1 /* 2 * util/module.h - DNS handling module interface 3 * 4 * Copyright (c) 2007, NLnet Labs. All rights reserved. 5 * 6 * This software is open source. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 12 * Redistributions of source code must retain the above copyright notice, 13 * this list of conditions and the following disclaimer. 14 * 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. 18 * 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. 22 * 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. 34 */ 35 36 /** 37 * \file 38 * 39 * This file contains the interface for DNS handling modules. 40 * 41 * The module interface uses the DNS modules as state machines. The 42 * state machines are activated in sequence to operate on queries. Once 43 * they are done, the reply is passed back. In the usual setup the mesh 44 * is the caller of the state machines and once things are done sends replies 45 * and invokes result callbacks. 46 * 47 * The module provides a number of functions, listed in the module_func_block. 48 * The module is inited and destroyed and memory usage queries, for the 49 * module as a whole, for entire-module state (such as a cache). And per-query 50 * functions are called, operate to move the state machine and cleanup of 51 * the per-query state. 52 * 53 * Most per-query state should simply be allocated in the query region. 54 * This is destroyed at the end of the query. 55 * 56 * The module environment contains services and information and caches 57 * shared by the modules and the rest of the system. It also contains 58 * function pointers for module-specific tasks (like sending queries). 59 * 60 * *** Example module calls for a normal query 61 * 62 * In this example, the query does not need recursion, all the other data 63 * can be found in the cache. This makes the example shorter. 64 * 65 * At the start of the program the iterator module is initialised. 66 * The iterator module sets up its global state, such as donotquery lists 67 * and private address trees. 68 * 69 * A query comes in, and a mesh entry is created for it. The mesh 70 * starts the resolution process. The validator module is the first 71 * in the list of modules, and it is started on this new query. The 72 * operate() function is called. The validator decides it needs not do 73 * anything yet until there is a result and returns wait_module, that 74 * causes the next module in the list to be started. 75 * 76 * The next module is the iterator. It is started on the passed query and 77 * decides to perform a lookup. For this simple example, the delegation 78 * point information is available, and all the iterator wants to do is 79 * send a UDP query. The iterator uses env.send_query() to send the 80 * query. Then the iterator suspends (returns from the operate call). 81 * 82 * When the UDP reply comes back (and on errors and timeouts), the 83 * operate function is called for the query, on the iterator module, 84 * with the event that there is a reply. The iterator decides that this 85 * is enough, the work is done. It returns the value finished from the 86 * operate call, which causes the previous module to be started. 87 * 88 * The previous module, the validator module, is started with the event 89 * that the iterator module is done. The validator decides to validate 90 * the query. Once it is done (which could take recursive lookups, but 91 * in this example no recursive lookups are needed), it returns from the 92 * operate function with finished. 93 * 94 * There is no previous module from the validator module, and the mesh 95 * takes this to mean that the query is finally done. The mesh invokes 96 * callbacks and sends packets to queriers. 97 * 98 * If other modules had been waiting (recursively) on the answer to this 99 * query, then the mesh will tell them about it. It calls the inform_super 100 * routine on all the waiting modules, and once that is done it calls all of 101 * them with the operate() call. During inform_super the query that is done 102 * still exists and information can be copied from it (but the module should 103 * not really re-entry codepoints and services). During the operate call 104 * the modules can use stored state to continue operation with the results. 105 * (network buffers are used to contain the answer packet during the 106 * inform_super phase, but after that the network buffers will be cleared 107 * of their contents so that other tasks can be performed). 108 * 109 * *** Example module calls for recursion 110 * 111 * A module is called in operate, and it decides that it wants to perform 112 * recursion. That is, it wants the full state-machine-list to operate on 113 * a different query. It calls env.attach_sub() to create a new query state. 114 * The routine returns the newly created state, and potentially the module 115 * can edit the module-states for the newly created query (i.e. pass along 116 * some information, like delegation points). The module then suspends, 117 * returns from the operate routine. 118 * 119 * The mesh meanwhile will have the newly created query (or queries) on 120 * a waiting list, and will call operate() on this query (or queries). 121 * It starts again at the start of the module list for them. The query 122 * (or queries) continue to operate their state machines, until they are 123 * done. When they are done the mesh calls inform_super on the module that 124 * wanted the recursion. After that the mesh calls operate() on the module 125 * that wanted to do the recursion, and during this phase the module could, 126 * for example, decide to create more recursions. 127 * 128 * If the module decides it no longer wants the recursive information 129 * it can call detach_subs. Those queries will still run to completion, 130 * potentially filling the cache with information. Inform_super is not 131 * called any more. 132 * 133 * The iterator module will fetch items from the cache, so a recursion 134 * attempt may complete very quickly if the item is in cache. The calling 135 * module has to wait for completion or eventual timeout. A recursive query 136 * that times out returns a servfail rcode (servfail is also returned for 137 * other errors during the lookup). 138 * 139 * Results are passed in the qstate, the rcode member is used to pass 140 * errors without requiring memory allocation, so that the code can continue 141 * in out-of-memory conditions. If the rcode member is 0 (NOERROR) then 142 * the dns_msg entry contains a filled out message. This message may 143 * also contain an rcode that is nonzero, but in this case additional 144 * information (query, additional) can be passed along. 145 * 146 * The rcode and dns_msg are used to pass the result from the the rightmost 147 * module towards the leftmost modules and then towards the user. 148 * 149 * If you want to avoid recursion-cycles where queries need other queries 150 * that need the first one, use detect_cycle() to see if that will happen. 151 * 152 */ 153 154 #ifndef UTIL_MODULE_H 155 #define UTIL_MODULE_H 156 #include "util/storage/lruhash.h" 157 #include "util/data/msgreply.h" 158 #include "util/data/msgparse.h" 159 struct sldns_buffer; 160 struct alloc_cache; 161 struct rrset_cache; 162 struct key_cache; 163 struct config_file; 164 struct slabhash; 165 struct query_info; 166 struct edns_data; 167 struct regional; 168 struct worker; 169 struct module_qstate; 170 struct ub_randstate; 171 struct mesh_area; 172 struct mesh_state; 173 struct val_anchors; 174 struct val_neg_cache; 175 struct iter_forwards; 176 struct iter_hints; 177 178 /** Maximum number of modules in operation */ 179 #define MAX_MODULE 5 180 181 /** 182 * Module environment. 183 * Services and data provided to the module. 184 */ 185 struct module_env { 186 /* --- data --- */ 187 /** config file with config options */ 188 struct config_file* cfg; 189 /** shared message cache */ 190 struct slabhash* msg_cache; 191 /** shared rrset cache */ 192 struct rrset_cache* rrset_cache; 193 /** shared infrastructure cache (edns, lameness) */ 194 struct infra_cache* infra_cache; 195 /** shared key cache */ 196 struct key_cache* key_cache; 197 198 /* --- services --- */ 199 /** 200 * Send serviced DNS query to server. UDP/TCP and EDNS is handled. 201 * operate() should return with wait_reply. Later on a callback 202 * will cause operate() to be called with event timeout or reply. 203 * The time until a timeout is calculated from roundtrip timing, 204 * several UDP retries are attempted. 205 * @param qname: query name. (host order) 206 * @param qnamelen: length in bytes of qname, including trailing 0. 207 * @param qtype: query type. (host order) 208 * @param qclass: query class. (host order) 209 * @param flags: host order flags word, with opcode and CD bit. 210 * @param dnssec: if set, EDNS record will have bits set. 211 * If EDNS_DO bit is set, DO bit is set in EDNS records. 212 * If BIT_CD is set, CD bit is set in queries with EDNS records. 213 * @param want_dnssec: if set, the validator wants DNSSEC. Without 214 * EDNS, the answer is likely to be useless for this domain. 215 * @param nocaps: do not use caps_for_id, use the qname as given. 216 * (ignored if caps_for_id is disabled). 217 * @param opt_list: set these EDNS options on the outgoing packet. 218 * or NULL if none (the list is deep-copied). 219 * @param addr: where to. 220 * @param addrlen: length of addr. 221 * @param zone: delegation point name. 222 * @param zonelen: length of zone name. 223 * @param q: wich query state to reactivate upon return. 224 * @return: false on failure (memory or socket related). no query was 225 * sent. Or returns an outbound entry with qsent and qstate set. 226 * This outbound_entry will be used on later module invocations 227 * that involve this query (timeout, error or reply). 228 */ 229 struct outbound_entry* (*send_query)(uint8_t* qname, size_t qnamelen, 230 uint16_t qtype, uint16_t qclass, uint16_t flags, int dnssec, 231 int want_dnssec, int nocaps, struct edns_option* opt_list, 232 struct sockaddr_storage* addr, socklen_t addrlen, 233 uint8_t* zone, size_t zonelen, struct module_qstate* q); 234 235 /** 236 * Detach-subqueries. 237 * Remove all sub-query references from this query state. 238 * Keeps super-references of those sub-queries correct. 239 * Updates stat items in mesh_area structure. 240 * @param qstate: used to find mesh state. 241 */ 242 void (*detach_subs)(struct module_qstate* qstate); 243 244 /** 245 * Attach subquery. 246 * Creates it if it does not exist already. 247 * Keeps sub and super references correct. 248 * Updates stat items in mesh_area structure. 249 * Pass if it is priming query or not. 250 * return: 251 * o if error (malloc) happened. 252 * o need to initialise the new state (module init; it is a new state). 253 * so that the next run of the query with this module is successful. 254 * o no init needed, attachment successful. 255 * 256 * @param qstate: the state to find mesh state, and that wants to 257 * receive the results from the new subquery. 258 * @param qinfo: what to query for (copied). 259 * @param qflags: what flags to use (RD, CD flag or not). 260 * @param prime: if it is a (stub) priming query. 261 * @param valrec: validation lookup recursion, does not need validation 262 * @param newq: If the new subquery needs initialisation, it is 263 * returned, otherwise NULL is returned. 264 * @return: false on error, true if success (and init may be needed). 265 */ 266 int (*attach_sub)(struct module_qstate* qstate, 267 struct query_info* qinfo, uint16_t qflags, int prime, 268 int valrec, struct module_qstate** newq); 269 270 /** 271 * Kill newly attached sub. If attach_sub returns newq for 272 * initialisation, but that fails, then this routine will cleanup and 273 * delete the fresly created sub. 274 * @param newq: the new subquery that is no longer needed. 275 * It is removed. 276 */ 277 void (*kill_sub)(struct module_qstate* newq); 278 279 /** 280 * Detect if adding a dependency for qstate on name,type,class will 281 * create a dependency cycle. 282 * @param qstate: given mesh querystate. 283 * @param qinfo: query info for dependency. 284 * @param flags: query flags of dependency, RD/CD flags. 285 * @param prime: if dependency is a priming query or not. 286 * @param valrec: validation lookup recursion, does not need validation 287 * @return true if the name,type,class exists and the given 288 * qstate mesh exists as a dependency of that name. Thus 289 * if qstate becomes dependent on name,type,class then a 290 * cycle is created. 291 */ 292 int (*detect_cycle)(struct module_qstate* qstate, 293 struct query_info* qinfo, uint16_t flags, int prime, 294 int valrec); 295 296 /** region for temporary usage. May be cleared after operate() call. */ 297 struct regional* scratch; 298 /** buffer for temporary usage. May be cleared after operate() call. */ 299 struct sldns_buffer* scratch_buffer; 300 /** internal data for daemon - worker thread. */ 301 struct worker* worker; 302 /** mesh area with query state dependencies */ 303 struct mesh_area* mesh; 304 /** allocation service */ 305 struct alloc_cache* alloc; 306 /** random table to generate random numbers */ 307 struct ub_randstate* rnd; 308 /** time in seconds, converted to integer */ 309 time_t* now; 310 /** time in microseconds. Relatively recent. */ 311 struct timeval* now_tv; 312 /** is validation required for messages, controls client-facing 313 * validation status (AD bits) and servfails */ 314 int need_to_validate; 315 /** trusted key storage; these are the configured keys, if not NULL, 316 * otherwise configured by validator. These are the trust anchors, 317 * and are not primed and ready for validation, but on the bright 318 * side, they are read only memory, thus no locks and fast. */ 319 struct val_anchors* anchors; 320 /** negative cache, configured by the validator. if not NULL, 321 * contains NSEC record lookup trees. */ 322 struct val_neg_cache* neg_cache; 323 /** the 5011-probe timer (if any) */ 324 struct comm_timer* probe_timer; 325 /** Mapping of forwarding zones to targets. 326 * iterator forwarder information. per-thread, created by worker */ 327 struct iter_forwards* fwds; 328 /** 329 * iterator forwarder information. per-thread, created by worker. 330 * The hints -- these aren't stored in the cache because they don't 331 * expire. The hints are always used to "prime" the cache. Note 332 * that both root hints and stub zone "hints" are stored in this 333 * data structure. 334 */ 335 struct iter_hints* hints; 336 /** module specific data. indexed by module id. */ 337 void* modinfo[MAX_MODULE]; 338 }; 339 340 /** 341 * External visible states of the module state machine 342 * Modules may also have an internal state. 343 * Modules are supposed to run to completion or until blocked. 344 */ 345 enum module_ext_state { 346 /** initial state - new query */ 347 module_state_initial = 0, 348 /** waiting for reply to outgoing network query */ 349 module_wait_reply, 350 /** module is waiting for another module */ 351 module_wait_module, 352 /** module is waiting for another module; that other is restarted */ 353 module_restart_next, 354 /** module is waiting for sub-query */ 355 module_wait_subquery, 356 /** module could not finish the query */ 357 module_error, 358 /** module is finished with query */ 359 module_finished 360 }; 361 362 /** 363 * Events that happen to modules, that start or wakeup modules. 364 */ 365 enum module_ev { 366 /** new query */ 367 module_event_new = 0, 368 /** query passed by other module */ 369 module_event_pass, 370 /** reply inbound from server */ 371 module_event_reply, 372 /** no reply, timeout or other error */ 373 module_event_noreply, 374 /** reply is there, but capitalisation check failed */ 375 module_event_capsfail, 376 /** next module is done, and its reply is awaiting you */ 377 module_event_moddone, 378 /** error */ 379 module_event_error 380 }; 381 382 /** 383 * Linked list of sockaddrs 384 * May be allocated such that only 'len' bytes of addr exist for the structure. 385 */ 386 struct sock_list { 387 /** next in list */ 388 struct sock_list* next; 389 /** length of addr */ 390 socklen_t len; 391 /** sockaddr */ 392 struct sockaddr_storage addr; 393 }; 394 395 /** 396 * Module state, per query. 397 */ 398 struct module_qstate { 399 /** which query is being answered: name, type, class */ 400 struct query_info qinfo; 401 /** flags uint16 from query */ 402 uint16_t query_flags; 403 /** if this is a (stub or root) priming query (with hints) */ 404 int is_priming; 405 /** if this is a validation recursion query that does not get 406 * validation itself */ 407 int is_valrec; 408 409 /** comm_reply contains server replies */ 410 struct comm_reply* reply; 411 /** the reply message, with message for client and calling module */ 412 struct dns_msg* return_msg; 413 /** the rcode, in case of error, instead of a reply message */ 414 int return_rcode; 415 /** origin of the reply (can be NULL from cache, list for cnames) */ 416 struct sock_list* reply_origin; 417 /** IP blacklist for queries */ 418 struct sock_list* blacklist; 419 /** region for this query. Cleared when query process finishes. */ 420 struct regional* region; 421 /** failure reason information if val-log-level is high */ 422 struct config_strlist* errinf; 423 424 /** which module is executing */ 425 int curmod; 426 /** module states */ 427 enum module_ext_state ext_state[MAX_MODULE]; 428 /** module specific data for query. indexed by module id. */ 429 void* minfo[MAX_MODULE]; 430 /** environment for this query */ 431 struct module_env* env; 432 /** mesh related information for this query */ 433 struct mesh_state* mesh_info; 434 /** how many seconds before expiry is this prefetched (0 if not) */ 435 time_t prefetch_leeway; 436 }; 437 438 /** 439 * Module functionality block 440 */ 441 struct module_func_block { 442 /** text string name of module */ 443 const char* name; 444 445 /** 446 * init the module. Called once for the global state. 447 * This is the place to apply settings from the config file. 448 * @param env: module environment. 449 * @param id: module id number. 450 * return: 0 on error 451 */ 452 int (*init)(struct module_env* env, int id); 453 454 /** 455 * de-init, delete, the module. Called once for the global state. 456 * @param env: module environment. 457 * @param id: module id number. 458 */ 459 void (*deinit)(struct module_env* env, int id); 460 461 /** 462 * accept a new query, or work further on existing query. 463 * Changes the qstate->ext_state to be correct on exit. 464 * @param ev: event that causes the module state machine to 465 * (re-)activate. 466 * @param qstate: the query state. 467 * Note that this method is not allowed to change the 468 * query state 'identity', that is query info, qflags, 469 * and priming status. 470 * Attach a subquery to get results to a different query. 471 * @param id: module id number that operate() is called on. 472 * @param outbound: if not NULL this event is due to the reply/timeout 473 * or error on this outbound query. 474 * @return: if at exit the ext_state is: 475 * o wait_module: next module is started. (with pass event). 476 * o error or finished: previous module is resumed. 477 * o otherwise it waits until that event happens (assumes 478 * the service routine to make subrequest or send message 479 * have been called. 480 */ 481 void (*operate)(struct module_qstate* qstate, enum module_ev event, 482 int id, struct outbound_entry* outbound); 483 484 /** 485 * inform super querystate about the results from this subquerystate. 486 * Is called when the querystate is finished. The method invoked is 487 * the one from the current module active in the super querystate. 488 * @param qstate: the query state that is finished. 489 * Examine return_rcode and return_reply in the qstate. 490 * @param id: module id for this module. 491 * This coincides with the current module for the super qstate. 492 * @param super: the super querystate that needs to be informed. 493 */ 494 void (*inform_super)(struct module_qstate* qstate, int id, 495 struct module_qstate* super); 496 497 /** 498 * clear module specific data 499 */ 500 void (*clear)(struct module_qstate* qstate, int id); 501 502 /** 503 * How much memory is the module specific data using. 504 * @param env: module environment. 505 * @param id: the module id. 506 * @return the number of bytes that are alloced. 507 */ 508 size_t (*get_mem)(struct module_env* env, int id); 509 }; 510 511 /** 512 * Debug utility: module external qstate to string 513 * @param s: the state value. 514 * @return descriptive string. 515 */ 516 const char* strextstate(enum module_ext_state s); 517 518 /** 519 * Debug utility: module event to string 520 * @param e: the module event value. 521 * @return descriptive string. 522 */ 523 const char* strmodulevent(enum module_ev e); 524 525 #endif /* UTIL_MODULE_H */ 526