]> git.ipfire.org Git - thirdparty/glibc.git/blob - resolv/res_send.c
Change most internal uses of __gettimeofday to __clock_gettime.
[thirdparty/glibc.git] / resolv / res_send.c
1 /* Copyright (C) 2016-2019 Free Software Foundation, Inc.
2 This file is part of the GNU C Library.
3
4 The GNU C Library is free software; you can redistribute it and/or
5 modify it under the terms of the GNU Lesser General Public
6 License as published by the Free Software Foundation; either
7 version 2.1 of the License, or (at your option) any later version.
8
9 The GNU C Library is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 Lesser General Public License for more details.
13
14 You should have received a copy of the GNU Lesser General Public
15 License along with the GNU C Library; if not, see
16 <https://www.gnu.org/licenses/>. */
17
18 /*
19 * Copyright (c) 1985, 1989, 1993
20 * The Regents of the University of California. All rights reserved.
21 *
22 * Redistribution and use in source and binary forms, with or without
23 * modification, are permitted provided that the following conditions
24 * are met:
25 * 1. Redistributions of source code must retain the above copyright
26 * notice, this list of conditions and the following disclaimer.
27 * 2. Redistributions in binary form must reproduce the above copyright
28 * notice, this list of conditions and the following disclaimer in the
29 * documentation and/or other materials provided with the distribution.
30 * 4. Neither the name of the University nor the names of its contributors
31 * may be used to endorse or promote products derived from this software
32 * without specific prior written permission.
33 *
34 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
35 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
36 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
37 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
38 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
39 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
40 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
41 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
42 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
43 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
44 * SUCH DAMAGE.
45 */
46
47 /*
48 * Portions Copyright (c) 1993 by Digital Equipment Corporation.
49 *
50 * Permission to use, copy, modify, and distribute this software for any
51 * purpose with or without fee is hereby granted, provided that the above
52 * copyright notice and this permission notice appear in all copies, and that
53 * the name of Digital Equipment Corporation not be used in advertising or
54 * publicity pertaining to distribution of the document or software without
55 * specific, written prior permission.
56 *
57 * THE SOFTWARE IS PROVIDED "AS IS" AND DIGITAL EQUIPMENT CORP. DISCLAIMS ALL
58 * WARRANTIES WITH REGARD TO THIS SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES
59 * OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL DIGITAL EQUIPMENT
60 * CORPORATION BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL
61 * DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR
62 * PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS
63 * ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS
64 * SOFTWARE.
65 */
66
67 /*
68 * Portions Copyright (c) 1996-1999 by Internet Software Consortium.
69 *
70 * Permission to use, copy, modify, and distribute this software for any
71 * purpose with or without fee is hereby granted, provided that the above
72 * copyright notice and this permission notice appear in all copies.
73 *
74 * THE SOFTWARE IS PROVIDED "AS IS" AND INTERNET SOFTWARE CONSORTIUM DISCLAIMS
75 * ALL WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES
76 * OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL INTERNET SOFTWARE
77 * CONSORTIUM BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL
78 * DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR
79 * PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS
80 * ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS
81 * SOFTWARE.
82 */
83
84 /*
85 * Send query to name server and wait for reply.
86 */
87
88 #include <assert.h>
89 #include <sys/types.h>
90 #include <sys/param.h>
91 #include <sys/time.h>
92 #include <sys/socket.h>
93 #include <sys/uio.h>
94 #include <sys/poll.h>
95
96 #include <netinet/in.h>
97 #include <arpa/nameser.h>
98 #include <arpa/inet.h>
99 #include <sys/ioctl.h>
100
101 #include <errno.h>
102 #include <fcntl.h>
103 #include <netdb.h>
104 #include <resolv/resolv-internal.h>
105 #include <resolv/resolv_context.h>
106 #include <signal.h>
107 #include <stdlib.h>
108 #include <string.h>
109 #include <unistd.h>
110 #include <kernel-features.h>
111 #include <libc-diag.h>
112 #include <random-bits.h>
113
114 #if PACKETSZ > 65536
115 #define MAXPACKET PACKETSZ
116 #else
117 #define MAXPACKET 65536
118 #endif
119
120 /* From ev_streams.c. */
121
122 static inline void
123 __attribute ((always_inline))
124 evConsIovec(void *buf, size_t cnt, struct iovec *vec) {
125 memset(vec, 0xf5, sizeof (*vec));
126 vec->iov_base = buf;
127 vec->iov_len = cnt;
128 }
129
130 /* From ev_timers.c. */
131
132 #define BILLION 1000000000
133
134 static inline void
135 evConsTime(struct timespec *res, time_t sec, long nsec) {
136 res->tv_sec = sec;
137 res->tv_nsec = nsec;
138 }
139
140 static inline void
141 evAddTime(struct timespec *res, const struct timespec *addend1,
142 const struct timespec *addend2) {
143 res->tv_sec = addend1->tv_sec + addend2->tv_sec;
144 res->tv_nsec = addend1->tv_nsec + addend2->tv_nsec;
145 if (res->tv_nsec >= BILLION) {
146 res->tv_sec++;
147 res->tv_nsec -= BILLION;
148 }
149 }
150
151 static inline void
152 evSubTime(struct timespec *res, const struct timespec *minuend,
153 const struct timespec *subtrahend) {
154 res->tv_sec = minuend->tv_sec - subtrahend->tv_sec;
155 if (minuend->tv_nsec >= subtrahend->tv_nsec)
156 res->tv_nsec = minuend->tv_nsec - subtrahend->tv_nsec;
157 else {
158 res->tv_nsec = (BILLION
159 - subtrahend->tv_nsec + minuend->tv_nsec);
160 res->tv_sec--;
161 }
162 }
163
164 static int
165 evCmpTime(struct timespec a, struct timespec b) {
166 long x = a.tv_sec - b.tv_sec;
167
168 if (x == 0L)
169 x = a.tv_nsec - b.tv_nsec;
170 return (x < 0L ? (-1) : x > 0L ? (1) : (0));
171 }
172
173 static void
174 evNowTime(struct timespec *res) {
175 __clock_gettime(CLOCK_REALTIME, res);
176 }
177
178
179 #define EXT(res) ((res)->_u._ext)
180
181 /* Forward. */
182
183 static struct sockaddr *get_nsaddr (res_state, unsigned int);
184 static int send_vc(res_state, const u_char *, int,
185 const u_char *, int,
186 u_char **, int *, int *, int, u_char **,
187 u_char **, int *, int *, int *);
188 static int send_dg(res_state, const u_char *, int,
189 const u_char *, int,
190 u_char **, int *, int *, int,
191 int *, int *, u_char **,
192 u_char **, int *, int *, int *);
193 static int sock_eq(struct sockaddr_in6 *, struct sockaddr_in6 *);
194
195 /* Public. */
196
197 /* int
198 * res_isourserver(ina)
199 * looks up "ina" in _res.ns_addr_list[]
200 * returns:
201 * 0 : not found
202 * >0 : found
203 * author:
204 * paul vixie, 29may94
205 */
206 int
207 res_ourserver_p(const res_state statp, const struct sockaddr_in6 *inp)
208 {
209 int ns;
210
211 if (inp->sin6_family == AF_INET) {
212 struct sockaddr_in *in4p = (struct sockaddr_in *) inp;
213 in_port_t port = in4p->sin_port;
214 in_addr_t addr = in4p->sin_addr.s_addr;
215
216 for (ns = 0; ns < statp->nscount; ns++) {
217 const struct sockaddr_in *srv =
218 (struct sockaddr_in *) get_nsaddr (statp, ns);
219
220 if ((srv->sin_family == AF_INET) &&
221 (srv->sin_port == port) &&
222 (srv->sin_addr.s_addr == INADDR_ANY ||
223 srv->sin_addr.s_addr == addr))
224 return (1);
225 }
226 } else if (inp->sin6_family == AF_INET6) {
227 for (ns = 0; ns < statp->nscount; ns++) {
228 const struct sockaddr_in6 *srv
229 = (struct sockaddr_in6 *) get_nsaddr (statp, ns);
230 if ((srv->sin6_family == AF_INET6) &&
231 (srv->sin6_port == inp->sin6_port) &&
232 !(memcmp(&srv->sin6_addr, &in6addr_any,
233 sizeof (struct in6_addr)) &&
234 memcmp(&srv->sin6_addr, &inp->sin6_addr,
235 sizeof (struct in6_addr))))
236 return (1);
237 }
238 }
239 return (0);
240 }
241
242 int
243 res_isourserver (const struct sockaddr_in *inp)
244 {
245 return res_ourserver_p (&_res, (const struct sockaddr_in6 *) inp);
246 }
247
248 /* int
249 * res_nameinquery(name, type, class, buf, eom)
250 * look for (name,type,class) in the query section of packet (buf,eom)
251 * requires:
252 * buf + HFIXEDSZ <= eom
253 * returns:
254 * -1 : format error
255 * 0 : not found
256 * >0 : found
257 * author:
258 * paul vixie, 29may94
259 */
260 int
261 res_nameinquery(const char *name, int type, int class,
262 const u_char *buf, const u_char *eom)
263 {
264 const u_char *cp = buf + HFIXEDSZ;
265 int qdcount = ntohs(((HEADER*)buf)->qdcount);
266
267 while (qdcount-- > 0) {
268 char tname[MAXDNAME+1];
269 int n, ttype, tclass;
270
271 n = dn_expand(buf, eom, cp, tname, sizeof tname);
272 if (n < 0)
273 return (-1);
274 cp += n;
275 if (cp + 2 * INT16SZ > eom)
276 return (-1);
277 NS_GET16(ttype, cp);
278 NS_GET16(tclass, cp);
279 if (ttype == type && tclass == class &&
280 ns_samename(tname, name) == 1)
281 return (1);
282 }
283 return (0);
284 }
285 libresolv_hidden_def (res_nameinquery)
286
287 /* Returns a shift value for the name server index. Used to implement
288 RES_ROTATE. */
289 static unsigned int
290 nameserver_offset (struct __res_state *statp)
291 {
292 /* If we only have one name server or rotation is disabled, return
293 offset 0 (no rotation). */
294 unsigned int nscount = statp->nscount;
295 if (nscount <= 1 || !(statp->options & RES_ROTATE))
296 return 0;
297
298 /* Global offset. The lowest bit indicates whether the offset has
299 been initialized with a random value. Use relaxed MO to access
300 global_offset because all we need is a sequence of roughly
301 sequential value. */
302 static unsigned int global_offset;
303 unsigned int offset = atomic_fetch_add_relaxed (&global_offset, 2);
304 if ((offset & 1) == 0)
305 {
306 /* Initialization is required. */
307 offset = random_bits ();
308 /* The lowest bit is the most random. Preserve it. */
309 offset <<= 1;
310
311 /* Store the new starting value. atomic_fetch_add_relaxed
312 returns the old value, so emulate that by storing the new
313 (incremented) value. Concurrent initialization with
314 different random values is harmless. */
315 atomic_store_relaxed (&global_offset, (offset | 1) + 2);
316 }
317
318 /* Remove the initialization bit. */
319 offset >>= 1;
320
321 /* Avoid the division in the most common cases. */
322 switch (nscount)
323 {
324 case 2:
325 return offset & 1;
326 case 3:
327 return offset % 3;
328 case 4:
329 return offset & 3;
330 default:
331 return offset % nscount;
332 }
333 }
334
335 /* int
336 * res_queriesmatch(buf1, eom1, buf2, eom2)
337 * is there a 1:1 mapping of (name,type,class)
338 * in (buf1,eom1) and (buf2,eom2)?
339 * returns:
340 * -1 : format error
341 * 0 : not a 1:1 mapping
342 * >0 : is a 1:1 mapping
343 * author:
344 * paul vixie, 29may94
345 */
346 int
347 res_queriesmatch(const u_char *buf1, const u_char *eom1,
348 const u_char *buf2, const u_char *eom2)
349 {
350 if (buf1 + HFIXEDSZ > eom1 || buf2 + HFIXEDSZ > eom2)
351 return (-1);
352
353 /*
354 * Only header section present in replies to
355 * dynamic update packets.
356 */
357 if ((((HEADER *)buf1)->opcode == ns_o_update) &&
358 (((HEADER *)buf2)->opcode == ns_o_update))
359 return (1);
360
361 /* Note that we initially do not convert QDCOUNT to the host byte
362 order. We can compare it with the second buffer's QDCOUNT
363 value without doing this. */
364 int qdcount = ((HEADER*)buf1)->qdcount;
365 if (qdcount != ((HEADER*)buf2)->qdcount)
366 return (0);
367
368 qdcount = htons (qdcount);
369 const u_char *cp = buf1 + HFIXEDSZ;
370
371 while (qdcount-- > 0) {
372 char tname[MAXDNAME+1];
373 int n, ttype, tclass;
374
375 n = dn_expand(buf1, eom1, cp, tname, sizeof tname);
376 if (n < 0)
377 return (-1);
378 cp += n;
379 if (cp + 2 * INT16SZ > eom1)
380 return (-1);
381 NS_GET16(ttype, cp);
382 NS_GET16(tclass, cp);
383 if (!res_nameinquery(tname, ttype, tclass, buf2, eom2))
384 return (0);
385 }
386 return (1);
387 }
388 libresolv_hidden_def (res_queriesmatch)
389
390 int
391 __res_context_send (struct resolv_context *ctx,
392 const unsigned char *buf, int buflen,
393 const unsigned char *buf2, int buflen2,
394 unsigned char *ans, int anssiz,
395 unsigned char **ansp, unsigned char **ansp2,
396 int *nansp2, int *resplen2, int *ansp2_malloced)
397 {
398 struct __res_state *statp = ctx->resp;
399 int gotsomewhere, terrno, try, v_circuit, resplen;
400 /* On some architectures send_vc is inlined and the compiler might emit
401 a warning indicating 'resplen' may be used uninitialized. Note that
402 the warning belongs to resplen in send_vc which is used as return
403 value! There the maybe-uninitialized warning is already ignored as
404 it is a false-positive - see comment in send_vc.
405 Here the variable n is set to the return value of send_vc.
406 See below. */
407 DIAG_PUSH_NEEDS_COMMENT;
408 DIAG_IGNORE_NEEDS_COMMENT (9, "-Wmaybe-uninitialized");
409 int n;
410 DIAG_POP_NEEDS_COMMENT;
411
412 if (statp->nscount == 0) {
413 __set_errno (ESRCH);
414 return (-1);
415 }
416
417 if (anssiz < (buf2 == NULL ? 1 : 2) * HFIXEDSZ) {
418 __set_errno (EINVAL);
419 return (-1);
420 }
421
422 v_circuit = ((statp->options & RES_USEVC)
423 || buflen > PACKETSZ
424 || buflen2 > PACKETSZ);
425 gotsomewhere = 0;
426 terrno = ETIMEDOUT;
427
428 /*
429 * If the ns_addr_list in the resolver context has changed, then
430 * invalidate our cached copy and the associated timing data.
431 */
432 if (EXT(statp).nscount != 0) {
433 int needclose = 0;
434
435 if (EXT(statp).nscount != statp->nscount)
436 needclose++;
437 else
438 for (unsigned int ns = 0; ns < statp->nscount; ns++) {
439 if (statp->nsaddr_list[ns].sin_family != 0
440 && !sock_eq((struct sockaddr_in6 *)
441 &statp->nsaddr_list[ns],
442 EXT(statp).nsaddrs[ns]))
443 {
444 needclose++;
445 break;
446 }
447 }
448 if (needclose) {
449 __res_iclose(statp, false);
450 EXT(statp).nscount = 0;
451 }
452 }
453
454 /*
455 * Maybe initialize our private copy of the ns_addr_list.
456 */
457 if (EXT(statp).nscount == 0) {
458 for (unsigned int ns = 0; ns < statp->nscount; ns++) {
459 EXT(statp).nssocks[ns] = -1;
460 if (statp->nsaddr_list[ns].sin_family == 0)
461 continue;
462 if (EXT(statp).nsaddrs[ns] == NULL)
463 EXT(statp).nsaddrs[ns] =
464 malloc(sizeof (struct sockaddr_in6));
465 if (EXT(statp).nsaddrs[ns] != NULL)
466 memset (mempcpy(EXT(statp).nsaddrs[ns],
467 &statp->nsaddr_list[ns],
468 sizeof (struct sockaddr_in)),
469 '\0',
470 sizeof (struct sockaddr_in6)
471 - sizeof (struct sockaddr_in));
472 else
473 return -1;
474 }
475 EXT(statp).nscount = statp->nscount;
476 }
477
478 /* Name server index offset. Used to implement
479 RES_ROTATE. */
480 unsigned int ns_offset = nameserver_offset (statp);
481
482 /*
483 * Send request, RETRY times, or until successful.
484 */
485 for (try = 0; try < statp->retry; try++) {
486 for (unsigned ns_shift = 0; ns_shift < statp->nscount; ns_shift++)
487 {
488 /* The actual name server index. This implements
489 RES_ROTATE. */
490 unsigned int ns = ns_shift + ns_offset;
491 if (ns >= statp->nscount)
492 ns -= statp->nscount;
493
494 same_ns:
495 if (__glibc_unlikely (v_circuit)) {
496 /* Use VC; at most one attempt per server. */
497 try = statp->retry;
498 n = send_vc(statp, buf, buflen, buf2, buflen2,
499 &ans, &anssiz, &terrno,
500 ns, ansp, ansp2, nansp2, resplen2,
501 ansp2_malloced);
502 if (n < 0)
503 return (-1);
504 /* See comment at the declaration of n. */
505 DIAG_PUSH_NEEDS_COMMENT;
506 DIAG_IGNORE_NEEDS_COMMENT (9, "-Wmaybe-uninitialized");
507 if (n == 0 && (buf2 == NULL || *resplen2 == 0))
508 goto next_ns;
509 DIAG_POP_NEEDS_COMMENT;
510 } else {
511 /* Use datagrams. */
512 n = send_dg(statp, buf, buflen, buf2, buflen2,
513 &ans, &anssiz, &terrno,
514 ns, &v_circuit, &gotsomewhere, ansp,
515 ansp2, nansp2, resplen2, ansp2_malloced);
516 if (n < 0)
517 return (-1);
518 if (n == 0 && (buf2 == NULL || *resplen2 == 0))
519 goto next_ns;
520 if (v_circuit)
521 // XXX Check whether both requests failed or
522 // XXX whether one has been answered successfully
523 goto same_ns;
524 }
525
526 resplen = n;
527
528 /*
529 * If we have temporarily opened a virtual circuit,
530 * or if we haven't been asked to keep a socket open,
531 * close the socket.
532 */
533 if ((v_circuit && (statp->options & RES_USEVC) == 0) ||
534 (statp->options & RES_STAYOPEN) == 0) {
535 __res_iclose(statp, false);
536 }
537 return (resplen);
538 next_ns: ;
539 } /*foreach ns*/
540 } /*foreach retry*/
541 __res_iclose(statp, false);
542 if (!v_circuit) {
543 if (!gotsomewhere)
544 __set_errno (ECONNREFUSED); /* no nameservers found */
545 else
546 __set_errno (ETIMEDOUT); /* no answer obtained */
547 } else
548 __set_errno (terrno);
549 return (-1);
550 }
551
552 /* Common part of res_nsend and res_send. */
553 static int
554 context_send_common (struct resolv_context *ctx,
555 const unsigned char *buf, int buflen,
556 unsigned char *ans, int anssiz)
557 {
558 if (ctx == NULL)
559 {
560 RES_SET_H_ERRNO (&_res, NETDB_INTERNAL);
561 return -1;
562 }
563 int result = __res_context_send (ctx, buf, buflen, NULL, 0, ans, anssiz,
564 NULL, NULL, NULL, NULL, NULL);
565 __resolv_context_put (ctx);
566 return result;
567 }
568
569 int
570 res_nsend (res_state statp, const unsigned char *buf, int buflen,
571 unsigned char *ans, int anssiz)
572 {
573 return context_send_common
574 (__resolv_context_get_override (statp), buf, buflen, ans, anssiz);
575 }
576
577 int
578 res_send (const unsigned char *buf, int buflen, unsigned char *ans, int anssiz)
579 {
580 return context_send_common
581 (__resolv_context_get (), buf, buflen, ans, anssiz);
582 }
583
584 /* Private */
585
586 static struct sockaddr *
587 get_nsaddr (res_state statp, unsigned int n)
588 {
589 assert (n < statp->nscount);
590
591 if (statp->nsaddr_list[n].sin_family == 0 && EXT(statp).nsaddrs[n] != NULL)
592 /* EXT(statp).nsaddrs[n] holds an address that is larger than
593 struct sockaddr, and user code did not update
594 statp->nsaddr_list[n]. */
595 return (struct sockaddr *) EXT(statp).nsaddrs[n];
596 else
597 /* User code updated statp->nsaddr_list[n], or statp->nsaddr_list[n]
598 has the same content as EXT(statp).nsaddrs[n]. */
599 return (struct sockaddr *) (void *) &statp->nsaddr_list[n];
600 }
601
602 /* Close the resolver structure, assign zero to *RESPLEN2 if RESPLEN2
603 is not NULL, and return zero. */
604 static int
605 __attribute__ ((warn_unused_result))
606 close_and_return_error (res_state statp, int *resplen2)
607 {
608 __res_iclose(statp, false);
609 if (resplen2 != NULL)
610 *resplen2 = 0;
611 return 0;
612 }
613
614 /* The send_vc function is responsible for sending a DNS query over TCP
615 to the nameserver numbered NS from the res_state STATP i.e.
616 EXT(statp).nssocks[ns]. The function supports sending both IPv4 and
617 IPv6 queries at the same serially on the same socket.
618
619 Please note that for TCP there is no way to disable sending both
620 queries, unlike UDP, which honours RES_SNGLKUP and RES_SNGLKUPREOP
621 and sends the queries serially and waits for the result after each
622 sent query. This implementation should be corrected to honour these
623 options.
624
625 Please also note that for TCP we send both queries over the same
626 socket one after another. This technically violates best practice
627 since the server is allowed to read the first query, respond, and
628 then close the socket (to service another client). If the server
629 does this, then the remaining second query in the socket data buffer
630 will cause the server to send the client an RST which will arrive
631 asynchronously and the client's OS will likely tear down the socket
632 receive buffer resulting in a potentially short read and lost
633 response data. This will force the client to retry the query again,
634 and this process may repeat until all servers and connection resets
635 are exhausted and then the query will fail. It's not known if this
636 happens with any frequency in real DNS server implementations. This
637 implementation should be corrected to use two sockets by default for
638 parallel queries.
639
640 The query stored in BUF of BUFLEN length is sent first followed by
641 the query stored in BUF2 of BUFLEN2 length. Queries are sent
642 serially on the same socket.
643
644 Answers to the query are stored firstly in *ANSP up to a max of
645 *ANSSIZP bytes. If more than *ANSSIZP bytes are needed and ANSCP
646 is non-NULL (to indicate that modifying the answer buffer is allowed)
647 then malloc is used to allocate a new response buffer and ANSCP and
648 ANSP will both point to the new buffer. If more than *ANSSIZP bytes
649 are needed but ANSCP is NULL, then as much of the response as
650 possible is read into the buffer, but the results will be truncated.
651 When truncation happens because of a small answer buffer the DNS
652 packets header field TC will bet set to 1, indicating a truncated
653 message and the rest of the socket data will be read and discarded.
654
655 Answers to the query are stored secondly in *ANSP2 up to a max of
656 *ANSSIZP2 bytes, with the actual response length stored in
657 *RESPLEN2. If more than *ANSSIZP bytes are needed and ANSP2
658 is non-NULL (required for a second query) then malloc is used to
659 allocate a new response buffer, *ANSSIZP2 is set to the new buffer
660 size and *ANSP2_MALLOCED is set to 1.
661
662 The ANSP2_MALLOCED argument will eventually be removed as the
663 change in buffer pointer can be used to detect the buffer has
664 changed and that the caller should use free on the new buffer.
665
666 Note that the answers may arrive in any order from the server and
667 therefore the first and second answer buffers may not correspond to
668 the first and second queries.
669
670 It is not supported to call this function with a non-NULL ANSP2
671 but a NULL ANSCP. Put another way, you can call send_vc with a
672 single unmodifiable buffer or two modifiable buffers, but no other
673 combination is supported.
674
675 It is the caller's responsibility to free the malloc allocated
676 buffers by detecting that the pointers have changed from their
677 original values i.e. *ANSCP or *ANSP2 has changed.
678
679 If errors are encountered then *TERRNO is set to an appropriate
680 errno value and a zero result is returned for a recoverable error,
681 and a less-than zero result is returned for a non-recoverable error.
682
683 If no errors are encountered then *TERRNO is left unmodified and
684 a the length of the first response in bytes is returned. */
685 static int
686 send_vc(res_state statp,
687 const u_char *buf, int buflen, const u_char *buf2, int buflen2,
688 u_char **ansp, int *anssizp,
689 int *terrno, int ns, u_char **anscp, u_char **ansp2, int *anssizp2,
690 int *resplen2, int *ansp2_malloced)
691 {
692 const HEADER *hp = (HEADER *) buf;
693 const HEADER *hp2 = (HEADER *) buf2;
694 HEADER *anhp = (HEADER *) *ansp;
695 struct sockaddr *nsap = get_nsaddr (statp, ns);
696 int truncating, connreset, n;
697 /* On some architectures compiler might emit a warning indicating
698 'resplen' may be used uninitialized. However if buf2 == NULL
699 then this code won't be executed; if buf2 != NULL, then first
700 time round the loop recvresp1 and recvresp2 will be 0 so this
701 code won't be executed but "thisresplenp = &resplen;" followed
702 by "*thisresplenp = rlen;" will be executed so that subsequent
703 times round the loop resplen has been initialized. So this is
704 a false-positive.
705 */
706 DIAG_PUSH_NEEDS_COMMENT;
707 DIAG_IGNORE_NEEDS_COMMENT (5, "-Wmaybe-uninitialized");
708 int resplen;
709 DIAG_POP_NEEDS_COMMENT;
710 struct iovec iov[4];
711 u_short len;
712 u_short len2;
713 u_char *cp;
714
715 connreset = 0;
716 same_ns:
717 truncating = 0;
718
719 /* Are we still talking to whom we want to talk to? */
720 if (statp->_vcsock >= 0 && (statp->_flags & RES_F_VC) != 0) {
721 struct sockaddr_in6 peer;
722 socklen_t size = sizeof peer;
723
724 if (getpeername(statp->_vcsock,
725 (struct sockaddr *)&peer, &size) < 0 ||
726 !sock_eq(&peer, (struct sockaddr_in6 *) nsap)) {
727 __res_iclose(statp, false);
728 statp->_flags &= ~RES_F_VC;
729 }
730 }
731
732 if (statp->_vcsock < 0 || (statp->_flags & RES_F_VC) == 0) {
733 if (statp->_vcsock >= 0)
734 __res_iclose(statp, false);
735
736 statp->_vcsock = socket
737 (nsap->sa_family, SOCK_STREAM | SOCK_CLOEXEC, 0);
738 if (statp->_vcsock < 0) {
739 *terrno = errno;
740 if (resplen2 != NULL)
741 *resplen2 = 0;
742 return (-1);
743 }
744 __set_errno (0);
745 if (connect(statp->_vcsock, nsap,
746 nsap->sa_family == AF_INET
747 ? sizeof (struct sockaddr_in)
748 : sizeof (struct sockaddr_in6)) < 0) {
749 *terrno = errno;
750 return close_and_return_error (statp, resplen2);
751 }
752 statp->_flags |= RES_F_VC;
753 }
754
755 /*
756 * Send length & message
757 */
758 len = htons ((u_short) buflen);
759 evConsIovec(&len, INT16SZ, &iov[0]);
760 evConsIovec((void*)buf, buflen, &iov[1]);
761 int niov = 2;
762 ssize_t explen = INT16SZ + buflen;
763 if (buf2 != NULL) {
764 len2 = htons ((u_short) buflen2);
765 evConsIovec(&len2, INT16SZ, &iov[2]);
766 evConsIovec((void*)buf2, buflen2, &iov[3]);
767 niov = 4;
768 explen += INT16SZ + buflen2;
769 }
770 if (TEMP_FAILURE_RETRY (writev(statp->_vcsock, iov, niov)) != explen) {
771 *terrno = errno;
772 return close_and_return_error (statp, resplen2);
773 }
774 /*
775 * Receive length & response
776 */
777 int recvresp1 = 0;
778 /* Skip the second response if there is no second query.
779 To do that we mark the second response as received. */
780 int recvresp2 = buf2 == NULL;
781 uint16_t rlen16;
782 read_len:
783 cp = (u_char *)&rlen16;
784 len = sizeof(rlen16);
785 while ((n = TEMP_FAILURE_RETRY (read(statp->_vcsock, cp,
786 (int)len))) > 0) {
787 cp += n;
788 if ((len -= n) <= 0)
789 break;
790 }
791 if (n <= 0) {
792 *terrno = errno;
793 /*
794 * A long running process might get its TCP
795 * connection reset if the remote server was
796 * restarted. Requery the server instead of
797 * trying a new one. When there is only one
798 * server, this means that a query might work
799 * instead of failing. We only allow one reset
800 * per query to prevent looping.
801 */
802 if (*terrno == ECONNRESET && !connreset)
803 {
804 __res_iclose (statp, false);
805 connreset = 1;
806 goto same_ns;
807 }
808 return close_and_return_error (statp, resplen2);
809 }
810 int rlen = ntohs (rlen16);
811
812 int *thisanssizp;
813 u_char **thisansp;
814 int *thisresplenp;
815 if ((recvresp1 | recvresp2) == 0 || buf2 == NULL) {
816 /* We have not received any responses
817 yet or we only have one response to
818 receive. */
819 thisanssizp = anssizp;
820 thisansp = anscp ?: ansp;
821 assert (anscp != NULL || ansp2 == NULL);
822 thisresplenp = &resplen;
823 } else {
824 thisanssizp = anssizp2;
825 thisansp = ansp2;
826 thisresplenp = resplen2;
827 }
828 anhp = (HEADER *) *thisansp;
829
830 *thisresplenp = rlen;
831 /* Is the answer buffer too small? */
832 if (*thisanssizp < rlen) {
833 /* If the current buffer is not the the static
834 user-supplied buffer then we can reallocate
835 it. */
836 if (thisansp != NULL && thisansp != ansp) {
837 /* Always allocate MAXPACKET, callers expect
838 this specific size. */
839 u_char *newp = malloc (MAXPACKET);
840 if (newp == NULL)
841 {
842 *terrno = ENOMEM;
843 return close_and_return_error (statp, resplen2);
844 }
845 *thisanssizp = MAXPACKET;
846 *thisansp = newp;
847 if (thisansp == ansp2)
848 *ansp2_malloced = 1;
849 anhp = (HEADER *) newp;
850 /* A uint16_t can't be larger than MAXPACKET
851 thus it's safe to allocate MAXPACKET but
852 read RLEN bytes instead. */
853 len = rlen;
854 } else {
855 truncating = 1;
856 len = *thisanssizp;
857 }
858 } else
859 len = rlen;
860
861 if (__glibc_unlikely (len < HFIXEDSZ)) {
862 /*
863 * Undersized message.
864 */
865 *terrno = EMSGSIZE;
866 return close_and_return_error (statp, resplen2);
867 }
868
869 cp = *thisansp;
870 while (len != 0 && (n = read(statp->_vcsock, (char *)cp, (int)len)) > 0){
871 cp += n;
872 len -= n;
873 }
874 if (__glibc_unlikely (n <= 0)) {
875 *terrno = errno;
876 return close_and_return_error (statp, resplen2);
877 }
878 if (__glibc_unlikely (truncating)) {
879 /*
880 * Flush rest of answer so connection stays in synch.
881 */
882 anhp->tc = 1;
883 len = rlen - *thisanssizp;
884 while (len != 0) {
885 char junk[PACKETSZ];
886
887 n = read(statp->_vcsock, junk,
888 (len > sizeof junk) ? sizeof junk : len);
889 if (n > 0)
890 len -= n;
891 else
892 break;
893 }
894 }
895 /*
896 * If the calling application has bailed out of
897 * a previous call and failed to arrange to have
898 * the circuit closed or the server has got
899 * itself confused, then drop the packet and
900 * wait for the correct one.
901 */
902 if ((recvresp1 || hp->id != anhp->id)
903 && (recvresp2 || hp2->id != anhp->id))
904 goto read_len;
905
906 /* Mark which reply we received. */
907 if (recvresp1 == 0 && hp->id == anhp->id)
908 recvresp1 = 1;
909 else
910 recvresp2 = 1;
911 /* Repeat waiting if we have a second answer to arrive. */
912 if ((recvresp1 & recvresp2) == 0)
913 goto read_len;
914
915 /*
916 * All is well, or the error is fatal. Signal that the
917 * next nameserver ought not be tried.
918 */
919 return resplen;
920 }
921
922 static int
923 reopen (res_state statp, int *terrno, int ns)
924 {
925 if (EXT(statp).nssocks[ns] == -1) {
926 struct sockaddr *nsap = get_nsaddr (statp, ns);
927 socklen_t slen;
928
929 /* only try IPv6 if IPv6 NS and if not failed before */
930 if (nsap->sa_family == AF_INET6 && !statp->ipv6_unavail) {
931 EXT(statp).nssocks[ns] = socket
932 (PF_INET6,
933 SOCK_DGRAM | SOCK_NONBLOCK | SOCK_CLOEXEC, 0);
934 if (EXT(statp).nssocks[ns] < 0)
935 statp->ipv6_unavail = errno == EAFNOSUPPORT;
936 slen = sizeof (struct sockaddr_in6);
937 } else if (nsap->sa_family == AF_INET) {
938 EXT(statp).nssocks[ns] = socket
939 (PF_INET,
940 SOCK_DGRAM | SOCK_NONBLOCK | SOCK_CLOEXEC, 0);
941 slen = sizeof (struct sockaddr_in);
942 }
943 if (EXT(statp).nssocks[ns] < 0) {
944 *terrno = errno;
945 return (-1);
946 }
947
948 /* Enable full ICMP error reporting for this
949 socket. */
950 if (__res_enable_icmp (nsap->sa_family,
951 EXT (statp).nssocks[ns]) < 0)
952 {
953 int saved_errno = errno;
954 __res_iclose (statp, false);
955 __set_errno (saved_errno);
956 *terrno = saved_errno;
957 return -1;
958 }
959
960 /*
961 * On a 4.3BSD+ machine (client and server,
962 * actually), sending to a nameserver datagram
963 * port with no nameserver will cause an
964 * ICMP port unreachable message to be returned.
965 * If our datagram socket is "connected" to the
966 * server, we get an ECONNREFUSED error on the next
967 * socket operation, and select returns if the
968 * error message is received. We can thus detect
969 * the absence of a nameserver without timing out.
970 */
971 /* With GCC 5.3 when compiling with -Os the compiler
972 emits a warning that slen may be used uninitialized,
973 but that is never true. Both slen and
974 EXT(statp).nssocks[ns] are initialized together or
975 the function return -1 before control flow reaches
976 the call to connect with slen. */
977 DIAG_PUSH_NEEDS_COMMENT;
978 DIAG_IGNORE_Os_NEEDS_COMMENT (5, "-Wmaybe-uninitialized");
979 if (connect(EXT(statp).nssocks[ns], nsap, slen) < 0) {
980 DIAG_POP_NEEDS_COMMENT;
981 __res_iclose(statp, false);
982 return (0);
983 }
984 }
985
986 return 1;
987 }
988
989 /* The send_dg function is responsible for sending a DNS query over UDP
990 to the nameserver numbered NS from the res_state STATP i.e.
991 EXT(statp).nssocks[ns]. The function supports IPv4 and IPv6 queries
992 along with the ability to send the query in parallel for both stacks
993 (default) or serially (RES_SINGLKUP). It also supports serial lookup
994 with a close and reopen of the socket used to talk to the server
995 (RES_SNGLKUPREOP) to work around broken name servers.
996
997 The query stored in BUF of BUFLEN length is sent first followed by
998 the query stored in BUF2 of BUFLEN2 length. Queries are sent
999 in parallel (default) or serially (RES_SINGLKUP or RES_SNGLKUPREOP).
1000
1001 Answers to the query are stored firstly in *ANSP up to a max of
1002 *ANSSIZP bytes. If more than *ANSSIZP bytes are needed and ANSCP
1003 is non-NULL (to indicate that modifying the answer buffer is allowed)
1004 then malloc is used to allocate a new response buffer and ANSCP and
1005 ANSP will both point to the new buffer. If more than *ANSSIZP bytes
1006 are needed but ANSCP is NULL, then as much of the response as
1007 possible is read into the buffer, but the results will be truncated.
1008 When truncation happens because of a small answer buffer the DNS
1009 packets header field TC will bet set to 1, indicating a truncated
1010 message, while the rest of the UDP packet is discarded.
1011
1012 Answers to the query are stored secondly in *ANSP2 up to a max of
1013 *ANSSIZP2 bytes, with the actual response length stored in
1014 *RESPLEN2. If more than *ANSSIZP bytes are needed and ANSP2
1015 is non-NULL (required for a second query) then malloc is used to
1016 allocate a new response buffer, *ANSSIZP2 is set to the new buffer
1017 size and *ANSP2_MALLOCED is set to 1.
1018
1019 The ANSP2_MALLOCED argument will eventually be removed as the
1020 change in buffer pointer can be used to detect the buffer has
1021 changed and that the caller should use free on the new buffer.
1022
1023 Note that the answers may arrive in any order from the server and
1024 therefore the first and second answer buffers may not correspond to
1025 the first and second queries.
1026
1027 It is not supported to call this function with a non-NULL ANSP2
1028 but a NULL ANSCP. Put another way, you can call send_vc with a
1029 single unmodifiable buffer or two modifiable buffers, but no other
1030 combination is supported.
1031
1032 It is the caller's responsibility to free the malloc allocated
1033 buffers by detecting that the pointers have changed from their
1034 original values i.e. *ANSCP or *ANSP2 has changed.
1035
1036 If an answer is truncated because of UDP datagram DNS limits then
1037 *V_CIRCUIT is set to 1 and the return value non-zero to indicate to
1038 the caller to retry with TCP. The value *GOTSOMEWHERE is set to 1
1039 if any progress was made reading a response from the nameserver and
1040 is used by the caller to distinguish between ECONNREFUSED and
1041 ETIMEDOUT (the latter if *GOTSOMEWHERE is 1).
1042
1043 If errors are encountered then *TERRNO is set to an appropriate
1044 errno value and a zero result is returned for a recoverable error,
1045 and a less-than zero result is returned for a non-recoverable error.
1046
1047 If no errors are encountered then *TERRNO is left unmodified and
1048 a the length of the first response in bytes is returned. */
1049 static int
1050 send_dg(res_state statp,
1051 const u_char *buf, int buflen, const u_char *buf2, int buflen2,
1052 u_char **ansp, int *anssizp,
1053 int *terrno, int ns, int *v_circuit, int *gotsomewhere, u_char **anscp,
1054 u_char **ansp2, int *anssizp2, int *resplen2, int *ansp2_malloced)
1055 {
1056 const HEADER *hp = (HEADER *) buf;
1057 const HEADER *hp2 = (HEADER *) buf2;
1058 struct timespec now, timeout, finish;
1059 struct pollfd pfd[1];
1060 int ptimeout;
1061 struct sockaddr_in6 from;
1062 int resplen = 0;
1063 int n;
1064
1065 /*
1066 * Compute time for the total operation.
1067 */
1068 int seconds = (statp->retrans << ns);
1069 if (ns > 0)
1070 seconds /= statp->nscount;
1071 if (seconds <= 0)
1072 seconds = 1;
1073 bool single_request_reopen = (statp->options & RES_SNGLKUPREOP) != 0;
1074 bool single_request = (((statp->options & RES_SNGLKUP) != 0)
1075 | single_request_reopen);
1076 int save_gotsomewhere = *gotsomewhere;
1077
1078 int retval;
1079 retry_reopen:
1080 retval = reopen (statp, terrno, ns);
1081 if (retval <= 0)
1082 {
1083 if (resplen2 != NULL)
1084 *resplen2 = 0;
1085 return retval;
1086 }
1087 retry:
1088 evNowTime(&now);
1089 evConsTime(&timeout, seconds, 0);
1090 evAddTime(&finish, &now, &timeout);
1091 int need_recompute = 0;
1092 int nwritten = 0;
1093 int recvresp1 = 0;
1094 /* Skip the second response if there is no second query.
1095 To do that we mark the second response as received. */
1096 int recvresp2 = buf2 == NULL;
1097 pfd[0].fd = EXT(statp).nssocks[ns];
1098 pfd[0].events = POLLOUT;
1099 wait:
1100 if (need_recompute) {
1101 recompute_resend:
1102 evNowTime(&now);
1103 if (evCmpTime(finish, now) <= 0) {
1104 poll_err_out:
1105 return close_and_return_error (statp, resplen2);
1106 }
1107 evSubTime(&timeout, &finish, &now);
1108 need_recompute = 0;
1109 }
1110 /* Convert struct timespec in milliseconds. */
1111 ptimeout = timeout.tv_sec * 1000 + timeout.tv_nsec / 1000000;
1112
1113 n = 0;
1114 if (nwritten == 0)
1115 n = __poll (pfd, 1, 0);
1116 if (__glibc_unlikely (n == 0)) {
1117 n = __poll (pfd, 1, ptimeout);
1118 need_recompute = 1;
1119 }
1120 if (n == 0) {
1121 if (resplen > 1 && (recvresp1 || (buf2 != NULL && recvresp2)))
1122 {
1123 /* There are quite a few broken name servers out
1124 there which don't handle two outstanding
1125 requests from the same source. There are also
1126 broken firewall settings. If we time out after
1127 having received one answer switch to the mode
1128 where we send the second request only once we
1129 have received the first answer. */
1130 if (!single_request)
1131 {
1132 statp->options |= RES_SNGLKUP;
1133 single_request = true;
1134 *gotsomewhere = save_gotsomewhere;
1135 goto retry;
1136 }
1137 else if (!single_request_reopen)
1138 {
1139 statp->options |= RES_SNGLKUPREOP;
1140 single_request_reopen = true;
1141 *gotsomewhere = save_gotsomewhere;
1142 __res_iclose (statp, false);
1143 goto retry_reopen;
1144 }
1145
1146 *resplen2 = 1;
1147 return resplen;
1148 }
1149
1150 *gotsomewhere = 1;
1151 if (resplen2 != NULL)
1152 *resplen2 = 0;
1153 return 0;
1154 }
1155 if (n < 0) {
1156 if (errno == EINTR)
1157 goto recompute_resend;
1158
1159 goto poll_err_out;
1160 }
1161 __set_errno (0);
1162 if (pfd[0].revents & POLLOUT) {
1163 #ifndef __ASSUME_SENDMMSG
1164 static int have_sendmmsg;
1165 #else
1166 # define have_sendmmsg 1
1167 #endif
1168 if (have_sendmmsg >= 0 && nwritten == 0 && buf2 != NULL
1169 && !single_request)
1170 {
1171 struct iovec iov =
1172 { .iov_base = (void *) buf, .iov_len = buflen };
1173 struct iovec iov2 =
1174 { .iov_base = (void *) buf2, .iov_len = buflen2 };
1175 struct mmsghdr reqs[2] =
1176 {
1177 {
1178 .msg_hdr =
1179 {
1180 .msg_iov = &iov,
1181 .msg_iovlen = 1,
1182 },
1183 },
1184 {
1185 .msg_hdr =
1186 {
1187 .msg_iov = &iov2,
1188 .msg_iovlen = 1,
1189 }
1190 },
1191 };
1192
1193 int ndg = __sendmmsg (pfd[0].fd, reqs, 2, MSG_NOSIGNAL);
1194 if (__glibc_likely (ndg == 2))
1195 {
1196 if (reqs[0].msg_len != buflen
1197 || reqs[1].msg_len != buflen2)
1198 goto fail_sendmmsg;
1199
1200 pfd[0].events = POLLIN;
1201 nwritten += 2;
1202 }
1203 else if (ndg == 1 && reqs[0].msg_len == buflen)
1204 goto just_one;
1205 else if (ndg < 0 && (errno == EINTR || errno == EAGAIN))
1206 goto recompute_resend;
1207 else
1208 {
1209 #ifndef __ASSUME_SENDMMSG
1210 if (__glibc_unlikely (have_sendmmsg == 0))
1211 {
1212 if (ndg < 0 && errno == ENOSYS)
1213 {
1214 have_sendmmsg = -1;
1215 goto try_send;
1216 }
1217 have_sendmmsg = 1;
1218 }
1219 #endif
1220
1221 fail_sendmmsg:
1222 return close_and_return_error (statp, resplen2);
1223 }
1224 }
1225 else
1226 {
1227 ssize_t sr;
1228 #ifndef __ASSUME_SENDMMSG
1229 try_send:
1230 #endif
1231 if (nwritten != 0)
1232 sr = send (pfd[0].fd, buf2, buflen2, MSG_NOSIGNAL);
1233 else
1234 sr = send (pfd[0].fd, buf, buflen, MSG_NOSIGNAL);
1235
1236 if (sr != (nwritten != 0 ? buflen2 : buflen)) {
1237 if (errno == EINTR || errno == EAGAIN)
1238 goto recompute_resend;
1239 return close_and_return_error (statp, resplen2);
1240 }
1241 just_one:
1242 if (nwritten != 0 || buf2 == NULL || single_request)
1243 pfd[0].events = POLLIN;
1244 else
1245 pfd[0].events = POLLIN | POLLOUT;
1246 ++nwritten;
1247 }
1248 goto wait;
1249 } else if (pfd[0].revents & POLLIN) {
1250 int *thisanssizp;
1251 u_char **thisansp;
1252 int *thisresplenp;
1253
1254 if ((recvresp1 | recvresp2) == 0 || buf2 == NULL) {
1255 /* We have not received any responses
1256 yet or we only have one response to
1257 receive. */
1258 thisanssizp = anssizp;
1259 thisansp = anscp ?: ansp;
1260 assert (anscp != NULL || ansp2 == NULL);
1261 thisresplenp = &resplen;
1262 } else {
1263 thisanssizp = anssizp2;
1264 thisansp = ansp2;
1265 thisresplenp = resplen2;
1266 }
1267
1268 if (*thisanssizp < MAXPACKET
1269 /* If the current buffer is not the the static
1270 user-supplied buffer then we can reallocate
1271 it. */
1272 && (thisansp != NULL && thisansp != ansp)
1273 #ifdef FIONREAD
1274 /* Is the size too small? */
1275 && (ioctl (pfd[0].fd, FIONREAD, thisresplenp) < 0
1276 || *thisanssizp < *thisresplenp)
1277 #endif
1278 ) {
1279 /* Always allocate MAXPACKET, callers expect
1280 this specific size. */
1281 u_char *newp = malloc (MAXPACKET);
1282 if (newp != NULL) {
1283 *thisanssizp = MAXPACKET;
1284 *thisansp = newp;
1285 if (thisansp == ansp2)
1286 *ansp2_malloced = 1;
1287 }
1288 }
1289 /* We could end up with truncation if anscp was NULL
1290 (not allowed to change caller's buffer) and the
1291 response buffer size is too small. This isn't a
1292 reliable way to detect truncation because the ioctl
1293 may be an inaccurate report of the UDP message size.
1294 Therefore we use this only to issue debug output.
1295 To do truncation accurately with UDP we need
1296 MSG_TRUNC which is only available on Linux. We
1297 can abstract out the Linux-specific feature in the
1298 future to detect truncation. */
1299 HEADER *anhp = (HEADER *) *thisansp;
1300 socklen_t fromlen = sizeof(struct sockaddr_in6);
1301 assert (sizeof(from) <= fromlen);
1302 *thisresplenp = recvfrom(pfd[0].fd, (char*)*thisansp,
1303 *thisanssizp, 0,
1304 (struct sockaddr *)&from, &fromlen);
1305 if (__glibc_unlikely (*thisresplenp <= 0)) {
1306 if (errno == EINTR || errno == EAGAIN) {
1307 need_recompute = 1;
1308 goto wait;
1309 }
1310 return close_and_return_error (statp, resplen2);
1311 }
1312 *gotsomewhere = 1;
1313 if (__glibc_unlikely (*thisresplenp < HFIXEDSZ)) {
1314 /*
1315 * Undersized message.
1316 */
1317 *terrno = EMSGSIZE;
1318 return close_and_return_error (statp, resplen2);
1319 }
1320 if ((recvresp1 || hp->id != anhp->id)
1321 && (recvresp2 || hp2->id != anhp->id)) {
1322 /*
1323 * response from old query, ignore it.
1324 * XXX - potential security hazard could
1325 * be detected here.
1326 */
1327 goto wait;
1328 }
1329
1330 /* Paranoia check. Due to the connected UDP socket,
1331 the kernel has already filtered invalid addresses
1332 for us. */
1333 if (!res_ourserver_p(statp, &from))
1334 goto wait;
1335
1336 /* Check for the correct header layout and a matching
1337 question. */
1338 if ((recvresp1 || !res_queriesmatch(buf, buf + buflen,
1339 *thisansp,
1340 *thisansp
1341 + *thisanssizp))
1342 && (recvresp2 || !res_queriesmatch(buf2, buf2 + buflen2,
1343 *thisansp,
1344 *thisansp
1345 + *thisanssizp)))
1346 goto wait;
1347
1348 if (anhp->rcode == SERVFAIL ||
1349 anhp->rcode == NOTIMP ||
1350 anhp->rcode == REFUSED) {
1351 next_ns:
1352 if (recvresp1 || (buf2 != NULL && recvresp2)) {
1353 *resplen2 = 0;
1354 return resplen;
1355 }
1356 if (buf2 != NULL)
1357 {
1358 /* No data from the first reply. */
1359 resplen = 0;
1360 /* We are waiting for a possible second reply. */
1361 if (hp->id == anhp->id)
1362 recvresp1 = 1;
1363 else
1364 recvresp2 = 1;
1365
1366 goto wait;
1367 }
1368
1369 /* don't retry if called from dig */
1370 if (!statp->pfcode)
1371 return close_and_return_error (statp, resplen2);
1372 __res_iclose(statp, false);
1373 }
1374 if (anhp->rcode == NOERROR && anhp->ancount == 0
1375 && anhp->aa == 0 && anhp->ra == 0 && anhp->arcount == 0) {
1376 goto next_ns;
1377 }
1378 if (!(statp->options & RES_IGNTC) && anhp->tc) {
1379 /*
1380 * To get the rest of answer,
1381 * use TCP with same server.
1382 */
1383 *v_circuit = 1;
1384 __res_iclose(statp, false);
1385 // XXX if we have received one reply we could
1386 // XXX use it and not repeat it over TCP...
1387 if (resplen2 != NULL)
1388 *resplen2 = 0;
1389 return (1);
1390 }
1391 /* Mark which reply we received. */
1392 if (recvresp1 == 0 && hp->id == anhp->id)
1393 recvresp1 = 1;
1394 else
1395 recvresp2 = 1;
1396 /* Repeat waiting if we have a second answer to arrive. */
1397 if ((recvresp1 & recvresp2) == 0) {
1398 if (single_request) {
1399 pfd[0].events = POLLOUT;
1400 if (single_request_reopen) {
1401 __res_iclose (statp, false);
1402 retval = reopen (statp, terrno, ns);
1403 if (retval <= 0)
1404 {
1405 if (resplen2 != NULL)
1406 *resplen2 = 0;
1407 return retval;
1408 }
1409 pfd[0].fd = EXT(statp).nssocks[ns];
1410 }
1411 }
1412 goto wait;
1413 }
1414 /* All is well. We have received both responses (if
1415 two responses were requested). */
1416 return (resplen);
1417 } else if (pfd[0].revents & (POLLERR | POLLHUP | POLLNVAL))
1418 /* Something went wrong. We can stop trying. */
1419 return close_and_return_error (statp, resplen2);
1420 else {
1421 /* poll should not have returned > 0 in this case. */
1422 abort ();
1423 }
1424 }
1425
1426 static int
1427 sock_eq(struct sockaddr_in6 *a1, struct sockaddr_in6 *a2) {
1428 if (a1->sin6_family == a2->sin6_family) {
1429 if (a1->sin6_family == AF_INET)
1430 return ((((struct sockaddr_in *)a1)->sin_port ==
1431 ((struct sockaddr_in *)a2)->sin_port) &&
1432 (((struct sockaddr_in *)a1)->sin_addr.s_addr ==
1433 ((struct sockaddr_in *)a2)->sin_addr.s_addr));
1434 else
1435 return ((a1->sin6_port == a2->sin6_port) &&
1436 !memcmp(&a1->sin6_addr, &a2->sin6_addr,
1437 sizeof (struct in6_addr)));
1438 }
1439 if (a1->sin6_family == AF_INET) {
1440 struct sockaddr_in6 *sap = a1;
1441 a1 = a2;
1442 a2 = sap;
1443 } /* assumes that AF_INET and AF_INET6 are the only possibilities */
1444 return ((a1->sin6_port == ((struct sockaddr_in *)a2)->sin_port) &&
1445 IN6_IS_ADDR_V4MAPPED(&a1->sin6_addr) &&
1446 (a1->sin6_addr.s6_addr32[3] ==
1447 ((struct sockaddr_in *)a2)->sin_addr.s_addr));
1448 }