1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
5 /* Copyright (c) University of Cambridge 1995 - 2017 */
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, especially 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 non-null iff TCP_FASTOPEN can be used; may indicate early-data to
179 be sent in SYN segment
181 Returns: 0 on success; -1 on failure, with errno set
185 ip_connect(int sock, int af, const uschar *address, int port, int timeout,
186 const blob * fastopen)
188 struct sockaddr_in s_in4;
189 struct sockaddr *s_ptr;
190 int s_len, rc, save_errno;
192 /* For an IPv6 address, use an IPv6 sockaddr structure. */
195 struct sockaddr_in6 s_in6;
198 memset(&s_in6, 0, sizeof(s_in6));
199 ip_addrinfo(address, &s_in6); /* Panic-dies on error */
200 s_in6.sin6_port = htons(port);
201 s_ptr = (struct sockaddr *)&s_in6;
202 s_len = sizeof(s_in6);
205 #else /* HAVE_IPV6 */
206 af = af; /* Avoid compiler warning */
207 #endif /* HAVE_IPV6 */
209 /* For an IPv4 address, use an IPv4 sockaddr structure, even on a system with
213 memset(&s_in4, 0, sizeof(s_in4));
214 s_in4.sin_family = AF_INET;
215 s_in4.sin_port = htons(port);
216 s_in4.sin_addr.s_addr = (S_ADDR_TYPE)inet_addr(CCS address);
217 s_ptr = (struct sockaddr *)&s_in4;
218 s_len = sizeof(s_in4);
221 /* If no connection timeout is set, just call connect() without setting a
222 timer, thereby allowing the inbuilt OS timeout to operate. */
224 callout_address = string_sprintf("[%s]:%d", address, port);
225 sigalrm_seen = FALSE;
226 if (timeout > 0) alarm(timeout);
228 #if defined(TCP_FASTOPEN) && defined(MSG_FASTOPEN)
229 /* TCP Fast Open, if the system has a cookie from a previous call to
230 this peer, can send data in the SYN packet. The peer can send data
231 before it gets our ACK of its SYN,ACK - the latter is useful for
232 the SMTP banner. Other (than SMTP) cases of TCP connections can
233 possibly use the data-on-syn, so support that too. */
237 if ((rc = sendto(sock, fastopen->data, fastopen->len,
238 MSG_FASTOPEN | MSG_DONTWAIT, s_ptr, s_len)) >= 0)
240 DEBUG(D_transport|D_v)
241 debug_printf("TCP_FASTOPEN mode connection, with data\n");
242 tcp_out_fastopen = TRUE;
244 else if (errno == EINPROGRESS) /* expected for nonready peer */
248 DEBUG(D_transport|D_v)
249 debug_printf("TCP_FASTOPEN mode connection, no data\n");
250 tcp_out_fastopen = TRUE;
253 else if ( (rc = send(sock, fastopen->data, fastopen->len, 0)) < 0
254 && errno == EINPROGRESS) /* expected for nonready peer */
257 else if(errno == EOPNOTSUPP)
260 debug_printf("Tried TCP Fast Open but apparently not enabled by sysctl\n");
268 if ((rc = connect(sock, s_ptr, s_len)) >= 0)
269 if ( fastopen && fastopen->data && fastopen->len
270 && send(sock, fastopen->data, fastopen->len, 0) < 0)
277 /* There is a testing facility for simulating a connection timeout, as I
278 can't think of any other way of doing this. It converts a connection refused
279 into a timeout if the timeout is set to 999999. */
281 if (running_in_test_harness && save_errno == ECONNREFUSED && timeout == 999999)
293 /* A failure whose error code is "Interrupted system call" is in fact
294 an externally applied timeout if the signal handler has been run. */
296 errno = save_errno == EINTR && sigalrm_seen ? ETIMEDOUT : save_errno;
302 /*************************************************
303 * Create connected socket to remote host *
304 *************************************************/
306 /* Create a socket and connect to host (name or number, ipv6 ok)
307 at one of port-range.
310 type SOCK_DGRAM or SOCK_STREAM
311 af AF_INET6 or AF_INET for the socket type
312 address the remote address, in text form
313 portlo,porthi the remote port range
315 connhost if not NULL, host_item to be filled in with connection details
316 errstr pointer for allocated string on error
317 fastopen with SOCK_STREAM, if non-null, request TCP Fast Open.
318 Additionally, optional early-data to send
321 socket fd, or -1 on failure (having allocated an error string)
324 ip_connectedsocket(int type, const uschar * hostname, int portlo, int porthi,
325 int timeout, host_item * connhost, uschar ** errstr, const blob * fastopen)
330 int af = 0, fd, fd4 = -1, fd6 = -1;
333 shost.address = NULL;
337 namelen = Ustrlen(hostname);
339 /* Anything enclosed in [] must be an IP address. */
341 if (hostname[0] == '[' &&
342 hostname[namelen - 1] == ']')
344 uschar * host = string_copyn(hostname+1, namelen-2);
345 if (string_is_ip_address(host, NULL) == 0)
347 *errstr = string_sprintf("malformed IP address \"%s\"", hostname);
350 shost.name = shost.address = host;
353 /* Otherwise check for an unadorned IP address */
355 else if (string_is_ip_address(hostname, NULL) != 0)
356 shost.name = shost.address = string_copyn(hostname, namelen);
358 /* Otherwise lookup IP address(es) from the name */
362 shost.name = string_copyn(hostname, namelen);
363 if (host_find_byname(&shost, NULL, HOST_FIND_QUALIFY_SINGLE,
364 NULL, FALSE) != HOST_FOUND)
366 *errstr = string_sprintf("no IP address found for host %s", shost.name);
371 /* Try to connect to the server - test each IP till one works */
373 for (h = &shost; h; h = h->next)
375 fd = Ustrchr(h->address, ':') != 0
376 ? fd6 < 0 ? (fd6 = ip_socket(type, af = AF_INET6)) : fd6
377 : fd4 < 0 ? (fd4 = ip_socket(type, af = AF_INET )) : fd4;
381 *errstr = string_sprintf("failed to create socket: %s", strerror(errno));
385 for(port = portlo; port <= porthi; port++)
386 if (ip_connect(fd, af, h->address, port, timeout, fastopen) == 0)
388 if (fd != fd6) close(fd6);
389 if (fd != fd4) close(fd4);
394 connhost->next = NULL;
400 *errstr = string_sprintf("failed to connect to any address for %s: %s",
401 hostname, strerror(errno));
404 close(fd4); close(fd6); return -1;
410 ip_tcpsocket(const uschar * hostport, uschar ** errstr, int tmo)
413 uschar hostname[256];
414 unsigned int portlow, porthigh;
416 /* extract host and port part */
417 scan = sscanf(CS hostport, "%255s %u-%u", hostname, &portlow, &porthigh);
422 *errstr = string_sprintf("invalid socket '%s'", hostport);
428 return ip_connectedsocket(SOCK_STREAM, hostname, portlow, porthigh,
429 tmo, NULL, errstr, NULL);
433 ip_unixsocket(const uschar * path, uschar ** errstr)
436 struct sockaddr_un server;
438 if ((sock = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
440 *errstr = US"can't open UNIX socket.";
444 callout_address = string_copy(path);
445 server.sun_family = AF_UNIX;
446 Ustrncpy(server.sun_path, path, sizeof(server.sun_path)-1);
447 server.sun_path[sizeof(server.sun_path)-1] = '\0';
448 if (connect(sock, (struct sockaddr *) &server, sizeof(server)) < 0)
452 *errstr = string_sprintf("unable to connect to UNIX socket (%s): %s",
453 path, strerror(err));
460 ip_streamsocket(const uschar * spec, uschar ** errstr, int tmo)
463 ? ip_unixsocket(spec, errstr) : ip_tcpsocket(spec, errstr, tmo);
466 /*************************************************
467 * Set keepalive on a socket *
468 *************************************************/
470 /* Can be called for both incoming and outgoing sockets.
474 address the remote host address, for failure logging
475 torf true for outgoing connection, false for incoming
481 ip_keepalive(int sock, const uschar *address, BOOL torf)
484 if (setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
485 US (&fodder), sizeof(fodder)) != 0)
486 log_write(0, LOG_MAIN, "setsockopt(SO_KEEPALIVE) on connection %s %s "
487 "failed: %s", torf? "to":"from", address, strerror(errno));
492 /*************************************************
493 * Receive from a socket with timeout *
494 *************************************************/
498 fd the file descriptor
499 timeout the timeout, seconds
500 Returns: TRUE => ready for i/o
501 FALSE => timed out, or other error
504 fd_ready(int fd, int timeout)
507 time_t start_recv = time(NULL);
508 int time_left = timeout;
516 /* Wait until the socket is ready */
520 struct timeval tv = { .tv_sec = time_left, .tv_usec = 0 };
521 FD_ZERO (&select_inset);
522 FD_SET (fd, &select_inset);
524 /*DEBUG(D_transport) debug_printf("waiting for data on fd\n");*/
525 rc = select(fd + 1, (SELECT_ARG2_TYPE *)&select_inset, NULL, NULL, &tv);
527 /* If some interrupt arrived, just retry. We presume this to be rare,
528 but it can happen (e.g. the SIGUSR1 signal sent by exiwhat causes
531 Aug 2004: Somebody set up a cron job that ran exiwhat every 2 minutes, making
532 the interrupt not at all rare. Since the timeout is typically more than 2
533 minutes, the effect was to block the timeout completely. To prevent this
534 happening again, we do an explicit time test and adjust the timeout
537 if (rc < 0 && errno == EINTR)
539 DEBUG(D_transport) debug_printf("EINTR while waiting for socket data\n");
541 /* Watch out, 'continue' jumps to the condition, not to the loops top */
542 time_left = timeout - (time(NULL) - start_recv);
543 if (time_left > 0) continue;
552 /* Checking the FD_ISSET is not enough, if we're interrupted, the
553 select_inset may still contain the 'input'. */
555 while (rc < 0 || !FD_ISSET(fd, &select_inset));
559 /* The timeout is implemented using select(), and we loop to cover select()
560 getting interrupted, and the possibility of select() returning with a positive
561 result but no ready descriptor. Is this in fact possible?
566 bufsize the buffer size
569 Returns: > 0 => that much data read
570 <= 0 on error or EOF; errno set - zero for EOF
574 ip_recv(int sock, uschar *buffer, int buffsize, int timeout)
578 if (!fd_ready(sock, timeout))
581 /* The socket is ready, read from it (via TLS if it's active). On EOF (i.e.
582 close down of the connection), set errno to zero; otherwise leave it alone. */
585 if (tls_out.active == sock)
586 rc = tls_read(FALSE, buffer, buffsize);
587 else if (tls_in.active == sock)
588 rc = tls_read(TRUE, buffer, buffsize);
591 rc = recv(sock, buffer, buffsize, 0);
593 if (rc > 0) return rc;
594 if (rc == 0) errno = 0;
601 /*************************************************
602 * Lookup address family of potential socket *
603 *************************************************/
605 /* Given a file-descriptor, check to see if it's a socket and, if so,
606 return the address family; detects IPv4 vs IPv6. If not a socket then
609 The value 0 is typically AF_UNSPEC, which should not be seen on a connected
610 fd. If the return is -1, the errno will be from getsockname(); probably
611 ENOTSOCK or ECONNRESET.
613 Arguments: socket-or-not fd
614 Returns: address family or -1
618 ip_get_address_family(int fd)
620 struct sockaddr_storage ss;
621 socklen_t sslen = sizeof(ss);
623 if (getsockname(fd, (struct sockaddr *) &ss, &sslen) < 0)
626 return (int) ss.ss_family;
632 /*************************************************
633 * Lookup DSCP settings for a socket *
634 *************************************************/
636 struct dscp_name_tableentry {
640 /* Keep both of these tables sorted! */
641 static struct dscp_name_tableentry dscp_table[] = {
642 #ifdef IPTOS_DSCP_AF11
643 { CUS"af11", IPTOS_DSCP_AF11 },
644 { CUS"af12", IPTOS_DSCP_AF12 },
645 { CUS"af13", IPTOS_DSCP_AF13 },
646 { CUS"af21", IPTOS_DSCP_AF21 },
647 { CUS"af22", IPTOS_DSCP_AF22 },
648 { CUS"af23", IPTOS_DSCP_AF23 },
649 { CUS"af31", IPTOS_DSCP_AF31 },
650 { CUS"af32", IPTOS_DSCP_AF32 },
651 { CUS"af33", IPTOS_DSCP_AF33 },
652 { CUS"af41", IPTOS_DSCP_AF41 },
653 { CUS"af42", IPTOS_DSCP_AF42 },
654 { CUS"af43", IPTOS_DSCP_AF43 },
655 { CUS"ef", IPTOS_DSCP_EF },
658 { CUS"lowcost", IPTOS_LOWCOST },
660 { CUS"lowdelay", IPTOS_LOWDELAY },
662 { CUS"mincost", IPTOS_MINCOST },
664 { CUS"reliability", IPTOS_RELIABILITY },
665 { CUS"throughput", IPTOS_THROUGHPUT }
667 static int dscp_table_size =
668 sizeof(dscp_table) / sizeof(struct dscp_name_tableentry);
670 /* DSCP values change by protocol family, and so do the options used for
671 setsockopt(); this utility does all the lookups. It takes an unexpanded
672 option string, expands it, strips off affix whitespace, then checks if it's
673 a number. If all of what's left is a number, then that's how the option will
674 be parsed and success/failure is a range check. If it's not all a number,
675 then it must be a supported keyword.
678 dscp_name a string, so far unvalidated
679 af address_family in use
680 level setsockopt level to use
681 optname setsockopt name to use
682 dscp_value value for dscp_name
684 Returns: TRUE if okay to setsockopt(), else FALSE
686 *level and *optname may be set even if FALSE is returned
690 dscp_lookup(const uschar *dscp_name, int af,
691 int *level, int *optname, int *dscp_value)
693 uschar *dscp_lookup, *p;
702 #if HAVE_IPV6 && defined(IPV6_TCLASS)
703 else if (af == AF_INET6)
705 *level = IPPROTO_IPV6;
706 *optname = IPV6_TCLASS;
712 debug_printf("Unhandled address family %d in dscp_lookup()\n", af);
718 debug_printf("[empty DSCP]\n");
721 dscp_lookup = expand_string(US dscp_name);
722 if (dscp_lookup == NULL || *dscp_lookup == '\0')
725 p = dscp_lookup + Ustrlen(dscp_lookup) - 1;
726 while (isspace(*p)) *p-- = '\0';
727 while (isspace(*dscp_lookup) && dscp_lookup < p) dscp_lookup++;
728 if (*dscp_lookup == '\0')
731 rawlong = Ustrtol(dscp_lookup, &p, 0);
732 if (p != dscp_lookup && *p == '\0')
734 /* We have six bits available, which will end up shifted to fit in 0xFC mask.
735 RFC 2597 defines the values unshifted. */
736 if (rawlong < 0 || rawlong > 0x3F)
739 debug_printf("DSCP value %ld out of range, ignored.\n", rawlong);
742 *dscp_value = rawlong << 2;
747 last = dscp_table_size;
750 int middle = (first + last)/2;
751 int c = Ustrcmp(dscp_lookup, dscp_table[middle].name);
754 *dscp_value = dscp_table[middle].value;
766 dscp_list_to_stream(FILE *stream)
769 for (i=0; i < dscp_table_size; ++i)
770 fprintf(stream, "%s\n", dscp_table[i].name);