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. Is there any usage where the former might be?
233 We might extend the ip_connect() args for data if so. For now,
234 connect in FASTOPEN mode but with zero data.
239 if ((rc
= sendto(sock
, fastopen
->data
, fastopen
->len
,
240 MSG_FASTOPEN
| MSG_DONTWAIT
, s_ptr
, s_len
)) < 0)
241 if (errno
== EINPROGRESS
) /* expected for nonready peer */
242 { /* queue the data */
243 if ( (rc
= send(sock
, fastopen
->data
, fastopen
->len
, 0)) < 0
244 && errno
== EINPROGRESS
) /* expected for nonready peer */
247 else if(errno
== EOPNOTSUPP
)
250 debug_printf("Tried TCP Fast Open but apparently not enabled by sysctl\n");
258 if ((rc
= connect(sock
, s_ptr
, s_len
)) >= 0)
259 if ( fastopen
&& fastopen
->data
&& fastopen
->len
260 && send(sock
, fastopen
->data
, fastopen
->len
, 0) < 0)
267 /* There is a testing facility for simulating a connection timeout, as I
268 can't think of any other way of doing this. It converts a connection refused
269 into a timeout if the timeout is set to 999999. */
271 if (running_in_test_harness
&& save_errno
== ECONNREFUSED
&& timeout
== 999999)
283 /* A failure whose error code is "Interrupted system call" is in fact
284 an externally applied timeout if the signal handler has been run. */
286 errno
= save_errno
== EINTR
&& sigalrm_seen
? ETIMEDOUT
: save_errno
;
292 /*************************************************
293 * Create connected socket to remote host *
294 *************************************************/
296 /* Create a socket and connect to host (name or number, ipv6 ok)
297 at one of port-range.
300 type SOCK_DGRAM or SOCK_STREAM
301 af AF_INET6 or AF_INET for the socket type
302 address the remote address, in text form
303 portlo,porthi the remote port range
305 connhost if not NULL, host_item filled in with connection details
306 errstr pointer for allocated string on error
307 XXX could add early-data support
310 socket fd, or -1 on failure (having allocated an error string)
313 ip_connectedsocket(int type
, const uschar
* hostname
, int portlo
, int porthi
,
314 int timeout
, host_item
* connhost
, uschar
** errstr
)
319 int af
= 0, fd
, fd4
= -1, fd6
= -1;
320 blob
* fastopen
= tcp_fastopen_ok
&& type
== SOCK_STREAM
321 ? &tcp_fastopen_nodata
: NULL
;
324 shost
.address
= NULL
;
328 namelen
= Ustrlen(hostname
);
330 /* Anything enclosed in [] must be an IP address. */
332 if (hostname
[0] == '[' &&
333 hostname
[namelen
- 1] == ']')
335 uschar
* host
= string_copyn(hostname
+1, namelen
-2);
336 if (string_is_ip_address(host
, NULL
) == 0)
338 *errstr
= string_sprintf("malformed IP address \"%s\"", hostname
);
341 shost
.name
= shost
.address
= host
;
344 /* Otherwise check for an unadorned IP address */
346 else if (string_is_ip_address(hostname
, NULL
) != 0)
347 shost
.name
= shost
.address
= string_copyn(hostname
, namelen
);
349 /* Otherwise lookup IP address(es) from the name */
353 shost
.name
= string_copyn(hostname
, namelen
);
354 if (host_find_byname(&shost
, NULL
, HOST_FIND_QUALIFY_SINGLE
,
355 NULL
, FALSE
) != HOST_FOUND
)
357 *errstr
= string_sprintf("no IP address found for host %s", shost
.name
);
362 /* Try to connect to the server - test each IP till one works */
364 for (h
= &shost
; h
; h
= h
->next
)
366 fd
= Ustrchr(h
->address
, ':') != 0
367 ? fd6
< 0 ? (fd6
= ip_socket(type
, af
= AF_INET6
)) : fd6
368 : fd4
< 0 ? (fd4
= ip_socket(type
, af
= AF_INET
)) : fd4
;
372 *errstr
= string_sprintf("failed to create socket: %s", strerror(errno
));
376 for(port
= portlo
; port
<= porthi
; port
++)
377 if (ip_connect(fd
, af
, h
->address
, port
, timeout
, fastopen
) == 0)
379 if (fd
!= fd6
) close(fd6
);
380 if (fd
!= fd4
) close(fd4
);
385 connhost
->next
= NULL
;
391 *errstr
= string_sprintf("failed to connect to any address for %s: %s",
392 hostname
, strerror(errno
));
395 close(fd4
); close(fd6
); return -1;
400 ip_tcpsocket(const uschar
* hostport
, uschar
** errstr
, int tmo
)
403 uschar hostname
[256];
404 unsigned int portlow
, porthigh
;
406 /* extract host and port part */
407 scan
= sscanf(CS hostport
, "%255s %u-%u", hostname
, &portlow
, &porthigh
);
412 *errstr
= string_sprintf("invalid socket '%s'", hostport
);
418 return ip_connectedsocket(SOCK_STREAM
, hostname
, portlow
, porthigh
,
423 ip_unixsocket(const uschar
* path
, uschar
** errstr
)
426 struct sockaddr_un server
;
428 if ((sock
= socket(AF_UNIX
, SOCK_STREAM
, 0)) < 0)
430 *errstr
= US
"can't open UNIX socket.";
434 callout_address
= string_copy(path
);
435 server
.sun_family
= AF_UNIX
;
436 Ustrncpy(server
.sun_path
, path
, sizeof(server
.sun_path
)-1);
437 server
.sun_path
[sizeof(server
.sun_path
)-1] = '\0';
438 if (connect(sock
, (struct sockaddr
*) &server
, sizeof(server
)) < 0)
442 *errstr
= string_sprintf("unable to connect to UNIX socket (%s): %s",
443 path
, strerror(err
));
450 ip_streamsocket(const uschar
* spec
, uschar
** errstr
, int tmo
)
453 ? ip_unixsocket(spec
, errstr
) : ip_tcpsocket(spec
, errstr
, tmo
);
456 /*************************************************
457 * Set keepalive on a socket *
458 *************************************************/
460 /* Can be called for both incoming and outgoing sockets.
464 address the remote host address, for failure logging
465 torf true for outgoing connection, false for incoming
471 ip_keepalive(int sock
, const uschar
*address
, BOOL torf
)
474 if (setsockopt(sock
, SOL_SOCKET
, SO_KEEPALIVE
,
475 US (&fodder
), sizeof(fodder
)) != 0)
476 log_write(0, LOG_MAIN
, "setsockopt(SO_KEEPALIVE) on connection %s %s "
477 "failed: %s", torf
? "to":"from", address
, strerror(errno
));
482 /*************************************************
483 * Receive from a socket with timeout *
484 *************************************************/
488 fd the file descriptor
489 timeout the timeout, seconds
490 Returns: TRUE => ready for i/o
491 FALSE => timed out, or other error
494 fd_ready(int fd
, int timeout
)
497 time_t start_recv
= time(NULL
);
498 int time_left
= timeout
;
506 /* Wait until the socket is ready */
510 struct timeval tv
= { .tv_sec
= time_left
, .tv_usec
= 0 };
511 FD_ZERO (&select_inset
);
512 FD_SET (fd
, &select_inset
);
514 /*DEBUG(D_transport) debug_printf("waiting for data on fd\n");*/
515 rc
= select(fd
+ 1, (SELECT_ARG2_TYPE
*)&select_inset
, NULL
, NULL
, &tv
);
517 /* If some interrupt arrived, just retry. We presume this to be rare,
518 but it can happen (e.g. the SIGUSR1 signal sent by exiwhat causes
521 Aug 2004: Somebody set up a cron job that ran exiwhat every 2 minutes, making
522 the interrupt not at all rare. Since the timeout is typically more than 2
523 minutes, the effect was to block the timeout completely. To prevent this
524 happening again, we do an explicit time test and adjust the timeout
527 if (rc
< 0 && errno
== EINTR
)
529 DEBUG(D_transport
) debug_printf("EINTR while waiting for socket data\n");
531 /* Watch out, 'continue' jumps to the condition, not to the loops top */
532 time_left
= timeout
- (time(NULL
) - start_recv
);
533 if (time_left
> 0) continue;
542 /* Checking the FD_ISSET is not enough, if we're interrupted, the
543 select_inset may still contain the 'input'. */
545 while (rc
< 0 || !FD_ISSET(fd
, &select_inset
));
549 /* The timeout is implemented using select(), and we loop to cover select()
550 getting interrupted, and the possibility of select() returning with a positive
551 result but no ready descriptor. Is this in fact possible?
556 bufsize the buffer size
559 Returns: > 0 => that much data read
560 <= 0 on error or EOF; errno set - zero for EOF
564 ip_recv(int sock
, uschar
*buffer
, int buffsize
, int timeout
)
568 if (!fd_ready(sock
, timeout
))
571 /* The socket is ready, read from it (via TLS if it's active). On EOF (i.e.
572 close down of the connection), set errno to zero; otherwise leave it alone. */
575 if (tls_out
.active
== sock
)
576 rc
= tls_read(FALSE
, buffer
, buffsize
);
577 else if (tls_in
.active
== sock
)
578 rc
= tls_read(TRUE
, buffer
, buffsize
);
581 rc
= recv(sock
, buffer
, buffsize
, 0);
583 if (rc
> 0) return rc
;
584 if (rc
== 0) errno
= 0;
591 /*************************************************
592 * Lookup address family of potential socket *
593 *************************************************/
595 /* Given a file-descriptor, check to see if it's a socket and, if so,
596 return the address family; detects IPv4 vs IPv6. If not a socket then
599 The value 0 is typically AF_UNSPEC, which should not be seen on a connected
600 fd. If the return is -1, the errno will be from getsockname(); probably
601 ENOTSOCK or ECONNRESET.
603 Arguments: socket-or-not fd
604 Returns: address family or -1
608 ip_get_address_family(int fd
)
610 struct sockaddr_storage ss
;
611 socklen_t sslen
= sizeof(ss
);
613 if (getsockname(fd
, (struct sockaddr
*) &ss
, &sslen
) < 0)
616 return (int) ss
.ss_family
;
622 /*************************************************
623 * Lookup DSCP settings for a socket *
624 *************************************************/
626 struct dscp_name_tableentry
{
630 /* Keep both of these tables sorted! */
631 static struct dscp_name_tableentry dscp_table
[] = {
632 #ifdef IPTOS_DSCP_AF11
633 { CUS
"af11", IPTOS_DSCP_AF11
},
634 { CUS
"af12", IPTOS_DSCP_AF12
},
635 { CUS
"af13", IPTOS_DSCP_AF13
},
636 { CUS
"af21", IPTOS_DSCP_AF21
},
637 { CUS
"af22", IPTOS_DSCP_AF22
},
638 { CUS
"af23", IPTOS_DSCP_AF23
},
639 { CUS
"af31", IPTOS_DSCP_AF31
},
640 { CUS
"af32", IPTOS_DSCP_AF32
},
641 { CUS
"af33", IPTOS_DSCP_AF33
},
642 { CUS
"af41", IPTOS_DSCP_AF41
},
643 { CUS
"af42", IPTOS_DSCP_AF42
},
644 { CUS
"af43", IPTOS_DSCP_AF43
},
645 { CUS
"ef", IPTOS_DSCP_EF
},
648 { CUS
"lowcost", IPTOS_LOWCOST
},
650 { CUS
"lowdelay", IPTOS_LOWDELAY
},
652 { CUS
"mincost", IPTOS_MINCOST
},
654 { CUS
"reliability", IPTOS_RELIABILITY
},
655 { CUS
"throughput", IPTOS_THROUGHPUT
}
657 static int dscp_table_size
=
658 sizeof(dscp_table
) / sizeof(struct dscp_name_tableentry
);
660 /* DSCP values change by protocol family, and so do the options used for
661 setsockopt(); this utility does all the lookups. It takes an unexpanded
662 option string, expands it, strips off affix whitespace, then checks if it's
663 a number. If all of what's left is a number, then that's how the option will
664 be parsed and success/failure is a range check. If it's not all a number,
665 then it must be a supported keyword.
668 dscp_name a string, so far unvalidated
669 af address_family in use
670 level setsockopt level to use
671 optname setsockopt name to use
672 dscp_value value for dscp_name
674 Returns: TRUE if okay to setsockopt(), else FALSE
676 *level and *optname may be set even if FALSE is returned
680 dscp_lookup(const uschar
*dscp_name
, int af
,
681 int *level
, int *optname
, int *dscp_value
)
683 uschar
*dscp_lookup
, *p
;
692 #if HAVE_IPV6 && defined(IPV6_TCLASS)
693 else if (af
== AF_INET6
)
695 *level
= IPPROTO_IPV6
;
696 *optname
= IPV6_TCLASS
;
702 debug_printf("Unhandled address family %d in dscp_lookup()\n", af
);
708 debug_printf("[empty DSCP]\n");
711 dscp_lookup
= expand_string(US dscp_name
);
712 if (dscp_lookup
== NULL
|| *dscp_lookup
== '\0')
715 p
= dscp_lookup
+ Ustrlen(dscp_lookup
) - 1;
716 while (isspace(*p
)) *p
-- = '\0';
717 while (isspace(*dscp_lookup
) && dscp_lookup
< p
) dscp_lookup
++;
718 if (*dscp_lookup
== '\0')
721 rawlong
= Ustrtol(dscp_lookup
, &p
, 0);
722 if (p
!= dscp_lookup
&& *p
== '\0')
724 /* We have six bits available, which will end up shifted to fit in 0xFC mask.
725 RFC 2597 defines the values unshifted. */
726 if (rawlong
< 0 || rawlong
> 0x3F)
729 debug_printf("DSCP value %ld out of range, ignored.\n", rawlong
);
732 *dscp_value
= rawlong
<< 2;
737 last
= dscp_table_size
;
740 int middle
= (first
+ last
)/2;
741 int c
= Ustrcmp(dscp_lookup
, dscp_table
[middle
].name
);
744 *dscp_value
= dscp_table
[middle
].value
;
756 dscp_list_to_stream(FILE *stream
)
759 for (i
=0; i
< dscp_table_size
; ++i
)
760 fprintf(stream
, "%s\n", dscp_table
[i
].name
);