1 /* Establishing and handling network connections.
2 Copyright (C) 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011 Free Software
6 This file is part of GNU Wget.
8 GNU Wget is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 3 of the License, or
11 (at your option) any later version.
13 GNU Wget is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with Wget. If not, see <http://www.gnu.org/licenses/>.
21 Additional permission under GNU GPL version 3 section 7
23 If you modify this program, or any covered work, by linking or
24 combining it with the OpenSSL project's OpenSSL library (or a
25 modified version of that library), containing parts covered by the
26 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
27 grants you additional permission to convey the resulting work.
28 Corresponding Source for a non-source form of such a combination
29 shall include the source code for the parts of OpenSSL used as well
30 as that of the covered work. */
39 #ifdef HAVE_SYS_SOCKET_H
40 # include <sys/socket.h>
41 #endif /* def HAVE_SYS_SOCKET_H */
43 #ifdef HAVE_SYS_SELECT_H
44 # include <sys/select.h>
45 #endif /* def HAVE_SYS_SELECT_H */
50 # else /* def __VMS */
52 # endif /* def __VMS [else] */
53 # include <netinet/in.h>
55 # include <arpa/inet.h>
57 #endif /* not WINDOWS */
67 /* Apparently needed for Interix: */
72 /* Define sockaddr_storage where unavailable (presumably on IPv4-only
76 # ifndef HAVE_STRUCT_SOCKADDR_STORAGE
77 # define sockaddr_storage sockaddr_in
79 #endif /* ENABLE_IPV6 */
81 /* Fill SA as per the data in IP and PORT. SA shoult point to struct
82 sockaddr_storage if ENABLE_IPV6 is defined, to struct sockaddr_in
86 sockaddr_set_data (struct sockaddr *sa, const ip_address *ip, int port)
92 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
94 sin->sin_family = AF_INET;
95 sin->sin_port = htons (port);
96 sin->sin_addr = ip->data.d4;
102 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
104 sin6->sin6_family = AF_INET6;
105 sin6->sin6_port = htons (port);
106 sin6->sin6_addr = ip->data.d6;
107 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
108 sin6->sin6_scope_id = ip->ipv6_scope;
112 #endif /* ENABLE_IPV6 */
118 /* Get the data of SA, specifically the IP address and the port. If
119 you're not interested in one or the other information, pass NULL as
123 sockaddr_get_data (const struct sockaddr *sa, ip_address *ip, int *port)
125 switch (sa->sa_family)
129 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
132 ip->family = AF_INET;
133 ip->data.d4 = sin->sin_addr;
136 *port = ntohs (sin->sin_port);
142 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
145 ip->family = AF_INET6;
146 ip->data.d6 = sin6->sin6_addr;
147 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
148 ip->ipv6_scope = sin6->sin6_scope_id;
152 *port = ntohs (sin6->sin6_port);
161 /* Return the size of the sockaddr structure depending on its
165 sockaddr_size (const struct sockaddr *sa)
167 switch (sa->sa_family)
170 return sizeof (struct sockaddr_in);
173 return sizeof (struct sockaddr_in6);
180 /* Resolve the bind address specified via --bind-address and store it
181 to SA. The resolved value is stored in a static variable and
182 reused after the first invocation of this function.
184 Returns true on success, false on failure. */
187 resolve_bind_address (struct sockaddr *sa)
189 struct address_list *al;
191 /* Make sure this is called only once. opt.bind_address doesn't
192 change during a Wget run. */
193 static bool called, should_bind;
194 static ip_address ip;
198 sockaddr_set_data (sa, &ip, 0);
203 al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
206 /* #### We should be able to print the error message here. */
207 logprintf (LOG_NOTQUIET,
208 _("%s: unable to resolve bind address %s; disabling bind.\n"),
209 exec_name, quote (opt.bind_address));
214 /* Pick the first address in the list and use it as bind address.
215 Perhaps we should try multiple addresses in succession, but I
216 don't think that's necessary in practice. */
217 ip = *address_list_address_at (al, 0);
218 address_list_release (al);
220 sockaddr_set_data (sa, &ip, 0);
227 const struct sockaddr *addr;
233 connect_with_timeout_callback (void *arg)
235 struct cwt_context *ctx = (struct cwt_context *)arg;
236 ctx->result = connect (ctx->fd, ctx->addr, ctx->addrlen);
239 /* Like connect, but specifies a timeout. If connecting takes longer
240 than TIMEOUT seconds, -1 is returned and errno is set to
244 connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
247 struct cwt_context ctx;
250 ctx.addrlen = addrlen;
252 if (run_with_timeout (timeout, connect_with_timeout_callback, &ctx))
257 if (ctx.result == -1 && errno == EINTR)
262 /* Connect via TCP to the specified address and port.
264 If PRINT is non-NULL, it is the host name to print that we're
268 connect_to_ip (const ip_address *ip, int port, const char *print)
270 struct sockaddr_storage ss;
271 struct sockaddr *sa = (struct sockaddr *)&ss;
274 /* If PRINT is non-NULL, print the "Connecting to..." line, with
275 PRINT being the host name we're connecting to. */
278 const char *txt_addr = print_address (ip);
279 if (0 != strcmp (print, txt_addr))
281 char *str = NULL, *name;
283 if (opt.enable_iri && (name = idn_decode ((char *) print)) != NULL)
285 int len = strlen (print) + strlen (name) + 4;
287 snprintf (str, len, "%s (%s)", name, print);
292 logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
293 str ? str : escnonprint_uri (print), txt_addr, port);
300 if (ip->family == AF_INET)
301 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
302 else if (ip->family == AF_INET6)
303 logprintf (LOG_VERBOSE, _("Connecting to [%s]:%d... "), txt_addr, port);
307 /* Store the sockaddr info to SA. */
308 sockaddr_set_data (sa, ip, port);
310 /* Create the socket of the family appropriate for the address. */
311 sock = socket (sa->sa_family, SOCK_STREAM, 0);
315 #if defined(ENABLE_IPV6) && defined(IPV6_V6ONLY)
318 /* In case of error, we will go on anyway... */
319 int err = setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof (on));
322 DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
326 /* For very small rate limits, set the buffer size (and hence,
327 hopefully, the kernel's TCP window size) to the per-second limit.
328 That way we should never have to sleep for more than 1s between
330 if (opt.limit_rate && opt.limit_rate < 8192)
332 int bufsize = opt.limit_rate;
334 bufsize = 512; /* avoid pathologically small values */
336 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
337 (void *)&bufsize, (socklen_t)sizeof (bufsize));
339 /* When we add limit_rate support for writing, which is useful
340 for POST, we should also set SO_SNDBUF here. */
343 if (opt.bind_address)
345 /* Bind the client side of the socket to the requested
347 struct sockaddr_storage bind_ss;
348 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
349 if (resolve_bind_address (bind_sa))
351 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
356 /* Connect the socket to the remote endpoint. */
357 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
358 opt.connect_timeout) < 0)
364 logprintf (LOG_VERBOSE, _("connected.\n"));
365 DEBUGP (("Created socket %d.\n", sock));
370 /* Protect errno from possible modifications by close and
372 int save_errno = errno;
376 logprintf (LOG_VERBOSE, _("failed: %s.\n"), strerror (errno));
382 /* Connect via TCP to a remote host on the specified port.
384 HOST is resolved as an Internet host name. If HOST resolves to
385 more than one IP address, they are tried in the order returned by
386 DNS until connecting to one of them succeeds. */
389 connect_to_host (const char *host, int port)
394 struct address_list *al = lookup_host (host, 0);
399 logprintf (LOG_NOTQUIET,
400 _("%s: unable to resolve host address %s\n"),
401 exec_name, quote (host));
405 address_list_get_bounds (al, &start, &end);
406 for (i = start; i < end; i++)
408 const ip_address *ip = address_list_address_at (al, i);
409 sock = connect_to_ip (ip, port, host);
413 address_list_set_connected (al);
414 address_list_release (al);
418 /* The attempt to connect has failed. Continue with the loop
419 and try next address. */
421 address_list_set_faulty (al, i);
424 /* Failed to connect to any of the addresses in AL. */
426 if (address_list_connected_p (al))
428 /* We connected to AL before, but cannot do so now. That might
429 indicate that our DNS cache entry for HOST has expired. */
430 address_list_release (al);
431 al = lookup_host (host, LH_REFRESH);
434 address_list_release (al);
439 /* Create a socket, bind it to local interface BIND_ADDRESS on port
440 *PORT, set up a listen backlog, and return the resulting socket, or
443 BIND_ADDRESS is the address of the interface to bind to. If it is
444 NULL, the socket is bound to the default address. PORT should
445 point to the port number that will be used for the binding. If
446 that number is 0, the system will choose a suitable port, and the
447 chosen value will be written to *PORT.
449 Calling accept() on such a socket waits for and accepts incoming
453 bind_local (const ip_address *bind_address, int *port)
456 struct sockaddr_storage ss;
457 struct sockaddr *sa = (struct sockaddr *)&ss;
459 /* For setting options with setsockopt. */
461 void *setopt_ptr = (void *)&setopt_val;
462 socklen_t setopt_size = sizeof (setopt_val);
464 sock = socket (bind_address->family, SOCK_STREAM, 0);
469 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
473 sockaddr_set_data (sa, bind_address, *port);
474 if (bind (sock, sa, sockaddr_size (sa)) < 0)
479 DEBUGP (("Local socket fd %d bound.\n", sock));
481 /* If *PORT is 0, find out which port we've bound to. */
484 socklen_t addrlen = sockaddr_size (sa);
485 if (getsockname (sock, sa, &addrlen) < 0)
487 /* If we can't find out the socket's local address ("name"),
488 something is seriously wrong with the socket, and it's
489 unusable for us anyway because we must know the chosen
494 sockaddr_get_data (sa, NULL, port);
495 DEBUGP (("binding to address %s using port %i.\n",
496 print_address (bind_address), *port));
498 if (listen (sock, 1) < 0)
506 /* Like a call to accept(), but with the added check for timeout.
508 In other words, accept a client connection on LOCAL_SOCK, and
509 return the new socket used for communication with the client.
510 LOCAL_SOCK should have been bound, e.g. using bind_local().
512 The caller is blocked until a connection is established. If no
513 connection is established for opt.connect_timeout seconds, the
514 function exits with an error status. */
517 accept_connection (int local_sock)
521 /* We don't need the values provided by accept, but accept
522 apparently requires them to be present. */
523 struct sockaddr_storage ss;
524 struct sockaddr *sa = (struct sockaddr *)&ss;
525 socklen_t addrlen = sizeof (ss);
527 if (opt.connect_timeout)
529 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
535 sock = accept (local_sock, sa, &addrlen);
536 DEBUGP (("Accepted client at socket %d.\n", sock));
540 /* Get the IP address associated with the connection on FD and store
541 it to IP. Return true on success, false otherwise.
543 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
544 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
545 returns the address of the remote (peer's) side of the socket. */
548 socket_ip_address (int sock, ip_address *ip, int endpoint)
550 struct sockaddr_storage storage;
551 struct sockaddr *sockaddr = (struct sockaddr *) &storage;
552 socklen_t addrlen = sizeof (storage);
555 memset (sockaddr, 0, addrlen);
556 if (endpoint == ENDPOINT_LOCAL)
557 ret = getsockname (sock, sockaddr, &addrlen);
558 else if (endpoint == ENDPOINT_PEER)
559 ret = getpeername (sock, sockaddr, &addrlen);
565 ip->family = sockaddr->sa_family;
566 switch (sockaddr->sa_family)
571 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
572 ip->data.d6 = sa6->sin6_addr;
573 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
574 ip->ipv6_scope = sa6->sin6_scope_id;
576 DEBUGP (("conaddr is: %s\n", print_address (ip)));
582 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
583 ip->data.d4 = sa->sin_addr;
584 DEBUGP (("conaddr is: %s\n", print_address (ip)));
592 /* Get the socket family of connection on FD and store
593 Return family type on success, -1 otherwise.
595 If ENDPOINT is ENDPOINT_LOCAL, it returns the sock family of the local
596 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
597 returns the sock family of the remote (peer's) side of the socket. */
600 socket_family (int sock, int endpoint)
602 struct sockaddr_storage storage;
603 struct sockaddr *sockaddr = (struct sockaddr *) &storage;
604 socklen_t addrlen = sizeof (storage);
607 memset (sockaddr, 0, addrlen);
609 if (endpoint == ENDPOINT_LOCAL)
610 ret = getsockname (sock, sockaddr, &addrlen);
611 else if (endpoint == ENDPOINT_PEER)
612 ret = getpeername (sock, sockaddr, &addrlen);
619 return sockaddr->sa_family;
622 /* Return true if the error from the connect code can be considered
623 retryable. Wget normally retries after errors, but the exception
624 are the "unsupported protocol" type errors (possible on IPv4/IPv6
625 dual family systems) and "connection refused". */
628 retryable_socket_connect_error (int err)
630 /* Have to guard against some of these values not being defined.
631 Cannot use a switch statement because some of the values might be
635 || err == EAFNOSUPPORT
638 || err == EPFNOSUPPORT
640 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
641 || err == ESOCKTNOSUPPORT
643 #ifdef EPROTONOSUPPORT
644 || err == EPROTONOSUPPORT
647 || err == ENOPROTOOPT
649 /* Apparently, older versions of Linux and BSD used EINVAL
650 instead of EAFNOSUPPORT and such. */
655 if (!opt.retry_connrefused)
656 if (err == ECONNREFUSED
658 || err == ENETUNREACH /* network is unreachable */
661 || err == EHOSTUNREACH /* host is unreachable */
669 /* Wait for a single descriptor to become available, timing out after
670 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
671 -1 for error. The argument WAIT_FOR can be a combination of
672 WAIT_FOR_READ and WAIT_FOR_WRITE.
674 This is a mere convenience wrapper around the select call, and
675 should be taken as such (for example, it doesn't implement Wget's
676 0-timeout-means-no-timeout semantics.) */
679 select_fd (int fd, double maxtime, int wait_for)
682 fd_set *rd = NULL, *wr = NULL;
683 struct timeval tmout;
688 if (wait_for & WAIT_FOR_READ)
690 if (wait_for & WAIT_FOR_WRITE)
693 tmout.tv_sec = (long) maxtime;
694 tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
698 result = select (fd + 1, rd, wr, NULL, &tmout);
700 /* gnulib select() converts blocking sockets to nonblocking in windows.
701 wget uses blocking sockets so we must convert them back to blocking. */
702 set_windows_fd_as_blocking_socket (fd);
705 while (result < 0 && errno == EINTR);
710 /* Return true iff the connection to the remote site established
711 through SOCK is still open.
713 Specifically, this function returns true if SOCK is not ready for
714 reading. This is because, when the connection closes, the socket
715 is ready for reading because EOF is about to be delivered. A side
716 effect of this method is that sockets that have pending data are
717 considered non-open. This is actually a good thing for callers of
718 this function, where such pending data can only be unwanted
719 leftover from a previous request. */
722 test_socket_open (int sock)
728 /* Check if we still have a valid (non-EOF) connection. From Andrew
729 * Maholski's code in the Unix Socket FAQ. */
731 FD_ZERO (&check_set);
732 FD_SET (sock, &check_set);
734 /* Wait one microsecond */
738 ret = select (sock + 1, &check_set, NULL, NULL, &to);
740 /* gnulib select() converts blocking sockets to nonblocking in windows.
741 wget uses blocking sockets so we must convert them back to blocking
743 set_windows_fd_as_blocking_socket ( sock );
747 /* We got a timeout, it means we're still connected. */
750 /* Read now would not wait, it means we have either pending data
755 /* Basic socket operations, mostly EINTR wrappers. */
758 sock_read (int fd, char *buf, int bufsize)
762 res = read (fd, buf, bufsize);
763 while (res == -1 && errno == EINTR);
768 sock_write (int fd, char *buf, int bufsize)
772 res = write (fd, buf, bufsize);
773 while (res == -1 && errno == EINTR);
778 sock_poll (int fd, double timeout, int wait_for)
780 return select_fd (fd, timeout, wait_for);
784 sock_peek (int fd, char *buf, int bufsize)
788 res = recv (fd, buf, bufsize, MSG_PEEK);
789 while (res == -1 && errno == EINTR);
797 DEBUGP (("Closed fd %d\n", fd));
803 /* Reading and writing from the network. We build around the socket
804 (file descriptor) API, but support "extended" operations for things
805 that are not mere file descriptors under the hood, such as SSL
808 That way the user code can call fd_read(fd, ...) and we'll run read
809 or SSL_read or whatever is necessary. */
811 static struct hash_table *transport_map;
812 static unsigned int transport_map_modified_tick;
814 struct transport_info {
815 struct transport_implementation *imp;
819 /* Register the transport layer operations that will be used when
820 reading, writing, and polling FD.
822 This should be used for transport layers like SSL that piggyback on
823 sockets. FD should otherwise be a real socket, on which you can
824 call getpeername, etc. */
827 fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
829 struct transport_info *info;
831 /* The file descriptor must be non-negative to be registered.
832 Negative values are ignored by fd_close(), and -1 cannot be used as
836 info = xnew (struct transport_info);
840 transport_map = hash_table_new (0, NULL, NULL);
841 hash_table_put (transport_map, (void *)(intptr_t) fd, info);
842 ++transport_map_modified_tick;
845 /* Return context of the transport registered with
846 fd_register_transport. This assumes fd_register_transport was
847 previously called on FD. */
850 fd_transport_context (int fd)
852 struct transport_info *info = hash_table_get (transport_map, (void *)(intptr_t) fd);
856 /* When fd_read/fd_write are called multiple times in a loop, they should
857 remember the INFO pointer instead of fetching it every time. It is
858 not enough to compare FD to LAST_FD because FD might have been
859 closed and reopened. modified_tick ensures that changes to
860 transport_map will not be unnoticed.
862 This is a macro because we want the static storage variables to be
865 #define LAZY_RETRIEVE_INFO(info) do { \
866 static struct transport_info *last_info; \
867 static int last_fd = -1; \
868 static unsigned int last_tick; \
869 if (!transport_map) \
871 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
875 info = hash_table_get (transport_map, (void *)(intptr_t) fd); \
878 last_tick = transport_map_modified_tick; \
883 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
886 timeout = opt.read_timeout;
890 if (info && info->imp->poller)
891 test = info->imp->poller (fd, timeout, wf, info->ctx);
893 test = sock_poll (fd, timeout, wf);
902 /* Read no more than BUFSIZE bytes of data from FD, storing them to
903 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
904 received after that many seconds. If TIMEOUT is -1, the value of
905 opt.timeout is used for TIMEOUT. */
908 fd_read (int fd, char *buf, int bufsize, double timeout)
910 struct transport_info *info;
911 LAZY_RETRIEVE_INFO (info);
912 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
914 if (info && info->imp->reader)
915 return info->imp->reader (fd, buf, bufsize, info->ctx);
917 return sock_read (fd, buf, bufsize);
920 /* Like fd_read, except it provides a "preview" of the data that will
921 be read by subsequent calls to fd_read. Specifically, it copies no
922 more than BUFSIZE bytes of the currently available data to BUF and
923 returns the number of bytes copied. Return values and timeout
924 semantics are the same as those of fd_read.
926 CAVEAT: Do not assume that the first subsequent call to fd_read
927 will retrieve the same amount of data. Reading can return more or
928 less data, depending on the TCP implementation and other
929 circumstances. However, barring an error, it can be expected that
930 all the peeked data will eventually be read by fd_read. */
933 fd_peek (int fd, char *buf, int bufsize, double timeout)
935 struct transport_info *info;
936 LAZY_RETRIEVE_INFO (info);
937 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
939 if (info && info->imp->peeker)
940 return info->imp->peeker (fd, buf, bufsize, info->ctx);
942 return sock_peek (fd, buf, bufsize);
945 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
946 the operation aborts if no data is received after that many
947 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
951 fd_write (int fd, char *buf, int bufsize, double timeout)
954 struct transport_info *info;
955 LAZY_RETRIEVE_INFO (info);
957 /* `write' may write less than LEN bytes, thus the loop keeps trying
958 it until all was written, or an error occurred. */
962 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
964 if (info && info->imp->writer)
965 res = info->imp->writer (fd, buf, bufsize, info->ctx);
967 res = sock_write (fd, buf, bufsize);
976 /* Report the most recent error(s) on FD. This should only be called
977 after fd_* functions, such as fd_read and fd_write, and only if
978 they return a negative result. For errors coming from other calls
979 such as setsockopt or fopen, strerror should continue to be
982 If the transport doesn't support error messages or doesn't supply
983 one, strerror(errno) is returned. The returned error message
984 should not be used after fd_close has been called. */
989 /* Don't bother with LAZY_RETRIEVE_INFO, as this will only be called
990 in case of error, never in a tight loop. */
991 struct transport_info *info = NULL;
993 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
995 if (info && info->imp->errstr)
997 const char *err = info->imp->errstr (fd, info->ctx);
1000 /* else, fall through and print the system error. */
1002 return strerror (errno);
1005 /* Close the file descriptor FD. */
1010 struct transport_info *info;
1014 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
1015 per socket, so that particular optimization wouldn't work. */
1018 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
1020 if (info && info->imp->closer)
1021 info->imp->closer (fd, info->ctx);
1027 hash_table_remove (transport_map, (void *)(intptr_t) fd);
1029 ++transport_map_modified_tick;