1#include <sys/socket.h>
2#include <netinet/in.h>
3#include <netinet/tcp.h>
4#include <netdb.h>
5#include <arpa/inet.h>
6#include <stdint.h>
7#include <string.h>
8#include <poll.h>
9#include <time.h>
10#include <ctype.h>
11#include <unistd.h>
12#include <errno.h>
13#include <pthread.h>
14#include "stdio_impl.h"
15#include "syscall.h"
16#include "lookup.h"
17
18static void cleanup(void *p)
19{
20 struct pollfd *pfd = p;
21 for (int i=0; pfd[i].fd >= -1; i++)
22 if (pfd[i].fd >= 0) __syscall(SYS_close, pfd[i].fd);
23}
24
25static unsigned long mtime()
26{
27 struct timespec ts;
28 if (clock_gettime(CLOCK_MONOTONIC, &ts) < 0 && errno == ENOSYS)
29 clock_gettime(CLOCK_REALTIME, &ts);
30 return (unsigned long)ts.tv_sec * 1000
31 + ts.tv_nsec / 1000000;
32}
33
34static int start_tcp(struct pollfd *pfd, int family, const void *sa, socklen_t sl, const unsigned char *q, int ql)
35{
36 struct msghdr mh = {
37 .msg_name = (void *)sa,
38 .msg_namelen = sl,
39 .msg_iovlen = 2,
40 .msg_iov = (struct iovec [2]){
41 { .iov_base = (uint8_t[]){ ql>>8, ql }, .iov_len = 2 },
42 { .iov_base = (void *)q, .iov_len = ql } }
43 };
44 int r;
45 int fd = socket(family, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
46 pfd->fd = fd;
47 pfd->events = POLLOUT;
48 if (!setsockopt(fd, IPPROTO_TCP, TCP_FASTOPEN_CONNECT,
49 &(int){1}, sizeof(int))) {
50 r = sendmsg(fd, &mh, MSG_FASTOPEN|MSG_NOSIGNAL);
51 if (r == ql+2) pfd->events = POLLIN;
52 if (r >= 0) return r;
53 if (errno == EINPROGRESS) return 0;
54 }
55 r = connect(fd, sa, sl);
56 if (!r || errno == EINPROGRESS) return 0;
57 close(fd);
58 pfd->fd = -1;
59 return -1;
60}
61
62static void step_mh(struct msghdr *mh, size_t n)
63{
64 /* Adjust iovec in msghdr to skip first n bytes. */
65 while (mh->msg_iovlen && n >= mh->msg_iov->iov_len) {
66 n -= mh->msg_iov->iov_len;
67 mh->msg_iov++;
68 mh->msg_iovlen--;
69 }
70 if (!mh->msg_iovlen) return;
71 mh->msg_iov->iov_base = (char *)mh->msg_iov->iov_base + n;
72 mh->msg_iov->iov_len -= n;
73}
74
75/* Internal contract for __res_msend[_rc]: asize must be >=512, nqueries
76 * must be sufficiently small to be safe as VLA size. In practice it's
77 * either 1 or 2, anyway. */
78
79int __res_msend_rc(int nqueries, const unsigned char *const *queries,
80 const int *qlens, unsigned char *const *answers, int *alens, int asize,
81 const struct resolvconf *conf)
82{
83 int fd;
84 int timeout, attempts, retry_interval, servfail_retry;
85 union {
86 struct sockaddr_in sin;
87 struct sockaddr_in6 sin6;
88 } sa = {0}, ns[MAXNS] = {{0}};
89 socklen_t sl = sizeof sa.sin;
90 int nns = 0;
91 int family = AF_INET;
92 int rlen;
93 int next;
94 int i, j;
95 int cs;
96 struct pollfd pfd[nqueries+2];
97 int qpos[nqueries], apos[nqueries];
98 unsigned char alen_buf[nqueries][2];
99 int r;
100 unsigned long t0, t1, t2;
101
102 pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &cs);
103
104 timeout = 1000*conf->timeout;
105 attempts = conf->attempts;
106
107 for (nns=0; nns<conf->nns; nns++) {
108 const struct address *iplit = &conf->ns[nns];
109 if (iplit->family == AF_INET) {
110 memcpy(&ns[nns].sin.sin_addr, iplit->addr, 4);
111 ns[nns].sin.sin_port = htons(53);
112 ns[nns].sin.sin_family = AF_INET;
113 } else {
114 sl = sizeof sa.sin6;
115 memcpy(&ns[nns].sin6.sin6_addr, iplit->addr, 16);
116 ns[nns].sin6.sin6_port = htons(53);
117 ns[nns].sin6.sin6_scope_id = iplit->scopeid;
118 ns[nns].sin6.sin6_family = family = AF_INET6;
119 }
120 }
121
122 /* Get local address and open/bind a socket */
123 fd = socket(family, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
124
125 /* Handle case where system lacks IPv6 support */
126 if (fd < 0 && family == AF_INET6 && errno == EAFNOSUPPORT) {
127 for (i=0; i<nns && conf->ns[nns].family == AF_INET6; i++);
128 if (i==nns) {
129 pthread_setcancelstate(cs, 0);
130 return -1;
131 }
132 fd = socket(AF_INET, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
133 family = AF_INET;
134 sl = sizeof sa.sin;
135 }
136
137 /* Convert any IPv4 addresses in a mixed environment to v4-mapped */
138 if (fd >= 0 && family == AF_INET6) {
139 setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, &(int){0}, sizeof 0);
140 for (i=0; i<nns; i++) {
141 if (ns[i].sin.sin_family != AF_INET) continue;
142 memcpy(ns[i].sin6.sin6_addr.s6_addr+12,
143 &ns[i].sin.sin_addr, 4);
144 memcpy(ns[i].sin6.sin6_addr.s6_addr,
145 "\0\0\0\0\0\0\0\0\0\0\xff\xff", 12);
146 ns[i].sin6.sin6_family = AF_INET6;
147 ns[i].sin6.sin6_flowinfo = 0;
148 ns[i].sin6.sin6_scope_id = 0;
149 }
150 }
151
152 sa.sin.sin_family = family;
153 if (fd < 0 || bind(fd, (void *)&sa, sl) < 0) {
154 if (fd >= 0) close(fd);
155 pthread_setcancelstate(cs, 0);
156 return -1;
157 }
158
159 /* Past this point, there are no errors. Each individual query will
160 * yield either no reply (indicated by zero length) or an answer
161 * packet which is up to the caller to interpret. */
162
163 for (i=0; i<nqueries; i++) pfd[i].fd = -1;
164 pfd[nqueries].fd = fd;
165 pfd[nqueries].events = POLLIN;
166 pfd[nqueries+1].fd = -2;
167
168 pthread_cleanup_push(cleanup, pfd);
169 pthread_setcancelstate(cs, 0);
170
171 memset(alens, 0, sizeof *alens * nqueries);
172
173 retry_interval = timeout / attempts;
174 next = 0;
175 t0 = t2 = mtime();
176 t1 = t2 - retry_interval;
177
178 for (; t2-t0 < timeout; t2=mtime()) {
179 /* This is the loop exit condition: that all queries
180 * have an accepted answer. */
181 for (i=0; i<nqueries && alens[i]>0; i++);
182 if (i==nqueries) break;
183
184 if (t2-t1 >= retry_interval) {
185 /* Query all configured namservers in parallel */
186 for (i=0; i<nqueries; i++)
187 if (!alens[i])
188 for (j=0; j<nns; j++)
189 sendto(fd, queries[i],
190 qlens[i], MSG_NOSIGNAL,
191 (void *)&ns[j], sl);
192 t1 = t2;
193 servfail_retry = 2 * nqueries;
194 }
195
196 /* Wait for a response, or until time to retry */
197 if (poll(pfd, nqueries+1, t1+retry_interval-t2) <= 0) continue;
198
199 while (next < nqueries) {
200 struct msghdr mh = {
201 .msg_name = (void *)&sa,
202 .msg_namelen = sl,
203 .msg_iovlen = 1,
204 .msg_iov = (struct iovec []){
205 { .iov_base = (void *)answers[next],
206 .iov_len = asize }
207 }
208 };
209 rlen = recvmsg(fd, &mh, 0);
210 if (rlen < 0) break;
211
212 /* Ignore non-identifiable packets */
213 if (rlen < 4) continue;
214
215 /* Ignore replies from addresses we didn't send to */
216 for (j=0; j<nns && memcmp(ns+j, &sa, sl); j++);
217 if (j==nns) continue;
218
219 /* Find which query this answer goes with, if any */
220 for (i=next; i<nqueries && (
221 answers[next][0] != queries[i][0] ||
222 answers[next][1] != queries[i][1] ); i++);
223 if (i==nqueries) continue;
224 if (alens[i]) continue;
225
226 /* Only accept positive or negative responses;
227 * retry immediately on server failure, and ignore
228 * all other codes such as refusal. */
229 switch (answers[next][3] & 15) {
230 case 0:
231 case 3:
232 break;
233 case 2:
234 if (servfail_retry && servfail_retry--)
235 sendto(fd, queries[i],
236 qlens[i], MSG_NOSIGNAL,
237 (void *)&ns[j], sl);
238 default:
239 continue;
240 }
241
242 /* Store answer in the right slot, or update next
243 * available temp slot if it's already in place. */
244 alens[i] = rlen;
245 if (i == next)
246 for (; next<nqueries && alens[next]; next++);
247 else
248 memcpy(answers[i], answers[next], rlen);
249
250 /* Ignore further UDP if all slots full or TCP-mode */
251 if (next == nqueries) pfd[nqueries].events = 0;
252
253 /* If answer is truncated (TC bit), fallback to TCP */
254 if ((answers[i][2] & 2) || (mh.msg_flags & MSG_TRUNC)) {
255 alens[i] = -1;
256 pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, 0);
257 r = start_tcp(pfd+i, family, ns+j, sl, queries[i], qlens[i]);
258 pthread_setcancelstate(cs, 0);
259 if (r >= 0) {
260 qpos[i] = r;
261 apos[i] = 0;
262 }
263 continue;
264 }
265 }
266
267 for (i=0; i<nqueries; i++) if (pfd[i].revents & POLLOUT) {
268 struct msghdr mh = {
269 .msg_iovlen = 2,
270 .msg_iov = (struct iovec [2]){
271 { .iov_base = (uint8_t[]){ qlens[i]>>8, qlens[i] }, .iov_len = 2 },
272 { .iov_base = (void *)queries[i], .iov_len = qlens[i] } }
273 };
274 step_mh(&mh, qpos[i]);
275 r = sendmsg(pfd[i].fd, &mh, MSG_NOSIGNAL);
276 if (r < 0) goto out;
277 qpos[i] += r;
278 if (qpos[i] == qlens[i]+2)
279 pfd[i].events = POLLIN;
280 }
281
282 for (i=0; i<nqueries; i++) if (pfd[i].revents & POLLIN) {
283 struct msghdr mh = {
284 .msg_iovlen = 2,
285 .msg_iov = (struct iovec [2]){
286 { .iov_base = alen_buf[i], .iov_len = 2 },
287 { .iov_base = answers[i], .iov_len = asize } }
288 };
289 step_mh(&mh, apos[i]);
290 r = recvmsg(pfd[i].fd, &mh, 0);
291 if (r <= 0) goto out;
292 apos[i] += r;
293 if (apos[i] < 2) continue;
294 int alen = alen_buf[i][0]*256 + alen_buf[i][1];
295 if (alen < 13) goto out;
296 if (apos[i] < alen+2 && apos[i] < asize+2)
297 continue;
298 int rcode = answers[i][3] & 15;
299 if (rcode != 0 && rcode != 3)
300 goto out;
301
302 /* Storing the length here commits the accepted answer.
303 * Immediately close TCP socket so as not to consume
304 * resources we no longer need. */
305 alens[i] = alen;
306 __syscall(SYS_close, pfd[i].fd);
307 pfd[i].fd = -1;
308 }
309 }
310out:
311 pthread_cleanup_pop(1);
312
313 /* Disregard any incomplete TCP results */
314 for (i=0; i<nqueries; i++) if (alens[i]<0) alens[i] = 0;
315
316 return 0;
317}
318
319int __res_msend(int nqueries, const unsigned char *const *queries,
320 const int *qlens, unsigned char *const *answers, int *alens, int asize)
321{
322 struct resolvconf conf;
323 if (__get_resolv_conf(&conf, 0, 0) < 0) return -1;
324 return __res_msend_rc(nqueries, queries, qlens, answers, alens, asize, &conf);
325}