1 /* $NetBSD: ntp_intres.c,v 1.9 2016/05/01 23:32:00 christos Exp $ */ 2 3 /* 4 * ntp_intres.c - Implements a generic blocking worker child or thread, 5 * initially to provide a nonblocking solution for DNS 6 * name to address lookups available with getaddrinfo(). 7 * 8 * This is a new implementation as of 2009 sharing the filename and 9 * very little else with the prior implementation, which used a 10 * temporary file to receive a single set of requests from the parent, 11 * and a NTP mode 7 authenticated request to push back responses. 12 * 13 * A primary goal in rewriting this code was the need to support the 14 * pool configuration directive's requirement to retrieve multiple 15 * addresses resolving a single name, which has previously been 16 * satisfied with blocking resolver calls from the ntpd mainline code. 17 * 18 * A secondary goal is to provide a generic mechanism for other 19 * blocking operations to be delegated to a worker using a common 20 * model for both Unix and Windows ntpd. ntp_worker.c, work_fork.c, 21 * and work_thread.c implement the generic mechanism. This file 22 * implements the two current consumers, getaddrinfo_sometime() and the 23 * presently unused getnameinfo_sometime(). 24 * 25 * Both routines deliver results to a callback and manage memory 26 * allocation, meaning there is no freeaddrinfo_sometime(). 27 * 28 * The initial implementation for Unix uses a pair of unidirectional 29 * pipes, one each for requests and responses, connecting the forked 30 * blocking child worker with the ntpd mainline. The threaded code 31 * uses arrays of pointers to queue requests and responses. 32 * 33 * The parent drives the process, including scheduling sleeps between 34 * retries. 35 * 36 * Memory is managed differently for a child process, which mallocs 37 * request buffers to read from the pipe into, whereas the threaded 38 * code mallocs a copy of the request to hand off to the worker via 39 * the queueing array. The resulting request buffer is free()d by 40 * platform-independent code. A wrinkle is the request needs to be 41 * available to the requestor during response processing. 42 * 43 * Response memory allocation is also platform-dependent. With a 44 * separate process and pipes, the response is free()d after being 45 * written to the pipe. With threads, the same memory is handed 46 * over and the requestor frees it after processing is completed. 47 * 48 * The code should be generalized to support threads on Unix using 49 * much of the same code used for Windows initially. 50 * 51 */ 52 #ifdef HAVE_CONFIG_H 53 # include <config.h> 54 #endif 55 56 #include "ntp_workimpl.h" 57 58 #ifdef WORKER 59 60 #include <stdio.h> 61 #include <ctype.h> 62 #include <signal.h> 63 64 /**/ 65 #ifdef HAVE_SYS_TYPES_H 66 # include <sys/types.h> 67 #endif 68 #ifdef HAVE_NETINET_IN_H 69 #include <netinet/in.h> 70 #endif 71 #include <arpa/inet.h> 72 /**/ 73 #ifdef HAVE_SYS_PARAM_H 74 # include <sys/param.h> 75 #endif 76 77 #if !defined(HAVE_RES_INIT) && defined(HAVE___RES_INIT) 78 # define HAVE_RES_INIT 79 #endif 80 81 #if defined(HAVE_RESOLV_H) && defined(HAVE_RES_INIT) 82 # ifdef HAVE_ARPA_NAMESER_H 83 # include <arpa/nameser.h> /* DNS HEADER struct */ 84 # endif 85 # ifdef HAVE_NETDB_H 86 # include <netdb.h> 87 # endif 88 # include <resolv.h> 89 # ifdef HAVE_INT32_ONLY_WITH_DNS 90 # define HAVE_INT32 91 # endif 92 # ifdef HAVE_U_INT32_ONLY_WITH_DNS 93 # define HAVE_U_INT32 94 # endif 95 #endif 96 97 #include "ntp.h" 98 #include "ntp_debug.h" 99 #include "ntp_malloc.h" 100 #include "ntp_syslog.h" 101 #include "ntp_unixtime.h" 102 #include "ntp_intres.h" 103 #include "intreswork.h" 104 105 106 /* 107 * Following are implementations of getaddrinfo_sometime() and 108 * getnameinfo_sometime(). Each is implemented in three routines: 109 * 110 * getaddrinfo_sometime() getnameinfo_sometime() 111 * blocking_getaddrinfo() blocking_getnameinfo() 112 * getaddrinfo_sometime_complete() getnameinfo_sometime_complete() 113 * 114 * The first runs in the parent and marshalls (or serializes) request 115 * parameters into a request blob which is processed in the child by 116 * the second routine, blocking_*(), which serializes the results into 117 * a response blob unpacked by the third routine, *_complete(), which 118 * calls the callback routine provided with the request and frees 119 * _request_ memory allocated by the first routine. Response memory 120 * is managed by the code which calls the *_complete routines. 121 */ 122 123 /* === typedefs === */ 124 typedef struct blocking_gai_req_tag { /* marshalled args */ 125 size_t octets; 126 u_int dns_idx; 127 time_t scheduled; 128 time_t earliest; 129 struct addrinfo hints; 130 int retry; 131 gai_sometime_callback callback; 132 void * context; 133 size_t nodesize; 134 size_t servsize; 135 } blocking_gai_req; 136 137 typedef struct blocking_gai_resp_tag { 138 size_t octets; 139 int retcode; 140 int retry; 141 int gai_errno; /* for EAI_SYSTEM case */ 142 int ai_count; 143 /* 144 * Followed by ai_count struct addrinfo and then ai_count 145 * sockaddr_u and finally the canonical name strings. 146 */ 147 } blocking_gai_resp; 148 149 typedef struct blocking_gni_req_tag { 150 size_t octets; 151 u_int dns_idx; 152 time_t scheduled; 153 time_t earliest; 154 int retry; 155 size_t hostoctets; 156 size_t servoctets; 157 int flags; 158 gni_sometime_callback callback; 159 void * context; 160 sockaddr_u socku; 161 } blocking_gni_req; 162 163 typedef struct blocking_gni_resp_tag { 164 size_t octets; 165 int retcode; 166 int gni_errno; /* for EAI_SYSTEM case */ 167 int retry; 168 size_t hostoctets; 169 size_t servoctets; 170 /* 171 * Followed by hostoctets bytes of null-terminated host, 172 * then servoctets bytes of null-terminated service. 173 */ 174 } blocking_gni_resp; 175 176 /* per-DNS-worker state in parent */ 177 typedef struct dnschild_ctx_tag { 178 u_int index; 179 time_t next_dns_timeslot; 180 } dnschild_ctx; 181 182 /* per-DNS-worker state in worker */ 183 typedef struct dnsworker_ctx_tag { 184 blocking_child * c; 185 time_t ignore_scheduled_before; 186 #ifdef HAVE_RES_INIT 187 time_t next_res_init; 188 #endif 189 } dnsworker_ctx; 190 191 192 /* === variables === */ 193 dnschild_ctx ** dnschild_contexts; /* parent */ 194 u_int dnschild_contexts_alloc; 195 dnsworker_ctx ** dnsworker_contexts; /* child */ 196 u_int dnsworker_contexts_alloc; 197 198 #ifdef HAVE_RES_INIT 199 static time_t next_res_init; 200 #endif 201 202 203 /* === forward declarations === */ 204 static u_int reserve_dnschild_ctx(void); 205 static u_int get_dnschild_ctx(void); 206 static dnsworker_ctx * get_worker_context(blocking_child *, u_int); 207 static void scheduled_sleep(time_t, time_t, 208 dnsworker_ctx *); 209 static void manage_dns_retry_interval(time_t *, time_t *, 210 int *, 211 time_t *); 212 static int should_retry_dns(int, int); 213 #ifdef HAVE_RES_INIT 214 static void reload_resolv_conf(dnsworker_ctx *); 215 #else 216 # define reload_resolv_conf(wc) \ 217 do { \ 218 (void)(wc); \ 219 } while (FALSE) 220 #endif 221 static void getaddrinfo_sometime_complete(blocking_work_req, 222 void *, size_t, 223 void *); 224 static void getnameinfo_sometime_complete(blocking_work_req, 225 void *, size_t, 226 void *); 227 228 229 /* === functions === */ 230 /* 231 * getaddrinfo_sometime - uses blocking child to call getaddrinfo then 232 * invokes provided callback completion function. 233 */ 234 int 235 getaddrinfo_sometime( 236 const char * node, 237 const char * service, 238 const struct addrinfo * hints, 239 int retry, 240 gai_sometime_callback callback, 241 void * context 242 ) 243 { 244 blocking_gai_req * gai_req; 245 u_int idx; 246 dnschild_ctx * child_ctx; 247 size_t req_size; 248 size_t nodesize; 249 size_t servsize; 250 time_t now; 251 252 REQUIRE(NULL != node); 253 if (NULL != hints) { 254 REQUIRE(0 == hints->ai_addrlen); 255 REQUIRE(NULL == hints->ai_addr); 256 REQUIRE(NULL == hints->ai_canonname); 257 REQUIRE(NULL == hints->ai_next); 258 } 259 260 idx = get_dnschild_ctx(); 261 child_ctx = dnschild_contexts[idx]; 262 263 nodesize = strlen(node) + 1; 264 servsize = strlen(service) + 1; 265 req_size = sizeof(*gai_req) + nodesize + servsize; 266 267 gai_req = emalloc_zero(req_size); 268 269 gai_req->octets = req_size; 270 gai_req->dns_idx = idx; 271 now = time(NULL); 272 gai_req->scheduled = now; 273 gai_req->earliest = max(now, child_ctx->next_dns_timeslot); 274 child_ctx->next_dns_timeslot = gai_req->earliest; 275 if (hints != NULL) 276 gai_req->hints = *hints; 277 gai_req->retry = retry; 278 gai_req->callback = callback; 279 gai_req->context = context; 280 gai_req->nodesize = nodesize; 281 gai_req->servsize = servsize; 282 283 memcpy((char *)gai_req + sizeof(*gai_req), node, nodesize); 284 memcpy((char *)gai_req + sizeof(*gai_req) + nodesize, service, 285 servsize); 286 287 if (queue_blocking_request( 288 BLOCKING_GETADDRINFO, 289 gai_req, 290 req_size, 291 &getaddrinfo_sometime_complete, 292 gai_req)) { 293 294 msyslog(LOG_ERR, "unable to queue getaddrinfo request"); 295 errno = EFAULT; 296 return -1; 297 } 298 299 return 0; 300 } 301 302 int 303 blocking_getaddrinfo( 304 blocking_child * c, 305 blocking_pipe_header * req 306 ) 307 { 308 blocking_gai_req * gai_req; 309 dnsworker_ctx * worker_ctx; 310 blocking_pipe_header * resp; 311 blocking_gai_resp * gai_resp; 312 char * node; 313 char * service; 314 struct addrinfo * ai_res; 315 struct addrinfo * ai; 316 struct addrinfo * serialized_ai; 317 size_t canons_octets; 318 size_t this_octets; 319 size_t resp_octets; 320 char * cp; 321 time_t time_now; 322 323 gai_req = (void *)((char *)req + sizeof(*req)); 324 node = (char *)gai_req + sizeof(*gai_req); 325 service = node + gai_req->nodesize; 326 327 worker_ctx = get_worker_context(c, gai_req->dns_idx); 328 scheduled_sleep(gai_req->scheduled, gai_req->earliest, 329 worker_ctx); 330 reload_resolv_conf(worker_ctx); 331 332 /* 333 * Take a shot at the final size, better to overestimate 334 * at first and then realloc to a smaller size. 335 */ 336 337 resp_octets = sizeof(*resp) + sizeof(*gai_resp) + 338 16 * (sizeof(struct addrinfo) + 339 sizeof(sockaddr_u)) + 340 256; 341 resp = emalloc_zero(resp_octets); 342 gai_resp = (void *)(resp + 1); 343 344 TRACE(2, ("blocking_getaddrinfo given node %s serv %s fam %d flags %x\n", 345 node, service, gai_req->hints.ai_family, 346 gai_req->hints.ai_flags)); 347 #ifdef DEBUG 348 if (debug >= 2) 349 fflush(stdout); 350 #endif 351 ai_res = NULL; 352 gai_resp->retcode = getaddrinfo(node, service, &gai_req->hints, 353 &ai_res); 354 gai_resp->retry = gai_req->retry; 355 #ifdef EAI_SYSTEM 356 if (EAI_SYSTEM == gai_resp->retcode) 357 gai_resp->gai_errno = errno; 358 #endif 359 canons_octets = 0; 360 361 if (0 == gai_resp->retcode) { 362 ai = ai_res; 363 while (NULL != ai) { 364 gai_resp->ai_count++; 365 if (ai->ai_canonname) 366 canons_octets += strlen(ai->ai_canonname) + 1; 367 ai = ai->ai_next; 368 } 369 /* 370 * If this query succeeded only after retrying, DNS may have 371 * just become responsive. Ignore previously-scheduled 372 * retry sleeps once for each pending request, similar to 373 * the way scheduled_sleep() does when its worker_sleep() 374 * is interrupted. 375 */ 376 if (gai_resp->retry > INITIAL_DNS_RETRY) { 377 time_now = time(NULL); 378 worker_ctx->ignore_scheduled_before = time_now; 379 TRACE(1, ("DNS success after retry, ignoring sleeps scheduled before now (%s)\n", 380 humantime(time_now))); 381 } 382 } 383 384 /* 385 * Our response consists of a header, followed by ai_count 386 * addrinfo structs followed by ai_count sockaddr_storage 387 * structs followed by the canonical names. 388 */ 389 gai_resp->octets = sizeof(*gai_resp) 390 + gai_resp->ai_count 391 * (sizeof(gai_req->hints) 392 + sizeof(sockaddr_u)) 393 + canons_octets; 394 395 resp_octets = sizeof(*resp) + gai_resp->octets; 396 resp = erealloc(resp, resp_octets); 397 gai_resp = (void *)(resp + 1); 398 399 /* cp serves as our current pointer while serializing */ 400 cp = (void *)(gai_resp + 1); 401 canons_octets = 0; 402 403 if (0 == gai_resp->retcode) { 404 ai = ai_res; 405 while (NULL != ai) { 406 memcpy(cp, ai, sizeof(*ai)); 407 serialized_ai = (void *)cp; 408 cp += sizeof(*ai); 409 410 /* transform ai_canonname into offset */ 411 if (NULL != serialized_ai->ai_canonname) { 412 serialized_ai->ai_canonname = (char *)canons_octets; 413 canons_octets += strlen(ai->ai_canonname) + 1; 414 } 415 416 /* leave fixup of ai_addr pointer for receiver */ 417 418 ai = ai->ai_next; 419 } 420 421 ai = ai_res; 422 while (NULL != ai) { 423 INSIST(ai->ai_addrlen <= sizeof(sockaddr_u)); 424 memcpy(cp, ai->ai_addr, ai->ai_addrlen); 425 cp += sizeof(sockaddr_u); 426 427 ai = ai->ai_next; 428 } 429 430 ai = ai_res; 431 while (NULL != ai) { 432 if (NULL != ai->ai_canonname) { 433 this_octets = strlen(ai->ai_canonname) + 1; 434 memcpy(cp, ai->ai_canonname, this_octets); 435 cp += this_octets; 436 } 437 438 ai = ai->ai_next; 439 } 440 freeaddrinfo(ai_res); 441 } 442 443 /* 444 * make sure our walk and earlier calc match 445 */ 446 DEBUG_INSIST((size_t)(cp - (char *)resp) == resp_octets); 447 448 if (queue_blocking_response(c, resp, resp_octets, req)) { 449 msyslog(LOG_ERR, "blocking_getaddrinfo can not queue response"); 450 return -1; 451 } 452 453 return 0; 454 } 455 456 457 static void 458 getaddrinfo_sometime_complete( 459 blocking_work_req rtype, 460 void * context, 461 size_t respsize, 462 void * resp 463 ) 464 { 465 blocking_gai_req * gai_req; 466 blocking_gai_resp * gai_resp; 467 dnschild_ctx * child_ctx; 468 struct addrinfo * ai; 469 struct addrinfo * next_ai; 470 sockaddr_u * psau; 471 char * node; 472 char * service; 473 char * canon_start; 474 time_t time_now; 475 int again; 476 int af; 477 const char * fam_spec; 478 int i; 479 480 gai_req = context; 481 gai_resp = resp; 482 483 DEBUG_REQUIRE(BLOCKING_GETADDRINFO == rtype); 484 DEBUG_REQUIRE(respsize == gai_resp->octets); 485 486 node = (char *)gai_req + sizeof(*gai_req); 487 service = node + gai_req->nodesize; 488 489 child_ctx = dnschild_contexts[gai_req->dns_idx]; 490 491 if (0 == gai_resp->retcode) { 492 /* 493 * If this query succeeded only after retrying, DNS may have 494 * just become responsive. 495 */ 496 if (gai_resp->retry > INITIAL_DNS_RETRY) { 497 time_now = time(NULL); 498 child_ctx->next_dns_timeslot = time_now; 499 TRACE(1, ("DNS success after retry, %u next_dns_timeslot reset (%s)\n", 500 gai_req->dns_idx, humantime(time_now))); 501 } 502 } else { 503 again = should_retry_dns(gai_resp->retcode, 504 gai_resp->gai_errno); 505 /* 506 * exponential backoff of DNS retries to 64s 507 */ 508 if (gai_req->retry > 0 && again) { 509 /* log the first retry only */ 510 if (INITIAL_DNS_RETRY == gai_req->retry) 511 NLOG(NLOG_SYSINFO) { 512 af = gai_req->hints.ai_family; 513 fam_spec = (AF_INET6 == af) 514 ? " (AAAA)" 515 : (AF_INET == af) 516 ? " (A)" 517 : ""; 518 #ifdef EAI_SYSTEM 519 if (EAI_SYSTEM == gai_resp->retcode) { 520 errno = gai_resp->gai_errno; 521 msyslog(LOG_INFO, 522 "retrying DNS %s%s: EAI_SYSTEM %d: %m", 523 node, fam_spec, 524 gai_resp->gai_errno); 525 } else 526 #endif 527 msyslog(LOG_INFO, 528 "retrying DNS %s%s: %s (%d)", 529 node, fam_spec, 530 gai_strerror(gai_resp->retcode), 531 gai_resp->retcode); 532 } 533 manage_dns_retry_interval(&gai_req->scheduled, 534 &gai_req->earliest, &gai_req->retry, 535 &child_ctx->next_dns_timeslot); 536 if (!queue_blocking_request( 537 BLOCKING_GETADDRINFO, 538 gai_req, 539 gai_req->octets, 540 &getaddrinfo_sometime_complete, 541 gai_req)) 542 return; 543 else 544 msyslog(LOG_ERR, 545 "unable to retry hostname %s", 546 node); 547 } 548 } 549 550 /* 551 * fixup pointers in returned addrinfo array 552 */ 553 ai = (void *)((char *)gai_resp + sizeof(*gai_resp)); 554 next_ai = NULL; 555 for (i = gai_resp->ai_count - 1; i >= 0; i--) { 556 ai[i].ai_next = next_ai; 557 next_ai = &ai[i]; 558 } 559 560 psau = (void *)((char *)ai + gai_resp->ai_count * sizeof(*ai)); 561 canon_start = (char *)psau + gai_resp->ai_count * sizeof(*psau); 562 563 for (i = 0; i < gai_resp->ai_count; i++) { 564 if (NULL != ai[i].ai_addr) 565 ai[i].ai_addr = &psau->sa; 566 psau++; 567 if (NULL != ai[i].ai_canonname) 568 ai[i].ai_canonname += (size_t)canon_start; 569 } 570 571 ENSURE((char *)psau == canon_start); 572 573 if (!gai_resp->ai_count) 574 ai = NULL; 575 576 (*gai_req->callback)(gai_resp->retcode, gai_resp->gai_errno, 577 gai_req->context, node, service, 578 &gai_req->hints, ai); 579 580 free(gai_req); 581 /* gai_resp is part of block freed by process_blocking_resp() */ 582 } 583 584 585 #ifdef TEST_BLOCKING_WORKER 586 void gai_test_callback(int rescode, int gai_errno, void *context, const char *name, const char *service, const struct addrinfo *hints, const struct addrinfo *ai_res) 587 { 588 sockaddr_u addr; 589 590 if (rescode) { 591 TRACE(1, ("gai_test_callback context %p error rescode %d %s serv %s\n", 592 context, rescode, name, service)); 593 return; 594 } 595 while (!rescode && NULL != ai_res) { 596 ZERO_SOCK(&addr); 597 memcpy(&addr, ai_res->ai_addr, ai_res->ai_addrlen); 598 TRACE(1, ("ctx %p fam %d addr %s canon '%s' type %s at %p ai_addr %p ai_next %p\n", 599 context, 600 AF(&addr), 601 stoa(&addr), 602 (ai_res->ai_canonname) 603 ? ai_res->ai_canonname 604 : "", 605 (SOCK_DGRAM == ai_res->ai_socktype) 606 ? "DGRAM" 607 : (SOCK_STREAM == ai_res->ai_socktype) 608 ? "STREAM" 609 : "(other)", 610 ai_res, 611 ai_res->ai_addr, 612 ai_res->ai_next)); 613 614 getnameinfo_sometime((sockaddr_u *)ai_res->ai_addr, 128, 32, 0, gni_test_callback, context); 615 616 ai_res = ai_res->ai_next; 617 } 618 } 619 #endif /* TEST_BLOCKING_WORKER */ 620 621 622 int 623 getnameinfo_sometime( 624 sockaddr_u * psau, 625 size_t hostoctets, 626 size_t servoctets, 627 int flags, 628 gni_sometime_callback callback, 629 void * context 630 ) 631 { 632 blocking_gni_req * gni_req; 633 u_int idx; 634 dnschild_ctx * child_ctx; 635 time_t time_now; 636 637 REQUIRE(hostoctets); 638 REQUIRE(hostoctets + servoctets < 1024); 639 640 idx = get_dnschild_ctx(); 641 child_ctx = dnschild_contexts[idx]; 642 643 gni_req = emalloc_zero(sizeof(*gni_req)); 644 645 gni_req->octets = sizeof(*gni_req); 646 gni_req->dns_idx = idx; 647 time_now = time(NULL); 648 gni_req->scheduled = time_now; 649 gni_req->earliest = max(time_now, child_ctx->next_dns_timeslot); 650 child_ctx->next_dns_timeslot = gni_req->earliest; 651 memcpy(&gni_req->socku, psau, SOCKLEN(psau)); 652 gni_req->hostoctets = hostoctets; 653 gni_req->servoctets = servoctets; 654 gni_req->flags = flags; 655 gni_req->retry = INITIAL_DNS_RETRY; 656 gni_req->callback = callback; 657 gni_req->context = context; 658 659 if (queue_blocking_request( 660 BLOCKING_GETNAMEINFO, 661 gni_req, 662 sizeof(*gni_req), 663 &getnameinfo_sometime_complete, 664 gni_req)) { 665 666 msyslog(LOG_ERR, "unable to queue getnameinfo request"); 667 errno = EFAULT; 668 return -1; 669 } 670 671 return 0; 672 } 673 674 675 int 676 blocking_getnameinfo( 677 blocking_child * c, 678 blocking_pipe_header * req 679 ) 680 { 681 blocking_gni_req * gni_req; 682 dnsworker_ctx * worker_ctx; 683 blocking_pipe_header * resp; 684 blocking_gni_resp * gni_resp; 685 size_t octets; 686 size_t resp_octets; 687 char * service; 688 char * cp; 689 int rc; 690 time_t time_now; 691 char host[1024]; 692 693 gni_req = (void *)((char *)req + sizeof(*req)); 694 695 octets = gni_req->hostoctets + gni_req->servoctets; 696 697 /* 698 * Some alloca() implementations are fragile regarding 699 * large allocations. We only need room for the host 700 * and service names. 701 */ 702 REQUIRE(octets < sizeof(host)); 703 service = host + gni_req->hostoctets; 704 705 worker_ctx = get_worker_context(c, gni_req->dns_idx); 706 scheduled_sleep(gni_req->scheduled, gni_req->earliest, 707 worker_ctx); 708 reload_resolv_conf(worker_ctx); 709 710 /* 711 * Take a shot at the final size, better to overestimate 712 * then realloc to a smaller size. 713 */ 714 715 resp_octets = sizeof(*resp) + sizeof(*gni_resp) + octets; 716 resp = emalloc_zero(resp_octets); 717 gni_resp = (void *)((char *)resp + sizeof(*resp)); 718 719 TRACE(2, ("blocking_getnameinfo given addr %s flags 0x%x hostlen %lu servlen %lu\n", 720 stoa(&gni_req->socku), gni_req->flags, 721 (u_long)gni_req->hostoctets, (u_long)gni_req->servoctets)); 722 723 gni_resp->retcode = getnameinfo(&gni_req->socku.sa, 724 SOCKLEN(&gni_req->socku), 725 host, 726 gni_req->hostoctets, 727 service, 728 gni_req->servoctets, 729 gni_req->flags); 730 gni_resp->retry = gni_req->retry; 731 #ifdef EAI_SYSTEM 732 if (EAI_SYSTEM == gni_resp->retcode) 733 gni_resp->gni_errno = errno; 734 #endif 735 736 if (0 != gni_resp->retcode) { 737 gni_resp->hostoctets = 0; 738 gni_resp->servoctets = 0; 739 } else { 740 gni_resp->hostoctets = strlen(host) + 1; 741 gni_resp->servoctets = strlen(service) + 1; 742 /* 743 * If this query succeeded only after retrying, DNS may have 744 * just become responsive. Ignore previously-scheduled 745 * retry sleeps once for each pending request, similar to 746 * the way scheduled_sleep() does when its worker_sleep() 747 * is interrupted. 748 */ 749 if (gni_req->retry > INITIAL_DNS_RETRY) { 750 time_now = time(NULL); 751 worker_ctx->ignore_scheduled_before = time_now; 752 TRACE(1, ("DNS success after retrying, ignoring sleeps scheduled before now (%s)\n", 753 humantime(time_now))); 754 } 755 } 756 octets = gni_resp->hostoctets + gni_resp->servoctets; 757 /* 758 * Our response consists of a header, followed by the host and 759 * service strings, each null-terminated. 760 */ 761 resp_octets = sizeof(*resp) + sizeof(*gni_resp) + octets; 762 763 resp = erealloc(resp, resp_octets); 764 gni_resp = (void *)(resp + 1); 765 766 gni_resp->octets = sizeof(*gni_resp) + octets; 767 768 /* cp serves as our current pointer while serializing */ 769 cp = (void *)(gni_resp + 1); 770 771 if (0 == gni_resp->retcode) { 772 memcpy(cp, host, gni_resp->hostoctets); 773 cp += gni_resp->hostoctets; 774 memcpy(cp, service, gni_resp->servoctets); 775 cp += gni_resp->servoctets; 776 } 777 778 INSIST((size_t)(cp - (char *)resp) == resp_octets); 779 INSIST(resp_octets - sizeof(*resp) == gni_resp->octets); 780 781 rc = queue_blocking_response(c, resp, resp_octets, req); 782 if (rc) 783 msyslog(LOG_ERR, "blocking_getnameinfo unable to queue response"); 784 return rc; 785 } 786 787 788 static void 789 getnameinfo_sometime_complete( 790 blocking_work_req rtype, 791 void * context, 792 size_t respsize, 793 void * resp 794 ) 795 { 796 blocking_gni_req * gni_req; 797 blocking_gni_resp * gni_resp; 798 dnschild_ctx * child_ctx; 799 char * host; 800 char * service; 801 time_t time_now; 802 int again; 803 804 gni_req = context; 805 gni_resp = resp; 806 807 DEBUG_REQUIRE(BLOCKING_GETNAMEINFO == rtype); 808 DEBUG_REQUIRE(respsize == gni_resp->octets); 809 810 child_ctx = dnschild_contexts[gni_req->dns_idx]; 811 812 if (0 == gni_resp->retcode) { 813 /* 814 * If this query succeeded only after retrying, DNS may have 815 * just become responsive. 816 */ 817 if (gni_resp->retry > INITIAL_DNS_RETRY) { 818 time_now = time(NULL); 819 child_ctx->next_dns_timeslot = time_now; 820 TRACE(1, ("DNS success after retry, %u next_dns_timeslot reset (%s)\n", 821 gni_req->dns_idx, humantime(time_now))); 822 } 823 } else { 824 again = should_retry_dns(gni_resp->retcode, gni_resp->gni_errno); 825 /* 826 * exponential backoff of DNS retries to 64s 827 */ 828 if (gni_req->retry > 0) 829 manage_dns_retry_interval(&gni_req->scheduled, 830 &gni_req->earliest, &gni_req->retry, 831 &child_ctx->next_dns_timeslot); 832 833 if (gni_req->retry > 0 && again) { 834 if (!queue_blocking_request( 835 BLOCKING_GETNAMEINFO, 836 gni_req, 837 gni_req->octets, 838 &getnameinfo_sometime_complete, 839 gni_req)) 840 return; 841 842 msyslog(LOG_ERR, "unable to retry reverse lookup of %s", stoa(&gni_req->socku)); 843 } 844 } 845 846 if (!gni_resp->hostoctets) { 847 host = NULL; 848 service = NULL; 849 } else { 850 host = (char *)gni_resp + sizeof(*gni_resp); 851 service = (gni_resp->servoctets) 852 ? host + gni_resp->hostoctets 853 : NULL; 854 } 855 856 (*gni_req->callback)(gni_resp->retcode, gni_resp->gni_errno, 857 &gni_req->socku, gni_req->flags, host, 858 service, gni_req->context); 859 860 free(gni_req); 861 /* gni_resp is part of block freed by process_blocking_resp() */ 862 } 863 864 865 #ifdef TEST_BLOCKING_WORKER 866 void gni_test_callback(int rescode, int gni_errno, sockaddr_u *psau, int flags, const char *host, const char *service, void *context) 867 { 868 if (!rescode) 869 TRACE(1, ("gni_test_callback got host '%s' serv '%s' for addr %s context %p\n", 870 host, service, stoa(psau), context)); 871 else 872 TRACE(1, ("gni_test_callback context %p rescode %d gni_errno %d flags 0x%x addr %s\n", 873 context, rescode, gni_errno, flags, stoa(psau))); 874 } 875 #endif /* TEST_BLOCKING_WORKER */ 876 877 878 #ifdef HAVE_RES_INIT 879 static void 880 reload_resolv_conf( 881 dnsworker_ctx * worker_ctx 882 ) 883 { 884 time_t time_now; 885 886 /* 887 * This is ad-hoc. Reload /etc/resolv.conf once per minute 888 * to pick up on changes from the DHCP client. [Bug 1226] 889 * When using threads for the workers, this needs to happen 890 * only once per minute process-wide. 891 */ 892 time_now = time(NULL); 893 # ifdef WORK_THREAD 894 worker_ctx->next_res_init = next_res_init; 895 # endif 896 if (worker_ctx->next_res_init <= time_now) { 897 if (worker_ctx->next_res_init != 0) 898 res_init(); 899 worker_ctx->next_res_init = time_now + 60; 900 # ifdef WORK_THREAD 901 next_res_init = worker_ctx->next_res_init; 902 # endif 903 } 904 } 905 #endif /* HAVE_RES_INIT */ 906 907 908 static u_int 909 reserve_dnschild_ctx(void) 910 { 911 const size_t ps = sizeof(dnschild_contexts[0]); 912 const size_t cs = sizeof(*dnschild_contexts[0]); 913 u_int c; 914 u_int new_alloc; 915 size_t octets; 916 size_t new_octets; 917 918 c = 0; 919 while (TRUE) { 920 for ( ; c < dnschild_contexts_alloc; c++) { 921 if (NULL == dnschild_contexts[c]) { 922 dnschild_contexts[c] = emalloc_zero(cs); 923 924 return c; 925 } 926 } 927 new_alloc = dnschild_contexts_alloc + 20; 928 new_octets = new_alloc * ps; 929 octets = dnschild_contexts_alloc * ps; 930 dnschild_contexts = erealloc_zero(dnschild_contexts, 931 new_octets, octets); 932 dnschild_contexts_alloc = new_alloc; 933 } 934 } 935 936 937 static u_int 938 get_dnschild_ctx(void) 939 { 940 static u_int shared_ctx = UINT_MAX; 941 942 if (worker_per_query) 943 return reserve_dnschild_ctx(); 944 945 if (UINT_MAX == shared_ctx) 946 shared_ctx = reserve_dnschild_ctx(); 947 948 return shared_ctx; 949 } 950 951 952 static dnsworker_ctx * 953 get_worker_context( 954 blocking_child * c, 955 u_int idx 956 ) 957 { 958 u_int min_new_alloc; 959 u_int new_alloc; 960 size_t octets; 961 size_t new_octets; 962 dnsworker_ctx * retv; 963 964 worker_global_lock(TRUE); 965 966 if (dnsworker_contexts_alloc <= idx) { 967 min_new_alloc = 1 + idx; 968 /* round new_alloc up to nearest multiple of 4 */ 969 new_alloc = (min_new_alloc + 4) & ~(4 - 1); 970 new_octets = new_alloc * sizeof(dnsworker_ctx*); 971 octets = dnsworker_contexts_alloc * sizeof(dnsworker_ctx*); 972 dnsworker_contexts = erealloc_zero(dnsworker_contexts, 973 new_octets, octets); 974 dnsworker_contexts_alloc = new_alloc; 975 retv = emalloc_zero(sizeof(dnsworker_ctx)); 976 dnsworker_contexts[idx] = retv; 977 } else if (NULL == (retv = dnsworker_contexts[idx])) { 978 retv = emalloc_zero(sizeof(dnsworker_ctx)); 979 dnsworker_contexts[idx] = retv; 980 } 981 982 worker_global_lock(FALSE); 983 984 ZERO(*retv); 985 retv->c = c; 986 return retv; 987 } 988 989 990 static void 991 scheduled_sleep( 992 time_t scheduled, 993 time_t earliest, 994 dnsworker_ctx * worker_ctx 995 ) 996 { 997 time_t now; 998 999 if (scheduled < worker_ctx->ignore_scheduled_before) { 1000 TRACE(1, ("ignoring sleep until %s scheduled at %s (before %s)\n", 1001 humantime(earliest), humantime(scheduled), 1002 humantime(worker_ctx->ignore_scheduled_before))); 1003 return; 1004 } 1005 1006 now = time(NULL); 1007 1008 if (now < earliest) { 1009 TRACE(1, ("sleep until %s scheduled at %s (>= %s)\n", 1010 humantime(earliest), humantime(scheduled), 1011 humantime(worker_ctx->ignore_scheduled_before))); 1012 if (-1 == worker_sleep(worker_ctx->c, earliest - now)) { 1013 /* our sleep was interrupted */ 1014 now = time(NULL); 1015 worker_ctx->ignore_scheduled_before = now; 1016 #ifdef HAVE_RES_INIT 1017 worker_ctx->next_res_init = now + 60; 1018 next_res_init = worker_ctx->next_res_init; 1019 res_init(); 1020 #endif 1021 TRACE(1, ("sleep interrupted by daemon, ignoring sleeps scheduled before now (%s)\n", 1022 humantime(worker_ctx->ignore_scheduled_before))); 1023 } 1024 } 1025 } 1026 1027 1028 /* 1029 * manage_dns_retry_interval is a helper used by 1030 * getaddrinfo_sometime_complete and getnameinfo_sometime_complete 1031 * to calculate the new retry interval and schedule the next query. 1032 */ 1033 static void 1034 manage_dns_retry_interval( 1035 time_t * pscheduled, 1036 time_t * pwhen, 1037 int * pretry, 1038 time_t * pnext_timeslot 1039 ) 1040 { 1041 time_t now; 1042 time_t when; 1043 int retry; 1044 1045 now = time(NULL); 1046 retry = *pretry; 1047 when = max(now + retry, *pnext_timeslot); 1048 *pnext_timeslot = when; 1049 retry = min(64, retry << 1); 1050 1051 *pscheduled = now; 1052 *pwhen = when; 1053 *pretry = retry; 1054 } 1055 1056 /* 1057 * should_retry_dns is a helper used by getaddrinfo_sometime_complete 1058 * and getnameinfo_sometime_complete which implements ntpd's DNS retry 1059 * policy. 1060 */ 1061 static int 1062 should_retry_dns( 1063 int rescode, 1064 int res_errno 1065 ) 1066 { 1067 static int eai_again_seen; 1068 int again; 1069 #if defined (EAI_SYSTEM) && defined(DEBUG) 1070 char msg[256]; 1071 #endif 1072 1073 /* 1074 * If the resolver failed, see if the failure is 1075 * temporary. If so, return success. 1076 */ 1077 again = 0; 1078 1079 switch (rescode) { 1080 1081 case EAI_FAIL: 1082 again = 1; 1083 break; 1084 1085 case EAI_AGAIN: 1086 again = 1; 1087 eai_again_seen = 1; /* [Bug 1178] */ 1088 break; 1089 1090 case EAI_NONAME: 1091 #if defined(EAI_NODATA) && (EAI_NODATA != EAI_NONAME) 1092 case EAI_NODATA: 1093 #endif 1094 again = !eai_again_seen; /* [Bug 1178] */ 1095 break; 1096 1097 #ifdef EAI_SYSTEM 1098 case EAI_SYSTEM: 1099 /* 1100 * EAI_SYSTEM means the real error is in errno. We should be more 1101 * discriminating about which errno values require retrying, but 1102 * this matches existing behavior. 1103 */ 1104 again = 1; 1105 # ifdef DEBUG 1106 errno_to_str(res_errno, msg, sizeof(msg)); 1107 TRACE(1, ("intres: EAI_SYSTEM errno %d (%s) means try again, right?\n", 1108 res_errno, msg)); 1109 # endif 1110 break; 1111 #endif 1112 } 1113 1114 TRACE(2, ("intres: resolver returned: %s (%d), %sretrying\n", 1115 gai_strerror(rescode), rescode, again ? "" : "not ")); 1116 1117 return again; 1118 } 1119 1120 #else /* !WORKER follows */ 1121 int ntp_intres_nonempty_compilation_unit; 1122 #endif 1123