Commit | Line | Data |
---|---|---|
420a0d19 CE |
1 | /************************************************* |
2 | * Exim - an Internet mail transport agent * | |
3 | *************************************************/ | |
4 | ||
2ea97746 | 5 | /* Copyright (c) University of Cambridge 1995 - 2018 */ |
420a0d19 CE |
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 | |
2ea97746 | 10 | that other parts of Exim can call, especially as there are now several |
420a0d19 CE |
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 | |
2ea97746 | 67 | ip_addrinfo(const uschar *address, struct sockaddr_in6 *saddr) |
420a0d19 CE |
68 | { |
69 | #ifdef IPV6_USE_INET_PTON | |
70 | ||
2ea97746 | 71 | if (inet_pton(AF_INET6, CCS address, &saddr->sin6_addr) != 1) |
420a0d19 CE |
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; | |
2ea97746 | 84 | if ((rc = getaddrinfo(CCS address, NULL, &hints, &res)) != 0 || res == NULL) |
420a0d19 CE |
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 | ||
420a0d19 | 100 | int |
2ea97746 | 101 | ip_addr(void * sin_, int af, const uschar * address, int port) |
420a0d19 | 102 | { |
2ea97746 CE |
103 | union sockaddr_46 * sin = sin_; |
104 | memset(sin, 0, sizeof(*sin)); | |
420a0d19 CE |
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 | { | |
2ea97746 CE |
114 | sin->v6.sin6_family = AF_INET6; |
115 | sin->v6.sin6_addr = in6addr_any; | |
420a0d19 CE |
116 | } |
117 | else | |
2ea97746 CE |
118 | ip_addrinfo(address, &sin->v6); /* Panic-dies on error */ |
119 | sin->v6.sin6_port = htons(port); | |
120 | return sizeof(sin->v6); | |
420a0d19 CE |
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 | { | |
2ea97746 CE |
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); | |
420a0d19 | 136 | } |
2ea97746 | 137 | } |
420a0d19 | 138 | |
420a0d19 | 139 | |
2ea97746 CE |
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); | |
420a0d19 CE |
158 | return bind(sock, (struct sockaddr *)&sin, s_len); |
159 | } | |
160 | ||
161 | ||
162 | ||
2ea97746 CE |
163 | /************************************************* |
164 | *************************************************/ | |
165 | ||
166 | #ifdef EXIM_TFO_PROBE | |
167 | void | |
168 | tfo_probe(void) | |
169 | { | |
170 | # ifdef TCP_FASTOPEN | |
171 | int sock, backlog = 5; | |
172 | ||
173 | if ( (sock = socket(SOCK_STREAM, AF_INET, 0)) < 0 | |
174 | && setsockopt(sock, IPPROTO_TCP, TCP_FASTOPEN, &backlog, sizeof(backlog)) | |
175 | ) | |
176 | f.tcp_fastopen_ok = TRUE; | |
177 | close(sock); | |
178 | # endif | |
179 | } | |
180 | #endif | |
181 | ||
182 | ||
420a0d19 CE |
183 | /************************************************* |
184 | * Connect socket to remote host * | |
185 | *************************************************/ | |
186 | ||
187 | /* This function connects a socket to a remote address and port. The socket may | |
188 | or may not have previously been bound to a local interface. The socket is not | |
189 | closed, even in cases of error. It is expected that the calling function, which | |
190 | created the socket, will be the one that closes it. | |
191 | ||
192 | Arguments: | |
193 | sock the socket | |
194 | af AF_INET6 or AF_INET for the socket type | |
195 | address the remote address, in text form | |
196 | port the remote port | |
197 | timeout a timeout (zero for indefinite timeout) | |
2ea97746 CE |
198 | fastopen_blob non-null iff TCP_FASTOPEN can be used; may indicate early-data to |
199 | be sent in SYN segment. Any such data must be idempotent. | |
420a0d19 CE |
200 | |
201 | Returns: 0 on success; -1 on failure, with errno set | |
202 | */ | |
203 | ||
204 | int | |
2ea97746 CE |
205 | ip_connect(int sock, int af, const uschar *address, int port, int timeout, |
206 | const blob * fastopen_blob) | |
420a0d19 CE |
207 | { |
208 | struct sockaddr_in s_in4; | |
209 | struct sockaddr *s_ptr; | |
210 | int s_len, rc, save_errno; | |
211 | ||
212 | /* For an IPv6 address, use an IPv6 sockaddr structure. */ | |
213 | ||
214 | #if HAVE_IPV6 | |
215 | struct sockaddr_in6 s_in6; | |
216 | if (af == AF_INET6) | |
217 | { | |
218 | memset(&s_in6, 0, sizeof(s_in6)); | |
219 | ip_addrinfo(address, &s_in6); /* Panic-dies on error */ | |
220 | s_in6.sin6_port = htons(port); | |
221 | s_ptr = (struct sockaddr *)&s_in6; | |
222 | s_len = sizeof(s_in6); | |
223 | } | |
224 | else | |
225 | #else /* HAVE_IPV6 */ | |
226 | af = af; /* Avoid compiler warning */ | |
227 | #endif /* HAVE_IPV6 */ | |
228 | ||
229 | /* For an IPv4 address, use an IPv4 sockaddr structure, even on a system with | |
230 | IPv6 support. */ | |
231 | ||
232 | { | |
233 | memset(&s_in4, 0, sizeof(s_in4)); | |
234 | s_in4.sin_family = AF_INET; | |
235 | s_in4.sin_port = htons(port); | |
2ea97746 | 236 | s_in4.sin_addr.s_addr = (S_ADDR_TYPE)inet_addr(CCS address); |
420a0d19 CE |
237 | s_ptr = (struct sockaddr *)&s_in4; |
238 | s_len = sizeof(s_in4); | |
239 | } | |
240 | ||
241 | /* If no connection timeout is set, just call connect() without setting a | |
242 | timer, thereby allowing the inbuilt OS timeout to operate. */ | |
243 | ||
2ea97746 | 244 | callout_address = string_sprintf("[%s]:%d", address, port); |
420a0d19 | 245 | sigalrm_seen = FALSE; |
2ea97746 CE |
246 | if (timeout > 0) ALARM(timeout); |
247 | ||
248 | #ifdef TCP_FASTOPEN | |
249 | /* TCP Fast Open, if the system has a cookie from a previous call to | |
250 | this peer, can send data in the SYN packet. The peer can send data | |
251 | before it gets our ACK of its SYN,ACK - the latter is useful for | |
252 | the SMTP banner. Other (than SMTP) cases of TCP connections can | |
253 | possibly use the data-on-syn, so support that too. */ | |
254 | ||
255 | if (fastopen_blob && f.tcp_fastopen_ok) | |
256 | { | |
257 | # ifdef MSG_FASTOPEN | |
258 | /* This is a Linux implementation. It might be useable on FreeBSD; I have | |
259 | not checked. */ | |
260 | ||
261 | if ((rc = sendto(sock, fastopen_blob->data, fastopen_blob->len, | |
262 | MSG_FASTOPEN | MSG_DONTWAIT, s_ptr, s_len)) >= 0) | |
263 | /* seen for with-data, experimental TFO option, with-cookie case */ | |
264 | /* seen for with-data, proper TFO opt, with-cookie case */ | |
265 | { | |
266 | DEBUG(D_transport|D_v) | |
267 | debug_printf("TFO mode connection attempt to %s, %lu data\n", | |
268 | address, (unsigned long)fastopen_blob->len); | |
269 | /*XXX also seen on successful TFO, sigh */ | |
270 | tcp_out_fastopen = fastopen_blob->len > 0 ? TFO_ATTEMPTED_DATA : TFO_ATTEMPTED_NODATA; | |
271 | } | |
272 | else if (errno == EINPROGRESS) /* expected if we had no cookie for peer */ | |
273 | /* seen for no-data, proper TFO option, both cookie-request and with-cookie cases */ | |
274 | /* apparently no visibility of the diffference at this point */ | |
275 | /* seen for with-data, proper TFO opt, cookie-req */ | |
276 | /* with netwk delay, post-conn tcp_info sees unacked 1 for R, 2 for C; code in smtp_out.c */ | |
277 | /* ? older Experimental TFO option behaviour ? */ | |
278 | { /* queue unsent data */ | |
279 | DEBUG(D_transport|D_v) debug_printf("TFO mode sendto, %s data: EINPROGRESS\n", | |
280 | fastopen_blob->len > 0 ? "with" : "no"); | |
281 | if (!fastopen_blob->data) | |
282 | { | |
283 | tcp_out_fastopen = TFO_ATTEMPTED_NODATA; /* we tried; unknown if useful yet */ | |
284 | rc = 0; | |
285 | } | |
286 | else | |
287 | rc = send(sock, fastopen_blob->data, fastopen_blob->len, 0); | |
288 | } | |
289 | else if(errno == EOPNOTSUPP) | |
290 | { | |
291 | DEBUG(D_transport) | |
292 | debug_printf("Tried TCP Fast Open but apparently not enabled by sysctl\n"); | |
293 | goto legacy_connect; | |
294 | } | |
295 | # endif | |
296 | # ifdef EXIM_TFO_CONNECTX | |
297 | /* MacOS */ | |
298 | sa_endpoints_t ends = { | |
299 | .sae_srcif = 0, .sae_srcaddr = NULL, .sae_srcaddrlen = 0, | |
300 | .sae_dstaddr = s_ptr, .sae_dstaddrlen = s_len }; | |
301 | struct iovec iov = { | |
302 | .iov_base = fastopen_blob->data, .iov_len = fastopen_blob->len }; | |
303 | size_t len; | |
304 | ||
305 | if ((rc = connectx(sock, &ends, SAE_ASSOCID_ANY, | |
306 | CONNECT_DATA_IDEMPOTENT, &iov, 1, &len, NULL)) == 0) | |
307 | { | |
308 | DEBUG(D_transport|D_v) | |
309 | debug_printf("TFO mode connection attempt to %s, %lu data\n", | |
310 | address, (unsigned long)fastopen_blob->len); | |
311 | tcp_out_fastopen = fastopen_blob->len > 0 ? TFO_ATTEMPTED_DATA : TFO_ATTEMPTED_NODATA; | |
312 | ||
313 | if (len != fastopen_blob->len) | |
314 | DEBUG(D_transport|D_v) | |
315 | debug_printf(" only queued %lu data!\n", (unsigned long)len); | |
316 | } | |
317 | else if (errno == EINPROGRESS) | |
318 | { | |
319 | DEBUG(D_transport|D_v) debug_printf("TFO mode sendto, %s data: EINPROGRESS\n", | |
320 | fastopen_blob->len > 0 ? "with" : "no"); | |
321 | if (!fastopen_blob->data) | |
322 | { | |
323 | tcp_out_fastopen = TFO_ATTEMPTED_NODATA; /* we tried; unknown if useful yet */ | |
324 | rc = 0; | |
325 | } | |
326 | else /* assume that no data was queued; block in send */ | |
327 | rc = send(sock, fastopen_blob->data, fastopen_blob->len, 0); | |
328 | } | |
329 | # endif | |
330 | } | |
331 | else | |
332 | #endif /*TCP_FASTOPEN*/ | |
333 | { | |
334 | legacy_connect: | |
335 | DEBUG(D_transport|D_v) if (fastopen_blob) | |
336 | debug_printf("non-TFO mode connection attempt to %s, %lu data\n", | |
337 | address, (unsigned long)fastopen_blob->len); | |
338 | if ((rc = connect(sock, s_ptr, s_len)) >= 0) | |
339 | if ( fastopen_blob && fastopen_blob->data && fastopen_blob->len | |
340 | && send(sock, fastopen_blob->data, fastopen_blob->len, 0) < 0) | |
341 | rc = -1; | |
342 | } | |
343 | ||
420a0d19 | 344 | save_errno = errno; |
2ea97746 | 345 | ALARM_CLR(0); |
420a0d19 CE |
346 | |
347 | /* There is a testing facility for simulating a connection timeout, as I | |
348 | can't think of any other way of doing this. It converts a connection refused | |
349 | into a timeout if the timeout is set to 999999. */ | |
350 | ||
2ea97746 | 351 | if (f.running_in_test_harness && save_errno == ECONNREFUSED && timeout == 999999) |
420a0d19 | 352 | { |
2ea97746 CE |
353 | rc = -1; |
354 | save_errno = EINTR; | |
355 | sigalrm_seen = TRUE; | |
420a0d19 CE |
356 | } |
357 | ||
358 | /* Success */ | |
359 | ||
2ea97746 CE |
360 | if (rc >= 0) |
361 | return 0; | |
420a0d19 CE |
362 | |
363 | /* A failure whose error code is "Interrupted system call" is in fact | |
364 | an externally applied timeout if the signal handler has been run. */ | |
365 | ||
2ea97746 | 366 | errno = save_errno == EINTR && sigalrm_seen ? ETIMEDOUT : save_errno; |
420a0d19 CE |
367 | return -1; |
368 | } | |
369 | ||
370 | ||
371 | ||
372 | /************************************************* | |
373 | * Create connected socket to remote host * | |
374 | *************************************************/ | |
375 | ||
376 | /* Create a socket and connect to host (name or number, ipv6 ok) | |
377 | at one of port-range. | |
378 | ||
379 | Arguments: | |
380 | type SOCK_DGRAM or SOCK_STREAM | |
381 | af AF_INET6 or AF_INET for the socket type | |
2ea97746 | 382 | hostname host name, or ip address (as text) |
420a0d19 CE |
383 | portlo,porthi the remote port range |
384 | timeout a timeout | |
2ea97746 | 385 | connhost if not NULL, host_item to be filled in with connection details |
420a0d19 | 386 | errstr pointer for allocated string on error |
2ea97746 CE |
387 | fastopen_blob with SOCK_STREAM, if non-null, request TCP Fast Open. |
388 | Additionally, optional idempotent early-data to send | |
420a0d19 CE |
389 | |
390 | Return: | |
391 | socket fd, or -1 on failure (having allocated an error string) | |
392 | */ | |
393 | int | |
394 | ip_connectedsocket(int type, const uschar * hostname, int portlo, int porthi, | |
2ea97746 | 395 | int timeout, host_item * connhost, uschar ** errstr, const blob * fastopen_blob) |
420a0d19 CE |
396 | { |
397 | int namelen, port; | |
398 | host_item shost; | |
399 | host_item *h; | |
400 | int af = 0, fd, fd4 = -1, fd6 = -1; | |
401 | ||
402 | shost.next = NULL; | |
403 | shost.address = NULL; | |
404 | shost.port = portlo; | |
405 | shost.mx = -1; | |
406 | ||
407 | namelen = Ustrlen(hostname); | |
408 | ||
409 | /* Anything enclosed in [] must be an IP address. */ | |
410 | ||
411 | if (hostname[0] == '[' && | |
412 | hostname[namelen - 1] == ']') | |
413 | { | |
2ea97746 | 414 | uschar * host = string_copyn(hostname+1, namelen-2); |
420a0d19 CE |
415 | if (string_is_ip_address(host, NULL) == 0) |
416 | { | |
417 | *errstr = string_sprintf("malformed IP address \"%s\"", hostname); | |
418 | return -1; | |
419 | } | |
420 | shost.name = shost.address = host; | |
421 | } | |
422 | ||
423 | /* Otherwise check for an unadorned IP address */ | |
424 | ||
425 | else if (string_is_ip_address(hostname, NULL) != 0) | |
2ea97746 | 426 | shost.name = shost.address = string_copyn(hostname, namelen); |
420a0d19 CE |
427 | |
428 | /* Otherwise lookup IP address(es) from the name */ | |
429 | ||
430 | else | |
431 | { | |
2ea97746 CE |
432 | shost.name = string_copyn(hostname, namelen); |
433 | if (host_find_byname(&shost, NULL, HOST_FIND_QUALIFY_SINGLE, | |
434 | NULL, FALSE) != HOST_FOUND) | |
420a0d19 CE |
435 | { |
436 | *errstr = string_sprintf("no IP address found for host %s", shost.name); | |
437 | return -1; | |
438 | } | |
439 | } | |
440 | ||
441 | /* Try to connect to the server - test each IP till one works */ | |
442 | ||
2ea97746 | 443 | for (h = &shost; h; h = h->next) |
420a0d19 | 444 | { |
2ea97746 CE |
445 | fd = Ustrchr(h->address, ':') != 0 |
446 | ? fd6 < 0 ? (fd6 = ip_socket(type, af = AF_INET6)) : fd6 | |
447 | : fd4 < 0 ? (fd4 = ip_socket(type, af = AF_INET )) : fd4; | |
420a0d19 CE |
448 | |
449 | if (fd < 0) | |
450 | { | |
451 | *errstr = string_sprintf("failed to create socket: %s", strerror(errno)); | |
452 | goto bad; | |
453 | } | |
454 | ||
455 | for(port = portlo; port <= porthi; port++) | |
2ea97746 | 456 | if (ip_connect(fd, af, h->address, port, timeout, fastopen_blob) == 0) |
420a0d19 CE |
457 | { |
458 | if (fd != fd6) close(fd6); | |
459 | if (fd != fd4) close(fd4); | |
2ea97746 CE |
460 | if (connhost) |
461 | { | |
420a0d19 CE |
462 | h->port = port; |
463 | *connhost = *h; | |
464 | connhost->next = NULL; | |
465 | } | |
466 | return fd; | |
467 | } | |
468 | } | |
469 | ||
2ea97746 CE |
470 | *errstr = string_sprintf("failed to connect to any address for %s: %s", |
471 | hostname, strerror(errno)); | |
420a0d19 CE |
472 | |
473 | bad: | |
474 | close(fd4); close(fd6); return -1; | |
475 | } | |
476 | ||
477 | ||
2ea97746 CE |
478 | /*XXX TFO? */ |
479 | int | |
480 | ip_tcpsocket(const uschar * hostport, uschar ** errstr, int tmo) | |
481 | { | |
482 | int scan; | |
483 | uschar hostname[256]; | |
484 | unsigned int portlow, porthigh; | |
485 | ||
486 | /* extract host and port part */ | |
487 | scan = sscanf(CS hostport, "%255s %u-%u", hostname, &portlow, &porthigh); | |
488 | if (scan != 3) | |
489 | { | |
490 | if (scan != 2) | |
491 | { | |
492 | *errstr = string_sprintf("invalid socket '%s'", hostport); | |
493 | return -1; | |
494 | } | |
495 | porthigh = portlow; | |
496 | } | |
497 | ||
498 | return ip_connectedsocket(SOCK_STREAM, hostname, portlow, porthigh, | |
499 | tmo, NULL, errstr, NULL); | |
500 | } | |
501 | ||
502 | int | |
503 | ip_unixsocket(const uschar * path, uschar ** errstr) | |
504 | { | |
505 | int sock; | |
506 | struct sockaddr_un server; | |
507 | ||
508 | if ((sock = socket(AF_UNIX, SOCK_STREAM, 0)) < 0) | |
509 | { | |
510 | *errstr = US"can't open UNIX socket."; | |
511 | return -1; | |
512 | } | |
513 | ||
514 | callout_address = string_copy(path); | |
515 | server.sun_family = AF_UNIX; | |
516 | Ustrncpy(server.sun_path, path, sizeof(server.sun_path)-1); | |
517 | server.sun_path[sizeof(server.sun_path)-1] = '\0'; | |
518 | if (connect(sock, (struct sockaddr *) &server, sizeof(server)) < 0) | |
519 | { | |
520 | int err = errno; | |
521 | (void)close(sock); | |
522 | *errstr = string_sprintf("unable to connect to UNIX socket (%s): %s", | |
523 | path, strerror(err)); | |
524 | return -1; | |
525 | } | |
526 | return sock; | |
527 | } | |
528 | ||
529 | int | |
530 | ip_streamsocket(const uschar * spec, uschar ** errstr, int tmo) | |
531 | { | |
532 | return *spec == '/' | |
533 | ? ip_unixsocket(spec, errstr) : ip_tcpsocket(spec, errstr, tmo); | |
534 | } | |
535 | ||
420a0d19 CE |
536 | /************************************************* |
537 | * Set keepalive on a socket * | |
538 | *************************************************/ | |
539 | ||
540 | /* Can be called for both incoming and outgoing sockets. | |
541 | ||
542 | Arguments: | |
543 | sock the socket | |
544 | address the remote host address, for failure logging | |
545 | torf true for outgoing connection, false for incoming | |
546 | ||
547 | Returns: nothing | |
548 | */ | |
549 | ||
550 | void | |
2ea97746 | 551 | ip_keepalive(int sock, const uschar *address, BOOL torf) |
420a0d19 CE |
552 | { |
553 | int fodder = 1; | |
554 | if (setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, | |
2ea97746 | 555 | US (&fodder), sizeof(fodder)) != 0) |
420a0d19 CE |
556 | log_write(0, LOG_MAIN, "setsockopt(SO_KEEPALIVE) on connection %s %s " |
557 | "failed: %s", torf? "to":"from", address, strerror(errno)); | |
558 | } | |
559 | ||
560 | ||
561 | ||
562 | /************************************************* | |
563 | * Receive from a socket with timeout * | |
564 | *************************************************/ | |
565 | ||
2ea97746 | 566 | /* |
420a0d19 | 567 | Arguments: |
2ea97746 CE |
568 | fd the file descriptor |
569 | timeout the timeout, seconds | |
570 | Returns: TRUE => ready for i/o | |
571 | FALSE => timed out, or other error | |
420a0d19 | 572 | */ |
2ea97746 CE |
573 | BOOL |
574 | fd_ready(int fd, int timeout) | |
420a0d19 CE |
575 | { |
576 | fd_set select_inset; | |
420a0d19 | 577 | time_t start_recv = time(NULL); |
2ea97746 | 578 | int time_left = timeout; |
420a0d19 CE |
579 | int rc; |
580 | ||
2ea97746 CE |
581 | if (time_left <= 0) |
582 | { | |
583 | errno = ETIMEDOUT; | |
584 | return FALSE; | |
585 | } | |
420a0d19 CE |
586 | /* Wait until the socket is ready */ |
587 | ||
2ea97746 | 588 | do |
420a0d19 | 589 | { |
2ea97746 | 590 | struct timeval tv = { .tv_sec = time_left, .tv_usec = 0 }; |
420a0d19 | 591 | FD_ZERO (&select_inset); |
2ea97746 | 592 | FD_SET (fd, &select_inset); |
420a0d19 | 593 | |
2ea97746 CE |
594 | /*DEBUG(D_transport) debug_printf("waiting for data on fd\n");*/ |
595 | rc = select(fd + 1, (SELECT_ARG2_TYPE *)&select_inset, NULL, NULL, &tv); | |
420a0d19 CE |
596 | |
597 | /* If some interrupt arrived, just retry. We presume this to be rare, | |
598 | but it can happen (e.g. the SIGUSR1 signal sent by exiwhat causes | |
599 | select() to exit). | |
600 | ||
601 | Aug 2004: Somebody set up a cron job that ran exiwhat every 2 minutes, making | |
602 | the interrupt not at all rare. Since the timeout is typically more than 2 | |
603 | minutes, the effect was to block the timeout completely. To prevent this | |
2ea97746 CE |
604 | happening again, we do an explicit time test and adjust the timeout |
605 | accordingly */ | |
420a0d19 CE |
606 | |
607 | if (rc < 0 && errno == EINTR) | |
608 | { | |
609 | DEBUG(D_transport) debug_printf("EINTR while waiting for socket data\n"); | |
420a0d19 | 610 | |
2ea97746 CE |
611 | /* Watch out, 'continue' jumps to the condition, not to the loops top */ |
612 | time_left = timeout - (time(NULL) - start_recv); | |
613 | if (time_left > 0) continue; | |
614 | } | |
420a0d19 CE |
615 | |
616 | if (rc <= 0) | |
617 | { | |
618 | errno = ETIMEDOUT; | |
2ea97746 | 619 | return FALSE; |
420a0d19 CE |
620 | } |
621 | ||
2ea97746 CE |
622 | /* Checking the FD_ISSET is not enough, if we're interrupted, the |
623 | select_inset may still contain the 'input'. */ | |
420a0d19 | 624 | } |
2ea97746 CE |
625 | while (rc < 0 || !FD_ISSET(fd, &select_inset)); |
626 | return TRUE; | |
627 | } | |
628 | ||
629 | /* The timeout is implemented using select(), and we loop to cover select() | |
630 | getting interrupted, and the possibility of select() returning with a positive | |
631 | result but no ready descriptor. Is this in fact possible? | |
632 | ||
633 | Arguments: | |
634 | cctx the connection context (socket fd, possibly TLS context) | |
635 | buffer to read into | |
636 | bufsize the buffer size | |
637 | timeout the timeout | |
638 | ||
639 | Returns: > 0 => that much data read | |
640 | <= 0 on error or EOF; errno set - zero for EOF | |
641 | */ | |
642 | ||
643 | int | |
644 | ip_recv(client_conn_ctx * cctx, uschar * buffer, int buffsize, int timeout) | |
645 | { | |
646 | int rc; | |
647 | ||
648 | if (!fd_ready(cctx->sock, timeout)) | |
649 | return -1; | |
420a0d19 CE |
650 | |
651 | /* The socket is ready, read from it (via TLS if it's active). On EOF (i.e. | |
652 | close down of the connection), set errno to zero; otherwise leave it alone. */ | |
653 | ||
654 | #ifdef SUPPORT_TLS | |
2ea97746 CE |
655 | if (cctx->tls_ctx) /* client TLS */ |
656 | rc = tls_read(cctx->tls_ctx, buffer, buffsize); | |
657 | else if (tls_in.active.sock == cctx->sock) /* server TLS */ | |
658 | rc = tls_read(NULL, buffer, buffsize); | |
420a0d19 CE |
659 | else |
660 | #endif | |
2ea97746 | 661 | rc = recv(cctx->sock, buffer, buffsize, 0); |
420a0d19 CE |
662 | |
663 | if (rc > 0) return rc; | |
664 | if (rc == 0) errno = 0; | |
665 | return -1; | |
666 | } | |
667 | ||
668 | ||
669 | ||
670 | ||
671 | /************************************************* | |
672 | * Lookup address family of potential socket * | |
673 | *************************************************/ | |
674 | ||
675 | /* Given a file-descriptor, check to see if it's a socket and, if so, | |
676 | return the address family; detects IPv4 vs IPv6. If not a socket then | |
677 | return -1. | |
678 | ||
679 | The value 0 is typically AF_UNSPEC, which should not be seen on a connected | |
680 | fd. If the return is -1, the errno will be from getsockname(); probably | |
681 | ENOTSOCK or ECONNRESET. | |
682 | ||
683 | Arguments: socket-or-not fd | |
684 | Returns: address family or -1 | |
685 | */ | |
686 | ||
687 | int | |
688 | ip_get_address_family(int fd) | |
689 | { | |
690 | struct sockaddr_storage ss; | |
691 | socklen_t sslen = sizeof(ss); | |
692 | ||
693 | if (getsockname(fd, (struct sockaddr *) &ss, &sslen) < 0) | |
694 | return -1; | |
695 | ||
696 | return (int) ss.ss_family; | |
697 | } | |
698 | ||
699 | ||
700 | ||
701 | ||
702 | /************************************************* | |
703 | * Lookup DSCP settings for a socket * | |
704 | *************************************************/ | |
705 | ||
706 | struct dscp_name_tableentry { | |
707 | const uschar *name; | |
708 | int value; | |
709 | }; | |
710 | /* Keep both of these tables sorted! */ | |
711 | static struct dscp_name_tableentry dscp_table[] = { | |
712 | #ifdef IPTOS_DSCP_AF11 | |
713 | { CUS"af11", IPTOS_DSCP_AF11 }, | |
714 | { CUS"af12", IPTOS_DSCP_AF12 }, | |
715 | { CUS"af13", IPTOS_DSCP_AF13 }, | |
716 | { CUS"af21", IPTOS_DSCP_AF21 }, | |
717 | { CUS"af22", IPTOS_DSCP_AF22 }, | |
718 | { CUS"af23", IPTOS_DSCP_AF23 }, | |
719 | { CUS"af31", IPTOS_DSCP_AF31 }, | |
720 | { CUS"af32", IPTOS_DSCP_AF32 }, | |
721 | { CUS"af33", IPTOS_DSCP_AF33 }, | |
722 | { CUS"af41", IPTOS_DSCP_AF41 }, | |
723 | { CUS"af42", IPTOS_DSCP_AF42 }, | |
724 | { CUS"af43", IPTOS_DSCP_AF43 }, | |
725 | { CUS"ef", IPTOS_DSCP_EF }, | |
726 | #endif | |
727 | #ifdef IPTOS_LOWCOST | |
728 | { CUS"lowcost", IPTOS_LOWCOST }, | |
729 | #endif | |
730 | { CUS"lowdelay", IPTOS_LOWDELAY }, | |
731 | #ifdef IPTOS_MINCOST | |
732 | { CUS"mincost", IPTOS_MINCOST }, | |
733 | #endif | |
734 | { CUS"reliability", IPTOS_RELIABILITY }, | |
735 | { CUS"throughput", IPTOS_THROUGHPUT } | |
736 | }; | |
737 | static int dscp_table_size = | |
738 | sizeof(dscp_table) / sizeof(struct dscp_name_tableentry); | |
739 | ||
740 | /* DSCP values change by protocol family, and so do the options used for | |
741 | setsockopt(); this utility does all the lookups. It takes an unexpanded | |
742 | option string, expands it, strips off affix whitespace, then checks if it's | |
743 | a number. If all of what's left is a number, then that's how the option will | |
744 | be parsed and success/failure is a range check. If it's not all a number, | |
745 | then it must be a supported keyword. | |
746 | ||
747 | Arguments: | |
748 | dscp_name a string, so far unvalidated | |
749 | af address_family in use | |
750 | level setsockopt level to use | |
751 | optname setsockopt name to use | |
752 | dscp_value value for dscp_name | |
753 | ||
754 | Returns: TRUE if okay to setsockopt(), else FALSE | |
755 | ||
756 | *level and *optname may be set even if FALSE is returned | |
757 | */ | |
758 | ||
759 | BOOL | |
760 | dscp_lookup(const uschar *dscp_name, int af, | |
761 | int *level, int *optname, int *dscp_value) | |
762 | { | |
763 | uschar *dscp_lookup, *p; | |
764 | int first, last; | |
765 | long rawlong; | |
766 | ||
767 | if (af == AF_INET) | |
768 | { | |
769 | *level = IPPROTO_IP; | |
770 | *optname = IP_TOS; | |
771 | } | |
772 | #if HAVE_IPV6 && defined(IPV6_TCLASS) | |
773 | else if (af == AF_INET6) | |
774 | { | |
775 | *level = IPPROTO_IPV6; | |
776 | *optname = IPV6_TCLASS; | |
777 | } | |
778 | #endif | |
779 | else | |
780 | { | |
781 | DEBUG(D_transport) | |
782 | debug_printf("Unhandled address family %d in dscp_lookup()\n", af); | |
783 | return FALSE; | |
784 | } | |
785 | if (!dscp_name) | |
786 | { | |
787 | DEBUG(D_transport) | |
788 | debug_printf("[empty DSCP]\n"); | |
789 | return FALSE; | |
790 | } | |
791 | dscp_lookup = expand_string(US dscp_name); | |
792 | if (dscp_lookup == NULL || *dscp_lookup == '\0') | |
793 | return FALSE; | |
794 | ||
795 | p = dscp_lookup + Ustrlen(dscp_lookup) - 1; | |
796 | while (isspace(*p)) *p-- = '\0'; | |
797 | while (isspace(*dscp_lookup) && dscp_lookup < p) dscp_lookup++; | |
798 | if (*dscp_lookup == '\0') | |
799 | return FALSE; | |
800 | ||
801 | rawlong = Ustrtol(dscp_lookup, &p, 0); | |
802 | if (p != dscp_lookup && *p == '\0') | |
803 | { | |
804 | /* We have six bits available, which will end up shifted to fit in 0xFC mask. | |
805 | RFC 2597 defines the values unshifted. */ | |
806 | if (rawlong < 0 || rawlong > 0x3F) | |
807 | { | |
808 | DEBUG(D_transport) | |
809 | debug_printf("DSCP value %ld out of range, ignored.\n", rawlong); | |
810 | return FALSE; | |
811 | } | |
812 | *dscp_value = rawlong << 2; | |
813 | return TRUE; | |
814 | } | |
815 | ||
816 | first = 0; | |
817 | last = dscp_table_size; | |
818 | while (last > first) | |
819 | { | |
820 | int middle = (first + last)/2; | |
821 | int c = Ustrcmp(dscp_lookup, dscp_table[middle].name); | |
822 | if (c == 0) | |
823 | { | |
824 | *dscp_value = dscp_table[middle].value; | |
825 | return TRUE; | |
826 | } | |
827 | else if (c > 0) | |
420a0d19 | 828 | first = middle + 1; |
420a0d19 | 829 | else |
420a0d19 | 830 | last = middle; |
420a0d19 CE |
831 | } |
832 | return FALSE; | |
833 | } | |
834 | ||
835 | void | |
836 | dscp_list_to_stream(FILE *stream) | |
837 | { | |
838 | int i; | |
839 | for (i=0; i < dscp_table_size; ++i) | |
840 | fprintf(stream, "%s\n", dscp_table[i].name); | |
841 | } | |
842 | ||
843 | ||
844 | /* End of ip.c */ | |
2ea97746 CE |
845 | /* vi: aw ai sw=2 |
846 | */ |