Merge pull request #30 from SpamExperts/master
[exim.git] / src / src / ip.c
CommitLineData
059ec3d9
PH
1/*************************************************
2* Exim - an Internet mail transport agent *
3*************************************************/
4
5a66c31b 5/* Copyright (c) University of Cambridge 1995 - 2014 */
059ec3d9
PH
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
9got messier, so that it's worth pulling out the code into separate functions
10that other parts of Exim can call, expecially as there are now several
11different 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
22convenience.
23
24Arguments:
25 type SOCK_DGRAM or SOCK_STREAM
26 af AF_INET or AF_INET6
27
28Returns: socket number or -1 on failure
29*/
30
31int
32ip_socket(int type, int af)
33{
34int sock = socket(af, type, 0);
35if (sock < 0)
36 log_write(0, LOG_MAIN, "IPv%c socket creation failed: %s",
37 (af == AF_INET6)? '6':'4', strerror(errno));
38return 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
50in an appropriate structure in an IPv6 environment. The getaddrinfo() function
51can (apparently) handle more complicated addresses (e.g. those containing
52scopes) than inet_pton() in some environments. We use hints to tell it that the
53input must be a numeric address.
54
55However, apparently some operating systems (or libraries) don't support
56getaddrinfo(), so there is a build-time option to revert to inet_pton() (which
57does not support scopes).
58
59Arguments:
60 address textual form of the address
61 addr where to copy back the answer
62
63Returns: nothing - failure provokes a panic-die
64*/
65
66static void
a56cc2b8 67ip_addrinfo(const uschar *address, struct sockaddr_in6 *saddr)
059ec3d9
PH
68{
69#ifdef IPV6_USE_INET_PTON
70
a56cc2b8 71 if (inet_pton(AF_INET6, CCS address, &saddr->sin6_addr) != 1)
059ec3d9
PH
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;
a56cc2b8 84 if ((rc = getaddrinfo(CCS address, NULL, &hints, &res)) != 0 || res == NULL)
059ec3d9
PH
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
101wildcard IPv6 bind, the address is ":".
102
103Arguments:
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
109Returns: the result of bind()
110*/
111
112int
113ip_bind(int sock, int af, uschar *address, int port)
114{
115int s_len;
116union sockaddr_46 sin;
117memset(&sin, 0, sizeof(sin));
118
119/* Setup code when using an IPv6 socket. The wildcard address is ":", to
120ensure an IPv6 socket is used. */
121
122#if HAVE_IPV6
123if (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 }
137else
138#else /* HAVE_IPv6 */
139af = 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
156return 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
d515a917
PH
166or may not have previously been bound to a local interface. The socket is not
167closed, even in cases of error. It is expected that the calling function, which
168created the socket, will be the one that closes it.
059ec3d9
PH
169
170Arguments:
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
b1f8e4f8 175 timeout a timeout (zero for indefinite timeout)
059ec3d9
PH
176
177Returns: 0 on success; -1 on failure, with errno set
178*/
179
180int
a56cc2b8 181ip_connect(int sock, int af, const uschar *address, int port, int timeout)
059ec3d9
PH
182{
183struct sockaddr_in s_in4;
184struct sockaddr *s_ptr;
185int s_len, rc, save_errno;
186
187/* For an IPv6 address, use an IPv6 sockaddr structure. */
188
189#if HAVE_IPV6
190struct sockaddr_in6 s_in6;
191if (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 }
199else
200#else /* HAVE_IPV6 */
201af = af; /* Avoid compiler warning */
202#endif /* HAVE_IPV6 */
203
204/* For an IPv4 address, use an IPv4 sockaddr structure, even on a system with
205IPv6 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);
a56cc2b8 211 s_in4.sin_addr.s_addr = (S_ADDR_TYPE)inet_addr(CCS address);
059ec3d9
PH
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
217timer, thereby allowing the inbuilt OS timeout to operate. */
218
219sigalrm_seen = FALSE;
220if (timeout > 0) alarm(timeout);
221rc = connect(sock, s_ptr, s_len);
222save_errno = errno;
223alarm(0);
224
225/* There is a testing facility for simulating a connection timeout, as I
226can't think of any other way of doing this. It converts a connection refused
75e0e026 227into a timeout if the timeout is set to 999999. */
059ec3d9
PH
228
229if (running_in_test_harness)
230 {
75e0e026 231 if (save_errno == ECONNREFUSED && timeout == 999999)
059ec3d9
PH
232 {
233 rc = -1;
234 save_errno = EINTR;
235 sigalrm_seen = TRUE;
236 }
237 }
238
239/* Success */
240
241if (rc >= 0) return 0;
242
243/* A failure whose error code is "Interrupted system call" is in fact
244an externally applied timeout if the signal handler has been run. */
245
059ec3d9
PH
246errno = (save_errno == EINTR && sigalrm_seen)? ETIMEDOUT : save_errno;
247return -1;
248}
249
250
a6d4c44e
TF
251
252/*************************************************
253* Create connected socket to remote host *
254*************************************************/
255
b1f8e4f8
JH
256/* Create a socket and connect to host (name or number, ipv6 ok)
257 at one of port-range.
a6d4c44e 258
b1f8e4f8
JH
259Arguments:
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
268Return:
269 socket fd, or -1 on failure (having allocated an error string)
270*/
271int
272ip_connectedsocket(int type, const uschar * hostname, int portlo, int porthi,
273 int timeout, host_item * connhost, uschar ** errstr)
274{
275int namelen, port;
276host_item shost;
277host_item *h;
278int af = 0, fd, fd4 = -1, fd6 = -1;
279
280shost.next = NULL;
281shost.address = NULL;
282shost.port = portlo;
283shost.mx = -1;
284
285namelen = Ustrlen(hostname);
286
287/* Anything enclosed in [] must be an IP address. */
288
289if (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
305else 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
310else
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
323for (h = &shost; h != NULL; h = h->next)
324 {
325 fd = (Ustrchr(h->address, ':') != 0)
a6d4c44e
TF
326 ? (fd6 < 0) ? (fd6 = ip_socket(type, af = AF_INET6)) : fd6
327 : (fd4 < 0) ? (fd4 = ip_socket(type, af = AF_INET )) : fd4;
b1f8e4f8
JH
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);
8a512ed5
JH
340 if (connhost)
341 {
b1f8e4f8
JH
342 h->port = port;
343 *connhost = *h;
344 connhost->next = NULL;
345 }
346 return fd;
347 }
348 }
349
a9764ac5
JH
350*errstr = string_sprintf("failed to connect to any address for %s: %s",
351 hostname, strerror(errno));
b1f8e4f8
JH
352
353bad:
354 close(fd4); close(fd6); return -1;
355}
356
059ec3d9 357
3e60dd41
JH
358int
359ip_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
379int
380ip_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
403int
404ip_streamsocket(const uschar * spec, uschar ** errstr, int tmo)
405{
406 return *spec == '/'
407 ? ip_unixsocket(spec, errstr) : ip_tcpsocket(spec, errstr, tmo);
408}
409
059ec3d9
PH
410/*************************************************
411* Set keepalive on a socket *
412*************************************************/
413
414/* Can be called for both incoming and outgoing sockets.
415
416Arguments:
417 sock the socket
418 address the remote host address, for failure logging
419 torf true for outgoing connection, false for incoming
420
421Returns: nothing
422*/
423
424void
55414b25 425ip_keepalive(int sock, const uschar *address, BOOL torf)
059ec3d9
PH
426{
427int fodder = 1;
428if (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
4e71661f 440/*
059ec3d9 441Arguments:
4e71661f
JH
442 fd the file descriptor
443 timeout the timeout, seconds
444Returns: TRUE => ready for i/o
445 FALSE => timed out, or other error
059ec3d9 446*/
4e71661f
JH
447BOOL
448fd_ready(int fd, int timeout)
059ec3d9
PH
449{
450fd_set select_inset;
451struct timeval tv;
19050083 452time_t start_recv = time(NULL);
059ec3d9
PH
453int rc;
454
4e71661f
JH
455if (timeout <= 0)
456 {
457 errno = ETIMEDOUT;
458 return FALSE;
459 }
059ec3d9
PH
460/* Wait until the socket is ready */
461
462for (;;)
463 {
464 FD_ZERO (&select_inset);
4e71661f 465 FD_SET (fd, &select_inset);
059ec3d9
PH
466 tv.tv_sec = timeout;
467 tv.tv_usec = 0;
468
0f0c8159 469 /*DEBUG(D_transport) debug_printf("waiting for data on fd\n");*/
4e71661f 470 rc = select(fd + 1, (SELECT_ARG2_TYPE *)&select_inset, NULL, NULL, &tv);
059ec3d9
PH
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;
4e71661f 494 return FALSE;
059ec3d9
PH
495 }
496
497 /* If the socket is ready, break out of the loop. */
498
4e71661f 499 if (FD_ISSET(fd, &select_inset)) break;
059ec3d9 500 }
4e71661f
JH
501return TRUE;
502}
503
504/* The timeout is implemented using select(), and we loop to cover select()
505getting interrupted, and the possibility of select() returning with a positive
506result but no ready descriptor. Is this in fact possible?
507
508Arguments:
509 sock the socket
510 buffer to read into
511 bufsize the buffer size
512 timeout the timeout
513
514Returns: > 0 => that much data read
515 <= 0 on error or EOF; errno set - zero for EOF
516*/
517
518int
519ip_recv(int sock, uschar *buffer, int buffsize, int timeout)
520{
521int rc;
522
523if (!fd_ready(sock, timeout))
524 return -1;
059ec3d9
PH
525
526/* The socket is ready, read from it (via TLS if it's active). On EOF (i.e.
527close down of the connection), set errno to zero; otherwise leave it alone. */
528
529#ifdef SUPPORT_TLS
817d9f57
JH
530if (tls_out.active == sock)
531 rc = tls_read(FALSE, buffer, buffsize);
532else if (tls_in.active == sock)
533 rc = tls_read(TRUE, buffer, buffsize);
059ec3d9
PH
534else
535#endif
536 rc = recv(sock, buffer, buffsize, 0);
537
538if (rc > 0) return rc;
539if (rc == 0) errno = 0;
540return -1;
541}
542
543
9e4f5962
PP
544
545
546/*************************************************
13363eba
PP
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,
551return the address family; detects IPv4 vs IPv6. If not a socket then
552return -1.
553
554The value 0 is typically AF_UNSPEC, which should not be seen on a connected
555fd. If the return is -1, the errno will be from getsockname(); probably
556ENOTSOCK or ECONNRESET.
557
558Arguments: socket-or-not fd
559Returns: address family or -1
560*/
561
562int
563ip_get_address_family(int fd)
564{
565struct sockaddr_storage ss;
566socklen_t sslen = sizeof(ss);
567
568if (getsockname(fd, (struct sockaddr *) &ss, &sslen) < 0)
569 return -1;
570
571return (int) ss.ss_family;
572}
573
574
575
576
577/*************************************************
9e4f5962
PP
578* Lookup DSCP settings for a socket *
579*************************************************/
580
581struct dscp_name_tableentry {
582 const uschar *name;
583 int value;
584};
585/* Keep both of these tables sorted! */
586static struct dscp_name_tableentry dscp_table[] = {
587#ifdef IPTOS_DSCP_AF11
36a3ae5f
PP
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 },
9e4f5962
PP
601#endif
602#ifdef IPTOS_LOWCOST
36a3ae5f 603 { CUS"lowcost", IPTOS_LOWCOST },
9e4f5962 604#endif
36a3ae5f 605 { CUS"lowdelay", IPTOS_LOWDELAY },
9e4f5962 606#ifdef IPTOS_MINCOST
36a3ae5f 607 { CUS"mincost", IPTOS_MINCOST },
9e4f5962 608#endif
36a3ae5f
PP
609 { CUS"reliability", IPTOS_RELIABILITY },
610 { CUS"throughput", IPTOS_THROUGHPUT }
9e4f5962
PP
611};
612static 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
2a1b36b3
PP
616setsockopt(); this utility does all the lookups. It takes an unexpanded
617option string, expands it, strips off affix whitespace, then checks if it's
618a number. If all of what's left is a number, then that's how the option will
619be parsed and success/failure is a range check. If it's not all a number,
620then it must be a supported keyword.
9e4f5962
PP
621
622Arguments:
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
629Returns: TRUE if okay to setsockopt(), else FALSE
2a1b36b3
PP
630
631*level and *optname may be set even if FALSE is returned
9e4f5962
PP
632*/
633
634BOOL
635dscp_lookup(const uschar *dscp_name, int af,
636 int *level, int *optname, int *dscp_value)
637{
2a1b36b3 638uschar *dscp_lookup, *p;
9e4f5962 639int first, last;
2a1b36b3 640long rawlong;
9e4f5962
PP
641
642if (af == AF_INET)
643 {
644 *level = IPPROTO_IP;
645 *optname = IP_TOS;
646 }
bb7b9411 647#if HAVE_IPV6 && defined(IPV6_TCLASS)
9e4f5962
PP
648else if (af == AF_INET6)
649 {
650 *level = IPPROTO_IPV6;
651 *optname = IPV6_TCLASS;
652 }
b301a50b 653#endif
9e4f5962
PP
654else
655 {
656 DEBUG(D_transport)
657 debug_printf("Unhandled address family %d in dscp_lookup()\n", af);
658 return FALSE;
659 }
660if (!dscp_name)
661 {
662 DEBUG(D_transport)
663 debug_printf("[empty DSCP]\n");
664 return FALSE;
665 }
666dscp_lookup = expand_string(US dscp_name);
667if (dscp_lookup == NULL || *dscp_lookup == '\0')
668 return FALSE;
669
2a1b36b3
PP
670p = dscp_lookup + Ustrlen(dscp_lookup) - 1;
671while (isspace(*p)) *p-- = '\0';
672while (isspace(*dscp_lookup) && dscp_lookup < p) dscp_lookup++;
673if (*dscp_lookup == '\0')
674 return FALSE;
675
676rawlong = Ustrtol(dscp_lookup, &p, 0);
677if (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
9e4f5962
PP
691first = 0;
692last = dscp_table_size;
693while (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 }
711return FALSE;
712}
713
36a3ae5f
PP
714void
715dscp_list_to_stream(FILE *stream)
716{
717int i;
718for (i=0; i < dscp_table_size; ++i)
719 fprintf(stream, "%s\n", dscp_table[i].name);
720}
721
9e4f5962 722
059ec3d9 723/* End of ip.c */
8a512ed5
JH
724/* vi: aw ai sw=2
725*/