f6072c2e80019e4df87e34f3d490f956bb8c39b9
[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(const uschar *address, struct sockaddr_in6 *saddr)
68 {
69 #ifdef IPV6_USE_INET_PTON
70
71 if (inet_pton(AF_INET6, CCS 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(CCS 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, const 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(CCS 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
252 /*************************************************
253 * Create connected socket to remote host *
254 *************************************************/
255
256 /* Create a socket and connect to host (name or number, ipv6 ok)
257 at one of port-range.
258
259 Arguments:
260 type SOCK_DGRAM or SOCK_STREAM
261 af AF_INET6 or AF_INET for the socket type
262 address the remote address, in text form
263 portlo,porthi the remote port range
264 timeout a timeout
265 connhost if not NULL, host_item filled in with connection details
266 errstr pointer for allocated string on error
267
268 Return:
269 socket fd, or -1 on failure (having allocated an error string)
270 */
271 int
272 ip_connectedsocket(int type, const uschar * hostname, int portlo, int porthi,
273 int timeout, host_item * connhost, uschar ** errstr)
274 {
275 int namelen, port;
276 host_item shost;
277 host_item *h;
278 int af = 0, fd, fd4 = -1, fd6 = -1;
279
280 shost.next = NULL;
281 shost.address = NULL;
282 shost.port = portlo;
283 shost.mx = -1;
284
285 namelen = Ustrlen(hostname);
286
287 /* Anything enclosed in [] must be an IP address. */
288
289 if (hostname[0] == '[' &&
290 hostname[namelen - 1] == ']')
291 {
292 uschar * host = string_copy(hostname);
293 host[namelen - 1] = 0;
294 host++;
295 if (string_is_ip_address(host, NULL) == 0)
296 {
297 *errstr = string_sprintf("malformed IP address \"%s\"", hostname);
298 return -1;
299 }
300 shost.name = shost.address = host;
301 }
302
303 /* Otherwise check for an unadorned IP address */
304
305 else if (string_is_ip_address(hostname, NULL) != 0)
306 shost.name = shost.address = string_copy(hostname);
307
308 /* Otherwise lookup IP address(es) from the name */
309
310 else
311 {
312 shost.name = string_copy(hostname);
313 if (host_find_byname(&shost, NULL, HOST_FIND_QUALIFY_SINGLE, NULL,
314 FALSE) != HOST_FOUND)
315 {
316 *errstr = string_sprintf("no IP address found for host %s", shost.name);
317 return -1;
318 }
319 }
320
321 /* Try to connect to the server - test each IP till one works */
322
323 for (h = &shost; h != NULL; h = h->next)
324 {
325 fd = (Ustrchr(h->address, ':') != 0)
326 ? (fd6 < 0) ? (fd6 = ip_socket(type, af = AF_INET6)) : fd6
327 : (fd4 < 0) ? (fd4 = ip_socket(type, af = AF_INET )) : fd4;
328
329 if (fd < 0)
330 {
331 *errstr = string_sprintf("failed to create socket: %s", strerror(errno));
332 goto bad;
333 }
334
335 for(port = portlo; port <= porthi; port++)
336 if (ip_connect(fd, af, h->address, port, timeout) == 0)
337 {
338 if (fd != fd6) close(fd6);
339 if (fd != fd4) close(fd4);
340 if (connhost)
341 {
342 h->port = port;
343 *connhost = *h;
344 connhost->next = NULL;
345 }
346 return fd;
347 }
348 }
349
350 *errstr = string_sprintf("failed to connect to any address for %s: %s",
351 hostname, strerror(errno));
352
353 bad:
354 close(fd4); close(fd6); return -1;
355 }
356
357
358 int
359 ip_tcpsocket(const uschar * hostport, uschar ** errstr, int tmo)
360 {
361 int scan;
362 uschar hostname[256];
363 unsigned int portlow, porthigh;
364
365 /* extract host and port part */
366 scan = sscanf(CS hostport, "%255s %u-%u", hostname, &portlow, &porthigh);
367 if ( scan != 3 ) {
368 if ( scan != 2 ) {
369 *errstr = string_sprintf("invalid socket '%s'", hostport);
370 return -1;
371 }
372 porthigh = portlow;
373 }
374
375 return ip_connectedsocket(SOCK_STREAM, hostname, portlow, porthigh,
376 tmo, NULL, errstr);
377 }
378
379 int
380 ip_unixsocket(const uschar * path, uschar ** errstr)
381 {
382 int sock;
383 struct sockaddr_un server;
384
385 if ((sock = socket(AF_UNIX, SOCK_STREAM, 0)) < 0) {
386 *errstr = US"can't open UNIX socket.";
387 return -1;
388 }
389
390 server.sun_family = AF_UNIX;
391 Ustrncpy(server.sun_path, path, sizeof(server.sun_path)-1);
392 server.sun_path[sizeof(server.sun_path)-1] = '\0';
393 if (connect(sock, (struct sockaddr *) &server, sizeof(server)) < 0) {
394 int err = errno;
395 (void)close(sock);
396 *errstr = string_sprintf("unable to connect to UNIX socket (%s): %s",
397 path, strerror(err));
398 return -1;
399 }
400 return sock;
401 }
402
403 int
404 ip_streamsocket(const uschar * spec, uschar ** errstr, int tmo)
405 {
406 return *spec == '/'
407 ? ip_unixsocket(spec, errstr) : ip_tcpsocket(spec, errstr, tmo);
408 }
409
410 /*************************************************
411 * Set keepalive on a socket *
412 *************************************************/
413
414 /* Can be called for both incoming and outgoing sockets.
415
416 Arguments:
417 sock the socket
418 address the remote host address, for failure logging
419 torf true for outgoing connection, false for incoming
420
421 Returns: nothing
422 */
423
424 void
425 ip_keepalive(int sock, const uschar *address, BOOL torf)
426 {
427 int fodder = 1;
428 if (setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
429 (uschar *)(&fodder), sizeof(fodder)) != 0)
430 log_write(0, LOG_MAIN, "setsockopt(SO_KEEPALIVE) on connection %s %s "
431 "failed: %s", torf? "to":"from", address, strerror(errno));
432 }
433
434
435
436 /*************************************************
437 * Receive from a socket with timeout *
438 *************************************************/
439
440 /*
441 Arguments:
442 fd the file descriptor
443 timeout the timeout, seconds
444 Returns: TRUE => ready for i/o
445 FALSE => timed out, or other error
446 */
447 BOOL
448 fd_ready(int fd, int timeout)
449 {
450 fd_set select_inset;
451 struct timeval tv;
452 time_t start_recv = time(NULL);
453 int rc;
454
455 if (timeout <= 0)
456 {
457 errno = ETIMEDOUT;
458 return FALSE;
459 }
460 /* Wait until the socket is ready */
461
462 for (;;)
463 {
464 FD_ZERO (&select_inset);
465 FD_SET (fd, &select_inset);
466 tv.tv_sec = timeout;
467 tv.tv_usec = 0;
468
469 /*DEBUG(D_transport) debug_printf("waiting for data on fd\n");*/
470 rc = select(fd + 1, (SELECT_ARG2_TYPE *)&select_inset, NULL, NULL, &tv);
471
472 /* If some interrupt arrived, just retry. We presume this to be rare,
473 but it can happen (e.g. the SIGUSR1 signal sent by exiwhat causes
474 select() to exit).
475
476 Aug 2004: Somebody set up a cron job that ran exiwhat every 2 minutes, making
477 the interrupt not at all rare. Since the timeout is typically more than 2
478 minutes, the effect was to block the timeout completely. To prevent this
479 happening again, we do an explicit time test. */
480
481 if (rc < 0 && errno == EINTR)
482 {
483 DEBUG(D_transport) debug_printf("EINTR while waiting for socket data\n");
484 if (time(NULL) - start_recv < timeout) continue;
485 DEBUG(D_transport) debug_printf("total wait time exceeds timeout\n");
486 }
487
488 /* Handle a timeout, and treat any other select error as a timeout, including
489 an EINTR when we have been in this loop for longer than timeout. */
490
491 if (rc <= 0)
492 {
493 errno = ETIMEDOUT;
494 return FALSE;
495 }
496
497 /* If the socket is ready, break out of the loop. */
498
499 if (FD_ISSET(fd, &select_inset)) break;
500 }
501 return TRUE;
502 }
503
504 /* The timeout is implemented using select(), and we loop to cover select()
505 getting interrupted, and the possibility of select() returning with a positive
506 result but no ready descriptor. Is this in fact possible?
507
508 Arguments:
509 sock the socket
510 buffer to read into
511 bufsize the buffer size
512 timeout the timeout
513
514 Returns: > 0 => that much data read
515 <= 0 on error or EOF; errno set - zero for EOF
516 */
517
518 int
519 ip_recv(int sock, uschar *buffer, int buffsize, int timeout)
520 {
521 int rc;
522
523 if (!fd_ready(sock, timeout))
524 return -1;
525
526 /* The socket is ready, read from it (via TLS if it's active). On EOF (i.e.
527 close down of the connection), set errno to zero; otherwise leave it alone. */
528
529 #ifdef SUPPORT_TLS
530 if (tls_out.active == sock)
531 rc = tls_read(FALSE, buffer, buffsize);
532 else if (tls_in.active == sock)
533 rc = tls_read(TRUE, buffer, buffsize);
534 else
535 #endif
536 rc = recv(sock, buffer, buffsize, 0);
537
538 if (rc > 0) return rc;
539 if (rc == 0) errno = 0;
540 return -1;
541 }
542
543
544
545
546 /*************************************************
547 * Lookup address family of potential socket *
548 *************************************************/
549
550 /* Given a file-descriptor, check to see if it's a socket and, if so,
551 return the address family; detects IPv4 vs IPv6. If not a socket then
552 return -1.
553
554 The value 0 is typically AF_UNSPEC, which should not be seen on a connected
555 fd. If the return is -1, the errno will be from getsockname(); probably
556 ENOTSOCK or ECONNRESET.
557
558 Arguments: socket-or-not fd
559 Returns: address family or -1
560 */
561
562 int
563 ip_get_address_family(int fd)
564 {
565 struct sockaddr_storage ss;
566 socklen_t sslen = sizeof(ss);
567
568 if (getsockname(fd, (struct sockaddr *) &ss, &sslen) < 0)
569 return -1;
570
571 return (int) ss.ss_family;
572 }
573
574
575
576
577 /*************************************************
578 * Lookup DSCP settings for a socket *
579 *************************************************/
580
581 struct dscp_name_tableentry {
582 const uschar *name;
583 int value;
584 };
585 /* Keep both of these tables sorted! */
586 static struct dscp_name_tableentry dscp_table[] = {
587 #ifdef IPTOS_DSCP_AF11
588 { CUS"af11", IPTOS_DSCP_AF11 },
589 { CUS"af12", IPTOS_DSCP_AF12 },
590 { CUS"af13", IPTOS_DSCP_AF13 },
591 { CUS"af21", IPTOS_DSCP_AF21 },
592 { CUS"af22", IPTOS_DSCP_AF22 },
593 { CUS"af23", IPTOS_DSCP_AF23 },
594 { CUS"af31", IPTOS_DSCP_AF31 },
595 { CUS"af32", IPTOS_DSCP_AF32 },
596 { CUS"af33", IPTOS_DSCP_AF33 },
597 { CUS"af41", IPTOS_DSCP_AF41 },
598 { CUS"af42", IPTOS_DSCP_AF42 },
599 { CUS"af43", IPTOS_DSCP_AF43 },
600 { CUS"ef", IPTOS_DSCP_EF },
601 #endif
602 #ifdef IPTOS_LOWCOST
603 { CUS"lowcost", IPTOS_LOWCOST },
604 #endif
605 { CUS"lowdelay", IPTOS_LOWDELAY },
606 #ifdef IPTOS_MINCOST
607 { CUS"mincost", IPTOS_MINCOST },
608 #endif
609 { CUS"reliability", IPTOS_RELIABILITY },
610 { CUS"throughput", IPTOS_THROUGHPUT }
611 };
612 static int dscp_table_size =
613 sizeof(dscp_table) / sizeof(struct dscp_name_tableentry);
614
615 /* DSCP values change by protocol family, and so do the options used for
616 setsockopt(); this utility does all the lookups. It takes an unexpanded
617 option string, expands it, strips off affix whitespace, then checks if it's
618 a number. If all of what's left is a number, then that's how the option will
619 be parsed and success/failure is a range check. If it's not all a number,
620 then it must be a supported keyword.
621
622 Arguments:
623 dscp_name a string, so far unvalidated
624 af address_family in use
625 level setsockopt level to use
626 optname setsockopt name to use
627 dscp_value value for dscp_name
628
629 Returns: TRUE if okay to setsockopt(), else FALSE
630
631 *level and *optname may be set even if FALSE is returned
632 */
633
634 BOOL
635 dscp_lookup(const uschar *dscp_name, int af,
636 int *level, int *optname, int *dscp_value)
637 {
638 uschar *dscp_lookup, *p;
639 int first, last;
640 long rawlong;
641
642 if (af == AF_INET)
643 {
644 *level = IPPROTO_IP;
645 *optname = IP_TOS;
646 }
647 #if HAVE_IPV6 && defined(IPV6_TCLASS)
648 else if (af == AF_INET6)
649 {
650 *level = IPPROTO_IPV6;
651 *optname = IPV6_TCLASS;
652 }
653 #endif
654 else
655 {
656 DEBUG(D_transport)
657 debug_printf("Unhandled address family %d in dscp_lookup()\n", af);
658 return FALSE;
659 }
660 if (!dscp_name)
661 {
662 DEBUG(D_transport)
663 debug_printf("[empty DSCP]\n");
664 return FALSE;
665 }
666 dscp_lookup = expand_string(US dscp_name);
667 if (dscp_lookup == NULL || *dscp_lookup == '\0')
668 return FALSE;
669
670 p = dscp_lookup + Ustrlen(dscp_lookup) - 1;
671 while (isspace(*p)) *p-- = '\0';
672 while (isspace(*dscp_lookup) && dscp_lookup < p) dscp_lookup++;
673 if (*dscp_lookup == '\0')
674 return FALSE;
675
676 rawlong = Ustrtol(dscp_lookup, &p, 0);
677 if (p != dscp_lookup && *p == '\0')
678 {
679 /* We have six bits available, which will end up shifted to fit in 0xFC mask.
680 RFC 2597 defines the values unshifted. */
681 if (rawlong < 0 || rawlong > 0x3F)
682 {
683 DEBUG(D_transport)
684 debug_printf("DSCP value %ld out of range, ignored.\n", rawlong);
685 return FALSE;
686 }
687 *dscp_value = rawlong << 2;
688 return TRUE;
689 }
690
691 first = 0;
692 last = dscp_table_size;
693 while (last > first)
694 {
695 int middle = (first + last)/2;
696 int c = Ustrcmp(dscp_lookup, dscp_table[middle].name);
697 if (c == 0)
698 {
699 *dscp_value = dscp_table[middle].value;
700 return TRUE;
701 }
702 else if (c > 0)
703 {
704 first = middle + 1;
705 }
706 else
707 {
708 last = middle;
709 }
710 }
711 return FALSE;
712 }
713
714 void
715 dscp_list_to_stream(FILE *stream)
716 {
717 int i;
718 for (i=0; i < dscp_table_size; ++i)
719 fprintf(stream, "%s\n", dscp_table[i].name);
720 }
721
722
723 /* End of ip.c */
724 /* vi: aw ai sw=2
725 */