1 /* $NetBSD: getaddrinfo.c,v 1.82 2006/03/25 12:09:40 rpaulo Exp $ */
2 /* $KAME: getaddrinfo.c,v 1.29 2000/08/31 17:26:57 itojun Exp $ */
3
4 /*
5 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of the project nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 */
32
33 #define LOG_TAG "getaddrinfo"
34
35 #include "getaddrinfo.h"
36
37 #include <arpa/inet.h>
38 #include <arpa/nameser.h>
39 #include <assert.h>
40 #include <ctype.h>
41 #include <errno.h>
42 #include <fcntl.h>
43 #include <net/if.h>
44 #include <netdb.h>
45 #include <netinet/in.h>
46 #include <stdbool.h>
47 #include <stddef.h>
48 #include <stdio.h>
49 #include <stdlib.h>
50 #include <string.h>
51 #include <sys/param.h>
52 #include <sys/socket.h>
53 #include <sys/stat.h>
54 #include <sys/types.h>
55 #include <sys/un.h>
56 #include <unistd.h>
57
58 #include <android-base/logging.h>
59
60 #include "netd_resolv/resolv.h"
61 #include "resolv_cache.h"
62 #include "resolv_private.h"
63
64 #define ANY 0
65
66 const char in_addrany[] = {0, 0, 0, 0};
67 const char in_loopback[] = {127, 0, 0, 1};
68 const char in6_addrany[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
69 const char in6_loopback[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1};
70
71 const struct afd {
72 int a_af;
73 int a_addrlen;
74 int a_socklen;
75 int a_off;
76 const char* a_addrany;
77 const char* a_loopback;
78 int a_scoped;
79 } afdl[] = {
80 {PF_INET6, sizeof(struct in6_addr), sizeof(struct sockaddr_in6),
81 offsetof(struct sockaddr_in6, sin6_addr), in6_addrany, in6_loopback, 1},
82 {PF_INET, sizeof(struct in_addr), sizeof(struct sockaddr_in),
83 offsetof(struct sockaddr_in, sin_addr), in_addrany, in_loopback, 0},
84 {0, 0, 0, 0, NULL, NULL, 0},
85 };
86
87 struct Explore {
88 int e_af;
89 int e_socktype;
90 int e_protocol;
91 int e_wild;
92 #define WILD_AF(ex) ((ex).e_wild & 0x01)
93 #define WILD_SOCKTYPE(ex) ((ex).e_wild & 0x02)
94 #define WILD_PROTOCOL(ex) ((ex).e_wild & 0x04)
95 };
96
97 const Explore explore_options[] = {
98 {PF_INET6, SOCK_DGRAM, IPPROTO_UDP, 0x07},
99 {PF_INET6, SOCK_STREAM, IPPROTO_TCP, 0x07},
100 {PF_INET6, SOCK_RAW, ANY, 0x05},
101 {PF_INET, SOCK_DGRAM, IPPROTO_UDP, 0x07},
102 {PF_INET, SOCK_STREAM, IPPROTO_TCP, 0x07},
103 {PF_INET, SOCK_RAW, ANY, 0x05},
104 {PF_UNSPEC, SOCK_DGRAM, IPPROTO_UDP, 0x07},
105 {PF_UNSPEC, SOCK_STREAM, IPPROTO_TCP, 0x07},
106 {PF_UNSPEC, SOCK_RAW, ANY, 0x05},
107 };
108
109 #define PTON_MAX 16
110 #define MAXPACKET (8 * 1024)
111
112 typedef union {
113 HEADER hdr;
114 u_char buf[MAXPACKET];
115 } querybuf;
116
117 struct res_target {
118 struct res_target* next;
119 const char* name; /* domain name */
120 int qclass, qtype; /* class and type of query */
121 u_char* answer; /* buffer to put answer */
122 int anslen; /* size of answer buffer */
123 int n; /* result length */
124 };
125
126 static int str2number(const char*);
127 static int explore_fqdn(const struct addrinfo*, const char*, const char*, struct addrinfo**,
128 const struct android_net_context*);
129 static int explore_null(const struct addrinfo*, const char*, struct addrinfo**);
130 static int explore_numeric(const struct addrinfo*, const char*, const char*, struct addrinfo**,
131 const char*);
132 static int explore_numeric_scope(const struct addrinfo*, const char*, const char*,
133 struct addrinfo**);
134 static int get_canonname(const struct addrinfo*, struct addrinfo*, const char*);
135 static struct addrinfo* get_ai(const struct addrinfo*, const struct afd*, const char*);
136 static int get_portmatch(const struct addrinfo*, const char*);
137 static int get_port(const struct addrinfo*, const char*, int);
138 static const struct afd* find_afd(int);
139 static int ip6_str2scopeid(const char*, struct sockaddr_in6*, u_int32_t*);
140
141 static struct addrinfo* getanswer(const querybuf*, int, const char*, int, const struct addrinfo*,
142 int* herrno);
143 static int dns_getaddrinfo(const char* name, const addrinfo* pai,
144 const android_net_context* netcontext, addrinfo** rv);
145 static void _sethtent(FILE**);
146 static void _endhtent(FILE**);
147 static struct addrinfo* _gethtent(FILE**, const char*, const struct addrinfo*);
148 static bool files_getaddrinfo(const char* name, const addrinfo* pai, addrinfo** res);
149 static int _find_src_addr(const struct sockaddr*, struct sockaddr*, unsigned, uid_t);
150
151 static int res_queryN(const char* name, res_target* target, res_state res, int* herrno);
152 static int res_searchN(const char* name, res_target* target, res_state res, int* herrno);
153 static int res_querydomainN(const char* name, const char* domain, res_target* target, res_state res,
154 int* herrno);
155
156 const char* const ai_errlist[] = {
157 "Success",
158 "Address family for hostname not supported", /* EAI_ADDRFAMILY */
159 "Temporary failure in name resolution", /* EAI_AGAIN */
160 "Invalid value for ai_flags", /* EAI_BADFLAGS */
161 "Non-recoverable failure in name resolution", /* EAI_FAIL */
162 "ai_family not supported", /* EAI_FAMILY */
163 "Memory allocation failure", /* EAI_MEMORY */
164 "No address associated with hostname", /* EAI_NODATA */
165 "hostname nor servname provided, or not known", /* EAI_NONAME */
166 "servname not supported for ai_socktype", /* EAI_SERVICE */
167 "ai_socktype not supported", /* EAI_SOCKTYPE */
168 "System error returned in errno", /* EAI_SYSTEM */
169 "Invalid value for hints", /* EAI_BADHINTS */
170 "Resolved protocol is unknown", /* EAI_PROTOCOL */
171 "Argument buffer overflow", /* EAI_OVERFLOW */
172 "Unknown error", /* EAI_MAX */
173 };
174
175 /* XXX macros that make external reference is BAD. */
176
177 #define GET_AI(ai, afd, addr) \
178 do { \
179 /* external reference: pai, error, and label free */ \
180 (ai) = get_ai(pai, (afd), (addr)); \
181 if ((ai) == NULL) { \
182 error = EAI_MEMORY; \
183 goto free; \
184 } \
185 } while (0)
186
187 #define GET_PORT(ai, serv) \
188 do { \
189 /* external reference: error and label free */ \
190 error = get_port((ai), (serv), 0); \
191 if (error != 0) goto free; \
192 } while (0)
193
194 #define MATCH_FAMILY(x, y, w) \
195 ((x) == (y) || ((w) && ((x) == PF_UNSPEC || (y) == PF_UNSPEC)))
196 #define MATCH(x, y, w) ((x) == (y) || ((w) && ((x) == ANY || (y) == ANY)))
197
gai_strerror(int ecode)198 const char* gai_strerror(int ecode) {
199 if (ecode < 0 || ecode > EAI_MAX) ecode = EAI_MAX;
200 return ai_errlist[ecode];
201 }
202
freeaddrinfo(struct addrinfo * ai)203 void freeaddrinfo(struct addrinfo* ai) {
204 while (ai) {
205 struct addrinfo* next = ai->ai_next;
206 if (ai->ai_canonname) free(ai->ai_canonname);
207 // Also frees ai->ai_addr which points to extra space beyond addrinfo
208 free(ai);
209 ai = next;
210 }
211 }
212
str2number(const char * p)213 static int str2number(const char* p) {
214 char* ep;
215 unsigned long v;
216
217 assert(p != NULL);
218
219 if (*p == '\0') return -1;
220 ep = NULL;
221 errno = 0;
222 v = strtoul(p, &ep, 10);
223 if (errno == 0 && ep && *ep == '\0' && v <= UINT_MAX)
224 return v;
225 else
226 return -1;
227 }
228
229 /*
230 * The following functions determine whether IPv4 or IPv6 connectivity is
231 * available in order to implement AI_ADDRCONFIG.
232 *
233 * Strictly speaking, AI_ADDRCONFIG should not look at whether connectivity is
234 * available, but whether addresses of the specified family are "configured
235 * on the local system". However, bionic doesn't currently support getifaddrs,
236 * so checking for connectivity is the next best thing.
237 */
have_ipv6(unsigned mark,uid_t uid)238 static int have_ipv6(unsigned mark, uid_t uid) {
239 static const struct sockaddr_in6 sin6_test = {
240 .sin6_family = AF_INET6,
241 .sin6_addr.s6_addr = {// 2000::
242 0x20, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}};
243 sockaddr_union addr = {.sin6 = sin6_test};
244 return _find_src_addr(&addr.sa, NULL, mark, uid) == 1;
245 }
246
have_ipv4(unsigned mark,uid_t uid)247 static int have_ipv4(unsigned mark, uid_t uid) {
248 static const struct sockaddr_in sin_test = {
249 .sin_family = AF_INET,
250 .sin_addr.s_addr = __constant_htonl(0x08080808L) // 8.8.8.8
251 };
252 sockaddr_union addr = {.sin = sin_test};
253 return _find_src_addr(&addr.sa, NULL, mark, uid) == 1;
254 }
255
256 // Internal version of getaddrinfo(), but limited to AI_NUMERICHOST.
257 // NOTE: also called by resolv_set_nameservers_for_net().
getaddrinfo_numeric(const char * hostname,const char * servname,addrinfo hints,addrinfo ** result)258 int getaddrinfo_numeric(const char* hostname, const char* servname, addrinfo hints,
259 addrinfo** result) {
260 hints.ai_flags = AI_NUMERICHOST;
261 const android_net_context netcontext = {
262 .app_netid = NETID_UNSET,
263 .app_mark = MARK_UNSET,
264 .dns_netid = NETID_UNSET,
265 .dns_mark = MARK_UNSET,
266 .uid = NET_CONTEXT_INVALID_UID,
267 };
268 return android_getaddrinfofornetcontext(hostname, servname, &hints, &netcontext, result);
269 }
270
android_getaddrinfofornetcontext(const char * hostname,const char * servname,const struct addrinfo * hints,const struct android_net_context * netcontext,struct addrinfo ** res)271 int android_getaddrinfofornetcontext(const char* hostname, const char* servname,
272 const struct addrinfo* hints,
273 const struct android_net_context* netcontext,
274 struct addrinfo** res) {
275 struct addrinfo sentinel = {};
276 struct addrinfo* cur = &sentinel;
277 int error = 0;
278
279 // hostname is allowed to be nullptr
280 // servname is allowed to be nullptr
281 // hints is allowed to be nullptr
282 assert(res != nullptr);
283 assert(netcontext != nullptr);
284
285 struct addrinfo ai = {
286 .ai_flags = 0,
287 .ai_family = PF_UNSPEC,
288 .ai_socktype = ANY,
289 .ai_protocol = ANY,
290 .ai_addrlen = 0,
291 .ai_canonname = nullptr,
292 .ai_addr = nullptr,
293 .ai_next = nullptr,
294 };
295
296 do {
297 if (hostname == NULL && servname == NULL) {
298 error = EAI_NONAME;
299 break;
300 }
301 if (hints) {
302 /* error check for hints */
303 if (hints->ai_addrlen || hints->ai_canonname || hints->ai_addr || hints->ai_next) {
304 error = EAI_BADHINTS;
305 break;
306 }
307 if (hints->ai_flags & ~AI_MASK) {
308 error = EAI_BADFLAGS;
309 break;
310 }
311
312 if (!(hints->ai_family == PF_UNSPEC || hints->ai_family == PF_INET ||
313 hints->ai_family == PF_INET6)) {
314 error = EAI_FAMILY;
315 break;
316 }
317
318 ai = *hints;
319
320 /*
321 * if both socktype/protocol are specified, check if they
322 * are meaningful combination.
323 */
324 if (ai.ai_socktype != ANY && ai.ai_protocol != ANY) {
325 for (const Explore& ex : explore_options) {
326 if (ai.ai_family != ex.e_af) continue;
327 if (ex.e_socktype == ANY) continue;
328 if (ex.e_protocol == ANY) continue;
329 if (ai.ai_socktype == ex.e_socktype && ai.ai_protocol != ex.e_protocol) {
330 error = EAI_BADHINTS;
331 break;
332 }
333 }
334 if (error) break;
335 }
336 }
337
338 /*
339 * Check for special cases:
340 * (1) numeric servname is disallowed if socktype/protocol are left unspecified.
341 * (2) servname is disallowed for raw and other inet{,6} sockets.
342 */
343 if (MATCH_FAMILY(ai.ai_family, PF_INET, 1) || MATCH_FAMILY(ai.ai_family, PF_INET6, 1)) {
344 struct addrinfo tmp = ai;
345 if (tmp.ai_family == PF_UNSPEC) {
346 tmp.ai_family = PF_INET6;
347 }
348 error = get_portmatch(&tmp, servname);
349 if (error) break;
350 }
351
352 // NULL hostname, or numeric hostname
353 for (const Explore& ex : explore_options) {
354 /* PF_UNSPEC entries are prepared for DNS queries only */
355 if (ex.e_af == PF_UNSPEC) continue;
356
357 if (!MATCH_FAMILY(ai.ai_family, ex.e_af, WILD_AF(ex))) continue;
358 if (!MATCH(ai.ai_socktype, ex.e_socktype, WILD_SOCKTYPE(ex))) continue;
359 if (!MATCH(ai.ai_protocol, ex.e_protocol, WILD_PROTOCOL(ex))) continue;
360
361 struct addrinfo tmp = ai;
362 if (tmp.ai_family == PF_UNSPEC) tmp.ai_family = ex.e_af;
363 if (tmp.ai_socktype == ANY && ex.e_socktype != ANY) tmp.ai_socktype = ex.e_socktype;
364 if (tmp.ai_protocol == ANY && ex.e_protocol != ANY) tmp.ai_protocol = ex.e_protocol;
365
366 LOG(DEBUG) << __func__ << ": explore_numeric: ai_family=" << tmp.ai_family
367 << " ai_socktype=" << tmp.ai_socktype << " ai_protocol=" << tmp.ai_protocol;
368 if (hostname == nullptr)
369 error = explore_null(&tmp, servname, &cur->ai_next);
370 else
371 error = explore_numeric_scope(&tmp, hostname, servname, &cur->ai_next);
372
373 if (error) break;
374
375 while (cur->ai_next) cur = cur->ai_next;
376 }
377 if (error) break;
378
379 /*
380 * XXX
381 * If numeric representation of AF1 can be interpreted as FQDN
382 * representation of AF2, we need to think again about the code below.
383 */
384 if (sentinel.ai_next) break;
385
386 if (hostname == nullptr) {
387 error = EAI_NODATA;
388 break;
389 }
390 if (ai.ai_flags & AI_NUMERICHOST) {
391 error = EAI_NONAME;
392 break;
393 }
394
395 /*
396 * hostname as alphabetical name.
397 * We would like to prefer AF_INET6 over AF_INET, so we'll make a outer loop by AFs.
398 */
399 for (const Explore& ex : explore_options) {
400 // Require exact match for family field
401 if (ai.ai_family != ex.e_af) continue;
402
403 if (!MATCH(ai.ai_socktype, ex.e_socktype, WILD_SOCKTYPE(ex))) {
404 continue;
405 }
406 if (!MATCH(ai.ai_protocol, ex.e_protocol, WILD_PROTOCOL(ex))) {
407 continue;
408 }
409
410 struct addrinfo tmp = ai;
411 if (tmp.ai_socktype == ANY && ex.e_socktype != ANY) tmp.ai_socktype = ex.e_socktype;
412 if (tmp.ai_protocol == ANY && ex.e_protocol != ANY) tmp.ai_protocol = ex.e_protocol;
413
414 LOG(DEBUG) << __func__ << ": explore_fqdn(): ai_family=" << tmp.ai_family
415 << " ai_socktype=" << tmp.ai_socktype << " ai_protocol=" << tmp.ai_protocol;
416 error = explore_fqdn(&tmp, hostname, servname, &cur->ai_next, netcontext);
417
418 while (cur->ai_next) cur = cur->ai_next;
419 }
420
421 if (sentinel.ai_next) {
422 error = 0;
423 } else if (error == 0) {
424 error = EAI_FAIL;
425 }
426 } while (0);
427
428 if (error) {
429 freeaddrinfo(sentinel.ai_next);
430 *res = nullptr;
431 } else {
432 *res = sentinel.ai_next;
433 }
434 return error;
435 }
436
437 // FQDN hostname, DNS lookup
explore_fqdn(const struct addrinfo * pai,const char * hostname,const char * servname,struct addrinfo ** res,const struct android_net_context * netcontext)438 static int explore_fqdn(const struct addrinfo* pai, const char* hostname, const char* servname,
439 struct addrinfo** res, const struct android_net_context* netcontext) {
440 struct addrinfo* result;
441 int error = 0;
442
443 assert(pai != NULL);
444 /* hostname may be NULL */
445 /* servname may be NULL */
446 assert(res != NULL);
447
448 result = NULL;
449
450 // If the servname does not match socktype/protocol, ignore it.
451 if (get_portmatch(pai, servname) != 0) return 0;
452
453 if (!files_getaddrinfo(hostname, pai, &result)) {
454 error = dns_getaddrinfo(hostname, pai, netcontext, &result);
455 }
456 if (!error) {
457 struct addrinfo* cur;
458 for (cur = result; cur; cur = cur->ai_next) {
459 GET_PORT(cur, servname);
460 /* canonname should be filled already */
461 }
462 *res = result;
463 return 0;
464 }
465
466 free:
467 freeaddrinfo(result);
468 return error;
469 }
470
471 /*
472 * hostname == NULL.
473 * passive socket -> anyaddr (0.0.0.0 or ::)
474 * non-passive socket -> localhost (127.0.0.1 or ::1)
475 */
explore_null(const struct addrinfo * pai,const char * servname,struct addrinfo ** res)476 static int explore_null(const struct addrinfo* pai, const char* servname, struct addrinfo** res) {
477 int s;
478 const struct afd* afd;
479 struct addrinfo* cur;
480 struct addrinfo sentinel;
481 int error;
482
483 LOG(DEBUG) << __func__;
484
485 assert(pai != NULL);
486 /* servname may be NULL */
487 assert(res != NULL);
488
489 *res = NULL;
490 sentinel.ai_next = NULL;
491 cur = &sentinel;
492
493 /*
494 * filter out AFs that are not supported by the kernel
495 * XXX errno?
496 */
497 s = socket(pai->ai_family, SOCK_DGRAM | SOCK_CLOEXEC, 0);
498 if (s < 0) {
499 if (errno != EMFILE) return 0;
500 } else
501 close(s);
502
503 /*
504 * if the servname does not match socktype/protocol, ignore it.
505 */
506 if (get_portmatch(pai, servname) != 0) return 0;
507
508 afd = find_afd(pai->ai_family);
509 if (afd == NULL) return 0;
510
511 if (pai->ai_flags & AI_PASSIVE) {
512 GET_AI(cur->ai_next, afd, afd->a_addrany);
513 GET_PORT(cur->ai_next, servname);
514 } else {
515 GET_AI(cur->ai_next, afd, afd->a_loopback);
516 GET_PORT(cur->ai_next, servname);
517 }
518 cur = cur->ai_next;
519
520 *res = sentinel.ai_next;
521 return 0;
522
523 free:
524 freeaddrinfo(sentinel.ai_next);
525 return error;
526 }
527
528 /*
529 * numeric hostname
530 */
explore_numeric(const struct addrinfo * pai,const char * hostname,const char * servname,struct addrinfo ** res,const char * canonname)531 static int explore_numeric(const struct addrinfo* pai, const char* hostname, const char* servname,
532 struct addrinfo** res, const char* canonname) {
533 const struct afd* afd;
534 struct addrinfo* cur;
535 struct addrinfo sentinel;
536 int error;
537 char pton[PTON_MAX];
538
539 assert(pai != NULL);
540 /* hostname may be NULL */
541 /* servname may be NULL */
542 assert(res != NULL);
543
544 *res = NULL;
545 sentinel.ai_next = NULL;
546 cur = &sentinel;
547
548 /*
549 * if the servname does not match socktype/protocol, ignore it.
550 */
551 if (get_portmatch(pai, servname) != 0) return 0;
552
553 afd = find_afd(pai->ai_family);
554 if (afd == NULL) return 0;
555
556 if (inet_pton(afd->a_af, hostname, pton) == 1) {
557 if (pai->ai_family == afd->a_af || pai->ai_family == PF_UNSPEC /*?*/) {
558 GET_AI(cur->ai_next, afd, pton);
559 GET_PORT(cur->ai_next, servname);
560 if ((pai->ai_flags & AI_CANONNAME)) {
561 /*
562 * Set the numeric address itself as
563 * the canonical name, based on a
564 * clarification in rfc2553bis-03.
565 */
566 error = get_canonname(pai, cur->ai_next, canonname);
567 if (error != 0) {
568 freeaddrinfo(sentinel.ai_next);
569 return error;
570 }
571 }
572 while (cur->ai_next) cur = cur->ai_next;
573 } else
574 return EAI_FAMILY;
575 }
576
577 *res = sentinel.ai_next;
578 return 0;
579
580 free:
581 freeaddrinfo(sentinel.ai_next);
582 return error;
583 }
584
585 /*
586 * numeric hostname with scope
587 */
explore_numeric_scope(const struct addrinfo * pai,const char * hostname,const char * servname,struct addrinfo ** res)588 static int explore_numeric_scope(const struct addrinfo* pai, const char* hostname,
589 const char* servname, struct addrinfo** res) {
590 const struct afd* afd;
591 struct addrinfo* cur;
592 int error;
593 const char *cp, *scope, *addr;
594 struct sockaddr_in6* sin6;
595
596 LOG(DEBUG) << __func__;
597
598 assert(pai != NULL);
599 /* hostname may be NULL */
600 /* servname may be NULL */
601 assert(res != NULL);
602
603 /*
604 * if the servname does not match socktype/protocol, ignore it.
605 */
606 if (get_portmatch(pai, servname) != 0) return 0;
607
608 afd = find_afd(pai->ai_family);
609 if (afd == NULL) return 0;
610
611 if (!afd->a_scoped) return explore_numeric(pai, hostname, servname, res, hostname);
612
613 cp = strchr(hostname, SCOPE_DELIMITER);
614 if (cp == NULL) return explore_numeric(pai, hostname, servname, res, hostname);
615
616 /*
617 * Handle special case of <scoped_address><delimiter><scope id>
618 */
619 char* hostname2 = strdup(hostname);
620 if (hostname2 == NULL) return EAI_MEMORY;
621 /* terminate at the delimiter */
622 hostname2[cp - hostname] = '\0';
623 addr = hostname2;
624 scope = cp + 1;
625
626 error = explore_numeric(pai, addr, servname, res, hostname);
627 if (error == 0) {
628 u_int32_t scopeid;
629
630 for (cur = *res; cur; cur = cur->ai_next) {
631 if (cur->ai_family != AF_INET6) continue;
632 sin6 = (struct sockaddr_in6*) (void*) cur->ai_addr;
633 if (ip6_str2scopeid(scope, sin6, &scopeid) == -1) {
634 free(hostname2);
635 return (EAI_NODATA); /* XXX: is return OK? */
636 }
637 sin6->sin6_scope_id = scopeid;
638 }
639 }
640
641 free(hostname2);
642
643 return error;
644 }
645
get_canonname(const struct addrinfo * pai,struct addrinfo * ai,const char * str)646 static int get_canonname(const struct addrinfo* pai, struct addrinfo* ai, const char* str) {
647 assert(pai != NULL);
648 assert(ai != NULL);
649 assert(str != NULL);
650
651 if ((pai->ai_flags & AI_CANONNAME) != 0) {
652 ai->ai_canonname = strdup(str);
653 if (ai->ai_canonname == NULL) return EAI_MEMORY;
654 }
655 return 0;
656 }
657
get_ai(const struct addrinfo * pai,const struct afd * afd,const char * addr)658 static struct addrinfo* get_ai(const struct addrinfo* pai, const struct afd* afd,
659 const char* addr) {
660 char* p;
661 struct addrinfo* ai;
662
663 assert(pai != NULL);
664 assert(afd != NULL);
665 assert(addr != NULL);
666
667 ai = (struct addrinfo*) malloc(sizeof(struct addrinfo) + sizeof(sockaddr_union));
668 if (ai == NULL) return NULL;
669
670 memcpy(ai, pai, sizeof(struct addrinfo));
671 ai->ai_addr = (struct sockaddr*) (void*) (ai + 1);
672 memset(ai->ai_addr, 0, sizeof(sockaddr_union));
673
674 ai->ai_addrlen = afd->a_socklen;
675 ai->ai_addr->sa_family = ai->ai_family = afd->a_af;
676 p = (char*) (void*) (ai->ai_addr);
677 memcpy(p + afd->a_off, addr, (size_t) afd->a_addrlen);
678 return ai;
679 }
680
get_portmatch(const struct addrinfo * ai,const char * servname)681 static int get_portmatch(const struct addrinfo* ai, const char* servname) {
682 assert(ai != NULL);
683 /* servname may be NULL */
684
685 return get_port(ai, servname, 1);
686 }
687
get_port(const struct addrinfo * ai,const char * servname,int matchonly)688 static int get_port(const struct addrinfo* ai, const char* servname, int matchonly) {
689 const char* proto;
690 struct servent* sp;
691 int port;
692 int allownumeric;
693
694 assert(ai != NULL);
695 /* servname may be NULL */
696
697 if (servname == NULL) return 0;
698 switch (ai->ai_family) {
699 case AF_INET:
700 case AF_INET6:
701 break;
702 default:
703 return 0;
704 }
705
706 switch (ai->ai_socktype) {
707 case SOCK_RAW:
708 return EAI_SERVICE;
709 case SOCK_DGRAM:
710 case SOCK_STREAM:
711 allownumeric = 1;
712 break;
713 case ANY:
714 allownumeric = 1;
715 break;
716 default:
717 return EAI_SOCKTYPE;
718 }
719
720 port = str2number(servname);
721 if (port >= 0) {
722 if (!allownumeric) return EAI_SERVICE;
723 if (port < 0 || port > 65535) return EAI_SERVICE;
724 port = htons(port);
725 } else {
726 if (ai->ai_flags & AI_NUMERICSERV) return EAI_NONAME;
727
728 switch (ai->ai_socktype) {
729 case SOCK_DGRAM:
730 proto = "udp";
731 break;
732 case SOCK_STREAM:
733 proto = "tcp";
734 break;
735 default:
736 proto = NULL;
737 break;
738 }
739
740 if ((sp = getservbyname(servname, proto)) == NULL) return EAI_SERVICE;
741 port = sp->s_port;
742 }
743
744 if (!matchonly) {
745 switch (ai->ai_family) {
746 case AF_INET:
747 ((struct sockaddr_in*) (void*) ai->ai_addr)->sin_port = port;
748 break;
749 case AF_INET6:
750 ((struct sockaddr_in6*) (void*) ai->ai_addr)->sin6_port = port;
751 break;
752 }
753 }
754
755 return 0;
756 }
757
find_afd(int af)758 static const struct afd* find_afd(int af) {
759 const struct afd* afd;
760
761 if (af == PF_UNSPEC) return NULL;
762 for (afd = afdl; afd->a_af; afd++) {
763 if (afd->a_af == af) return afd;
764 }
765 return NULL;
766 }
767
768 // Convert a string to a scope identifier.
ip6_str2scopeid(const char * scope,struct sockaddr_in6 * sin6,u_int32_t * scopeid)769 static int ip6_str2scopeid(const char* scope, struct sockaddr_in6* sin6, u_int32_t* scopeid) {
770 u_long lscopeid;
771 struct in6_addr* a6;
772 char* ep;
773
774 assert(scope != NULL);
775 assert(sin6 != NULL);
776 assert(scopeid != NULL);
777
778 a6 = &sin6->sin6_addr;
779
780 /* empty scopeid portion is invalid */
781 if (*scope == '\0') return -1;
782
783 if (IN6_IS_ADDR_LINKLOCAL(a6) || IN6_IS_ADDR_MC_LINKLOCAL(a6)) {
784 /*
785 * We currently assume a one-to-one mapping between links
786 * and interfaces, so we simply use interface indices for
787 * like-local scopes.
788 */
789 *scopeid = if_nametoindex(scope);
790 if (*scopeid == 0) goto trynumeric;
791 return 0;
792 }
793
794 /* still unclear about literal, allow numeric only - placeholder */
795 if (IN6_IS_ADDR_SITELOCAL(a6) || IN6_IS_ADDR_MC_SITELOCAL(a6)) goto trynumeric;
796 if (IN6_IS_ADDR_MC_ORGLOCAL(a6))
797 goto trynumeric;
798 else
799 goto trynumeric; /* global */
800
801 /* try to convert to a numeric id as a last resort */
802 trynumeric:
803 errno = 0;
804 lscopeid = strtoul(scope, &ep, 10);
805 *scopeid = (u_int32_t)(lscopeid & 0xffffffffUL);
806 if (errno == 0 && ep && *ep == '\0' && *scopeid == lscopeid)
807 return 0;
808 else
809 return -1;
810 }
811
812 /* code duplicate with gethnamaddr.c */
813
814 #define BOUNDED_INCR(x) \
815 do { \
816 BOUNDS_CHECK(cp, x); \
817 cp += (x); \
818 } while (0)
819
820 #define BOUNDS_CHECK(ptr, count) \
821 do { \
822 if (eom - (ptr) < (count)) { \
823 *herrno = NO_RECOVERY; \
824 return NULL; \
825 } \
826 } while (0)
827
getanswer(const querybuf * answer,int anslen,const char * qname,int qtype,const struct addrinfo * pai,int * herrno)828 static struct addrinfo* getanswer(const querybuf* answer, int anslen, const char* qname, int qtype,
829 const struct addrinfo* pai, int* herrno) {
830 struct addrinfo sentinel = {};
831 struct addrinfo *cur;
832 struct addrinfo ai;
833 const struct afd* afd;
834 char* canonname;
835 const HEADER* hp;
836 const u_char* cp;
837 int n;
838 const u_char* eom;
839 char *bp, *ep;
840 int type, ancount, qdcount;
841 int haveanswer, had_error;
842 char tbuf[MAXDNAME];
843 int (*name_ok)(const char*);
844 char hostbuf[8 * 1024];
845
846 assert(answer != NULL);
847 assert(qname != NULL);
848 assert(pai != NULL);
849
850 cur = &sentinel;
851
852 canonname = NULL;
853 eom = answer->buf + anslen;
854 switch (qtype) {
855 case T_A:
856 case T_AAAA:
857 case T_ANY: /*use T_ANY only for T_A/T_AAAA lookup*/
858 name_ok = res_hnok;
859 break;
860 default:
861 return NULL; /* XXX should be abort(); */
862 }
863 /*
864 * find first satisfactory answer
865 */
866 hp = &answer->hdr;
867 ancount = ntohs(hp->ancount);
868 qdcount = ntohs(hp->qdcount);
869 bp = hostbuf;
870 ep = hostbuf + sizeof hostbuf;
871 cp = answer->buf;
872 BOUNDED_INCR(HFIXEDSZ);
873 if (qdcount != 1) {
874 *herrno = NO_RECOVERY;
875 return (NULL);
876 }
877 n = dn_expand(answer->buf, eom, cp, bp, ep - bp);
878 if ((n < 0) || !(*name_ok)(bp)) {
879 *herrno = NO_RECOVERY;
880 return (NULL);
881 }
882 BOUNDED_INCR(n + QFIXEDSZ);
883 if (qtype == T_A || qtype == T_AAAA || qtype == T_ANY) {
884 /* res_send() has already verified that the query name is the
885 * same as the one we sent; this just gets the expanded name
886 * (i.e., with the succeeding search-domain tacked on).
887 */
888 n = strlen(bp) + 1; /* for the \0 */
889 if (n >= MAXHOSTNAMELEN) {
890 *herrno = NO_RECOVERY;
891 return (NULL);
892 }
893 canonname = bp;
894 bp += n;
895 /* The qname can be abbreviated, but h_name is now absolute. */
896 qname = canonname;
897 }
898 haveanswer = 0;
899 had_error = 0;
900 while (ancount-- > 0 && cp < eom && !had_error) {
901 n = dn_expand(answer->buf, eom, cp, bp, ep - bp);
902 if ((n < 0) || !(*name_ok)(bp)) {
903 had_error++;
904 continue;
905 }
906 cp += n; /* name */
907 BOUNDS_CHECK(cp, 3 * INT16SZ + INT32SZ);
908 type = ntohs(*reinterpret_cast<const uint16_t*>(cp));
909 cp += INT16SZ; /* type */
910 int cl = ntohs(*reinterpret_cast<const uint16_t*>(cp));
911 cp += INT16SZ + INT32SZ; /* class, TTL */
912 n = ntohs(*reinterpret_cast<const uint16_t*>(cp));
913 cp += INT16SZ; /* len */
914 BOUNDS_CHECK(cp, n);
915 if (cl != C_IN) {
916 /* XXX - debug? syslog? */
917 cp += n;
918 continue; /* XXX - had_error++ ? */
919 }
920 if ((qtype == T_A || qtype == T_AAAA || qtype == T_ANY) && type == T_CNAME) {
921 n = dn_expand(answer->buf, eom, cp, tbuf, sizeof tbuf);
922 if ((n < 0) || !(*name_ok)(tbuf)) {
923 had_error++;
924 continue;
925 }
926 cp += n;
927 /* Get canonical name. */
928 n = strlen(tbuf) + 1; /* for the \0 */
929 if (n > ep - bp || n >= MAXHOSTNAMELEN) {
930 had_error++;
931 continue;
932 }
933 strlcpy(bp, tbuf, (size_t)(ep - bp));
934 canonname = bp;
935 bp += n;
936 continue;
937 }
938 if (qtype == T_ANY) {
939 if (!(type == T_A || type == T_AAAA)) {
940 cp += n;
941 continue;
942 }
943 } else if (type != qtype) {
944 if (type != T_KEY && type != T_SIG)
945 LOG(DEBUG) << __func__ << ": asked for \"" << qname << " " << p_class(C_IN) << " "
946 << p_type(qtype) << "\", got type \"" << p_type(type) << "\"";
947 cp += n;
948 continue; /* XXX - had_error++ ? */
949 }
950 switch (type) {
951 case T_A:
952 case T_AAAA:
953 if (strcasecmp(canonname, bp) != 0) {
954 LOG(DEBUG) << __func__ << ": asked for \"" << canonname << "\", got \"" << bp
955 << "\"";
956 cp += n;
957 continue; /* XXX - had_error++ ? */
958 }
959 if (type == T_A && n != INADDRSZ) {
960 cp += n;
961 continue;
962 }
963 if (type == T_AAAA && n != IN6ADDRSZ) {
964 cp += n;
965 continue;
966 }
967 if (type == T_AAAA) {
968 struct in6_addr in6;
969 memcpy(&in6, cp, IN6ADDRSZ);
970 if (IN6_IS_ADDR_V4MAPPED(&in6)) {
971 cp += n;
972 continue;
973 }
974 }
975 if (!haveanswer) {
976 int nn;
977
978 canonname = bp;
979 nn = strlen(bp) + 1; /* for the \0 */
980 bp += nn;
981 }
982
983 /* don't overwrite pai */
984 ai = *pai;
985 ai.ai_family = (type == T_A) ? AF_INET : AF_INET6;
986 afd = find_afd(ai.ai_family);
987 if (afd == NULL) {
988 cp += n;
989 continue;
990 }
991 cur->ai_next = get_ai(&ai, afd, (const char*) cp);
992 if (cur->ai_next == NULL) had_error++;
993 while (cur && cur->ai_next) cur = cur->ai_next;
994 cp += n;
995 break;
996 default:
997 abort();
998 }
999 if (!had_error) haveanswer++;
1000 }
1001 if (haveanswer) {
1002 if (!canonname)
1003 (void) get_canonname(pai, sentinel.ai_next, qname);
1004 else
1005 (void) get_canonname(pai, sentinel.ai_next, canonname);
1006 *herrno = NETDB_SUCCESS;
1007 return sentinel.ai_next;
1008 }
1009
1010 *herrno = NO_RECOVERY;
1011 return NULL;
1012 }
1013
1014 struct addrinfo_sort_elem {
1015 struct addrinfo* ai;
1016 int has_src_addr;
1017 sockaddr_union src_addr;
1018 int original_order;
1019 };
1020
_get_scope(const struct sockaddr * addr)1021 static int _get_scope(const struct sockaddr* addr) {
1022 if (addr->sa_family == AF_INET6) {
1023 const struct sockaddr_in6* addr6 = (const struct sockaddr_in6*) addr;
1024 if (IN6_IS_ADDR_MULTICAST(&addr6->sin6_addr)) {
1025 return IPV6_ADDR_MC_SCOPE(&addr6->sin6_addr);
1026 } else if (IN6_IS_ADDR_LOOPBACK(&addr6->sin6_addr) ||
1027 IN6_IS_ADDR_LINKLOCAL(&addr6->sin6_addr)) {
1028 /*
1029 * RFC 4291 section 2.5.3 says loopback is to be treated as having
1030 * link-local scope.
1031 */
1032 return IPV6_ADDR_SCOPE_LINKLOCAL;
1033 } else if (IN6_IS_ADDR_SITELOCAL(&addr6->sin6_addr)) {
1034 return IPV6_ADDR_SCOPE_SITELOCAL;
1035 } else {
1036 return IPV6_ADDR_SCOPE_GLOBAL;
1037 }
1038 } else if (addr->sa_family == AF_INET) {
1039 const struct sockaddr_in* addr4 = (const struct sockaddr_in*) addr;
1040 unsigned long int na = ntohl(addr4->sin_addr.s_addr);
1041
1042 if (IN_LOOPBACK(na) || /* 127.0.0.0/8 */
1043 (na & 0xffff0000) == 0xa9fe0000) { /* 169.254.0.0/16 */
1044 return IPV6_ADDR_SCOPE_LINKLOCAL;
1045 } else {
1046 /*
1047 * RFC 6724 section 3.2. Other IPv4 addresses, including private addresses
1048 * and shared addresses (100.64.0.0/10), are assigned global scope.
1049 */
1050 return IPV6_ADDR_SCOPE_GLOBAL;
1051 }
1052 } else {
1053 /*
1054 * This should never happen.
1055 * Return a scope with low priority as a last resort.
1056 */
1057 return IPV6_ADDR_SCOPE_NODELOCAL;
1058 }
1059 }
1060
1061 /* These macros are modelled after the ones in <netinet/in6.h>. */
1062
1063 /* RFC 4380, section 2.6 */
1064 #define IN6_IS_ADDR_TEREDO(a) \
1065 ((*(const uint32_t*) (const void*) (&(a)->s6_addr[0]) == ntohl(0x20010000)))
1066
1067 /* RFC 3056, section 2. */
1068 #define IN6_IS_ADDR_6TO4(a) (((a)->s6_addr[0] == 0x20) && ((a)->s6_addr[1] == 0x02))
1069
1070 /* 6bone testing address area (3ffe::/16), deprecated in RFC 3701. */
1071 #define IN6_IS_ADDR_6BONE(a) (((a)->s6_addr[0] == 0x3f) && ((a)->s6_addr[1] == 0xfe))
1072
1073 /*
1074 * Get the label for a given IPv4/IPv6 address.
1075 * RFC 6724, section 2.1.
1076 */
1077
_get_label(const struct sockaddr * addr)1078 static int _get_label(const struct sockaddr* addr) {
1079 if (addr->sa_family == AF_INET) {
1080 return 4;
1081 } else if (addr->sa_family == AF_INET6) {
1082 const struct sockaddr_in6* addr6 = (const struct sockaddr_in6*) addr;
1083 if (IN6_IS_ADDR_LOOPBACK(&addr6->sin6_addr)) {
1084 return 0;
1085 } else if (IN6_IS_ADDR_V4MAPPED(&addr6->sin6_addr)) {
1086 return 4;
1087 } else if (IN6_IS_ADDR_6TO4(&addr6->sin6_addr)) {
1088 return 2;
1089 } else if (IN6_IS_ADDR_TEREDO(&addr6->sin6_addr)) {
1090 return 5;
1091 } else if (IN6_IS_ADDR_ULA(&addr6->sin6_addr)) {
1092 return 13;
1093 } else if (IN6_IS_ADDR_V4COMPAT(&addr6->sin6_addr)) {
1094 return 3;
1095 } else if (IN6_IS_ADDR_SITELOCAL(&addr6->sin6_addr)) {
1096 return 11;
1097 } else if (IN6_IS_ADDR_6BONE(&addr6->sin6_addr)) {
1098 return 12;
1099 } else {
1100 /* All other IPv6 addresses, including global unicast addresses. */
1101 return 1;
1102 }
1103 } else {
1104 /*
1105 * This should never happen.
1106 * Return a semi-random label as a last resort.
1107 */
1108 return 1;
1109 }
1110 }
1111
1112 /*
1113 * Get the precedence for a given IPv4/IPv6 address.
1114 * RFC 6724, section 2.1.
1115 */
1116
_get_precedence(const struct sockaddr * addr)1117 static int _get_precedence(const struct sockaddr* addr) {
1118 if (addr->sa_family == AF_INET) {
1119 return 35;
1120 } else if (addr->sa_family == AF_INET6) {
1121 const struct sockaddr_in6* addr6 = (const struct sockaddr_in6*) addr;
1122 if (IN6_IS_ADDR_LOOPBACK(&addr6->sin6_addr)) {
1123 return 50;
1124 } else if (IN6_IS_ADDR_V4MAPPED(&addr6->sin6_addr)) {
1125 return 35;
1126 } else if (IN6_IS_ADDR_6TO4(&addr6->sin6_addr)) {
1127 return 30;
1128 } else if (IN6_IS_ADDR_TEREDO(&addr6->sin6_addr)) {
1129 return 5;
1130 } else if (IN6_IS_ADDR_ULA(&addr6->sin6_addr)) {
1131 return 3;
1132 } else if (IN6_IS_ADDR_V4COMPAT(&addr6->sin6_addr) ||
1133 IN6_IS_ADDR_SITELOCAL(&addr6->sin6_addr) ||
1134 IN6_IS_ADDR_6BONE(&addr6->sin6_addr)) {
1135 return 1;
1136 } else {
1137 /* All other IPv6 addresses, including global unicast addresses. */
1138 return 40;
1139 }
1140 } else {
1141 return 1;
1142 }
1143 }
1144
1145 /*
1146 * Find number of matching initial bits between the two addresses a1 and a2.
1147 */
1148
_common_prefix_len(const struct in6_addr * a1,const struct in6_addr * a2)1149 static int _common_prefix_len(const struct in6_addr* a1, const struct in6_addr* a2) {
1150 const char* p1 = (const char*) a1;
1151 const char* p2 = (const char*) a2;
1152 unsigned i;
1153
1154 for (i = 0; i < sizeof(*a1); ++i) {
1155 int x, j;
1156
1157 if (p1[i] == p2[i]) {
1158 continue;
1159 }
1160 x = p1[i] ^ p2[i];
1161 for (j = 0; j < CHAR_BIT; ++j) {
1162 if (x & (1 << (CHAR_BIT - 1))) {
1163 return i * CHAR_BIT + j;
1164 }
1165 x <<= 1;
1166 }
1167 }
1168 return sizeof(*a1) * CHAR_BIT;
1169 }
1170
1171 /*
1172 * Compare two source/destination address pairs.
1173 * RFC 6724, section 6.
1174 */
1175
_rfc6724_compare(const void * ptr1,const void * ptr2)1176 static int _rfc6724_compare(const void* ptr1, const void* ptr2) {
1177 const struct addrinfo_sort_elem* a1 = (const struct addrinfo_sort_elem*) ptr1;
1178 const struct addrinfo_sort_elem* a2 = (const struct addrinfo_sort_elem*) ptr2;
1179 int scope_src1, scope_dst1, scope_match1;
1180 int scope_src2, scope_dst2, scope_match2;
1181 int label_src1, label_dst1, label_match1;
1182 int label_src2, label_dst2, label_match2;
1183 int precedence1, precedence2;
1184 int prefixlen1, prefixlen2;
1185
1186 /* Rule 1: Avoid unusable destinations. */
1187 if (a1->has_src_addr != a2->has_src_addr) {
1188 return a2->has_src_addr - a1->has_src_addr;
1189 }
1190
1191 /* Rule 2: Prefer matching scope. */
1192 scope_src1 = _get_scope(&a1->src_addr.sa);
1193 scope_dst1 = _get_scope(a1->ai->ai_addr);
1194 scope_match1 = (scope_src1 == scope_dst1);
1195
1196 scope_src2 = _get_scope(&a2->src_addr.sa);
1197 scope_dst2 = _get_scope(a2->ai->ai_addr);
1198 scope_match2 = (scope_src2 == scope_dst2);
1199
1200 if (scope_match1 != scope_match2) {
1201 return scope_match2 - scope_match1;
1202 }
1203
1204 /*
1205 * Rule 3: Avoid deprecated addresses.
1206 * TODO(sesse): We don't currently have a good way of finding this.
1207 */
1208
1209 /*
1210 * Rule 4: Prefer home addresses.
1211 * TODO(sesse): We don't currently have a good way of finding this.
1212 */
1213
1214 /* Rule 5: Prefer matching label. */
1215 label_src1 = _get_label(&a1->src_addr.sa);
1216 label_dst1 = _get_label(a1->ai->ai_addr);
1217 label_match1 = (label_src1 == label_dst1);
1218
1219 label_src2 = _get_label(&a2->src_addr.sa);
1220 label_dst2 = _get_label(a2->ai->ai_addr);
1221 label_match2 = (label_src2 == label_dst2);
1222
1223 if (label_match1 != label_match2) {
1224 return label_match2 - label_match1;
1225 }
1226
1227 /* Rule 6: Prefer higher precedence. */
1228 precedence1 = _get_precedence(a1->ai->ai_addr);
1229 precedence2 = _get_precedence(a2->ai->ai_addr);
1230 if (precedence1 != precedence2) {
1231 return precedence2 - precedence1;
1232 }
1233
1234 /*
1235 * Rule 7: Prefer native transport.
1236 * TODO(sesse): We don't currently have a good way of finding this.
1237 */
1238
1239 /* Rule 8: Prefer smaller scope. */
1240 if (scope_dst1 != scope_dst2) {
1241 return scope_dst1 - scope_dst2;
1242 }
1243
1244 /*
1245 * Rule 9: Use longest matching prefix.
1246 * We implement this for IPv6 only, as the rules in RFC 6724 don't seem
1247 * to work very well directly applied to IPv4. (glibc uses information from
1248 * the routing table for a custom IPv4 implementation here.)
1249 */
1250 if (a1->has_src_addr && a1->ai->ai_addr->sa_family == AF_INET6 && a2->has_src_addr &&
1251 a2->ai->ai_addr->sa_family == AF_INET6) {
1252 const struct sockaddr_in6* a1_src = &a1->src_addr.sin6;
1253 const struct sockaddr_in6* a1_dst = (const struct sockaddr_in6*) a1->ai->ai_addr;
1254 const struct sockaddr_in6* a2_src = &a2->src_addr.sin6;
1255 const struct sockaddr_in6* a2_dst = (const struct sockaddr_in6*) a2->ai->ai_addr;
1256 prefixlen1 = _common_prefix_len(&a1_src->sin6_addr, &a1_dst->sin6_addr);
1257 prefixlen2 = _common_prefix_len(&a2_src->sin6_addr, &a2_dst->sin6_addr);
1258 if (prefixlen1 != prefixlen2) {
1259 return prefixlen2 - prefixlen1;
1260 }
1261 }
1262
1263 /*
1264 * Rule 10: Leave the order unchanged.
1265 * We need this since qsort() is not necessarily stable.
1266 */
1267 return a1->original_order - a2->original_order;
1268 }
1269
1270 /*
1271 * Find the source address that will be used if trying to connect to the given
1272 * address. src_addr must be large enough to hold a struct sockaddr_in6.
1273 *
1274 * Returns 1 if a source address was found, 0 if the address is unreachable,
1275 * and -1 if a fatal error occurred. If 0 or -1, the contents of src_addr are
1276 * undefined.
1277 */
1278
_find_src_addr(const struct sockaddr * addr,struct sockaddr * src_addr,unsigned mark,uid_t uid)1279 static int _find_src_addr(const struct sockaddr* addr, struct sockaddr* src_addr, unsigned mark,
1280 uid_t uid) {
1281 int sock;
1282 int ret;
1283 socklen_t len;
1284
1285 switch (addr->sa_family) {
1286 case AF_INET:
1287 len = sizeof(struct sockaddr_in);
1288 break;
1289 case AF_INET6:
1290 len = sizeof(struct sockaddr_in6);
1291 break;
1292 default:
1293 /* No known usable source address for non-INET families. */
1294 return 0;
1295 }
1296
1297 sock = socket(addr->sa_family, SOCK_DGRAM | SOCK_CLOEXEC, IPPROTO_UDP);
1298 if (sock == -1) {
1299 if (errno == EAFNOSUPPORT) {
1300 return 0;
1301 } else {
1302 return -1;
1303 }
1304 }
1305 if (mark != MARK_UNSET && setsockopt(sock, SOL_SOCKET, SO_MARK, &mark, sizeof(mark)) < 0) {
1306 close(sock);
1307 return 0;
1308 }
1309 if (uid > 0 && uid != NET_CONTEXT_INVALID_UID && fchown(sock, uid, (gid_t) -1) < 0) {
1310 close(sock);
1311 return 0;
1312 }
1313 do {
1314 ret = connect(sock, addr, len);
1315 } while (ret == -1 && errno == EINTR);
1316
1317 if (ret == -1) {
1318 close(sock);
1319 return 0;
1320 }
1321
1322 if (src_addr && getsockname(sock, src_addr, &len) == -1) {
1323 close(sock);
1324 return -1;
1325 }
1326 close(sock);
1327 return 1;
1328 }
1329
1330 /*
1331 * Sort the linked list starting at sentinel->ai_next in RFC6724 order.
1332 * Will leave the list unchanged if an error occurs.
1333 */
1334
_rfc6724_sort(struct addrinfo * list_sentinel,unsigned mark,uid_t uid)1335 static void _rfc6724_sort(struct addrinfo* list_sentinel, unsigned mark, uid_t uid) {
1336 struct addrinfo* cur;
1337 int nelem = 0, i;
1338 struct addrinfo_sort_elem* elems;
1339
1340 cur = list_sentinel->ai_next;
1341 while (cur) {
1342 ++nelem;
1343 cur = cur->ai_next;
1344 }
1345
1346 elems = (struct addrinfo_sort_elem*) malloc(nelem * sizeof(struct addrinfo_sort_elem));
1347 if (elems == NULL) {
1348 goto error;
1349 }
1350
1351 /*
1352 * Convert the linked list to an array that also contains the candidate
1353 * source address for each destination address.
1354 */
1355 for (i = 0, cur = list_sentinel->ai_next; i < nelem; ++i, cur = cur->ai_next) {
1356 int has_src_addr;
1357 assert(cur != NULL);
1358 elems[i].ai = cur;
1359 elems[i].original_order = i;
1360
1361 has_src_addr = _find_src_addr(cur->ai_addr, &elems[i].src_addr.sa, mark, uid);
1362 if (has_src_addr == -1) {
1363 goto error;
1364 }
1365 elems[i].has_src_addr = has_src_addr;
1366 }
1367
1368 /* Sort the addresses, and rearrange the linked list so it matches the sorted order. */
1369 qsort((void*) elems, nelem, sizeof(struct addrinfo_sort_elem), _rfc6724_compare);
1370
1371 list_sentinel->ai_next = elems[0].ai;
1372 for (i = 0; i < nelem - 1; ++i) {
1373 elems[i].ai->ai_next = elems[i + 1].ai;
1374 }
1375 elems[nelem - 1].ai->ai_next = NULL;
1376
1377 error:
1378 free(elems);
1379 }
1380
dns_getaddrinfo(const char * name,const addrinfo * pai,const android_net_context * netcontext,addrinfo ** rv)1381 static int dns_getaddrinfo(const char* name, const addrinfo* pai,
1382 const android_net_context* netcontext, addrinfo** rv) {
1383 res_target q = {};
1384 res_target q2 = {};
1385
1386 auto buf = std::make_unique<querybuf>();
1387 auto buf2 = std::make_unique<querybuf>();
1388
1389 switch (pai->ai_family) {
1390 case AF_UNSPEC: {
1391 /* prefer IPv6 */
1392 q.name = name;
1393 q.qclass = C_IN;
1394 q.answer = buf->buf;
1395 q.anslen = sizeof(buf->buf);
1396 int query_ipv6 = 1, query_ipv4 = 1;
1397 if (pai->ai_flags & AI_ADDRCONFIG) {
1398 query_ipv6 = have_ipv6(netcontext->app_mark, netcontext->uid);
1399 query_ipv4 = have_ipv4(netcontext->app_mark, netcontext->uid);
1400 }
1401 if (query_ipv6) {
1402 q.qtype = T_AAAA;
1403 if (query_ipv4) {
1404 q.next = &q2;
1405 q2.name = name;
1406 q2.qclass = C_IN;
1407 q2.qtype = T_A;
1408 q2.answer = buf2->buf;
1409 q2.anslen = sizeof(buf2->buf);
1410 }
1411 } else if (query_ipv4) {
1412 q.qtype = T_A;
1413 } else {
1414 return EAI_NODATA;
1415 }
1416 break;
1417 }
1418 case AF_INET:
1419 q.name = name;
1420 q.qclass = C_IN;
1421 q.qtype = T_A;
1422 q.answer = buf->buf;
1423 q.anslen = sizeof(buf->buf);
1424 break;
1425 case AF_INET6:
1426 q.name = name;
1427 q.qclass = C_IN;
1428 q.qtype = T_AAAA;
1429 q.answer = buf->buf;
1430 q.anslen = sizeof(buf->buf);
1431 break;
1432 default:
1433 return EAI_FAMILY;
1434 }
1435
1436 res_state res = res_get_state();
1437 if (!res) return EAI_MEMORY;
1438
1439 /* this just sets our netid val in the thread private data so we don't have to
1440 * modify the api's all the way down to res_send.c's res_nsend. We could
1441 * fully populate the thread private data here, but if we get down there
1442 * and have a cache hit that would be wasted, so we do the rest there on miss
1443 */
1444 res_setnetcontext(res, netcontext);
1445
1446 int he;
1447 if (res_searchN(name, &q, res, &he) < 0) {
1448 // Return h_errno (he) to catch more detailed errors rather than EAI_NODATA.
1449 // Note that res_searchN() doesn't set the pair NETDB_INTERNAL and errno.
1450 // See also herrnoToAiErrno().
1451 return herrnoToAiErrno(he);
1452 }
1453
1454 addrinfo sentinel = {};
1455 addrinfo* cur = &sentinel;
1456 addrinfo* ai = getanswer(buf.get(), q.n, q.name, q.qtype, pai, &he);
1457 if (ai) {
1458 cur->ai_next = ai;
1459 while (cur && cur->ai_next) cur = cur->ai_next;
1460 }
1461 if (q.next) {
1462 ai = getanswer(buf2.get(), q2.n, q2.name, q2.qtype, pai, &he);
1463 if (ai) cur->ai_next = ai;
1464 }
1465 if (sentinel.ai_next == NULL) {
1466 // Note that getanswer() doesn't set the pair NETDB_INTERNAL and errno.
1467 // See also herrnoToAiErrno().
1468 return herrnoToAiErrno(he);
1469 }
1470
1471 _rfc6724_sort(&sentinel, netcontext->app_mark, netcontext->uid);
1472
1473 *rv = sentinel.ai_next;
1474 return 0;
1475 }
1476
_sethtent(FILE ** hostf)1477 static void _sethtent(FILE** hostf) {
1478 if (!*hostf)
1479 *hostf = fopen(_PATH_HOSTS, "re");
1480 else
1481 rewind(*hostf);
1482 }
1483
_endhtent(FILE ** hostf)1484 static void _endhtent(FILE** hostf) {
1485 if (*hostf) {
1486 (void) fclose(*hostf);
1487 *hostf = NULL;
1488 }
1489 }
1490
_gethtent(FILE ** hostf,const char * name,const struct addrinfo * pai)1491 static struct addrinfo* _gethtent(FILE** hostf, const char* name, const struct addrinfo* pai) {
1492 char* p;
1493 char *cp, *tname, *cname;
1494 struct addrinfo *res0, *res;
1495 int error;
1496 const char* addr;
1497 char hostbuf[8 * 1024];
1498
1499 assert(name != NULL);
1500 assert(pai != NULL);
1501
1502 if (!*hostf && !(*hostf = fopen(_PATH_HOSTS, "re"))) return (NULL);
1503 again:
1504 if (!(p = fgets(hostbuf, sizeof hostbuf, *hostf))) return (NULL);
1505 if (*p == '#') goto again;
1506 if (!(cp = strpbrk(p, "#\n"))) goto again;
1507 *cp = '\0';
1508 if (!(cp = strpbrk(p, " \t"))) goto again;
1509 *cp++ = '\0';
1510 addr = p;
1511 /* if this is not something we're looking for, skip it. */
1512 cname = NULL;
1513 while (cp && *cp) {
1514 if (*cp == ' ' || *cp == '\t') {
1515 cp++;
1516 continue;
1517 }
1518 if (!cname) cname = cp;
1519 tname = cp;
1520 if ((cp = strpbrk(cp, " \t")) != NULL) *cp++ = '\0';
1521 if (strcasecmp(name, tname) == 0) goto found;
1522 }
1523 goto again;
1524
1525 found:
1526 error = getaddrinfo_numeric(addr, nullptr, *pai, &res0);
1527 if (error) goto again;
1528 for (res = res0; res; res = res->ai_next) {
1529 /* cover it up */
1530 res->ai_flags = pai->ai_flags;
1531
1532 if (pai->ai_flags & AI_CANONNAME) {
1533 if (get_canonname(pai, res, cname) != 0) {
1534 freeaddrinfo(res0);
1535 goto again;
1536 }
1537 }
1538 }
1539 return res0;
1540 }
1541
files_getaddrinfo(const char * name,const addrinfo * pai,addrinfo ** res)1542 static bool files_getaddrinfo(const char* name, const addrinfo* pai, addrinfo** res) {
1543 struct addrinfo sentinel = {};
1544 struct addrinfo *p, *cur;
1545 FILE* hostf = NULL;
1546
1547 cur = &sentinel;
1548
1549 _sethtent(&hostf);
1550 while ((p = _gethtent(&hostf, name, pai)) != NULL) {
1551 cur->ai_next = p;
1552 while (cur && cur->ai_next) cur = cur->ai_next;
1553 }
1554 _endhtent(&hostf);
1555
1556 *res = sentinel.ai_next;
1557 return sentinel.ai_next != NULL;
1558 }
1559
1560 /* resolver logic */
1561
1562 /*
1563 * Formulate a normal query, send, and await answer.
1564 * Returned answer is placed in supplied buffer "answer".
1565 * Perform preliminary check of answer, returning success only
1566 * if no error is indicated and the answer count is nonzero.
1567 * Return the size of the response on success, -1 on error.
1568 * Error number is left in *herrno.
1569 *
1570 * Caller must parse answer and determine whether it answers the question.
1571 */
res_queryN(const char * name,res_target * target,res_state res,int * herrno)1572 static int res_queryN(const char* name, res_target* target, res_state res, int* herrno) {
1573 u_char buf[MAXPACKET];
1574 HEADER* hp;
1575 int n;
1576 struct res_target* t;
1577 int rcode;
1578 int ancount;
1579
1580 assert(name != NULL);
1581 /* XXX: target may be NULL??? */
1582
1583 rcode = NOERROR;
1584 ancount = 0;
1585
1586 for (t = target; t; t = t->next) {
1587 u_char* answer;
1588 int anslen;
1589
1590 hp = (HEADER*) (void*) t->answer;
1591 bool retried = false;
1592 again:
1593 hp->rcode = NOERROR; /* default */
1594
1595 /* make it easier... */
1596 int cl = t->qclass;
1597 int type = t->qtype;
1598 answer = t->answer;
1599 anslen = t->anslen;
1600
1601 LOG(DEBUG) << __func__ << ": (" << cl << ", " << type << ")";
1602
1603 n = res_nmkquery(res, QUERY, name, cl, type, NULL, 0, NULL, buf, sizeof(buf));
1604 if (n > 0 && (res->options & (RES_USE_EDNS0 | RES_USE_DNSSEC)) != 0 && !retried)
1605 n = res_nopt(res, n, buf, sizeof(buf), anslen);
1606 if (n <= 0) {
1607 LOG(ERROR) << __func__ << ": res_nmkquery failed";
1608 *herrno = NO_RECOVERY;
1609 return n;
1610 }
1611
1612 n = res_nsend(res, buf, n, answer, anslen, &rcode, 0);
1613 if (n < 0 || hp->rcode != NOERROR || ntohs(hp->ancount) == 0) {
1614 // Record rcode from DNS response header only if no timeout.
1615 // Keep rcode timeout for reporting later if any.
1616 if (rcode != RCODE_TIMEOUT) rcode = hp->rcode; /* record most recent error */
1617 /* if the query choked with EDNS0, retry without EDNS0 */
1618 if ((res->options & (RES_USE_EDNS0 | RES_USE_DNSSEC)) != 0 &&
1619 (res->_flags & RES_F_EDNS0ERR) && !retried) {
1620 LOG(DEBUG) << __func__ << ": retry without EDNS0";
1621 retried = true;
1622 goto again;
1623 }
1624 LOG(DEBUG) << __func__ << ": rcode=" << hp->rcode << ", ancount=" << ntohs(hp->ancount);
1625 continue;
1626 }
1627
1628 ancount += ntohs(hp->ancount);
1629
1630 t->n = n;
1631 }
1632
1633 if (ancount == 0) {
1634 switch (rcode) {
1635 // Not defined in RFC.
1636 case RCODE_TIMEOUT:
1637 // DNS metrics monitors DNS query timeout.
1638 *herrno = NETD_RESOLV_H_ERRNO_EXT_TIMEOUT; // extended h_errno.
1639 break;
1640 // Defined in RFC 1035 section 4.1.1.
1641 case NXDOMAIN:
1642 *herrno = HOST_NOT_FOUND;
1643 break;
1644 case SERVFAIL:
1645 *herrno = TRY_AGAIN;
1646 break;
1647 case NOERROR:
1648 *herrno = NO_DATA;
1649 break;
1650 case FORMERR:
1651 case NOTIMP:
1652 case REFUSED:
1653 default:
1654 *herrno = NO_RECOVERY;
1655 break;
1656 }
1657 return -1;
1658 }
1659 return ancount;
1660 }
1661
1662 /*
1663 * Formulate a normal query, send, and retrieve answer in supplied buffer.
1664 * Return the size of the response on success, -1 on error.
1665 * If enabled, implement search rules until answer or unrecoverable failure
1666 * is detected. Error code, if any, is left in *herrno.
1667 */
res_searchN(const char * name,res_target * target,res_state res,int * herrno)1668 static int res_searchN(const char* name, res_target* target, res_state res, int* herrno) {
1669 const char *cp, *const *domain;
1670 HEADER* hp;
1671 u_int dots;
1672 int trailing_dot, ret, saved_herrno;
1673 int got_nodata = 0, got_servfail = 0, tried_as_is = 0;
1674
1675 assert(name != NULL);
1676 assert(target != NULL);
1677
1678 hp = (HEADER*) (void*) target->answer; /*XXX*/
1679
1680 errno = 0;
1681 *herrno = HOST_NOT_FOUND; /* default, if we never query */
1682 dots = 0;
1683 for (cp = name; *cp; cp++) dots += (*cp == '.');
1684 trailing_dot = 0;
1685 if (cp > name && *--cp == '.') trailing_dot++;
1686
1687 /*
1688 * If there are dots in the name already, let's just give it a try
1689 * 'as is'. The threshold can be set with the "ndots" option.
1690 */
1691 saved_herrno = -1;
1692 if (dots >= res->ndots) {
1693 ret = res_querydomainN(name, NULL, target, res, herrno);
1694 if (ret > 0) return (ret);
1695 saved_herrno = *herrno;
1696 tried_as_is++;
1697 }
1698
1699 /*
1700 * We do at least one level of search if
1701 * - there is no dot and RES_DEFNAME is set, or
1702 * - there is at least one dot, there is no trailing dot,
1703 * and RES_DNSRCH is set.
1704 */
1705 if ((!dots && (res->options & RES_DEFNAMES)) ||
1706 (dots && !trailing_dot && (res->options & RES_DNSRCH))) {
1707 int done = 0;
1708
1709 /* Unfortunately we need to set stuff up before
1710 * the domain stuff is tried. Will have a better
1711 * fix after thread pools are used.
1712 */
1713 _resolv_populate_res_for_net(res);
1714
1715 for (domain = (const char* const*) res->dnsrch; *domain && !done; domain++) {
1716 ret = res_querydomainN(name, *domain, target, res, herrno);
1717 if (ret > 0) return ret;
1718
1719 /*
1720 * If no server present, give up.
1721 * If name isn't found in this domain,
1722 * keep trying higher domains in the search list
1723 * (if that's enabled).
1724 * On a NO_DATA error, keep trying, otherwise
1725 * a wildcard entry of another type could keep us
1726 * from finding this entry higher in the domain.
1727 * If we get some other error (negative answer or
1728 * server failure), then stop searching up,
1729 * but try the input name below in case it's
1730 * fully-qualified.
1731 */
1732 if (errno == ECONNREFUSED) {
1733 *herrno = TRY_AGAIN;
1734 return -1;
1735 }
1736
1737 switch (*herrno) {
1738 case NO_DATA:
1739 got_nodata++;
1740 [[fallthrough]];
1741 case HOST_NOT_FOUND:
1742 /* keep trying */
1743 break;
1744 case TRY_AGAIN:
1745 if (hp->rcode == SERVFAIL) {
1746 /* try next search element, if any */
1747 got_servfail++;
1748 break;
1749 }
1750 [[fallthrough]];
1751 default:
1752 /* anything else implies that we're done */
1753 done++;
1754 }
1755 /*
1756 * if we got here for some reason other than DNSRCH,
1757 * we only wanted one iteration of the loop, so stop.
1758 */
1759 if (!(res->options & RES_DNSRCH)) done++;
1760 }
1761 }
1762
1763 /*
1764 * if we have not already tried the name "as is", do that now.
1765 * note that we do this regardless of how many dots were in the
1766 * name or whether it ends with a dot.
1767 */
1768 if (!tried_as_is) {
1769 ret = res_querydomainN(name, NULL, target, res, herrno);
1770 if (ret > 0) return ret;
1771 }
1772
1773 /*
1774 * if we got here, we didn't satisfy the search.
1775 * if we did an initial full query, return that query's h_errno
1776 * (note that we wouldn't be here if that query had succeeded).
1777 * else if we ever got a nodata, send that back as the reason.
1778 * else send back meaningless h_errno, that being the one from
1779 * the last DNSRCH we did.
1780 */
1781 if (saved_herrno != -1)
1782 *herrno = saved_herrno;
1783 else if (got_nodata)
1784 *herrno = NO_DATA;
1785 else if (got_servfail)
1786 *herrno = TRY_AGAIN;
1787 return -1;
1788 }
1789
1790 /*
1791 * Perform a call on res_query on the concatenation of name and domain,
1792 * removing a trailing dot from name if domain is NULL.
1793 */
res_querydomainN(const char * name,const char * domain,res_target * target,res_state res,int * herrno)1794 static int res_querydomainN(const char* name, const char* domain, res_target* target, res_state res,
1795 int* herrno) {
1796 char nbuf[MAXDNAME];
1797 const char* longname = nbuf;
1798 size_t n, d;
1799
1800 assert(name != NULL);
1801
1802 if (domain == NULL) {
1803 // Check for trailing '.'; copy without '.' if present.
1804 n = strlen(name);
1805 if (n + 1 > sizeof(nbuf)) {
1806 *herrno = NO_RECOVERY;
1807 return -1;
1808 }
1809 if (n > 0 && name[--n] == '.') {
1810 strncpy(nbuf, name, n);
1811 nbuf[n] = '\0';
1812 } else
1813 longname = name;
1814 } else {
1815 n = strlen(name);
1816 d = strlen(domain);
1817 if (n + 1 + d + 1 > sizeof(nbuf)) {
1818 *herrno = NO_RECOVERY;
1819 return -1;
1820 }
1821 snprintf(nbuf, sizeof(nbuf), "%s.%s", name, domain);
1822 }
1823 return res_queryN(longname, target, res, herrno);
1824 }
1825