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 #include <sys/socket.h>
40 #include <sys/select.h>
45 # else /* def __VMS */
47 # endif /* def __VMS [else] */
48 # include <netinet/in.h>
50 # include <arpa/inet.h>
52 #endif /* not WINDOWS */
62 /* Apparently needed for Interix: */
67 /* Define sockaddr_storage where unavailable (presumably on IPv4-only
71 # ifndef HAVE_STRUCT_SOCKADDR_STORAGE
72 # define sockaddr_storage sockaddr_in
74 #endif /* ENABLE_IPV6 */
76 /* Fill SA as per the data in IP and PORT. SA shoult point to struct
77 sockaddr_storage if ENABLE_IPV6 is defined, to struct sockaddr_in
81 sockaddr_set_data (struct sockaddr *sa, const ip_address *ip, int port)
87 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
89 sin->sin_family = AF_INET;
90 sin->sin_port = htons (port);
91 sin->sin_addr = ip->data.d4;
97 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
99 sin6->sin6_family = AF_INET6;
100 sin6->sin6_port = htons (port);
101 sin6->sin6_addr = ip->data.d6;
102 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
103 sin6->sin6_scope_id = ip->ipv6_scope;
107 #endif /* ENABLE_IPV6 */
113 /* Get the data of SA, specifically the IP address and the port. If
114 you're not interested in one or the other information, pass NULL as
118 sockaddr_get_data (const struct sockaddr *sa, ip_address *ip, int *port)
120 switch (sa->sa_family)
124 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
127 ip->family = AF_INET;
128 ip->data.d4 = sin->sin_addr;
131 *port = ntohs (sin->sin_port);
137 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
140 ip->family = AF_INET6;
141 ip->data.d6 = sin6->sin6_addr;
142 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
143 ip->ipv6_scope = sin6->sin6_scope_id;
147 *port = ntohs (sin6->sin6_port);
156 /* Return the size of the sockaddr structure depending on its
160 sockaddr_size (const struct sockaddr *sa)
162 switch (sa->sa_family)
165 return sizeof (struct sockaddr_in);
168 return sizeof (struct sockaddr_in6);
175 /* Resolve the bind address specified via --bind-address and store it
176 to SA. The resolved value is stored in a static variable and
177 reused after the first invocation of this function.
179 Returns true on success, false on failure. */
182 resolve_bind_address (struct sockaddr *sa)
184 struct address_list *al;
186 /* Make sure this is called only once. opt.bind_address doesn't
187 change during a Wget run. */
188 static bool called, should_bind;
189 static ip_address ip;
193 sockaddr_set_data (sa, &ip, 0);
198 al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
201 /* #### We should be able to print the error message here. */
202 logprintf (LOG_NOTQUIET,
203 _("%s: unable to resolve bind address %s; disabling bind.\n"),
204 exec_name, quote (opt.bind_address));
209 /* Pick the first address in the list and use it as bind address.
210 Perhaps we should try multiple addresses in succession, but I
211 don't think that's necessary in practice. */
212 ip = *address_list_address_at (al, 0);
213 address_list_release (al);
215 sockaddr_set_data (sa, &ip, 0);
222 const struct sockaddr *addr;
228 connect_with_timeout_callback (void *arg)
230 struct cwt_context *ctx = (struct cwt_context *)arg;
231 ctx->result = connect (ctx->fd, ctx->addr, ctx->addrlen);
234 /* Like connect, but specifies a timeout. If connecting takes longer
235 than TIMEOUT seconds, -1 is returned and errno is set to
239 connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
242 struct cwt_context ctx;
245 ctx.addrlen = addrlen;
247 if (run_with_timeout (timeout, connect_with_timeout_callback, &ctx))
252 if (ctx.result == -1 && errno == EINTR)
257 /* Connect via TCP to the specified address and port.
259 If PRINT is non-NULL, it is the host name to print that we're
263 connect_to_ip (const ip_address *ip, int port, const char *print)
265 struct sockaddr_storage ss;
266 struct sockaddr *sa = (struct sockaddr *)&ss;
269 /* If PRINT is non-NULL, print the "Connecting to..." line, with
270 PRINT being the host name we're connecting to. */
273 const char *txt_addr = print_address (ip);
274 if (0 != strcmp (print, txt_addr))
276 char *str = NULL, *name;
278 if (opt.enable_iri && (name = idn_decode ((char *) print)) != NULL)
280 int len = strlen (print) + strlen (name) + 4;
282 snprintf (str, len, "%s (%s)", name, print);
287 logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
288 str ? str : escnonprint_uri (print), txt_addr, port);
295 if (ip->family == AF_INET)
296 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
298 else if (ip->family == AF_INET6)
299 logprintf (LOG_VERBOSE, _("Connecting to [%s]:%d... "), txt_addr, port);
304 /* Store the sockaddr info to SA. */
305 sockaddr_set_data (sa, ip, port);
307 /* Create the socket of the family appropriate for the address. */
308 sock = socket (sa->sa_family, SOCK_STREAM, 0);
312 #if defined(ENABLE_IPV6) && defined(IPV6_V6ONLY)
315 /* In case of error, we will go on anyway... */
316 int err = setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof (on));
319 DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
323 /* For very small rate limits, set the buffer size (and hence,
324 hopefully, the kernel's TCP window size) to the per-second limit.
325 That way we should never have to sleep for more than 1s between
327 if (opt.limit_rate && opt.limit_rate < 8192)
329 int bufsize = opt.limit_rate;
331 bufsize = 512; /* avoid pathologically small values */
333 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
334 (void *)&bufsize, (socklen_t)sizeof (bufsize));
336 /* When we add limit_rate support for writing, which is useful
337 for POST, we should also set SO_SNDBUF here. */
340 if (opt.bind_address)
342 /* Bind the client side of the socket to the requested
344 struct sockaddr_storage bind_ss;
345 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
346 if (resolve_bind_address (bind_sa))
348 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
353 /* Connect the socket to the remote endpoint. */
354 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
355 opt.connect_timeout) < 0)
361 logprintf (LOG_VERBOSE, _("connected.\n"));
362 DEBUGP (("Created socket %d.\n", sock));
367 /* Protect errno from possible modifications by close and
369 int save_errno = errno;
373 logprintf (LOG_VERBOSE, _("failed: %s.\n"), strerror (errno));
379 /* Connect via TCP to a remote host on the specified port.
381 HOST is resolved as an Internet host name. If HOST resolves to
382 more than one IP address, they are tried in the order returned by
383 DNS until connecting to one of them succeeds. */
386 connect_to_host (const char *host, int port)
391 struct address_list *al = lookup_host (host, 0);
396 logprintf (LOG_NOTQUIET,
397 _("%s: unable to resolve host address %s\n"),
398 exec_name, quote (host));
402 address_list_get_bounds (al, &start, &end);
403 for (i = start; i < end; i++)
405 const ip_address *ip = address_list_address_at (al, i);
406 sock = connect_to_ip (ip, port, host);
410 address_list_set_connected (al);
411 address_list_release (al);
415 /* The attempt to connect has failed. Continue with the loop
416 and try next address. */
418 address_list_set_faulty (al, i);
421 /* Failed to connect to any of the addresses in AL. */
423 if (address_list_connected_p (al))
425 /* We connected to AL before, but cannot do so now. That might
426 indicate that our DNS cache entry for HOST has expired. */
427 address_list_release (al);
428 al = lookup_host (host, LH_REFRESH);
431 address_list_release (al);
436 /* Create a socket, bind it to local interface BIND_ADDRESS on port
437 *PORT, set up a listen backlog, and return the resulting socket, or
440 BIND_ADDRESS is the address of the interface to bind to. If it is
441 NULL, the socket is bound to the default address. PORT should
442 point to the port number that will be used for the binding. If
443 that number is 0, the system will choose a suitable port, and the
444 chosen value will be written to *PORT.
446 Calling accept() on such a socket waits for and accepts incoming
450 bind_local (const ip_address *bind_address, int *port)
453 struct sockaddr_storage ss;
454 struct sockaddr *sa = (struct sockaddr *)&ss;
456 /* For setting options with setsockopt. */
458 void *setopt_ptr = (void *)&setopt_val;
459 socklen_t setopt_size = sizeof (setopt_val);
461 sock = socket (bind_address->family, SOCK_STREAM, 0);
466 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
470 sockaddr_set_data (sa, bind_address, *port);
471 if (bind (sock, sa, sockaddr_size (sa)) < 0)
476 DEBUGP (("Local socket fd %d bound.\n", sock));
478 /* If *PORT is 0, find out which port we've bound to. */
481 socklen_t addrlen = sockaddr_size (sa);
482 if (getsockname (sock, sa, &addrlen) < 0)
484 /* If we can't find out the socket's local address ("name"),
485 something is seriously wrong with the socket, and it's
486 unusable for us anyway because we must know the chosen
491 sockaddr_get_data (sa, NULL, port);
492 DEBUGP (("binding to address %s using port %i.\n",
493 print_address (bind_address), *port));
495 if (listen (sock, 1) < 0)
503 /* Like a call to accept(), but with the added check for timeout.
505 In other words, accept a client connection on LOCAL_SOCK, and
506 return the new socket used for communication with the client.
507 LOCAL_SOCK should have been bound, e.g. using bind_local().
509 The caller is blocked until a connection is established. If no
510 connection is established for opt.connect_timeout seconds, the
511 function exits with an error status. */
514 accept_connection (int local_sock)
518 /* We don't need the values provided by accept, but accept
519 apparently requires them to be present. */
520 struct sockaddr_storage ss;
521 struct sockaddr *sa = (struct sockaddr *)&ss;
522 socklen_t addrlen = sizeof (ss);
524 if (opt.connect_timeout)
526 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
532 sock = accept (local_sock, sa, &addrlen);
533 DEBUGP (("Accepted client at socket %d.\n", sock));
537 /* Get the IP address associated with the connection on FD and store
538 it to IP. Return true on success, false otherwise.
540 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
541 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
542 returns the address of the remote (peer's) side of the socket. */
545 socket_ip_address (int sock, ip_address *ip, int endpoint)
547 struct sockaddr_storage storage;
548 struct sockaddr *sockaddr = (struct sockaddr *) &storage;
549 socklen_t addrlen = sizeof (storage);
552 memset (sockaddr, 0, addrlen);
553 if (endpoint == ENDPOINT_LOCAL)
554 ret = getsockname (sock, sockaddr, &addrlen);
555 else if (endpoint == ENDPOINT_PEER)
556 ret = getpeername (sock, sockaddr, &addrlen);
562 memset(ip, 0, sizeof(ip_address));
563 ip->family = sockaddr->sa_family;
564 switch (sockaddr->sa_family)
569 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
570 ip->data.d6 = sa6->sin6_addr;
571 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
572 ip->ipv6_scope = sa6->sin6_scope_id;
574 DEBUGP (("conaddr is: %s\n", print_address (ip)));
580 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
581 ip->data.d4 = sa->sin_addr;
582 DEBUGP (("conaddr is: %s\n", print_address (ip)));
590 /* Get the socket family of connection on FD and store
591 Return family type on success, -1 otherwise.
593 If ENDPOINT is ENDPOINT_LOCAL, it returns the sock family of the local
594 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
595 returns the sock family of the remote (peer's) side of the socket. */
598 socket_family (int sock, int endpoint)
600 struct sockaddr_storage storage;
601 struct sockaddr *sockaddr = (struct sockaddr *) &storage;
602 socklen_t addrlen = sizeof (storage);
605 memset (sockaddr, 0, addrlen);
607 if (endpoint == ENDPOINT_LOCAL)
608 ret = getsockname (sock, sockaddr, &addrlen);
609 else if (endpoint == ENDPOINT_PEER)
610 ret = getpeername (sock, sockaddr, &addrlen);
617 return sockaddr->sa_family;
620 /* Return true if the error from the connect code can be considered
621 retryable. Wget normally retries after errors, but the exception
622 are the "unsupported protocol" type errors (possible on IPv4/IPv6
623 dual family systems) and "connection refused". */
626 retryable_socket_connect_error (int err)
628 /* Have to guard against some of these values not being defined.
629 Cannot use a switch statement because some of the values might be
633 || err == EAFNOSUPPORT
636 || err == EPFNOSUPPORT
638 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
639 || err == ESOCKTNOSUPPORT
641 #ifdef EPROTONOSUPPORT
642 || err == EPROTONOSUPPORT
645 || err == ENOPROTOOPT
647 /* Apparently, older versions of Linux and BSD used EINVAL
648 instead of EAFNOSUPPORT and such. */
653 if (!opt.retry_connrefused)
654 if (err == ECONNREFUSED
656 || err == ENETUNREACH /* network is unreachable */
659 || err == EHOSTUNREACH /* host is unreachable */
667 /* Wait for a single descriptor to become available, timing out after
668 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
669 -1 for error. The argument WAIT_FOR can be a combination of
670 WAIT_FOR_READ and WAIT_FOR_WRITE.
672 This is a mere convenience wrapper around the select call, and
673 should be taken as such (for example, it doesn't implement Wget's
674 0-timeout-means-no-timeout semantics.) */
677 select_fd (int fd, double maxtime, int wait_for)
680 fd_set *rd = NULL, *wr = NULL;
681 struct timeval tmout;
686 if (wait_for & WAIT_FOR_READ)
688 if (wait_for & WAIT_FOR_WRITE)
691 tmout.tv_sec = (long) maxtime;
692 tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
696 result = select (fd + 1, rd, wr, NULL, &tmout);
698 /* gnulib select() converts blocking sockets to nonblocking in windows.
699 wget uses blocking sockets so we must convert them back to blocking. */
700 set_windows_fd_as_blocking_socket (fd);
703 while (result < 0 && errno == EINTR);
708 /* Return true iff the connection to the remote site established
709 through SOCK is still open.
711 Specifically, this function returns true if SOCK is not ready for
712 reading. This is because, when the connection closes, the socket
713 is ready for reading because EOF is about to be delivered. A side
714 effect of this method is that sockets that have pending data are
715 considered non-open. This is actually a good thing for callers of
716 this function, where such pending data can only be unwanted
717 leftover from a previous request. */
720 test_socket_open (int sock)
726 /* Check if we still have a valid (non-EOF) connection. From Andrew
727 * Maholski's code in the Unix Socket FAQ. */
729 FD_ZERO (&check_set);
730 FD_SET (sock, &check_set);
732 /* Wait one microsecond */
736 ret = select (sock + 1, &check_set, NULL, NULL, &to);
738 /* gnulib select() converts blocking sockets to nonblocking in windows.
739 wget uses blocking sockets so we must convert them back to blocking
741 set_windows_fd_as_blocking_socket ( sock );
745 /* We got a timeout, it means we're still connected. */
748 /* Read now would not wait, it means we have either pending data
753 /* Basic socket operations, mostly EINTR wrappers. */
756 sock_read (int fd, char *buf, int bufsize)
760 res = read (fd, buf, bufsize);
761 while (res == -1 && errno == EINTR);
766 sock_write (int fd, char *buf, int bufsize)
770 res = write (fd, buf, bufsize);
771 while (res == -1 && errno == EINTR);
776 sock_poll (int fd, double timeout, int wait_for)
778 return select_fd (fd, timeout, wait_for);
782 sock_peek (int fd, char *buf, int bufsize)
786 res = recv (fd, buf, bufsize, MSG_PEEK);
787 while (res == -1 && errno == EINTR);
795 DEBUGP (("Closed fd %d\n", fd));
801 /* Reading and writing from the network. We build around the socket
802 (file descriptor) API, but support "extended" operations for things
803 that are not mere file descriptors under the hood, such as SSL
806 That way the user code can call fd_read(fd, ...) and we'll run read
807 or SSL_read or whatever is necessary. */
809 static struct hash_table *transport_map;
810 static unsigned int transport_map_modified_tick;
812 struct transport_info {
813 struct transport_implementation *imp;
817 /* Register the transport layer operations that will be used when
818 reading, writing, and polling FD.
820 This should be used for transport layers like SSL that piggyback on
821 sockets. FD should otherwise be a real socket, on which you can
822 call getpeername, etc. */
825 fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
827 struct transport_info *info;
829 /* The file descriptor must be non-negative to be registered.
830 Negative values are ignored by fd_close(), and -1 cannot be used as
834 info = xnew (struct transport_info);
838 transport_map = hash_table_new (0, NULL, NULL);
839 hash_table_put (transport_map, (void *)(intptr_t) fd, info);
840 ++transport_map_modified_tick;
843 /* Return context of the transport registered with
844 fd_register_transport. This assumes fd_register_transport was
845 previously called on FD. */
848 fd_transport_context (int fd)
850 struct transport_info *info = hash_table_get (transport_map, (void *)(intptr_t) fd);
854 /* When fd_read/fd_write are called multiple times in a loop, they should
855 remember the INFO pointer instead of fetching it every time. It is
856 not enough to compare FD to LAST_FD because FD might have been
857 closed and reopened. modified_tick ensures that changes to
858 transport_map will not be unnoticed.
860 This is a macro because we want the static storage variables to be
863 #define LAZY_RETRIEVE_INFO(info) do { \
864 static struct transport_info *last_info; \
865 static int last_fd = -1; \
866 static unsigned int last_tick; \
867 if (!transport_map) \
869 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
873 info = hash_table_get (transport_map, (void *)(intptr_t) fd); \
876 last_tick = transport_map_modified_tick; \
881 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
884 timeout = opt.read_timeout;
888 if (info && info->imp->poller)
889 test = info->imp->poller (fd, timeout, wf, info->ctx);
891 test = sock_poll (fd, timeout, wf);
900 /* Read no more than BUFSIZE bytes of data from FD, storing them to
901 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
902 received after that many seconds. If TIMEOUT is -1, the value of
903 opt.timeout is used for TIMEOUT. */
906 fd_read (int fd, char *buf, int bufsize, double timeout)
908 struct transport_info *info;
909 LAZY_RETRIEVE_INFO (info);
910 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
912 if (info && info->imp->reader)
913 return info->imp->reader (fd, buf, bufsize, info->ctx);
915 return sock_read (fd, buf, bufsize);
918 /* Like fd_read, except it provides a "preview" of the data that will
919 be read by subsequent calls to fd_read. Specifically, it copies no
920 more than BUFSIZE bytes of the currently available data to BUF and
921 returns the number of bytes copied. Return values and timeout
922 semantics are the same as those of fd_read.
924 CAVEAT: Do not assume that the first subsequent call to fd_read
925 will retrieve the same amount of data. Reading can return more or
926 less data, depending on the TCP implementation and other
927 circumstances. However, barring an error, it can be expected that
928 all the peeked data will eventually be read by fd_read. */
931 fd_peek (int fd, char *buf, int bufsize, double timeout)
933 struct transport_info *info;
934 LAZY_RETRIEVE_INFO (info);
935 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
937 if (info && info->imp->peeker)
938 return info->imp->peeker (fd, buf, bufsize, info->ctx);
940 return sock_peek (fd, buf, bufsize);
943 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
944 the operation aborts if no data is received after that many
945 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
949 fd_write (int fd, char *buf, int bufsize, double timeout)
952 struct transport_info *info;
953 LAZY_RETRIEVE_INFO (info);
955 /* `write' may write less than LEN bytes, thus the loop keeps trying
956 it until all was written, or an error occurred. */
960 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
962 if (info && info->imp->writer)
963 res = info->imp->writer (fd, buf, bufsize, info->ctx);
965 res = sock_write (fd, buf, bufsize);
974 /* Report the most recent error(s) on FD. This should only be called
975 after fd_* functions, such as fd_read and fd_write, and only if
976 they return a negative result. For errors coming from other calls
977 such as setsockopt or fopen, strerror should continue to be
980 If the transport doesn't support error messages or doesn't supply
981 one, strerror(errno) is returned. The returned error message
982 should not be used after fd_close has been called. */
987 /* Don't bother with LAZY_RETRIEVE_INFO, as this will only be called
988 in case of error, never in a tight loop. */
989 struct transport_info *info = NULL;
991 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
993 if (info && info->imp->errstr)
995 const char *err = info->imp->errstr (fd, info->ctx);
998 /* else, fall through and print the system error. */
1000 return strerror (errno);
1003 /* Close the file descriptor FD. */
1008 struct transport_info *info;
1012 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
1013 per socket, so that particular optimization wouldn't work. */
1016 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
1018 if (info && info->imp->closer)
1019 info->imp->closer (fd, info->ctx);
1025 hash_table_remove (transport_map, (void *)(intptr_t) fd);
1027 ++transport_map_modified_tick;