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 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget. If not, see <http://www.gnu.org/licenses/>.
20 Additional permission under GNU GPL version 3 section 7
22 If you modify this program, or any covered work, by linking or
23 combining it with the OpenSSL project's OpenSSL library (or a
24 modified version of that library), containing parts covered by the
25 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
26 grants you additional permission to convey the resulting work.
27 Corresponding Source for a non-source form of such a combination
28 shall include the source code for the parts of OpenSSL used as well
29 as that of the covered work. */
38 #include <sys/socket.h>
39 #include <sys/select.h>
44 # else /* def __VMS */
46 # endif /* def __VMS [else] */
47 # include <netinet/in.h>
49 # include <arpa/inet.h>
51 #endif /* not WINDOWS */
55 #ifdef HAVE_SYS_TIME_H
56 # include <sys/time.h>
63 /* Apparently needed for Interix: */
68 /* Define sockaddr_storage where unavailable (presumably on IPv4-only
72 # ifndef HAVE_STRUCT_SOCKADDR_STORAGE
73 # define sockaddr_storage sockaddr_in
75 #endif /* ENABLE_IPV6 */
77 /* Fill SA as per the data in IP and PORT. SA shoult point to struct
78 sockaddr_storage if ENABLE_IPV6 is defined, to struct sockaddr_in
82 sockaddr_set_data (struct sockaddr *sa, const ip_address *ip, int port)
88 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
90 sin->sin_family = AF_INET;
91 sin->sin_port = htons (port);
92 sin->sin_addr = ip->data.d4;
98 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
100 sin6->sin6_family = AF_INET6;
101 sin6->sin6_port = htons (port);
102 sin6->sin6_addr = ip->data.d6;
103 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
104 sin6->sin6_scope_id = ip->ipv6_scope;
108 #endif /* ENABLE_IPV6 */
114 /* Get the data of SA, specifically the IP address and the port. If
115 you're not interested in one or the other information, pass NULL as
119 sockaddr_get_data (const struct sockaddr *sa, ip_address *ip, int *port)
121 switch (sa->sa_family)
125 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
128 ip->family = AF_INET;
129 ip->data.d4 = sin->sin_addr;
132 *port = ntohs (sin->sin_port);
138 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
141 ip->family = AF_INET6;
142 ip->data.d6 = sin6->sin6_addr;
143 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
144 ip->ipv6_scope = sin6->sin6_scope_id;
148 *port = ntohs (sin6->sin6_port);
157 /* Return the size of the sockaddr structure depending on its
161 sockaddr_size (const struct sockaddr *sa)
163 switch (sa->sa_family)
166 return sizeof (struct sockaddr_in);
169 return sizeof (struct sockaddr_in6);
176 /* Resolve the bind address specified via --bind-address and store it
177 to SA. The resolved value is stored in a static variable and
178 reused after the first invocation of this function.
180 Returns true on success, false on failure. */
183 resolve_bind_address (struct sockaddr *sa)
185 struct address_list *al;
187 /* Make sure this is called only once. opt.bind_address doesn't
188 change during a Wget run. */
189 static bool called, should_bind;
190 static ip_address ip;
194 sockaddr_set_data (sa, &ip, 0);
199 al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
202 /* #### We should be able to print the error message here. */
203 logprintf (LOG_NOTQUIET,
204 _("%s: unable to resolve bind address %s; disabling bind.\n"),
205 exec_name, quote (opt.bind_address));
210 /* Pick the first address in the list and use it as bind address.
211 Perhaps we should try multiple addresses in succession, but I
212 don't think that's necessary in practice. */
213 ip = *address_list_address_at (al, 0);
214 address_list_release (al);
216 sockaddr_set_data (sa, &ip, 0);
223 const struct sockaddr *addr;
229 connect_with_timeout_callback (void *arg)
231 struct cwt_context *ctx = (struct cwt_context *)arg;
232 ctx->result = connect (ctx->fd, ctx->addr, ctx->addrlen);
235 /* Like connect, but specifies a timeout. If connecting takes longer
236 than TIMEOUT seconds, -1 is returned and errno is set to
240 connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
243 struct cwt_context ctx;
246 ctx.addrlen = addrlen;
248 if (run_with_timeout (timeout, connect_with_timeout_callback, &ctx))
253 if (ctx.result == -1 && errno == EINTR)
258 /* Connect via TCP to the specified address and port.
260 If PRINT is non-NULL, it is the host name to print that we're
264 connect_to_ip (const ip_address *ip, int port, const char *print)
266 struct sockaddr_storage ss;
267 struct sockaddr *sa = (struct sockaddr *)&ss;
270 /* If PRINT is non-NULL, print the "Connecting to..." line, with
271 PRINT being the host name we're connecting to. */
274 const char *txt_addr = print_address (ip);
275 if (0 != strcmp (print, txt_addr))
277 char *str = NULL, *name;
279 if (opt.enable_iri && (name = idn_decode ((char *) print)) != NULL)
281 int len = strlen (print) + strlen (name) + 4;
283 snprintf (str, len, "%s (%s)", name, print);
288 logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
289 str ? str : escnonprint_uri (print), txt_addr, port);
295 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
298 /* Store the sockaddr info to SA. */
299 sockaddr_set_data (sa, ip, port);
301 /* Create the socket of the family appropriate for the address. */
302 sock = socket (sa->sa_family, SOCK_STREAM, 0);
306 #if defined(ENABLE_IPV6) && defined(IPV6_V6ONLY)
309 /* In case of error, we will go on anyway... */
310 int err = setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof (on));
313 DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
317 /* For very small rate limits, set the buffer size (and hence,
318 hopefully, the kernel's TCP window size) to the per-second limit.
319 That way we should never have to sleep for more than 1s between
321 if (opt.limit_rate && opt.limit_rate < 8192)
323 int bufsize = opt.limit_rate;
325 bufsize = 512; /* avoid pathologically small values */
327 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
328 (void *)&bufsize, (socklen_t)sizeof (bufsize));
330 /* When we add limit_rate support for writing, which is useful
331 for POST, we should also set SO_SNDBUF here. */
334 if (opt.bind_address)
336 /* Bind the client side of the socket to the requested
338 struct sockaddr_storage bind_ss;
339 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
340 if (resolve_bind_address (bind_sa))
342 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
347 /* Connect the socket to the remote endpoint. */
348 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
349 opt.connect_timeout) < 0)
355 logprintf (LOG_VERBOSE, _("connected.\n"));
356 DEBUGP (("Created socket %d.\n", sock));
361 /* Protect errno from possible modifications by close and
363 int save_errno = errno;
367 logprintf (LOG_VERBOSE, _("failed: %s.\n"), strerror (errno));
373 /* Connect via TCP to a remote host on the specified port.
375 HOST is resolved as an Internet host name. If HOST resolves to
376 more than one IP address, they are tried in the order returned by
377 DNS until connecting to one of them succeeds. */
380 connect_to_host (const char *host, int port)
385 struct address_list *al = lookup_host (host, 0);
390 logprintf (LOG_NOTQUIET,
391 _("%s: unable to resolve host address %s\n"),
392 exec_name, quote (host));
396 address_list_get_bounds (al, &start, &end);
397 for (i = start; i < end; i++)
399 const ip_address *ip = address_list_address_at (al, i);
400 sock = connect_to_ip (ip, port, host);
404 address_list_set_connected (al);
405 address_list_release (al);
409 /* The attempt to connect has failed. Continue with the loop
410 and try next address. */
412 address_list_set_faulty (al, i);
415 /* Failed to connect to any of the addresses in AL. */
417 if (address_list_connected_p (al))
419 /* We connected to AL before, but cannot do so now. That might
420 indicate that our DNS cache entry for HOST has expired. */
421 address_list_release (al);
422 al = lookup_host (host, LH_REFRESH);
425 address_list_release (al);
430 /* Create a socket, bind it to local interface BIND_ADDRESS on port
431 *PORT, set up a listen backlog, and return the resulting socket, or
434 BIND_ADDRESS is the address of the interface to bind to. If it is
435 NULL, the socket is bound to the default address. PORT should
436 point to the port number that will be used for the binding. If
437 that number is 0, the system will choose a suitable port, and the
438 chosen value will be written to *PORT.
440 Calling accept() on such a socket waits for and accepts incoming
444 bind_local (const ip_address *bind_address, int *port)
447 struct sockaddr_storage ss;
448 struct sockaddr *sa = (struct sockaddr *)&ss;
450 /* For setting options with setsockopt. */
452 void *setopt_ptr = (void *)&setopt_val;
453 socklen_t setopt_size = sizeof (setopt_val);
455 sock = socket (bind_address->family, SOCK_STREAM, 0);
460 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
464 sockaddr_set_data (sa, bind_address, *port);
465 if (bind (sock, sa, sockaddr_size (sa)) < 0)
470 DEBUGP (("Local socket fd %d bound.\n", sock));
472 /* If *PORT is 0, find out which port we've bound to. */
475 socklen_t addrlen = sockaddr_size (sa);
476 if (getsockname (sock, sa, &addrlen) < 0)
478 /* If we can't find out the socket's local address ("name"),
479 something is seriously wrong with the socket, and it's
480 unusable for us anyway because we must know the chosen
485 sockaddr_get_data (sa, NULL, port);
486 DEBUGP (("binding to address %s using port %i.\n",
487 print_address (bind_address), *port));
489 if (listen (sock, 1) < 0)
497 /* Like a call to accept(), but with the added check for timeout.
499 In other words, accept a client connection on LOCAL_SOCK, and
500 return the new socket used for communication with the client.
501 LOCAL_SOCK should have been bound, e.g. using bind_local().
503 The caller is blocked until a connection is established. If no
504 connection is established for opt.connect_timeout seconds, the
505 function exits with an error status. */
508 accept_connection (int local_sock)
512 /* We don't need the values provided by accept, but accept
513 apparently requires them to be present. */
514 struct sockaddr_storage ss;
515 struct sockaddr *sa = (struct sockaddr *)&ss;
516 socklen_t addrlen = sizeof (ss);
518 if (opt.connect_timeout)
520 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
526 sock = accept (local_sock, sa, &addrlen);
527 DEBUGP (("Accepted client at socket %d.\n", sock));
531 /* Get the IP address associated with the connection on FD and store
532 it to IP. Return true on success, false otherwise.
534 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
535 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
536 returns the address of the remote (peer's) side of the socket. */
539 socket_ip_address (int sock, ip_address *ip, int endpoint)
541 struct sockaddr_storage storage;
542 struct sockaddr *sockaddr = (struct sockaddr *) &storage;
543 socklen_t addrlen = sizeof (storage);
546 memset (sockaddr, 0, addrlen);
547 if (endpoint == ENDPOINT_LOCAL)
548 ret = getsockname (sock, sockaddr, &addrlen);
549 else if (endpoint == ENDPOINT_PEER)
550 ret = getpeername (sock, sockaddr, &addrlen);
556 ip->family = sockaddr->sa_family;
557 switch (sockaddr->sa_family)
562 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
563 ip->data.d6 = sa6->sin6_addr;
564 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
565 ip->ipv6_scope = sa6->sin6_scope_id;
567 DEBUGP (("conaddr is: %s\n", print_address (ip)));
573 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
574 ip->data.d4 = sa->sin_addr;
575 DEBUGP (("conaddr is: %s\n", print_address (ip)));
583 /* Return true if the error from the connect code can be considered
584 retryable. Wget normally retries after errors, but the exception
585 are the "unsupported protocol" type errors (possible on IPv4/IPv6
586 dual family systems) and "connection refused". */
589 retryable_socket_connect_error (int err)
591 /* Have to guard against some of these values not being defined.
592 Cannot use a switch statement because some of the values might be
596 || err == EAFNOSUPPORT
599 || err == EPFNOSUPPORT
601 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
602 || err == ESOCKTNOSUPPORT
604 #ifdef EPROTONOSUPPORT
605 || err == EPROTONOSUPPORT
608 || err == ENOPROTOOPT
610 /* Apparently, older versions of Linux and BSD used EINVAL
611 instead of EAFNOSUPPORT and such. */
616 if (!opt.retry_connrefused)
617 if (err == ECONNREFUSED
619 || err == ENETUNREACH /* network is unreachable */
622 || err == EHOSTUNREACH /* host is unreachable */
630 /* Wait for a single descriptor to become available, timing out after
631 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
632 -1 for error. The argument WAIT_FOR can be a combination of
633 WAIT_FOR_READ and WAIT_FOR_WRITE.
635 This is a mere convenience wrapper around the select call, and
636 should be taken as such (for example, it doesn't implement Wget's
637 0-timeout-means-no-timeout semantics.) */
640 select_fd (int fd, double maxtime, int wait_for)
643 fd_set *rd = NULL, *wr = NULL;
644 struct timeval tmout;
649 if (wait_for & WAIT_FOR_READ)
651 if (wait_for & WAIT_FOR_WRITE)
654 tmout.tv_sec = (long) maxtime;
655 tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
658 result = select (fd + 1, rd, wr, NULL, &tmout);
659 while (result < 0 && errno == EINTR);
664 /* Return true iff the connection to the remote site established
665 through SOCK is still open.
667 Specifically, this function returns true if SOCK is not ready for
668 reading. This is because, when the connection closes, the socket
669 is ready for reading because EOF is about to be delivered. A side
670 effect of this method is that sockets that have pending data are
671 considered non-open. This is actually a good thing for callers of
672 this function, where such pending data can only be unwanted
673 leftover from a previous request. */
676 test_socket_open (int sock)
681 /* Check if we still have a valid (non-EOF) connection. From Andrew
682 * Maholski's code in the Unix Socket FAQ. */
684 FD_ZERO (&check_set);
685 FD_SET (sock, &check_set);
687 /* Wait one microsecond */
691 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
692 /* We got a timeout, it means we're still connected. */
695 /* Read now would not wait, it means we have either pending data
700 /* Basic socket operations, mostly EINTR wrappers. */
703 sock_read (int fd, char *buf, int bufsize)
707 res = read (fd, buf, bufsize);
708 while (res == -1 && errno == EINTR);
713 sock_write (int fd, char *buf, int bufsize)
717 res = write (fd, buf, bufsize);
718 while (res == -1 && errno == EINTR);
723 sock_poll (int fd, double timeout, int wait_for)
725 return select_fd (fd, timeout, wait_for);
729 sock_peek (int fd, char *buf, int bufsize)
733 res = recv (fd, buf, bufsize, MSG_PEEK);
734 while (res == -1 && errno == EINTR);
742 DEBUGP (("Closed fd %d\n", fd));
748 /* Reading and writing from the network. We build around the socket
749 (file descriptor) API, but support "extended" operations for things
750 that are not mere file descriptors under the hood, such as SSL
753 That way the user code can call fd_read(fd, ...) and we'll run read
754 or SSL_read or whatever is necessary. */
756 static struct hash_table *transport_map;
757 static unsigned int transport_map_modified_tick;
759 struct transport_info {
760 struct transport_implementation *imp;
764 /* Register the transport layer operations that will be used when
765 reading, writing, and polling FD.
767 This should be used for transport layers like SSL that piggyback on
768 sockets. FD should otherwise be a real socket, on which you can
769 call getpeername, etc. */
772 fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
774 struct transport_info *info;
776 /* The file descriptor must be non-negative to be registered.
777 Negative values are ignored by fd_close(), and -1 cannot be used as
781 info = xnew (struct transport_info);
785 transport_map = hash_table_new (0, NULL, NULL);
786 hash_table_put (transport_map, (void *)(intptr_t) fd, info);
787 ++transport_map_modified_tick;
790 /* Return context of the transport registered with
791 fd_register_transport. This assumes fd_register_transport was
792 previously called on FD. */
795 fd_transport_context (int fd)
797 struct transport_info *info = hash_table_get (transport_map, (void *)(intptr_t) fd);
801 /* When fd_read/fd_write are called multiple times in a loop, they should
802 remember the INFO pointer instead of fetching it every time. It is
803 not enough to compare FD to LAST_FD because FD might have been
804 closed and reopened. modified_tick ensures that changes to
805 transport_map will not be unnoticed.
807 This is a macro because we want the static storage variables to be
810 #define LAZY_RETRIEVE_INFO(info) do { \
811 static struct transport_info *last_info; \
812 static int last_fd = -1; \
813 static unsigned int last_tick; \
814 if (!transport_map) \
816 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
820 info = hash_table_get (transport_map, (void *)(intptr_t) fd); \
823 last_tick = transport_map_modified_tick; \
828 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
831 timeout = opt.read_timeout;
835 if (info && info->imp->poller)
836 test = info->imp->poller (fd, timeout, wf, info->ctx);
838 test = sock_poll (fd, timeout, wf);
847 /* Read no more than BUFSIZE bytes of data from FD, storing them to
848 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
849 received after that many seconds. If TIMEOUT is -1, the value of
850 opt.timeout is used for TIMEOUT. */
853 fd_read (int fd, char *buf, int bufsize, double timeout)
855 struct transport_info *info;
856 LAZY_RETRIEVE_INFO (info);
857 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
859 if (info && info->imp->reader)
860 return info->imp->reader (fd, buf, bufsize, info->ctx);
862 return sock_read (fd, buf, bufsize);
865 /* Like fd_read, except it provides a "preview" of the data that will
866 be read by subsequent calls to fd_read. Specifically, it copies no
867 more than BUFSIZE bytes of the currently available data to BUF and
868 returns the number of bytes copied. Return values and timeout
869 semantics are the same as those of fd_read.
871 CAVEAT: Do not assume that the first subsequent call to fd_read
872 will retrieve the same amount of data. Reading can return more or
873 less data, depending on the TCP implementation and other
874 circumstances. However, barring an error, it can be expected that
875 all the peeked data will eventually be read by fd_read. */
878 fd_peek (int fd, char *buf, int bufsize, double timeout)
880 struct transport_info *info;
881 LAZY_RETRIEVE_INFO (info);
882 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
884 if (info && info->imp->peeker)
885 return info->imp->peeker (fd, buf, bufsize, info->ctx);
887 return sock_peek (fd, buf, bufsize);
890 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
891 the operation aborts if no data is received after that many
892 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
896 fd_write (int fd, char *buf, int bufsize, double timeout)
899 struct transport_info *info;
900 LAZY_RETRIEVE_INFO (info);
902 /* `write' may write less than LEN bytes, thus the loop keeps trying
903 it until all was written, or an error occurred. */
907 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
909 if (info && info->imp->writer)
910 res = info->imp->writer (fd, buf, bufsize, info->ctx);
912 res = sock_write (fd, buf, bufsize);
921 /* Report the most recent error(s) on FD. This should only be called
922 after fd_* functions, such as fd_read and fd_write, and only if
923 they return a negative result. For errors coming from other calls
924 such as setsockopt or fopen, strerror should continue to be
927 If the transport doesn't support error messages or doesn't supply
928 one, strerror(errno) is returned. The returned error message
929 should not be used after fd_close has been called. */
934 /* Don't bother with LAZY_RETRIEVE_INFO, as this will only be called
935 in case of error, never in a tight loop. */
936 struct transport_info *info = NULL;
938 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
940 if (info && info->imp->errstr)
942 const char *err = info->imp->errstr (fd, info->ctx);
945 /* else, fall through and print the system error. */
947 return strerror (errno);
950 /* Close the file descriptor FD. */
955 struct transport_info *info;
959 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
960 per socket, so that particular optimization wouldn't work. */
963 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
965 if (info && info->imp->closer)
966 info->imp->closer (fd, info->ctx);
972 hash_table_remove (transport_map, (void *)(intptr_t) fd);
974 ++transport_map_modified_tick;