Merge tag 'exim-4_82_1'
[exim.git] / src / src / ip.c
1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
4
5 /* Copyright (c) University of Cambridge 1995 - 2014 */
6 /* See the file NOTICE for conditions of use and distribution. */
7
8 /* Functions for doing things with sockets. With the advent of IPv6 this has
9 got messier, so that it's worth pulling out the code into separate functions
10 that other parts of Exim can call, expecially as there are now several
11 different places in the code where sockets are used. */
12
13
14 #include "exim.h"
15
16
17 /*************************************************
18 * Create a socket *
19 *************************************************/
20
21 /* Socket creation happens in a number of places so it's packaged here for
22 convenience.
23
24 Arguments:
25 type SOCK_DGRAM or SOCK_STREAM
26 af AF_INET or AF_INET6
27
28 Returns: socket number or -1 on failure
29 */
30
31 int
32 ip_socket(int type, int af)
33 {
34 int sock = socket(af, type, 0);
35 if (sock < 0)
36 log_write(0, LOG_MAIN, "IPv%c socket creation failed: %s",
37 (af == AF_INET6)? '6':'4', strerror(errno));
38 return sock;
39 }
40
41
42
43
44 #if HAVE_IPV6
45 /*************************************************
46 * Convert printing address to numeric *
47 *************************************************/
48
49 /* This function converts the textual form of an IP address into a numeric form
50 in an appropriate structure in an IPv6 environment. The getaddrinfo() function
51 can (apparently) handle more complicated addresses (e.g. those containing
52 scopes) than inet_pton() in some environments. We use hints to tell it that the
53 input must be a numeric address.
54
55 However, apparently some operating systems (or libraries) don't support
56 getaddrinfo(), so there is a build-time option to revert to inet_pton() (which
57 does not support scopes).
58
59 Arguments:
60 address textual form of the address
61 addr where to copy back the answer
62
63 Returns: nothing - failure provokes a panic-die
64 */
65
66 static void
67 ip_addrinfo(uschar *address, struct sockaddr_in6 *saddr)
68 {
69 #ifdef IPV6_USE_INET_PTON
70
71 if (inet_pton(AF_INET6, CS address, &saddr->sin6_addr) != 1)
72 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "unable to parse \"%s\" as an "
73 "IP address", address);
74 saddr->sin6_family = AF_INET6;
75
76 #else
77
78 int rc;
79 struct addrinfo hints, *res;
80 memset(&hints, 0, sizeof(hints));
81 hints.ai_family = AF_INET6;
82 hints.ai_socktype = SOCK_STREAM;
83 hints.ai_flags = AI_NUMERICHOST;
84 if ((rc = getaddrinfo(CS address, NULL, &hints, &res)) != 0 || res == NULL)
85 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "unable to parse \"%s\" as an "
86 "IP address: %s", address,
87 (rc == 0)? "NULL result returned" : gai_strerror(rc));
88 memcpy(saddr, res->ai_addr, res->ai_addrlen);
89 freeaddrinfo(res);
90
91 #endif
92 }
93 #endif /* HAVE_IPV6 */
94
95
96 /*************************************************
97 * Bind socket to interface and port *
98 *************************************************/
99
100 /* This function binds a socket to a local interface address and port. For a
101 wildcard IPv6 bind, the address is ":".
102
103 Arguments:
104 sock the socket
105 af AF_INET or AF_INET6 - the socket type
106 address the IP address, in text form
107 port the IP port (host order)
108
109 Returns: the result of bind()
110 */
111
112 int
113 ip_bind(int sock, int af, uschar *address, int port)
114 {
115 int s_len;
116 union sockaddr_46 sin;
117 memset(&sin, 0, sizeof(sin));
118
119 /* Setup code when using an IPv6 socket. The wildcard address is ":", to
120 ensure an IPv6 socket is used. */
121
122 #if HAVE_IPV6
123 if (af == AF_INET6)
124 {
125 if (address[0] == ':' && address[1] == 0)
126 {
127 sin.v6.sin6_family = AF_INET6;
128 sin.v6.sin6_addr = in6addr_any;
129 }
130 else
131 {
132 ip_addrinfo(address, &sin.v6); /* Panic-dies on error */
133 }
134 sin.v6.sin6_port = htons(port);
135 s_len = sizeof(sin.v6);
136 }
137 else
138 #else /* HAVE_IPv6 */
139 af = af; /* Avoid compiler warning */
140 #endif /* HAVE_IPV6 */
141
142 /* Setup code when using IPv4 socket. The wildcard address is "". */
143
144 {
145 sin.v4.sin_family = AF_INET;
146 sin.v4.sin_port = htons(port);
147 s_len = sizeof(sin.v4);
148 if (address[0] == 0)
149 sin.v4.sin_addr.s_addr = (S_ADDR_TYPE)INADDR_ANY;
150 else
151 sin.v4.sin_addr.s_addr = (S_ADDR_TYPE)inet_addr(CS address);
152 }
153
154 /* Now we can call the bind() function */
155
156 return bind(sock, (struct sockaddr *)&sin, s_len);
157 }
158
159
160
161 /*************************************************
162 * Connect socket to remote host *
163 *************************************************/
164
165 /* This function connects a socket to a remote address and port. The socket may
166 or may not have previously been bound to a local interface. The socket is not
167 closed, even in cases of error. It is expected that the calling function, which
168 created the socket, will be the one that closes it.
169
170 Arguments:
171 sock the socket
172 af AF_INET6 or AF_INET for the socket type
173 address the remote address, in text form
174 port the remote port
175 timeout a timeout (zero for indefinite timeout)
176
177 Returns: 0 on success; -1 on failure, with errno set
178 */
179
180 int
181 ip_connect(int sock, int af, uschar *address, int port, int timeout)
182 {
183 struct sockaddr_in s_in4;
184 struct sockaddr *s_ptr;
185 int s_len, rc, save_errno;
186
187 /* For an IPv6 address, use an IPv6 sockaddr structure. */
188
189 #if HAVE_IPV6
190 struct sockaddr_in6 s_in6;
191 if (af == AF_INET6)
192 {
193 memset(&s_in6, 0, sizeof(s_in6));
194 ip_addrinfo(address, &s_in6); /* Panic-dies on error */
195 s_in6.sin6_port = htons(port);
196 s_ptr = (struct sockaddr *)&s_in6;
197 s_len = sizeof(s_in6);
198 }
199 else
200 #else /* HAVE_IPV6 */
201 af = af; /* Avoid compiler warning */
202 #endif /* HAVE_IPV6 */
203
204 /* For an IPv4 address, use an IPv4 sockaddr structure, even on a system with
205 IPv6 support. */
206
207 {
208 memset(&s_in4, 0, sizeof(s_in4));
209 s_in4.sin_family = AF_INET;
210 s_in4.sin_port = htons(port);
211 s_in4.sin_addr.s_addr = (S_ADDR_TYPE)inet_addr(CS address);
212 s_ptr = (struct sockaddr *)&s_in4;
213 s_len = sizeof(s_in4);
214 }
215
216 /* If no connection timeout is set, just call connect() without setting a
217 timer, thereby allowing the inbuilt OS timeout to operate. */
218
219 sigalrm_seen = FALSE;
220 if (timeout > 0) alarm(timeout);
221 rc = connect(sock, s_ptr, s_len);
222 save_errno = errno;
223 alarm(0);
224
225 /* There is a testing facility for simulating a connection timeout, as I
226 can't think of any other way of doing this. It converts a connection refused
227 into a timeout if the timeout is set to 999999. */
228
229 if (running_in_test_harness)
230 {
231 if (save_errno == ECONNREFUSED && timeout == 999999)
232 {
233 rc = -1;
234 save_errno = EINTR;
235 sigalrm_seen = TRUE;
236 }
237 }
238
239 /* Success */
240
241 if (rc >= 0) return 0;
242
243 /* A failure whose error code is "Interrupted system call" is in fact
244 an externally applied timeout if the signal handler has been run. */
245
246 errno = (save_errno == EINTR && sigalrm_seen)? ETIMEDOUT : save_errno;
247 return -1;
248 }
249
250
251 /* Create a socket and connect to host (name or number, ipv6 ok)
252 at one of port-range.
253 Arguments:
254 type SOCK_DGRAM or SOCK_STREAM
255 af AF_INET6 or AF_INET for the socket type
256 address the remote address, in text form
257 portlo,porthi the remote port range
258 timeout a timeout
259 connhost if not NULL, host_item filled in with connection details
260 errstr pointer for allocated string on error
261
262 Return:
263 socket fd, or -1 on failure (having allocated an error string)
264 */
265 int
266 ip_connectedsocket(int type, const uschar * hostname, int portlo, int porthi,
267 int timeout, host_item * connhost, uschar ** errstr)
268 {
269 int namelen, port;
270 host_item shost;
271 host_item *h;
272 int af = 0, fd, fd4 = -1, fd6 = -1;
273
274 shost.next = NULL;
275 shost.address = NULL;
276 shost.port = portlo;
277 shost.mx = -1;
278
279 namelen = Ustrlen(hostname);
280
281 /* Anything enclosed in [] must be an IP address. */
282
283 if (hostname[0] == '[' &&
284 hostname[namelen - 1] == ']')
285 {
286 uschar * host = string_copy(hostname);
287 host[namelen - 1] = 0;
288 host++;
289 if (string_is_ip_address(host, NULL) == 0)
290 {
291 *errstr = string_sprintf("malformed IP address \"%s\"", hostname);
292 return -1;
293 }
294 shost.name = shost.address = host;
295 }
296
297 /* Otherwise check for an unadorned IP address */
298
299 else if (string_is_ip_address(hostname, NULL) != 0)
300 shost.name = shost.address = string_copy(hostname);
301
302 /* Otherwise lookup IP address(es) from the name */
303
304 else
305 {
306 shost.name = string_copy(hostname);
307 if (host_find_byname(&shost, NULL, HOST_FIND_QUALIFY_SINGLE, NULL,
308 FALSE) != HOST_FOUND)
309 {
310 *errstr = string_sprintf("no IP address found for host %s", shost.name);
311 return -1;
312 }
313 }
314
315 /* Try to connect to the server - test each IP till one works */
316
317 for (h = &shost; h != NULL; h = h->next)
318 {
319 fd = (Ustrchr(h->address, ':') != 0)
320 ? (fd6 < 0) ? (fd6 = ip_socket(SOCK_STREAM, af = AF_INET6)) : fd6
321 : (fd4 < 0) ? (fd4 = ip_socket(SOCK_STREAM, af = AF_INET )) : fd4;
322
323 if (fd < 0)
324 {
325 *errstr = string_sprintf("failed to create socket: %s", strerror(errno));
326 goto bad;
327 }
328
329 for(port = portlo; port <= porthi; port++)
330 if (ip_connect(fd, af, h->address, port, timeout) == 0)
331 {
332 if (fd != fd6) close(fd6);
333 if (fd != fd4) close(fd4);
334 if (connhost) {
335 h->port = port;
336 *connhost = *h;
337 connhost->next = NULL;
338 }
339 return fd;
340 }
341 }
342
343 *errstr = string_sprintf("failed to connect to "
344 "%s: couldn't connect to any host", hostname, strerror(errno));
345
346 bad:
347 close(fd4); close(fd6); return -1;
348 }
349
350
351 /*************************************************
352 * Set keepalive on a socket *
353 *************************************************/
354
355 /* Can be called for both incoming and outgoing sockets.
356
357 Arguments:
358 sock the socket
359 address the remote host address, for failure logging
360 torf true for outgoing connection, false for incoming
361
362 Returns: nothing
363 */
364
365 void
366 ip_keepalive(int sock, uschar *address, BOOL torf)
367 {
368 int fodder = 1;
369 if (setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
370 (uschar *)(&fodder), sizeof(fodder)) != 0)
371 log_write(0, LOG_MAIN, "setsockopt(SO_KEEPALIVE) on connection %s %s "
372 "failed: %s", torf? "to":"from", address, strerror(errno));
373 }
374
375
376
377 /*************************************************
378 * Receive from a socket with timeout *
379 *************************************************/
380
381 /* The timeout is implemented using select(), and we loop to cover select()
382 getting interrupted, and the possibility of select() returning with a positive
383 result but no ready descriptor. Is this in fact possible?
384
385 Arguments:
386 sock the socket
387 buffer to read into
388 bufsize the buffer size
389 timeout the timeout
390
391 Returns: > 0 => that much data read
392 <= 0 on error or EOF; errno set - zero for EOF
393 */
394
395 int
396 ip_recv(int sock, uschar *buffer, int buffsize, int timeout)
397 {
398 fd_set select_inset;
399 struct timeval tv;
400 int start_recv = time(NULL);
401 int rc;
402
403 /* Wait until the socket is ready */
404
405 for (;;)
406 {
407 FD_ZERO (&select_inset);
408 FD_SET (sock, &select_inset);
409 tv.tv_sec = timeout;
410 tv.tv_usec = 0;
411
412 DEBUG(D_transport) debug_printf("waiting for data on socket\n");
413 rc = select(sock + 1, (SELECT_ARG2_TYPE *)&select_inset, NULL, NULL, &tv);
414
415 /* If some interrupt arrived, just retry. We presume this to be rare,
416 but it can happen (e.g. the SIGUSR1 signal sent by exiwhat causes
417 select() to exit).
418
419 Aug 2004: Somebody set up a cron job that ran exiwhat every 2 minutes, making
420 the interrupt not at all rare. Since the timeout is typically more than 2
421 minutes, the effect was to block the timeout completely. To prevent this
422 happening again, we do an explicit time test. */
423
424 if (rc < 0 && errno == EINTR)
425 {
426 DEBUG(D_transport) debug_printf("EINTR while waiting for socket data\n");
427 if (time(NULL) - start_recv < timeout) continue;
428 DEBUG(D_transport) debug_printf("total wait time exceeds timeout\n");
429 }
430
431 /* Handle a timeout, and treat any other select error as a timeout, including
432 an EINTR when we have been in this loop for longer than timeout. */
433
434 if (rc <= 0)
435 {
436 errno = ETIMEDOUT;
437 return -1;
438 }
439
440 /* If the socket is ready, break out of the loop. */
441
442 if (FD_ISSET(sock, &select_inset)) break;
443 }
444
445 /* The socket is ready, read from it (via TLS if it's active). On EOF (i.e.
446 close down of the connection), set errno to zero; otherwise leave it alone. */
447
448 #ifdef SUPPORT_TLS
449 if (tls_out.active == sock)
450 rc = tls_read(FALSE, buffer, buffsize);
451 else if (tls_in.active == sock)
452 rc = tls_read(TRUE, buffer, buffsize);
453 else
454 #endif
455 rc = recv(sock, buffer, buffsize, 0);
456
457 if (rc > 0) return rc;
458 if (rc == 0) errno = 0;
459 return -1;
460 }
461
462
463
464
465 /*************************************************
466 * Lookup address family of potential socket *
467 *************************************************/
468
469 /* Given a file-descriptor, check to see if it's a socket and, if so,
470 return the address family; detects IPv4 vs IPv6. If not a socket then
471 return -1.
472
473 The value 0 is typically AF_UNSPEC, which should not be seen on a connected
474 fd. If the return is -1, the errno will be from getsockname(); probably
475 ENOTSOCK or ECONNRESET.
476
477 Arguments: socket-or-not fd
478 Returns: address family or -1
479 */
480
481 int
482 ip_get_address_family(int fd)
483 {
484 struct sockaddr_storage ss;
485 socklen_t sslen = sizeof(ss);
486
487 if (getsockname(fd, (struct sockaddr *) &ss, &sslen) < 0)
488 return -1;
489
490 return (int) ss.ss_family;
491 }
492
493
494
495
496 /*************************************************
497 * Lookup DSCP settings for a socket *
498 *************************************************/
499
500 struct dscp_name_tableentry {
501 const uschar *name;
502 int value;
503 };
504 /* Keep both of these tables sorted! */
505 static struct dscp_name_tableentry dscp_table[] = {
506 #ifdef IPTOS_DSCP_AF11
507 { CUS"af11", IPTOS_DSCP_AF11 },
508 { CUS"af12", IPTOS_DSCP_AF12 },
509 { CUS"af13", IPTOS_DSCP_AF13 },
510 { CUS"af21", IPTOS_DSCP_AF21 },
511 { CUS"af22", IPTOS_DSCP_AF22 },
512 { CUS"af23", IPTOS_DSCP_AF23 },
513 { CUS"af31", IPTOS_DSCP_AF31 },
514 { CUS"af32", IPTOS_DSCP_AF32 },
515 { CUS"af33", IPTOS_DSCP_AF33 },
516 { CUS"af41", IPTOS_DSCP_AF41 },
517 { CUS"af42", IPTOS_DSCP_AF42 },
518 { CUS"af43", IPTOS_DSCP_AF43 },
519 { CUS"ef", IPTOS_DSCP_EF },
520 #endif
521 #ifdef IPTOS_LOWCOST
522 { CUS"lowcost", IPTOS_LOWCOST },
523 #endif
524 { CUS"lowdelay", IPTOS_LOWDELAY },
525 #ifdef IPTOS_MINCOST
526 { CUS"mincost", IPTOS_MINCOST },
527 #endif
528 { CUS"reliability", IPTOS_RELIABILITY },
529 { CUS"throughput", IPTOS_THROUGHPUT }
530 };
531 static int dscp_table_size =
532 sizeof(dscp_table) / sizeof(struct dscp_name_tableentry);
533
534 /* DSCP values change by protocol family, and so do the options used for
535 setsockopt(); this utility does all the lookups. It takes an unexpanded
536 option string, expands it, strips off affix whitespace, then checks if it's
537 a number. If all of what's left is a number, then that's how the option will
538 be parsed and success/failure is a range check. If it's not all a number,
539 then it must be a supported keyword.
540
541 Arguments:
542 dscp_name a string, so far unvalidated
543 af address_family in use
544 level setsockopt level to use
545 optname setsockopt name to use
546 dscp_value value for dscp_name
547
548 Returns: TRUE if okay to setsockopt(), else FALSE
549
550 *level and *optname may be set even if FALSE is returned
551 */
552
553 BOOL
554 dscp_lookup(const uschar *dscp_name, int af,
555 int *level, int *optname, int *dscp_value)
556 {
557 uschar *dscp_lookup, *p;
558 int first, last;
559 long rawlong;
560
561 if (af == AF_INET)
562 {
563 *level = IPPROTO_IP;
564 *optname = IP_TOS;
565 }
566 #if HAVE_IPV6 && defined(IPV6_TCLASS)
567 else if (af == AF_INET6)
568 {
569 *level = IPPROTO_IPV6;
570 *optname = IPV6_TCLASS;
571 }
572 #endif
573 else
574 {
575 DEBUG(D_transport)
576 debug_printf("Unhandled address family %d in dscp_lookup()\n", af);
577 return FALSE;
578 }
579 if (!dscp_name)
580 {
581 DEBUG(D_transport)
582 debug_printf("[empty DSCP]\n");
583 return FALSE;
584 }
585 dscp_lookup = expand_string(US dscp_name);
586 if (dscp_lookup == NULL || *dscp_lookup == '\0')
587 return FALSE;
588
589 p = dscp_lookup + Ustrlen(dscp_lookup) - 1;
590 while (isspace(*p)) *p-- = '\0';
591 while (isspace(*dscp_lookup) && dscp_lookup < p) dscp_lookup++;
592 if (*dscp_lookup == '\0')
593 return FALSE;
594
595 rawlong = Ustrtol(dscp_lookup, &p, 0);
596 if (p != dscp_lookup && *p == '\0')
597 {
598 /* We have six bits available, which will end up shifted to fit in 0xFC mask.
599 RFC 2597 defines the values unshifted. */
600 if (rawlong < 0 || rawlong > 0x3F)
601 {
602 DEBUG(D_transport)
603 debug_printf("DSCP value %ld out of range, ignored.\n", rawlong);
604 return FALSE;
605 }
606 *dscp_value = rawlong << 2;
607 return TRUE;
608 }
609
610 first = 0;
611 last = dscp_table_size;
612 while (last > first)
613 {
614 int middle = (first + last)/2;
615 int c = Ustrcmp(dscp_lookup, dscp_table[middle].name);
616 if (c == 0)
617 {
618 *dscp_value = dscp_table[middle].value;
619 return TRUE;
620 }
621 else if (c > 0)
622 {
623 first = middle + 1;
624 }
625 else
626 {
627 last = middle;
628 }
629 }
630 return FALSE;
631 }
632
633 void
634 dscp_list_to_stream(FILE *stream)
635 {
636 int i;
637 for (i=0; i < dscp_table_size; ++i)
638 fprintf(stream, "%s\n", dscp_table[i].name);
639 }
640
641
642 /* End of ip.c */