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"
54 #include "services/rpz.h"
55 #include "libunbound/unbound.h"
62 struct outbound_entry;
64 struct respip_client_info;
67 * Maximum number of mesh state activations. Any more is likely an
68 * infinite loop in the module. It is then terminated.
70 #define MESH_MAX_ACTIVATION 10000
73 * Max number of references-to-references-to-references.. search size.
74 * Any more is treated like 'too large', and the creation of a new
75 * dependency is failed (so that no loops can be created).
77 #define MESH_MAX_SUBSUB 1024
80 * Mesh of query states
83 /** active module stack */
84 struct module_stack mods;
85 /** environment for new states */
86 struct module_env* env;
88 /** set of runnable queries (mesh_state.run_node) */
90 /** rbtree of all current queries (mesh_state.node)*/
93 /** count of the total number of mesh_reply entries */
94 size_t num_reply_addrs;
95 /** count of the number of mesh_states that have mesh_replies
96 * Because a state can send results to multiple reply addresses,
97 * this number must be equal or lower than num_reply_addrs. */
98 size_t num_reply_states;
99 /** number of mesh_states that have no mesh_replies, and also
100 * an empty set of super-states, thus are 'toplevel' or detached
101 * internal opportunistic queries */
102 size_t num_detached_states;
103 /** number of reply states in the forever list */
104 size_t num_forever_states;
106 /** max total number of reply states to have */
107 size_t max_reply_states;
108 /** max forever number of reply states to have */
109 size_t max_forever_states;
111 /** stats, cumulative number of reply states jostled out */
112 size_t stats_jostled;
113 /** stats, cumulative number of incoming client msgs dropped */
114 size_t stats_dropped;
115 /** stats, number of expired replies sent */
117 /** stats, number of cached replies from cachedb */
119 /** number of replies sent */
121 /** sum of waiting times for the replies */
122 struct timeval replies_sum_wait;
123 /** histogram of time values */
124 struct timehist* histogram;
125 /** (extended stats) secure replies */
127 /** (extended stats) bogus replies */
129 /** (extended stats) rcodes in replies */
130 size_t ans_rcode[UB_STATS_RCODE_NUM];
131 /** (extended stats) rcode nodata in replies */
133 /** (extended stats) type of applied RPZ action */
134 size_t rpz_action[UB_STATS_RPZ_ACTION_NUM];
136 /** backup of query if other operations recurse and need the
138 struct sldns_buffer* qbuf_bak;
140 /** double linked list of the run-to-completion query states.
141 * These are query states with a reply */
142 struct mesh_state* forever_first;
143 /** last entry in run forever list */
144 struct mesh_state* forever_last;
146 /** double linked list of the query states that can be jostled out
147 * by new queries if too old. These are query states with a reply */
148 struct mesh_state* jostle_first;
149 /** last entry in jostle list - this is the entry that is newest */
150 struct mesh_state* jostle_last;
151 /** timeout for jostling. if age is lower, it does not get jostled. */
152 struct timeval jostle_max;
154 /** If we need to use response ip (value passed from daemon)*/
156 /** If we need to use RPZ (value passed from daemon) */
162 * Unique per qname, qtype, qclass (from the qstate).
163 * And RD / CD flag; in case a client turns it off.
164 * And priming queries are different from ordinary queries (because of hints).
166 * The entire structure is allocated in a region, this region is the qstate
167 * region. All parts (rbtree nodes etc) are also allocated in the region.
170 /** node in mesh_area all tree, key is this struct. Must be first. */
172 /** node in mesh_area runnable tree, key is this struct */
173 rbnode_type run_node;
174 /** the query state. Note that the qinfo and query_flags
176 struct module_qstate s;
177 /** the list of replies to clients for the results */
178 struct mesh_reply* reply_list;
179 /** the list of callbacks for the results */
180 struct mesh_cb* cb_list;
181 /** set of superstates (that want this state's result)
182 * contains struct mesh_state_ref* */
183 rbtree_type super_set;
184 /** set of substates (that this state needs to continue)
185 * contains struct mesh_state_ref* */
187 /** number of activations for the mesh state */
188 size_t num_activated;
190 /** previous in linked list for reply states */
191 struct mesh_state* prev;
192 /** next in linked list for reply states */
193 struct mesh_state* next;
194 /** if this state is in the forever list, jostle list, or neither */
195 enum mesh_list_select { mesh_no_list, mesh_forever_list,
196 mesh_jostle_list } list_select;
197 /** pointer to this state for uniqueness or NULL */
198 struct mesh_state* unique;
200 /** true if replies have been sent out (at end for alignment) */
201 uint8_t replies_sent;
205 * Rbtree reference to a mesh_state.
206 * Used in super_set and sub_set.
208 struct mesh_state_ref {
209 /** node in rbtree for set, key is this structure */
211 /** the mesh state */
212 struct mesh_state* s;
219 /** next in reply list */
220 struct mesh_reply* next;
221 /** the query reply destination, packet buffer and where to send. */
222 struct comm_reply query_reply;
223 /** edns data from query */
224 struct edns_data edns;
225 /** the time when request was entered */
226 struct timeval start_time;
227 /** id of query, in network byteorder. */
229 /** flags of query, for reply flags */
231 /** qname from this query. len same as mesh qinfo. */
233 /** same as that in query_info. */
234 struct local_rrset* local_alias;
235 /** send query to this http2 stream, if set */
236 struct http2_stream* h2_stream;
240 * Mesh result callback func.
241 * called as func(cb_arg, rcode, buffer_with_reply, security, why_bogus,
244 typedef void (*mesh_cb_func_type)(void* cb_arg, int rcode, struct sldns_buffer*,
245 enum sec_status, char* why_bogus, int was_ratelimited);
248 * Callback to result routine
252 struct mesh_cb* next;
253 /** edns data from query */
254 struct edns_data edns;
255 /** id of query, in network byteorder. */
257 /** flags of query, for reply flags */
259 /** buffer for reply */
260 struct sldns_buffer* buf;
261 /** callback routine for results. if rcode != 0 buf has message.
262 * called as cb(cb_arg, rcode, buf, sec_state, why_bogus, was_ratelimited);
264 mesh_cb_func_type cb;
265 /** user arg for callback */
269 /* ------------------- Functions for worker -------------------- */
272 * Allocate mesh, to empty.
273 * @param stack: module stack to activate, copied (as readonly reference).
274 * @param env: environment for new queries.
275 * @return mesh: the new mesh or NULL on error.
277 struct mesh_area* mesh_create(struct module_stack* stack,
278 struct module_env* env);
281 * Delete mesh, and all query states and replies in it.
282 * @param mesh: the mesh to delete.
284 void mesh_delete(struct mesh_area* mesh);
287 * New query incoming from clients. Create new query state if needed, and
288 * add mesh_reply to it. Returns error to client on malloc failures.
289 * Will run the mesh area queries to process if a new query state is created.
291 * @param mesh: the mesh.
292 * @param qinfo: query from client.
293 * @param cinfo: additional information associated with the query client.
294 * 'cinfo' itself is ephemeral but data pointed to by its members
295 * can be assumed to be valid and unchanged until the query processing is
297 * @param qflags: flags from client query.
298 * @param edns: edns data from client query.
299 * @param rep: where to reply to.
300 * @param qid: query id to reply with.
301 * @param rpz_passthru: if true, the rpz passthru was previously found and
302 * further rpz processing is stopped.
304 void mesh_new_client(struct mesh_area* mesh, struct query_info* qinfo,
305 struct respip_client_info* cinfo, uint16_t qflags,
306 struct edns_data* edns, struct comm_reply* rep, uint16_t qid,
310 * New query with callback. Create new query state if needed, and
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 edns: edns data from client query.
318 * @param buf: buffer for reply contents.
319 * @param qid: query id to reply with.
320 * @param cb: callback function.
321 * @param cb_arg: callback user arg.
322 * @param rpz_passthru: if true, the rpz passthru was previously found and
323 * further rpz processing is stopped.
324 * @return 0 on error.
326 int mesh_new_callback(struct mesh_area* mesh, struct query_info* qinfo,
327 uint16_t qflags, struct edns_data* edns, struct sldns_buffer* buf,
328 uint16_t qid, mesh_cb_func_type cb, void* cb_arg, int rpz_passthru);
331 * New prefetch message. Create new query state if needed.
332 * Will run the mesh area queries to process if a new query state is created.
334 * @param mesh: the mesh.
335 * @param qinfo: query from client.
336 * @param qflags: flags from client query.
337 * @param leeway: TTL leeway what to expire earlier for this update.
338 * @param rpz_passthru: if true, the rpz passthru was previously found and
339 * further rpz processing is stopped.
340 * @param addr: sockaddr_storage for the client; to be used with subnet.
341 * @param opt_list: edns opt_list from the client; to be used when subnet is
344 void mesh_new_prefetch(struct mesh_area* mesh, struct query_info* qinfo,
345 uint16_t qflags, time_t leeway, int rpz_passthru,
346 struct sockaddr_storage* addr, struct edns_option* opt_list);
349 * Handle new event from the wire. A serviced query has returned.
350 * The query state will be made runnable, and the mesh_area will process
351 * query states until processing is complete.
353 * @param mesh: the query mesh.
354 * @param e: outbound entry, with query state to run and reply pointer.
355 * @param reply: the comm point reply info.
356 * @param what: NETEVENT_* error code (if not 0, what is wrong, TIMEOUT).
358 void mesh_report_reply(struct mesh_area* mesh, struct outbound_entry* e,
359 struct comm_reply* reply, int what);
361 /* ------------------- Functions for module environment --------------- */
365 * Remove all sub-query references from this query state.
366 * Keeps super-references of those sub-queries correct.
367 * Updates stat items in mesh_area structure.
368 * @param qstate: used to find mesh state.
370 void mesh_detach_subs(struct module_qstate* qstate);
374 * Creates it if it does not exist already.
375 * Keeps sub and super references correct.
376 * Performs a cycle detection - for double check - and fails if there is one.
377 * Also fails if the sub-sub-references become too large.
378 * Updates stat items in mesh_area structure.
379 * Pass if it is priming query or not.
381 * o if error (malloc) happened.
382 * o need to initialise the new state (module init; it is a new state).
383 * so that the next run of the query with this module is successful.
384 * o no init needed, attachment successful.
386 * @param qstate: the state to find mesh state, and that wants to receive
387 * the results from the new subquery.
388 * @param qinfo: what to query for (copied).
389 * @param qflags: what flags to use (RD / CD flag or not).
390 * @param prime: if it is a (stub) priming query.
391 * @param valrec: if it is a validation recursion query (lookup of key, DS).
392 * @param newq: If the new subquery needs initialisation, it is returned,
393 * otherwise NULL is returned.
394 * @return: false on error, true if success (and init may be needed).
396 int mesh_attach_sub(struct module_qstate* qstate, struct query_info* qinfo,
397 uint16_t qflags, int prime, int valrec, struct module_qstate** newq);
400 * Add detached query.
401 * Creates it if it does not exist already.
402 * Does not make super/sub references.
403 * Performs a cycle detection - for double check - and fails if there is one.
404 * Updates stat items in mesh_area structure.
405 * Pass if it is priming query or not.
407 * o if error (malloc) happened.
408 * o need to initialise the new state (module init; it is a new state).
409 * so that the next run of the query with this module is successful.
410 * o no init needed, attachment successful.
411 * o added subquery, created if it did not exist already.
413 * @param qstate: the state to find mesh state, and that wants to receive
414 * the results from the new subquery.
415 * @param qinfo: what to query for (copied).
416 * @param qflags: what flags to use (RD / CD flag or not).
417 * @param prime: if it is a (stub) priming query.
418 * @param valrec: if it is a validation recursion query (lookup of key, DS).
419 * @param newq: If the new subquery needs initialisation, it is returned,
420 * otherwise NULL is returned.
421 * @param sub: The added mesh state, created if it did not exist already.
422 * @return: false on error, true if success (and init may be needed).
424 int mesh_add_sub(struct module_qstate* qstate, struct query_info* qinfo,
425 uint16_t qflags, int prime, int valrec, struct module_qstate** newq,
426 struct mesh_state** sub);
429 * Query state is done, send messages to reply entries.
430 * Encode messages using reply entry values and the querystate (with original
431 * qinfo), using given reply_info.
432 * Pass errcode != 0 if an error reply is needed.
433 * If no reply entries, nothing is done.
434 * Must be called before a module can module_finished or return module_error.
435 * The module must handle the super query states itself as well.
437 * @param mstate: mesh state that is done. return_rcode and return_msg
438 * are used for replies.
439 * return_rcode: if not 0 (NOERROR) an error is sent back (and
440 * return_msg is ignored).
441 * return_msg: reply to encode and send back to clients.
443 void mesh_query_done(struct mesh_state* mstate);
446 * Call inform_super for the super query states that are interested in the
447 * results from this query state. These can then be changed for error
449 * Called when a module is module_finished or returns module_error.
450 * The super query states become runnable with event module_event_pass,
451 * it calls the current module for the super with the inform_super event.
453 * @param mesh: mesh area to add newly runnable modules to.
454 * @param mstate: the state that has results, used to find mesh state.
456 void mesh_walk_supers(struct mesh_area* mesh, struct mesh_state* mstate);
459 * Delete mesh state, cleanup and also rbtrees and so on.
460 * Will detach from all super/subnodes.
461 * @param qstate: to remove.
463 void mesh_state_delete(struct module_qstate* qstate);
465 /* ------------------- Functions for mesh -------------------- */
468 * Create and initialize a new mesh state and its query state
469 * Does not put the mesh state into rbtrees and so on.
470 * @param env: module environment to set.
471 * @param qinfo: query info that the mesh is for.
472 * @param cinfo: control info for the query client (can be NULL).
473 * @param qflags: flags for query (RD / CD flag).
474 * @param prime: if true, it is a priming query, set is_priming on mesh state.
475 * @param valrec: if true, it is a validation recursion query, and sets
476 * is_valrec on the mesh state.
477 * @return: new mesh state or NULL on allocation error.
479 struct mesh_state* mesh_state_create(struct module_env* env,
480 struct query_info* qinfo, struct respip_client_info* cinfo,
481 uint16_t qflags, int prime, int valrec);
484 * Make a mesh state unique.
485 * A unique mesh state uses it's unique member to point to itself.
486 * @param mstate: mesh state to check.
488 void mesh_state_make_unique(struct mesh_state* mstate);
491 * Cleanup a mesh state and its query state. Does not do rbtree or
493 * @param mstate: mesh state to cleanup. Its pointer may no longer be used
494 * afterwards. Cleanup rbtrees before calling this function.
496 void mesh_state_cleanup(struct mesh_state* mstate);
499 * Delete all mesh states from the mesh.
500 * @param mesh: the mesh area to clear
502 void mesh_delete_all(struct mesh_area* mesh);
505 * Find a mesh state in the mesh area. Pass relevant flags.
507 * @param mesh: the mesh area to look in.
508 * @param cinfo: if non-NULL client specific info that may affect IP-based
509 * actions that apply to the query result.
510 * @param qinfo: what query
511 * @param qflags: if RD / CD bit is set or not.
512 * @param prime: if it is a priming query.
513 * @param valrec: if it is a validation-recursion query.
514 * @return: mesh state or NULL if not found.
516 struct mesh_state* mesh_area_find(struct mesh_area* mesh,
517 struct respip_client_info* cinfo, struct query_info* qinfo,
518 uint16_t qflags, int prime, int valrec);
521 * Setup attachment super/sub relation between super and sub mesh state.
522 * The relation must not be present when calling the function.
523 * Does not update stat items in mesh_area.
524 * @param super: super state.
525 * @param sub: sub state.
526 * @return: 0 on alloc error.
528 int mesh_state_attachment(struct mesh_state* super, struct mesh_state* sub);
531 * Create new reply structure and attach it to a mesh state.
532 * Does not update stat items in mesh area.
533 * @param s: the mesh state.
534 * @param edns: edns data for reply (bufsize).
535 * @param rep: comm point reply info.
536 * @param qid: ID of reply.
537 * @param qflags: original query flags.
538 * @param qinfo: original query info.
539 * @return: 0 on alloc error.
541 int mesh_state_add_reply(struct mesh_state* s, struct edns_data* edns,
542 struct comm_reply* rep, uint16_t qid, uint16_t qflags,
543 const struct query_info* qinfo);
546 * Create new callback structure and attach it to a mesh state.
547 * Does not update stat items in mesh area.
548 * @param s: the mesh state.
549 * @param edns: edns data for reply (bufsize).
550 * @param buf: buffer for reply
551 * @param cb: callback to call with results.
552 * @param cb_arg: callback user arg.
553 * @param qid: ID of reply.
554 * @param qflags: original query flags.
555 * @return: 0 on alloc error.
557 int mesh_state_add_cb(struct mesh_state* s, struct edns_data* edns,
558 struct sldns_buffer* buf, mesh_cb_func_type cb, void* cb_arg,
559 uint16_t qid, uint16_t qflags);
562 * Run the mesh. Run all runnable mesh states. Which can create new
563 * runnable mesh states. Until completion. Automatically called by
564 * mesh_report_reply and mesh_new_client as needed.
565 * @param mesh: mesh area.
566 * @param mstate: first mesh state to run.
567 * @param ev: event the mstate. Others get event_pass.
568 * @param e: if a reply, its outbound entry.
570 void mesh_run(struct mesh_area* mesh, struct mesh_state* mstate,
571 enum module_ev ev, struct outbound_entry* e);
574 * Print some stats about the mesh to the log.
575 * @param mesh: the mesh to print it for.
576 * @param str: descriptive string to go with it.
578 void mesh_stats(struct mesh_area* mesh, const char* str);
581 * Clear the stats that the mesh keeps (number of queries serviced)
582 * @param mesh: the mesh
584 void mesh_stats_clear(struct mesh_area* mesh);
587 * Print all the states in the mesh to the log.
588 * @param mesh: the mesh to print all states of.
590 void mesh_log_list(struct mesh_area* mesh);
593 * Calculate memory size in use by mesh and all queries inside it.
594 * @param mesh: the mesh to examine.
595 * @return size in bytes.
597 size_t mesh_get_mem(struct mesh_area* mesh);
600 * Find cycle; see if the given mesh is in the targets sub, or sub-sub, ...
602 * If the sub-sub structure is too large, it returns 'a cycle'=2.
603 * @param qstate: given mesh querystate.
604 * @param qinfo: query info for dependency.
605 * @param flags: query flags of dependency.
606 * @param prime: if dependency is a priming query or not.
607 * @param valrec: if it is a validation recursion query (lookup of key, DS).
608 * @return true if the name,type,class exists and the given qstate mesh exists
609 * as a dependency of that name. Thus if qstate becomes dependent on
610 * name,type,class then a cycle is created, this is return value 1.
611 * Too large to search is value 2 (also true).
613 int mesh_detect_cycle(struct module_qstate* qstate, struct query_info* qinfo,
614 uint16_t flags, int prime, int valrec);
616 /** compare two mesh_states */
617 int mesh_state_compare(const void* ap, const void* bp);
619 /** compare two mesh references */
620 int mesh_state_ref_compare(const void* ap, const void* bp);
623 * Make space for another recursion state for a reply in the mesh
624 * @param mesh: mesh area
625 * @param qbuf: query buffer to save if recursion is invoked to make space.
626 * This buffer is necessary, because the following sequence in calls
627 * can result in an overwrite of the incoming query:
628 * delete_other_mesh_query - iter_clean - serviced_delete - waiting
629 * udp query is sent - on error callback - callback sends SERVFAIL reply
630 * over the same network channel, and shared UDP buffer is overwritten.
631 * You can pass NULL if there is no buffer that must be backed up.
632 * @return false if no space is available.
634 int mesh_make_new_space(struct mesh_area* mesh, struct sldns_buffer* qbuf);
637 * Insert mesh state into a double linked list. Inserted at end.
638 * @param m: mesh state.
639 * @param fp: pointer to the first-elem-pointer of the list.
640 * @param lp: pointer to the last-elem-pointer of the list.
642 void mesh_list_insert(struct mesh_state* m, struct mesh_state** fp,
643 struct mesh_state** lp);
646 * Remove mesh state from a double linked list. Remove from any position.
647 * @param m: mesh state.
648 * @param fp: pointer to the first-elem-pointer of the list.
649 * @param lp: pointer to the last-elem-pointer of the list.
651 void mesh_list_remove(struct mesh_state* m, struct mesh_state** fp,
652 struct mesh_state** lp);
655 * Remove mesh reply entry from the reply entry list. Searches for
656 * the comm_point pointer.
657 * @param mesh: to update the counters.
658 * @param m: the mesh state.
659 * @param cp: the comm_point to remove from the list.
661 void mesh_state_remove_reply(struct mesh_area* mesh, struct mesh_state* m,
662 struct comm_point* cp);
664 /** Callback for when the serve expired client timer has run out. Tries to
665 * find an expired answer in the cache and reply that to the client.
666 * @param arg: the argument passed to the callback.
668 void mesh_serve_expired_callback(void* arg);
671 * Try to get a (expired) cached answer.
672 * This needs to behave like the worker's answer_from_cache() in order to have
673 * the same behavior as when replying from cache.
674 * @param qstate: the module qstate.
675 * @param lookup_qinfo: the query info to look for in the cache.
676 * @return dns_msg if a cached answer was found, otherwise NULL.
679 mesh_serve_expired_lookup(struct module_qstate* qstate,
680 struct query_info* lookup_qinfo);
683 * See if the mesh has space for more queries. You can allocate queries
684 * anyway, but this checks for the allocated space.
685 * @param mesh: mesh area.
686 * @return true if the query list is full.
687 * It checks the number of all queries, not just number of reply states,
688 * that have a client address. So that spawned queries count too,
689 * that were created by the iterator, or other modules.
691 int mesh_jostle_exceeded(struct mesh_area* mesh);
693 #endif /* SERVICES_MESH_H */