jostle list - this is the entry that is newest */
142 struct mesh_state* jostle_last;
143 /** timeout for jostling. if age is lower, it does not get jostled. */
144 struct timeval jostle_max;
149 * Unique per qname, qtype, qclass (from the qstate).
150 * And RD / CD flag; in case a client turns it off.
151 * And priming queries are different from ordinary queries (because of hints).
153 * The entire structure is allocated in a region, this region is the qstate
154 * region. All parts (rbtree nodes etc) are also allocated in the region.
157 /** node in mesh_area all tree, key is this struct. Must be first. */
159 /** node in mesh_area runnable tree, key is this struct */
160 rbnode_type run_node;
161 /** the query state. Note that the qinfo and query_flags
163 struct module_qstate s;
164 /** the list of replies to clients for the results */
165 struct mesh_reply* reply_list;
166 /** the list of callbacks for the results */
167 struct mesh_cb* cb_list;
168 /** set of superstates (that want this state's result)
169 * contains struct mesh_state_ref* */
170 rbtree_type super_set;
171 /** set of substates (that this state needs to continue)
172 * contains struct mesh_state_ref* */
174 /** number of activations for the mesh state */
175 size_t num_activated;
177 /** previous in linked list for reply states */
178 struct mesh_state* prev;
179 /** next in linked list for reply states */
180 struct mesh_state* next;
181 /** if this state is in the forever list, jostle list, or neither */
182 enum mesh_list_select { mesh_no_list, mesh_forever_list,
183 mesh_jostle_list } list_select;
184 /** pointer to this state for uniqueness or NULL */
185 struct mesh_state* unique;
187 /** true if replies have been sent out (at end for alignment) */
188 uint8_t replies_sent;
192 * Rbtree reference to a mesh_state.
193 * Used in super_set and sub_set.
195 struct mesh_state_ref {
196 /** node in rbtree for set, key is this structure */
198 /** the mesh state */
199 struct mesh_state* s;
206 /** next in reply list */
207 struct mesh_reply* next;
208 /** the query reply destination, packet buffer and where to send. */
209 struct comm_reply query_reply;
210 /** edns data from query */
211 struct edns_data edns;
212 /** the time when request was entered */
213 struct timeval start_time;
214 /** id of query, in network byteorder. */
216 /** flags of query, for reply flags */
218 /** qname from this query. len same as mesh qinfo. */
220 /** same as that in query_info. */
221 struct local_rrset* local_alias;
225 * Mesh result callback func.
226 * called as func(cb_arg, rcode, buffer_with_reply, security, why_bogus);
228 typedef void (*mesh_cb_func_type)(void*, int, struct sldns_buffer*, enum sec_status,
232 * Callback to result routine
236 struct mesh_cb* next;
237 /** edns data from query */
238 struct edns_data edns;
239 /** id of query, in network byteorder. */
241 /** flags of query, for reply flags */
243 /** buffer for reply */
244 struct sldns_buffer* buf;
246 /** callback routine for results. if rcode != 0 buf has message.
247 * called as cb(cb_arg, rcode, buf, sec_state);
249 mesh_cb_func_type cb;
250 /** user arg for callback */
254 /* ------------------- Functions for worker -------------------- */
257 * Allocate mesh, to empty.
258 * @param stack: module stack to activate, copied (as readonly reference).
259 * @param env: environment for new queries.
260 * @return mesh: the new mesh or NULL on error.
262 struct mesh_area* mesh_create(struct module_stack* stack,
263 struct module_env* env);
266 * Delete mesh, and all query states and replies in it.
267 * @param mesh: the mesh to delete.
269 void mesh_delete(struct mesh_area* mesh);
272 * New query incoming from clients. Create new query state if needed, and
273 * add mesh_reply to it. Returns error to client on malloc failures.
274 * Will run the mesh area queries to process if a new query state is created.
276 * @param mesh: the mesh.
277 * @param qinfo: query from client.
278 * @param cinfo: additional information associated with the query client.
279 * 'cinfo' itself is ephemeral but data pointed to by its members
280 * can be assumed to be valid and unchanged until the query processing is
282 * @param qflags: flags from client query.
283 * @param edns: edns data from client query.
284 * @param rep: where to reply to.
285 * @param qid: query id to reply with.
287 void mesh_new_client(struct mesh_area* mesh, struct query_info* qinfo,
288 struct respip_client_info* cinfo, uint16_t qflags,
289 struct edns_data* edns, struct comm_reply* rep, uint16_t qid);
292 * New query with callback. Create new query state if needed, and
294 * Will run the mesh area queries to process if a new query state is created.
296 * @param mesh: the mesh.
297 * @param qinfo: query from client.
298 * @param qflags: flags from client query.
299 * @param edns: edns data from client query.
300 * @param buf: buffer for reply contents.
301 * @param qid: query id to reply with.
302 * @param cb: callback function.
303 * @param cb_arg: callback user arg.
304 * @return 0 on error.
306 int mesh_new_callback(struct mesh_area* mesh, struct query_info* qinfo,
307 uint16_t qflags, struct edns_data* edns, struct sldns_buffer* buf,
308 uint16_t qid, mesh_cb_func_type cb, void* cb_arg);
311 * New prefetch message. Create new query state if needed.
312 * Will run the mesh area queries to process if a new query state is created.
314 * @param mesh: the mesh.
315 * @param qinfo: query from client.
316 * @param qflags: flags from client query.
317 * @param leeway: TTL leeway what to expire earlier for this update.
319 void mesh_new_prefetch(struct mesh_area* mesh, struct query_info* qinfo,
320 uint16_t qflags, time_t leeway);
323 * Handle new event from the wire. A serviced query has returned.
324 * The query state will be made runnable, and the mesh_area will process
325 * query states until processing is complete.
327 * @param mesh: the query mesh.
328 * @param e: outbound entry, with query state to run and reply pointer.
329 * @param reply: the comm point reply info.
330 * @param what: NETEVENT_* error code (if not 0, what is wrong, TIMEOUT).
332 void mesh_report_reply(struct mesh_area* mesh, struct outbound_entry* e,
333 struct comm_reply* reply, int what);
335 /* ------------------- Functions for module environment --------------- */
339 * Remove all sub-query references from this query state.
340 * Keeps super-references of those sub-queries correct.
341 * Updates stat items in mesh_area structure.
342 * @param qstate: used to find mesh state.
344 void mesh_detach_subs(struct module_qstate* qstate);
348 * Creates it if it does not exist already.
349 * Keeps sub and super references correct.
350 * Performs a cycle detection - for double check - and fails if there is one.
351 * Also fails if the sub-sub-references become too large.
352 * Updates stat items in mesh_area structure.
353 * Pass if it is priming query or not.
355 * o if error (malloc) happened.
356 * o need to initialise the new state (module init; it is a new state).
357 * so that the next run of the query with this module is successful.
358 * o no init needed, attachment successful.
360 * @param qstate: the state to find mesh state, and that wants to receive
361 * the results from the new subquery.
362 * @param qinfo: what to query for (copied).
363 * @param qflags: what flags to use (RD / CD flag or not).
364 * @param prime: if it is a (stub) priming query.
365 * @param valrec: if it is a validation recursion query (lookup of key, DS).
366 * @param newq: If the new subquery needs initialisation, it is returned,
367 * otherwise NULL is returned.
368 * @return: false on error, true if success (and init may be needed).
370 int mesh_attach_sub(struct module_qstate* qstate, struct query_info* qinfo,
371 uint16_t qflags, int prime, int valrec, struct module_qstate** newq);
374 * Query state is done, send messages to reply entries.
375 * Encode messages using reply entry values and the querystate (with original
376 * qinfo), using given reply_info.
377 * Pass errcode != 0 if an error reply is needed.
378 * If no reply entries, nothing is done.
379 * Must be called before a module can module_finished or return module_error.
380 * The module must handle the super query states itself as well.
382 * @param mstate: mesh state that is done. return_rcode and return_msg
383 * are used for replies.
384 * return_rcode: if not 0 (NOERROR) an error is sent back (and
385 * return_msg is ignored).
386 * return_msg: reply to encode and send back to clients.
388 void mesh_query_done(struct mesh_state* mstate);
391 * Call inform_super for the super query states that are interested in the
392 * results from this query state. These can then be changed for error
394 * Called when a module is module_finished or returns module_error.
395 * The super query states become runnable with event module_event_pass,
396 * it calls the current module for the super with the inform_super event.
398 * @param mesh: mesh area to add newly runnable modules to.
399 * @param mstate: the state that has results, used to find mesh state.
401 void mesh_walk_supers(struct mesh_area* mesh, struct mesh_state* mstate);
404 * Delete mesh state, cleanup and also rbtrees and so on.
405 * Will detach from all super/subnodes.
406 * @param qstate: to remove.
408 void mesh_state_delete(struct module_qstate* qstate);
410 /* ------------------- Functions for mesh -------------------- */
413 * Create and initialize a new mesh state and its query state
414 * Does not put the mesh state into rbtrees and so on.
415 * @param env: module environment to set.
416 * @param qinfo: query info that the mesh is for.
417 * @param cinfo: control info for the query client (can be NULL).
418 * @param qflags: flags for query (RD / CD flag).
419 * @param prime: if true, it is a priming query, set is_priming on mesh state.
420 * @param valrec: if true, it is a validation recursion query, and sets
421 * is_valrec on the mesh state.
422 * @return: new mesh state or NULL on allocation error.
424 struct mesh_state* mesh_state_create(struct module_env* env,
425 struct query_info* qinfo, struct respip_client_info* cinfo,
426 uint16_t qflags, int prime, int valrec);
429 * Check if the mesh state is unique.
430 * A unique mesh state uses it's unique member to point to itself, else NULL.
431 * @param mstate: mesh state to check.
432 * @return true if the mesh state is unique, false otherwise.
434 int mesh_state_is_unique(struct mesh_state* mstate);
437 * Make a mesh state unique.
438 * A unique mesh state uses it's unique member to point to itself.
439 * @param mstate: mesh state to check.
441 void mesh_state_make_unique(struct mesh_state* mstate);
444 * Cleanup a mesh state and its query state. Does not do rbtree or
446 * @param mstate: mesh state to cleanup. Its pointer may no longer be used
447 * afterwards. Cleanup rbtrees before calling this function.
449 void mesh_state_cleanup(struct mesh_state* mstate);
452 * Delete all mesh states from the mesh.
453 * @param mesh: the mesh area to clear
455 void mesh_delete_all(struct mesh_area* mesh);
458 * Find a mesh state in the mesh area. Pass relevant flags.
460 * @param mesh: the mesh area to look in.
461 * @param cinfo: if non-NULL client specific info that may affect IP-based
462 * actions that apply to the query result.
463 * @param qinfo: what query
464 * @param qflags: if RD / CD bit is set or not.
465 * @param prime: if it is a priming query.
466 * @param valrec: if it is a validation-recursion query.
467 * @return: mesh state or NULL if not found.
469 struct mesh_state* mesh_area_find(struct mesh_area* mesh,
470 struct respip_client_info* cinfo, struct query_info* qinfo,
471 uint16_t qflags, int prime, int valrec);
474 * Setup attachment super/sub relation between super and sub mesh state.
475 * The relation must not be present when calling the function.
476 * Does not update stat items in mesh_area.
477 * @param super: super state.
478 * @param sub: sub state.
479 * @return: 0 on alloc error.
481 int mesh_state_attachment(struct mesh_state* super, struct mesh_state* sub);
484 * Create new reply structure and attach it to a mesh state.
485 * Does not update stat items in mesh area.
486 * @param s: the mesh state.
487 * @param edns: edns data for reply (bufsize).
488 * @param rep: comm point reply info.
489 * @param qid: ID of reply.
490 * @param qflags: original query flags.
491 * @param qinfo: original query info.
492 * @return: 0 on alloc error.
494 int mesh_state_add_reply(struct mesh_state* s, struct edns_data* edns,
495 struct comm_reply* rep, uint16_t qid, uint16_t qflags,
496 const struct query_info* qinfo);
499 * Create new callback structure and attach it to a mesh state.
500 * Does not update stat items in mesh area.
501 * @param s: the mesh state.
502 * @param edns: edns data for reply (bufsize).
503 * @param buf: buffer for reply
504 * @param cb: callback to call with results.
505 * @param cb_arg: callback user arg.
506 * @param qid: ID of reply.
507 * @param qflags: original query flags.
508 * @return: 0 on alloc error.
510 int mesh_state_add_cb(struct mesh_state* s, struct edns_data* edns,
511 struct sldns_buffer* buf, mesh_cb_func_type cb, void* cb_arg,
512 uint16_t qid, uint16_t qflags);
515 * Run the mesh. Run all runnable mesh states. Which can create new
516 * runnable mesh states. Until completion. Automatically called by
517 * mesh_report_reply and mesh_new_client as needed.
518 * @param mesh: mesh area.
519 * @param mstate: first mesh state to run.
520 * @param ev: event the mstate. Others get event_pass.
521 * @param e: if a reply, its outbound entry.
523 void mesh_run(struct mesh_area* mesh, struct mesh_state* mstate,
524 enum module_ev ev, struct outbound_entry* e);
527 * Print some stats about the mesh to the log.
528 * @param mesh: the mesh to print it for.
529 * @param str: descriptive string to go with it.
531 void mesh_stats(struct mesh_area* mesh, const char* str);
534 * Clear the stats that the mesh keeps (number of queries serviced)
535 * @param mesh: the mesh
537 void mesh_stats_clear(struct mesh_area* mesh);
540 * Print all the states in the mesh to the log.
541 * @param mesh: the mesh to print all states of.
543 void mesh_log_list(struct mesh_area* mesh);
546 * Calculate memory size in use by mesh and all queries inside it.
547 * @param mesh: the mesh to examine.
548 * @return size in bytes.
550 size_t mesh_get_mem(struct mesh_area* mesh);
553 * Find cycle; see if the given mesh is in the targets sub, or sub-sub, ...
555 * If the sub-sub structure is too large, it returns 'a cycle'=2.
556 * @param qstate: given mesh querystate.
557 * @param qinfo: query info for dependency.
558 * @param flags: query flags of dependency.
559 * @param prime: if dependency is a priming query or not.
560 * @param valrec: if it is a validation recursion query (lookup of key, DS).
561 * @return true if the name,type,class exists and the given qstate mesh exists
562 * as a dependency of that name. Thus if qstate becomes dependent on
563 * name,type,class then a cycle is created, this is return value 1.
564 * Too large to search is value 2 (also true).
566 int mesh_detect_cycle(struct module_qstate* qstate, struct query_info* qinfo,
567 uint16_t flags, int prime, int valrec);
569 /** compare two mesh_states */
570 int mesh_state_compare(const void* ap, const void* bp);
572 /** compare two mesh references */
573 int mesh_state_ref_compare(const void* ap, const void* bp);
576 * Make space for another recursion state for a reply in the mesh
577 * @param mesh: mesh area
578 * @param qbuf: query buffer to save if recursion is invoked to make space.
579 * This buffer is necessary, because the following sequence in calls
580 * can result in an overwrite of the incoming query:
581 * delete_other_mesh_query - iter_clean - serviced_delete - waiting
582 * udp query is sent - on error callback - callback sends SERVFAIL reply
583 * over the same network channel, and shared UDP buffer is overwritten.
584 * You can pass NULL if there is no buffer that must be backed up.
585 * @return false if no space is available.
587 int mesh_make_new_space(struct mesh_area* mesh, struct sldns_buffer* qbuf);
590 * Insert mesh state into a double linked list. Inserted at end.
591 * @param m: mesh state.
592 * @param fp: pointer to the first-elem-pointer of the list.
593 * @param lp: pointer to the last-elem-pointer of the list.
595 void mesh_list_insert(struct mesh_state* m, struct mesh_state** fp,
596 struct mesh_state** lp);
599 * Remove mesh state from a double linked list. Remove from any position.
600 * @param m: mesh state.
601 * @param fp: pointer to the first-elem-pointer of the list.
602 * @param lp: pointer to the last-elem-pointer of the list.
604 void mesh_list_remove(struct mesh_state* m, struct mesh_state** fp,
605 struct mesh_state** lp);
607 #endif /* SERVICES_MESH_H */