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 ((flags & ANDROID_RESOLV_NO_RETRY) && usableServersCount > 1) {
501 auto hp = reinterpret_cast<const HEADER*>(buf);
502
503 // Select a random server based on the query id
504 int selectedServer = (hp->id % usableServersCount) + 1;
505 res_set_usable_server(selectedServer, statp->nameserverCount(), usable_servers);
506 }
507
508 // Send request, RETRY times, or until successful.
509 int retryTimes = (flags & ANDROID_RESOLV_NO_RETRY) ? 1 : params.retry_count;
510 int useTcp = buflen > PACKETSZ;
511 int gotsomewhere = 0;
512 // Use an impossible error code as default value
513 int terrno = ETIME;
514
515 for (int attempt = 0; attempt < retryTimes; ++attempt) {
516 for (size_t ns = 0; ns < statp->nsaddrs.size(); ++ns) {
517 if (!usable_servers[ns]) continue;
518
519 *rcode = RCODE_INTERNAL_ERROR;
520
521 // Get server addr
522 const IPSockAddr& serverSockAddr = statp->nsaddrs[ns];
523 LOG(DEBUG) << __func__ << ": Querying server (# " << ns + 1
524 << ") address = " << serverSockAddr.toString();
525
526 ::android::net::Protocol query_proto = useTcp ? PROTO_TCP : PROTO_UDP;
527 time_t query_time = 0;
528 int delay = 0;
529 bool fallbackTCP = false;
530 const bool shouldRecordStats = (attempt == 0);
531 int resplen;
532 Stopwatch queryStopwatch;
533 int retry_count_for_event = 0;
534 size_t actualNs = ns;
535 // Use an impossible error code as default value
536 terrno = ETIME;
537 if (useTcp) {
538 // TCP; at most one attempt per server.
539 attempt = retryTimes;
540 resplen = send_vc(statp, ¶ms, buf, buflen, ans, anssiz, &terrno, ns,
541 &query_time, rcode, &delay);
542
543 if (buflen <= PACKETSZ && resplen <= 0 &&
544 statp->tc_mode == aidl::android::net::IDnsResolver::TC_MODE_UDP_TCP) {
545 // reset to UDP for next query on next DNS server if resolver is currently doing
546 // TCP fallback retry and current server does not support TCP connectin
547 useTcp = false;
548 }
549 LOG(INFO) << __func__ << ": used send_vc " << resplen << " terrno: " << terrno;
550 } else {
551 // UDP
552 resplen = send_dg(statp, ¶ms, buf, buflen, ans, anssiz, &terrno, &actualNs,
553 &useTcp, &gotsomewhere, &query_time, rcode, &delay);
554 fallbackTCP = useTcp ? true : false;
555 retry_count_for_event = attempt;
556 LOG(INFO) << __func__ << ": used send_dg " << resplen << " terrno: " << terrno;
557 }
558
559 const IPSockAddr& receivedServerAddr = statp->nsaddrs[actualNs];
560 DnsQueryEvent* dnsQueryEvent = addDnsQueryEvent(statp->event);
561 dnsQueryEvent->set_cache_hit(static_cast<CacheStatus>(cache_status));
562 // When |retryTimes| > 1, we cannot actually know the correct latency value if we
563 // received the answer from the previous server. So temporarily set the latency as -1 if
564 // that condition happened.
565 // TODO: make the latency value accurate.
566 dnsQueryEvent->set_latency_micros(
567 (actualNs == ns) ? saturate_cast<int32_t>(queryStopwatch.timeTakenUs()) : -1);
568 dnsQueryEvent->set_dns_server_index(actualNs);
569 dnsQueryEvent->set_ip_version(ipFamilyToIPVersion(receivedServerAddr.family()));
570 dnsQueryEvent->set_retry_times(retry_count_for_event);
571 dnsQueryEvent->set_rcode(static_cast<NsRcode>(*rcode));
572 dnsQueryEvent->set_protocol(query_proto);
573 dnsQueryEvent->set_type(getQueryType(buf, buflen));
574 dnsQueryEvent->set_linux_errno(static_cast<LinuxErrno>(terrno));
575
576 // Only record stats the first time we try a query. This ensures that
577 // queries that deterministically fail (e.g., a name that always returns
578 // SERVFAIL or times out) do not unduly affect the stats.
579 if (shouldRecordStats) {
580 // (b/151166599): This is a workaround to prevent that DnsResolver calculates the
581 // reliability of DNS servers from being broken when network restricted mode is
582 // enabled.
583 // TODO: Introduce the new server selection instead of skipping stats recording.
584 if (!isNetworkRestricted(terrno)) {
585 res_sample sample;
586 res_stats_set_sample(&sample, query_time, *rcode, delay);
587 // KeepListening UDP mechanism is incompatible with usable_servers of legacy
588 // stats, so keep the old logic for now.
589 // TODO: Replace usable_servers of legacy stats with new one.
590 resolv_cache_add_resolver_stats_sample(
591 statp->netid, revision_id, serverSockAddr, sample, params.max_samples);
592 }
593 resolv_stats_add(statp->netid, receivedServerAddr, dnsQueryEvent);
594 }
595
596 if (resplen == 0) continue;
597 if (fallbackTCP) {
598 ns--;
599 continue;
600 }
601 if (resplen < 0) {
602 _resolv_cache_query_failed(statp->netid, buf, buflen, flags);
603 statp->closeSockets();
604 return -terrno;
605 };
606
607 LOG(DEBUG) << __func__ << ": got answer:";
608 res_pquery(ans, (resplen > anssiz) ? anssiz : resplen);
609
610 if (cache_status == RESOLV_CACHE_NOTFOUND) {
611 resolv_cache_add(statp->netid, buf, buflen, ans, resplen);
612 }
613 statp->closeSockets();
614 return (resplen);
615 } // for each ns
616 } // for each retry
617 statp->closeSockets();
618 terrno = useTcp ? terrno : gotsomewhere ? ETIMEDOUT : ECONNREFUSED;
619 // TODO: Remove errno once callers stop using it
620 errno = useTcp ? terrno
621 : gotsomewhere ? ETIMEDOUT /* no answer obtained */
622 : ECONNREFUSED /* no nameservers found */;
623
624 _resolv_cache_query_failed(statp->netid, buf, buflen, flags);
625 return -terrno;
626 }
627
get_timeout(res_state statp,const res_params * params,const int ns)628 static struct timespec get_timeout(res_state statp, const res_params* params, const int ns) {
629 int msec;
630 // Legacy algorithm which scales the timeout by nameserver number.
631 // For instance, with 4 nameservers: 5s, 2.5s, 5s, 10s
632 // This has no effect with 1 or 2 nameservers
633 msec = params->base_timeout_msec << ns;
634 if (ns > 0) {
635 msec /= statp->nameserverCount();
636 }
637 // For safety, don't allow OEMs and experiments to configure a timeout shorter than 1s.
638 if (msec < 1000) {
639 msec = 1000; // Use at least 1000ms
640 }
641 LOG(INFO) << __func__ << ": using timeout of " << msec << " msec";
642
643 struct timespec result;
644 result.tv_sec = msec / 1000;
645 result.tv_nsec = (msec % 1000) * 1000000;
646 return result;
647 }
648
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)649 static int send_vc(res_state statp, res_params* params, const uint8_t* buf, int buflen,
650 uint8_t* ans, int anssiz, int* terrno, size_t ns, time_t* at, int* rcode,
651 int* delay) {
652 *at = time(NULL);
653 *delay = 0;
654 const HEADER* hp = (const HEADER*) (const void*) buf;
655 HEADER* anhp = (HEADER*) (void*) ans;
656 struct sockaddr* nsap;
657 int nsaplen;
658 int truncating, connreset, n;
659 uint8_t* cp;
660
661 LOG(INFO) << __func__ << ": using send_vc";
662
663 // It should never happen, but just in case.
664 if (ns >= statp->nsaddrs.size()) {
665 LOG(ERROR) << __func__ << ": Out-of-bound indexing: " << ns;
666 *terrno = EINVAL;
667 return -1;
668 }
669
670 sockaddr_storage ss = statp->nsaddrs[ns];
671 nsap = reinterpret_cast<sockaddr*>(&ss);
672 nsaplen = sockaddrSize(nsap);
673
674 connreset = 0;
675 same_ns:
676 truncating = 0;
677
678 struct timespec start_time = evNowTime();
679
680 /* Are we still talking to whom we want to talk to? */
681 if (statp->tcp_nssock >= 0 && (statp->_flags & RES_F_VC) != 0) {
682 struct sockaddr_storage peer;
683 socklen_t size = sizeof peer;
684 unsigned old_mark;
685 socklen_t mark_size = sizeof(old_mark);
686 if (getpeername(statp->tcp_nssock, (struct sockaddr*)(void*)&peer, &size) < 0 ||
687 !sock_eq((struct sockaddr*)(void*)&peer, nsap) ||
688 getsockopt(statp->tcp_nssock, SOL_SOCKET, SO_MARK, &old_mark, &mark_size) < 0 ||
689 old_mark != statp->_mark) {
690 statp->closeSockets();
691 }
692 }
693
694 if (statp->tcp_nssock < 0 || (statp->_flags & RES_F_VC) == 0) {
695 if (statp->tcp_nssock >= 0) statp->closeSockets();
696
697 statp->tcp_nssock.reset(socket(nsap->sa_family, SOCK_STREAM | SOCK_CLOEXEC, 0));
698 if (statp->tcp_nssock < 0) {
699 *terrno = errno;
700 PLOG(DEBUG) << __func__ << ": socket(vc): ";
701 switch (errno) {
702 case EPROTONOSUPPORT:
703 case EPFNOSUPPORT:
704 case EAFNOSUPPORT:
705 return 0;
706 default:
707 return -1;
708 }
709 }
710 const uid_t uid = statp->enforce_dns_uid ? AID_DNS : statp->uid;
711 resolv_tag_socket(statp->tcp_nssock, uid, statp->pid);
712 if (statp->_mark != MARK_UNSET) {
713 if (setsockopt(statp->tcp_nssock, SOL_SOCKET, SO_MARK, &statp->_mark,
714 sizeof(statp->_mark)) < 0) {
715 *terrno = errno;
716 PLOG(DEBUG) << __func__ << ": setsockopt: ";
717 return -1;
718 }
719 }
720 errno = 0;
721 if (random_bind(statp->tcp_nssock, nsap->sa_family) < 0) {
722 *terrno = errno;
723 dump_error("bind/vc", nsap, nsaplen);
724 statp->closeSockets();
725 return (0);
726 }
727 if (connect_with_timeout(statp->tcp_nssock, nsap, (socklen_t)nsaplen,
728 get_timeout(statp, params, ns)) < 0) {
729 *terrno = errno;
730 dump_error("connect/vc", nsap, nsaplen);
731 statp->closeSockets();
732 /*
733 * The way connect_with_timeout() is implemented prevents us from reliably
734 * determining whether this was really a timeout or e.g. ECONNREFUSED. Since
735 * currently both cases are handled in the same way, there is no need to
736 * change this (yet). If we ever need to reliably distinguish between these
737 * cases, both connect_with_timeout() and retrying_poll() need to be
738 * modified, though.
739 */
740 *rcode = RCODE_TIMEOUT;
741 return (0);
742 }
743 statp->_flags |= RES_F_VC;
744 }
745
746 /*
747 * Send length & message
748 */
749 uint16_t len = htons(static_cast<uint16_t>(buflen));
750 const iovec iov[] = {
751 {.iov_base = &len, .iov_len = INT16SZ},
752 {.iov_base = const_cast<uint8_t*>(buf), .iov_len = static_cast<size_t>(buflen)},
753 };
754 if (writev(statp->tcp_nssock, iov, 2) != (INT16SZ + buflen)) {
755 *terrno = errno;
756 PLOG(DEBUG) << __func__ << ": write failed: ";
757 statp->closeSockets();
758 return (0);
759 }
760 /*
761 * Receive length & response
762 */
763 read_len:
764 cp = ans;
765 len = INT16SZ;
766 while ((n = read(statp->tcp_nssock, (char*)cp, (size_t)len)) > 0) {
767 cp += n;
768 if ((len -= n) == 0) break;
769 }
770 if (n <= 0) {
771 *terrno = errno;
772 PLOG(DEBUG) << __func__ << ": read failed: ";
773 statp->closeSockets();
774 /*
775 * A long running process might get its TCP
776 * connection reset if the remote server was
777 * restarted. Requery the server instead of
778 * trying a new one. When there is only one
779 * server, this means that a query might work
780 * instead of failing. We only allow one reset
781 * per query to prevent looping.
782 */
783 if (*terrno == ECONNRESET && !connreset) {
784 connreset = 1;
785 goto same_ns;
786 }
787 return (0);
788 }
789 uint16_t resplen = ntohs(*reinterpret_cast<const uint16_t*>(ans));
790 if (resplen > anssiz) {
791 LOG(DEBUG) << __func__ << ": response truncated";
792 truncating = 1;
793 len = anssiz;
794 } else
795 len = resplen;
796 if (len < HFIXEDSZ) {
797 /*
798 * Undersized message.
799 */
800 LOG(DEBUG) << __func__ << ": undersized: " << len;
801 *terrno = EMSGSIZE;
802 statp->closeSockets();
803 return (0);
804 }
805 cp = ans;
806 while (len != 0 && (n = read(statp->tcp_nssock, (char*)cp, (size_t)len)) > 0) {
807 cp += n;
808 len -= n;
809 }
810 if (n <= 0) {
811 *terrno = errno;
812 PLOG(DEBUG) << __func__ << ": read(vc): ";
813 statp->closeSockets();
814 return (0);
815 }
816
817 if (truncating) {
818 /*
819 * Flush rest of answer so connection stays in synch.
820 */
821 anhp->tc = 1;
822 len = resplen - anssiz;
823 while (len != 0) {
824 char junk[PACKETSZ];
825
826 n = read(statp->tcp_nssock, junk, (len > sizeof junk) ? sizeof junk : len);
827 if (n > 0)
828 len -= n;
829 else
830 break;
831 }
832 LOG(WARNING) << __func__ << ": resplen " << resplen << " exceeds buf size " << anssiz;
833 // return size should never exceed container size
834 resplen = anssiz;
835 }
836 /*
837 * If the calling application has bailed out of
838 * a previous call and failed to arrange to have
839 * the circuit closed or the server has got
840 * itself confused, then drop the packet and
841 * wait for the correct one.
842 */
843 if (hp->id != anhp->id) {
844 LOG(DEBUG) << __func__ << ": ld answer (unexpected):";
845 res_pquery(ans, resplen);
846 goto read_len;
847 }
848
849 /*
850 * All is well, or the error is fatal. Signal that the
851 * next nameserver ought not be tried.
852 */
853 if (resplen > 0) {
854 struct timespec done = evNowTime();
855 *delay = res_stats_calculate_rtt(&done, &start_time);
856 *rcode = anhp->rcode;
857 }
858 *terrno = 0;
859 return (resplen);
860 }
861
862 /* return -1 on error (errno set), 0 on success */
connect_with_timeout(int sock,const sockaddr * nsap,socklen_t salen,const timespec timeout)863 static int connect_with_timeout(int sock, const sockaddr* nsap, socklen_t salen,
864 const timespec timeout) {
865 int res, origflags;
866
867 origflags = fcntl(sock, F_GETFL, 0);
868 fcntl(sock, F_SETFL, origflags | O_NONBLOCK);
869
870 res = connect(sock, nsap, salen);
871 if (res < 0 && errno != EINPROGRESS) {
872 res = -1;
873 goto done;
874 }
875 if (res != 0) {
876 timespec now = evNowTime();
877 timespec finish = evAddTime(now, timeout);
878 LOG(INFO) << __func__ << ": " << sock << " send_vc";
879 res = retrying_poll(sock, POLLIN | POLLOUT, &finish);
880 if (res <= 0) {
881 res = -1;
882 }
883 }
884 done:
885 fcntl(sock, F_SETFL, origflags);
886 LOG(INFO) << __func__ << ": " << sock << " connect_with_const timeout returning " << res;
887 return res;
888 }
889
retrying_poll(const int sock,const short events,const struct timespec * finish)890 static int retrying_poll(const int sock, const short events, const struct timespec* finish) {
891 struct timespec now, timeout;
892
893 retry:
894 LOG(INFO) << __func__ << ": " << sock << " retrying_poll";
895
896 now = evNowTime();
897 if (evCmpTime(*finish, now) > 0)
898 timeout = evSubTime(*finish, now);
899 else
900 timeout = evConsTime(0L, 0L);
901 struct pollfd fds = {.fd = sock, .events = events};
902 int n = ppoll(&fds, 1, &timeout, /*__mask=*/NULL);
903 if (n == 0) {
904 LOG(INFO) << __func__ << ": " << sock << " retrying_poll timeout";
905 errno = ETIMEDOUT;
906 return 0;
907 }
908 if (n < 0) {
909 if (errno == EINTR) goto retry;
910 PLOG(INFO) << __func__ << ": " << sock << " retrying_poll failed";
911 return n;
912 }
913 if (fds.revents & (POLLIN | POLLOUT | POLLERR)) {
914 int error;
915 socklen_t len = sizeof(error);
916 if (getsockopt(sock, SOL_SOCKET, SO_ERROR, &error, &len) < 0 || error) {
917 errno = error;
918 PLOG(INFO) << __func__ << ": " << sock << " retrying_poll getsockopt failed";
919 return -1;
920 }
921 }
922 LOG(INFO) << __func__ << ": " << sock << " retrying_poll returning " << n;
923 return n;
924 }
925
extractUdpFdset(res_state statp,const short events=POLLIN)926 static std::vector<pollfd> extractUdpFdset(res_state statp, const short events = POLLIN) {
927 std::vector<pollfd> fdset(statp->nsaddrs.size());
928 for (size_t i = 0; i < statp->nsaddrs.size(); ++i) {
929 fdset[i] = {.fd = statp->nssocks[i], .events = events};
930 }
931 return fdset;
932 }
933
udpRetryingPoll(res_state statp,const timespec * finish)934 static Result<std::vector<int>> udpRetryingPoll(res_state statp, const timespec* finish) {
935 for (;;) {
936 LOG(DEBUG) << __func__ << ": poll";
937 timespec start_time = evNowTime();
938 timespec timeout = (evCmpTime(*finish, start_time) > 0) ? evSubTime(*finish, start_time)
939 : evConsTime(0L, 0L);
940 std::vector<pollfd> fdset = extractUdpFdset(statp);
941 const int n = ppoll(fdset.data(), fdset.size(), &timeout, /*__mask=*/nullptr);
942 if (n <= 0) {
943 if (errno == EINTR && n < 0) continue;
944 if (n == 0) errno = ETIMEDOUT;
945 PLOG(INFO) << __func__ << ": failed";
946 return ErrnoError();
947 }
948 std::vector<int> fdsToRead;
949 for (const auto& pollfd : fdset) {
950 if (pollfd.revents & (POLLIN | POLLERR)) {
951 fdsToRead.push_back(pollfd.fd);
952 }
953 }
954 LOG(DEBUG) << __func__ << ": "
955 << " returning fd size: " << fdsToRead.size();
956 return fdsToRead;
957 }
958 }
959
udpRetryingPollWrapper(res_state statp,int ns,const timespec * finish)960 static Result<std::vector<int>> udpRetryingPollWrapper(res_state statp, int ns,
961 const timespec* finish) {
962 const bool keepListeningUdp =
963 android::net::Experiments::getInstance()->getFlag("keep_listening_udp", 0);
964 if (keepListeningUdp) return udpRetryingPoll(statp, finish);
965
966 if (int n = retrying_poll(statp->nssocks[ns], POLLIN, finish); n <= 0) {
967 return ErrnoError();
968 }
969 return std::vector<int>{statp->nssocks[ns]};
970 }
971
ignoreInvalidAnswer(res_state statp,const sockaddr_storage & from,const uint8_t * buf,int buflen,uint8_t * ans,int anssiz,int * receivedFromNs)972 bool ignoreInvalidAnswer(res_state statp, const sockaddr_storage& from, const uint8_t* buf,
973 int buflen, uint8_t* ans, int anssiz, int* receivedFromNs) {
974 const HEADER* hp = (const HEADER*)(const void*)buf;
975 HEADER* anhp = (HEADER*)(void*)ans;
976 if (hp->id != anhp->id) {
977 // response from old query, ignore it.
978 LOG(DEBUG) << __func__ << ": old answer:";
979 return true;
980 }
981 if (*receivedFromNs = res_ourserver_p(statp, (sockaddr*)(void*)&from); *receivedFromNs < 0) {
982 // response from wrong server? ignore it.
983 LOG(DEBUG) << __func__ << ": not our server:";
984 return true;
985 }
986 if (!res_queriesmatch(buf, buf + buflen, ans, ans + anssiz)) {
987 // response contains wrong query? ignore it.
988 LOG(DEBUG) << __func__ << ": wrong query name:";
989 return true;
990 }
991 return false;
992 }
993
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)994 static int send_dg(res_state statp, res_params* params, const uint8_t* buf, int buflen,
995 uint8_t* ans, int anssiz, int* terrno, size_t* ns, int* v_circuit,
996 int* gotsomewhere, time_t* at, int* rcode, int* delay) {
997 // It should never happen, but just in case.
998 if (*ns >= statp->nsaddrs.size()) {
999 LOG(ERROR) << __func__ << ": Out-of-bound indexing: " << ns;
1000 *terrno = EINVAL;
1001 return -1;
1002 }
1003
1004 *at = time(nullptr);
1005 *delay = 0;
1006 const sockaddr_storage ss = statp->nsaddrs[*ns];
1007 const sockaddr* nsap = reinterpret_cast<const sockaddr*>(&ss);
1008 const int nsaplen = sockaddrSize(nsap);
1009
1010 if (statp->nssocks[*ns] == -1) {
1011 statp->nssocks[*ns].reset(socket(nsap->sa_family, SOCK_DGRAM | SOCK_CLOEXEC, 0));
1012 if (statp->nssocks[*ns] < 0) {
1013 *terrno = errno;
1014 PLOG(DEBUG) << __func__ << ": socket(dg): ";
1015 switch (errno) {
1016 case EPROTONOSUPPORT:
1017 case EPFNOSUPPORT:
1018 case EAFNOSUPPORT:
1019 return (0);
1020 default:
1021 return (-1);
1022 }
1023 }
1024
1025 const uid_t uid = statp->enforce_dns_uid ? AID_DNS : statp->uid;
1026 resolv_tag_socket(statp->nssocks[*ns], uid, statp->pid);
1027 if (statp->_mark != MARK_UNSET) {
1028 if (setsockopt(statp->nssocks[*ns], SOL_SOCKET, SO_MARK, &(statp->_mark),
1029 sizeof(statp->_mark)) < 0) {
1030 *terrno = errno;
1031 statp->closeSockets();
1032 return -1;
1033 }
1034 }
1035 // Use a "connected" datagram socket to receive an ECONNREFUSED error
1036 // on the next socket operation when the server responds with an
1037 // ICMP port-unreachable error. This way we can detect the absence of
1038 // a nameserver without timing out.
1039 if (random_bind(statp->nssocks[*ns], nsap->sa_family) < 0) {
1040 *terrno = errno;
1041 dump_error("bind(dg)", nsap, nsaplen);
1042 statp->closeSockets();
1043 return (0);
1044 }
1045 if (connect(statp->nssocks[*ns], nsap, (socklen_t)nsaplen) < 0) {
1046 *terrno = errno;
1047 dump_error("connect(dg)", nsap, nsaplen);
1048 statp->closeSockets();
1049 return (0);
1050 }
1051 LOG(DEBUG) << __func__ << ": new DG socket";
1052 }
1053 if (send(statp->nssocks[*ns], (const char*)buf, (size_t)buflen, 0) != buflen) {
1054 *terrno = errno;
1055 PLOG(DEBUG) << __func__ << ": send: ";
1056 statp->closeSockets();
1057 return 0;
1058 }
1059
1060 timespec timeout = get_timeout(statp, params, *ns);
1061 timespec start_time = evNowTime();
1062 timespec finish = evAddTime(start_time, timeout);
1063 for (;;) {
1064 // Wait for reply.
1065 auto result = udpRetryingPollWrapper(statp, *ns, &finish);
1066
1067 if (!result.has_value()) {
1068 const bool isTimeout = (result.error().code() == ETIMEDOUT);
1069 *rcode = (isTimeout) ? RCODE_TIMEOUT : *rcode;
1070 *terrno = (isTimeout) ? ETIMEDOUT : errno;
1071 *gotsomewhere = (isTimeout) ? 1 : *gotsomewhere;
1072 // Leave the UDP sockets open on timeout so we can keep listening for
1073 // a late response from this server while retrying on the next server.
1074 if (!isTimeout) statp->closeSockets();
1075 LOG(DEBUG) << __func__ << ": " << (isTimeout) ? "timeout" : "poll";
1076 return 0;
1077 }
1078 bool needRetry = false;
1079 for (int fd : result.value()) {
1080 needRetry = false;
1081 sockaddr_storage from;
1082 socklen_t fromlen = sizeof(from);
1083 int resplen =
1084 recvfrom(fd, (char*)ans, (size_t)anssiz, 0, (sockaddr*)(void*)&from, &fromlen);
1085 if (resplen <= 0) {
1086 *terrno = errno;
1087 PLOG(DEBUG) << __func__ << ": recvfrom: ";
1088 continue;
1089 }
1090 *gotsomewhere = 1;
1091 if (resplen < HFIXEDSZ) {
1092 // Undersized message.
1093 LOG(DEBUG) << __func__ << ": undersized: " << resplen;
1094 *terrno = EMSGSIZE;
1095 continue;
1096 }
1097
1098 int receivedFromNs = *ns;
1099 if (needRetry =
1100 ignoreInvalidAnswer(statp, from, buf, buflen, ans, anssiz, &receivedFromNs);
1101 needRetry) {
1102 res_pquery(ans, (resplen > anssiz) ? anssiz : resplen);
1103 continue;
1104 }
1105
1106 HEADER* anhp = (HEADER*)(void*)ans;
1107 if (anhp->rcode == FORMERR && (statp->netcontext_flags & NET_CONTEXT_FLAG_USE_EDNS)) {
1108 // Do not retry if the server do not understand EDNS0.
1109 // The case has to be captured here, as FORMERR packet do not
1110 // carry query section, hence res_queriesmatch() returns 0.
1111 LOG(DEBUG) << __func__ << ": server rejected query with EDNS0:";
1112 res_pquery(ans, (resplen > anssiz) ? anssiz : resplen);
1113 // record the error
1114 statp->_flags |= RES_F_EDNS0ERR;
1115 *terrno = EREMOTEIO;
1116 continue;
1117 }
1118
1119 timespec done = evNowTime();
1120 *delay = res_stats_calculate_rtt(&done, &start_time);
1121 if (anhp->rcode == SERVFAIL || anhp->rcode == NOTIMP || anhp->rcode == REFUSED) {
1122 LOG(DEBUG) << __func__ << ": server rejected query:";
1123 res_pquery(ans, (resplen > anssiz) ? anssiz : resplen);
1124 *rcode = anhp->rcode;
1125 continue;
1126 }
1127 if (anhp->tc) {
1128 // To get the rest of answer,
1129 // use TCP with same server.
1130 LOG(DEBUG) << __func__ << ": truncated answer";
1131 *terrno = E2BIG;
1132 *v_circuit = 1;
1133 return 1;
1134 }
1135 // All is well, or the error is fatal. Signal that the
1136 // next nameserver ought not be tried.
1137
1138 *rcode = anhp->rcode;
1139 *ns = receivedFromNs;
1140 *terrno = 0;
1141 return resplen;
1142 }
1143 if (!needRetry) return 0;
1144 }
1145 }
1146
dump_error(const char * str,const struct sockaddr * address,int alen)1147 static void dump_error(const char* str, const struct sockaddr* address, int alen) {
1148 char hbuf[NI_MAXHOST];
1149 char sbuf[NI_MAXSERV];
1150 constexpr int niflags = NI_NUMERICHOST | NI_NUMERICSERV;
1151 const int err = errno;
1152
1153 if (!WOULD_LOG(DEBUG)) return;
1154
1155 if (getnameinfo(address, (socklen_t)alen, hbuf, sizeof(hbuf), sbuf, sizeof(sbuf), niflags)) {
1156 strncpy(hbuf, "?", sizeof(hbuf) - 1);
1157 hbuf[sizeof(hbuf) - 1] = '\0';
1158 strncpy(sbuf, "?", sizeof(sbuf) - 1);
1159 sbuf[sizeof(sbuf) - 1] = '\0';
1160 }
1161 errno = err;
1162 PLOG(DEBUG) << __func__ << ": " << str << " ([" << hbuf << "]." << sbuf << "): ";
1163 }
1164
sock_eq(struct sockaddr * a,struct sockaddr * b)1165 static int sock_eq(struct sockaddr* a, struct sockaddr* b) {
1166 struct sockaddr_in *a4, *b4;
1167 struct sockaddr_in6 *a6, *b6;
1168
1169 if (a->sa_family != b->sa_family) return 0;
1170 switch (a->sa_family) {
1171 case AF_INET:
1172 a4 = (struct sockaddr_in*) (void*) a;
1173 b4 = (struct sockaddr_in*) (void*) b;
1174 return a4->sin_port == b4->sin_port && a4->sin_addr.s_addr == b4->sin_addr.s_addr;
1175 case AF_INET6:
1176 a6 = (struct sockaddr_in6*) (void*) a;
1177 b6 = (struct sockaddr_in6*) (void*) b;
1178 return a6->sin6_port == b6->sin6_port &&
1179 #ifdef HAVE_SIN6_SCOPE_ID
1180 a6->sin6_scope_id == b6->sin6_scope_id &&
1181 #endif
1182 IN6_ARE_ADDR_EQUAL(&a6->sin6_addr, &b6->sin6_addr);
1183 default:
1184 return 0;
1185 }
1186 }
1187
convertEnumType(PrivateDnsMode privateDnsmode)1188 PrivateDnsModes convertEnumType(PrivateDnsMode privateDnsmode) {
1189 switch (privateDnsmode) {
1190 case PrivateDnsMode::OFF:
1191 return PrivateDnsModes::PDM_OFF;
1192 case PrivateDnsMode::OPPORTUNISTIC:
1193 return PrivateDnsModes::PDM_OPPORTUNISTIC;
1194 case PrivateDnsMode::STRICT:
1195 return PrivateDnsModes::PDM_STRICT;
1196 default:
1197 return PrivateDnsModes::PDM_UNKNOWN;
1198 }
1199 }
1200
res_tls_send(res_state statp,const Slice query,const Slice answer,int * rcode,bool * fallback)1201 static int res_tls_send(res_state statp, const Slice query, const Slice answer, int* rcode,
1202 bool* fallback) {
1203 int resplen = 0;
1204 const unsigned netId = statp->netid;
1205
1206 PrivateDnsStatus privateDnsStatus = gPrivateDnsConfiguration.getStatus(netId);
1207 statp->event->set_private_dns_modes(convertEnumType(privateDnsStatus.mode));
1208
1209 if (privateDnsStatus.mode == PrivateDnsMode::OFF) {
1210 *fallback = true;
1211 return -1;
1212 }
1213
1214 if (privateDnsStatus.validatedServers().empty()) {
1215 if (privateDnsStatus.mode == PrivateDnsMode::OPPORTUNISTIC) {
1216 *fallback = true;
1217 return -1;
1218 } else {
1219 // Sleep and iterate some small number of times checking for the
1220 // arrival of resolved and validated server IP addresses, instead
1221 // of returning an immediate error.
1222 // This is needed because as soon as a network becomes the default network, apps will
1223 // send DNS queries on that network. If no servers have yet validated, and we do not
1224 // block those queries, they would immediately fail, causing application-visible errors.
1225 // Note that this can happen even before the network validates, since an unvalidated
1226 // network can become the default network if no validated networks are available.
1227 //
1228 // TODO: see if there is a better way to address this problem, such as buffering the
1229 // queries in a queue or only blocking queries for the first few seconds after a default
1230 // network change.
1231 for (int i = 0; i < 42; i++) {
1232 std::this_thread::sleep_for(std::chrono::milliseconds(100));
1233 // Calling getStatus() to merely check if there's any validated server seems
1234 // wasteful. Consider adding a new method in PrivateDnsConfiguration for speed ups.
1235 if (!gPrivateDnsConfiguration.getStatus(netId).validatedServers().empty()) {
1236 privateDnsStatus = gPrivateDnsConfiguration.getStatus(netId);
1237 break;
1238 }
1239 }
1240 if (privateDnsStatus.validatedServers().empty()) {
1241 return -1;
1242 }
1243 }
1244 }
1245
1246 LOG(INFO) << __func__ << ": performing query over TLS";
1247
1248 const auto response = sDnsTlsDispatcher.query(privateDnsStatus.validatedServers(), statp, query,
1249 answer, &resplen);
1250
1251 LOG(INFO) << __func__ << ": TLS query result: " << static_cast<int>(response);
1252
1253 if (privateDnsStatus.mode == PrivateDnsMode::OPPORTUNISTIC) {
1254 // In opportunistic mode, handle falling back to cleartext in some
1255 // cases (DNS shouldn't fail if a validated opportunistic mode server
1256 // becomes unreachable for some reason).
1257 switch (response) {
1258 case DnsTlsTransport::Response::success:
1259 *rcode = reinterpret_cast<HEADER*>(answer.base())->rcode;
1260 return resplen;
1261 case DnsTlsTransport::Response::network_error:
1262 // No need to set the error timeout here since it will fallback to UDP.
1263 case DnsTlsTransport::Response::internal_error:
1264 // Note: this will cause cleartext queries to be emitted, with
1265 // all of the EDNS0 goodness enabled. Fingers crossed. :-/
1266 *fallback = true;
1267 [[fallthrough]];
1268 default:
1269 return -1;
1270 }
1271 } else {
1272 // Strict mode
1273 switch (response) {
1274 case DnsTlsTransport::Response::success:
1275 *rcode = reinterpret_cast<HEADER*>(answer.base())->rcode;
1276 return resplen;
1277 case DnsTlsTransport::Response::network_error:
1278 // This case happens when the query stored in DnsTlsTransport is expired since
1279 // either 1) the query has been tried for 3 times but no response or 2) fail to
1280 // establish the connection with the server.
1281 *rcode = RCODE_TIMEOUT;
1282 [[fallthrough]];
1283 default:
1284 return -1;
1285 }
1286 }
1287 }
1288
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)1289 int resolv_res_nsend(const android_net_context* netContext, const uint8_t* msg, int msgLen,
1290 uint8_t* ans, int ansLen, int* rcode, uint32_t flags,
1291 NetworkDnsEventReported* event) {
1292 assert(event != nullptr);
1293 ResState res;
1294 res_init(&res, netContext, event);
1295 resolv_populate_res_for_net(&res);
1296 *rcode = NOERROR;
1297 return res_nsend(&res, msg, msgLen, ans, ansLen, rcode, flags);
1298 }
1299