Really re-select() when interrupted.
[exim.git] / src / src / ip.c
1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
4
5 /* Copyright (c) University of Cambridge 1995 - 2015 */
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 int
101 ip_addr(void * sin_, int af, const uschar * address, int port)
102 {
103 union sockaddr_46 * sin = sin_;
104 memset(sin, 0, sizeof(*sin));
105
106 /* Setup code when using an IPv6 socket. The wildcard address is ":", to
107 ensure an IPv6 socket is used. */
108
109 #if HAVE_IPV6
110 if (af == AF_INET6)
111 {
112 if (address[0] == ':' && address[1] == 0)
113 {
114 sin->v6.sin6_family = AF_INET6;
115 sin->v6.sin6_addr = in6addr_any;
116 }
117 else
118 ip_addrinfo(address, &sin->v6); /* Panic-dies on error */
119 sin->v6.sin6_port = htons(port);
120 return sizeof(sin->v6);
121 }
122 else
123 #else /* HAVE_IPv6 */
124 af = af; /* Avoid compiler warning */
125 #endif /* HAVE_IPV6 */
126
127 /* Setup code when using IPv4 socket. The wildcard address is "". */
128
129 {
130 sin->v4.sin_family = AF_INET;
131 sin->v4.sin_port = htons(port);
132 sin->v4.sin_addr.s_addr = address[0] == 0
133 ? (S_ADDR_TYPE)INADDR_ANY
134 : (S_ADDR_TYPE)inet_addr(CS address);
135 return sizeof(sin->v4);
136 }
137 }
138
139
140
141 /* This function binds a socket to a local interface address and port. For a
142 wildcard IPv6 bind, the address is ":".
143
144 Arguments:
145 sock the socket
146 af AF_INET or AF_INET6 - the socket type
147 address the IP address, in text form
148 port the IP port (host order)
149
150 Returns: the result of bind()
151 */
152
153 int
154 ip_bind(int sock, int af, uschar *address, int port)
155 {
156 union sockaddr_46 sin;
157 int s_len = ip_addr(&sin, af, address, port);
158 return bind(sock, (struct sockaddr *)&sin, s_len);
159 }
160
161
162
163 /*************************************************
164 * Connect socket to remote host *
165 *************************************************/
166
167 /* This function connects a socket to a remote address and port. The socket may
168 or may not have previously been bound to a local interface. The socket is not
169 closed, even in cases of error. It is expected that the calling function, which
170 created the socket, will be the one that closes it.
171
172 Arguments:
173 sock the socket
174 af AF_INET6 or AF_INET for the socket type
175 address the remote address, in text form
176 port the remote port
177 timeout a timeout (zero for indefinite timeout)
178
179 Returns: 0 on success; -1 on failure, with errno set
180 */
181
182 int
183 ip_connect(int sock, int af, const uschar *address, int port, int timeout)
184 {
185 struct sockaddr_in s_in4;
186 struct sockaddr *s_ptr;
187 int s_len, rc, save_errno;
188
189 /* For an IPv6 address, use an IPv6 sockaddr structure. */
190
191 #if HAVE_IPV6
192 struct sockaddr_in6 s_in6;
193 if (af == AF_INET6)
194 {
195 memset(&s_in6, 0, sizeof(s_in6));
196 ip_addrinfo(address, &s_in6); /* Panic-dies on error */
197 s_in6.sin6_port = htons(port);
198 s_ptr = (struct sockaddr *)&s_in6;
199 s_len = sizeof(s_in6);
200 }
201 else
202 #else /* HAVE_IPV6 */
203 af = af; /* Avoid compiler warning */
204 #endif /* HAVE_IPV6 */
205
206 /* For an IPv4 address, use an IPv4 sockaddr structure, even on a system with
207 IPv6 support. */
208
209 {
210 memset(&s_in4, 0, sizeof(s_in4));
211 s_in4.sin_family = AF_INET;
212 s_in4.sin_port = htons(port);
213 s_in4.sin_addr.s_addr = (S_ADDR_TYPE)inet_addr(CCS address);
214 s_ptr = (struct sockaddr *)&s_in4;
215 s_len = sizeof(s_in4);
216 }
217
218 /* If no connection timeout is set, just call connect() without setting a
219 timer, thereby allowing the inbuilt OS timeout to operate. */
220
221 sigalrm_seen = FALSE;
222 if (timeout > 0) alarm(timeout);
223 rc = connect(sock, s_ptr, s_len);
224 save_errno = errno;
225 alarm(0);
226
227 /* There is a testing facility for simulating a connection timeout, as I
228 can't think of any other way of doing this. It converts a connection refused
229 into a timeout if the timeout is set to 999999. */
230
231 if (running_in_test_harness && save_errno == ECONNREFUSED && timeout == 999999)
232 {
233 rc = -1;
234 save_errno = EINTR;
235 sigalrm_seen = TRUE;
236 }
237
238 /* Success */
239
240 if (rc >= 0) return 0;
241
242 /* A failure whose error code is "Interrupted system call" is in fact
243 an externally applied timeout if the signal handler has been run. */
244
245 errno = save_errno == EINTR && sigalrm_seen ? ETIMEDOUT : save_errno;
246 return -1;
247 }
248
249
250
251 /*************************************************
252 * Create connected socket to remote host *
253 *************************************************/
254
255 /* Create a socket and connect to host (name or number, ipv6 ok)
256 at one of port-range.
257
258 Arguments:
259 type SOCK_DGRAM or SOCK_STREAM
260 af AF_INET6 or AF_INET for the socket type
261 address the remote address, in text form
262 portlo,porthi the remote port range
263 timeout a timeout
264 connhost if not NULL, host_item filled in with connection details
265 errstr pointer for allocated string on error
266
267 Return:
268 socket fd, or -1 on failure (having allocated an error string)
269 */
270 int
271 ip_connectedsocket(int type, const uschar * hostname, int portlo, int porthi,
272 int timeout, host_item * connhost, uschar ** errstr)
273 {
274 int namelen, port;
275 host_item shost;
276 host_item *h;
277 int af = 0, fd, fd4 = -1, fd6 = -1;
278
279 shost.next = NULL;
280 shost.address = NULL;
281 shost.port = portlo;
282 shost.mx = -1;
283
284 namelen = Ustrlen(hostname);
285
286 /* Anything enclosed in [] must be an IP address. */
287
288 if (hostname[0] == '[' &&
289 hostname[namelen - 1] == ']')
290 {
291 uschar * host = string_copy(hostname);
292 host[namelen - 1] = 0;
293 host++;
294 if (string_is_ip_address(host, NULL) == 0)
295 {
296 *errstr = string_sprintf("malformed IP address \"%s\"", hostname);
297 return -1;
298 }
299 shost.name = shost.address = host;
300 }
301
302 /* Otherwise check for an unadorned IP address */
303
304 else if (string_is_ip_address(hostname, NULL) != 0)
305 shost.name = shost.address = string_copy(hostname);
306
307 /* Otherwise lookup IP address(es) from the name */
308
309 else
310 {
311 shost.name = string_copy(hostname);
312 if (host_find_byname(&shost, NULL, HOST_FIND_QUALIFY_SINGLE,
313 NULL, FALSE) != HOST_FOUND)
314 {
315 *errstr = string_sprintf("no IP address found for host %s", shost.name);
316 return -1;
317 }
318 }
319
320 /* Try to connect to the server - test each IP till one works */
321
322 for (h = &shost; h != NULL; h = h->next)
323 {
324 fd = (Ustrchr(h->address, ':') != 0)
325 ? (fd6 < 0) ? (fd6 = ip_socket(type, af = AF_INET6)) : fd6
326 : (fd4 < 0) ? (fd4 = ip_socket(type, af = AF_INET )) : fd4;
327
328 if (fd < 0)
329 {
330 *errstr = string_sprintf("failed to create socket: %s", strerror(errno));
331 goto bad;
332 }
333
334 for(port = portlo; port <= porthi; port++)
335 if (ip_connect(fd, af, h->address, port, timeout) == 0)
336 {
337 if (fd != fd6) close(fd6);
338 if (fd != fd4) close(fd4);
339 if (connhost)
340 {
341 h->port = port;
342 *connhost = *h;
343 connhost->next = NULL;
344 }
345 return fd;
346 }
347 }
348
349 *errstr = string_sprintf("failed to connect to any address for %s: %s",
350 hostname, strerror(errno));
351
352 bad:
353 close(fd4); close(fd6); return -1;
354 }
355
356
357 int
358 ip_tcpsocket(const uschar * hostport, uschar ** errstr, int tmo)
359 {
360 int scan;
361 uschar hostname[256];
362 unsigned int portlow, porthigh;
363
364 /* extract host and port part */
365 scan = sscanf(CS hostport, "%255s %u-%u", hostname, &portlow, &porthigh);
366 if (scan != 3)
367 {
368 if (scan != 2)
369 {
370 *errstr = string_sprintf("invalid socket '%s'", hostport);
371 return -1;
372 }
373 porthigh = portlow;
374 }
375
376 return ip_connectedsocket(SOCK_STREAM, hostname, portlow, porthigh,
377 tmo, NULL, errstr);
378 }
379
380 int
381 ip_unixsocket(const uschar * path, uschar ** errstr)
382 {
383 int sock;
384 struct sockaddr_un server;
385
386 if ((sock = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
387 {
388 *errstr = US"can't open UNIX socket.";
389 return -1;
390 }
391
392 server.sun_family = AF_UNIX;
393 Ustrncpy(server.sun_path, path, sizeof(server.sun_path)-1);
394 server.sun_path[sizeof(server.sun_path)-1] = '\0';
395 if (connect(sock, (struct sockaddr *) &server, sizeof(server)) < 0)
396 {
397 int err = errno;
398 (void)close(sock);
399 *errstr = string_sprintf("unable to connect to UNIX socket (%s): %s",
400 path, strerror(err));
401 return -1;
402 }
403 return sock;
404 }
405
406 int
407 ip_streamsocket(const uschar * spec, uschar ** errstr, int tmo)
408 {
409 return *spec == '/'
410 ? ip_unixsocket(spec, errstr) : ip_tcpsocket(spec, errstr, tmo);
411 }
412
413 /*************************************************
414 * Set keepalive on a socket *
415 *************************************************/
416
417 /* Can be called for both incoming and outgoing sockets.
418
419 Arguments:
420 sock the socket
421 address the remote host address, for failure logging
422 torf true for outgoing connection, false for incoming
423
424 Returns: nothing
425 */
426
427 void
428 ip_keepalive(int sock, const uschar *address, BOOL torf)
429 {
430 int fodder = 1;
431 if (setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
432 (uschar *)(&fodder), sizeof(fodder)) != 0)
433 log_write(0, LOG_MAIN, "setsockopt(SO_KEEPALIVE) on connection %s %s "
434 "failed: %s", torf? "to":"from", address, strerror(errno));
435 }
436
437
438
439 /*************************************************
440 * Receive from a socket with timeout *
441 *************************************************/
442
443 /*
444 Arguments:
445 fd the file descriptor
446 timeout the timeout, seconds
447 Returns: TRUE => ready for i/o
448 FALSE => timed out, or other error
449 */
450 BOOL
451 fd_ready(int fd, int timeout)
452 {
453 fd_set select_inset;
454 struct timeval tv;
455 time_t start_recv = time(NULL);
456 int rc;
457
458 if (timeout <= 0)
459 {
460 errno = ETIMEDOUT;
461 return FALSE;
462 }
463 /* Wait until the socket is ready */
464
465 do
466 {
467 FD_ZERO (&select_inset);
468 FD_SET (fd, &select_inset);
469 tv.tv_sec = timeout;
470 tv.tv_usec = 0;
471
472 /*DEBUG(D_transport) debug_printf("waiting for data on fd\n");*/
473 rc = select(fd + 1, (SELECT_ARG2_TYPE *)&select_inset, NULL, NULL, &tv);
474
475 /* If some interrupt arrived, just retry. We presume this to be rare,
476 but it can happen (e.g. the SIGUSR1 signal sent by exiwhat causes
477 select() to exit).
478
479 Aug 2004: Somebody set up a cron job that ran exiwhat every 2 minutes, making
480 the interrupt not at all rare. Since the timeout is typically more than 2
481 minutes, the effect was to block the timeout completely. To prevent this
482 happening again, we do an explicit time test. */
483
484 if (rc < 0 && errno == EINTR)
485 {
486 DEBUG(D_transport) debug_printf("EINTR while waiting for socket data\n");
487 if (time(NULL) - start_recv < timeout) continue;
488 DEBUG(D_transport) debug_printf("total wait time exceeds timeout\n");
489 }
490
491 /* Handle a timeout, and treat any other select error as a timeout, including
492 an EINTR when we have been in this loop for longer than timeout. */
493
494 if (rc <= 0)
495 {
496 errno = ETIMEDOUT;
497 return FALSE;
498 }
499
500 /* If the socket is ready, break out of the loop. */
501 }
502 while (rc < 0 || !FD_ISSET(fd, &select_inset));
503 return TRUE;
504 }
505
506 /* The timeout is implemented using select(), and we loop to cover select()
507 getting interrupted, and the possibility of select() returning with a positive
508 result but no ready descriptor. Is this in fact possible?
509
510 Arguments:
511 sock the socket
512 buffer to read into
513 bufsize the buffer size
514 timeout the timeout
515
516 Returns: > 0 => that much data read
517 <= 0 on error or EOF; errno set - zero for EOF
518 */
519
520 int
521 ip_recv(int sock, uschar *buffer, int buffsize, int timeout)
522 {
523 int rc;
524
525 if (!fd_ready(sock, timeout))
526 return -1;
527
528 /* The socket is ready, read from it (via TLS if it's active). On EOF (i.e.
529 close down of the connection), set errno to zero; otherwise leave it alone. */
530
531 #ifdef SUPPORT_TLS
532 if (tls_out.active == sock)
533 rc = tls_read(FALSE, buffer, buffsize);
534 else if (tls_in.active == sock)
535 rc = tls_read(TRUE, buffer, buffsize);
536 else
537 #endif
538 rc = recv(sock, buffer, buffsize, 0);
539
540 if (rc > 0) return rc;
541 if (rc == 0) errno = 0;
542 return -1;
543 }
544
545
546
547
548 /*************************************************
549 * Lookup address family of potential socket *
550 *************************************************/
551
552 /* Given a file-descriptor, check to see if it's a socket and, if so,
553 return the address family; detects IPv4 vs IPv6. If not a socket then
554 return -1.
555
556 The value 0 is typically AF_UNSPEC, which should not be seen on a connected
557 fd. If the return is -1, the errno will be from getsockname(); probably
558 ENOTSOCK or ECONNRESET.
559
560 Arguments: socket-or-not fd
561 Returns: address family or -1
562 */
563
564 int
565 ip_get_address_family(int fd)
566 {
567 struct sockaddr_storage ss;
568 socklen_t sslen = sizeof(ss);
569
570 if (getsockname(fd, (struct sockaddr *) &ss, &sslen) < 0)
571 return -1;
572
573 return (int) ss.ss_family;
574 }
575
576
577
578
579 /*************************************************
580 * Lookup DSCP settings for a socket *
581 *************************************************/
582
583 struct dscp_name_tableentry {
584 const uschar *name;
585 int value;
586 };
587 /* Keep both of these tables sorted! */
588 static struct dscp_name_tableentry dscp_table[] = {
589 #ifdef IPTOS_DSCP_AF11
590 { CUS"af11", IPTOS_DSCP_AF11 },
591 { CUS"af12", IPTOS_DSCP_AF12 },
592 { CUS"af13", IPTOS_DSCP_AF13 },
593 { CUS"af21", IPTOS_DSCP_AF21 },
594 { CUS"af22", IPTOS_DSCP_AF22 },
595 { CUS"af23", IPTOS_DSCP_AF23 },
596 { CUS"af31", IPTOS_DSCP_AF31 },
597 { CUS"af32", IPTOS_DSCP_AF32 },
598 { CUS"af33", IPTOS_DSCP_AF33 },
599 { CUS"af41", IPTOS_DSCP_AF41 },
600 { CUS"af42", IPTOS_DSCP_AF42 },
601 { CUS"af43", IPTOS_DSCP_AF43 },
602 { CUS"ef", IPTOS_DSCP_EF },
603 #endif
604 #ifdef IPTOS_LOWCOST
605 { CUS"lowcost", IPTOS_LOWCOST },
606 #endif
607 { CUS"lowdelay", IPTOS_LOWDELAY },
608 #ifdef IPTOS_MINCOST
609 { CUS"mincost", IPTOS_MINCOST },
610 #endif
611 { CUS"reliability", IPTOS_RELIABILITY },
612 { CUS"throughput", IPTOS_THROUGHPUT }
613 };
614 static int dscp_table_size =
615 sizeof(dscp_table) / sizeof(struct dscp_name_tableentry);
616
617 /* DSCP values change by protocol family, and so do the options used for
618 setsockopt(); this utility does all the lookups. It takes an unexpanded
619 option string, expands it, strips off affix whitespace, then checks if it's
620 a number. If all of what's left is a number, then that's how the option will
621 be parsed and success/failure is a range check. If it's not all a number,
622 then it must be a supported keyword.
623
624 Arguments:
625 dscp_name a string, so far unvalidated
626 af address_family in use
627 level setsockopt level to use
628 optname setsockopt name to use
629 dscp_value value for dscp_name
630
631 Returns: TRUE if okay to setsockopt(), else FALSE
632
633 *level and *optname may be set even if FALSE is returned
634 */
635
636 BOOL
637 dscp_lookup(const uschar *dscp_name, int af,
638 int *level, int *optname, int *dscp_value)
639 {
640 uschar *dscp_lookup, *p;
641 int first, last;
642 long rawlong;
643
644 if (af == AF_INET)
645 {
646 *level = IPPROTO_IP;
647 *optname = IP_TOS;
648 }
649 #if HAVE_IPV6 && defined(IPV6_TCLASS)
650 else if (af == AF_INET6)
651 {
652 *level = IPPROTO_IPV6;
653 *optname = IPV6_TCLASS;
654 }
655 #endif
656 else
657 {
658 DEBUG(D_transport)
659 debug_printf("Unhandled address family %d in dscp_lookup()\n", af);
660 return FALSE;
661 }
662 if (!dscp_name)
663 {
664 DEBUG(D_transport)
665 debug_printf("[empty DSCP]\n");
666 return FALSE;
667 }
668 dscp_lookup = expand_string(US dscp_name);
669 if (dscp_lookup == NULL || *dscp_lookup == '\0')
670 return FALSE;
671
672 p = dscp_lookup + Ustrlen(dscp_lookup) - 1;
673 while (isspace(*p)) *p-- = '\0';
674 while (isspace(*dscp_lookup) && dscp_lookup < p) dscp_lookup++;
675 if (*dscp_lookup == '\0')
676 return FALSE;
677
678 rawlong = Ustrtol(dscp_lookup, &p, 0);
679 if (p != dscp_lookup && *p == '\0')
680 {
681 /* We have six bits available, which will end up shifted to fit in 0xFC mask.
682 RFC 2597 defines the values unshifted. */
683 if (rawlong < 0 || rawlong > 0x3F)
684 {
685 DEBUG(D_transport)
686 debug_printf("DSCP value %ld out of range, ignored.\n", rawlong);
687 return FALSE;
688 }
689 *dscp_value = rawlong << 2;
690 return TRUE;
691 }
692
693 first = 0;
694 last = dscp_table_size;
695 while (last > first)
696 {
697 int middle = (first + last)/2;
698 int c = Ustrcmp(dscp_lookup, dscp_table[middle].name);
699 if (c == 0)
700 {
701 *dscp_value = dscp_table[middle].value;
702 return TRUE;
703 }
704 else if (c > 0)
705 first = middle + 1;
706 else
707 last = middle;
708 }
709 return FALSE;
710 }
711
712 void
713 dscp_list_to_stream(FILE *stream)
714 {
715 int i;
716 for (i=0; i < dscp_table_size; ++i)
717 fprintf(stream, "%s\n", dscp_table[i].name);
718 }
719
720
721 /* End of ip.c */
722 /* vi: aw ai sw=2
723 */