1 /* Establishing and handling network connections.
2 Copyright (C) 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007, 2008 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 /* Define sockaddr_storage where unavailable (presumably on IPv4-only
70 # ifndef HAVE_STRUCT_SOCKADDR_STORAGE
71 # define sockaddr_storage sockaddr_in
73 #endif /* ENABLE_IPV6 */
75 /* Fill SA as per the data in IP and PORT. SA shoult point to struct
76 sockaddr_storage if ENABLE_IPV6 is defined, to struct sockaddr_in
80 sockaddr_set_data (struct sockaddr *sa, const ip_address *ip, int port)
86 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
88 sin->sin_family = AF_INET;
89 sin->sin_port = htons (port);
90 sin->sin_addr = ip->data.d4;
96 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
98 sin6->sin6_family = AF_INET6;
99 sin6->sin6_port = htons (port);
100 sin6->sin6_addr = ip->data.d6;
101 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
102 sin6->sin6_scope_id = ip->ipv6_scope;
106 #endif /* ENABLE_IPV6 */
112 /* Get the data of SA, specifically the IP address and the port. If
113 you're not interested in one or the other information, pass NULL as
117 sockaddr_get_data (const struct sockaddr *sa, ip_address *ip, int *port)
119 switch (sa->sa_family)
123 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
126 ip->family = AF_INET;
127 ip->data.d4 = sin->sin_addr;
130 *port = ntohs (sin->sin_port);
136 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
139 ip->family = AF_INET6;
140 ip->data.d6 = sin6->sin6_addr;
141 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
142 ip->ipv6_scope = sin6->sin6_scope_id;
146 *port = ntohs (sin6->sin6_port);
155 /* Return the size of the sockaddr structure depending on its
159 sockaddr_size (const struct sockaddr *sa)
161 switch (sa->sa_family)
164 return sizeof (struct sockaddr_in);
167 return sizeof (struct sockaddr_in6);
174 /* Resolve the bind address specified via --bind-address and store it
175 to SA. The resolved value is stored in a static variable and
176 reused after the first invocation of this function.
178 Returns true on success, false on failure. */
181 resolve_bind_address (struct sockaddr *sa)
183 struct address_list *al;
185 /* Make sure this is called only once. opt.bind_address doesn't
186 change during a Wget run. */
187 static bool called, should_bind;
188 static ip_address ip;
192 sockaddr_set_data (sa, &ip, 0);
197 al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
200 /* #### We should be able to print the error message here. */
201 logprintf (LOG_NOTQUIET,
202 _("%s: unable to resolve bind address `%s'; disabling bind.\n"),
203 exec_name, opt.bind_address);
208 /* Pick the first address in the list and use it as bind address.
209 Perhaps we should try multiple addresses in succession, but I
210 don't think that's necessary in practice. */
211 ip = *address_list_address_at (al, 0);
212 address_list_release (al);
214 sockaddr_set_data (sa, &ip, 0);
221 const struct sockaddr *addr;
227 connect_with_timeout_callback (void *arg)
229 struct cwt_context *ctx = (struct cwt_context *)arg;
230 ctx->result = connect (ctx->fd, ctx->addr, ctx->addrlen);
233 /* Like connect, but specifies a timeout. If connecting takes longer
234 than TIMEOUT seconds, -1 is returned and errno is set to
238 connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
241 struct cwt_context ctx;
244 ctx.addrlen = addrlen;
246 if (run_with_timeout (timeout, connect_with_timeout_callback, &ctx))
251 if (ctx.result == -1 && errno == EINTR)
256 /* Connect via TCP to the specified address and port.
258 If PRINT is non-NULL, it is the host name to print that we're
262 connect_to_ip (const ip_address *ip, int port, const char *print)
264 struct sockaddr_storage ss;
265 struct sockaddr *sa = (struct sockaddr *)&ss;
268 /* If PRINT is non-NULL, print the "Connecting to..." line, with
269 PRINT being the host name we're connecting to. */
272 const char *txt_addr = print_address (ip);
273 if (print && 0 != strcmp (print, txt_addr))
274 logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
275 escnonprint (print), txt_addr, port);
277 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
280 /* Store the sockaddr info to SA. */
281 sockaddr_set_data (sa, ip, port);
283 /* Create the socket of the family appropriate for the address. */
284 sock = socket (sa->sa_family, SOCK_STREAM, 0);
288 #if defined(ENABLE_IPV6) && defined(IPV6_V6ONLY)
291 /* In case of error, we will go on anyway... */
292 int err = setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof (on));
295 DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
299 /* For very small rate limits, set the buffer size (and hence,
300 hopefully, the kernel's TCP window size) to the per-second limit.
301 That way we should never have to sleep for more than 1s between
303 if (opt.limit_rate && opt.limit_rate < 8192)
305 int bufsize = opt.limit_rate;
307 bufsize = 512; /* avoid pathologically small values */
309 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
310 (void *)&bufsize, (socklen_t)sizeof (bufsize));
312 /* When we add limit_rate support for writing, which is useful
313 for POST, we should also set SO_SNDBUF here. */
316 if (opt.bind_address)
318 /* Bind the client side of the socket to the requested
320 struct sockaddr_storage bind_ss;
321 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
322 if (resolve_bind_address (bind_sa))
324 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
329 /* Connect the socket to the remote endpoint. */
330 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
331 opt.connect_timeout) < 0)
337 logprintf (LOG_VERBOSE, _("connected.\n"));
338 DEBUGP (("Created socket %d.\n", sock));
343 /* Protect errno from possible modifications by close and
345 int save_errno = errno;
349 logprintf (LOG_VERBOSE, _("failed: %s.\n"), strerror (errno));
355 /* Connect via TCP to a remote host on the specified port.
357 HOST is resolved as an Internet host name. If HOST resolves to
358 more than one IP address, they are tried in the order returned by
359 DNS until connecting to one of them succeeds. */
362 connect_to_host (const char *host, int port)
367 struct address_list *al = lookup_host (host, 0);
372 logprintf (LOG_NOTQUIET,
373 _("%s: unable to resolve host address `%s'\n"),
378 address_list_get_bounds (al, &start, &end);
379 for (i = start; i < end; i++)
381 const ip_address *ip = address_list_address_at (al, i);
382 sock = connect_to_ip (ip, port, host);
386 address_list_set_connected (al);
387 address_list_release (al);
391 /* The attempt to connect has failed. Continue with the loop
392 and try next address. */
394 address_list_set_faulty (al, i);
397 /* Failed to connect to any of the addresses in AL. */
399 if (address_list_connected_p (al))
401 /* We connected to AL before, but cannot do so now. That might
402 indicate that our DNS cache entry for HOST has expired. */
403 address_list_release (al);
404 al = lookup_host (host, LH_REFRESH);
407 address_list_release (al);
412 /* Create a socket, bind it to local interface BIND_ADDRESS on port
413 *PORT, set up a listen backlog, and return the resulting socket, or
416 BIND_ADDRESS is the address of the interface to bind to. If it is
417 NULL, the socket is bound to the default address. PORT should
418 point to the port number that will be used for the binding. If
419 that number is 0, the system will choose a suitable port, and the
420 chosen value will be written to *PORT.
422 Calling accept() on such a socket waits for and accepts incoming
426 bind_local (const ip_address *bind_address, int *port)
429 struct sockaddr_storage ss;
430 struct sockaddr *sa = (struct sockaddr *)&ss;
432 /* For setting options with setsockopt. */
434 void *setopt_ptr = (void *)&setopt_val;
435 socklen_t setopt_size = sizeof (setopt_val);
437 sock = socket (bind_address->family, SOCK_STREAM, 0);
442 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
446 sockaddr_set_data (sa, bind_address, *port);
447 if (bind (sock, sa, sockaddr_size (sa)) < 0)
452 DEBUGP (("Local socket fd %d bound.\n", sock));
454 /* If *PORT is 0, find out which port we've bound to. */
457 socklen_t addrlen = sockaddr_size (sa);
458 if (getsockname (sock, sa, &addrlen) < 0)
460 /* If we can't find out the socket's local address ("name"),
461 something is seriously wrong with the socket, and it's
462 unusable for us anyway because we must know the chosen
467 sockaddr_get_data (sa, NULL, port);
468 DEBUGP (("binding to address %s using port %i.\n",
469 print_address (bind_address), *port));
471 if (listen (sock, 1) < 0)
479 /* Like a call to accept(), but with the added check for timeout.
481 In other words, accept a client connection on LOCAL_SOCK, and
482 return the new socket used for communication with the client.
483 LOCAL_SOCK should have been bound, e.g. using bind_local().
485 The caller is blocked until a connection is established. If no
486 connection is established for opt.connect_timeout seconds, the
487 function exits with an error status. */
490 accept_connection (int local_sock)
494 /* We don't need the values provided by accept, but accept
495 apparently requires them to be present. */
496 struct sockaddr_storage ss;
497 struct sockaddr *sa = (struct sockaddr *)&ss;
498 socklen_t addrlen = sizeof (ss);
500 if (opt.connect_timeout)
502 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
508 sock = accept (local_sock, sa, &addrlen);
509 DEBUGP (("Accepted client at socket %d.\n", sock));
513 /* Get the IP address associated with the connection on FD and store
514 it to IP. Return true on success, false otherwise.
516 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
517 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
518 returns the address of the remote (peer's) side of the socket. */
521 socket_ip_address (int sock, ip_address *ip, int endpoint)
523 struct sockaddr_storage storage;
524 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
525 socklen_t addrlen = sizeof (storage);
528 if (endpoint == ENDPOINT_LOCAL)
529 ret = getsockname (sock, sockaddr, &addrlen);
530 else if (endpoint == ENDPOINT_PEER)
531 ret = getpeername (sock, sockaddr, &addrlen);
537 ip->family = sockaddr->sa_family;
538 switch (sockaddr->sa_family)
543 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
544 ip->data.d6 = sa6->sin6_addr;
545 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
546 ip->ipv6_scope = sa6->sin6_scope_id;
548 DEBUGP (("conaddr is: %s\n", print_address (ip)));
554 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
555 ip->data.d4 = sa->sin_addr;
556 DEBUGP (("conaddr is: %s\n", print_address (ip)));
564 /* Return true if the error from the connect code can be considered
565 retryable. Wget normally retries after errors, but the exception
566 are the "unsupported protocol" type errors (possible on IPv4/IPv6
567 dual family systems) and "connection refused". */
570 retryable_socket_connect_error (int err)
572 /* Have to guard against some of these values not being defined.
573 Cannot use a switch statement because some of the values might be
577 || err == EAFNOSUPPORT
580 || err == EPFNOSUPPORT
582 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
583 || err == ESOCKTNOSUPPORT
585 #ifdef EPROTONOSUPPORT
586 || err == EPROTONOSUPPORT
589 || err == ENOPROTOOPT
591 /* Apparently, older versions of Linux and BSD used EINVAL
592 instead of EAFNOSUPPORT and such. */
597 if (!opt.retry_connrefused)
598 if (err == ECONNREFUSED
600 || err == ENETUNREACH /* network is unreachable */
603 || err == EHOSTUNREACH /* host is unreachable */
611 /* Wait for a single descriptor to become available, timing out after
612 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
613 -1 for error. The argument WAIT_FOR can be a combination of
614 WAIT_FOR_READ and WAIT_FOR_WRITE.
616 This is a mere convenience wrapper around the select call, and
617 should be taken as such (for example, it doesn't implement Wget's
618 0-timeout-means-no-timeout semantics.) */
621 select_fd (int fd, double maxtime, int wait_for)
624 fd_set *rd = NULL, *wr = NULL;
625 struct timeval tmout;
630 if (wait_for & WAIT_FOR_READ)
632 if (wait_for & WAIT_FOR_WRITE)
635 tmout.tv_sec = (long) maxtime;
636 tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
639 result = select (fd + 1, rd, wr, NULL, &tmout);
640 while (result < 0 && errno == EINTR);
645 /* Return true iff the connection to the remote site established
646 through SOCK is still open.
648 Specifically, this function returns true if SOCK is not ready for
649 reading. This is because, when the connection closes, the socket
650 is ready for reading because EOF is about to be delivered. A side
651 effect of this method is that sockets that have pending data are
652 considered non-open. This is actually a good thing for callers of
653 this function, where such pending data can only be unwanted
654 leftover from a previous request. */
657 test_socket_open (int sock)
662 /* Check if we still have a valid (non-EOF) connection. From Andrew
663 * Maholski's code in the Unix Socket FAQ. */
665 FD_ZERO (&check_set);
666 FD_SET (sock, &check_set);
668 /* Wait one microsecond */
672 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
673 /* We got a timeout, it means we're still connected. */
676 /* Read now would not wait, it means we have either pending data
681 /* Basic socket operations, mostly EINTR wrappers. */
683 #if defined(WINDOWS) || defined(MSDOS)
684 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
685 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
686 # define close(fd) closesocket (fd)
690 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
691 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
695 sock_read (int fd, char *buf, int bufsize)
699 res = read (fd, buf, bufsize);
700 while (res == -1 && errno == EINTR);
705 sock_write (int fd, char *buf, int bufsize)
709 res = write (fd, buf, bufsize);
710 while (res == -1 && errno == EINTR);
715 sock_poll (int fd, double timeout, int wait_for)
717 return select_fd (fd, timeout, wait_for);
721 sock_peek (int fd, char *buf, int bufsize)
725 res = recv (fd, buf, bufsize, MSG_PEEK);
726 while (res == -1 && errno == EINTR);
734 DEBUGP (("Closed fd %d\n", fd));
740 /* Reading and writing from the network. We build around the socket
741 (file descriptor) API, but support "extended" operations for things
742 that are not mere file descriptors under the hood, such as SSL
745 That way the user code can call fd_read(fd, ...) and we'll run read
746 or SSL_read or whatever is necessary. */
748 static struct hash_table *transport_map;
749 static unsigned int transport_map_modified_tick;
751 struct transport_info {
752 struct transport_implementation *imp;
756 /* Register the transport layer operations that will be used when
757 reading, writing, and polling FD.
759 This should be used for transport layers like SSL that piggyback on
760 sockets. FD should otherwise be a real socket, on which you can
761 call getpeername, etc. */
764 fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
766 struct transport_info *info;
768 /* The file descriptor must be non-negative to be registered.
769 Negative values are ignored by fd_close(), and -1 cannot be used as
773 info = xnew (struct transport_info);
777 transport_map = hash_table_new (0, NULL, NULL);
778 hash_table_put (transport_map, (void *)(intptr_t) fd, info);
779 ++transport_map_modified_tick;
782 /* Return context of the transport registered with
783 fd_register_transport. This assumes fd_register_transport was
784 previously called on FD. */
787 fd_transport_context (int fd)
789 struct transport_info *info = hash_table_get (transport_map, (void *)(intptr_t) fd);
793 /* When fd_read/fd_write are called multiple times in a loop, they should
794 remember the INFO pointer instead of fetching it every time. It is
795 not enough to compare FD to LAST_FD because FD might have been
796 closed and reopened. modified_tick ensures that changes to
797 transport_map will not be unnoticed.
799 This is a macro because we want the static storage variables to be
802 #define LAZY_RETRIEVE_INFO(info) do { \
803 static struct transport_info *last_info; \
804 static int last_fd = -1; \
805 static unsigned int last_tick; \
806 if (!transport_map) \
808 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
812 info = hash_table_get (transport_map, (void *)(intptr_t) fd); \
815 last_tick = transport_map_modified_tick; \
820 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
823 timeout = opt.read_timeout;
827 if (info && info->imp->poller)
828 test = info->imp->poller (fd, timeout, wf, info->ctx);
830 test = sock_poll (fd, timeout, wf);
839 /* Read no more than BUFSIZE bytes of data from FD, storing them to
840 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
841 received after that many seconds. If TIMEOUT is -1, the value of
842 opt.timeout is used for TIMEOUT. */
845 fd_read (int fd, char *buf, int bufsize, double timeout)
847 struct transport_info *info;
848 LAZY_RETRIEVE_INFO (info);
849 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
851 if (info && info->imp->reader)
852 return info->imp->reader (fd, buf, bufsize, info->ctx);
854 return sock_read (fd, buf, bufsize);
857 /* Like fd_read, except it provides a "preview" of the data that will
858 be read by subsequent calls to fd_read. Specifically, it copies no
859 more than BUFSIZE bytes of the currently available data to BUF and
860 returns the number of bytes copied. Return values and timeout
861 semantics are the same as those of fd_read.
863 CAVEAT: Do not assume that the first subsequent call to fd_read
864 will retrieve the same amount of data. Reading can return more or
865 less data, depending on the TCP implementation and other
866 circumstances. However, barring an error, it can be expected that
867 all the peeked data will eventually be read by fd_read. */
870 fd_peek (int fd, char *buf, int bufsize, double timeout)
872 struct transport_info *info;
873 LAZY_RETRIEVE_INFO (info);
874 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
876 if (info && info->imp->peeker)
877 return info->imp->peeker (fd, buf, bufsize, info->ctx);
879 return sock_peek (fd, buf, bufsize);
882 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
883 the operation aborts if no data is received after that many
884 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
888 fd_write (int fd, char *buf, int bufsize, double timeout)
891 struct transport_info *info;
892 LAZY_RETRIEVE_INFO (info);
894 /* `write' may write less than LEN bytes, thus the loop keeps trying
895 it until all was written, or an error occurred. */
899 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
901 if (info && info->imp->writer)
902 res = info->imp->writer (fd, buf, bufsize, info->ctx);
904 res = sock_write (fd, buf, bufsize);
913 /* Report the most recent error(s) on FD. This should only be called
914 after fd_* functions, such as fd_read and fd_write, and only if
915 they return a negative result. For errors coming from other calls
916 such as setsockopt or fopen, strerror should continue to be
919 If the transport doesn't support error messages or doesn't supply
920 one, strerror(errno) is returned. The returned error message
921 should not be used after fd_close has been called. */
926 /* Don't bother with LAZY_RETRIEVE_INFO, as this will only be called
927 in case of error, never in a tight loop. */
928 struct transport_info *info = NULL;
930 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
932 if (info && info->imp->errstr)
934 const char *err = info->imp->errstr (fd, info->ctx);
937 /* else, fall through and print the system error. */
939 return strerror (errno);
942 /* Close the file descriptor FD. */
947 struct transport_info *info;
951 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
952 per socket, so that particular optimization wouldn't work. */
955 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
957 if (info && info->imp->closer)
958 info->imp->closer (fd, info->ctx);
964 hash_table_remove (transport_map, (void *)(intptr_t) fd);
966 ++transport_map_modified_tick;