1 /* $NetBSD: res_send.c,v 1.9 2006/01/24 17:41:25 christos Exp $ */
2
3 /*
4 * Copyright (c) 1985, 1989, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. All advertising materials mentioning features or use of this software
16 * must display the following acknowledgement:
17 * This product includes software developed by the University of
18 * California, Berkeley and its contributors.
19 * 4. Neither the name of the University nor the names of its contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * SUCH DAMAGE.
34 */
35
36 /*
37 * Portions Copyright (c) 1993 by Digital Equipment Corporation.
38 *
39 * Permission to use, copy, modify, and distribute this software for any
40 * purpose with or without fee is hereby granted, provided that the above
41 * copyright notice and this permission notice appear in all copies, and that
42 * the name of Digital Equipment Corporation not be used in advertising or
43 * publicity pertaining to distribution of the document or software without
44 * specific, written prior permission.
45 *
46 * THE SOFTWARE IS PROVIDED "AS IS" AND DIGITAL EQUIPMENT CORP. DISCLAIMS ALL
47 * WARRANTIES WITH REGARD TO THIS SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES
48 * OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL DIGITAL EQUIPMENT
49 * CORPORATION BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL
50 * DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR
51 * PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS
52 * ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS
53 * SOFTWARE.
54 */
55
56 /*
57 * Copyright (c) 2004 by Internet Systems Consortium, Inc. ("ISC")
58 * Portions Copyright (c) 1996-1999 by Internet Software Consortium.
59 *
60 * Permission to use, copy, modify, and distribute this software for any
61 * purpose with or without fee is hereby granted, provided that the above
62 * copyright notice and this permission notice appear in all copies.
63 *
64 * THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES
65 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
66 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL ISC BE LIABLE FOR
67 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
68 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
69 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT
70 * OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
71 */
72
73 /*
74 * Send query to name server and wait for reply.
75 */
76
77 #define LOG_TAG "resolv"
78
79 #include <chrono>
80
81 #include <sys/param.h>
82 #include <sys/socket.h>
83 #include <sys/time.h>
84 #include <sys/uio.h>
85
86 #include <arpa/inet.h>
87 #include <arpa/nameser.h>
88
89 #include <errno.h>
90 #include <fcntl.h>
91 #include <netdb.h>
92 #include <poll.h>
93 #include <signal.h>
94 #include <stdlib.h>
95 #include <string.h>
96 #include <time.h>
97 #include <unistd.h>
98
99 #include <android-base/logging.h>
100 #include <android-base/result.h>
101 #include <android/multinetwork.h> // ResNsendFlags
102
103 #include <netdutils/Slice.h>
104 #include <netdutils/Stopwatch.h>
105 #include "DnsTlsDispatcher.h"
106 #include "DnsTlsTransport.h"
107 #include "Experiments.h"
108 #include "PrivateDnsConfiguration.h"
109 #include "netd_resolv/resolv.h"
110 #include "private/android_filesystem_config.h"
111
112 #include "res_comp.h"
113 #include "res_debug.h"
114 #include "res_init.h"
115 #include "resolv_cache.h"
116 #include "stats.h"
117 #include "stats.pb.h"
118 #include "util.h"
119
120 using namespace std::chrono_literals;
121 // TODO: use the namespace something like android::netd_resolv for libnetd_resolv
122 using android::base::ErrnoError;
123 using android::base::Result;
124 using android::net::CacheStatus;
125 using android::net::DnsQueryEvent;
126 using android::net::DnsTlsDispatcher;
127 using android::net::DnsTlsTransport;
128 using android::net::gPrivateDnsConfiguration;
129 using android::net::IpVersion;
130 using android::net::IV_IPV4;
131 using android::net::IV_IPV6;
132 using android::net::IV_UNKNOWN;
133 using android::net::LinuxErrno;
134 using android::net::NetworkDnsEventReported;
135 using android::net::NS_T_INVALID;
136 using android::net::NsRcode;
137 using android::net::NsType;
138 using android::net::PrivateDnsMode;
139 using android::net::PrivateDnsModes;
140 using android::net::PrivateDnsStatus;
141 using android::net::PROTO_TCP;
142 using android::net::PROTO_UDP;
143 using android::netdutils::IPSockAddr;
144 using android::netdutils::Slice;
145 using android::netdutils::Stopwatch;
146
147 static DnsTlsDispatcher sDnsTlsDispatcher;
148
149 static int send_vc(res_state statp, res_params* params, const uint8_t* buf, int buflen,
150 uint8_t* ans, int anssiz, int* terrno, size_t ns, time_t* at, int* rcode,
151 int* delay);
152 static int send_dg(res_state statp, res_params* params, const uint8_t* buf, int buflen,
153 uint8_t* ans, int anssiz, int* terrno, size_t* ns, int* v_circuit,
154 int* gotsomewhere, time_t* at, int* rcode, int* delay);
155
156 static void dump_error(const char*, const struct sockaddr*, int);
157
158 static int sock_eq(struct sockaddr*, struct sockaddr*);
159 static int connect_with_timeout(int sock, const struct sockaddr* nsap, socklen_t salen,
160 const struct timespec timeout);
161 static int retrying_poll(const int sock, short events, const struct timespec* finish);
162 static int res_tls_send(res_state, const Slice query, const Slice answer, int* rcode,
163 bool* fallback);
164
getQueryType(const uint8_t * msg,size_t msgLen)165 NsType getQueryType(const uint8_t* msg, size_t msgLen) {
166 ns_msg handle;
167 ns_rr rr;
168 if (ns_initparse((const uint8_t*)msg, msgLen, &handle) < 0 ||
169 ns_parserr(&handle, ns_s_qd, 0, &rr) < 0) {
170 return NS_T_INVALID;
171 }
172 return static_cast<NsType>(ns_rr_type(rr));
173 }
174
ipFamilyToIPVersion(const int ipFamily)175 IpVersion ipFamilyToIPVersion(const int ipFamily) {
176 switch (ipFamily) {
177 case AF_INET:
178 return IV_IPV4;
179 case AF_INET6:
180 return IV_IPV6;
181 default:
182 return IV_UNKNOWN;
183 }
184 }
185
186 // BEGIN: Code copied from ISC eventlib
187 // TODO: move away from this code
188 #define BILLION 1000000000
189
evConsTime(time_t sec,long nsec)190 static struct timespec evConsTime(time_t sec, long nsec) {
191 struct timespec x;
192
193 x.tv_sec = sec;
194 x.tv_nsec = nsec;
195 return (x);
196 }
197
evAddTime(struct timespec addend1,struct timespec addend2)198 static struct timespec evAddTime(struct timespec addend1, struct timespec addend2) {
199 struct timespec x;
200
201 x.tv_sec = addend1.tv_sec + addend2.tv_sec;
202 x.tv_nsec = addend1.tv_nsec + addend2.tv_nsec;
203 if (x.tv_nsec >= BILLION) {
204 x.tv_sec++;
205 x.tv_nsec -= BILLION;
206 }
207 return (x);
208 }
209
evSubTime(struct timespec minuend,struct timespec subtrahend)210 static struct timespec evSubTime(struct timespec minuend, struct timespec subtrahend) {
211 struct timespec x;
212
213 x.tv_sec = minuend.tv_sec - subtrahend.tv_sec;
214 if (minuend.tv_nsec >= subtrahend.tv_nsec)
215 x.tv_nsec = minuend.tv_nsec - subtrahend.tv_nsec;
216 else {
217 x.tv_nsec = BILLION - subtrahend.tv_nsec + minuend.tv_nsec;
218 x.tv_sec--;
219 }
220 return (x);
221 }
222
evCmpTime(struct timespec a,struct timespec b)223 static int evCmpTime(struct timespec a, struct timespec b) {
224 #define SGN(x) ((x) < 0 ? (-1) : (x) > 0 ? (1) : (0));
225 time_t s = a.tv_sec - b.tv_sec;
226 long n;
227
228 if (s != 0) return SGN(s);
229
230 n = a.tv_nsec - b.tv_nsec;
231 return SGN(n);
232 }
233
evNowTime(void)234 static struct timespec evNowTime(void) {
235 struct timespec tsnow;
236 clock_gettime(CLOCK_REALTIME, &tsnow);
237 return tsnow;
238 }
239
240 // END: Code copied from ISC eventlib
241
242 /* BIONIC-BEGIN: implement source port randomization */
random_bind(int s,int family)243 static int random_bind(int s, int family) {
244 sockaddr_union u;
245 int j;
246 socklen_t slen;
247
248 /* clear all, this also sets the IP4/6 address to 'any' */
249 memset(&u, 0, sizeof u);
250
251 switch (family) {
252 case AF_INET:
253 u.sin.sin_family = family;
254 slen = sizeof u.sin;
255 break;
256 case AF_INET6:
257 u.sin6.sin6_family = family;
258 slen = sizeof u.sin6;
259 break;
260 default:
261 errno = EPROTO;
262 return -1;
263 }
264
265 /* first try to bind to a random source port a few times */
266 for (j = 0; j < 10; j++) {
267 /* find a random port between 1025 .. 65534 */
268 int port = 1025 + (arc4random_uniform(65535 - 1025));
269 if (family == AF_INET)
270 u.sin.sin_port = htons(port);
271 else
272 u.sin6.sin6_port = htons(port);
273
274 if (!bind(s, &u.sa, slen)) return 0;
275 }
276
277 // nothing after 10 attempts, our network table is probably busy
278 // let the system decide which port is best
279 if (family == AF_INET)
280 u.sin.sin_port = 0;
281 else
282 u.sin6.sin6_port = 0;
283
284 return bind(s, &u.sa, slen);
285 }
286 /* BIONIC-END */
287
288 // Disables all nameservers other than selectedServer
res_set_usable_server(int selectedServer,int nscount,bool usable_servers[])289 static void res_set_usable_server(int selectedServer, int nscount, bool usable_servers[]) {
290 int usableIndex = 0;
291 for (int ns = 0; ns < nscount; ns++) {
292 if (usable_servers[ns]) ++usableIndex;
293 if (usableIndex != selectedServer) usable_servers[ns] = false;
294 }
295 }
296
297 // Looks up the nameserver address in res.nsaddrs[], returns the ns number if found, otherwise -1.
res_ourserver_p(res_state statp,const sockaddr * sa)298 static int res_ourserver_p(res_state statp, const sockaddr* sa) {
299 const sockaddr_in *inp, *srv;
300 const sockaddr_in6 *in6p, *srv6;
301 int ns = 0;
302 switch (sa->sa_family) {
303 case AF_INET:
304 inp = (const struct sockaddr_in*) (const void*) sa;
305
306 for (const IPSockAddr& ipsa : statp->nsaddrs) {
307 sockaddr_storage ss = ipsa;
308 srv = reinterpret_cast<sockaddr_in*>(&ss);
309 if (srv->sin_family == inp->sin_family && srv->sin_port == inp->sin_port &&
310 (srv->sin_addr.s_addr == INADDR_ANY ||
311 srv->sin_addr.s_addr == inp->sin_addr.s_addr))
312 return ns;
313 ++ns;
314 }
315 break;
316 case AF_INET6:
317 in6p = (const struct sockaddr_in6*) (const void*) sa;
318 for (const IPSockAddr& ipsa : statp->nsaddrs) {
319 sockaddr_storage ss = ipsa;
320 srv6 = reinterpret_cast<sockaddr_in6*>(&ss);
321 if (srv6->sin6_family == in6p->sin6_family && srv6->sin6_port == in6p->sin6_port &&
322 #ifdef HAVE_SIN6_SCOPE_ID
323 (srv6->sin6_scope_id == 0 || srv6->sin6_scope_id == in6p->sin6_scope_id) &&
324 #endif
325 (IN6_IS_ADDR_UNSPECIFIED(&srv6->sin6_addr) ||
326 IN6_ARE_ADDR_EQUAL(&srv6->sin6_addr, &in6p->sin6_addr)))
327 return ns;
328 ++ns;
329 }
330 break;
331 default:
332 break;
333 }
334 return -1;
335 }
336
337 /* int
338 * res_nameinquery(name, type, cl, buf, eom)
339 * look for (name, type, cl) in the query section of packet (buf, eom)
340 * requires:
341 * buf + HFIXEDSZ <= eom
342 * returns:
343 * -1 : format error
344 * 0 : not found
345 * >0 : found
346 * author:
347 * paul vixie, 29may94
348 */
res_nameinquery(const char * name,int type,int cl,const uint8_t * buf,const uint8_t * eom)349 int res_nameinquery(const char* name, int type, int cl, const uint8_t* buf, const uint8_t* eom) {
350 const uint8_t* cp = buf + HFIXEDSZ;
351 int qdcount = ntohs(((const HEADER*) (const void*) buf)->qdcount);
352
353 while (qdcount-- > 0) {
354 char tname[MAXDNAME + 1];
355 int n = dn_expand(buf, eom, cp, tname, sizeof tname);
356 if (n < 0) return (-1);
357 cp += n;
358 if (cp + 2 * INT16SZ > eom) return (-1);
359 int ttype = ntohs(*reinterpret_cast<const uint16_t*>(cp));
360 cp += INT16SZ;
361 int tclass = ntohs(*reinterpret_cast<const uint16_t*>(cp));
362 cp += INT16SZ;
363 if (ttype == type && tclass == cl && ns_samename(tname, name) == 1) return (1);
364 }
365 return (0);
366 }
367
368 /* int
369 * res_queriesmatch(buf1, eom1, buf2, eom2)
370 * is there a 1:1 mapping of (name,type,class)
371 * in (buf1,eom1) and (buf2,eom2)?
372 * returns:
373 * -1 : format error
374 * 0 : not a 1:1 mapping
375 * >0 : is a 1:1 mapping
376 * author:
377 * paul vixie, 29may94
378 */
res_queriesmatch(const uint8_t * buf1,const uint8_t * eom1,const uint8_t * buf2,const uint8_t * eom2)379 int res_queriesmatch(const uint8_t* buf1, const uint8_t* eom1, const uint8_t* buf2,
380 const uint8_t* eom2) {
381 const uint8_t* cp = buf1 + HFIXEDSZ;
382 int qdcount = ntohs(((const HEADER*) (const void*) buf1)->qdcount);
383
384 if (buf1 + HFIXEDSZ > eom1 || buf2 + HFIXEDSZ > eom2) return (-1);
385
386 /*
387 * Only header section present in replies to
388 * dynamic update packets.
389 */
390 if ((((const HEADER*) (const void*) buf1)->opcode == ns_o_update) &&
391 (((const HEADER*) (const void*) buf2)->opcode == ns_o_update))
392 return (1);
393
394 if (qdcount != ntohs(((const HEADER*) (const void*) buf2)->qdcount)) return (0);
395 while (qdcount-- > 0) {
396 char tname[MAXDNAME + 1];
397 int n = dn_expand(buf1, eom1, cp, tname, sizeof tname);
398 if (n < 0) return (-1);
399 cp += n;
400 if (cp + 2 * INT16SZ > eom1) return (-1);
401 int ttype = ntohs(*reinterpret_cast<const uint16_t*>(cp));
402 cp += INT16SZ;
403 int tclass = ntohs(*reinterpret_cast<const uint16_t*>(cp));
404 cp += INT16SZ;
405 if (!res_nameinquery(tname, ttype, tclass, buf2, eom2)) return (0);
406 }
407 return (1);
408 }
409
addDnsQueryEvent(NetworkDnsEventReported * event)410 static DnsQueryEvent* addDnsQueryEvent(NetworkDnsEventReported* event) {
411 return event->mutable_dns_query_events()->add_dns_query_event();
412 }
413
isNetworkRestricted(int terrno)414 static bool isNetworkRestricted(int terrno) {
415 // It's possible that system was in some network restricted mode, which blocked
416 // the operation of sending packet and resulted in EPERM errno.
417 // It would be no reason to keep retrying on that case.
418 // TODO: Check the system status to know if network restricted mode is
419 // enabled.
420 return (terrno == EPERM);
421 }
422
res_nsend(res_state statp,const uint8_t * buf,int buflen,uint8_t * ans,int anssiz,int * rcode,uint32_t flags,std::chrono::milliseconds sleepTimeMs)423 int res_nsend(res_state statp, const uint8_t* buf, int buflen, uint8_t* ans, int anssiz, int* rcode,
424 uint32_t flags, std::chrono::milliseconds sleepTimeMs) {
425 LOG(DEBUG) << __func__;
426
427 // Should not happen
428 if (anssiz < HFIXEDSZ) {
429 // TODO: Remove errno once callers stop using it
430 errno = EINVAL;
431 return -EINVAL;
432 }
433 res_pquery(buf, buflen);
434
435 int anslen = 0;
436 Stopwatch cacheStopwatch;
437 ResolvCacheStatus cache_status =
438 resolv_cache_lookup(statp->netid, buf, buflen, ans, anssiz, &anslen, flags);
439 const int32_t cacheLatencyUs = saturate_cast<int32_t>(cacheStopwatch.timeTakenUs());
440 if (cache_status == RESOLV_CACHE_FOUND) {
441 HEADER* hp = (HEADER*)(void*)ans;
442 *rcode = hp->rcode;
443 DnsQueryEvent* dnsQueryEvent = addDnsQueryEvent(statp->event);
444 dnsQueryEvent->set_latency_micros(cacheLatencyUs);
445 dnsQueryEvent->set_cache_hit(static_cast<CacheStatus>(cache_status));
446 dnsQueryEvent->set_type(getQueryType(buf, buflen));
447 return anslen;
448 } else if (cache_status != RESOLV_CACHE_UNSUPPORTED) {
449 // had a cache miss for a known network, so populate the thread private
450 // data so the normal resolve path can do its thing
451 resolv_populate_res_for_net(statp);
452 }
453 if (statp->nameserverCount() == 0) {
454 // We have no nameservers configured, so there's no point trying.
455 // Tell the cache the query failed, or any retries and anyone else asking the same
456 // question will block for PENDING_REQUEST_TIMEOUT seconds instead of failing fast.
457 _resolv_cache_query_failed(statp->netid, buf, buflen, flags);
458
459 // TODO: Remove errno once callers stop using it
460 errno = ESRCH;
461 return -ESRCH;
462 }
463
464 // If parallel_lookup is enabled, it might be required to wait some time to avoid
465 // gateways drop packets if queries are sent too close together
466 if (sleepTimeMs != 0ms) {
467 std::this_thread::sleep_for(sleepTimeMs);
468 }
469 // DoT
470 if (!(statp->netcontext_flags & NET_CONTEXT_FLAG_USE_LOCAL_NAMESERVERS)) {
471 bool fallback = false;
472 int resplen = res_tls_send(statp, Slice(const_cast<uint8_t*>(buf), buflen),
473 Slice(ans, anssiz), rcode, &fallback);
474 if (resplen > 0) {
475 LOG(DEBUG) << __func__ << ": got answer from DoT";
476 res_pquery(ans, resplen);
477 if (cache_status == RESOLV_CACHE_NOTFOUND) {
478 resolv_cache_add(statp->netid, buf, buflen, ans, resplen);
479 }
480 return resplen;
481 }
482 if (!fallback) {
483 _resolv_cache_query_failed(statp->netid, buf, buflen, flags);
484 return -ETIMEDOUT;
485 }
486 }
487
488 res_stats stats[MAXNS]{};
489 res_params params;
490 int revision_id = resolv_cache_get_resolver_stats(statp->netid, ¶ms, stats, statp->nsaddrs);
491 if (revision_id < 0) {
492 // TODO: Remove errno once callers stop using it
493 errno = ESRCH;
494 return -ESRCH;
495 }
496 bool usable_servers[MAXNS];
497 int usableServersCount = android_net_res_stats_get_usable_servers(
498 ¶ms, stats, statp->nameserverCount(), usable_servers);
499
500 if (statp->sort_nameservers) {
501 // It's unnecessary to mark a DNS server as unusable since broken servers will be less
502 // likely to be chosen.
503 for (int i = 0; i < statp->nameserverCount(); i++) {
504 usable_servers[i] = true;
505 }
506 }
507
508 // TODO: Let it always choose the first nameserver when sort_nameservers is enabled.
509 if ((flags & ANDROID_RESOLV_NO_RETRY) && usableServersCount > 1) {
510 auto hp = reinterpret_cast<const HEADER*>(buf);
511
512 // Select a random server based on the query id
513 int selectedServer = (hp->id % usableServersCount) + 1;
514 res_set_usable_server(selectedServer, statp->nameserverCount(), usable_servers);
515 }
516
517 // Send request, RETRY times, or until successful.
518 int retryTimes = (flags & ANDROID_RESOLV_NO_RETRY) ? 1 : params.retry_count;
519 int useTcp = buflen > PACKETSZ;
520 int gotsomewhere = 0;
521 // Use an impossible error code as default value
522 int terrno = ETIME;
523
524 for (int attempt = 0; attempt < retryTimes; ++attempt) {
525 for (size_t ns = 0; ns < statp->nsaddrs.size(); ++ns) {
526 if (!usable_servers[ns]) continue;
527
528 *rcode = RCODE_INTERNAL_ERROR;
529
530 // Get server addr
531 const IPSockAddr& serverSockAddr = statp->nsaddrs[ns];
532 LOG(DEBUG) << __func__ << ": Querying server (# " << ns + 1
533 << ") address = " << serverSockAddr.toString();
534
535 ::android::net::Protocol query_proto = useTcp ? PROTO_TCP : PROTO_UDP;
536 time_t query_time = 0;
537 int delay = 0;
538 bool fallbackTCP = false;
539 const bool shouldRecordStats = (attempt == 0);
540 int resplen;
541 Stopwatch queryStopwatch;
542 int retry_count_for_event = 0;
543 size_t actualNs = ns;
544 // Use an impossible error code as default value
545 terrno = ETIME;
546 if (useTcp) {
547 // TCP; at most one attempt per server.
548 attempt = retryTimes;
549 resplen = send_vc(statp, ¶ms, buf, buflen, ans, anssiz, &terrno, ns,
550 &query_time, rcode, &delay);
551
552 if (buflen <= PACKETSZ && resplen <= 0 &&
553 statp->tc_mode == aidl::android::net::IDnsResolver::TC_MODE_UDP_TCP) {
554 // reset to UDP for next query on next DNS server if resolver is currently doing
555 // TCP fallback retry and current server does not support TCP connectin
556 useTcp = false;
557 }
558 LOG(INFO) << __func__ << ": used send_vc " << resplen << " terrno: " << terrno;
559 } else {
560 // UDP
561 resplen = send_dg(statp, ¶ms, buf, buflen, ans, anssiz, &terrno, &actualNs,
562 &useTcp, &gotsomewhere, &query_time, rcode, &delay);
563 fallbackTCP = useTcp ? true : false;
564 retry_count_for_event = attempt;
565 LOG(INFO) << __func__ << ": used send_dg " << resplen << " terrno: " << terrno;
566 }
567
568 const IPSockAddr& receivedServerAddr = statp->nsaddrs[actualNs];
569 DnsQueryEvent* dnsQueryEvent = addDnsQueryEvent(statp->event);
570 dnsQueryEvent->set_cache_hit(static_cast<CacheStatus>(cache_status));
571 // When |retryTimes| > 1, we cannot actually know the correct latency value if we
572 // received the answer from the previous server. So temporarily set the latency as -1 if
573 // that condition happened.
574 // TODO: make the latency value accurate.
575 dnsQueryEvent->set_latency_micros(
576 (actualNs == ns) ? saturate_cast<int32_t>(queryStopwatch.timeTakenUs()) : -1);
577 dnsQueryEvent->set_dns_server_index(actualNs);
578 dnsQueryEvent->set_ip_version(ipFamilyToIPVersion(receivedServerAddr.family()));
579 dnsQueryEvent->set_retry_times(retry_count_for_event);
580 dnsQueryEvent->set_rcode(static_cast<NsRcode>(*rcode));
581 dnsQueryEvent->set_protocol(query_proto);
582 dnsQueryEvent->set_type(getQueryType(buf, buflen));
583 dnsQueryEvent->set_linux_errno(static_cast<LinuxErrno>(terrno));
584
585 // Only record stats the first time we try a query. This ensures that
586 // queries that deterministically fail (e.g., a name that always returns
587 // SERVFAIL or times out) do not unduly affect the stats.
588 if (shouldRecordStats) {
589 // (b/151166599): This is a workaround to prevent that DnsResolver calculates the
590 // reliability of DNS servers from being broken when network restricted mode is
591 // enabled.
592 // TODO: Introduce the new server selection instead of skipping stats recording.
593 if (!isNetworkRestricted(terrno)) {
594 res_sample sample;
595 res_stats_set_sample(&sample, query_time, *rcode, delay);
596 // KeepListening UDP mechanism is incompatible with usable_servers of legacy
597 // stats, so keep the old logic for now.
598 // TODO: Replace usable_servers of legacy stats with new one.
599 resolv_cache_add_resolver_stats_sample(
600 statp->netid, revision_id, serverSockAddr, sample, params.max_samples);
601 }
602 resolv_stats_add(statp->netid, receivedServerAddr, dnsQueryEvent);
603 }
604
605 if (resplen == 0) continue;
606 if (fallbackTCP) {
607 ns--;
608 continue;
609 }
610 if (resplen < 0) {
611 _resolv_cache_query_failed(statp->netid, buf, buflen, flags);
612 statp->closeSockets();
613 return -terrno;
614 };
615
616 LOG(DEBUG) << __func__ << ": got answer:";
617 res_pquery(ans, (resplen > anssiz) ? anssiz : resplen);
618
619 if (cache_status == RESOLV_CACHE_NOTFOUND) {
620 resolv_cache_add(statp->netid, buf, buflen, ans, resplen);
621 }
622 statp->closeSockets();
623 return (resplen);
624 } // for each ns
625 } // for each retry
626 statp->closeSockets();
627 terrno = useTcp ? terrno : gotsomewhere ? ETIMEDOUT : ECONNREFUSED;
628 // TODO: Remove errno once callers stop using it
629 errno = useTcp ? terrno
630 : gotsomewhere ? ETIMEDOUT /* no answer obtained */
631 : ECONNREFUSED /* no nameservers found */;
632
633 _resolv_cache_query_failed(statp->netid, buf, buflen, flags);
634 return -terrno;
635 }
636
get_timeout(res_state statp,const res_params * params,const int ns)637 static struct timespec get_timeout(res_state statp, const res_params* params, const int ns) {
638 int msec;
639 // Legacy algorithm which scales the timeout by nameserver number.
640 // For instance, with 4 nameservers: 5s, 2.5s, 5s, 10s
641 // This has no effect with 1 or 2 nameservers
642 msec = params->base_timeout_msec << ns;
643 if (ns > 0) {
644 msec /= statp->nameserverCount();
645 }
646 // For safety, don't allow OEMs and experiments to configure a timeout shorter than 1s.
647 if (msec < 1000) {
648 msec = 1000; // Use at least 1000ms
649 }
650 LOG(INFO) << __func__ << ": using timeout of " << msec << " msec";
651
652 struct timespec result;
653 result.tv_sec = msec / 1000;
654 result.tv_nsec = (msec % 1000) * 1000000;
655 return result;
656 }
657
send_vc(res_state statp,res_params * params,const uint8_t * buf,int buflen,uint8_t * ans,int anssiz,int * terrno,size_t ns,time_t * at,int * rcode,int * delay)658 static int send_vc(res_state statp, res_params* params, const uint8_t* buf, int buflen,
659 uint8_t* ans, int anssiz, int* terrno, size_t ns, time_t* at, int* rcode,
660 int* delay) {
661 *at = time(NULL);
662 *delay = 0;
663 const HEADER* hp = (const HEADER*) (const void*) buf;
664 HEADER* anhp = (HEADER*) (void*) ans;
665 struct sockaddr* nsap;
666 int nsaplen;
667 int truncating, connreset, n;
668 uint8_t* cp;
669
670 LOG(INFO) << __func__ << ": using send_vc";
671
672 // It should never happen, but just in case.
673 if (ns >= statp->nsaddrs.size()) {
674 LOG(ERROR) << __func__ << ": Out-of-bound indexing: " << ns;
675 *terrno = EINVAL;
676 return -1;
677 }
678
679 sockaddr_storage ss = statp->nsaddrs[ns];
680 nsap = reinterpret_cast<sockaddr*>(&ss);
681 nsaplen = sockaddrSize(nsap);
682
683 connreset = 0;
684 same_ns:
685 truncating = 0;
686
687 struct timespec start_time = evNowTime();
688
689 /* Are we still talking to whom we want to talk to? */
690 if (statp->tcp_nssock >= 0 && (statp->_flags & RES_F_VC) != 0) {
691 struct sockaddr_storage peer;
692 socklen_t size = sizeof peer;
693 unsigned old_mark;
694 socklen_t mark_size = sizeof(old_mark);
695 if (getpeername(statp->tcp_nssock, (struct sockaddr*)(void*)&peer, &size) < 0 ||
696 !sock_eq((struct sockaddr*)(void*)&peer, nsap) ||
697 getsockopt(statp->tcp_nssock, SOL_SOCKET, SO_MARK, &old_mark, &mark_size) < 0 ||
698 old_mark != statp->_mark) {
699 statp->closeSockets();
700 }
701 }
702
703 if (statp->tcp_nssock < 0 || (statp->_flags & RES_F_VC) == 0) {
704 if (statp->tcp_nssock >= 0) statp->closeSockets();
705
706 statp->tcp_nssock.reset(socket(nsap->sa_family, SOCK_STREAM | SOCK_CLOEXEC, 0));
707 if (statp->tcp_nssock < 0) {
708 *terrno = errno;
709 PLOG(DEBUG) << __func__ << ": socket(vc): ";
710 switch (errno) {
711 case EPROTONOSUPPORT:
712 case EPFNOSUPPORT:
713 case EAFNOSUPPORT:
714 return 0;
715 default:
716 return -1;
717 }
718 }
719 const uid_t uid = statp->enforce_dns_uid ? AID_DNS : statp->uid;
720 resolv_tag_socket(statp->tcp_nssock, uid, statp->pid);
721 if (statp->_mark != MARK_UNSET) {
722 if (setsockopt(statp->tcp_nssock, SOL_SOCKET, SO_MARK, &statp->_mark,
723 sizeof(statp->_mark)) < 0) {
724 *terrno = errno;
725 PLOG(DEBUG) << __func__ << ": setsockopt: ";
726 return -1;
727 }
728 }
729 errno = 0;
730 if (random_bind(statp->tcp_nssock, nsap->sa_family) < 0) {
731 *terrno = errno;
732 dump_error("bind/vc", nsap, nsaplen);
733 statp->closeSockets();
734 return (0);
735 }
736 if (connect_with_timeout(statp->tcp_nssock, nsap, (socklen_t)nsaplen,
737 get_timeout(statp, params, ns)) < 0) {
738 *terrno = errno;
739 dump_error("connect/vc", nsap, nsaplen);
740 statp->closeSockets();
741 /*
742 * The way connect_with_timeout() is implemented prevents us from reliably
743 * determining whether this was really a timeout or e.g. ECONNREFUSED. Since
744 * currently both cases are handled in the same way, there is no need to
745 * change this (yet). If we ever need to reliably distinguish between these
746 * cases, both connect_with_timeout() and retrying_poll() need to be
747 * modified, though.
748 */
749 *rcode = RCODE_TIMEOUT;
750 return (0);
751 }
752 statp->_flags |= RES_F_VC;
753 }
754
755 /*
756 * Send length & message
757 */
758 uint16_t len = htons(static_cast<uint16_t>(buflen));
759 const iovec iov[] = {
760 {.iov_base = &len, .iov_len = INT16SZ},
761 {.iov_base = const_cast<uint8_t*>(buf), .iov_len = static_cast<size_t>(buflen)},
762 };
763 if (writev(statp->tcp_nssock, iov, 2) != (INT16SZ + buflen)) {
764 *terrno = errno;
765 PLOG(DEBUG) << __func__ << ": write failed: ";
766 statp->closeSockets();
767 return (0);
768 }
769 /*
770 * Receive length & response
771 */
772 read_len:
773 cp = ans;
774 len = INT16SZ;
775 while ((n = read(statp->tcp_nssock, (char*)cp, (size_t)len)) > 0) {
776 cp += n;
777 if ((len -= n) == 0) break;
778 }
779 if (n <= 0) {
780 *terrno = errno;
781 PLOG(DEBUG) << __func__ << ": read failed: ";
782 statp->closeSockets();
783 /*
784 * A long running process might get its TCP
785 * connection reset if the remote server was
786 * restarted. Requery the server instead of
787 * trying a new one. When there is only one
788 * server, this means that a query might work
789 * instead of failing. We only allow one reset
790 * per query to prevent looping.
791 */
792 if (*terrno == ECONNRESET && !connreset) {
793 connreset = 1;
794 goto same_ns;
795 }
796 return (0);
797 }
798 uint16_t resplen = ntohs(*reinterpret_cast<const uint16_t*>(ans));
799 if (resplen > anssiz) {
800 LOG(DEBUG) << __func__ << ": response truncated";
801 truncating = 1;
802 len = anssiz;
803 } else
804 len = resplen;
805 if (len < HFIXEDSZ) {
806 /*
807 * Undersized message.
808 */
809 LOG(DEBUG) << __func__ << ": undersized: " << len;
810 *terrno = EMSGSIZE;
811 statp->closeSockets();
812 return (0);
813 }
814 cp = ans;
815 while (len != 0 && (n = read(statp->tcp_nssock, (char*)cp, (size_t)len)) > 0) {
816 cp += n;
817 len -= n;
818 }
819 if (n <= 0) {
820 *terrno = errno;
821 PLOG(DEBUG) << __func__ << ": read(vc): ";
822 statp->closeSockets();
823 return (0);
824 }
825
826 if (truncating) {
827 /*
828 * Flush rest of answer so connection stays in synch.
829 */
830 anhp->tc = 1;
831 len = resplen - anssiz;
832 while (len != 0) {
833 char junk[PACKETSZ];
834
835 n = read(statp->tcp_nssock, junk, (len > sizeof junk) ? sizeof junk : len);
836 if (n > 0)
837 len -= n;
838 else
839 break;
840 }
841 }
842 /*
843 * If the calling application has bailed out of
844 * a previous call and failed to arrange to have
845 * the circuit closed or the server has got
846 * itself confused, then drop the packet and
847 * wait for the correct one.
848 */
849 if (hp->id != anhp->id) {
850 LOG(DEBUG) << __func__ << ": ld answer (unexpected):";
851 res_pquery(ans, (resplen > anssiz) ? anssiz : resplen);
852 goto read_len;
853 }
854
855 /*
856 * All is well, or the error is fatal. Signal that the
857 * next nameserver ought not be tried.
858 */
859 if (resplen > 0) {
860 struct timespec done = evNowTime();
861 *delay = res_stats_calculate_rtt(&done, &start_time);
862 *rcode = anhp->rcode;
863 }
864 *terrno = 0;
865 return (resplen);
866 }
867
868 /* return -1 on error (errno set), 0 on success */
connect_with_timeout(int sock,const sockaddr * nsap,socklen_t salen,const timespec timeout)869 static int connect_with_timeout(int sock, const sockaddr* nsap, socklen_t salen,
870 const timespec timeout) {
871 int res, origflags;
872
873 origflags = fcntl(sock, F_GETFL, 0);
874 fcntl(sock, F_SETFL, origflags | O_NONBLOCK);
875
876 res = connect(sock, nsap, salen);
877 if (res < 0 && errno != EINPROGRESS) {
878 res = -1;
879 goto done;
880 }
881 if (res != 0) {
882 timespec now = evNowTime();
883 timespec finish = evAddTime(now, timeout);
884 LOG(INFO) << __func__ << ": " << sock << " send_vc";
885 res = retrying_poll(sock, POLLIN | POLLOUT, &finish);
886 if (res <= 0) {
887 res = -1;
888 }
889 }
890 done:
891 fcntl(sock, F_SETFL, origflags);
892 LOG(INFO) << __func__ << ": " << sock << " connect_with_const timeout returning " << res;
893 return res;
894 }
895
retrying_poll(const int sock,const short events,const struct timespec * finish)896 static int retrying_poll(const int sock, const short events, const struct timespec* finish) {
897 struct timespec now, timeout;
898
899 retry:
900 LOG(INFO) << __func__ << ": " << sock << " retrying_poll";
901
902 now = evNowTime();
903 if (evCmpTime(*finish, now) > 0)
904 timeout = evSubTime(*finish, now);
905 else
906 timeout = evConsTime(0L, 0L);
907 struct pollfd fds = {.fd = sock, .events = events};
908 int n = ppoll(&fds, 1, &timeout, /*__mask=*/NULL);
909 if (n == 0) {
910 LOG(INFO) << __func__ << ": " << sock << " retrying_poll timeout";
911 errno = ETIMEDOUT;
912 return 0;
913 }
914 if (n < 0) {
915 if (errno == EINTR) goto retry;
916 PLOG(INFO) << __func__ << ": " << sock << " retrying_poll failed";
917 return n;
918 }
919 if (fds.revents & (POLLIN | POLLOUT | POLLERR)) {
920 int error;
921 socklen_t len = sizeof(error);
922 if (getsockopt(sock, SOL_SOCKET, SO_ERROR, &error, &len) < 0 || error) {
923 errno = error;
924 PLOG(INFO) << __func__ << ": " << sock << " retrying_poll getsockopt failed";
925 return -1;
926 }
927 }
928 LOG(INFO) << __func__ << ": " << sock << " retrying_poll returning " << n;
929 return n;
930 }
931
extractUdpFdset(res_state statp,const short events=POLLIN)932 static std::vector<pollfd> extractUdpFdset(res_state statp, const short events = POLLIN) {
933 std::vector<pollfd> fdset(statp->nsaddrs.size());
934 for (size_t i = 0; i < statp->nsaddrs.size(); ++i) {
935 fdset[i] = {.fd = statp->nssocks[i], .events = events};
936 }
937 return fdset;
938 }
939
udpRetryingPoll(res_state statp,const timespec * finish)940 static Result<std::vector<int>> udpRetryingPoll(res_state statp, const timespec* finish) {
941 for (;;) {
942 LOG(DEBUG) << __func__ << ": poll";
943 timespec start_time = evNowTime();
944 timespec timeout = (evCmpTime(*finish, start_time) > 0) ? evSubTime(*finish, start_time)
945 : evConsTime(0L, 0L);
946 std::vector<pollfd> fdset = extractUdpFdset(statp);
947 const int n = ppoll(fdset.data(), fdset.size(), &timeout, /*__mask=*/nullptr);
948 if (n <= 0) {
949 if (errno == EINTR && n < 0) continue;
950 if (n == 0) errno = ETIMEDOUT;
951 PLOG(INFO) << __func__ << ": failed";
952 return ErrnoError();
953 }
954 std::vector<int> fdsToRead;
955 for (const auto& pollfd : fdset) {
956 if (pollfd.revents & (POLLIN | POLLERR)) {
957 fdsToRead.push_back(pollfd.fd);
958 }
959 }
960 LOG(DEBUG) << __func__ << ": "
961 << " returning fd size: " << fdsToRead.size();
962 return fdsToRead;
963 }
964 }
965
udpRetryingPollWrapper(res_state statp,int ns,const timespec * finish)966 static Result<std::vector<int>> udpRetryingPollWrapper(res_state statp, int ns,
967 const timespec* finish) {
968 const bool keepListeningUdp =
969 android::net::Experiments::getInstance()->getFlag("keep_listening_udp", 0);
970 if (keepListeningUdp) return udpRetryingPoll(statp, finish);
971
972 if (int n = retrying_poll(statp->nssocks[ns], POLLIN, finish); n <= 0) {
973 return ErrnoError();
974 }
975 return std::vector<int>{statp->nssocks[ns]};
976 }
977
ignoreInvalidAnswer(res_state statp,const sockaddr_storage & from,const uint8_t * buf,int buflen,uint8_t * ans,int anssiz,int * receivedFromNs)978 bool ignoreInvalidAnswer(res_state statp, const sockaddr_storage& from, const uint8_t* buf,
979 int buflen, uint8_t* ans, int anssiz, int* receivedFromNs) {
980 const HEADER* hp = (const HEADER*)(const void*)buf;
981 HEADER* anhp = (HEADER*)(void*)ans;
982 if (hp->id != anhp->id) {
983 // response from old query, ignore it.
984 LOG(DEBUG) << __func__ << ": old answer:";
985 return true;
986 }
987 if (*receivedFromNs = res_ourserver_p(statp, (sockaddr*)(void*)&from); *receivedFromNs < 0) {
988 // response from wrong server? ignore it.
989 LOG(DEBUG) << __func__ << ": not our server:";
990 return true;
991 }
992 if (!res_queriesmatch(buf, buf + buflen, ans, ans + anssiz)) {
993 // response contains wrong query? ignore it.
994 LOG(DEBUG) << __func__ << ": wrong query name:";
995 return true;
996 }
997 return false;
998 }
999
send_dg(res_state statp,res_params * params,const uint8_t * buf,int buflen,uint8_t * ans,int anssiz,int * terrno,size_t * ns,int * v_circuit,int * gotsomewhere,time_t * at,int * rcode,int * delay)1000 static int send_dg(res_state statp, res_params* params, const uint8_t* buf, int buflen,
1001 uint8_t* ans, int anssiz, int* terrno, size_t* ns, int* v_circuit,
1002 int* gotsomewhere, time_t* at, int* rcode, int* delay) {
1003 // It should never happen, but just in case.
1004 if (*ns >= statp->nsaddrs.size()) {
1005 LOG(ERROR) << __func__ << ": Out-of-bound indexing: " << ns;
1006 *terrno = EINVAL;
1007 return -1;
1008 }
1009
1010 *at = time(nullptr);
1011 *delay = 0;
1012 const sockaddr_storage ss = statp->nsaddrs[*ns];
1013 const sockaddr* nsap = reinterpret_cast<const sockaddr*>(&ss);
1014 const int nsaplen = sockaddrSize(nsap);
1015
1016 if (statp->nssocks[*ns] == -1) {
1017 statp->nssocks[*ns].reset(socket(nsap->sa_family, SOCK_DGRAM | SOCK_CLOEXEC, 0));
1018 if (statp->nssocks[*ns] < 0) {
1019 *terrno = errno;
1020 PLOG(DEBUG) << __func__ << ": socket(dg): ";
1021 switch (errno) {
1022 case EPROTONOSUPPORT:
1023 case EPFNOSUPPORT:
1024 case EAFNOSUPPORT:
1025 return (0);
1026 default:
1027 return (-1);
1028 }
1029 }
1030
1031 const uid_t uid = statp->enforce_dns_uid ? AID_DNS : statp->uid;
1032 resolv_tag_socket(statp->nssocks[*ns], uid, statp->pid);
1033 if (statp->_mark != MARK_UNSET) {
1034 if (setsockopt(statp->nssocks[*ns], SOL_SOCKET, SO_MARK, &(statp->_mark),
1035 sizeof(statp->_mark)) < 0) {
1036 *terrno = errno;
1037 statp->closeSockets();
1038 return -1;
1039 }
1040 }
1041 // Use a "connected" datagram socket to receive an ECONNREFUSED error
1042 // on the next socket operation when the server responds with an
1043 // ICMP port-unreachable error. This way we can detect the absence of
1044 // a nameserver without timing out.
1045 if (random_bind(statp->nssocks[*ns], nsap->sa_family) < 0) {
1046 *terrno = errno;
1047 dump_error("bind(dg)", nsap, nsaplen);
1048 statp->closeSockets();
1049 return (0);
1050 }
1051 if (connect(statp->nssocks[*ns], nsap, (socklen_t)nsaplen) < 0) {
1052 *terrno = errno;
1053 dump_error("connect(dg)", nsap, nsaplen);
1054 statp->closeSockets();
1055 return (0);
1056 }
1057 LOG(DEBUG) << __func__ << ": new DG socket";
1058 }
1059 if (send(statp->nssocks[*ns], (const char*)buf, (size_t)buflen, 0) != buflen) {
1060 *terrno = errno;
1061 PLOG(DEBUG) << __func__ << ": send: ";
1062 statp->closeSockets();
1063 return 0;
1064 }
1065
1066 timespec timeout = get_timeout(statp, params, *ns);
1067 timespec start_time = evNowTime();
1068 timespec finish = evAddTime(start_time, timeout);
1069 for (;;) {
1070 // Wait for reply.
1071 auto result = udpRetryingPollWrapper(statp, *ns, &finish);
1072
1073 if (!result.has_value()) {
1074 const bool isTimeout = (result.error().code() == ETIMEDOUT);
1075 *rcode = (isTimeout) ? RCODE_TIMEOUT : *rcode;
1076 *terrno = (isTimeout) ? ETIMEDOUT : errno;
1077 *gotsomewhere = (isTimeout) ? 1 : *gotsomewhere;
1078 // Leave the UDP sockets open on timeout so we can keep listening for
1079 // a late response from this server while retrying on the next server.
1080 if (!isTimeout) statp->closeSockets();
1081 LOG(DEBUG) << __func__ << ": " << (isTimeout) ? "timeout" : "poll";
1082 return 0;
1083 }
1084 bool needRetry = false;
1085 for (int fd : result.value()) {
1086 needRetry = false;
1087 sockaddr_storage from;
1088 socklen_t fromlen = sizeof(from);
1089 int resplen =
1090 recvfrom(fd, (char*)ans, (size_t)anssiz, 0, (sockaddr*)(void*)&from, &fromlen);
1091 if (resplen <= 0) {
1092 *terrno = errno;
1093 PLOG(DEBUG) << __func__ << ": recvfrom: ";
1094 continue;
1095 }
1096 *gotsomewhere = 1;
1097 if (resplen < HFIXEDSZ) {
1098 // Undersized message.
1099 LOG(DEBUG) << __func__ << ": undersized: " << resplen;
1100 *terrno = EMSGSIZE;
1101 continue;
1102 }
1103
1104 int receivedFromNs = *ns;
1105 if (needRetry =
1106 ignoreInvalidAnswer(statp, from, buf, buflen, ans, anssiz, &receivedFromNs);
1107 needRetry) {
1108 res_pquery(ans, (resplen > anssiz) ? anssiz : resplen);
1109 continue;
1110 }
1111
1112 HEADER* anhp = (HEADER*)(void*)ans;
1113 if (anhp->rcode == FORMERR && (statp->netcontext_flags & NET_CONTEXT_FLAG_USE_EDNS)) {
1114 // Do not retry if the server do not understand EDNS0.
1115 // The case has to be captured here, as FORMERR packet do not
1116 // carry query section, hence res_queriesmatch() returns 0.
1117 LOG(DEBUG) << __func__ << ": server rejected query with EDNS0:";
1118 res_pquery(ans, (resplen > anssiz) ? anssiz : resplen);
1119 // record the error
1120 statp->_flags |= RES_F_EDNS0ERR;
1121 *terrno = EREMOTEIO;
1122 continue;
1123 }
1124
1125 timespec done = evNowTime();
1126 *delay = res_stats_calculate_rtt(&done, &start_time);
1127 if (anhp->rcode == SERVFAIL || anhp->rcode == NOTIMP || anhp->rcode == REFUSED) {
1128 LOG(DEBUG) << __func__ << ": server rejected query:";
1129 res_pquery(ans, (resplen > anssiz) ? anssiz : resplen);
1130 *rcode = anhp->rcode;
1131 continue;
1132 }
1133 if (anhp->tc) {
1134 // To get the rest of answer,
1135 // use TCP with same server.
1136 LOG(DEBUG) << __func__ << ": truncated answer";
1137 *terrno = E2BIG;
1138 *v_circuit = 1;
1139 return 1;
1140 }
1141 // All is well, or the error is fatal. Signal that the
1142 // next nameserver ought not be tried.
1143
1144 *rcode = anhp->rcode;
1145 *ns = receivedFromNs;
1146 *terrno = 0;
1147 return resplen;
1148 }
1149 if (!needRetry) return 0;
1150 }
1151 }
1152
dump_error(const char * str,const struct sockaddr * address,int alen)1153 static void dump_error(const char* str, const struct sockaddr* address, int alen) {
1154 char hbuf[NI_MAXHOST];
1155 char sbuf[NI_MAXSERV];
1156 constexpr int niflags = NI_NUMERICHOST | NI_NUMERICSERV;
1157 const int err = errno;
1158
1159 if (!WOULD_LOG(DEBUG)) return;
1160
1161 if (getnameinfo(address, (socklen_t)alen, hbuf, sizeof(hbuf), sbuf, sizeof(sbuf), niflags)) {
1162 strncpy(hbuf, "?", sizeof(hbuf) - 1);
1163 hbuf[sizeof(hbuf) - 1] = '\0';
1164 strncpy(sbuf, "?", sizeof(sbuf) - 1);
1165 sbuf[sizeof(sbuf) - 1] = '\0';
1166 }
1167 errno = err;
1168 PLOG(DEBUG) << __func__ << ": " << str << " ([" << hbuf << "]." << sbuf << "): ";
1169 }
1170
sock_eq(struct sockaddr * a,struct sockaddr * b)1171 static int sock_eq(struct sockaddr* a, struct sockaddr* b) {
1172 struct sockaddr_in *a4, *b4;
1173 struct sockaddr_in6 *a6, *b6;
1174
1175 if (a->sa_family != b->sa_family) return 0;
1176 switch (a->sa_family) {
1177 case AF_INET:
1178 a4 = (struct sockaddr_in*) (void*) a;
1179 b4 = (struct sockaddr_in*) (void*) b;
1180 return a4->sin_port == b4->sin_port && a4->sin_addr.s_addr == b4->sin_addr.s_addr;
1181 case AF_INET6:
1182 a6 = (struct sockaddr_in6*) (void*) a;
1183 b6 = (struct sockaddr_in6*) (void*) b;
1184 return a6->sin6_port == b6->sin6_port &&
1185 #ifdef HAVE_SIN6_SCOPE_ID
1186 a6->sin6_scope_id == b6->sin6_scope_id &&
1187 #endif
1188 IN6_ARE_ADDR_EQUAL(&a6->sin6_addr, &b6->sin6_addr);
1189 default:
1190 return 0;
1191 }
1192 }
1193
convertEnumType(PrivateDnsMode privateDnsmode)1194 PrivateDnsModes convertEnumType(PrivateDnsMode privateDnsmode) {
1195 switch (privateDnsmode) {
1196 case PrivateDnsMode::OFF:
1197 return PrivateDnsModes::PDM_OFF;
1198 case PrivateDnsMode::OPPORTUNISTIC:
1199 return PrivateDnsModes::PDM_OPPORTUNISTIC;
1200 case PrivateDnsMode::STRICT:
1201 return PrivateDnsModes::PDM_STRICT;
1202 default:
1203 return PrivateDnsModes::PDM_UNKNOWN;
1204 }
1205 }
1206
res_tls_send(res_state statp,const Slice query,const Slice answer,int * rcode,bool * fallback)1207 static int res_tls_send(res_state statp, const Slice query, const Slice answer, int* rcode,
1208 bool* fallback) {
1209 int resplen = 0;
1210 const unsigned netId = statp->netid;
1211
1212 PrivateDnsStatus privateDnsStatus = gPrivateDnsConfiguration.getStatus(netId);
1213 statp->event->set_private_dns_modes(convertEnumType(privateDnsStatus.mode));
1214
1215 if (privateDnsStatus.mode == PrivateDnsMode::OFF) {
1216 *fallback = true;
1217 return -1;
1218 }
1219
1220 if (privateDnsStatus.validatedServers().empty()) {
1221 if (privateDnsStatus.mode == PrivateDnsMode::OPPORTUNISTIC) {
1222 *fallback = true;
1223 return -1;
1224 } else {
1225 // Sleep and iterate some small number of times checking for the
1226 // arrival of resolved and validated server IP addresses, instead
1227 // of returning an immediate error.
1228 // This is needed because as soon as a network becomes the default network, apps will
1229 // send DNS queries on that network. If no servers have yet validated, and we do not
1230 // block those queries, they would immediately fail, causing application-visible errors.
1231 // Note that this can happen even before the network validates, since an unvalidated
1232 // network can become the default network if no validated networks are available.
1233 //
1234 // TODO: see if there is a better way to address this problem, such as buffering the
1235 // queries in a queue or only blocking queries for the first few seconds after a default
1236 // network change.
1237 for (int i = 0; i < 42; i++) {
1238 std::this_thread::sleep_for(std::chrono::milliseconds(100));
1239 // Calling getStatus() to merely check if there's any validated server seems
1240 // wasteful. Consider adding a new method in PrivateDnsConfiguration for speed ups.
1241 if (!gPrivateDnsConfiguration.getStatus(netId).validatedServers().empty()) {
1242 privateDnsStatus = gPrivateDnsConfiguration.getStatus(netId);
1243 break;
1244 }
1245 }
1246 if (privateDnsStatus.validatedServers().empty()) {
1247 return -1;
1248 }
1249 }
1250 }
1251
1252 LOG(INFO) << __func__ << ": performing query over TLS";
1253
1254 const auto response = sDnsTlsDispatcher.query(privateDnsStatus.validatedServers(), statp, query,
1255 answer, &resplen);
1256
1257 LOG(INFO) << __func__ << ": TLS query result: " << static_cast<int>(response);
1258
1259 if (privateDnsStatus.mode == PrivateDnsMode::OPPORTUNISTIC) {
1260 // In opportunistic mode, handle falling back to cleartext in some
1261 // cases (DNS shouldn't fail if a validated opportunistic mode server
1262 // becomes unreachable for some reason).
1263 switch (response) {
1264 case DnsTlsTransport::Response::success:
1265 *rcode = reinterpret_cast<HEADER*>(answer.base())->rcode;
1266 return resplen;
1267 case DnsTlsTransport::Response::network_error:
1268 // No need to set the error timeout here since it will fallback to UDP.
1269 case DnsTlsTransport::Response::internal_error:
1270 // Note: this will cause cleartext queries to be emitted, with
1271 // all of the EDNS0 goodness enabled. Fingers crossed. :-/
1272 *fallback = true;
1273 [[fallthrough]];
1274 default:
1275 return -1;
1276 }
1277 } else {
1278 // Strict mode
1279 switch (response) {
1280 case DnsTlsTransport::Response::success:
1281 *rcode = reinterpret_cast<HEADER*>(answer.base())->rcode;
1282 return resplen;
1283 case DnsTlsTransport::Response::network_error:
1284 // This case happens when the query stored in DnsTlsTransport is expired since
1285 // either 1) the query has been tried for 3 times but no response or 2) fail to
1286 // establish the connection with the server.
1287 *rcode = RCODE_TIMEOUT;
1288 [[fallthrough]];
1289 default:
1290 return -1;
1291 }
1292 }
1293 }
1294
resolv_res_nsend(const android_net_context * netContext,const uint8_t * msg,int msgLen,uint8_t * ans,int ansLen,int * rcode,uint32_t flags,NetworkDnsEventReported * event)1295 int resolv_res_nsend(const android_net_context* netContext, const uint8_t* msg, int msgLen,
1296 uint8_t* ans, int ansLen, int* rcode, uint32_t flags,
1297 NetworkDnsEventReported* event) {
1298 assert(event != nullptr);
1299 ResState res;
1300 res_init(&res, netContext, event);
1301 resolv_populate_res_for_net(&res);
1302 *rcode = NOERROR;
1303 return res_nsend(&res, msg, msgLen, ans, ansLen, rcode, flags);
1304 }
1305