xref: /freebsd/contrib/ntp/libntp/ntp_intres.c (revision 09a53ad8f1318c5daae6cfb19d97f4f6459f0013)
1 /*
2  * ntp_intres.c - Implements a generic blocking worker child or thread,
3  *		  initially to provide a nonblocking solution for DNS
4  *		  name to address lookups available with getaddrinfo().
5  *
6  * This is a new implementation as of 2009 sharing the filename and
7  * very little else with the prior implementation, which used a
8  * temporary file to receive a single set of requests from the parent,
9  * and a NTP mode 7 authenticated request to push back responses.
10  *
11  * A primary goal in rewriting this code was the need to support the
12  * pool configuration directive's requirement to retrieve multiple
13  * addresses resolving a single name, which has previously been
14  * satisfied with blocking resolver calls from the ntpd mainline code.
15  *
16  * A secondary goal is to provide a generic mechanism for other
17  * blocking operations to be delegated to a worker using a common
18  * model for both Unix and Windows ntpd.  ntp_worker.c, work_fork.c,
19  * and work_thread.c implement the generic mechanism.  This file
20  * implements the two current consumers, getaddrinfo_sometime() and the
21  * presently unused getnameinfo_sometime().
22  *
23  * Both routines deliver results to a callback and manage memory
24  * allocation, meaning there is no freeaddrinfo_sometime().
25  *
26  * The initial implementation for Unix uses a pair of unidirectional
27  * pipes, one each for requests and responses, connecting the forked
28  * blocking child worker with the ntpd mainline.  The threaded code
29  * uses arrays of pointers to queue requests and responses.
30  *
31  * The parent drives the process, including scheduling sleeps between
32  * retries.
33  *
34  * Memory is managed differently for a child process, which mallocs
35  * request buffers to read from the pipe into, whereas the threaded
36  * code mallocs a copy of the request to hand off to the worker via
37  * the queueing array.  The resulting request buffer is free()d by
38  * platform-independent code.  A wrinkle is the request needs to be
39  * available to the requestor during response processing.
40  *
41  * Response memory allocation is also platform-dependent.  With a
42  * separate process and pipes, the response is free()d after being
43  * written to the pipe.  With threads, the same memory is handed
44  * over and the requestor frees it after processing is completed.
45  *
46  * The code should be generalized to support threads on Unix using
47  * much of the same code used for Windows initially.
48  *
49  */
50 #ifdef HAVE_CONFIG_H
51 # include <config.h>
52 #endif
53 
54 #include "ntp_workimpl.h"
55 
56 #ifdef WORKER
57 
58 #include <stdio.h>
59 #include <ctype.h>
60 #include <signal.h>
61 
62 /**/
63 #ifdef HAVE_SYS_TYPES_H
64 # include <sys/types.h>
65 #endif
66 #ifdef HAVE_NETINET_IN_H
67 #include <netinet/in.h>
68 #endif
69 #include <arpa/inet.h>
70 /**/
71 #ifdef HAVE_SYS_PARAM_H
72 # include <sys/param.h>
73 #endif
74 
75 #if !defined(HAVE_RES_INIT) && defined(HAVE___RES_INIT)
76 # define HAVE_RES_INIT
77 #endif
78 
79 #if defined(HAVE_RESOLV_H) && defined(HAVE_RES_INIT)
80 # ifdef HAVE_ARPA_NAMESER_H
81 #  include <arpa/nameser.h> /* DNS HEADER struct */
82 # endif
83 # ifdef HAVE_NETDB_H
84 #  include <netdb.h>
85 # endif
86 # include <resolv.h>
87 # ifdef HAVE_INT32_ONLY_WITH_DNS
88 #  define HAVE_INT32
89 # endif
90 # ifdef HAVE_U_INT32_ONLY_WITH_DNS
91 #  define HAVE_U_INT32
92 # endif
93 #endif
94 
95 #include "ntp.h"
96 #include "ntp_debug.h"
97 #include "ntp_malloc.h"
98 #include "ntp_syslog.h"
99 #include "ntp_unixtime.h"
100 #include "ntp_intres.h"
101 #include "intreswork.h"
102 
103 
104 /*
105  * Following are implementations of getaddrinfo_sometime() and
106  * getnameinfo_sometime().  Each is implemented in three routines:
107  *
108  * getaddrinfo_sometime()		getnameinfo_sometime()
109  * blocking_getaddrinfo()		blocking_getnameinfo()
110  * getaddrinfo_sometime_complete()	getnameinfo_sometime_complete()
111  *
112  * The first runs in the parent and marshalls (or serializes) request
113  * parameters into a request blob which is processed in the child by
114  * the second routine, blocking_*(), which serializes the results into
115  * a response blob unpacked by the third routine, *_complete(), which
116  * calls the callback routine provided with the request and frees
117  * _request_ memory allocated by the first routine.  Response memory
118  * is managed by the code which calls the *_complete routines.
119  */
120 
121 /* === typedefs === */
122 typedef struct blocking_gai_req_tag {	/* marshalled args */
123 	size_t			octets;
124 	u_int			dns_idx;
125 	time_t			scheduled;
126 	time_t			earliest;
127 	struct addrinfo		hints;
128 	int			retry;
129 	gai_sometime_callback	callback;
130 	void *			context;
131 	size_t			nodesize;
132 	size_t			servsize;
133 } blocking_gai_req;
134 
135 typedef struct blocking_gai_resp_tag {
136 	size_t			octets;
137 	int			retcode;
138 	int			retry;
139 	int			gai_errno; /* for EAI_SYSTEM case */
140 	int			ai_count;
141 	/*
142 	 * Followed by ai_count struct addrinfo and then ai_count
143 	 * sockaddr_u and finally the canonical name strings.
144 	 */
145 } blocking_gai_resp;
146 
147 typedef struct blocking_gni_req_tag {
148 	size_t			octets;
149 	u_int			dns_idx;
150 	time_t			scheduled;
151 	time_t			earliest;
152 	int			retry;
153 	size_t			hostoctets;
154 	size_t			servoctets;
155 	int			flags;
156 	gni_sometime_callback	callback;
157 	void *			context;
158 	sockaddr_u		socku;
159 } blocking_gni_req;
160 
161 typedef struct blocking_gni_resp_tag {
162 	size_t			octets;
163 	int			retcode;
164 	int			gni_errno; /* for EAI_SYSTEM case */
165 	int			retry;
166 	size_t			hostoctets;
167 	size_t			servoctets;
168 	/*
169 	 * Followed by hostoctets bytes of null-terminated host,
170 	 * then servoctets bytes of null-terminated service.
171 	 */
172 } blocking_gni_resp;
173 
174 /* per-DNS-worker state in parent */
175 typedef struct dnschild_ctx_tag {
176 	u_int	index;
177 	time_t	next_dns_timeslot;
178 } dnschild_ctx;
179 
180 /* per-DNS-worker state in worker */
181 typedef struct dnsworker_ctx_tag {
182 	blocking_child *	c;
183 	time_t			ignore_scheduled_before;
184 #ifdef HAVE_RES_INIT
185 	time_t	next_res_init;
186 #endif
187 } dnsworker_ctx;
188 
189 
190 /* === variables === */
191 dnschild_ctx **		dnschild_contexts;		/* parent */
192 u_int			dnschild_contexts_alloc;
193 dnsworker_ctx **	dnsworker_contexts;		/* child */
194 u_int			dnsworker_contexts_alloc;
195 
196 #ifdef HAVE_RES_INIT
197 static	time_t		next_res_init;
198 #endif
199 
200 
201 /* === forward declarations === */
202 static	u_int		reserve_dnschild_ctx(void);
203 static	u_int		get_dnschild_ctx(void);
204 static	dnsworker_ctx *	get_worker_context(blocking_child *, u_int);
205 static	void		scheduled_sleep(time_t, time_t,
206 					dnsworker_ctx *);
207 static	void		manage_dns_retry_interval(time_t *, time_t *,
208 						  int *,
209 						  time_t *);
210 static	int		should_retry_dns(int, int);
211 #ifdef HAVE_RES_INIT
212 static	void		reload_resolv_conf(dnsworker_ctx *);
213 #else
214 # define		reload_resolv_conf(wc)		\
215 	do {						\
216 		(void)(wc);				\
217 	} while (FALSE)
218 #endif
219 static	void		getaddrinfo_sometime_complete(blocking_work_req,
220 						      void *, size_t,
221 						      void *);
222 static	void		getnameinfo_sometime_complete(blocking_work_req,
223 						      void *, size_t,
224 						      void *);
225 
226 
227 /* === functions === */
228 /*
229  * getaddrinfo_sometime - uses blocking child to call getaddrinfo then
230  *			  invokes provided callback completion function.
231  */
232 int
233 getaddrinfo_sometime(
234 	const char *		node,
235 	const char *		service,
236 	const struct addrinfo *	hints,
237 	int			retry,
238 	gai_sometime_callback	callback,
239 	void *			context
240 	)
241 {
242 	blocking_gai_req *	gai_req;
243 	u_int			idx;
244 	dnschild_ctx *		child_ctx;
245 	size_t			req_size;
246 	size_t			nodesize;
247 	size_t			servsize;
248 	time_t			now;
249 
250 	REQUIRE(NULL != node);
251 	if (NULL != hints) {
252 		REQUIRE(0 == hints->ai_addrlen);
253 		REQUIRE(NULL == hints->ai_addr);
254 		REQUIRE(NULL == hints->ai_canonname);
255 		REQUIRE(NULL == hints->ai_next);
256 	}
257 
258 	idx = get_dnschild_ctx();
259 	child_ctx = dnschild_contexts[idx];
260 
261 	nodesize = strlen(node) + 1;
262 	servsize = strlen(service) + 1;
263 	req_size = sizeof(*gai_req) + nodesize + servsize;
264 
265 	gai_req = emalloc_zero(req_size);
266 
267 	gai_req->octets = req_size;
268 	gai_req->dns_idx = idx;
269 	now = time(NULL);
270 	gai_req->scheduled = now;
271 	gai_req->earliest = max(now, child_ctx->next_dns_timeslot);
272 	child_ctx->next_dns_timeslot = gai_req->earliest;
273 	if (hints != NULL)
274 		gai_req->hints = *hints;
275 	gai_req->retry = retry;
276 	gai_req->callback = callback;
277 	gai_req->context = context;
278 	gai_req->nodesize = nodesize;
279 	gai_req->servsize = servsize;
280 
281 	memcpy((char *)gai_req + sizeof(*gai_req), node, nodesize);
282 	memcpy((char *)gai_req + sizeof(*gai_req) + nodesize, service,
283 	       servsize);
284 
285 	if (queue_blocking_request(
286 		BLOCKING_GETADDRINFO,
287 		gai_req,
288 		req_size,
289 		&getaddrinfo_sometime_complete,
290 		gai_req)) {
291 
292 		msyslog(LOG_ERR, "unable to queue getaddrinfo request");
293 		errno = EFAULT;
294 		return -1;
295 	}
296 
297 	return 0;
298 }
299 
300 int
301 blocking_getaddrinfo(
302 	blocking_child *	c,
303 	blocking_pipe_header *	req
304 	)
305 {
306 	blocking_gai_req *	gai_req;
307 	dnsworker_ctx *		worker_ctx;
308 	blocking_pipe_header *	resp;
309 	blocking_gai_resp *	gai_resp;
310 	char *			node;
311 	char *			service;
312 	struct addrinfo *	ai_res;
313 	struct addrinfo *	ai;
314 	struct addrinfo *	serialized_ai;
315 	size_t			canons_octets;
316 	size_t			this_octets;
317 	size_t			resp_octets;
318 	char *			cp;
319 	time_t			time_now;
320 
321 	gai_req = (void *)((char *)req + sizeof(*req));
322 	node = (char *)gai_req + sizeof(*gai_req);
323 	service = node + gai_req->nodesize;
324 
325 	worker_ctx = get_worker_context(c, gai_req->dns_idx);
326 	scheduled_sleep(gai_req->scheduled, gai_req->earliest,
327 			worker_ctx);
328 	reload_resolv_conf(worker_ctx);
329 
330 	/*
331 	 * Take a shot at the final size, better to overestimate
332 	 * at first and then realloc to a smaller size.
333 	 */
334 
335 	resp_octets = sizeof(*resp) + sizeof(*gai_resp) +
336 		      16 * (sizeof(struct addrinfo) +
337 			    sizeof(sockaddr_u)) +
338 		      256;
339 	resp = emalloc_zero(resp_octets);
340 	gai_resp = (void *)(resp + 1);
341 
342 	TRACE(2, ("blocking_getaddrinfo given node %s serv %s fam %d flags %x\n",
343 		  node, service, gai_req->hints.ai_family,
344 		  gai_req->hints.ai_flags));
345 #ifdef DEBUG
346 	if (debug >= 2)
347 		fflush(stdout);
348 #endif
349 	ai_res = NULL;
350 	gai_resp->retcode = getaddrinfo(node, service, &gai_req->hints,
351 					&ai_res);
352 	gai_resp->retry = gai_req->retry;
353 #ifdef EAI_SYSTEM
354 	if (EAI_SYSTEM == gai_resp->retcode)
355 		gai_resp->gai_errno = errno;
356 #endif
357 	canons_octets = 0;
358 
359 	if (0 == gai_resp->retcode) {
360 		ai = ai_res;
361 		while (NULL != ai) {
362 			gai_resp->ai_count++;
363 			if (ai->ai_canonname)
364 				canons_octets += strlen(ai->ai_canonname) + 1;
365 			ai = ai->ai_next;
366 		}
367 		/*
368 		 * If this query succeeded only after retrying, DNS may have
369 		 * just become responsive.  Ignore previously-scheduled
370 		 * retry sleeps once for each pending request, similar to
371 		 * the way scheduled_sleep() does when its worker_sleep()
372 		 * is interrupted.
373 		 */
374 		if (gai_resp->retry > INITIAL_DNS_RETRY) {
375 			time_now = time(NULL);
376 			worker_ctx->ignore_scheduled_before = time_now;
377 			TRACE(1, ("DNS success after retry, ignoring sleeps scheduled before now (%s)\n",
378 				  humantime(time_now)));
379 		}
380 	}
381 
382 	/*
383 	 * Our response consists of a header, followed by ai_count
384 	 * addrinfo structs followed by ai_count sockaddr_storage
385 	 * structs followed by the canonical names.
386 	 */
387 	gai_resp->octets = sizeof(*gai_resp)
388 			    + gai_resp->ai_count
389 				* (sizeof(gai_req->hints)
390 				   + sizeof(sockaddr_u))
391 			    + canons_octets;
392 
393 	resp_octets = sizeof(*resp) + gai_resp->octets;
394 	resp = erealloc(resp, resp_octets);
395 	gai_resp = (void *)(resp + 1);
396 
397 	/* cp serves as our current pointer while serializing */
398 	cp = (void *)(gai_resp + 1);
399 	canons_octets = 0;
400 
401 	if (0 == gai_resp->retcode) {
402 		ai = ai_res;
403 		while (NULL != ai) {
404 			memcpy(cp, ai, sizeof(*ai));
405 			serialized_ai = (void *)cp;
406 			cp += sizeof(*ai);
407 
408 			/* transform ai_canonname into offset */
409 			if (NULL != serialized_ai->ai_canonname) {
410 				serialized_ai->ai_canonname = (char *)canons_octets;
411 				canons_octets += strlen(ai->ai_canonname) + 1;
412 			}
413 
414 			/* leave fixup of ai_addr pointer for receiver */
415 
416 			ai = ai->ai_next;
417 		}
418 
419 		ai = ai_res;
420 		while (NULL != ai) {
421 			INSIST(ai->ai_addrlen <= sizeof(sockaddr_u));
422 			memcpy(cp, ai->ai_addr, ai->ai_addrlen);
423 			cp += sizeof(sockaddr_u);
424 
425 			ai = ai->ai_next;
426 		}
427 
428 		ai = ai_res;
429 		while (NULL != ai) {
430 			if (NULL != ai->ai_canonname) {
431 				this_octets = strlen(ai->ai_canonname) + 1;
432 				memcpy(cp, ai->ai_canonname, this_octets);
433 				cp += this_octets;
434 			}
435 
436 			ai = ai->ai_next;
437 		}
438 		freeaddrinfo(ai_res);
439 	}
440 
441 	/*
442 	 * make sure our walk and earlier calc match
443 	 */
444 	DEBUG_INSIST((size_t)(cp - (char *)resp) == resp_octets);
445 
446 	if (queue_blocking_response(c, resp, resp_octets, req)) {
447 		msyslog(LOG_ERR, "blocking_getaddrinfo can not queue response");
448 		return -1;
449 	}
450 
451 	return 0;
452 }
453 
454 
455 static void
456 getaddrinfo_sometime_complete(
457 	blocking_work_req	rtype,
458 	void *			context,
459 	size_t			respsize,
460 	void *			resp
461 	)
462 {
463 	blocking_gai_req *	gai_req;
464 	blocking_gai_resp *	gai_resp;
465 	dnschild_ctx *		child_ctx;
466 	struct addrinfo *	ai;
467 	struct addrinfo *	next_ai;
468 	sockaddr_u *		psau;
469 	char *			node;
470 	char *			service;
471 	char *			canon_start;
472 	time_t			time_now;
473 	int			again;
474 	int			af;
475 	const char *		fam_spec;
476 	int			i;
477 
478 	gai_req = context;
479 	gai_resp = resp;
480 
481 	DEBUG_REQUIRE(BLOCKING_GETADDRINFO == rtype);
482 	DEBUG_REQUIRE(respsize == gai_resp->octets);
483 
484 	node = (char *)gai_req + sizeof(*gai_req);
485 	service = node + gai_req->nodesize;
486 
487 	child_ctx = dnschild_contexts[gai_req->dns_idx];
488 
489 	if (0 == gai_resp->retcode) {
490 		/*
491 		 * If this query succeeded only after retrying, DNS may have
492 		 * just become responsive.
493 		 */
494 		if (gai_resp->retry > INITIAL_DNS_RETRY) {
495 			time_now = time(NULL);
496 			child_ctx->next_dns_timeslot = time_now;
497 			TRACE(1, ("DNS success after retry, %u next_dns_timeslot reset (%s)\n",
498 				  gai_req->dns_idx, humantime(time_now)));
499 		}
500 	} else {
501 		again = should_retry_dns(gai_resp->retcode,
502 					 gai_resp->gai_errno);
503 		/*
504 		 * exponential backoff of DNS retries to 64s
505 		 */
506 		if (gai_req->retry > 0 && again) {
507 			/* log the first retry only */
508 			if (INITIAL_DNS_RETRY == gai_req->retry)
509 				NLOG(NLOG_SYSINFO) {
510 					af = gai_req->hints.ai_family;
511 					fam_spec = (AF_INET6 == af)
512 						       ? " (AAAA)"
513 						       : (AF_INET == af)
514 							     ? " (A)"
515 							     : "";
516 #ifdef EAI_SYSTEM
517 					if (EAI_SYSTEM == gai_resp->retcode) {
518 						errno = gai_resp->gai_errno;
519 						msyslog(LOG_INFO,
520 							"retrying DNS %s%s: EAI_SYSTEM %d: %m",
521 							node, fam_spec,
522 							gai_resp->gai_errno);
523 					} else
524 #endif
525 						msyslog(LOG_INFO,
526 							"retrying DNS %s%s: %s (%d)",
527 							node, fam_spec,
528 							gai_strerror(gai_resp->retcode),
529 							gai_resp->retcode);
530 				}
531 			manage_dns_retry_interval(&gai_req->scheduled,
532 			    &gai_req->earliest, &gai_req->retry,
533 			    &child_ctx->next_dns_timeslot);
534 			if (!queue_blocking_request(
535 					BLOCKING_GETADDRINFO,
536 					gai_req,
537 					gai_req->octets,
538 					&getaddrinfo_sometime_complete,
539 					gai_req))
540 				return;
541 			else
542 				msyslog(LOG_ERR,
543 					"unable to retry hostname %s",
544 					node);
545 		}
546 	}
547 
548 	/*
549 	 * fixup pointers in returned addrinfo array
550 	 */
551 	ai = (void *)((char *)gai_resp + sizeof(*gai_resp));
552 	next_ai = NULL;
553 	for (i = gai_resp->ai_count - 1; i >= 0; i--) {
554 		ai[i].ai_next = next_ai;
555 		next_ai = &ai[i];
556 	}
557 
558 	psau = (void *)((char *)ai + gai_resp->ai_count * sizeof(*ai));
559 	canon_start = (char *)psau + gai_resp->ai_count * sizeof(*psau);
560 
561 	for (i = 0; i < gai_resp->ai_count; i++) {
562 		if (NULL != ai[i].ai_addr)
563 			ai[i].ai_addr = &psau->sa;
564 		psau++;
565 		if (NULL != ai[i].ai_canonname)
566 			ai[i].ai_canonname += (size_t)canon_start;
567 	}
568 
569 	ENSURE((char *)psau == canon_start);
570 
571 	if (!gai_resp->ai_count)
572 		ai = NULL;
573 
574 	(*gai_req->callback)(gai_resp->retcode, gai_resp->gai_errno,
575 			     gai_req->context, node, service,
576 			     &gai_req->hints, ai);
577 
578 	free(gai_req);
579 	/* gai_resp is part of block freed by process_blocking_resp() */
580 }
581 
582 
583 #ifdef TEST_BLOCKING_WORKER
584 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)
585 {
586 	sockaddr_u addr;
587 
588 	if (rescode) {
589 		TRACE(1, ("gai_test_callback context %p error rescode %d %s serv %s\n",
590 			  context, rescode, name, service));
591 		return;
592 	}
593 	while (!rescode && NULL != ai_res) {
594 		ZERO_SOCK(&addr);
595 		memcpy(&addr, ai_res->ai_addr, ai_res->ai_addrlen);
596 		TRACE(1, ("ctx %p fam %d addr %s canon '%s' type %s at %p ai_addr %p ai_next %p\n",
597 			  context,
598 			  AF(&addr),
599 			  stoa(&addr),
600 			  (ai_res->ai_canonname)
601 			      ? ai_res->ai_canonname
602 			      : "",
603 			  (SOCK_DGRAM == ai_res->ai_socktype)
604 			      ? "DGRAM"
605 			      : (SOCK_STREAM == ai_res->ai_socktype)
606 				    ? "STREAM"
607 				    : "(other)",
608 			  ai_res,
609 			  ai_res->ai_addr,
610 			  ai_res->ai_next));
611 
612 		getnameinfo_sometime((sockaddr_u *)ai_res->ai_addr, 128, 32, 0, gni_test_callback, context);
613 
614 		ai_res = ai_res->ai_next;
615 	}
616 }
617 #endif	/* TEST_BLOCKING_WORKER */
618 
619 
620 int
621 getnameinfo_sometime(
622 	sockaddr_u *		psau,
623 	size_t			hostoctets,
624 	size_t			servoctets,
625 	int			flags,
626 	gni_sometime_callback	callback,
627 	void *			context
628 	)
629 {
630 	blocking_gni_req *	gni_req;
631 	u_int			idx;
632 	dnschild_ctx *		child_ctx;
633 	time_t			time_now;
634 
635 	REQUIRE(hostoctets);
636 	REQUIRE(hostoctets + servoctets < 1024);
637 
638 	idx = get_dnschild_ctx();
639 	child_ctx = dnschild_contexts[idx];
640 
641 	gni_req = emalloc_zero(sizeof(*gni_req));
642 
643 	gni_req->octets = sizeof(*gni_req);
644 	gni_req->dns_idx = idx;
645 	time_now = time(NULL);
646 	gni_req->scheduled = time_now;
647 	gni_req->earliest = max(time_now, child_ctx->next_dns_timeslot);
648 	child_ctx->next_dns_timeslot = gni_req->earliest;
649 	memcpy(&gni_req->socku, psau, SOCKLEN(psau));
650 	gni_req->hostoctets = hostoctets;
651 	gni_req->servoctets = servoctets;
652 	gni_req->flags = flags;
653 	gni_req->retry = INITIAL_DNS_RETRY;
654 	gni_req->callback = callback;
655 	gni_req->context = context;
656 
657 	if (queue_blocking_request(
658 		BLOCKING_GETNAMEINFO,
659 		gni_req,
660 		sizeof(*gni_req),
661 		&getnameinfo_sometime_complete,
662 		gni_req)) {
663 
664 		msyslog(LOG_ERR, "unable to queue getnameinfo request");
665 		errno = EFAULT;
666 		return -1;
667 	}
668 
669 	return 0;
670 }
671 
672 
673 int
674 blocking_getnameinfo(
675 	blocking_child *	c,
676 	blocking_pipe_header *	req
677 	)
678 {
679 	blocking_gni_req *	gni_req;
680 	dnsworker_ctx *		worker_ctx;
681 	blocking_pipe_header *	resp;
682 	blocking_gni_resp *	gni_resp;
683 	size_t			octets;
684 	size_t			resp_octets;
685 	char *			service;
686 	char *			cp;
687 	int			rc;
688 	time_t			time_now;
689 	char			host[1024];
690 
691 	gni_req = (void *)((char *)req + sizeof(*req));
692 
693 	octets = gni_req->hostoctets + gni_req->servoctets;
694 
695 	/*
696 	 * Some alloca() implementations are fragile regarding
697 	 * large allocations.  We only need room for the host
698 	 * and service names.
699 	 */
700 	REQUIRE(octets < sizeof(host));
701 	service = host + gni_req->hostoctets;
702 
703 	worker_ctx = get_worker_context(c, gni_req->dns_idx);
704 	scheduled_sleep(gni_req->scheduled, gni_req->earliest,
705 			worker_ctx);
706 	reload_resolv_conf(worker_ctx);
707 
708 	/*
709 	 * Take a shot at the final size, better to overestimate
710 	 * then realloc to a smaller size.
711 	 */
712 
713 	resp_octets = sizeof(*resp) + sizeof(*gni_resp) + octets;
714 	resp = emalloc_zero(resp_octets);
715 	gni_resp = (void *)((char *)resp + sizeof(*resp));
716 
717 	TRACE(2, ("blocking_getnameinfo given addr %s flags 0x%x hostlen %lu servlen %lu\n",
718 		  stoa(&gni_req->socku), gni_req->flags,
719 		  (u_long)gni_req->hostoctets, (u_long)gni_req->servoctets));
720 
721 	gni_resp->retcode = getnameinfo(&gni_req->socku.sa,
722 					SOCKLEN(&gni_req->socku),
723 					host,
724 					gni_req->hostoctets,
725 					service,
726 					gni_req->servoctets,
727 					gni_req->flags);
728 	gni_resp->retry = gni_req->retry;
729 #ifdef EAI_SYSTEM
730 	if (EAI_SYSTEM == gni_resp->retcode)
731 		gni_resp->gni_errno = errno;
732 #endif
733 
734 	if (0 != gni_resp->retcode) {
735 		gni_resp->hostoctets = 0;
736 		gni_resp->servoctets = 0;
737 	} else {
738 		gni_resp->hostoctets = strlen(host) + 1;
739 		gni_resp->servoctets = strlen(service) + 1;
740 		/*
741 		 * If this query succeeded only after retrying, DNS may have
742 		 * just become responsive.  Ignore previously-scheduled
743 		 * retry sleeps once for each pending request, similar to
744 		 * the way scheduled_sleep() does when its worker_sleep()
745 		 * is interrupted.
746 		 */
747 		if (gni_req->retry > INITIAL_DNS_RETRY) {
748 			time_now = time(NULL);
749 			worker_ctx->ignore_scheduled_before = time_now;
750 			TRACE(1, ("DNS success after retrying, ignoring sleeps scheduled before now (%s)\n",
751 				humantime(time_now)));
752 		}
753 	}
754 	octets = gni_resp->hostoctets + gni_resp->servoctets;
755 	/*
756 	 * Our response consists of a header, followed by the host and
757 	 * service strings, each null-terminated.
758 	 */
759 	resp_octets = sizeof(*resp) + sizeof(*gni_resp) + octets;
760 
761 	resp = erealloc(resp, resp_octets);
762 	gni_resp = (void *)(resp + 1);
763 
764 	gni_resp->octets = sizeof(*gni_resp) + octets;
765 
766 	/* cp serves as our current pointer while serializing */
767 	cp = (void *)(gni_resp + 1);
768 
769 	if (0 == gni_resp->retcode) {
770 		memcpy(cp, host, gni_resp->hostoctets);
771 		cp += gni_resp->hostoctets;
772 		memcpy(cp, service, gni_resp->servoctets);
773 		cp += gni_resp->servoctets;
774 	}
775 
776 	INSIST((size_t)(cp - (char *)resp) == resp_octets);
777 	INSIST(resp_octets - sizeof(*resp) == gni_resp->octets);
778 
779 	rc = queue_blocking_response(c, resp, resp_octets, req);
780 	if (rc)
781 		msyslog(LOG_ERR, "blocking_getnameinfo unable to queue response");
782 	return rc;
783 }
784 
785 
786 static void
787 getnameinfo_sometime_complete(
788 	blocking_work_req	rtype,
789 	void *			context,
790 	size_t			respsize,
791 	void *			resp
792 	)
793 {
794 	blocking_gni_req *	gni_req;
795 	blocking_gni_resp *	gni_resp;
796 	dnschild_ctx *		child_ctx;
797 	char *			host;
798 	char *			service;
799 	time_t			time_now;
800 	int			again;
801 
802 	gni_req = context;
803 	gni_resp = resp;
804 
805 	DEBUG_REQUIRE(BLOCKING_GETNAMEINFO == rtype);
806 	DEBUG_REQUIRE(respsize == gni_resp->octets);
807 
808 	child_ctx = dnschild_contexts[gni_req->dns_idx];
809 
810 	if (0 == gni_resp->retcode) {
811 		/*
812 		 * If this query succeeded only after retrying, DNS may have
813 		 * just become responsive.
814 		 */
815 		if (gni_resp->retry > INITIAL_DNS_RETRY) {
816 			time_now = time(NULL);
817 			child_ctx->next_dns_timeslot = time_now;
818 			TRACE(1, ("DNS success after retry, %u next_dns_timeslot reset (%s)\n",
819 				  gni_req->dns_idx, humantime(time_now)));
820 		}
821 	} else {
822 		again = should_retry_dns(gni_resp->retcode, gni_resp->gni_errno);
823 		/*
824 		 * exponential backoff of DNS retries to 64s
825 		 */
826 		if (gni_req->retry > 0)
827 			manage_dns_retry_interval(&gni_req->scheduled,
828 			    &gni_req->earliest, &gni_req->retry,
829 			    &child_ctx->next_dns_timeslot);
830 
831 		if (gni_req->retry > 0 && again) {
832 			if (!queue_blocking_request(
833 				BLOCKING_GETNAMEINFO,
834 				gni_req,
835 				gni_req->octets,
836 				&getnameinfo_sometime_complete,
837 				gni_req))
838 				return;
839 
840 			msyslog(LOG_ERR, "unable to retry reverse lookup of %s", stoa(&gni_req->socku));
841 		}
842 	}
843 
844 	if (!gni_resp->hostoctets) {
845 		host = NULL;
846 		service = NULL;
847 	} else {
848 		host = (char *)gni_resp + sizeof(*gni_resp);
849 		service = (gni_resp->servoctets)
850 			      ? host + gni_resp->hostoctets
851 			      : NULL;
852 	}
853 
854 	(*gni_req->callback)(gni_resp->retcode, gni_resp->gni_errno,
855 			     &gni_req->socku, gni_req->flags, host,
856 			     service, gni_req->context);
857 
858 	free(gni_req);
859 	/* gni_resp is part of block freed by process_blocking_resp() */
860 }
861 
862 
863 #ifdef TEST_BLOCKING_WORKER
864 void gni_test_callback(int rescode, int gni_errno, sockaddr_u *psau, int flags, const char *host, const char *service, void *context)
865 {
866 	if (!rescode)
867 		TRACE(1, ("gni_test_callback got host '%s' serv '%s' for addr %s context %p\n",
868 			  host, service, stoa(psau), context));
869 	else
870 		TRACE(1, ("gni_test_callback context %p rescode %d gni_errno %d flags 0x%x addr %s\n",
871 			  context, rescode, gni_errno, flags, stoa(psau)));
872 }
873 #endif	/* TEST_BLOCKING_WORKER */
874 
875 
876 #ifdef HAVE_RES_INIT
877 static void
878 reload_resolv_conf(
879 	dnsworker_ctx *	worker_ctx
880 	)
881 {
882 	time_t	time_now;
883 
884 	/*
885 	 * This is ad-hoc.  Reload /etc/resolv.conf once per minute
886 	 * to pick up on changes from the DHCP client.  [Bug 1226]
887 	 * When using threads for the workers, this needs to happen
888 	 * only once per minute process-wide.
889 	 */
890 	time_now = time(NULL);
891 # ifdef WORK_THREAD
892 	worker_ctx->next_res_init = next_res_init;
893 # endif
894 	if (worker_ctx->next_res_init <= time_now) {
895 		if (worker_ctx->next_res_init != 0)
896 			res_init();
897 		worker_ctx->next_res_init = time_now + 60;
898 # ifdef WORK_THREAD
899 		next_res_init = worker_ctx->next_res_init;
900 # endif
901 	}
902 }
903 #endif	/* HAVE_RES_INIT */
904 
905 
906 static u_int
907 reserve_dnschild_ctx(void)
908 {
909 	const size_t	ps = sizeof(dnschild_contexts[0]);
910 	const size_t	cs = sizeof(*dnschild_contexts[0]);
911 	u_int		c;
912 	u_int		new_alloc;
913 	size_t		octets;
914 	size_t		new_octets;
915 
916 	c = 0;
917 	while (TRUE) {
918 		for ( ; c < dnschild_contexts_alloc; c++) {
919 			if (NULL == dnschild_contexts[c]) {
920 				dnschild_contexts[c] = emalloc_zero(cs);
921 
922 				return c;
923 			}
924 		}
925 		new_alloc = dnschild_contexts_alloc + 20;
926 		new_octets = new_alloc * ps;
927 		octets = dnschild_contexts_alloc * ps;
928 		dnschild_contexts = erealloc_zero(dnschild_contexts,
929 						  new_octets, octets);
930 		dnschild_contexts_alloc = new_alloc;
931 	}
932 }
933 
934 
935 static u_int
936 get_dnschild_ctx(void)
937 {
938 	static u_int	shared_ctx = UINT_MAX;
939 
940 	if (worker_per_query)
941 		return reserve_dnschild_ctx();
942 
943 	if (UINT_MAX == shared_ctx)
944 		shared_ctx = reserve_dnschild_ctx();
945 
946 	return shared_ctx;
947 }
948 
949 
950 static dnsworker_ctx *
951 get_worker_context(
952 	blocking_child *	c,
953 	u_int			idx
954 	)
955 {
956 	u_int		min_new_alloc;
957 	u_int		new_alloc;
958 	size_t		octets;
959 	size_t		new_octets;
960 	dnsworker_ctx *	retv;
961 
962 	worker_global_lock(TRUE);
963 
964 	if (dnsworker_contexts_alloc <= idx) {
965 		min_new_alloc = 1 + idx;
966 		/* round new_alloc up to nearest multiple of 4 */
967 		new_alloc = (min_new_alloc + 4) & ~(4 - 1);
968 		new_octets = new_alloc * sizeof(dnsworker_ctx*);
969 		octets = dnsworker_contexts_alloc * sizeof(dnsworker_ctx*);
970 		dnsworker_contexts = erealloc_zero(dnsworker_contexts,
971 						   new_octets, octets);
972 		dnsworker_contexts_alloc = new_alloc;
973 		retv = emalloc_zero(sizeof(dnsworker_ctx));
974 		dnsworker_contexts[idx] = retv;
975 	} else if (NULL == (retv = dnsworker_contexts[idx])) {
976 		retv = emalloc_zero(sizeof(dnsworker_ctx));
977 		dnsworker_contexts[idx] = retv;
978 	}
979 
980 	worker_global_lock(FALSE);
981 
982 	ZERO(*retv);
983 	retv->c = c;
984 	return retv;
985 }
986 
987 
988 static void
989 scheduled_sleep(
990 	time_t		scheduled,
991 	time_t		earliest,
992 	dnsworker_ctx *	worker_ctx
993 	)
994 {
995 	time_t now;
996 
997 	if (scheduled < worker_ctx->ignore_scheduled_before) {
998 		TRACE(1, ("ignoring sleep until %s scheduled at %s (before %s)\n",
999 			  humantime(earliest), humantime(scheduled),
1000 			  humantime(worker_ctx->ignore_scheduled_before)));
1001 		return;
1002 	}
1003 
1004 	now = time(NULL);
1005 
1006 	if (now < earliest) {
1007 		TRACE(1, ("sleep until %s scheduled at %s (>= %s)\n",
1008 			  humantime(earliest), humantime(scheduled),
1009 			  humantime(worker_ctx->ignore_scheduled_before)));
1010 		if (-1 == worker_sleep(worker_ctx->c, earliest - now)) {
1011 			/* our sleep was interrupted */
1012 			now = time(NULL);
1013 			worker_ctx->ignore_scheduled_before = now;
1014 #ifdef HAVE_RES_INIT
1015 			worker_ctx->next_res_init = now + 60;
1016 			next_res_init = worker_ctx->next_res_init;
1017 			res_init();
1018 #endif
1019 			TRACE(1, ("sleep interrupted by daemon, ignoring sleeps scheduled before now (%s)\n",
1020 				  humantime(worker_ctx->ignore_scheduled_before)));
1021 		}
1022 	}
1023 }
1024 
1025 
1026 /*
1027  * manage_dns_retry_interval is a helper used by
1028  * getaddrinfo_sometime_complete and getnameinfo_sometime_complete
1029  * to calculate the new retry interval and schedule the next query.
1030  */
1031 static void
1032 manage_dns_retry_interval(
1033 	time_t *	pscheduled,
1034 	time_t *	pwhen,
1035 	int *		pretry,
1036 	time_t *	pnext_timeslot
1037 	)
1038 {
1039 	time_t	now;
1040 	time_t	when;
1041 	int	retry;
1042 
1043 	now = time(NULL);
1044 	retry = *pretry;
1045 	when = max(now + retry, *pnext_timeslot);
1046 	*pnext_timeslot = when;
1047 	retry = min(64, retry << 1);
1048 
1049 	*pscheduled = now;
1050 	*pwhen = when;
1051 	*pretry = retry;
1052 }
1053 
1054 /*
1055  * should_retry_dns is a helper used by getaddrinfo_sometime_complete
1056  * and getnameinfo_sometime_complete which implements ntpd's DNS retry
1057  * policy.
1058  */
1059 static int
1060 should_retry_dns(
1061 	int	rescode,
1062 	int	res_errno
1063 	)
1064 {
1065 	static int	eai_again_seen;
1066 	int		again;
1067 #if defined (EAI_SYSTEM) && defined(DEBUG)
1068 	char		msg[256];
1069 #endif
1070 
1071 	/*
1072 	 * If the resolver failed, see if the failure is
1073 	 * temporary. If so, return success.
1074 	 */
1075 	again = 0;
1076 
1077 	switch (rescode) {
1078 
1079 	case EAI_FAIL:
1080 		again = 1;
1081 		break;
1082 
1083 	case EAI_AGAIN:
1084 		again = 1;
1085 		eai_again_seen = 1;		/* [Bug 1178] */
1086 		break;
1087 
1088 	case EAI_NONAME:
1089 #if defined(EAI_NODATA) && (EAI_NODATA != EAI_NONAME)
1090 	case EAI_NODATA:
1091 #endif
1092 		again = !eai_again_seen;	/* [Bug 1178] */
1093 		break;
1094 
1095 #ifdef EAI_SYSTEM
1096 	case EAI_SYSTEM:
1097 		/*
1098 		 * EAI_SYSTEM means the real error is in errno.  We should be more
1099 		 * discriminating about which errno values require retrying, but
1100 		 * this matches existing behavior.
1101 		 */
1102 		again = 1;
1103 # ifdef DEBUG
1104 		errno_to_str(res_errno, msg, sizeof(msg));
1105 		TRACE(1, ("intres: EAI_SYSTEM errno %d (%s) means try again, right?\n",
1106 			  res_errno, msg));
1107 # endif
1108 		break;
1109 #endif
1110 	}
1111 
1112 	TRACE(2, ("intres: resolver returned: %s (%d), %sretrying\n",
1113 		  gai_strerror(rescode), rescode, again ? "" : "not "));
1114 
1115 	return again;
1116 }
1117 
1118 #else	/* !WORKER follows */
1119 int ntp_intres_nonempty_compilation_unit;
1120 #endif
1121