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 sigalrm_seen = FALSE;
222 if (timeout > 0) alarm(timeout);
223 rc = connect(sock, s_ptr, s_len);
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. */
231 if (running_in_test_harness && save_errno == ECONNREFUSED && timeout == 999999)
242 callout_address = string_sprintf("[%s]:%d", address, port);
246 /* A failure whose error code is "Interrupted system call" is in fact
247 an externally applied timeout if the signal handler has been run. */
249 errno = save_errno == EINTR && sigalrm_seen ? ETIMEDOUT : save_errno;
255 /*************************************************
256 * Create connected socket to remote host *
257 *************************************************/
259 /* Create a socket and connect to host (name or number, ipv6 ok)
260 at one of port-range.
263 type SOCK_DGRAM or SOCK_STREAM
264 af AF_INET6 or AF_INET for the socket type
265 address the remote address, in text form
266 portlo,porthi the remote port range
268 connhost if not NULL, host_item filled in with connection details
269 errstr pointer for allocated string on error
272 socket fd, or -1 on failure (having allocated an error string)
275 ip_connectedsocket(int type, const uschar * hostname, int portlo, int porthi,
276 int timeout, host_item * connhost, uschar ** errstr)
281 int af = 0, fd, fd4 = -1, fd6 = -1;
284 shost.address = NULL;
288 namelen = Ustrlen(hostname);
290 /* Anything enclosed in [] must be an IP address. */
292 if (hostname[0] == '[' &&
293 hostname[namelen - 1] == ']')
295 uschar * host = string_copyn(hostname+1, namelen-2);
296 debug_printf("%s: 1\n", __FUNCTION__);
297 if (string_is_ip_address(host, NULL) == 0)
299 *errstr = string_sprintf("malformed IP address \"%s\"", hostname);
302 shost.name = shost.address = host;
305 /* Otherwise check for an unadorned IP address */
307 else if (string_is_ip_address(hostname, NULL) != 0)
309 debug_printf("%s: 2\n", __FUNCTION__);
310 shost.name = shost.address = string_copyn(hostname, namelen);
313 /* Otherwise lookup IP address(es) from the name */
317 debug_printf("%s: 3\n", __FUNCTION__);
318 shost.name = string_copyn(hostname, namelen);
319 if (host_find_byname(&shost, NULL, HOST_FIND_QUALIFY_SINGLE,
320 NULL, FALSE) != HOST_FOUND)
322 *errstr = string_sprintf("no IP address found for host %s", shost.name);
327 /* Try to connect to the server - test each IP till one works */
329 for (h = &shost; h; h = h->next)
331 debug_printf("%s: 4 '%s'\n", __FUNCTION__, h->address);
332 fd = Ustrchr(h->address, ':') != 0
333 ? fd6 < 0 ? (fd6 = ip_socket(type, af = AF_INET6)) : fd6
334 : fd4 < 0 ? (fd4 = ip_socket(type, af = AF_INET )) : fd4;
338 *errstr = string_sprintf("failed to create socket: %s", strerror(errno));
342 for(port = portlo; port <= porthi; port++)
343 if (ip_connect(fd, af, h->address, port, timeout) == 0)
345 if (fd != fd6) close(fd6);
346 if (fd != fd4) close(fd4);
351 connhost->next = NULL;
357 *errstr = string_sprintf("failed to connect to any address for %s: %s",
358 hostname, strerror(errno));
361 close(fd4); close(fd6); return -1;
366 ip_tcpsocket(const uschar * hostport, uschar ** errstr, int tmo)
369 uschar hostname[256];
370 unsigned int portlow, porthigh;
372 /* extract host and port part */
373 scan = sscanf(CS hostport, "%255s %u-%u", hostname, &portlow, &porthigh);
378 *errstr = string_sprintf("invalid socket '%s'", hostport);
384 return ip_connectedsocket(SOCK_STREAM, hostname, portlow, porthigh,
389 ip_unixsocket(const uschar * path, uschar ** errstr)
392 struct sockaddr_un server;
394 if ((sock = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
396 *errstr = US"can't open UNIX socket.";
400 server.sun_family = AF_UNIX;
401 Ustrncpy(server.sun_path, path, sizeof(server.sun_path)-1);
402 server.sun_path[sizeof(server.sun_path)-1] = '\0';
403 if (connect(sock, (struct sockaddr *) &server, sizeof(server)) < 0)
407 *errstr = string_sprintf("unable to connect to UNIX socket (%s): %s",
408 path, strerror(err));
411 callout_address = string_copy(path);
416 ip_streamsocket(const uschar * spec, uschar ** errstr, int tmo)
419 ? ip_unixsocket(spec, errstr) : ip_tcpsocket(spec, errstr, tmo);
422 /*************************************************
423 * Set keepalive on a socket *
424 *************************************************/
426 /* Can be called for both incoming and outgoing sockets.
430 address the remote host address, for failure logging
431 torf true for outgoing connection, false for incoming
437 ip_keepalive(int sock, const uschar *address, BOOL torf)
440 if (setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
441 (uschar *)(&fodder), sizeof(fodder)) != 0)
442 log_write(0, LOG_MAIN, "setsockopt(SO_KEEPALIVE) on connection %s %s "
443 "failed: %s", torf? "to":"from", address, strerror(errno));
448 /*************************************************
449 * Receive from a socket with timeout *
450 *************************************************/
454 fd the file descriptor
455 timeout the timeout, seconds
456 Returns: TRUE => ready for i/o
457 FALSE => timed out, or other error
460 fd_ready(int fd, int timeout)
463 time_t start_recv = time(NULL);
464 int time_left = timeout;
472 /* Wait until the socket is ready */
476 struct timeval tv = { time_left, 0 };
477 FD_ZERO (&select_inset);
478 FD_SET (fd, &select_inset);
480 /*DEBUG(D_transport) debug_printf("waiting for data on fd\n");*/
481 rc = select(fd + 1, (SELECT_ARG2_TYPE *)&select_inset, NULL, NULL, &tv);
483 /* If some interrupt arrived, just retry. We presume this to be rare,
484 but it can happen (e.g. the SIGUSR1 signal sent by exiwhat causes
487 Aug 2004: Somebody set up a cron job that ran exiwhat every 2 minutes, making
488 the interrupt not at all rare. Since the timeout is typically more than 2
489 minutes, the effect was to block the timeout completely. To prevent this
490 happening again, we do an explicit time test and adjust the timeout
493 if (rc < 0 && errno == EINTR)
495 DEBUG(D_transport) debug_printf("EINTR while waiting for socket data\n");
497 /* Watch out, 'continue' jumps to the condition, not to the loops top */
498 time_left = timeout - (time(NULL) - start_recv);
499 if (time_left > 0) continue;
508 /* Checking the FD_ISSET is not enough, if we're interrupted, the
509 select_inset may still contain the 'input'. */
511 while (rc < 0 || !FD_ISSET(fd, &select_inset));
515 /* The timeout is implemented using select(), and we loop to cover select()
516 getting interrupted, and the possibility of select() returning with a positive
517 result but no ready descriptor. Is this in fact possible?
522 bufsize the buffer size
525 Returns: > 0 => that much data read
526 <= 0 on error or EOF; errno set - zero for EOF
530 ip_recv(int sock, uschar *buffer, int buffsize, int timeout)
534 if (!fd_ready(sock, timeout))
537 /* The socket is ready, read from it (via TLS if it's active). On EOF (i.e.
538 close down of the connection), set errno to zero; otherwise leave it alone. */
541 if (tls_out.active == sock)
542 rc = tls_read(FALSE, buffer, buffsize);
543 else if (tls_in.active == sock)
544 rc = tls_read(TRUE, buffer, buffsize);
547 rc = recv(sock, buffer, buffsize, 0);
549 if (rc > 0) return rc;
550 if (rc == 0) errno = 0;
557 /*************************************************
558 * Lookup address family of potential socket *
559 *************************************************/
561 /* Given a file-descriptor, check to see if it's a socket and, if so,
562 return the address family; detects IPv4 vs IPv6. If not a socket then
565 The value 0 is typically AF_UNSPEC, which should not be seen on a connected
566 fd. If the return is -1, the errno will be from getsockname(); probably
567 ENOTSOCK or ECONNRESET.
569 Arguments: socket-or-not fd
570 Returns: address family or -1
574 ip_get_address_family(int fd)
576 struct sockaddr_storage ss;
577 socklen_t sslen = sizeof(ss);
579 if (getsockname(fd, (struct sockaddr *) &ss, &sslen) < 0)
582 return (int) ss.ss_family;
588 /*************************************************
589 * Lookup DSCP settings for a socket *
590 *************************************************/
592 struct dscp_name_tableentry {
596 /* Keep both of these tables sorted! */
597 static struct dscp_name_tableentry dscp_table[] = {
598 #ifdef IPTOS_DSCP_AF11
599 { CUS"af11", IPTOS_DSCP_AF11 },
600 { CUS"af12", IPTOS_DSCP_AF12 },
601 { CUS"af13", IPTOS_DSCP_AF13 },
602 { CUS"af21", IPTOS_DSCP_AF21 },
603 { CUS"af22", IPTOS_DSCP_AF22 },
604 { CUS"af23", IPTOS_DSCP_AF23 },
605 { CUS"af31", IPTOS_DSCP_AF31 },
606 { CUS"af32", IPTOS_DSCP_AF32 },
607 { CUS"af33", IPTOS_DSCP_AF33 },
608 { CUS"af41", IPTOS_DSCP_AF41 },
609 { CUS"af42", IPTOS_DSCP_AF42 },
610 { CUS"af43", IPTOS_DSCP_AF43 },
611 { CUS"ef", IPTOS_DSCP_EF },
614 { CUS"lowcost", IPTOS_LOWCOST },
616 { CUS"lowdelay", IPTOS_LOWDELAY },
618 { CUS"mincost", IPTOS_MINCOST },
620 { CUS"reliability", IPTOS_RELIABILITY },
621 { CUS"throughput", IPTOS_THROUGHPUT }
623 static int dscp_table_size =
624 sizeof(dscp_table) / sizeof(struct dscp_name_tableentry);
626 /* DSCP values change by protocol family, and so do the options used for
627 setsockopt(); this utility does all the lookups. It takes an unexpanded
628 option string, expands it, strips off affix whitespace, then checks if it's
629 a number. If all of what's left is a number, then that's how the option will
630 be parsed and success/failure is a range check. If it's not all a number,
631 then it must be a supported keyword.
634 dscp_name a string, so far unvalidated
635 af address_family in use
636 level setsockopt level to use
637 optname setsockopt name to use
638 dscp_value value for dscp_name
640 Returns: TRUE if okay to setsockopt(), else FALSE
642 *level and *optname may be set even if FALSE is returned
646 dscp_lookup(const uschar *dscp_name, int af,
647 int *level, int *optname, int *dscp_value)
649 uschar *dscp_lookup, *p;
658 #if HAVE_IPV6 && defined(IPV6_TCLASS)
659 else if (af == AF_INET6)
661 *level = IPPROTO_IPV6;
662 *optname = IPV6_TCLASS;
668 debug_printf("Unhandled address family %d in dscp_lookup()\n", af);
674 debug_printf("[empty DSCP]\n");
677 dscp_lookup = expand_string(US dscp_name);
678 if (dscp_lookup == NULL || *dscp_lookup == '\0')
681 p = dscp_lookup + Ustrlen(dscp_lookup) - 1;
682 while (isspace(*p)) *p-- = '\0';
683 while (isspace(*dscp_lookup) && dscp_lookup < p) dscp_lookup++;
684 if (*dscp_lookup == '\0')
687 rawlong = Ustrtol(dscp_lookup, &p, 0);
688 if (p != dscp_lookup && *p == '\0')
690 /* We have six bits available, which will end up shifted to fit in 0xFC mask.
691 RFC 2597 defines the values unshifted. */
692 if (rawlong < 0 || rawlong > 0x3F)
695 debug_printf("DSCP value %ld out of range, ignored.\n", rawlong);
698 *dscp_value = rawlong << 2;
703 last = dscp_table_size;
706 int middle = (first + last)/2;
707 int c = Ustrcmp(dscp_lookup, dscp_table[middle].name);
710 *dscp_value = dscp_table[middle].value;
722 dscp_list_to_stream(FILE *stream)
725 for (i=0; i < dscp_table_size; ++i)
726 fprintf(stream, "%s\n", dscp_table[i].name);