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 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. */
41 # include <sys/socket.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_SELECT_H
56 # include <sys/select.h>
57 #endif /* HAVE_SYS_SELECT_H */
58 #ifdef HAVE_SYS_TIME_H
59 # include <sys/time.h>
66 /* Apparently needed for Interix: */
71 /* Define sockaddr_storage where unavailable (presumably on IPv4-only
75 # ifndef HAVE_STRUCT_SOCKADDR_STORAGE
76 # define sockaddr_storage sockaddr_in
78 #endif /* ENABLE_IPV6 */
80 /* Fill SA as per the data in IP and PORT. SA shoult point to struct
81 sockaddr_storage if ENABLE_IPV6 is defined, to struct sockaddr_in
85 sockaddr_set_data (struct sockaddr *sa, const ip_address *ip, int port)
91 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
93 sin->sin_family = AF_INET;
94 sin->sin_port = htons (port);
95 sin->sin_addr = ip->data.d4;
101 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
103 sin6->sin6_family = AF_INET6;
104 sin6->sin6_port = htons (port);
105 sin6->sin6_addr = ip->data.d6;
106 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
107 sin6->sin6_scope_id = ip->ipv6_scope;
111 #endif /* ENABLE_IPV6 */
117 /* Get the data of SA, specifically the IP address and the port. If
118 you're not interested in one or the other information, pass NULL as
122 sockaddr_get_data (const struct sockaddr *sa, ip_address *ip, int *port)
124 switch (sa->sa_family)
128 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
131 ip->family = AF_INET;
132 ip->data.d4 = sin->sin_addr;
135 *port = ntohs (sin->sin_port);
141 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
144 ip->family = AF_INET6;
145 ip->data.d6 = sin6->sin6_addr;
146 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
147 ip->ipv6_scope = sin6->sin6_scope_id;
151 *port = ntohs (sin6->sin6_port);
160 /* Return the size of the sockaddr structure depending on its
164 sockaddr_size (const struct sockaddr *sa)
166 switch (sa->sa_family)
169 return sizeof (struct sockaddr_in);
172 return sizeof (struct sockaddr_in6);
179 /* Resolve the bind address specified via --bind-address and store it
180 to SA. The resolved value is stored in a static variable and
181 reused after the first invocation of this function.
183 Returns true on success, false on failure. */
186 resolve_bind_address (struct sockaddr *sa)
188 struct address_list *al;
190 /* Make sure this is called only once. opt.bind_address doesn't
191 change during a Wget run. */
192 static bool called, should_bind;
193 static ip_address ip;
197 sockaddr_set_data (sa, &ip, 0);
202 al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
205 /* #### We should be able to print the error message here. */
206 logprintf (LOG_NOTQUIET,
207 _("%s: unable to resolve bind address %s; disabling bind.\n"),
208 exec_name, quote (opt.bind_address));
213 /* Pick the first address in the list and use it as bind address.
214 Perhaps we should try multiple addresses in succession, but I
215 don't think that's necessary in practice. */
216 ip = *address_list_address_at (al, 0);
217 address_list_release (al);
219 sockaddr_set_data (sa, &ip, 0);
226 const struct sockaddr *addr;
232 connect_with_timeout_callback (void *arg)
234 struct cwt_context *ctx = (struct cwt_context *)arg;
235 ctx->result = connect (ctx->fd, ctx->addr, ctx->addrlen);
238 /* Like connect, but specifies a timeout. If connecting takes longer
239 than TIMEOUT seconds, -1 is returned and errno is set to
243 connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
246 struct cwt_context ctx;
249 ctx.addrlen = addrlen;
251 if (run_with_timeout (timeout, connect_with_timeout_callback, &ctx))
256 if (ctx.result == -1 && errno == EINTR)
261 /* Connect via TCP to the specified address and port.
263 If PRINT is non-NULL, it is the host name to print that we're
267 connect_to_ip (const ip_address *ip, int port, const char *print)
269 struct sockaddr_storage ss;
270 struct sockaddr *sa = (struct sockaddr *)&ss;
273 /* If PRINT is non-NULL, print the "Connecting to..." line, with
274 PRINT being the host name we're connecting to. */
277 const char *txt_addr = print_address (ip);
278 if (0 != strcmp (print, txt_addr))
280 char *str = NULL, *name;
282 if (opt.enable_iri && (name = idn_decode ((char *) print)) != NULL)
284 int len = strlen (print) + strlen (name) + 4;
286 snprintf (str, len, "%s (%s)", name, print);
291 logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
292 str ? str : escnonprint_uri (print), txt_addr, port);
298 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
301 /* Store the sockaddr info to SA. */
302 sockaddr_set_data (sa, ip, port);
304 /* Create the socket of the family appropriate for the address. */
305 sock = socket (sa->sa_family, SOCK_STREAM, 0);
309 #if defined(ENABLE_IPV6) && defined(IPV6_V6ONLY)
312 /* In case of error, we will go on anyway... */
313 int err = setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof (on));
316 DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
320 /* For very small rate limits, set the buffer size (and hence,
321 hopefully, the kernel's TCP window size) to the per-second limit.
322 That way we should never have to sleep for more than 1s between
324 if (opt.limit_rate && opt.limit_rate < 8192)
326 int bufsize = opt.limit_rate;
328 bufsize = 512; /* avoid pathologically small values */
330 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
331 (void *)&bufsize, (socklen_t)sizeof (bufsize));
333 /* When we add limit_rate support for writing, which is useful
334 for POST, we should also set SO_SNDBUF here. */
337 if (opt.bind_address)
339 /* Bind the client side of the socket to the requested
341 struct sockaddr_storage bind_ss;
342 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
343 if (resolve_bind_address (bind_sa))
345 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
350 /* Connect the socket to the remote endpoint. */
351 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
352 opt.connect_timeout) < 0)
358 logprintf (LOG_VERBOSE, _("connected.\n"));
359 DEBUGP (("Created socket %d.\n", sock));
364 /* Protect errno from possible modifications by close and
366 int save_errno = errno;
370 logprintf (LOG_VERBOSE, _("failed: %s.\n"), strerror (errno));
376 /* Connect via TCP to a remote host on the specified port.
378 HOST is resolved as an Internet host name. If HOST resolves to
379 more than one IP address, they are tried in the order returned by
380 DNS until connecting to one of them succeeds. */
383 connect_to_host (const char *host, int port)
388 struct address_list *al = lookup_host (host, 0);
393 logprintf (LOG_NOTQUIET,
394 _("%s: unable to resolve host address %s\n"),
395 exec_name, quote (host));
399 address_list_get_bounds (al, &start, &end);
400 for (i = start; i < end; i++)
402 const ip_address *ip = address_list_address_at (al, i);
403 sock = connect_to_ip (ip, port, host);
407 address_list_set_connected (al);
408 address_list_release (al);
412 /* The attempt to connect has failed. Continue with the loop
413 and try next address. */
415 address_list_set_faulty (al, i);
418 /* Failed to connect to any of the addresses in AL. */
420 if (address_list_connected_p (al))
422 /* We connected to AL before, but cannot do so now. That might
423 indicate that our DNS cache entry for HOST has expired. */
424 address_list_release (al);
425 al = lookup_host (host, LH_REFRESH);
428 address_list_release (al);
433 /* Create a socket, bind it to local interface BIND_ADDRESS on port
434 *PORT, set up a listen backlog, and return the resulting socket, or
437 BIND_ADDRESS is the address of the interface to bind to. If it is
438 NULL, the socket is bound to the default address. PORT should
439 point to the port number that will be used for the binding. If
440 that number is 0, the system will choose a suitable port, and the
441 chosen value will be written to *PORT.
443 Calling accept() on such a socket waits for and accepts incoming
447 bind_local (const ip_address *bind_address, int *port)
450 struct sockaddr_storage ss;
451 struct sockaddr *sa = (struct sockaddr *)&ss;
453 /* For setting options with setsockopt. */
455 void *setopt_ptr = (void *)&setopt_val;
456 socklen_t setopt_size = sizeof (setopt_val);
458 sock = socket (bind_address->family, SOCK_STREAM, 0);
463 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
467 sockaddr_set_data (sa, bind_address, *port);
468 if (bind (sock, sa, sockaddr_size (sa)) < 0)
473 DEBUGP (("Local socket fd %d bound.\n", sock));
475 /* If *PORT is 0, find out which port we've bound to. */
478 socklen_t addrlen = sockaddr_size (sa);
479 if (getsockname (sock, sa, &addrlen) < 0)
481 /* If we can't find out the socket's local address ("name"),
482 something is seriously wrong with the socket, and it's
483 unusable for us anyway because we must know the chosen
488 sockaddr_get_data (sa, NULL, port);
489 DEBUGP (("binding to address %s using port %i.\n",
490 print_address (bind_address), *port));
492 if (listen (sock, 1) < 0)
500 /* Like a call to accept(), but with the added check for timeout.
502 In other words, accept a client connection on LOCAL_SOCK, and
503 return the new socket used for communication with the client.
504 LOCAL_SOCK should have been bound, e.g. using bind_local().
506 The caller is blocked until a connection is established. If no
507 connection is established for opt.connect_timeout seconds, the
508 function exits with an error status. */
511 accept_connection (int local_sock)
515 /* We don't need the values provided by accept, but accept
516 apparently requires them to be present. */
517 struct sockaddr_storage ss;
518 struct sockaddr *sa = (struct sockaddr *)&ss;
519 socklen_t addrlen = sizeof (ss);
521 if (opt.connect_timeout)
523 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
529 sock = accept (local_sock, sa, &addrlen);
530 DEBUGP (("Accepted client at socket %d.\n", sock));
534 /* Get the IP address associated with the connection on FD and store
535 it to IP. Return true on success, false otherwise.
537 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
538 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
539 returns the address of the remote (peer's) side of the socket. */
542 socket_ip_address (int sock, ip_address *ip, int endpoint)
544 struct sockaddr_storage storage;
545 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
546 socklen_t addrlen = sizeof (storage);
549 if (endpoint == ENDPOINT_LOCAL)
550 ret = getsockname (sock, sockaddr, &addrlen);
551 else if (endpoint == ENDPOINT_PEER)
552 ret = getpeername (sock, sockaddr, &addrlen);
558 ip->family = sockaddr->sa_family;
559 switch (sockaddr->sa_family)
564 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
565 ip->data.d6 = sa6->sin6_addr;
566 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
567 ip->ipv6_scope = sa6->sin6_scope_id;
569 DEBUGP (("conaddr is: %s\n", print_address (ip)));
575 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
576 ip->data.d4 = sa->sin_addr;
577 DEBUGP (("conaddr is: %s\n", print_address (ip)));
585 /* Return true if the error from the connect code can be considered
586 retryable. Wget normally retries after errors, but the exception
587 are the "unsupported protocol" type errors (possible on IPv4/IPv6
588 dual family systems) and "connection refused". */
591 retryable_socket_connect_error (int err)
593 /* Have to guard against some of these values not being defined.
594 Cannot use a switch statement because some of the values might be
598 || err == EAFNOSUPPORT
601 || err == EPFNOSUPPORT
603 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
604 || err == ESOCKTNOSUPPORT
606 #ifdef EPROTONOSUPPORT
607 || err == EPROTONOSUPPORT
610 || err == ENOPROTOOPT
612 /* Apparently, older versions of Linux and BSD used EINVAL
613 instead of EAFNOSUPPORT and such. */
618 if (!opt.retry_connrefused)
619 if (err == ECONNREFUSED
621 || err == ENETUNREACH /* network is unreachable */
624 || err == EHOSTUNREACH /* host is unreachable */
632 /* Wait for a single descriptor to become available, timing out after
633 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
634 -1 for error. The argument WAIT_FOR can be a combination of
635 WAIT_FOR_READ and WAIT_FOR_WRITE.
637 This is a mere convenience wrapper around the select call, and
638 should be taken as such (for example, it doesn't implement Wget's
639 0-timeout-means-no-timeout semantics.) */
642 select_fd (int fd, double maxtime, int wait_for)
645 fd_set *rd = NULL, *wr = NULL;
646 struct timeval tmout;
651 if (wait_for & WAIT_FOR_READ)
653 if (wait_for & WAIT_FOR_WRITE)
656 tmout.tv_sec = (long) maxtime;
657 tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
660 result = select (fd + 1, rd, wr, NULL, &tmout);
661 while (result < 0 && errno == EINTR);
666 /* Return true iff the connection to the remote site established
667 through SOCK is still open.
669 Specifically, this function returns true if SOCK is not ready for
670 reading. This is because, when the connection closes, the socket
671 is ready for reading because EOF is about to be delivered. A side
672 effect of this method is that sockets that have pending data are
673 considered non-open. This is actually a good thing for callers of
674 this function, where such pending data can only be unwanted
675 leftover from a previous request. */
678 test_socket_open (int sock)
683 /* Check if we still have a valid (non-EOF) connection. From Andrew
684 * Maholski's code in the Unix Socket FAQ. */
686 FD_ZERO (&check_set);
687 FD_SET (sock, &check_set);
689 /* Wait one microsecond */
693 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
694 /* We got a timeout, it means we're still connected. */
697 /* Read now would not wait, it means we have either pending data
702 /* Basic socket operations, mostly EINTR wrappers. */
704 #if defined(WINDOWS) || defined(USE_WATT32)
705 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
706 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
707 # define close(fd) closesocket (fd)
711 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
712 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
716 sock_read (int fd, char *buf, int bufsize)
720 res = read (fd, buf, bufsize);
721 while (res == -1 && errno == EINTR);
726 sock_write (int fd, char *buf, int bufsize)
730 res = write (fd, buf, bufsize);
731 while (res == -1 && errno == EINTR);
736 sock_poll (int fd, double timeout, int wait_for)
738 return select_fd (fd, timeout, wait_for);
742 sock_peek (int fd, char *buf, int bufsize)
746 res = recv (fd, buf, bufsize, MSG_PEEK);
747 while (res == -1 && errno == EINTR);
755 DEBUGP (("Closed fd %d\n", fd));
761 /* Reading and writing from the network. We build around the socket
762 (file descriptor) API, but support "extended" operations for things
763 that are not mere file descriptors under the hood, such as SSL
766 That way the user code can call fd_read(fd, ...) and we'll run read
767 or SSL_read or whatever is necessary. */
769 static struct hash_table *transport_map;
770 static unsigned int transport_map_modified_tick;
772 struct transport_info {
773 struct transport_implementation *imp;
777 /* Register the transport layer operations that will be used when
778 reading, writing, and polling FD.
780 This should be used for transport layers like SSL that piggyback on
781 sockets. FD should otherwise be a real socket, on which you can
782 call getpeername, etc. */
785 fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
787 struct transport_info *info;
789 /* The file descriptor must be non-negative to be registered.
790 Negative values are ignored by fd_close(), and -1 cannot be used as
794 info = xnew (struct transport_info);
798 transport_map = hash_table_new (0, NULL, NULL);
799 hash_table_put (transport_map, (void *)(intptr_t) fd, info);
800 ++transport_map_modified_tick;
803 /* Return context of the transport registered with
804 fd_register_transport. This assumes fd_register_transport was
805 previously called on FD. */
808 fd_transport_context (int fd)
810 struct transport_info *info = hash_table_get (transport_map, (void *)(intptr_t) fd);
814 /* When fd_read/fd_write are called multiple times in a loop, they should
815 remember the INFO pointer instead of fetching it every time. It is
816 not enough to compare FD to LAST_FD because FD might have been
817 closed and reopened. modified_tick ensures that changes to
818 transport_map will not be unnoticed.
820 This is a macro because we want the static storage variables to be
823 #define LAZY_RETRIEVE_INFO(info) do { \
824 static struct transport_info *last_info; \
825 static int last_fd = -1; \
826 static unsigned int last_tick; \
827 if (!transport_map) \
829 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
833 info = hash_table_get (transport_map, (void *)(intptr_t) fd); \
836 last_tick = transport_map_modified_tick; \
841 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
844 timeout = opt.read_timeout;
848 if (info && info->imp->poller)
849 test = info->imp->poller (fd, timeout, wf, info->ctx);
851 test = sock_poll (fd, timeout, wf);
860 /* Read no more than BUFSIZE bytes of data from FD, storing them to
861 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
862 received after that many seconds. If TIMEOUT is -1, the value of
863 opt.timeout is used for TIMEOUT. */
866 fd_read (int fd, char *buf, int bufsize, double timeout)
868 struct transport_info *info;
869 LAZY_RETRIEVE_INFO (info);
870 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
872 if (info && info->imp->reader)
873 return info->imp->reader (fd, buf, bufsize, info->ctx);
875 return sock_read (fd, buf, bufsize);
878 /* Like fd_read, except it provides a "preview" of the data that will
879 be read by subsequent calls to fd_read. Specifically, it copies no
880 more than BUFSIZE bytes of the currently available data to BUF and
881 returns the number of bytes copied. Return values and timeout
882 semantics are the same as those of fd_read.
884 CAVEAT: Do not assume that the first subsequent call to fd_read
885 will retrieve the same amount of data. Reading can return more or
886 less data, depending on the TCP implementation and other
887 circumstances. However, barring an error, it can be expected that
888 all the peeked data will eventually be read by fd_read. */
891 fd_peek (int fd, char *buf, int bufsize, double timeout)
893 struct transport_info *info;
894 LAZY_RETRIEVE_INFO (info);
895 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
897 if (info && info->imp->peeker)
898 return info->imp->peeker (fd, buf, bufsize, info->ctx);
900 return sock_peek (fd, buf, bufsize);
903 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
904 the operation aborts if no data is received after that many
905 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
909 fd_write (int fd, char *buf, int bufsize, double timeout)
912 struct transport_info *info;
913 LAZY_RETRIEVE_INFO (info);
915 /* `write' may write less than LEN bytes, thus the loop keeps trying
916 it until all was written, or an error occurred. */
920 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
922 if (info && info->imp->writer)
923 res = info->imp->writer (fd, buf, bufsize, info->ctx);
925 res = sock_write (fd, buf, bufsize);
934 /* Report the most recent error(s) on FD. This should only be called
935 after fd_* functions, such as fd_read and fd_write, and only if
936 they return a negative result. For errors coming from other calls
937 such as setsockopt or fopen, strerror should continue to be
940 If the transport doesn't support error messages or doesn't supply
941 one, strerror(errno) is returned. The returned error message
942 should not be used after fd_close has been called. */
947 /* Don't bother with LAZY_RETRIEVE_INFO, as this will only be called
948 in case of error, never in a tight loop. */
949 struct transport_info *info = NULL;
951 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
953 if (info && info->imp->errstr)
955 const char *err = info->imp->errstr (fd, info->ctx);
958 /* else, fall through and print the system error. */
960 return strerror (errno);
963 /* Close the file descriptor FD. */
968 struct transport_info *info;
972 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
973 per socket, so that particular optimization wouldn't work. */
976 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
978 if (info && info->imp->closer)
979 info->imp->closer (fd, info->ctx);
985 hash_table_remove (transport_map, (void *)(intptr_t) fd);
987 ++transport_map_modified_tick;