Fix non-IPv6 compile on v6-less system
[exim.git] / src / src / ip.c
CommitLineData
059ec3d9
PH
1/*************************************************
2* Exim - an Internet mail transport agent *
3*************************************************/
4
0a49a7a4 5/* Copyright (c) University of Cambridge 1995 - 2009 */
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
67ip_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
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
175 timeout a timeout
176
177Returns: 0 on success; -1 on failure, with errno set
178*/
179
180int
181ip_connect(int sock, int af, uschar *address, int port, int timeout)
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);
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
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
251
252/*************************************************
253* Set keepalive on a socket *
254*************************************************/
255
256/* Can be called for both incoming and outgoing sockets.
257
258Arguments:
259 sock the socket
260 address the remote host address, for failure logging
261 torf true for outgoing connection, false for incoming
262
263Returns: nothing
264*/
265
266void
267ip_keepalive(int sock, uschar *address, BOOL torf)
268{
269int fodder = 1;
270if (setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
271 (uschar *)(&fodder), sizeof(fodder)) != 0)
272 log_write(0, LOG_MAIN, "setsockopt(SO_KEEPALIVE) on connection %s %s "
273 "failed: %s", torf? "to":"from", address, strerror(errno));
274}
275
276
277
278/*************************************************
279* Receive from a socket with timeout *
280*************************************************/
281
282/* The timeout is implemented using select(), and we loop to cover select()
283getting interrupted, and the possibility of select() returning with a positive
284result but no ready descriptor. Is this in fact possible?
285
286Arguments:
287 sock the socket
288 buffer to read into
289 bufsize the buffer size
290 timeout the timeout
291
292Returns: > 0 => that much data read
293 <= 0 on error or EOF; errno set - zero for EOF
294*/
295
296int
297ip_recv(int sock, uschar *buffer, int buffsize, int timeout)
298{
299fd_set select_inset;
300struct timeval tv;
301int start_recv = time(NULL);
302int rc;
303
304/* Wait until the socket is ready */
305
306for (;;)
307 {
308 FD_ZERO (&select_inset);
309 FD_SET (sock, &select_inset);
310 tv.tv_sec = timeout;
311 tv.tv_usec = 0;
312
313 DEBUG(D_transport) debug_printf("waiting for data on socket\n");
314 rc = select(sock + 1, (SELECT_ARG2_TYPE *)&select_inset, NULL, NULL, &tv);
315
316 /* If some interrupt arrived, just retry. We presume this to be rare,
317 but it can happen (e.g. the SIGUSR1 signal sent by exiwhat causes
318 select() to exit).
319
320 Aug 2004: Somebody set up a cron job that ran exiwhat every 2 minutes, making
321 the interrupt not at all rare. Since the timeout is typically more than 2
322 minutes, the effect was to block the timeout completely. To prevent this
323 happening again, we do an explicit time test. */
324
325 if (rc < 0 && errno == EINTR)
326 {
327 DEBUG(D_transport) debug_printf("EINTR while waiting for socket data\n");
328 if (time(NULL) - start_recv < timeout) continue;
329 DEBUG(D_transport) debug_printf("total wait time exceeds timeout\n");
330 }
331
332 /* Handle a timeout, and treat any other select error as a timeout, including
333 an EINTR when we have been in this loop for longer than timeout. */
334
335 if (rc <= 0)
336 {
337 errno = ETIMEDOUT;
338 return -1;
339 }
340
341 /* If the socket is ready, break out of the loop. */
342
343 if (FD_ISSET(sock, &select_inset)) break;
344 }
345
346/* The socket is ready, read from it (via TLS if it's active). On EOF (i.e.
347close down of the connection), set errno to zero; otherwise leave it alone. */
348
349#ifdef SUPPORT_TLS
817d9f57
JH
350if (tls_out.active == sock)
351 rc = tls_read(FALSE, buffer, buffsize);
352else if (tls_in.active == sock)
353 rc = tls_read(TRUE, buffer, buffsize);
059ec3d9
PH
354else
355#endif
356 rc = recv(sock, buffer, buffsize, 0);
357
358if (rc > 0) return rc;
359if (rc == 0) errno = 0;
360return -1;
361}
362
363
9e4f5962
PP
364
365
366/*************************************************
13363eba
PP
367* Lookup address family of potential socket *
368*************************************************/
369
370/* Given a file-descriptor, check to see if it's a socket and, if so,
371return the address family; detects IPv4 vs IPv6. If not a socket then
372return -1.
373
374The value 0 is typically AF_UNSPEC, which should not be seen on a connected
375fd. If the return is -1, the errno will be from getsockname(); probably
376ENOTSOCK or ECONNRESET.
377
378Arguments: socket-or-not fd
379Returns: address family or -1
380*/
381
382int
383ip_get_address_family(int fd)
384{
385struct sockaddr_storage ss;
386socklen_t sslen = sizeof(ss);
387
388if (getsockname(fd, (struct sockaddr *) &ss, &sslen) < 0)
389 return -1;
390
391return (int) ss.ss_family;
392}
393
394
395
396
397/*************************************************
9e4f5962
PP
398* Lookup DSCP settings for a socket *
399*************************************************/
400
401struct dscp_name_tableentry {
402 const uschar *name;
403 int value;
404};
405/* Keep both of these tables sorted! */
406static struct dscp_name_tableentry dscp_table[] = {
407#ifdef IPTOS_DSCP_AF11
36a3ae5f
PP
408 { CUS"af11", IPTOS_DSCP_AF11 },
409 { CUS"af12", IPTOS_DSCP_AF12 },
410 { CUS"af13", IPTOS_DSCP_AF13 },
411 { CUS"af21", IPTOS_DSCP_AF21 },
412 { CUS"af22", IPTOS_DSCP_AF22 },
413 { CUS"af23", IPTOS_DSCP_AF23 },
414 { CUS"af31", IPTOS_DSCP_AF31 },
415 { CUS"af32", IPTOS_DSCP_AF32 },
416 { CUS"af33", IPTOS_DSCP_AF33 },
417 { CUS"af41", IPTOS_DSCP_AF41 },
418 { CUS"af42", IPTOS_DSCP_AF42 },
419 { CUS"af43", IPTOS_DSCP_AF43 },
420 { CUS"ef", IPTOS_DSCP_EF },
9e4f5962
PP
421#endif
422#ifdef IPTOS_LOWCOST
36a3ae5f 423 { CUS"lowcost", IPTOS_LOWCOST },
9e4f5962 424#endif
36a3ae5f 425 { CUS"lowdelay", IPTOS_LOWDELAY },
9e4f5962 426#ifdef IPTOS_MINCOST
36a3ae5f 427 { CUS"mincost", IPTOS_MINCOST },
9e4f5962 428#endif
36a3ae5f
PP
429 { CUS"reliability", IPTOS_RELIABILITY },
430 { CUS"throughput", IPTOS_THROUGHPUT }
9e4f5962
PP
431};
432static int dscp_table_size =
433 sizeof(dscp_table) / sizeof(struct dscp_name_tableentry);
434
435/* DSCP values change by protocol family, and so do the options used for
2a1b36b3
PP
436setsockopt(); this utility does all the lookups. It takes an unexpanded
437option string, expands it, strips off affix whitespace, then checks if it's
438a number. If all of what's left is a number, then that's how the option will
439be parsed and success/failure is a range check. If it's not all a number,
440then it must be a supported keyword.
9e4f5962
PP
441
442Arguments:
443 dscp_name a string, so far unvalidated
444 af address_family in use
445 level setsockopt level to use
446 optname setsockopt name to use
447 dscp_value value for dscp_name
448
449Returns: TRUE if okay to setsockopt(), else FALSE
2a1b36b3
PP
450
451*level and *optname may be set even if FALSE is returned
9e4f5962
PP
452*/
453
454BOOL
455dscp_lookup(const uschar *dscp_name, int af,
456 int *level, int *optname, int *dscp_value)
457{
2a1b36b3 458uschar *dscp_lookup, *p;
9e4f5962 459int first, last;
2a1b36b3 460long rawlong;
9e4f5962
PP
461
462if (af == AF_INET)
463 {
464 *level = IPPROTO_IP;
465 *optname = IP_TOS;
466 }
b301a50b 467#if HAVE_IPV6
9e4f5962
PP
468else if (af == AF_INET6)
469 {
470 *level = IPPROTO_IPV6;
471 *optname = IPV6_TCLASS;
472 }
b301a50b 473#endif
9e4f5962
PP
474else
475 {
476 DEBUG(D_transport)
477 debug_printf("Unhandled address family %d in dscp_lookup()\n", af);
478 return FALSE;
479 }
480if (!dscp_name)
481 {
482 DEBUG(D_transport)
483 debug_printf("[empty DSCP]\n");
484 return FALSE;
485 }
486dscp_lookup = expand_string(US dscp_name);
487if (dscp_lookup == NULL || *dscp_lookup == '\0')
488 return FALSE;
489
2a1b36b3
PP
490p = dscp_lookup + Ustrlen(dscp_lookup) - 1;
491while (isspace(*p)) *p-- = '\0';
492while (isspace(*dscp_lookup) && dscp_lookup < p) dscp_lookup++;
493if (*dscp_lookup == '\0')
494 return FALSE;
495
496rawlong = Ustrtol(dscp_lookup, &p, 0);
497if (p != dscp_lookup && *p == '\0')
498 {
499 /* We have six bits available, which will end up shifted to fit in 0xFC mask.
500 RFC 2597 defines the values unshifted. */
501 if (rawlong < 0 || rawlong > 0x3F)
502 {
503 DEBUG(D_transport)
504 debug_printf("DSCP value %ld out of range, ignored.\n", rawlong);
505 return FALSE;
506 }
507 *dscp_value = rawlong << 2;
508 return TRUE;
509 }
510
9e4f5962
PP
511first = 0;
512last = dscp_table_size;
513while (last > first)
514 {
515 int middle = (first + last)/2;
516 int c = Ustrcmp(dscp_lookup, dscp_table[middle].name);
517 if (c == 0)
518 {
519 *dscp_value = dscp_table[middle].value;
520 return TRUE;
521 }
522 else if (c > 0)
523 {
524 first = middle + 1;
525 }
526 else
527 {
528 last = middle;
529 }
530 }
531return FALSE;
532}
533
36a3ae5f
PP
534void
535dscp_list_to_stream(FILE *stream)
536{
537int i;
538for (i=0; i < dscp_table_size; ++i)
539 fprintf(stream, "%s\n", dscp_table[i].name);
540}
541
9e4f5962 542
059ec3d9 543/* End of ip.c */