1 /* Establishing and handling network connections.
2 Copyright (C) 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007 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 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
40 # include <sys/socket.h>
42 # include <netinet/in.h>
44 # include <arpa/inet.h>
46 #endif /* not WINDOWS */
50 #ifdef HAVE_SYS_SELECT_H
51 # include <sys/select.h>
52 #endif /* HAVE_SYS_SELECT_H */
58 /* Define sockaddr_storage where unavailable (presumably on IPv4-only
62 # ifndef HAVE_STRUCT_SOCKADDR_STORAGE
63 # define sockaddr_storage sockaddr_in
65 #endif /* ENABLE_IPV6 */
67 /* Fill SA as per the data in IP and PORT. SA shoult point to struct
68 sockaddr_storage if ENABLE_IPV6 is defined, to struct sockaddr_in
72 sockaddr_set_data (struct sockaddr *sa, const ip_address *ip, int port)
78 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
80 sin->sin_family = AF_INET;
81 sin->sin_port = htons (port);
82 sin->sin_addr = ip->data.d4;
88 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
90 sin6->sin6_family = AF_INET6;
91 sin6->sin6_port = htons (port);
92 sin6->sin6_addr = ip->data.d6;
93 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
94 sin6->sin6_scope_id = ip->ipv6_scope;
98 #endif /* ENABLE_IPV6 */
104 /* Get the data of SA, specifically the IP address and the port. If
105 you're not interested in one or the other information, pass NULL as
109 sockaddr_get_data (const struct sockaddr *sa, ip_address *ip, int *port)
111 switch (sa->sa_family)
115 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
118 ip->family = AF_INET;
119 ip->data.d4 = sin->sin_addr;
122 *port = ntohs (sin->sin_port);
128 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
131 ip->family = AF_INET6;
132 ip->data.d6 = sin6->sin6_addr;
133 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
134 ip->ipv6_scope = sin6->sin6_scope_id;
138 *port = ntohs (sin6->sin6_port);
147 /* Return the size of the sockaddr structure depending on its
151 sockaddr_size (const struct sockaddr *sa)
153 switch (sa->sa_family)
156 return sizeof (struct sockaddr_in);
159 return sizeof (struct sockaddr_in6);
166 /* Resolve the bind address specified via --bind-address and store it
167 to SA. The resolved value is stored in a static variable and
168 reused after the first invocation of this function.
170 Returns true on success, false on failure. */
173 resolve_bind_address (struct sockaddr *sa)
175 struct address_list *al;
177 /* Make sure this is called only once. opt.bind_address doesn't
178 change during a Wget run. */
179 static bool called, should_bind;
180 static ip_address ip;
184 sockaddr_set_data (sa, &ip, 0);
189 al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
192 /* #### We should be able to print the error message here. */
193 logprintf (LOG_NOTQUIET,
194 _("%s: unable to resolve bind address `%s'; disabling bind.\n"),
195 exec_name, opt.bind_address);
200 /* Pick the first address in the list and use it as bind address.
201 Perhaps we should try multiple addresses in succession, but I
202 don't think that's necessary in practice. */
203 ip = *address_list_address_at (al, 0);
204 address_list_release (al);
206 sockaddr_set_data (sa, &ip, 0);
213 const struct sockaddr *addr;
219 connect_with_timeout_callback (void *arg)
221 struct cwt_context *ctx = (struct cwt_context *)arg;
222 ctx->result = connect (ctx->fd, ctx->addr, ctx->addrlen);
225 /* Like connect, but specifies a timeout. If connecting takes longer
226 than TIMEOUT seconds, -1 is returned and errno is set to
230 connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
233 struct cwt_context ctx;
236 ctx.addrlen = addrlen;
238 if (run_with_timeout (timeout, connect_with_timeout_callback, &ctx))
243 if (ctx.result == -1 && errno == EINTR)
248 /* Connect via TCP to the specified address and port.
250 If PRINT is non-NULL, it is the host name to print that we're
254 connect_to_ip (const ip_address *ip, int port, const char *print)
256 struct sockaddr_storage ss;
257 struct sockaddr *sa = (struct sockaddr *)&ss;
260 /* If PRINT is non-NULL, print the "Connecting to..." line, with
261 PRINT being the host name we're connecting to. */
264 const char *txt_addr = print_address (ip);
265 if (print && 0 != strcmp (print, txt_addr))
266 logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
267 escnonprint (print), txt_addr, port);
269 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
272 /* Store the sockaddr info to SA. */
273 sockaddr_set_data (sa, ip, port);
275 /* Create the socket of the family appropriate for the address. */
276 sock = socket (sa->sa_family, SOCK_STREAM, 0);
280 #if defined(ENABLE_IPV6) && defined(IPV6_V6ONLY)
283 /* In case of error, we will go on anyway... */
284 int err = setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof (on));
287 DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
291 /* For very small rate limits, set the buffer size (and hence,
292 hopefully, the kernel's TCP window size) to the per-second limit.
293 That way we should never have to sleep for more than 1s between
295 if (opt.limit_rate && opt.limit_rate < 8192)
297 int bufsize = opt.limit_rate;
299 bufsize = 512; /* avoid pathologically small values */
301 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
302 (void *)&bufsize, (socklen_t)sizeof (bufsize));
304 /* When we add limit_rate support for writing, which is useful
305 for POST, we should also set SO_SNDBUF here. */
308 if (opt.bind_address)
310 /* Bind the client side of the socket to the requested
312 struct sockaddr_storage bind_ss;
313 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
314 if (resolve_bind_address (bind_sa))
316 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
321 /* Connect the socket to the remote endpoint. */
322 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
323 opt.connect_timeout) < 0)
329 logprintf (LOG_VERBOSE, _("connected.\n"));
330 DEBUGP (("Created socket %d.\n", sock));
335 /* Protect errno from possible modifications by close and
337 int save_errno = errno;
341 logprintf (LOG_VERBOSE, _("failed: %s.\n"), strerror (errno));
347 /* Connect via TCP to a remote host on the specified port.
349 HOST is resolved as an Internet host name. If HOST resolves to
350 more than one IP address, they are tried in the order returned by
351 DNS until connecting to one of them succeeds. */
354 connect_to_host (const char *host, int port)
359 struct address_list *al = lookup_host (host, 0);
364 logprintf (LOG_NOTQUIET,
365 _("%s: unable to resolve host address `%s'\n"),
370 address_list_get_bounds (al, &start, &end);
371 for (i = start; i < end; i++)
373 const ip_address *ip = address_list_address_at (al, i);
374 sock = connect_to_ip (ip, port, host);
378 address_list_set_connected (al);
379 address_list_release (al);
383 /* The attempt to connect has failed. Continue with the loop
384 and try next address. */
386 address_list_set_faulty (al, i);
389 /* Failed to connect to any of the addresses in AL. */
391 if (address_list_connected_p (al))
393 /* We connected to AL before, but cannot do so now. That might
394 indicate that our DNS cache entry for HOST has expired. */
395 address_list_release (al);
396 al = lookup_host (host, LH_REFRESH);
399 address_list_release (al);
404 /* Create a socket, bind it to local interface BIND_ADDRESS on port
405 *PORT, set up a listen backlog, and return the resulting socket, or
408 BIND_ADDRESS is the address of the interface to bind to. If it is
409 NULL, the socket is bound to the default address. PORT should
410 point to the port number that will be used for the binding. If
411 that number is 0, the system will choose a suitable port, and the
412 chosen value will be written to *PORT.
414 Calling accept() on such a socket waits for and accepts incoming
418 bind_local (const ip_address *bind_address, int *port)
421 struct sockaddr_storage ss;
422 struct sockaddr *sa = (struct sockaddr *)&ss;
424 /* For setting options with setsockopt. */
426 void *setopt_ptr = (void *)&setopt_val;
427 socklen_t setopt_size = sizeof (setopt_val);
429 sock = socket (bind_address->family, SOCK_STREAM, 0);
434 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
438 sockaddr_set_data (sa, bind_address, *port);
439 if (bind (sock, sa, sockaddr_size (sa)) < 0)
444 DEBUGP (("Local socket fd %d bound.\n", sock));
446 /* If *PORT is 0, find out which port we've bound to. */
449 socklen_t addrlen = sockaddr_size (sa);
450 if (getsockname (sock, sa, &addrlen) < 0)
452 /* If we can't find out the socket's local address ("name"),
453 something is seriously wrong with the socket, and it's
454 unusable for us anyway because we must know the chosen
459 sockaddr_get_data (sa, NULL, port);
460 DEBUGP (("binding to address %s using port %i.\n",
461 print_address (bind_address), *port));
463 if (listen (sock, 1) < 0)
471 /* Like a call to accept(), but with the added check for timeout.
473 In other words, accept a client connection on LOCAL_SOCK, and
474 return the new socket used for communication with the client.
475 LOCAL_SOCK should have been bound, e.g. using bind_local().
477 The caller is blocked until a connection is established. If no
478 connection is established for opt.connect_timeout seconds, the
479 function exits with an error status. */
482 accept_connection (int local_sock)
486 /* We don't need the values provided by accept, but accept
487 apparently requires them to be present. */
488 struct sockaddr_storage ss;
489 struct sockaddr *sa = (struct sockaddr *)&ss;
490 socklen_t addrlen = sizeof (ss);
492 if (opt.connect_timeout)
494 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
500 sock = accept (local_sock, sa, &addrlen);
501 DEBUGP (("Accepted client at socket %d.\n", sock));
505 /* Get the IP address associated with the connection on FD and store
506 it to IP. Return true on success, false otherwise.
508 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
509 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
510 returns the address of the remote (peer's) side of the socket. */
513 socket_ip_address (int sock, ip_address *ip, int endpoint)
515 struct sockaddr_storage storage;
516 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
517 socklen_t addrlen = sizeof (storage);
520 if (endpoint == ENDPOINT_LOCAL)
521 ret = getsockname (sock, sockaddr, &addrlen);
522 else if (endpoint == ENDPOINT_PEER)
523 ret = getpeername (sock, sockaddr, &addrlen);
529 ip->family = sockaddr->sa_family;
530 switch (sockaddr->sa_family)
535 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
536 ip->data.d6 = sa6->sin6_addr;
537 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
538 ip->ipv6_scope = sa6->sin6_scope_id;
540 DEBUGP (("conaddr is: %s\n", print_address (ip)));
546 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
547 ip->data.d4 = sa->sin_addr;
548 DEBUGP (("conaddr is: %s\n", print_address (ip)));
556 /* Return true if the error from the connect code can be considered
557 retryable. Wget normally retries after errors, but the exception
558 are the "unsupported protocol" type errors (possible on IPv4/IPv6
559 dual family systems) and "connection refused". */
562 retryable_socket_connect_error (int err)
564 /* Have to guard against some of these values not being defined.
565 Cannot use a switch statement because some of the values might be
569 || err == EAFNOSUPPORT
572 || err == EPFNOSUPPORT
574 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
575 || err == ESOCKTNOSUPPORT
577 #ifdef EPROTONOSUPPORT
578 || err == EPROTONOSUPPORT
581 || err == ENOPROTOOPT
583 /* Apparently, older versions of Linux and BSD used EINVAL
584 instead of EAFNOSUPPORT and such. */
589 if (!opt.retry_connrefused)
590 if (err == ECONNREFUSED
592 || err == ENETUNREACH /* network is unreachable */
595 || err == EHOSTUNREACH /* host is unreachable */
603 /* Wait for a single descriptor to become available, timing out after
604 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
605 -1 for error. The argument WAIT_FOR can be a combination of
606 WAIT_FOR_READ and WAIT_FOR_WRITE.
608 This is a mere convenience wrapper around the select call, and
609 should be taken as such (for example, it doesn't implement Wget's
610 0-timeout-means-no-timeout semantics.) */
613 select_fd (int fd, double maxtime, int wait_for)
616 fd_set *rd = NULL, *wr = NULL;
617 struct timeval tmout;
622 if (wait_for & WAIT_FOR_READ)
624 if (wait_for & WAIT_FOR_WRITE)
627 tmout.tv_sec = (long) maxtime;
628 tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
631 result = select (fd + 1, rd, wr, NULL, &tmout);
632 while (result < 0 && errno == EINTR);
637 /* Return true iff the connection to the remote site established
638 through SOCK is still open.
640 Specifically, this function returns true if SOCK is not ready for
641 reading. This is because, when the connection closes, the socket
642 is ready for reading because EOF is about to be delivered. A side
643 effect of this method is that sockets that have pending data are
644 considered non-open. This is actually a good thing for callers of
645 this function, where such pending data can only be unwanted
646 leftover from a previous request. */
649 test_socket_open (int sock)
654 /* Check if we still have a valid (non-EOF) connection. From Andrew
655 * Maholski's code in the Unix Socket FAQ. */
657 FD_ZERO (&check_set);
658 FD_SET (sock, &check_set);
660 /* Wait one microsecond */
664 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
665 /* We got a timeout, it means we're still connected. */
668 /* Read now would not wait, it means we have either pending data
673 /* Basic socket operations, mostly EINTR wrappers. */
675 #if defined(WINDOWS) || defined(MSDOS)
676 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
677 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
678 # define close(fd) closesocket (fd)
682 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
683 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
687 sock_read (int fd, char *buf, int bufsize)
691 res = read (fd, buf, bufsize);
692 while (res == -1 && errno == EINTR);
697 sock_write (int fd, char *buf, int bufsize)
701 res = write (fd, buf, bufsize);
702 while (res == -1 && errno == EINTR);
707 sock_poll (int fd, double timeout, int wait_for)
709 return select_fd (fd, timeout, wait_for);
713 sock_peek (int fd, char *buf, int bufsize)
717 res = recv (fd, buf, bufsize, MSG_PEEK);
718 while (res == -1 && errno == EINTR);
726 DEBUGP (("Closed fd %d\n", fd));
732 /* Reading and writing from the network. We build around the socket
733 (file descriptor) API, but support "extended" operations for things
734 that are not mere file descriptors under the hood, such as SSL
737 That way the user code can call fd_read(fd, ...) and we'll run read
738 or SSL_read or whatever is necessary. */
740 static struct hash_table *transport_map;
741 static unsigned int transport_map_modified_tick;
743 struct transport_info {
744 struct transport_implementation *imp;
748 /* Register the transport layer operations that will be used when
749 reading, writing, and polling FD.
751 This should be used for transport layers like SSL that piggyback on
752 sockets. FD should otherwise be a real socket, on which you can
753 call getpeername, etc. */
756 fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
758 struct transport_info *info;
760 /* The file descriptor must be non-negative to be registered.
761 Negative values are ignored by fd_close(), and -1 cannot be used as
765 info = xnew (struct transport_info);
769 transport_map = hash_table_new (0, NULL, NULL);
770 hash_table_put (transport_map, (void *)(intptr_t) fd, info);
771 ++transport_map_modified_tick;
774 /* Return context of the transport registered with
775 fd_register_transport. This assumes fd_register_transport was
776 previously called on FD. */
779 fd_transport_context (int fd)
781 struct transport_info *info = hash_table_get (transport_map, (void *)(intptr_t) fd);
785 /* When fd_read/fd_write are called multiple times in a loop, they should
786 remember the INFO pointer instead of fetching it every time. It is
787 not enough to compare FD to LAST_FD because FD might have been
788 closed and reopened. modified_tick ensures that changes to
789 transport_map will not be unnoticed.
791 This is a macro because we want the static storage variables to be
794 #define LAZY_RETRIEVE_INFO(info) do { \
795 static struct transport_info *last_info; \
796 static int last_fd = -1; \
797 static unsigned int last_tick; \
798 if (!transport_map) \
800 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
804 info = hash_table_get (transport_map, (void *)(intptr_t) fd); \
807 last_tick = transport_map_modified_tick; \
812 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
815 timeout = opt.read_timeout;
819 if (info && info->imp->poller)
820 test = info->imp->poller (fd, timeout, wf, info->ctx);
822 test = sock_poll (fd, timeout, wf);
831 /* Read no more than BUFSIZE bytes of data from FD, storing them to
832 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
833 received after that many seconds. If TIMEOUT is -1, the value of
834 opt.timeout is used for TIMEOUT. */
837 fd_read (int fd, char *buf, int bufsize, double timeout)
839 struct transport_info *info;
840 LAZY_RETRIEVE_INFO (info);
841 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
843 if (info && info->imp->reader)
844 return info->imp->reader (fd, buf, bufsize, info->ctx);
846 return sock_read (fd, buf, bufsize);
849 /* Like fd_read, except it provides a "preview" of the data that will
850 be read by subsequent calls to fd_read. Specifically, it copies no
851 more than BUFSIZE bytes of the currently available data to BUF and
852 returns the number of bytes copied. Return values and timeout
853 semantics are the same as those of fd_read.
855 CAVEAT: Do not assume that the first subsequent call to fd_read
856 will retrieve the same amount of data. Reading can return more or
857 less data, depending on the TCP implementation and other
858 circumstances. However, barring an error, it can be expected that
859 all the peeked data will eventually be read by fd_read. */
862 fd_peek (int fd, char *buf, int bufsize, double timeout)
864 struct transport_info *info;
865 LAZY_RETRIEVE_INFO (info);
866 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
868 if (info && info->imp->peeker)
869 return info->imp->peeker (fd, buf, bufsize, info->ctx);
871 return sock_peek (fd, buf, bufsize);
874 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
875 the operation aborts if no data is received after that many
876 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
880 fd_write (int fd, char *buf, int bufsize, double timeout)
883 struct transport_info *info;
884 LAZY_RETRIEVE_INFO (info);
886 /* `write' may write less than LEN bytes, thus the loop keeps trying
887 it until all was written, or an error occurred. */
891 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
893 if (info && info->imp->writer)
894 res = info->imp->writer (fd, buf, bufsize, info->ctx);
896 res = sock_write (fd, buf, bufsize);
905 /* Report the most recent error(s) on FD. This should only be called
906 after fd_* functions, such as fd_read and fd_write, and only if
907 they return a negative result. For errors coming from other calls
908 such as setsockopt or fopen, strerror should continue to be
911 If the transport doesn't support error messages or doesn't supply
912 one, strerror(errno) is returned. The returned error message
913 should not be used after fd_close has been called. */
918 /* Don't bother with LAZY_RETRIEVE_INFO, as this will only be called
919 in case of error, never in a tight loop. */
920 struct transport_info *info = NULL;
922 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
924 if (info && info->imp->errstr)
926 const char *err = info->imp->errstr (fd, info->ctx);
929 /* else, fall through and print the system error. */
931 return strerror (errno);
934 /* Close the file descriptor FD. */
939 struct transport_info *info;
943 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
944 per socket, so that particular optimization wouldn't work. */
947 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
949 if (info && info->imp->closer)
950 info->imp->closer (fd, info->ctx);
956 hash_table_remove (transport_map, (void *)(intptr_t) fd);
958 ++transport_map_modified_tick;