1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
5 /* Copyright (c) University of Cambridge 1995 - 2015 */
6 /* See the file NOTICE for conditions of use and distribution. */
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. */
17 /*************************************************
19 *************************************************/
21 /* Socket creation happens in a number of places so it's packaged here for
25 type SOCK_DGRAM or SOCK_STREAM
26 af AF_INET or AF_INET6
28 Returns: socket number or -1 on failure
32 ip_socket(int type, int af)
34 int sock = socket(af, type, 0);
36 log_write(0, LOG_MAIN, "IPv%c socket creation failed: %s",
37 (af == AF_INET6)? '6':'4', strerror(errno));
45 /*************************************************
46 * Convert printing address to numeric *
47 *************************************************/
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.
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).
60 address textual form of the address
61 addr where to copy back the answer
63 Returns: nothing - failure provokes a panic-die
67 ip_addrinfo(const uschar *address, struct sockaddr_in6 *saddr)
69 #ifdef IPV6_USE_INET_PTON
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;
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);
93 #endif /* HAVE_IPV6 */
96 /*************************************************
97 * Bind socket to interface and port *
98 *************************************************/
101 ip_addr(void * sin_, int af, const uschar * address, int port)
103 union sockaddr_46 * sin = sin_;
104 memset(sin, 0, sizeof(*sin));
106 /* Setup code when using an IPv6 socket. The wildcard address is ":", to
107 ensure an IPv6 socket is used. */
112 if (address[0] == ':' && address[1] == 0)
114 sin->v6.sin6_family = AF_INET6;
115 sin->v6.sin6_addr = in6addr_any;
118 ip_addrinfo(address, &sin->v6); /* Panic-dies on error */
119 sin->v6.sin6_port = htons(port);
120 return sizeof(sin->v6);
123 #else /* HAVE_IPv6 */
124 af = af; /* Avoid compiler warning */
125 #endif /* HAVE_IPV6 */
127 /* Setup code when using IPv4 socket. The wildcard address is "". */
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);
141 /* This function binds a socket to a local interface address and port. For a
142 wildcard IPv6 bind, the address is ":".
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)
150 Returns: the result of bind()
154 ip_bind(int sock, int af, uschar *address, int port)
156 union sockaddr_46 sin;
157 int s_len = ip_addr(&sin, af, address, port);
158 return bind(sock, (struct sockaddr *)&sin, s_len);
163 /*************************************************
164 * Connect socket to remote host *
165 *************************************************/
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.
174 af AF_INET6 or AF_INET for the socket type
175 address the remote address, in text form
177 timeout a timeout (zero for indefinite timeout)
179 Returns: 0 on success; -1 on failure, with errno set
183 ip_connect(int sock, int af, const uschar *address, int port, int timeout)
185 struct sockaddr_in s_in4;
186 struct sockaddr *s_ptr;
187 int s_len, rc, save_errno;
189 /* For an IPv6 address, use an IPv6 sockaddr structure. */
192 struct sockaddr_in6 s_in6;
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);
202 #else /* HAVE_IPV6 */
203 af = af; /* Avoid compiler warning */
204 #endif /* HAVE_IPV6 */
206 /* For an IPv4 address, use an IPv4 sockaddr structure, even on a system with
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);
218 /* If no connection timeout is set, just call connect() without setting a
219 timer, thereby allowing the inbuilt OS timeout to operate. */
221 callout_address = string_sprintf("[%s]:%d", address, port);
222 sigalrm_seen = FALSE;
223 if (timeout > 0) alarm(timeout);
224 rc = connect(sock, s_ptr, s_len);
228 /* There is a testing facility for simulating a connection timeout, as I
229 can't think of any other way of doing this. It converts a connection refused
230 into a timeout if the timeout is set to 999999. */
232 if (running_in_test_harness && save_errno == ECONNREFUSED && timeout == 999999)
244 /* A failure whose error code is "Interrupted system call" is in fact
245 an externally applied timeout if the signal handler has been run. */
247 errno = save_errno == EINTR && sigalrm_seen ? ETIMEDOUT : save_errno;
253 /*************************************************
254 * Create connected socket to remote host *
255 *************************************************/
257 /* Create a socket and connect to host (name or number, ipv6 ok)
258 at one of port-range.
261 type SOCK_DGRAM or SOCK_STREAM
262 af AF_INET6 or AF_INET for the socket type
263 address the remote address, in text form
264 portlo,porthi the remote port range
266 connhost if not NULL, host_item filled in with connection details
267 errstr pointer for allocated string on error
270 socket fd, or -1 on failure (having allocated an error string)
273 ip_connectedsocket(int type, const uschar * hostname, int portlo, int porthi,
274 int timeout, host_item * connhost, uschar ** errstr)
279 int af = 0, fd, fd4 = -1, fd6 = -1;
282 shost.address = NULL;
286 namelen = Ustrlen(hostname);
288 /* Anything enclosed in [] must be an IP address. */
290 if (hostname[0] == '[' &&
291 hostname[namelen - 1] == ']')
293 uschar * host = string_copyn(hostname+1, namelen-2);
294 if (string_is_ip_address(host, NULL) == 0)
296 *errstr = string_sprintf("malformed IP address \"%s\"", hostname);
299 shost.name = shost.address = host;
302 /* Otherwise check for an unadorned IP address */
304 else if (string_is_ip_address(hostname, NULL) != 0)
305 shost.name = shost.address = string_copyn(hostname, namelen);
307 /* Otherwise lookup IP address(es) from the name */
311 shost.name = string_copyn(hostname, namelen);
312 if (host_find_byname(&shost, NULL, HOST_FIND_QUALIFY_SINGLE,
313 NULL, FALSE) != HOST_FOUND)
315 *errstr = string_sprintf("no IP address found for host %s", shost.name);
320 /* Try to connect to the server - test each IP till one works */
322 for (h = &shost; h != NULL; h = h->next)
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;
330 *errstr = string_sprintf("failed to create socket: %s", strerror(errno));
334 for(port = portlo; port <= porthi; port++)
335 if (ip_connect(fd, af, h->address, port, timeout) == 0)
337 if (fd != fd6) close(fd6);
338 if (fd != fd4) close(fd4);
343 connhost->next = NULL;
349 *errstr = string_sprintf("failed to connect to any address for %s: %s",
350 hostname, strerror(errno));
353 close(fd4); close(fd6); return -1;
358 ip_tcpsocket(const uschar * hostport, uschar ** errstr, int tmo)
361 uschar hostname[256];
362 unsigned int portlow, porthigh;
364 /* extract host and port part */
365 scan = sscanf(CS hostport, "%255s %u-%u", hostname, &portlow, &porthigh);
370 *errstr = string_sprintf("invalid socket '%s'", hostport);
376 return ip_connectedsocket(SOCK_STREAM, hostname, portlow, porthigh,
381 ip_unixsocket(const uschar * path, uschar ** errstr)
384 struct sockaddr_un server;
386 if ((sock = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
388 *errstr = US"can't open UNIX socket.";
392 callout_address = string_copy(path);
393 server.sun_family = AF_UNIX;
394 Ustrncpy(server.sun_path, path, sizeof(server.sun_path)-1);
395 server.sun_path[sizeof(server.sun_path)-1] = '\0';
396 if (connect(sock, (struct sockaddr *) &server, sizeof(server)) < 0)
400 *errstr = string_sprintf("unable to connect to UNIX socket (%s): %s",
401 path, strerror(err));
408 ip_streamsocket(const uschar * spec, uschar ** errstr, int tmo)
411 ? ip_unixsocket(spec, errstr) : ip_tcpsocket(spec, errstr, tmo);
414 /*************************************************
415 * Set keepalive on a socket *
416 *************************************************/
418 /* Can be called for both incoming and outgoing sockets.
422 address the remote host address, for failure logging
423 torf true for outgoing connection, false for incoming
429 ip_keepalive(int sock, const uschar *address, BOOL torf)
432 if (setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
433 (uschar *)(&fodder), sizeof(fodder)) != 0)
434 log_write(0, LOG_MAIN, "setsockopt(SO_KEEPALIVE) on connection %s %s "
435 "failed: %s", torf? "to":"from", address, strerror(errno));
440 /*************************************************
441 * Receive from a socket with timeout *
442 *************************************************/
446 fd the file descriptor
447 timeout the timeout, seconds
448 Returns: TRUE => ready for i/o
449 FALSE => timed out, or other error
452 fd_ready(int fd, int timeout)
455 time_t start_recv = time(NULL);
456 int time_left = timeout;
464 /* Wait until the socket is ready */
468 struct timeval tv = { time_left, 0 };
469 FD_ZERO (&select_inset);
470 FD_SET (fd, &select_inset);
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);
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
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 and adjust the timeout
485 if (rc < 0 && errno == EINTR)
487 DEBUG(D_transport) debug_printf("EINTR while waiting for socket data\n");
489 /* Watch out, 'continue' jumps to the condition, not to the loops top */
490 time_left = timeout - (time(NULL) - start_recv);
491 if (time_left > 0) continue;
500 /* Checking the FD_ISSET is not enough, if we're interrupted, the
501 select_inset may still contain the 'input'. */
503 while (rc < 0 || !FD_ISSET(fd, &select_inset));
507 /* The timeout is implemented using select(), and we loop to cover select()
508 getting interrupted, and the possibility of select() returning with a positive
509 result but no ready descriptor. Is this in fact possible?
514 bufsize the buffer size
517 Returns: > 0 => that much data read
518 <= 0 on error or EOF; errno set - zero for EOF
522 ip_recv(int sock, uschar *buffer, int buffsize, int timeout)
526 if (!fd_ready(sock, timeout))
529 /* The socket is ready, read from it (via TLS if it's active). On EOF (i.e.
530 close down of the connection), set errno to zero; otherwise leave it alone. */
533 if (tls_out.active == sock)
534 rc = tls_read(FALSE, buffer, buffsize);
535 else if (tls_in.active == sock)
536 rc = tls_read(TRUE, buffer, buffsize);
539 rc = recv(sock, buffer, buffsize, 0);
541 if (rc > 0) return rc;
542 if (rc == 0) errno = 0;
549 /*************************************************
550 * Lookup address family of potential socket *
551 *************************************************/
553 /* Given a file-descriptor, check to see if it's a socket and, if so,
554 return the address family; detects IPv4 vs IPv6. If not a socket then
557 The value 0 is typically AF_UNSPEC, which should not be seen on a connected
558 fd. If the return is -1, the errno will be from getsockname(); probably
559 ENOTSOCK or ECONNRESET.
561 Arguments: socket-or-not fd
562 Returns: address family or -1
566 ip_get_address_family(int fd)
568 struct sockaddr_storage ss;
569 socklen_t sslen = sizeof(ss);
571 if (getsockname(fd, (struct sockaddr *) &ss, &sslen) < 0)
574 return (int) ss.ss_family;
580 /*************************************************
581 * Lookup DSCP settings for a socket *
582 *************************************************/
584 struct dscp_name_tableentry {
588 /* Keep both of these tables sorted! */
589 static struct dscp_name_tableentry dscp_table[] = {
590 #ifdef IPTOS_DSCP_AF11
591 { CUS"af11", IPTOS_DSCP_AF11 },
592 { CUS"af12", IPTOS_DSCP_AF12 },
593 { CUS"af13", IPTOS_DSCP_AF13 },
594 { CUS"af21", IPTOS_DSCP_AF21 },
595 { CUS"af22", IPTOS_DSCP_AF22 },
596 { CUS"af23", IPTOS_DSCP_AF23 },
597 { CUS"af31", IPTOS_DSCP_AF31 },
598 { CUS"af32", IPTOS_DSCP_AF32 },
599 { CUS"af33", IPTOS_DSCP_AF33 },
600 { CUS"af41", IPTOS_DSCP_AF41 },
601 { CUS"af42", IPTOS_DSCP_AF42 },
602 { CUS"af43", IPTOS_DSCP_AF43 },
603 { CUS"ef", IPTOS_DSCP_EF },
606 { CUS"lowcost", IPTOS_LOWCOST },
608 { CUS"lowdelay", IPTOS_LOWDELAY },
610 { CUS"mincost", IPTOS_MINCOST },
612 { CUS"reliability", IPTOS_RELIABILITY },
613 { CUS"throughput", IPTOS_THROUGHPUT }
615 static int dscp_table_size =
616 sizeof(dscp_table) / sizeof(struct dscp_name_tableentry);
618 /* DSCP values change by protocol family, and so do the options used for
619 setsockopt(); this utility does all the lookups. It takes an unexpanded
620 option string, expands it, strips off affix whitespace, then checks if it's
621 a number. If all of what's left is a number, then that's how the option will
622 be parsed and success/failure is a range check. If it's not all a number,
623 then it must be a supported keyword.
626 dscp_name a string, so far unvalidated
627 af address_family in use
628 level setsockopt level to use
629 optname setsockopt name to use
630 dscp_value value for dscp_name
632 Returns: TRUE if okay to setsockopt(), else FALSE
634 *level and *optname may be set even if FALSE is returned
638 dscp_lookup(const uschar *dscp_name, int af,
639 int *level, int *optname, int *dscp_value)
641 uschar *dscp_lookup, *p;
650 #if HAVE_IPV6 && defined(IPV6_TCLASS)
651 else if (af == AF_INET6)
653 *level = IPPROTO_IPV6;
654 *optname = IPV6_TCLASS;
660 debug_printf("Unhandled address family %d in dscp_lookup()\n", af);
666 debug_printf("[empty DSCP]\n");
669 dscp_lookup = expand_string(US dscp_name);
670 if (dscp_lookup == NULL || *dscp_lookup == '\0')
673 p = dscp_lookup + Ustrlen(dscp_lookup) - 1;
674 while (isspace(*p)) *p-- = '\0';
675 while (isspace(*dscp_lookup) && dscp_lookup < p) dscp_lookup++;
676 if (*dscp_lookup == '\0')
679 rawlong = Ustrtol(dscp_lookup, &p, 0);
680 if (p != dscp_lookup && *p == '\0')
682 /* We have six bits available, which will end up shifted to fit in 0xFC mask.
683 RFC 2597 defines the values unshifted. */
684 if (rawlong < 0 || rawlong > 0x3F)
687 debug_printf("DSCP value %ld out of range, ignored.\n", rawlong);
690 *dscp_value = rawlong << 2;
695 last = dscp_table_size;
698 int middle = (first + last)/2;
699 int c = Ustrcmp(dscp_lookup, dscp_table[middle].name);
702 *dscp_value = dscp_table[middle].value;
714 dscp_list_to_stream(FILE *stream)
717 for (i=0; i < dscp_table_size; ++i)
718 fprintf(stream, "%s\n", dscp_table[i].name);