| 1 | /* Copyright (C) 2016-2022 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 | ((HEADER *) 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(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, *ansp); |
| 398 | else |
| 399 | mask_ad_bit (ctx, ans); |
| 400 | } |
| 401 | DIAG_POP_NEEDS_COMMENT; |
| 402 | if (resplen2 != NULL && *resplen2 > HFIXEDSZ) |
| 403 | mask_ad_bit (ctx, *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 | int result = __res_context_send (ctx, buf, buflen, NULL, 0, ans, anssiz, |
| 442 | NULL, NULL, NULL, NULL, NULL); |
| 443 | __resolv_context_put (ctx); |
| 444 | return result; |
| 445 | } |
| 446 | |
| 447 | int |
| 448 | ___res_nsend (res_state statp, const unsigned char *buf, int buflen, |
| 449 | unsigned char *ans, int anssiz) |
| 450 | { |
| 451 | return context_send_common |
| 452 | (__resolv_context_get_override (statp), buf, buflen, ans, anssiz); |
| 453 | } |
| 454 | versioned_symbol (libc, ___res_nsend, res_nsend, GLIBC_2_34); |
| 455 | #if OTHER_SHLIB_COMPAT (libresolv, GLIBC_2_2, GLIBC_2_34) |
| 456 | compat_symbol (libresolv, ___res_nsend, __res_nsend, GLIBC_2_2); |
| 457 | #endif |
| 458 | |
| 459 | int |
| 460 | ___res_send (const unsigned char *buf, int buflen, unsigned char *ans, |
| 461 | int anssiz) |
| 462 | { |
| 463 | return context_send_common |
| 464 | (__resolv_context_get (), buf, buflen, ans, anssiz); |
| 465 | } |
| 466 | versioned_symbol (libc, ___res_send, res_send, GLIBC_2_34); |
| 467 | #if OTHER_SHLIB_COMPAT (libresolv, GLIBC_2_0, GLIBC_2_34) |
| 468 | compat_symbol (libresolv, ___res_send, __res_send, GLIBC_2_0); |
| 469 | #endif |
| 470 | |
| 471 | /* Private */ |
| 472 | |
| 473 | /* Close the resolver structure, assign zero to *RESPLEN2 if RESPLEN2 |
| 474 | is not NULL, and return zero. */ |
| 475 | static int |
| 476 | __attribute__ ((warn_unused_result)) |
| 477 | close_and_return_error (res_state statp, int *resplen2) |
| 478 | { |
| 479 | __res_iclose(statp, false); |
| 480 | if (resplen2 != NULL) |
| 481 | *resplen2 = 0; |
| 482 | return 0; |
| 483 | } |
| 484 | |
| 485 | /* The send_vc function is responsible for sending a DNS query over TCP |
| 486 | to the nameserver numbered NS from the res_state STATP i.e. |
| 487 | EXT(statp).nssocks[ns]. The function supports sending both IPv4 and |
| 488 | IPv6 queries at the same serially on the same socket. |
| 489 | |
| 490 | Please note that for TCP there is no way to disable sending both |
| 491 | queries, unlike UDP, which honours RES_SNGLKUP and RES_SNGLKUPREOP |
| 492 | and sends the queries serially and waits for the result after each |
| 493 | sent query. This implementation should be corrected to honour these |
| 494 | options. |
| 495 | |
| 496 | Please also note that for TCP we send both queries over the same |
| 497 | socket one after another. This technically violates best practice |
| 498 | since the server is allowed to read the first query, respond, and |
| 499 | then close the socket (to service another client). If the server |
| 500 | does this, then the remaining second query in the socket data buffer |
| 501 | will cause the server to send the client an RST which will arrive |
| 502 | asynchronously and the client's OS will likely tear down the socket |
| 503 | receive buffer resulting in a potentially short read and lost |
| 504 | response data. This will force the client to retry the query again, |
| 505 | and this process may repeat until all servers and connection resets |
| 506 | are exhausted and then the query will fail. It's not known if this |
| 507 | happens with any frequency in real DNS server implementations. This |
| 508 | implementation should be corrected to use two sockets by default for |
| 509 | parallel queries. |
| 510 | |
| 511 | The query stored in BUF of BUFLEN length is sent first followed by |
| 512 | the query stored in BUF2 of BUFLEN2 length. Queries are sent |
| 513 | serially on the same socket. |
| 514 | |
| 515 | Answers to the query are stored firstly in *ANSP up to a max of |
| 516 | *ANSSIZP bytes. If more than *ANSSIZP bytes are needed and ANSCP |
| 517 | is non-NULL (to indicate that modifying the answer buffer is allowed) |
| 518 | then malloc is used to allocate a new response buffer and ANSCP and |
| 519 | ANSP will both point to the new buffer. If more than *ANSSIZP bytes |
| 520 | are needed but ANSCP is NULL, then as much of the response as |
| 521 | possible is read into the buffer, but the results will be truncated. |
| 522 | When truncation happens because of a small answer buffer the DNS |
| 523 | packets header field TC will bet set to 1, indicating a truncated |
| 524 | message and the rest of the socket data will be read and discarded. |
| 525 | |
| 526 | Answers to the query are stored secondly in *ANSP2 up to a max of |
| 527 | *ANSSIZP2 bytes, with the actual response length stored in |
| 528 | *RESPLEN2. If more than *ANSSIZP bytes are needed and ANSP2 |
| 529 | is non-NULL (required for a second query) then malloc is used to |
| 530 | allocate a new response buffer, *ANSSIZP2 is set to the new buffer |
| 531 | size and *ANSP2_MALLOCED is set to 1. |
| 532 | |
| 533 | The ANSP2_MALLOCED argument will eventually be removed as the |
| 534 | change in buffer pointer can be used to detect the buffer has |
| 535 | changed and that the caller should use free on the new buffer. |
| 536 | |
| 537 | Note that the answers may arrive in any order from the server and |
| 538 | therefore the first and second answer buffers may not correspond to |
| 539 | the first and second queries. |
| 540 | |
| 541 | It is not supported to call this function with a non-NULL ANSP2 |
| 542 | but a NULL ANSCP. Put another way, you can call send_vc with a |
| 543 | single unmodifiable buffer or two modifiable buffers, but no other |
| 544 | combination is supported. |
| 545 | |
| 546 | It is the caller's responsibility to free the malloc allocated |
| 547 | buffers by detecting that the pointers have changed from their |
| 548 | original values i.e. *ANSCP or *ANSP2 has changed. |
| 549 | |
| 550 | If errors are encountered then *TERRNO is set to an appropriate |
| 551 | errno value and a zero result is returned for a recoverable error, |
| 552 | and a less-than zero result is returned for a non-recoverable error. |
| 553 | |
| 554 | If no errors are encountered then *TERRNO is left unmodified and |
| 555 | a the length of the first response in bytes is returned. */ |
| 556 | static int |
| 557 | send_vc(res_state statp, |
| 558 | const u_char *buf, int buflen, const u_char *buf2, int buflen2, |
| 559 | u_char **ansp, int *anssizp, |
| 560 | int *terrno, int ns, u_char **anscp, u_char **ansp2, int *anssizp2, |
| 561 | int *resplen2, int *ansp2_malloced) |
| 562 | { |
| 563 | const HEADER *hp = (HEADER *) buf; |
| 564 | const HEADER *hp2 = (HEADER *) buf2; |
| 565 | HEADER *anhp = (HEADER *) *ansp; |
| 566 | struct sockaddr *nsap = __res_get_nsaddr (statp, ns); |
| 567 | int truncating, connreset, n; |
| 568 | /* On some architectures compiler might emit a warning indicating |
| 569 | 'resplen' may be used uninitialized. However if buf2 == NULL |
| 570 | then this code won't be executed; if buf2 != NULL, then first |
| 571 | time round the loop recvresp1 and recvresp2 will be 0 so this |
| 572 | code won't be executed but "thisresplenp = &resplen;" followed |
| 573 | by "*thisresplenp = rlen;" will be executed so that subsequent |
| 574 | times round the loop resplen has been initialized. So this is |
| 575 | a false-positive. |
| 576 | */ |
| 577 | DIAG_PUSH_NEEDS_COMMENT; |
| 578 | DIAG_IGNORE_NEEDS_COMMENT (5, "-Wmaybe-uninitialized" ); |
| 579 | int resplen; |
| 580 | DIAG_POP_NEEDS_COMMENT; |
| 581 | struct iovec iov[4]; |
| 582 | u_short len; |
| 583 | u_short len2; |
| 584 | u_char *cp; |
| 585 | |
| 586 | connreset = 0; |
| 587 | same_ns: |
| 588 | truncating = 0; |
| 589 | |
| 590 | /* Are we still talking to whom we want to talk to? */ |
| 591 | if (statp->_vcsock >= 0 && (statp->_flags & RES_F_VC) != 0) { |
| 592 | struct sockaddr_in6 peer; |
| 593 | socklen_t size = sizeof peer; |
| 594 | |
| 595 | if (__getpeername (statp->_vcsock, |
| 596 | (struct sockaddr *) &peer, &size) < 0 |
| 597 | || !sock_eq (&peer, (struct sockaddr_in6 *) nsap)) { |
| 598 | __res_iclose(statp, false); |
| 599 | statp->_flags &= ~RES_F_VC; |
| 600 | } |
| 601 | } |
| 602 | |
| 603 | if (statp->_vcsock < 0 || (statp->_flags & RES_F_VC) == 0) { |
| 604 | if (statp->_vcsock >= 0) |
| 605 | __res_iclose(statp, false); |
| 606 | |
| 607 | statp->_vcsock = __socket |
| 608 | (nsap->sa_family, SOCK_STREAM | SOCK_CLOEXEC, 0); |
| 609 | if (statp->_vcsock < 0) { |
| 610 | *terrno = errno; |
| 611 | if (resplen2 != NULL) |
| 612 | *resplen2 = 0; |
| 613 | return (-1); |
| 614 | } |
| 615 | __set_errno (0); |
| 616 | if (__connect (statp->_vcsock, nsap, |
| 617 | nsap->sa_family == AF_INET |
| 618 | ? sizeof (struct sockaddr_in) |
| 619 | : sizeof (struct sockaddr_in6)) < 0) { |
| 620 | *terrno = errno; |
| 621 | return close_and_return_error (statp, resplen2); |
| 622 | } |
| 623 | statp->_flags |= RES_F_VC; |
| 624 | } |
| 625 | |
| 626 | /* |
| 627 | * Send length & message |
| 628 | */ |
| 629 | len = htons ((u_short) buflen); |
| 630 | evConsIovec(&len, INT16SZ, &iov[0]); |
| 631 | evConsIovec((void*)buf, buflen, &iov[1]); |
| 632 | int niov = 2; |
| 633 | ssize_t explen = INT16SZ + buflen; |
| 634 | if (buf2 != NULL) { |
| 635 | len2 = htons ((u_short) buflen2); |
| 636 | evConsIovec(&len2, INT16SZ, &iov[2]); |
| 637 | evConsIovec((void*)buf2, buflen2, &iov[3]); |
| 638 | niov = 4; |
| 639 | explen += INT16SZ + buflen2; |
| 640 | } |
| 641 | if (TEMP_FAILURE_RETRY (__writev (statp->_vcsock, iov, niov)) |
| 642 | != explen) { |
| 643 | *terrno = errno; |
| 644 | return close_and_return_error (statp, resplen2); |
| 645 | } |
| 646 | /* |
| 647 | * Receive length & response |
| 648 | */ |
| 649 | int recvresp1 = 0; |
| 650 | /* Skip the second response if there is no second query. |
| 651 | To do that we mark the second response as received. */ |
| 652 | int recvresp2 = buf2 == NULL; |
| 653 | uint16_t rlen16; |
| 654 | read_len: |
| 655 | cp = (u_char *)&rlen16; |
| 656 | len = sizeof(rlen16); |
| 657 | while ((n = TEMP_FAILURE_RETRY (read(statp->_vcsock, cp, |
| 658 | (int)len))) > 0) { |
| 659 | cp += n; |
| 660 | if ((len -= n) <= 0) |
| 661 | break; |
| 662 | } |
| 663 | if (n <= 0) { |
| 664 | *terrno = errno; |
| 665 | /* |
| 666 | * A long running process might get its TCP |
| 667 | * connection reset if the remote server was |
| 668 | * restarted. Requery the server instead of |
| 669 | * trying a new one. When there is only one |
| 670 | * server, this means that a query might work |
| 671 | * instead of failing. We only allow one reset |
| 672 | * per query to prevent looping. |
| 673 | */ |
| 674 | if (*terrno == ECONNRESET && !connreset) |
| 675 | { |
| 676 | __res_iclose (statp, false); |
| 677 | connreset = 1; |
| 678 | goto same_ns; |
| 679 | } |
| 680 | return close_and_return_error (statp, resplen2); |
| 681 | } |
| 682 | int rlen = ntohs (rlen16); |
| 683 | |
| 684 | int *thisanssizp; |
| 685 | u_char **thisansp; |
| 686 | int *thisresplenp; |
| 687 | if ((recvresp1 | recvresp2) == 0 || buf2 == NULL) { |
| 688 | /* We have not received any responses |
| 689 | yet or we only have one response to |
| 690 | receive. */ |
| 691 | thisanssizp = anssizp; |
| 692 | thisansp = anscp ?: ansp; |
| 693 | assert (anscp != NULL || ansp2 == NULL); |
| 694 | thisresplenp = &resplen; |
| 695 | } else { |
| 696 | thisanssizp = anssizp2; |
| 697 | thisansp = ansp2; |
| 698 | thisresplenp = resplen2; |
| 699 | } |
| 700 | anhp = (HEADER *) *thisansp; |
| 701 | |
| 702 | *thisresplenp = rlen; |
| 703 | /* Is the answer buffer too small? */ |
| 704 | if (*thisanssizp < rlen) { |
| 705 | /* If the current buffer is not the the static |
| 706 | user-supplied buffer then we can reallocate |
| 707 | it. */ |
| 708 | if (thisansp != NULL && thisansp != ansp) { |
| 709 | /* Always allocate MAXPACKET, callers expect |
| 710 | this specific size. */ |
| 711 | u_char *newp = malloc (MAXPACKET); |
| 712 | if (newp == NULL) |
| 713 | { |
| 714 | *terrno = ENOMEM; |
| 715 | return close_and_return_error (statp, resplen2); |
| 716 | } |
| 717 | *thisanssizp = MAXPACKET; |
| 718 | *thisansp = newp; |
| 719 | if (thisansp == ansp2) |
| 720 | *ansp2_malloced = 1; |
| 721 | anhp = (HEADER *) newp; |
| 722 | /* A uint16_t can't be larger than MAXPACKET |
| 723 | thus it's safe to allocate MAXPACKET but |
| 724 | read RLEN bytes instead. */ |
| 725 | len = rlen; |
| 726 | } else { |
| 727 | truncating = 1; |
| 728 | len = *thisanssizp; |
| 729 | } |
| 730 | } else |
| 731 | len = rlen; |
| 732 | |
| 733 | if (__glibc_unlikely (len < HFIXEDSZ)) { |
| 734 | /* |
| 735 | * Undersized message. |
| 736 | */ |
| 737 | *terrno = EMSGSIZE; |
| 738 | return close_and_return_error (statp, resplen2); |
| 739 | } |
| 740 | |
| 741 | cp = *thisansp; |
| 742 | while (len != 0 && (n = read(statp->_vcsock, (char *)cp, (int)len)) > 0){ |
| 743 | cp += n; |
| 744 | len -= n; |
| 745 | } |
| 746 | if (__glibc_unlikely (n <= 0)) { |
| 747 | *terrno = errno; |
| 748 | return close_and_return_error (statp, resplen2); |
| 749 | } |
| 750 | if (__glibc_unlikely (truncating)) { |
| 751 | /* |
| 752 | * Flush rest of answer so connection stays in synch. |
| 753 | */ |
| 754 | anhp->tc = 1; |
| 755 | len = rlen - *thisanssizp; |
| 756 | while (len != 0) { |
| 757 | char junk[PACKETSZ]; |
| 758 | |
| 759 | n = read(statp->_vcsock, junk, |
| 760 | (len > sizeof junk) ? sizeof junk : len); |
| 761 | if (n > 0) |
| 762 | len -= n; |
| 763 | else |
| 764 | break; |
| 765 | } |
| 766 | } |
| 767 | /* |
| 768 | * If the calling application has bailed out of |
| 769 | * a previous call and failed to arrange to have |
| 770 | * the circuit closed or the server has got |
| 771 | * itself confused, then drop the packet and |
| 772 | * wait for the correct one. |
| 773 | */ |
| 774 | if ((recvresp1 || hp->id != anhp->id) |
| 775 | && (recvresp2 || hp2->id != anhp->id)) |
| 776 | goto read_len; |
| 777 | |
| 778 | /* Mark which reply we received. */ |
| 779 | if (recvresp1 == 0 && hp->id == anhp->id) |
| 780 | recvresp1 = 1; |
| 781 | else |
| 782 | recvresp2 = 1; |
| 783 | /* Repeat waiting if we have a second answer to arrive. */ |
| 784 | if ((recvresp1 & recvresp2) == 0) |
| 785 | goto read_len; |
| 786 | |
| 787 | /* |
| 788 | * All is well, or the error is fatal. Signal that the |
| 789 | * next nameserver ought not be tried. |
| 790 | */ |
| 791 | return resplen; |
| 792 | } |
| 793 | |
| 794 | static int |
| 795 | reopen (res_state statp, int *terrno, int ns) |
| 796 | { |
| 797 | if (EXT(statp).nssocks[ns] == -1) { |
| 798 | struct sockaddr *nsap = __res_get_nsaddr (statp, ns); |
| 799 | socklen_t slen; |
| 800 | |
| 801 | /* only try IPv6 if IPv6 NS and if not failed before */ |
| 802 | if (nsap->sa_family == AF_INET6 && !statp->ipv6_unavail) { |
| 803 | EXT (statp).nssocks[ns] = __socket |
| 804 | (PF_INET6, |
| 805 | SOCK_DGRAM | SOCK_NONBLOCK | SOCK_CLOEXEC, 0); |
| 806 | if (EXT(statp).nssocks[ns] < 0) |
| 807 | statp->ipv6_unavail = errno == EAFNOSUPPORT; |
| 808 | slen = sizeof (struct sockaddr_in6); |
| 809 | } else if (nsap->sa_family == AF_INET) { |
| 810 | EXT (statp).nssocks[ns] = __socket |
| 811 | (PF_INET, |
| 812 | SOCK_DGRAM | SOCK_NONBLOCK | SOCK_CLOEXEC, 0); |
| 813 | slen = sizeof (struct sockaddr_in); |
| 814 | } |
| 815 | if (EXT(statp).nssocks[ns] < 0) { |
| 816 | *terrno = errno; |
| 817 | return (-1); |
| 818 | } |
| 819 | |
| 820 | /* Enable full ICMP error reporting for this |
| 821 | socket. */ |
| 822 | if (__res_enable_icmp (nsap->sa_family, |
| 823 | EXT (statp).nssocks[ns]) < 0) |
| 824 | { |
| 825 | int saved_errno = errno; |
| 826 | __res_iclose (statp, false); |
| 827 | __set_errno (saved_errno); |
| 828 | *terrno = saved_errno; |
| 829 | return -1; |
| 830 | } |
| 831 | |
| 832 | /* |
| 833 | * On a 4.3BSD+ machine (client and server, |
| 834 | * actually), sending to a nameserver datagram |
| 835 | * port with no nameserver will cause an |
| 836 | * ICMP port unreachable message to be returned. |
| 837 | * If our datagram socket is "connected" to the |
| 838 | * server, we get an ECONNREFUSED error on the next |
| 839 | * socket operation, and select returns if the |
| 840 | * error message is received. We can thus detect |
| 841 | * the absence of a nameserver without timing out. |
| 842 | */ |
| 843 | /* With GCC 5.3 when compiling with -Os the compiler |
| 844 | emits a warning that slen may be used uninitialized, |
| 845 | but that is never true. Both slen and |
| 846 | EXT(statp).nssocks[ns] are initialized together or |
| 847 | the function return -1 before control flow reaches |
| 848 | the call to connect with slen. */ |
| 849 | DIAG_PUSH_NEEDS_COMMENT; |
| 850 | DIAG_IGNORE_Os_NEEDS_COMMENT (5, "-Wmaybe-uninitialized" ); |
| 851 | if (__connect (EXT (statp).nssocks[ns], nsap, slen) < 0) { |
| 852 | DIAG_POP_NEEDS_COMMENT; |
| 853 | __res_iclose(statp, false); |
| 854 | return (0); |
| 855 | } |
| 856 | } |
| 857 | |
| 858 | return 1; |
| 859 | } |
| 860 | |
| 861 | /* The send_dg function is responsible for sending a DNS query over UDP |
| 862 | to the nameserver numbered NS from the res_state STATP i.e. |
| 863 | EXT(statp).nssocks[ns]. The function supports IPv4 and IPv6 queries |
| 864 | along with the ability to send the query in parallel for both stacks |
| 865 | (default) or serially (RES_SINGLKUP). It also supports serial lookup |
| 866 | with a close and reopen of the socket used to talk to the server |
| 867 | (RES_SNGLKUPREOP) to work around broken name servers. |
| 868 | |
| 869 | The query stored in BUF of BUFLEN length is sent first followed by |
| 870 | the query stored in BUF2 of BUFLEN2 length. Queries are sent |
| 871 | in parallel (default) or serially (RES_SINGLKUP or RES_SNGLKUPREOP). |
| 872 | |
| 873 | Answers to the query are stored firstly in *ANSP up to a max of |
| 874 | *ANSSIZP bytes. If more than *ANSSIZP bytes are needed and ANSCP |
| 875 | is non-NULL (to indicate that modifying the answer buffer is allowed) |
| 876 | then malloc is used to allocate a new response buffer and ANSCP and |
| 877 | ANSP will both point to the new buffer. If more than *ANSSIZP bytes |
| 878 | are needed but ANSCP is NULL, then as much of the response as |
| 879 | possible is read into the buffer, but the results will be truncated. |
| 880 | When truncation happens because of a small answer buffer the DNS |
| 881 | packets header field TC will bet set to 1, indicating a truncated |
| 882 | message, while the rest of the UDP packet is discarded. |
| 883 | |
| 884 | Answers to the query are stored secondly in *ANSP2 up to a max of |
| 885 | *ANSSIZP2 bytes, with the actual response length stored in |
| 886 | *RESPLEN2. If more than *ANSSIZP bytes are needed and ANSP2 |
| 887 | is non-NULL (required for a second query) then malloc is used to |
| 888 | allocate a new response buffer, *ANSSIZP2 is set to the new buffer |
| 889 | size and *ANSP2_MALLOCED is set to 1. |
| 890 | |
| 891 | The ANSP2_MALLOCED argument will eventually be removed as the |
| 892 | change in buffer pointer can be used to detect the buffer has |
| 893 | changed and that the caller should use free on the new buffer. |
| 894 | |
| 895 | Note that the answers may arrive in any order from the server and |
| 896 | therefore the first and second answer buffers may not correspond to |
| 897 | the first and second queries. |
| 898 | |
| 899 | It is not supported to call this function with a non-NULL ANSP2 |
| 900 | but a NULL ANSCP. Put another way, you can call send_vc with a |
| 901 | single unmodifiable buffer or two modifiable buffers, but no other |
| 902 | combination is supported. |
| 903 | |
| 904 | It is the caller's responsibility to free the malloc allocated |
| 905 | buffers by detecting that the pointers have changed from their |
| 906 | original values i.e. *ANSCP or *ANSP2 has changed. |
| 907 | |
| 908 | If an answer is truncated because of UDP datagram DNS limits then |
| 909 | *V_CIRCUIT is set to 1 and the return value non-zero to indicate to |
| 910 | the caller to retry with TCP. The value *GOTSOMEWHERE is set to 1 |
| 911 | if any progress was made reading a response from the nameserver and |
| 912 | is used by the caller to distinguish between ECONNREFUSED and |
| 913 | ETIMEDOUT (the latter if *GOTSOMEWHERE is 1). |
| 914 | |
| 915 | If errors are encountered then *TERRNO is set to an appropriate |
| 916 | errno value and a zero result is returned for a recoverable error, |
| 917 | and a less-than zero result is returned for a non-recoverable error. |
| 918 | |
| 919 | If no errors are encountered then *TERRNO is left unmodified and |
| 920 | a the length of the first response in bytes is returned. */ |
| 921 | static int |
| 922 | send_dg(res_state statp, |
| 923 | const u_char *buf, int buflen, const u_char *buf2, int buflen2, |
| 924 | u_char **ansp, int *anssizp, |
| 925 | int *terrno, int ns, int *v_circuit, int *gotsomewhere, u_char **anscp, |
| 926 | u_char **ansp2, int *anssizp2, int *resplen2, int *ansp2_malloced) |
| 927 | { |
| 928 | const HEADER *hp = (HEADER *) buf; |
| 929 | const HEADER *hp2 = (HEADER *) buf2; |
| 930 | struct timespec now, timeout, finish; |
| 931 | struct pollfd pfd[1]; |
| 932 | int ptimeout; |
| 933 | struct sockaddr_in6 from; |
| 934 | int resplen = 0; |
| 935 | int n; |
| 936 | |
| 937 | /* |
| 938 | * Compute time for the total operation. |
| 939 | */ |
| 940 | int seconds = (statp->retrans << ns); |
| 941 | if (ns > 0) |
| 942 | seconds /= statp->nscount; |
| 943 | if (seconds <= 0) |
| 944 | seconds = 1; |
| 945 | bool single_request_reopen = (statp->options & RES_SNGLKUPREOP) != 0; |
| 946 | bool single_request = (((statp->options & RES_SNGLKUP) != 0) |
| 947 | | single_request_reopen); |
| 948 | int save_gotsomewhere = *gotsomewhere; |
| 949 | |
| 950 | int retval; |
| 951 | retry_reopen: |
| 952 | retval = reopen (statp, terrno, ns); |
| 953 | if (retval <= 0) |
| 954 | { |
| 955 | if (resplen2 != NULL) |
| 956 | *resplen2 = 0; |
| 957 | return retval; |
| 958 | } |
| 959 | retry: |
| 960 | evNowTime(&now); |
| 961 | evConsTime(&timeout, seconds, 0); |
| 962 | evAddTime(&finish, &now, &timeout); |
| 963 | int need_recompute = 0; |
| 964 | int nwritten = 0; |
| 965 | int recvresp1 = 0; |
| 966 | /* Skip the second response if there is no second query. |
| 967 | To do that we mark the second response as received. */ |
| 968 | int recvresp2 = buf2 == NULL; |
| 969 | pfd[0].fd = EXT(statp).nssocks[ns]; |
| 970 | pfd[0].events = POLLOUT; |
| 971 | wait: |
| 972 | if (need_recompute) { |
| 973 | recompute_resend: |
| 974 | evNowTime(&now); |
| 975 | if (evCmpTime(finish, now) <= 0) { |
| 976 | poll_err_out: |
| 977 | return close_and_return_error (statp, resplen2); |
| 978 | } |
| 979 | evSubTime(&timeout, &finish, &now); |
| 980 | need_recompute = 0; |
| 981 | } |
| 982 | /* Convert struct timespec in milliseconds. */ |
| 983 | ptimeout = timeout.tv_sec * 1000 + timeout.tv_nsec / 1000000; |
| 984 | |
| 985 | n = 0; |
| 986 | if (nwritten == 0) |
| 987 | n = __poll (pfd, 1, 0); |
| 988 | if (__glibc_unlikely (n == 0)) { |
| 989 | n = __poll (pfd, 1, ptimeout); |
| 990 | need_recompute = 1; |
| 991 | } |
| 992 | if (n == 0) { |
| 993 | if (resplen > 1 && (recvresp1 || (buf2 != NULL && recvresp2))) |
| 994 | { |
| 995 | /* There are quite a few broken name servers out |
| 996 | there which don't handle two outstanding |
| 997 | requests from the same source. There are also |
| 998 | broken firewall settings. If we time out after |
| 999 | having received one answer switch to the mode |
| 1000 | where we send the second request only once we |
| 1001 | have received the first answer. */ |
| 1002 | if (!single_request) |
| 1003 | { |
| 1004 | statp->options |= RES_SNGLKUP; |
| 1005 | single_request = true; |
| 1006 | *gotsomewhere = save_gotsomewhere; |
| 1007 | goto retry; |
| 1008 | } |
| 1009 | else if (!single_request_reopen) |
| 1010 | { |
| 1011 | statp->options |= RES_SNGLKUPREOP; |
| 1012 | single_request_reopen = true; |
| 1013 | *gotsomewhere = save_gotsomewhere; |
| 1014 | __res_iclose (statp, false); |
| 1015 | goto retry_reopen; |
| 1016 | } |
| 1017 | |
| 1018 | *resplen2 = 1; |
| 1019 | return resplen; |
| 1020 | } |
| 1021 | |
| 1022 | *gotsomewhere = 1; |
| 1023 | if (resplen2 != NULL) |
| 1024 | *resplen2 = 0; |
| 1025 | return 0; |
| 1026 | } |
| 1027 | if (n < 0) { |
| 1028 | if (errno == EINTR) |
| 1029 | goto recompute_resend; |
| 1030 | |
| 1031 | goto poll_err_out; |
| 1032 | } |
| 1033 | __set_errno (0); |
| 1034 | if (pfd[0].revents & POLLOUT) { |
| 1035 | #ifndef __ASSUME_SENDMMSG |
| 1036 | static int have_sendmmsg; |
| 1037 | #else |
| 1038 | # define have_sendmmsg 1 |
| 1039 | #endif |
| 1040 | if (have_sendmmsg >= 0 && nwritten == 0 && buf2 != NULL |
| 1041 | && !single_request) |
| 1042 | { |
| 1043 | struct iovec iov = |
| 1044 | { .iov_base = (void *) buf, .iov_len = buflen }; |
| 1045 | struct iovec iov2 = |
| 1046 | { .iov_base = (void *) buf2, .iov_len = buflen2 }; |
| 1047 | struct mmsghdr reqs[2] = |
| 1048 | { |
| 1049 | { |
| 1050 | .msg_hdr = |
| 1051 | { |
| 1052 | .msg_iov = &iov, |
| 1053 | .msg_iovlen = 1, |
| 1054 | }, |
| 1055 | }, |
| 1056 | { |
| 1057 | .msg_hdr = |
| 1058 | { |
| 1059 | .msg_iov = &iov2, |
| 1060 | .msg_iovlen = 1, |
| 1061 | } |
| 1062 | }, |
| 1063 | }; |
| 1064 | |
| 1065 | int ndg = __sendmmsg (pfd[0].fd, reqs, 2, MSG_NOSIGNAL); |
| 1066 | if (__glibc_likely (ndg == 2)) |
| 1067 | { |
| 1068 | if (reqs[0].msg_len != buflen |
| 1069 | || reqs[1].msg_len != buflen2) |
| 1070 | goto fail_sendmmsg; |
| 1071 | |
| 1072 | pfd[0].events = POLLIN; |
| 1073 | nwritten += 2; |
| 1074 | } |
| 1075 | else if (ndg == 1 && reqs[0].msg_len == buflen) |
| 1076 | goto just_one; |
| 1077 | else if (ndg < 0 && (errno == EINTR || errno == EAGAIN)) |
| 1078 | goto recompute_resend; |
| 1079 | else |
| 1080 | { |
| 1081 | #ifndef __ASSUME_SENDMMSG |
| 1082 | if (__glibc_unlikely (have_sendmmsg == 0)) |
| 1083 | { |
| 1084 | if (ndg < 0 && errno == ENOSYS) |
| 1085 | { |
| 1086 | have_sendmmsg = -1; |
| 1087 | goto try_send; |
| 1088 | } |
| 1089 | have_sendmmsg = 1; |
| 1090 | } |
| 1091 | #endif |
| 1092 | |
| 1093 | fail_sendmmsg: |
| 1094 | return close_and_return_error (statp, resplen2); |
| 1095 | } |
| 1096 | } |
| 1097 | else |
| 1098 | { |
| 1099 | ssize_t sr; |
| 1100 | #ifndef __ASSUME_SENDMMSG |
| 1101 | try_send: |
| 1102 | #endif |
| 1103 | if (nwritten != 0) |
| 1104 | sr = __send (pfd[0].fd, buf2, buflen2, MSG_NOSIGNAL); |
| 1105 | else |
| 1106 | sr = __send (pfd[0].fd, buf, buflen, MSG_NOSIGNAL); |
| 1107 | |
| 1108 | if (sr != (nwritten != 0 ? buflen2 : buflen)) { |
| 1109 | if (errno == EINTR || errno == EAGAIN) |
| 1110 | goto recompute_resend; |
| 1111 | return close_and_return_error (statp, resplen2); |
| 1112 | } |
| 1113 | just_one: |
| 1114 | if (nwritten != 0 || buf2 == NULL || single_request) |
| 1115 | pfd[0].events = POLLIN; |
| 1116 | else |
| 1117 | pfd[0].events = POLLIN | POLLOUT; |
| 1118 | ++nwritten; |
| 1119 | } |
| 1120 | goto wait; |
| 1121 | } else if (pfd[0].revents & POLLIN) { |
| 1122 | int *thisanssizp; |
| 1123 | u_char **thisansp; |
| 1124 | int *thisresplenp; |
| 1125 | |
| 1126 | if ((recvresp1 | recvresp2) == 0 || buf2 == NULL) { |
| 1127 | /* We have not received any responses |
| 1128 | yet or we only have one response to |
| 1129 | receive. */ |
| 1130 | thisanssizp = anssizp; |
| 1131 | thisansp = anscp ?: ansp; |
| 1132 | assert (anscp != NULL || ansp2 == NULL); |
| 1133 | thisresplenp = &resplen; |
| 1134 | } else { |
| 1135 | thisanssizp = anssizp2; |
| 1136 | thisansp = ansp2; |
| 1137 | thisresplenp = resplen2; |
| 1138 | } |
| 1139 | |
| 1140 | if (*thisanssizp < MAXPACKET |
| 1141 | /* If the current buffer is not the the static |
| 1142 | user-supplied buffer then we can reallocate |
| 1143 | it. */ |
| 1144 | && (thisansp != NULL && thisansp != ansp) |
| 1145 | #ifdef FIONREAD |
| 1146 | /* Is the size too small? */ |
| 1147 | && (__ioctl (pfd[0].fd, FIONREAD, thisresplenp) < 0 |
| 1148 | || *thisanssizp < *thisresplenp) |
| 1149 | #endif |
| 1150 | ) { |
| 1151 | /* Always allocate MAXPACKET, callers expect |
| 1152 | this specific size. */ |
| 1153 | u_char *newp = malloc (MAXPACKET); |
| 1154 | if (newp != NULL) { |
| 1155 | *thisanssizp = MAXPACKET; |
| 1156 | *thisansp = newp; |
| 1157 | if (thisansp == ansp2) |
| 1158 | *ansp2_malloced = 1; |
| 1159 | } |
| 1160 | } |
| 1161 | /* We could end up with truncation if anscp was NULL |
| 1162 | (not allowed to change caller's buffer) and the |
| 1163 | response buffer size is too small. This isn't a |
| 1164 | reliable way to detect truncation because the ioctl |
| 1165 | may be an inaccurate report of the UDP message size. |
| 1166 | Therefore we use this only to issue debug output. |
| 1167 | To do truncation accurately with UDP we need |
| 1168 | MSG_TRUNC which is only available on Linux. We |
| 1169 | can abstract out the Linux-specific feature in the |
| 1170 | future to detect truncation. */ |
| 1171 | HEADER *anhp = (HEADER *) *thisansp; |
| 1172 | socklen_t fromlen = sizeof(struct sockaddr_in6); |
| 1173 | assert (sizeof(from) <= fromlen); |
| 1174 | *thisresplenp = __recvfrom (pfd[0].fd, (char *) *thisansp, |
| 1175 | *thisanssizp, 0, |
| 1176 | (struct sockaddr *) &from, |
| 1177 | &fromlen); |
| 1178 | if (__glibc_unlikely (*thisresplenp <= 0)) { |
| 1179 | if (errno == EINTR || errno == EAGAIN) { |
| 1180 | need_recompute = 1; |
| 1181 | goto wait; |
| 1182 | } |
| 1183 | return close_and_return_error (statp, resplen2); |
| 1184 | } |
| 1185 | *gotsomewhere = 1; |
| 1186 | if (__glibc_unlikely (*thisresplenp < HFIXEDSZ)) { |
| 1187 | /* |
| 1188 | * Undersized message. |
| 1189 | */ |
| 1190 | *terrno = EMSGSIZE; |
| 1191 | return close_and_return_error (statp, resplen2); |
| 1192 | } |
| 1193 | |
| 1194 | /* Check for the correct header layout and a matching |
| 1195 | question. */ |
| 1196 | int matching_query = 0; /* Default to no matching query. */ |
| 1197 | if (!recvresp1 |
| 1198 | && anhp->id == hp->id |
| 1199 | && __libc_res_queriesmatch (buf, buf + buflen, |
| 1200 | *thisansp, |
| 1201 | *thisansp + *thisanssizp)) |
| 1202 | matching_query = 1; |
| 1203 | if (!recvresp2 |
| 1204 | && anhp->id == hp2->id |
| 1205 | && __libc_res_queriesmatch (buf2, buf2 + buflen2, |
| 1206 | *thisansp, |
| 1207 | *thisansp + *thisanssizp)) |
| 1208 | matching_query = 2; |
| 1209 | if (matching_query == 0) |
| 1210 | /* Spurious UDP packet. Drop it and continue |
| 1211 | waiting. */ |
| 1212 | { |
| 1213 | need_recompute = 1; |
| 1214 | goto wait; |
| 1215 | } |
| 1216 | |
| 1217 | if (anhp->rcode == SERVFAIL || |
| 1218 | anhp->rcode == NOTIMP || |
| 1219 | anhp->rcode == REFUSED) { |
| 1220 | next_ns: |
| 1221 | if (recvresp1 || (buf2 != NULL && recvresp2)) { |
| 1222 | *resplen2 = 0; |
| 1223 | return resplen; |
| 1224 | } |
| 1225 | if (buf2 != NULL) |
| 1226 | { |
| 1227 | /* No data from the first reply. */ |
| 1228 | resplen = 0; |
| 1229 | /* We are waiting for a possible second reply. */ |
| 1230 | if (matching_query == 1) |
| 1231 | recvresp1 = 1; |
| 1232 | else |
| 1233 | recvresp2 = 1; |
| 1234 | |
| 1235 | goto wait; |
| 1236 | } |
| 1237 | |
| 1238 | /* don't retry if called from dig */ |
| 1239 | if (!statp->pfcode) |
| 1240 | return close_and_return_error (statp, resplen2); |
| 1241 | __res_iclose(statp, false); |
| 1242 | } |
| 1243 | if (anhp->rcode == NOERROR && anhp->ancount == 0 |
| 1244 | && anhp->aa == 0 && anhp->ra == 0 && anhp->arcount == 0) { |
| 1245 | goto next_ns; |
| 1246 | } |
| 1247 | if (!(statp->options & RES_IGNTC) && anhp->tc) { |
| 1248 | /* |
| 1249 | * To get the rest of answer, |
| 1250 | * use TCP with same server. |
| 1251 | */ |
| 1252 | *v_circuit = 1; |
| 1253 | __res_iclose(statp, false); |
| 1254 | // XXX if we have received one reply we could |
| 1255 | // XXX use it and not repeat it over TCP... |
| 1256 | if (resplen2 != NULL) |
| 1257 | *resplen2 = 0; |
| 1258 | return (1); |
| 1259 | } |
| 1260 | /* Mark which reply we received. */ |
| 1261 | if (matching_query == 1) |
| 1262 | recvresp1 = 1; |
| 1263 | else |
| 1264 | recvresp2 = 1; |
| 1265 | /* Repeat waiting if we have a second answer to arrive. */ |
| 1266 | if ((recvresp1 & recvresp2) == 0) { |
| 1267 | if (single_request) { |
| 1268 | pfd[0].events = POLLOUT; |
| 1269 | if (single_request_reopen) { |
| 1270 | __res_iclose (statp, false); |
| 1271 | retval = reopen (statp, terrno, ns); |
| 1272 | if (retval <= 0) |
| 1273 | { |
| 1274 | if (resplen2 != NULL) |
| 1275 | *resplen2 = 0; |
| 1276 | return retval; |
| 1277 | } |
| 1278 | pfd[0].fd = EXT(statp).nssocks[ns]; |
| 1279 | } |
| 1280 | } |
| 1281 | goto wait; |
| 1282 | } |
| 1283 | /* All is well. We have received both responses (if |
| 1284 | two responses were requested). */ |
| 1285 | return (resplen); |
| 1286 | } else if (pfd[0].revents & (POLLERR | POLLHUP | POLLNVAL)) |
| 1287 | /* Something went wrong. We can stop trying. */ |
| 1288 | return close_and_return_error (statp, resplen2); |
| 1289 | else { |
| 1290 | /* poll should not have returned > 0 in this case. */ |
| 1291 | abort (); |
| 1292 | } |
| 1293 | } |
| 1294 | |
| 1295 | static int |
| 1296 | sock_eq(struct sockaddr_in6 *a1, struct sockaddr_in6 *a2) { |
| 1297 | if (a1->sin6_family == a2->sin6_family) { |
| 1298 | if (a1->sin6_family == AF_INET) |
| 1299 | return ((((struct sockaddr_in *)a1)->sin_port == |
| 1300 | ((struct sockaddr_in *)a2)->sin_port) && |
| 1301 | (((struct sockaddr_in *)a1)->sin_addr.s_addr == |
| 1302 | ((struct sockaddr_in *)a2)->sin_addr.s_addr)); |
| 1303 | else |
| 1304 | return ((a1->sin6_port == a2->sin6_port) && |
| 1305 | !memcmp(&a1->sin6_addr, &a2->sin6_addr, |
| 1306 | sizeof (struct in6_addr))); |
| 1307 | } |
| 1308 | if (a1->sin6_family == AF_INET) { |
| 1309 | struct sockaddr_in6 *sap = a1; |
| 1310 | a1 = a2; |
| 1311 | a2 = sap; |
| 1312 | } /* assumes that AF_INET and AF_INET6 are the only possibilities */ |
| 1313 | return ((a1->sin6_port == ((struct sockaddr_in *)a2)->sin_port) && |
| 1314 | IN6_IS_ADDR_V4MAPPED(&a1->sin6_addr) && |
| 1315 | (a1->sin6_addr.s6_addr32[3] == |
| 1316 | ((struct sockaddr_in *)a2)->sin_addr.s_addr)); |
| 1317 | } |
| 1318 | |