1 | /* Copyright (C) 2016-2024 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 int send_vc(res_state, const u_char *, int, |
184 | const u_char *, int, |
185 | u_char **, int *, int *, int, u_char **, |
186 | u_char **, int *, int *, int *); |
187 | static int send_dg(res_state, const u_char *, int, |
188 | const u_char *, int, |
189 | u_char **, int *, int *, int, |
190 | int *, int *, u_char **, |
191 | u_char **, int *, int *, int *); |
192 | static int sock_eq(struct sockaddr_in6 *, struct sockaddr_in6 *); |
193 | |
194 | /* Returns a shift value for the name server index. Used to implement |
195 | RES_ROTATE. */ |
196 | static unsigned int |
197 | nameserver_offset (struct __res_state *statp) |
198 | { |
199 | /* If we only have one name server or rotation is disabled, return |
200 | offset 0 (no rotation). */ |
201 | unsigned int nscount = statp->nscount; |
202 | if (nscount <= 1 || !(statp->options & RES_ROTATE)) |
203 | return 0; |
204 | |
205 | /* Global offset. The lowest bit indicates whether the offset has |
206 | been initialized with a random value. Use relaxed MO to access |
207 | global_offset because all we need is a sequence of roughly |
208 | sequential value. */ |
209 | static unsigned int global_offset; |
210 | unsigned int offset = atomic_fetch_add_relaxed (&global_offset, 2); |
211 | if ((offset & 1) == 0) |
212 | { |
213 | /* Initialization is required. */ |
214 | offset = random_bits (); |
215 | /* The lowest bit is the most random. Preserve it. */ |
216 | offset <<= 1; |
217 | |
218 | /* Store the new starting value. atomic_fetch_add_relaxed |
219 | returns the old value, so emulate that by storing the new |
220 | (incremented) value. Concurrent initialization with |
221 | different random values is harmless. */ |
222 | atomic_store_relaxed (&global_offset, (offset | 1) + 2); |
223 | } |
224 | |
225 | /* Remove the initialization bit. */ |
226 | offset >>= 1; |
227 | |
228 | /* Avoid the division in the most common cases. */ |
229 | switch (nscount) |
230 | { |
231 | case 2: |
232 | return offset & 1; |
233 | case 3: |
234 | return offset % 3; |
235 | case 4: |
236 | return offset & 3; |
237 | default: |
238 | return offset % nscount; |
239 | } |
240 | } |
241 | |
242 | /* Clear the AD bit unless the trust-ad option was specified in the |
243 | resolver configuration. */ |
244 | static void |
245 | mask_ad_bit (struct resolv_context *ctx, void *buf) |
246 | { |
247 | if (!(ctx->resp->options & RES_TRUSTAD)) |
248 | ((UHEADER *) buf)->ad = 0; |
249 | } |
250 | |
251 | int |
252 | __res_context_send (struct resolv_context *ctx, |
253 | const unsigned char *buf, int buflen, |
254 | const unsigned char *buf2, int buflen2, |
255 | unsigned char *ans, int anssiz, |
256 | unsigned char **ansp, unsigned char **ansp2, |
257 | int *nansp2, int *resplen2, int *ansp2_malloced) |
258 | { |
259 | struct __res_state *statp = ctx->resp; |
260 | int gotsomewhere, terrno, try, v_circuit, resplen; |
261 | /* On some architectures send_vc is inlined and the compiler might emit |
262 | a warning indicating 'resplen' may be used uninitialized. Note that |
263 | the warning belongs to resplen in send_vc which is used as return |
264 | value! There the maybe-uninitialized warning is already ignored as |
265 | it is a false-positive - see comment in send_vc. |
266 | Here the variable n is set to the return value of send_vc. |
267 | See below. */ |
268 | DIAG_PUSH_NEEDS_COMMENT; |
269 | DIAG_IGNORE_NEEDS_COMMENT (9, "-Wmaybe-uninitialized" ); |
270 | int n; |
271 | DIAG_POP_NEEDS_COMMENT; |
272 | |
273 | if (statp->nscount == 0) { |
274 | __set_errno (ESRCH); |
275 | return (-1); |
276 | } |
277 | |
278 | if (anssiz < (buf2 == NULL ? 1 : 2) * HFIXEDSZ) { |
279 | __set_errno (EINVAL); |
280 | return (-1); |
281 | } |
282 | |
283 | v_circuit = ((statp->options & RES_USEVC) |
284 | || buflen > PACKETSZ |
285 | || buflen2 > PACKETSZ); |
286 | gotsomewhere = 0; |
287 | terrno = ETIMEDOUT; |
288 | |
289 | /* |
290 | * If the ns_addr_list in the resolver context has changed, then |
291 | * invalidate our cached copy and the associated timing data. |
292 | */ |
293 | if (EXT(statp).nscount != 0) { |
294 | int needclose = 0; |
295 | |
296 | if (EXT(statp).nscount != statp->nscount) |
297 | needclose++; |
298 | else |
299 | for (unsigned int ns = 0; ns < statp->nscount; ns++) { |
300 | if (statp->nsaddr_list[ns].sin_family != 0 |
301 | && !sock_eq((struct sockaddr_in6 *) |
302 | &statp->nsaddr_list[ns], |
303 | EXT(statp).nsaddrs[ns])) |
304 | { |
305 | needclose++; |
306 | break; |
307 | } |
308 | } |
309 | if (needclose) { |
310 | __res_iclose(statp, false); |
311 | EXT(statp).nscount = 0; |
312 | } |
313 | } |
314 | |
315 | /* |
316 | * Maybe initialize our private copy of the ns_addr_list. |
317 | */ |
318 | if (EXT(statp).nscount == 0) { |
319 | for (unsigned int ns = 0; ns < statp->nscount; ns++) { |
320 | EXT(statp).nssocks[ns] = -1; |
321 | if (statp->nsaddr_list[ns].sin_family == 0) |
322 | continue; |
323 | if (EXT(statp).nsaddrs[ns] == NULL) |
324 | EXT(statp).nsaddrs[ns] = |
325 | malloc(size: sizeof (struct sockaddr_in6)); |
326 | if (EXT(statp).nsaddrs[ns] != NULL) |
327 | memset (mempcpy(EXT(statp).nsaddrs[ns], |
328 | &statp->nsaddr_list[ns], |
329 | sizeof (struct sockaddr_in)), |
330 | '\0', |
331 | sizeof (struct sockaddr_in6) |
332 | - sizeof (struct sockaddr_in)); |
333 | else |
334 | return -1; |
335 | } |
336 | EXT(statp).nscount = statp->nscount; |
337 | } |
338 | |
339 | /* Name server index offset. Used to implement |
340 | RES_ROTATE. */ |
341 | unsigned int ns_offset = nameserver_offset (statp); |
342 | |
343 | /* |
344 | * Send request, RETRY times, or until successful. |
345 | */ |
346 | for (try = 0; try < statp->retry; try++) { |
347 | for (unsigned ns_shift = 0; ns_shift < statp->nscount; ns_shift++) |
348 | { |
349 | /* The actual name server index. This implements |
350 | RES_ROTATE. */ |
351 | unsigned int ns = ns_shift + ns_offset; |
352 | if (ns >= statp->nscount) |
353 | ns -= statp->nscount; |
354 | |
355 | same_ns: |
356 | if (__glibc_unlikely (v_circuit)) { |
357 | /* Use VC; at most one attempt per server. */ |
358 | try = statp->retry; |
359 | n = send_vc(statp, buf, buflen, buf2, buflen2, |
360 | &ans, &anssiz, &terrno, |
361 | ns, ansp, ansp2, nansp2, resplen2, |
362 | ansp2_malloced); |
363 | if (n < 0) |
364 | return (-1); |
365 | /* See comment at the declaration of n. */ |
366 | DIAG_PUSH_NEEDS_COMMENT; |
367 | DIAG_IGNORE_NEEDS_COMMENT (9, "-Wmaybe-uninitialized" ); |
368 | if (n == 0 && (buf2 == NULL || *resplen2 == 0)) |
369 | goto next_ns; |
370 | DIAG_POP_NEEDS_COMMENT; |
371 | } else { |
372 | /* Use datagrams. */ |
373 | n = send_dg(statp, buf, buflen, buf2, buflen2, |
374 | &ans, &anssiz, &terrno, |
375 | ns, &v_circuit, &gotsomewhere, ansp, |
376 | ansp2, nansp2, resplen2, ansp2_malloced); |
377 | if (n < 0) |
378 | return (-1); |
379 | if (n == 0 && (buf2 == NULL || *resplen2 == 0)) |
380 | goto next_ns; |
381 | if (v_circuit) |
382 | // XXX Check whether both requests failed or |
383 | // XXX whether one has been answered successfully |
384 | goto same_ns; |
385 | } |
386 | |
387 | resplen = n; |
388 | |
389 | /* See comment at the declaration of n. Note: resplen = n; */ |
390 | DIAG_PUSH_NEEDS_COMMENT; |
391 | DIAG_IGNORE_NEEDS_COMMENT (9, "-Wmaybe-uninitialized" ); |
392 | /* Mask the AD bit in both responses unless it is |
393 | marked trusted. */ |
394 | if (resplen > HFIXEDSZ) |
395 | { |
396 | if (ansp != NULL) |
397 | mask_ad_bit (ctx, buf: *ansp); |
398 | else |
399 | mask_ad_bit (ctx, buf: ans); |
400 | } |
401 | DIAG_POP_NEEDS_COMMENT; |
402 | if (resplen2 != NULL && *resplen2 > HFIXEDSZ) |
403 | mask_ad_bit (ctx, buf: *ansp2); |
404 | |
405 | /* |
406 | * If we have temporarily opened a virtual circuit, |
407 | * or if we haven't been asked to keep a socket open, |
408 | * close the socket. |
409 | */ |
410 | if ((v_circuit && (statp->options & RES_USEVC) == 0) || |
411 | (statp->options & RES_STAYOPEN) == 0) { |
412 | __res_iclose(statp, false); |
413 | } |
414 | return (resplen); |
415 | next_ns: ; |
416 | } /*foreach ns*/ |
417 | } /*foreach retry*/ |
418 | __res_iclose(statp, false); |
419 | if (!v_circuit) { |
420 | if (!gotsomewhere) |
421 | __set_errno (ECONNREFUSED); /* no nameservers found */ |
422 | else |
423 | __set_errno (ETIMEDOUT); /* no answer obtained */ |
424 | } else |
425 | __set_errno (terrno); |
426 | return (-1); |
427 | } |
428 | libc_hidden_def (__res_context_send) |
429 | |
430 | /* Common part of res_nsend and res_send. */ |
431 | static int |
432 | context_send_common (struct resolv_context *ctx, |
433 | const unsigned char *buf, int buflen, |
434 | unsigned char *ans, int anssiz) |
435 | { |
436 | if (ctx == NULL) |
437 | { |
438 | RES_SET_H_ERRNO (&_res, NETDB_INTERNAL); |
439 | return -1; |
440 | } |
441 | |
442 | int result; |
443 | if (__res_handle_no_aaaa (ctx, buf, buflen, ans, anssiz, result: &result)) |
444 | return result; |
445 | |
446 | result = __res_context_send (ctx, buf, buflen, NULL, buflen2: 0, ans, anssiz, |
447 | NULL, NULL, NULL, NULL, NULL); |
448 | __resolv_context_put (ctx); |
449 | return result; |
450 | } |
451 | |
452 | int |
453 | ___res_nsend (res_state statp, const unsigned char *buf, int buflen, |
454 | unsigned char *ans, int anssiz) |
455 | { |
456 | return context_send_common |
457 | (ctx: __resolv_context_get_override (statp), buf, buflen, ans, anssiz); |
458 | } |
459 | versioned_symbol (libc, ___res_nsend, res_nsend, GLIBC_2_34); |
460 | #if OTHER_SHLIB_COMPAT (libresolv, GLIBC_2_2, GLIBC_2_34) |
461 | compat_symbol (libresolv, ___res_nsend, __res_nsend, GLIBC_2_2); |
462 | #endif |
463 | |
464 | int |
465 | ___res_send (const unsigned char *buf, int buflen, unsigned char *ans, |
466 | int anssiz) |
467 | { |
468 | return context_send_common |
469 | (ctx: __resolv_context_get (), buf, buflen, ans, anssiz); |
470 | } |
471 | versioned_symbol (libc, ___res_send, res_send, GLIBC_2_34); |
472 | #if OTHER_SHLIB_COMPAT (libresolv, GLIBC_2_0, GLIBC_2_34) |
473 | compat_symbol (libresolv, ___res_send, __res_send, GLIBC_2_0); |
474 | #endif |
475 | |
476 | /* Private */ |
477 | |
478 | /* Close the resolver structure, assign zero to *RESPLEN2 if RESPLEN2 |
479 | is not NULL, and return zero. */ |
480 | static int |
481 | __attribute__ ((warn_unused_result)) |
482 | close_and_return_error (res_state statp, int *resplen2) |
483 | { |
484 | __res_iclose(statp, false); |
485 | if (resplen2 != NULL) |
486 | *resplen2 = 0; |
487 | return 0; |
488 | } |
489 | |
490 | /* The send_vc function is responsible for sending a DNS query over TCP |
491 | to the nameserver numbered NS from the res_state STATP i.e. |
492 | EXT(statp).nssocks[ns]. The function supports sending both IPv4 and |
493 | IPv6 queries at the same serially on the same socket. |
494 | |
495 | Please note that for TCP there is no way to disable sending both |
496 | queries, unlike UDP, which honours RES_SNGLKUP and RES_SNGLKUPREOP |
497 | and sends the queries serially and waits for the result after each |
498 | sent query. This implementation should be corrected to honour these |
499 | options. |
500 | |
501 | Please also note that for TCP we send both queries over the same |
502 | socket one after another. This technically violates best practice |
503 | since the server is allowed to read the first query, respond, and |
504 | then close the socket (to service another client). If the server |
505 | does this, then the remaining second query in the socket data buffer |
506 | will cause the server to send the client an RST which will arrive |
507 | asynchronously and the client's OS will likely tear down the socket |
508 | receive buffer resulting in a potentially short read and lost |
509 | response data. This will force the client to retry the query again, |
510 | and this process may repeat until all servers and connection resets |
511 | are exhausted and then the query will fail. It's not known if this |
512 | happens with any frequency in real DNS server implementations. This |
513 | implementation should be corrected to use two sockets by default for |
514 | parallel queries. |
515 | |
516 | The query stored in BUF of BUFLEN length is sent first followed by |
517 | the query stored in BUF2 of BUFLEN2 length. Queries are sent |
518 | serially on the same socket. |
519 | |
520 | Answers to the query are stored firstly in *ANSP up to a max of |
521 | *ANSSIZP bytes. If more than *ANSSIZP bytes are needed and ANSCP |
522 | is non-NULL (to indicate that modifying the answer buffer is allowed) |
523 | then malloc is used to allocate a new response buffer and ANSCP and |
524 | ANSP will both point to the new buffer. If more than *ANSSIZP bytes |
525 | are needed but ANSCP is NULL, then as much of the response as |
526 | possible is read into the buffer, but the results will be truncated. |
527 | When truncation happens because of a small answer buffer the DNS |
528 | packets header field TC will bet set to 1, indicating a truncated |
529 | message and the rest of the socket data will be read and discarded. |
530 | |
531 | Answers to the query are stored secondly in *ANSP2 up to a max of |
532 | *ANSSIZP2 bytes, with the actual response length stored in |
533 | *RESPLEN2. If more than *ANSSIZP bytes are needed and ANSP2 |
534 | is non-NULL (required for a second query) then malloc is used to |
535 | allocate a new response buffer, *ANSSIZP2 is set to the new buffer |
536 | size and *ANSP2_MALLOCED is set to 1. |
537 | |
538 | The ANSP2_MALLOCED argument will eventually be removed as the |
539 | change in buffer pointer can be used to detect the buffer has |
540 | changed and that the caller should use free on the new buffer. |
541 | |
542 | Note that the answers may arrive in any order from the server and |
543 | therefore the first and second answer buffers may not correspond to |
544 | the first and second queries. |
545 | |
546 | It is not supported to call this function with a non-NULL ANSP2 |
547 | but a NULL ANSCP. Put another way, you can call send_vc with a |
548 | single unmodifiable buffer or two modifiable buffers, but no other |
549 | combination is supported. |
550 | |
551 | It is the caller's responsibility to free the malloc allocated |
552 | buffers by detecting that the pointers have changed from their |
553 | original values i.e. *ANSCP or *ANSP2 has changed. |
554 | |
555 | If errors are encountered then *TERRNO is set to an appropriate |
556 | errno value and a zero result is returned for a recoverable error, |
557 | and a less-than zero result is returned for a non-recoverable error. |
558 | |
559 | If no errors are encountered then *TERRNO is left unmodified and |
560 | a the length of the first response in bytes is returned. */ |
561 | static int |
562 | send_vc(res_state statp, |
563 | const u_char *buf, int buflen, const u_char *buf2, int buflen2, |
564 | u_char **ansp, int *anssizp, |
565 | int *terrno, int ns, u_char **anscp, u_char **ansp2, int *anssizp2, |
566 | int *resplen2, int *ansp2_malloced) |
567 | { |
568 | const UHEADER *hp = (UHEADER *) buf; |
569 | const UHEADER *hp2 = (UHEADER *) buf2; |
570 | UHEADER *anhp = (UHEADER *) *ansp; |
571 | struct sockaddr *nsap = __res_get_nsaddr (statp, ns); |
572 | int truncating, connreset, n; |
573 | /* On some architectures compiler might emit a warning indicating |
574 | 'resplen' may be used uninitialized. However if buf2 == NULL |
575 | then this code won't be executed; if buf2 != NULL, then first |
576 | time round the loop recvresp1 and recvresp2 will be 0 so this |
577 | code won't be executed but "thisresplenp = &resplen;" followed |
578 | by "*thisresplenp = rlen;" will be executed so that subsequent |
579 | times round the loop resplen has been initialized. So this is |
580 | a false-positive. |
581 | */ |
582 | DIAG_PUSH_NEEDS_COMMENT; |
583 | DIAG_IGNORE_NEEDS_COMMENT (5, "-Wmaybe-uninitialized" ); |
584 | int resplen; |
585 | DIAG_POP_NEEDS_COMMENT; |
586 | struct iovec iov[4]; |
587 | u_short len; |
588 | u_short len2; |
589 | u_char *cp; |
590 | |
591 | connreset = 0; |
592 | same_ns: |
593 | truncating = 0; |
594 | |
595 | /* Are we still talking to whom we want to talk to? */ |
596 | if (statp->_vcsock >= 0 && (statp->_flags & RES_F_VC) != 0) { |
597 | struct sockaddr_in6 peer; |
598 | socklen_t size = sizeof peer; |
599 | |
600 | if (__getpeername (fd: statp->_vcsock, |
601 | addr: (struct sockaddr *) &peer, len: &size) < 0 |
602 | || !sock_eq (&peer, (struct sockaddr_in6 *) nsap)) { |
603 | __res_iclose(statp, false); |
604 | statp->_flags &= ~RES_F_VC; |
605 | } |
606 | } |
607 | |
608 | if (statp->_vcsock < 0 || (statp->_flags & RES_F_VC) == 0) { |
609 | if (statp->_vcsock >= 0) |
610 | __res_iclose(statp, false); |
611 | |
612 | statp->_vcsock = __socket |
613 | (nsap->sa_family, SOCK_STREAM | SOCK_CLOEXEC, 0); |
614 | if (statp->_vcsock < 0) { |
615 | *terrno = errno; |
616 | if (resplen2 != NULL) |
617 | *resplen2 = 0; |
618 | return (-1); |
619 | } |
620 | __set_errno (0); |
621 | if (__connect (statp->_vcsock, nsap, |
622 | nsap->sa_family == AF_INET |
623 | ? sizeof (struct sockaddr_in) |
624 | : sizeof (struct sockaddr_in6)) < 0) { |
625 | *terrno = errno; |
626 | return close_and_return_error (statp, resplen2); |
627 | } |
628 | statp->_flags |= RES_F_VC; |
629 | } |
630 | |
631 | /* |
632 | * Send length & message |
633 | */ |
634 | len = htons ((u_short) buflen); |
635 | evConsIovec(buf: &len, INT16SZ, vec: &iov[0]); |
636 | evConsIovec(buf: (void*)buf, cnt: buflen, vec: &iov[1]); |
637 | int niov = 2; |
638 | ssize_t explen = INT16SZ + buflen; |
639 | if (buf2 != NULL) { |
640 | len2 = htons ((u_short) buflen2); |
641 | evConsIovec(buf: &len2, INT16SZ, vec: &iov[2]); |
642 | evConsIovec(buf: (void*)buf2, cnt: buflen2, vec: &iov[3]); |
643 | niov = 4; |
644 | explen += INT16SZ + buflen2; |
645 | } |
646 | if (TEMP_FAILURE_RETRY (__writev (statp->_vcsock, iov, niov)) |
647 | != explen) { |
648 | *terrno = errno; |
649 | return close_and_return_error (statp, resplen2); |
650 | } |
651 | /* |
652 | * Receive length & response |
653 | */ |
654 | int recvresp1 = 0; |
655 | /* Skip the second response if there is no second query. |
656 | To do that we mark the second response as received. */ |
657 | int recvresp2 = buf2 == NULL; |
658 | uint16_t rlen16; |
659 | read_len: |
660 | cp = (u_char *)&rlen16; |
661 | len = sizeof(rlen16); |
662 | while ((n = TEMP_FAILURE_RETRY (read(statp->_vcsock, cp, |
663 | (int)len))) > 0) { |
664 | cp += n; |
665 | if ((len -= n) <= 0) |
666 | break; |
667 | } |
668 | if (n <= 0) { |
669 | *terrno = errno; |
670 | /* |
671 | * A long running process might get its TCP |
672 | * connection reset if the remote server was |
673 | * restarted. Requery the server instead of |
674 | * trying a new one. When there is only one |
675 | * server, this means that a query might work |
676 | * instead of failing. We only allow one reset |
677 | * per query to prevent looping. |
678 | */ |
679 | if (*terrno == ECONNRESET && !connreset) |
680 | { |
681 | __res_iclose (statp, false); |
682 | connreset = 1; |
683 | goto same_ns; |
684 | } |
685 | return close_and_return_error (statp, resplen2); |
686 | } |
687 | int rlen = ntohs (rlen16); |
688 | |
689 | int *thisanssizp; |
690 | u_char **thisansp; |
691 | int *thisresplenp; |
692 | if ((recvresp1 | recvresp2) == 0 || buf2 == NULL) { |
693 | /* We have not received any responses |
694 | yet or we only have one response to |
695 | receive. */ |
696 | thisanssizp = anssizp; |
697 | thisansp = anscp ?: ansp; |
698 | assert (anscp != NULL || ansp2 == NULL); |
699 | thisresplenp = &resplen; |
700 | } else { |
701 | thisanssizp = anssizp2; |
702 | thisansp = ansp2; |
703 | thisresplenp = resplen2; |
704 | } |
705 | anhp = (UHEADER *) *thisansp; |
706 | |
707 | *thisresplenp = rlen; |
708 | /* Is the answer buffer too small? */ |
709 | if (*thisanssizp < rlen) { |
710 | /* If the current buffer is not the the static |
711 | user-supplied buffer then we can reallocate |
712 | it. */ |
713 | if (thisansp != NULL && thisansp != ansp) { |
714 | /* Always allocate MAXPACKET, callers expect |
715 | this specific size. */ |
716 | u_char *newp = malloc (MAXPACKET); |
717 | if (newp == NULL) |
718 | { |
719 | *terrno = ENOMEM; |
720 | return close_and_return_error (statp, resplen2); |
721 | } |
722 | *thisanssizp = MAXPACKET; |
723 | *thisansp = newp; |
724 | if (thisansp == ansp2) |
725 | *ansp2_malloced = 1; |
726 | anhp = (UHEADER *) newp; |
727 | /* A uint16_t can't be larger than MAXPACKET |
728 | thus it's safe to allocate MAXPACKET but |
729 | read RLEN bytes instead. */ |
730 | len = rlen; |
731 | } else { |
732 | truncating = 1; |
733 | len = *thisanssizp; |
734 | } |
735 | } else |
736 | len = rlen; |
737 | |
738 | if (__glibc_unlikely (len < HFIXEDSZ)) { |
739 | /* |
740 | * Undersized message. |
741 | */ |
742 | *terrno = EMSGSIZE; |
743 | return close_and_return_error (statp, resplen2); |
744 | } |
745 | |
746 | cp = *thisansp; |
747 | while (len != 0 && (n = read(statp->_vcsock, (char *)cp, (int)len)) > 0){ |
748 | cp += n; |
749 | len -= n; |
750 | } |
751 | if (__glibc_unlikely (n <= 0)) { |
752 | *terrno = errno; |
753 | return close_and_return_error (statp, resplen2); |
754 | } |
755 | if (__glibc_unlikely (truncating)) { |
756 | /* |
757 | * Flush rest of answer so connection stays in synch. |
758 | */ |
759 | anhp->tc = 1; |
760 | len = rlen - *thisanssizp; |
761 | while (len != 0) { |
762 | char junk[PACKETSZ]; |
763 | |
764 | n = read(statp->_vcsock, junk, |
765 | (len > sizeof junk) ? sizeof junk : len); |
766 | if (n > 0) |
767 | len -= n; |
768 | else |
769 | break; |
770 | } |
771 | } |
772 | /* |
773 | * If the calling application has bailed out of |
774 | * a previous call and failed to arrange to have |
775 | * the circuit closed or the server has got |
776 | * itself confused, then drop the packet and |
777 | * wait for the correct one. |
778 | */ |
779 | if ((recvresp1 || hp->id != anhp->id) |
780 | && (recvresp2 || hp2->id != anhp->id)) |
781 | goto read_len; |
782 | |
783 | /* Mark which reply we received. */ |
784 | if (recvresp1 == 0 && hp->id == anhp->id) |
785 | recvresp1 = 1; |
786 | else |
787 | recvresp2 = 1; |
788 | /* Repeat waiting if we have a second answer to arrive. */ |
789 | if ((recvresp1 & recvresp2) == 0) |
790 | goto read_len; |
791 | |
792 | /* |
793 | * All is well, or the error is fatal. Signal that the |
794 | * next nameserver ought not be tried. |
795 | */ |
796 | return resplen; |
797 | } |
798 | |
799 | static int |
800 | reopen (res_state statp, int *terrno, int ns) |
801 | { |
802 | if (EXT(statp).nssocks[ns] == -1) { |
803 | struct sockaddr *nsap = __res_get_nsaddr (statp, ns); |
804 | socklen_t slen; |
805 | |
806 | /* only try IPv6 if IPv6 NS and if not failed before */ |
807 | if (nsap->sa_family == AF_INET6 && !statp->ipv6_unavail) { |
808 | EXT (statp).nssocks[ns] = __socket |
809 | (PF_INET6, |
810 | SOCK_DGRAM | SOCK_NONBLOCK | SOCK_CLOEXEC, 0); |
811 | if (EXT(statp).nssocks[ns] < 0) |
812 | statp->ipv6_unavail = errno == EAFNOSUPPORT; |
813 | slen = sizeof (struct sockaddr_in6); |
814 | } else if (nsap->sa_family == AF_INET) { |
815 | EXT (statp).nssocks[ns] = __socket |
816 | (PF_INET, |
817 | SOCK_DGRAM | SOCK_NONBLOCK | SOCK_CLOEXEC, 0); |
818 | slen = sizeof (struct sockaddr_in); |
819 | } |
820 | if (EXT(statp).nssocks[ns] < 0) { |
821 | *terrno = errno; |
822 | return (-1); |
823 | } |
824 | |
825 | /* Enable full ICMP error reporting for this |
826 | socket. */ |
827 | if (__res_enable_icmp (family: nsap->sa_family, |
828 | EXT (statp).nssocks[ns]) < 0) |
829 | { |
830 | int saved_errno = errno; |
831 | __res_iclose (statp, false); |
832 | __set_errno (saved_errno); |
833 | *terrno = saved_errno; |
834 | return -1; |
835 | } |
836 | |
837 | /* |
838 | * On a 4.3BSD+ machine (client and server, |
839 | * actually), sending to a nameserver datagram |
840 | * port with no nameserver will cause an |
841 | * ICMP port unreachable message to be returned. |
842 | * If our datagram socket is "connected" to the |
843 | * server, we get an ECONNREFUSED error on the next |
844 | * socket operation, and select returns if the |
845 | * error message is received. We can thus detect |
846 | * the absence of a nameserver without timing out. |
847 | */ |
848 | /* With GCC 5.3 when compiling with -Os the compiler |
849 | emits a warning that slen may be used uninitialized, |
850 | but that is never true. Both slen and |
851 | EXT(statp).nssocks[ns] are initialized together or |
852 | the function return -1 before control flow reaches |
853 | the call to connect with slen. */ |
854 | DIAG_PUSH_NEEDS_COMMENT; |
855 | DIAG_IGNORE_Os_NEEDS_COMMENT (5, "-Wmaybe-uninitialized" ); |
856 | if (__connect (EXT (statp).nssocks[ns], nsap, slen) < 0) { |
857 | DIAG_POP_NEEDS_COMMENT; |
858 | __res_iclose(statp, false); |
859 | return (0); |
860 | } |
861 | } |
862 | |
863 | return 1; |
864 | } |
865 | |
866 | /* The send_dg function is responsible for sending a DNS query over UDP |
867 | to the nameserver numbered NS from the res_state STATP i.e. |
868 | EXT(statp).nssocks[ns]. The function supports IPv4 and IPv6 queries |
869 | along with the ability to send the query in parallel for both stacks |
870 | (default) or serially (RES_SINGLKUP). It also supports serial lookup |
871 | with a close and reopen of the socket used to talk to the server |
872 | (RES_SNGLKUPREOP) to work around broken name servers. |
873 | |
874 | The query stored in BUF of BUFLEN length is sent first followed by |
875 | the query stored in BUF2 of BUFLEN2 length. Queries are sent |
876 | in parallel (default) or serially (RES_SINGLKUP or RES_SNGLKUPREOP). |
877 | |
878 | Answers to the query are stored firstly in *ANSP up to a max of |
879 | *ANSSIZP bytes. If more than *ANSSIZP bytes are needed and ANSCP |
880 | is non-NULL (to indicate that modifying the answer buffer is allowed) |
881 | then malloc is used to allocate a new response buffer and ANSCP and |
882 | ANSP will both point to the new buffer. If more than *ANSSIZP bytes |
883 | are needed but ANSCP is NULL, then as much of the response as |
884 | possible is read into the buffer, but the results will be truncated. |
885 | When truncation happens because of a small answer buffer the DNS |
886 | packets header field TC will bet set to 1, indicating a truncated |
887 | message, while the rest of the UDP packet is discarded. |
888 | |
889 | Answers to the query are stored secondly in *ANSP2 up to a max of |
890 | *ANSSIZP2 bytes, with the actual response length stored in |
891 | *RESPLEN2. If more than *ANSSIZP bytes are needed and ANSP2 |
892 | is non-NULL (required for a second query) then malloc is used to |
893 | allocate a new response buffer, *ANSSIZP2 is set to the new buffer |
894 | size and *ANSP2_MALLOCED is set to 1. |
895 | |
896 | The ANSP2_MALLOCED argument will eventually be removed as the |
897 | change in buffer pointer can be used to detect the buffer has |
898 | changed and that the caller should use free on the new buffer. |
899 | |
900 | Note that the answers may arrive in any order from the server and |
901 | therefore the first and second answer buffers may not correspond to |
902 | the first and second queries. |
903 | |
904 | It is not supported to call this function with a non-NULL ANSP2 |
905 | but a NULL ANSCP. Put another way, you can call send_vc with a |
906 | single unmodifiable buffer or two modifiable buffers, but no other |
907 | combination is supported. |
908 | |
909 | It is the caller's responsibility to free the malloc allocated |
910 | buffers by detecting that the pointers have changed from their |
911 | original values i.e. *ANSCP or *ANSP2 has changed. |
912 | |
913 | If an answer is truncated because of UDP datagram DNS limits then |
914 | *V_CIRCUIT is set to 1 and the return value non-zero to indicate to |
915 | the caller to retry with TCP. The value *GOTSOMEWHERE is set to 1 |
916 | if any progress was made reading a response from the nameserver and |
917 | is used by the caller to distinguish between ECONNREFUSED and |
918 | ETIMEDOUT (the latter if *GOTSOMEWHERE is 1). |
919 | |
920 | If errors are encountered then *TERRNO is set to an appropriate |
921 | errno value and a zero result is returned for a recoverable error, |
922 | and a less-than zero result is returned for a non-recoverable error. |
923 | |
924 | If no errors are encountered then *TERRNO is left unmodified and |
925 | a the length of the first response in bytes is returned. */ |
926 | static int |
927 | send_dg(res_state statp, |
928 | const u_char *buf, int buflen, const u_char *buf2, int buflen2, |
929 | u_char **ansp, int *anssizp, |
930 | int *terrno, int ns, int *v_circuit, int *gotsomewhere, u_char **anscp, |
931 | u_char **ansp2, int *anssizp2, int *resplen2, int *ansp2_malloced) |
932 | { |
933 | const UHEADER *hp = (UHEADER *) buf; |
934 | const UHEADER *hp2 = (UHEADER *) buf2; |
935 | struct timespec now, timeout, finish; |
936 | struct pollfd pfd[1]; |
937 | int ptimeout; |
938 | struct sockaddr_in6 from; |
939 | int resplen = 0; |
940 | int n; |
941 | |
942 | /* |
943 | * Compute time for the total operation. |
944 | */ |
945 | int seconds = (statp->retrans << ns); |
946 | if (ns > 0) |
947 | seconds /= statp->nscount; |
948 | if (seconds <= 0) |
949 | seconds = 1; |
950 | bool single_request_reopen = (statp->options & RES_SNGLKUPREOP) != 0; |
951 | bool single_request = (((statp->options & RES_SNGLKUP) != 0) |
952 | | single_request_reopen); |
953 | int save_gotsomewhere = *gotsomewhere; |
954 | |
955 | int retval; |
956 | retry_reopen: |
957 | retval = reopen (statp, terrno, ns); |
958 | if (retval <= 0) |
959 | { |
960 | if (resplen2 != NULL) |
961 | *resplen2 = 0; |
962 | return retval; |
963 | } |
964 | retry: |
965 | evNowTime(res: &now); |
966 | evConsTime(res: &timeout, sec: seconds, nsec: 0); |
967 | evAddTime(res: &finish, addend1: &now, addend2: &timeout); |
968 | int need_recompute = 0; |
969 | int nwritten = 0; |
970 | int recvresp1 = 0; |
971 | /* Skip the second response if there is no second query. |
972 | To do that we mark the second response as received. */ |
973 | int recvresp2 = buf2 == NULL; |
974 | pfd[0].fd = EXT(statp).nssocks[ns]; |
975 | pfd[0].events = POLLOUT; |
976 | wait: |
977 | if (need_recompute) { |
978 | recompute_resend: |
979 | evNowTime(res: &now); |
980 | if (evCmpTime(a: finish, b: now) <= 0) { |
981 | poll_err_out: |
982 | return close_and_return_error (statp, resplen2); |
983 | } |
984 | evSubTime(res: &timeout, minuend: &finish, subtrahend: &now); |
985 | need_recompute = 0; |
986 | } |
987 | /* Convert struct timespec in milliseconds. */ |
988 | ptimeout = timeout.tv_sec * 1000 + timeout.tv_nsec / 1000000; |
989 | |
990 | n = 0; |
991 | if (nwritten == 0) |
992 | n = __poll (pfd, 1, 0); |
993 | if (__glibc_unlikely (n == 0)) { |
994 | n = __poll (pfd, 1, ptimeout); |
995 | need_recompute = 1; |
996 | } |
997 | if (n == 0) { |
998 | if (resplen > 1 && (recvresp1 || (buf2 != NULL && recvresp2))) |
999 | { |
1000 | /* There are quite a few broken name servers out |
1001 | there which don't handle two outstanding |
1002 | requests from the same source. There are also |
1003 | broken firewall settings. If we time out after |
1004 | having received one answer switch to the mode |
1005 | where we send the second request only once we |
1006 | have received the first answer. */ |
1007 | if (!single_request) |
1008 | { |
1009 | statp->options |= RES_SNGLKUP; |
1010 | single_request = true; |
1011 | *gotsomewhere = save_gotsomewhere; |
1012 | goto retry; |
1013 | } |
1014 | else if (!single_request_reopen) |
1015 | { |
1016 | statp->options |= RES_SNGLKUPREOP; |
1017 | single_request_reopen = true; |
1018 | *gotsomewhere = save_gotsomewhere; |
1019 | __res_iclose (statp, false); |
1020 | goto retry_reopen; |
1021 | } |
1022 | |
1023 | *resplen2 = 1; |
1024 | return resplen; |
1025 | } |
1026 | |
1027 | *gotsomewhere = 1; |
1028 | if (resplen2 != NULL) |
1029 | *resplen2 = 0; |
1030 | return 0; |
1031 | } |
1032 | if (n < 0) { |
1033 | if (errno == EINTR) |
1034 | goto recompute_resend; |
1035 | |
1036 | goto poll_err_out; |
1037 | } |
1038 | __set_errno (0); |
1039 | if (pfd[0].revents & POLLOUT) { |
1040 | #ifndef __ASSUME_SENDMMSG |
1041 | static int have_sendmmsg; |
1042 | #else |
1043 | # define have_sendmmsg 1 |
1044 | #endif |
1045 | if (have_sendmmsg >= 0 && nwritten == 0 && buf2 != NULL |
1046 | && !single_request) |
1047 | { |
1048 | struct iovec iov = |
1049 | { .iov_base = (void *) buf, .iov_len = buflen }; |
1050 | struct iovec iov2 = |
1051 | { .iov_base = (void *) buf2, .iov_len = buflen2 }; |
1052 | struct mmsghdr reqs[2] = |
1053 | { |
1054 | { |
1055 | .msg_hdr = |
1056 | { |
1057 | .msg_iov = &iov, |
1058 | .msg_iovlen = 1, |
1059 | }, |
1060 | }, |
1061 | { |
1062 | .msg_hdr = |
1063 | { |
1064 | .msg_iov = &iov2, |
1065 | .msg_iovlen = 1, |
1066 | } |
1067 | }, |
1068 | }; |
1069 | |
1070 | int ndg = __sendmmsg (pfd[0].fd, reqs, 2, MSG_NOSIGNAL); |
1071 | if (__glibc_likely (ndg == 2)) |
1072 | { |
1073 | if (reqs[0].msg_len != buflen |
1074 | || reqs[1].msg_len != buflen2) |
1075 | goto fail_sendmmsg; |
1076 | |
1077 | pfd[0].events = POLLIN; |
1078 | nwritten += 2; |
1079 | } |
1080 | else if (ndg == 1 && reqs[0].msg_len == buflen) |
1081 | goto just_one; |
1082 | else if (ndg < 0 && (errno == EINTR || errno == EAGAIN)) |
1083 | goto recompute_resend; |
1084 | else |
1085 | { |
1086 | #ifndef __ASSUME_SENDMMSG |
1087 | if (__glibc_unlikely (have_sendmmsg == 0)) |
1088 | { |
1089 | if (ndg < 0 && errno == ENOSYS) |
1090 | { |
1091 | have_sendmmsg = -1; |
1092 | goto try_send; |
1093 | } |
1094 | have_sendmmsg = 1; |
1095 | } |
1096 | #endif |
1097 | |
1098 | fail_sendmmsg: |
1099 | return close_and_return_error (statp, resplen2); |
1100 | } |
1101 | } |
1102 | else |
1103 | { |
1104 | ssize_t sr; |
1105 | #ifndef __ASSUME_SENDMMSG |
1106 | try_send: |
1107 | #endif |
1108 | if (nwritten != 0) |
1109 | sr = __send (pfd[0].fd, buf2, buflen2, MSG_NOSIGNAL); |
1110 | else |
1111 | sr = __send (pfd[0].fd, buf, buflen, MSG_NOSIGNAL); |
1112 | |
1113 | if (sr != (nwritten != 0 ? buflen2 : buflen)) { |
1114 | if (errno == EINTR || errno == EAGAIN) |
1115 | goto recompute_resend; |
1116 | return close_and_return_error (statp, resplen2); |
1117 | } |
1118 | just_one: |
1119 | if (nwritten != 0 || buf2 == NULL || single_request) |
1120 | pfd[0].events = POLLIN; |
1121 | else |
1122 | pfd[0].events = POLLIN | POLLOUT; |
1123 | ++nwritten; |
1124 | } |
1125 | goto wait; |
1126 | } else if (pfd[0].revents & POLLIN) { |
1127 | int *thisanssizp; |
1128 | u_char **thisansp; |
1129 | int *thisresplenp; |
1130 | |
1131 | if ((recvresp1 | recvresp2) == 0 || buf2 == NULL) { |
1132 | /* We have not received any responses |
1133 | yet or we only have one response to |
1134 | receive. */ |
1135 | thisanssizp = anssizp; |
1136 | thisansp = anscp ?: ansp; |
1137 | assert (anscp != NULL || ansp2 == NULL); |
1138 | thisresplenp = &resplen; |
1139 | } else { |
1140 | thisanssizp = anssizp2; |
1141 | thisansp = ansp2; |
1142 | thisresplenp = resplen2; |
1143 | } |
1144 | |
1145 | if (*thisanssizp < MAXPACKET |
1146 | /* If the current buffer is not the the static |
1147 | user-supplied buffer then we can reallocate |
1148 | it. */ |
1149 | && (thisansp != NULL && thisansp != ansp) |
1150 | #ifdef FIONREAD |
1151 | /* Is the size too small? */ |
1152 | && (__ioctl (pfd[0].fd, FIONREAD, thisresplenp) < 0 |
1153 | || *thisanssizp < *thisresplenp) |
1154 | #endif |
1155 | ) { |
1156 | /* Always allocate MAXPACKET, callers expect |
1157 | this specific size. */ |
1158 | u_char *newp = malloc (MAXPACKET); |
1159 | if (newp != NULL) { |
1160 | *thisanssizp = MAXPACKET; |
1161 | *thisansp = newp; |
1162 | if (thisansp == ansp2) |
1163 | *ansp2_malloced = 1; |
1164 | } |
1165 | } |
1166 | /* We could end up with truncation if anscp was NULL |
1167 | (not allowed to change caller's buffer) and the |
1168 | response buffer size is too small. This isn't a |
1169 | reliable way to detect truncation because the ioctl |
1170 | may be an inaccurate report of the UDP message size. |
1171 | Therefore we use this only to issue debug output. |
1172 | To do truncation accurately with UDP we need |
1173 | MSG_TRUNC which is only available on Linux. We |
1174 | can abstract out the Linux-specific feature in the |
1175 | future to detect truncation. */ |
1176 | UHEADER *anhp = (UHEADER *) *thisansp; |
1177 | socklen_t fromlen = sizeof(struct sockaddr_in6); |
1178 | assert (sizeof(from) <= fromlen); |
1179 | *thisresplenp = __recvfrom (fd: pfd[0].fd, buf: (char *) *thisansp, |
1180 | n: *thisanssizp, flags: 0, |
1181 | addr: (struct sockaddr *) &from, |
1182 | addr_len: &fromlen); |
1183 | if (__glibc_unlikely (*thisresplenp <= 0)) { |
1184 | if (errno == EINTR || errno == EAGAIN) { |
1185 | need_recompute = 1; |
1186 | goto wait; |
1187 | } |
1188 | return close_and_return_error (statp, resplen2); |
1189 | } |
1190 | *gotsomewhere = 1; |
1191 | if (__glibc_unlikely (*thisresplenp < HFIXEDSZ)) { |
1192 | /* |
1193 | * Undersized message. |
1194 | */ |
1195 | *terrno = EMSGSIZE; |
1196 | return close_and_return_error (statp, resplen2); |
1197 | } |
1198 | |
1199 | /* Check for the correct header layout and a matching |
1200 | question. */ |
1201 | int matching_query = 0; /* Default to no matching query. */ |
1202 | if (!recvresp1 |
1203 | && anhp->id == hp->id |
1204 | && __libc_res_queriesmatch (buf, buf + buflen, |
1205 | *thisansp, |
1206 | *thisansp + *thisanssizp)) |
1207 | matching_query = 1; |
1208 | if (!recvresp2 |
1209 | && anhp->id == hp2->id |
1210 | && __libc_res_queriesmatch (buf2, buf2 + buflen2, |
1211 | *thisansp, |
1212 | *thisansp + *thisanssizp)) |
1213 | matching_query = 2; |
1214 | if (matching_query == 0) |
1215 | /* Spurious UDP packet. Drop it and continue |
1216 | waiting. */ |
1217 | { |
1218 | need_recompute = 1; |
1219 | goto wait; |
1220 | } |
1221 | |
1222 | if (anhp->rcode == SERVFAIL || |
1223 | anhp->rcode == NOTIMP || |
1224 | anhp->rcode == REFUSED) { |
1225 | next_ns: |
1226 | if (recvresp1 || (buf2 != NULL && recvresp2)) { |
1227 | *resplen2 = 0; |
1228 | return resplen; |
1229 | } |
1230 | if (buf2 != NULL) |
1231 | { |
1232 | /* No data from the first reply. */ |
1233 | resplen = 0; |
1234 | /* We are waiting for a possible second reply. */ |
1235 | if (matching_query == 1) |
1236 | recvresp1 = 1; |
1237 | else |
1238 | recvresp2 = 1; |
1239 | |
1240 | goto wait; |
1241 | } |
1242 | |
1243 | /* don't retry if called from dig */ |
1244 | if (!statp->pfcode) |
1245 | return close_and_return_error (statp, resplen2); |
1246 | __res_iclose(statp, false); |
1247 | } |
1248 | if (anhp->rcode == NOERROR && anhp->ancount == 0 |
1249 | && anhp->aa == 0 && anhp->ra == 0 && anhp->arcount == 0) { |
1250 | goto next_ns; |
1251 | } |
1252 | if (!(statp->options & RES_IGNTC) && anhp->tc) { |
1253 | /* |
1254 | * To get the rest of answer, |
1255 | * use TCP with same server. |
1256 | */ |
1257 | *v_circuit = 1; |
1258 | __res_iclose(statp, false); |
1259 | // XXX if we have received one reply we could |
1260 | // XXX use it and not repeat it over TCP... |
1261 | if (resplen2 != NULL) |
1262 | *resplen2 = 0; |
1263 | return (1); |
1264 | } |
1265 | /* Mark which reply we received. */ |
1266 | if (matching_query == 1) |
1267 | recvresp1 = 1; |
1268 | else |
1269 | recvresp2 = 1; |
1270 | /* Repeat waiting if we have a second answer to arrive. */ |
1271 | if ((recvresp1 & recvresp2) == 0) { |
1272 | if (single_request) { |
1273 | pfd[0].events = POLLOUT; |
1274 | if (single_request_reopen) { |
1275 | __res_iclose (statp, false); |
1276 | retval = reopen (statp, terrno, ns); |
1277 | if (retval <= 0) |
1278 | { |
1279 | if (resplen2 != NULL) |
1280 | *resplen2 = 0; |
1281 | return retval; |
1282 | } |
1283 | pfd[0].fd = EXT(statp).nssocks[ns]; |
1284 | } |
1285 | } |
1286 | goto wait; |
1287 | } |
1288 | /* All is well. We have received both responses (if |
1289 | two responses were requested). */ |
1290 | return (resplen); |
1291 | } else if (pfd[0].revents & (POLLERR | POLLHUP | POLLNVAL)) |
1292 | /* Something went wrong. We can stop trying. */ |
1293 | return close_and_return_error (statp, resplen2); |
1294 | else { |
1295 | /* poll should not have returned > 0 in this case. */ |
1296 | abort (); |
1297 | } |
1298 | } |
1299 | |
1300 | static int |
1301 | sock_eq(struct sockaddr_in6 *a1, struct sockaddr_in6 *a2) { |
1302 | if (a1->sin6_family == a2->sin6_family) { |
1303 | if (a1->sin6_family == AF_INET) |
1304 | return ((((struct sockaddr_in *)a1)->sin_port == |
1305 | ((struct sockaddr_in *)a2)->sin_port) && |
1306 | (((struct sockaddr_in *)a1)->sin_addr.s_addr == |
1307 | ((struct sockaddr_in *)a2)->sin_addr.s_addr)); |
1308 | else |
1309 | return ((a1->sin6_port == a2->sin6_port) && |
1310 | !memcmp(&a1->sin6_addr, &a2->sin6_addr, |
1311 | sizeof (struct in6_addr))); |
1312 | } |
1313 | if (a1->sin6_family == AF_INET) { |
1314 | struct sockaddr_in6 *sap = a1; |
1315 | a1 = a2; |
1316 | a2 = sap; |
1317 | } /* assumes that AF_INET and AF_INET6 are the only possibilities */ |
1318 | return ((a1->sin6_port == ((struct sockaddr_in *)a2)->sin_port) && |
1319 | IN6_IS_ADDR_V4MAPPED(&a1->sin6_addr) && |
1320 | (a1->sin6_addr.s6_addr32[3] == |
1321 | ((struct sockaddr_in *)a2)->sin_addr.s_addr)); |
1322 | } |
1323 | |