1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
5 /* Copyright (c) University of Cambridge 1995 - 2016 */
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)
178 fastopen TRUE iff TCP_FASTOPEN can be used
180 Returns: 0 on success; -1 on failure, with errno set
184 ip_connect(int sock, int af, const uschar *address, int port, int timeout,
187 struct sockaddr_in s_in4;
188 struct sockaddr *s_ptr;
189 int s_len, rc, save_errno;
191 /* For an IPv6 address, use an IPv6 sockaddr structure. */
194 struct sockaddr_in6 s_in6;
197 memset(&s_in6, 0, sizeof(s_in6));
198 ip_addrinfo(address, &s_in6); /* Panic-dies on error */
199 s_in6.sin6_port = htons(port);
200 s_ptr = (struct sockaddr *)&s_in6;
201 s_len = sizeof(s_in6);
204 #else /* HAVE_IPV6 */
205 af = af; /* Avoid compiler warning */
206 #endif /* HAVE_IPV6 */
208 /* For an IPv4 address, use an IPv4 sockaddr structure, even on a system with
212 memset(&s_in4, 0, sizeof(s_in4));
213 s_in4.sin_family = AF_INET;
214 s_in4.sin_port = htons(port);
215 s_in4.sin_addr.s_addr = (S_ADDR_TYPE)inet_addr(CCS address);
216 s_ptr = (struct sockaddr *)&s_in4;
217 s_len = sizeof(s_in4);
220 /* If no connection timeout is set, just call connect() without setting a
221 timer, thereby allowing the inbuilt OS timeout to operate. */
223 callout_address = string_sprintf("[%s]:%d", address, port);
224 sigalrm_seen = FALSE;
225 if (timeout > 0) alarm(timeout);
228 /* TCP Fast Open, if the system has a cookie from a previous call to
229 this peer, can send data in the SYN packet. The peer can send data
230 before it gets our ACK of its SYN,ACK - the latter is useful for
231 the SMTP banner. Is there any usage where the former might be?
232 We might extend the ip_connect() args for data if so. For now,
233 connect in FASTOPEN mode but with zero data.
238 if ( (rc = sendto(sock, NULL, 0, MSG_FASTOPEN, s_ptr, s_len)) < 0
239 && errno == EOPNOTSUPP
242 log_write(0, LOG_MAIN|LOG_PANIC,
243 "Tried TCP Fast Open but apparently not enabled by sysctl");
244 rc = connect(sock, s_ptr, s_len);
249 rc = connect(sock, s_ptr, s_len);
254 /* There is a testing facility for simulating a connection timeout, as I
255 can't think of any other way of doing this. It converts a connection refused
256 into a timeout if the timeout is set to 999999. */
258 if (running_in_test_harness && save_errno == ECONNREFUSED && timeout == 999999)
270 /* A failure whose error code is "Interrupted system call" is in fact
271 an externally applied timeout if the signal handler has been run. */
273 errno = save_errno == EINTR && sigalrm_seen ? ETIMEDOUT : save_errno;
279 /*************************************************
280 * Create connected socket to remote host *
281 *************************************************/
283 /* Create a socket and connect to host (name or number, ipv6 ok)
284 at one of port-range.
287 type SOCK_DGRAM or SOCK_STREAM
288 af AF_INET6 or AF_INET for the socket type
289 address the remote address, in text form
290 portlo,porthi the remote port range
292 connhost if not NULL, host_item filled in with connection details
293 errstr pointer for allocated string on error
296 socket fd, or -1 on failure (having allocated an error string)
299 ip_connectedsocket(int type, const uschar * hostname, int portlo, int porthi,
300 int timeout, host_item * connhost, uschar ** errstr)
305 int af = 0, fd, fd4 = -1, fd6 = -1;
308 shost.address = NULL;
312 namelen = Ustrlen(hostname);
314 /* Anything enclosed in [] must be an IP address. */
316 if (hostname[0] == '[' &&
317 hostname[namelen - 1] == ']')
319 uschar * host = string_copyn(hostname+1, namelen-2);
320 if (string_is_ip_address(host, NULL) == 0)
322 *errstr = string_sprintf("malformed IP address \"%s\"", hostname);
325 shost.name = shost.address = host;
328 /* Otherwise check for an unadorned IP address */
330 else if (string_is_ip_address(hostname, NULL) != 0)
331 shost.name = shost.address = string_copyn(hostname, namelen);
333 /* Otherwise lookup IP address(es) from the name */
337 shost.name = string_copyn(hostname, namelen);
338 if (host_find_byname(&shost, NULL, HOST_FIND_QUALIFY_SINGLE,
339 NULL, FALSE) != HOST_FOUND)
341 *errstr = string_sprintf("no IP address found for host %s", shost.name);
346 /* Try to connect to the server - test each IP till one works */
348 for (h = &shost; h; h = h->next)
350 fd = Ustrchr(h->address, ':') != 0
351 ? fd6 < 0 ? (fd6 = ip_socket(type, af = AF_INET6)) : fd6
352 : fd4 < 0 ? (fd4 = ip_socket(type, af = AF_INET )) : fd4;
356 *errstr = string_sprintf("failed to create socket: %s", strerror(errno));
360 for(port = portlo; port <= porthi; port++)
361 if (ip_connect(fd, af, h->address, port, timeout, type == SOCK_STREAM) == 0)
363 if (fd != fd6) close(fd6);
364 if (fd != fd4) close(fd4);
369 connhost->next = NULL;
375 *errstr = string_sprintf("failed to connect to any address for %s: %s",
376 hostname, strerror(errno));
379 close(fd4); close(fd6); return -1;
384 ip_tcpsocket(const uschar * hostport, uschar ** errstr, int tmo)
387 uschar hostname[256];
388 unsigned int portlow, porthigh;
390 /* extract host and port part */
391 scan = sscanf(CS hostport, "%255s %u-%u", hostname, &portlow, &porthigh);
396 *errstr = string_sprintf("invalid socket '%s'", hostport);
402 return ip_connectedsocket(SOCK_STREAM, hostname, portlow, porthigh,
407 ip_unixsocket(const uschar * path, uschar ** errstr)
410 struct sockaddr_un server;
412 if ((sock = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
414 *errstr = US"can't open UNIX socket.";
418 callout_address = string_copy(path);
419 server.sun_family = AF_UNIX;
420 Ustrncpy(server.sun_path, path, sizeof(server.sun_path)-1);
421 server.sun_path[sizeof(server.sun_path)-1] = '\0';
422 if (connect(sock, (struct sockaddr *) &server, sizeof(server)) < 0)
426 *errstr = string_sprintf("unable to connect to UNIX socket (%s): %s",
427 path, strerror(err));
434 ip_streamsocket(const uschar * spec, uschar ** errstr, int tmo)
437 ? ip_unixsocket(spec, errstr) : ip_tcpsocket(spec, errstr, tmo);
440 /*************************************************
441 * Set keepalive on a socket *
442 *************************************************/
444 /* Can be called for both incoming and outgoing sockets.
448 address the remote host address, for failure logging
449 torf true for outgoing connection, false for incoming
455 ip_keepalive(int sock, const uschar *address, BOOL torf)
458 if (setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
459 (uschar *)(&fodder), sizeof(fodder)) != 0)
460 log_write(0, LOG_MAIN, "setsockopt(SO_KEEPALIVE) on connection %s %s "
461 "failed: %s", torf? "to":"from", address, strerror(errno));
466 /*************************************************
467 * Receive from a socket with timeout *
468 *************************************************/
472 fd the file descriptor
473 timeout the timeout, seconds
474 Returns: TRUE => ready for i/o
475 FALSE => timed out, or other error
478 fd_ready(int fd, int timeout)
481 time_t start_recv = time(NULL);
482 int time_left = timeout;
490 /* Wait until the socket is ready */
494 struct timeval tv = { time_left, 0 };
495 FD_ZERO (&select_inset);
496 FD_SET (fd, &select_inset);
498 /*DEBUG(D_transport) debug_printf("waiting for data on fd\n");*/
499 rc = select(fd + 1, (SELECT_ARG2_TYPE *)&select_inset, NULL, NULL, &tv);
501 /* If some interrupt arrived, just retry. We presume this to be rare,
502 but it can happen (e.g. the SIGUSR1 signal sent by exiwhat causes
505 Aug 2004: Somebody set up a cron job that ran exiwhat every 2 minutes, making
506 the interrupt not at all rare. Since the timeout is typically more than 2
507 minutes, the effect was to block the timeout completely. To prevent this
508 happening again, we do an explicit time test and adjust the timeout
511 if (rc < 0 && errno == EINTR)
513 DEBUG(D_transport) debug_printf("EINTR while waiting for socket data\n");
515 /* Watch out, 'continue' jumps to the condition, not to the loops top */
516 time_left = timeout - (time(NULL) - start_recv);
517 if (time_left > 0) continue;
526 /* Checking the FD_ISSET is not enough, if we're interrupted, the
527 select_inset may still contain the 'input'. */
529 while (rc < 0 || !FD_ISSET(fd, &select_inset));
533 /* The timeout is implemented using select(), and we loop to cover select()
534 getting interrupted, and the possibility of select() returning with a positive
535 result but no ready descriptor. Is this in fact possible?
540 bufsize the buffer size
543 Returns: > 0 => that much data read
544 <= 0 on error or EOF; errno set - zero for EOF
548 ip_recv(int sock, uschar *buffer, int buffsize, int timeout)
552 if (!fd_ready(sock, timeout))
555 /* The socket is ready, read from it (via TLS if it's active). On EOF (i.e.
556 close down of the connection), set errno to zero; otherwise leave it alone. */
559 if (tls_out.active == sock)
560 rc = tls_read(FALSE, buffer, buffsize);
561 else if (tls_in.active == sock)
562 rc = tls_read(TRUE, buffer, buffsize);
565 rc = recv(sock, buffer, buffsize, 0);
567 if (rc > 0) return rc;
568 if (rc == 0) errno = 0;
575 /*************************************************
576 * Lookup address family of potential socket *
577 *************************************************/
579 /* Given a file-descriptor, check to see if it's a socket and, if so,
580 return the address family; detects IPv4 vs IPv6. If not a socket then
583 The value 0 is typically AF_UNSPEC, which should not be seen on a connected
584 fd. If the return is -1, the errno will be from getsockname(); probably
585 ENOTSOCK or ECONNRESET.
587 Arguments: socket-or-not fd
588 Returns: address family or -1
592 ip_get_address_family(int fd)
594 struct sockaddr_storage ss;
595 socklen_t sslen = sizeof(ss);
597 if (getsockname(fd, (struct sockaddr *) &ss, &sslen) < 0)
600 return (int) ss.ss_family;
606 /*************************************************
607 * Lookup DSCP settings for a socket *
608 *************************************************/
610 struct dscp_name_tableentry {
614 /* Keep both of these tables sorted! */
615 static struct dscp_name_tableentry dscp_table[] = {
616 #ifdef IPTOS_DSCP_AF11
617 { CUS"af11", IPTOS_DSCP_AF11 },
618 { CUS"af12", IPTOS_DSCP_AF12 },
619 { CUS"af13", IPTOS_DSCP_AF13 },
620 { CUS"af21", IPTOS_DSCP_AF21 },
621 { CUS"af22", IPTOS_DSCP_AF22 },
622 { CUS"af23", IPTOS_DSCP_AF23 },
623 { CUS"af31", IPTOS_DSCP_AF31 },
624 { CUS"af32", IPTOS_DSCP_AF32 },
625 { CUS"af33", IPTOS_DSCP_AF33 },
626 { CUS"af41", IPTOS_DSCP_AF41 },
627 { CUS"af42", IPTOS_DSCP_AF42 },
628 { CUS"af43", IPTOS_DSCP_AF43 },
629 { CUS"ef", IPTOS_DSCP_EF },
632 { CUS"lowcost", IPTOS_LOWCOST },
634 { CUS"lowdelay", IPTOS_LOWDELAY },
636 { CUS"mincost", IPTOS_MINCOST },
638 { CUS"reliability", IPTOS_RELIABILITY },
639 { CUS"throughput", IPTOS_THROUGHPUT }
641 static int dscp_table_size =
642 sizeof(dscp_table) / sizeof(struct dscp_name_tableentry);
644 /* DSCP values change by protocol family, and so do the options used for
645 setsockopt(); this utility does all the lookups. It takes an unexpanded
646 option string, expands it, strips off affix whitespace, then checks if it's
647 a number. If all of what's left is a number, then that's how the option will
648 be parsed and success/failure is a range check. If it's not all a number,
649 then it must be a supported keyword.
652 dscp_name a string, so far unvalidated
653 af address_family in use
654 level setsockopt level to use
655 optname setsockopt name to use
656 dscp_value value for dscp_name
658 Returns: TRUE if okay to setsockopt(), else FALSE
660 *level and *optname may be set even if FALSE is returned
664 dscp_lookup(const uschar *dscp_name, int af,
665 int *level, int *optname, int *dscp_value)
667 uschar *dscp_lookup, *p;
676 #if HAVE_IPV6 && defined(IPV6_TCLASS)
677 else if (af == AF_INET6)
679 *level = IPPROTO_IPV6;
680 *optname = IPV6_TCLASS;
686 debug_printf("Unhandled address family %d in dscp_lookup()\n", af);
692 debug_printf("[empty DSCP]\n");
695 dscp_lookup = expand_string(US dscp_name);
696 if (dscp_lookup == NULL || *dscp_lookup == '\0')
699 p = dscp_lookup + Ustrlen(dscp_lookup) - 1;
700 while (isspace(*p)) *p-- = '\0';
701 while (isspace(*dscp_lookup) && dscp_lookup < p) dscp_lookup++;
702 if (*dscp_lookup == '\0')
705 rawlong = Ustrtol(dscp_lookup, &p, 0);
706 if (p != dscp_lookup && *p == '\0')
708 /* We have six bits available, which will end up shifted to fit in 0xFC mask.
709 RFC 2597 defines the values unshifted. */
710 if (rawlong < 0 || rawlong > 0x3F)
713 debug_printf("DSCP value %ld out of range, ignored.\n", rawlong);
716 *dscp_value = rawlong << 2;
721 last = dscp_table_size;
724 int middle = (first + last)/2;
725 int c = Ustrcmp(dscp_lookup, dscp_table[middle].name);
728 *dscp_value = dscp_table[middle].value;
740 dscp_list_to_stream(FILE *stream)
743 for (i=0; i < dscp_table_size; ++i)
744 fprintf(stream, "%s\n", dscp_table[i].name);