b0c98878b6736b9f3365129d50a11ee10a22af4a
[exim.git] / src / src / ip.c
1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
4
5 /* Copyright (c) University of Cambridge 1995 - 2009 */
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
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
252 /*************************************************
253 * Set keepalive on a socket *
254 *************************************************/
255
256 /* Can be called for both incoming and outgoing sockets.
257
258 Arguments:
259 sock the socket
260 address the remote host address, for failure logging
261 torf true for outgoing connection, false for incoming
262
263 Returns: nothing
264 */
265
266 void
267 ip_keepalive(int sock, uschar *address, BOOL torf)
268 {
269 int fodder = 1;
270 if (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()
283 getting interrupted, and the possibility of select() returning with a positive
284 result but no ready descriptor. Is this in fact possible?
285
286 Arguments:
287 sock the socket
288 buffer to read into
289 bufsize the buffer size
290 timeout the timeout
291
292 Returns: > 0 => that much data read
293 <= 0 on error or EOF; errno set - zero for EOF
294 */
295
296 int
297 ip_recv(int sock, uschar *buffer, int buffsize, int timeout)
298 {
299 fd_set select_inset;
300 struct timeval tv;
301 int start_recv = time(NULL);
302 int rc;
303
304 /* Wait until the socket is ready */
305
306 for (;;)
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.
347 close down of the connection), set errno to zero; otherwise leave it alone. */
348
349 #ifdef SUPPORT_TLS
350 if (tls_out.active == sock)
351 rc = tls_read(FALSE, buffer, buffsize);
352 else if (tls_in.active == sock)
353 rc = tls_read(TRUE, buffer, buffsize);
354 else
355 #endif
356 rc = recv(sock, buffer, buffsize, 0);
357
358 if (rc > 0) return rc;
359 if (rc == 0) errno = 0;
360 return -1;
361 }
362
363
364
365
366 /*************************************************
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,
371 return the address family; detects IPv4 vs IPv6. If not a socket then
372 return -1.
373
374 The value 0 is typically AF_UNSPEC, which should not be seen on a connected
375 fd. If the return is -1, the errno will be from getsockname(); probably
376 ENOTSOCK or ECONNRESET.
377
378 Arguments: socket-or-not fd
379 Returns: address family or -1
380 */
381
382 int
383 ip_get_address_family(int fd)
384 {
385 struct sockaddr_storage ss;
386 socklen_t sslen = sizeof(ss);
387
388 if (getsockname(fd, (struct sockaddr *) &ss, &sslen) < 0)
389 return -1;
390
391 return (int) ss.ss_family;
392 }
393
394
395
396
397 /*************************************************
398 * Lookup DSCP settings for a socket *
399 *************************************************/
400
401 struct dscp_name_tableentry {
402 const uschar *name;
403 int value;
404 };
405 /* Keep both of these tables sorted! */
406 static struct dscp_name_tableentry dscp_table[] = {
407 #ifdef IPTOS_DSCP_AF11
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 },
421 #endif
422 #ifdef IPTOS_LOWCOST
423 { CUS"lowcost", IPTOS_LOWCOST },
424 #endif
425 { CUS"lowdelay", IPTOS_LOWDELAY },
426 #ifdef IPTOS_MINCOST
427 { CUS"mincost", IPTOS_MINCOST },
428 #endif
429 { CUS"reliability", IPTOS_RELIABILITY },
430 { CUS"throughput", IPTOS_THROUGHPUT }
431 };
432 static 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
436 setsockopt(); this utility does all the lookups. It takes an unexpanded
437 option string, expands it, strips off affix whitespace, then checks if it's
438 a number. If all of what's left is a number, then that's how the option will
439 be parsed and success/failure is a range check. If it's not all a number,
440 then it must be a supported keyword.
441
442 Arguments:
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
449 Returns: TRUE if okay to setsockopt(), else FALSE
450
451 *level and *optname may be set even if FALSE is returned
452 */
453
454 BOOL
455 dscp_lookup(const uschar *dscp_name, int af,
456 int *level, int *optname, int *dscp_value)
457 {
458 uschar *dscp_lookup, *p;
459 int first, last;
460 long rawlong;
461
462 if (af == AF_INET)
463 {
464 *level = IPPROTO_IP;
465 *optname = IP_TOS;
466 }
467 else if (af == AF_INET6)
468 {
469 *level = IPPROTO_IPV6;
470 *optname = IPV6_TCLASS;
471 }
472 else
473 {
474 DEBUG(D_transport)
475 debug_printf("Unhandled address family %d in dscp_lookup()\n", af);
476 return FALSE;
477 }
478 if (!dscp_name)
479 {
480 DEBUG(D_transport)
481 debug_printf("[empty DSCP]\n");
482 return FALSE;
483 }
484 dscp_lookup = expand_string(US dscp_name);
485 if (dscp_lookup == NULL || *dscp_lookup == '\0')
486 return FALSE;
487
488 p = dscp_lookup + Ustrlen(dscp_lookup) - 1;
489 while (isspace(*p)) *p-- = '\0';
490 while (isspace(*dscp_lookup) && dscp_lookup < p) dscp_lookup++;
491 if (*dscp_lookup == '\0')
492 return FALSE;
493
494 rawlong = Ustrtol(dscp_lookup, &p, 0);
495 if (p != dscp_lookup && *p == '\0')
496 {
497 /* We have six bits available, which will end up shifted to fit in 0xFC mask.
498 RFC 2597 defines the values unshifted. */
499 if (rawlong < 0 || rawlong > 0x3F)
500 {
501 DEBUG(D_transport)
502 debug_printf("DSCP value %ld out of range, ignored.\n", rawlong);
503 return FALSE;
504 }
505 *dscp_value = rawlong << 2;
506 return TRUE;
507 }
508
509 first = 0;
510 last = dscp_table_size;
511 while (last > first)
512 {
513 int middle = (first + last)/2;
514 int c = Ustrcmp(dscp_lookup, dscp_table[middle].name);
515 if (c == 0)
516 {
517 *dscp_value = dscp_table[middle].value;
518 return TRUE;
519 }
520 else if (c > 0)
521 {
522 first = middle + 1;
523 }
524 else
525 {
526 last = middle;
527 }
528 }
529 return FALSE;
530 }
531
532 void
533 dscp_list_to_stream(FILE *stream)
534 {
535 int i;
536 for (i=0; i < dscp_table_size; ++i)
537 fprintf(stream, "%s\n", dscp_table[i].name);
538 }
539
540
541 /* End of ip.c */