1 /* Establishing and handling network connections.
2 Copyright (C) 1995, 1996, 1997, 2001, 2002 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 GNU Wget is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with Wget; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
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. */
34 #include <sys/types.h>
41 # include <sys/socket.h>
43 # include <netinet/in.h>
45 # include <arpa/inet.h>
47 #endif /* not WINDOWS */
54 #endif /* HAVE_STRING_H */
55 #ifdef HAVE_SYS_SELECT_H
56 # include <sys/select.h>
57 #endif /* HAVE_SYS_SELECT_H */
69 /* Define sockaddr_storage where unavailable (presumably on IPv4-only
73 # ifndef HAVE_STRUCT_SOCKADDR_STORAGE
74 # define sockaddr_storage sockaddr_in
76 #endif /* ENABLE_IPV6 */
78 /* Fill SA as per the data in IP and PORT. SA shoult point to struct
79 sockaddr_storage if ENABLE_IPV6 is defined, to struct sockaddr_in
83 sockaddr_set_data (struct sockaddr *sa, const ip_address *ip, int port)
89 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
91 sin->sin_family = AF_INET;
92 sin->sin_port = htons (port);
93 sin->sin_addr = ADDRESS_IPV4_IN_ADDR (ip);
99 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
101 sin6->sin6_family = AF_INET6;
102 sin6->sin6_port = htons (port);
103 sin6->sin6_addr = ADDRESS_IPV6_IN6_ADDR (ip);
104 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
105 sin6->sin6_scope_id = ADDRESS_IPV6_SCOPE (ip);
109 #endif /* ENABLE_IPV6 */
115 /* Get the data of SA, specifically the IP address and the port. If
116 you're not interested in one or the other information, pass NULL as
120 sockaddr_get_data (const struct sockaddr *sa, ip_address *ip, int *port)
122 switch (sa->sa_family)
126 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
129 ip->type = IPV4_ADDRESS;
130 ADDRESS_IPV4_IN_ADDR (ip) = sin->sin_addr;
133 *port = ntohs (sin->sin_port);
139 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
142 ip->type = IPV6_ADDRESS;
143 ADDRESS_IPV6_IN6_ADDR (ip) = sin6->sin6_addr;
144 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
145 ADDRESS_IPV6_SCOPE (ip) = sin6->sin6_scope_id;
149 *port = ntohs (sin6->sin6_port);
158 /* Return the size of the sockaddr structure depending on its
162 sockaddr_size (const struct sockaddr *sa)
164 switch (sa->sa_family)
167 return sizeof (struct sockaddr_in);
170 return sizeof (struct sockaddr_in6);
174 return 0; /* so the compiler shuts up. */
179 resolve_bind_address (struct sockaddr *sa)
181 struct address_list *al;
183 /* Make sure this is called only once. opt.bind_address doesn't
184 change during a Wget run. */
185 static int called, should_bind;
186 static ip_address ip;
190 sockaddr_set_data (sa, &ip, 0);
195 al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
198 /* #### We should be able to print the error message here. */
199 logprintf (LOG_NOTQUIET,
200 _("%s: unable to resolve bind address `%s'; disabling bind.\n"),
201 exec_name, opt.bind_address);
206 /* Pick the first address in the list and use it as bind address.
207 Perhaps we should try multiple addresses in succession, but I
208 don't think that's necessary in practice. */
209 ip = *address_list_address_at (al, 0);
210 address_list_release (al);
212 sockaddr_set_data (sa, &ip, 0);
219 const struct sockaddr *addr;
225 connect_with_timeout_callback (void *arg)
227 struct cwt_context *ctx = (struct cwt_context *)arg;
228 ctx->result = connect (ctx->fd, ctx->addr, ctx->addrlen);
231 /* Like connect, but specifies a timeout. If connecting takes longer
232 than TIMEOUT seconds, -1 is returned and errno is set to
236 connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
239 struct cwt_context ctx;
242 ctx.addrlen = addrlen;
244 if (run_with_timeout (timeout, connect_with_timeout_callback, &ctx))
249 if (ctx.result == -1 && errno == EINTR)
254 /* Connect via TCP to the specified address and port.
256 If PRINT is non-NULL, it is the host name to print that we're
260 connect_to_ip (const ip_address *ip, int port, const char *print)
262 struct sockaddr_storage ss;
263 struct sockaddr *sa = (struct sockaddr *)&ss;
266 /* If PRINT is non-NULL, print the "Connecting to..." line, with
267 PRINT being the host name we're connecting to. */
270 const char *txt_addr = pretty_print_address (ip);
271 if (print && 0 != strcmp (print, txt_addr))
272 logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
273 escnonprint (print), txt_addr, port);
275 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
278 /* Store the sockaddr info to SA. */
279 sockaddr_set_data (sa, ip, port);
281 /* Create the socket of the family appropriate for the address. */
282 sock = socket (sa->sa_family, SOCK_STREAM, 0);
286 /* For very small rate limits, set the buffer size (and hence,
287 hopefully, the kernel's TCP window size) to the per-second limit.
288 That way we should never have to sleep for more than 1s between
290 if (opt.limit_rate && opt.limit_rate < 8192)
292 int bufsize = opt.limit_rate;
294 bufsize = 512; /* avoid pathologically small values */
296 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
297 (void *)&bufsize, (socklen_t)sizeof (bufsize));
299 /* When we add limit_rate support for writing, which is useful
300 for POST, we should also set SO_SNDBUF here. */
303 if (opt.bind_address)
305 /* Bind the client side of the socket to the requested
307 struct sockaddr_storage bind_ss;
308 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
309 if (resolve_bind_address (bind_sa))
311 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
316 /* Connect the socket to the remote endpoint. */
317 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
318 opt.connect_timeout) < 0)
324 logprintf (LOG_VERBOSE, _("connected.\n"));
325 DEBUGP (("Created socket %d.\n", sock));
330 /* Protect errno from possible modifications by close and
332 int save_errno = errno;
336 logprintf (LOG_VERBOSE, "failed: %s.\n", strerror (errno));
342 /* Connect via TCP to a remote host on the specified port.
344 HOST is resolved as an Internet host name. If HOST resolves to
345 more than one IP address, they are tried in the order returned by
346 DNS until connecting to one of them succeeds. */
349 connect_to_host (const char *host, int port)
354 struct address_list *al = lookup_host (host, 0);
360 address_list_get_bounds (al, &start, &end);
361 for (i = start; i < end; i++)
363 const ip_address *ip = address_list_address_at (al, i);
364 sock = connect_to_ip (ip, port, host);
368 address_list_set_connected (al);
369 address_list_release (al);
373 /* The attempt to connect has failed. Continue with the loop
374 and try next address. */
376 address_list_set_faulty (al, i);
379 /* Failed to connect to any of the addresses in AL. */
381 if (address_list_connected_p (al))
383 /* We connected to AL before, but cannot do so now. That might
384 indicate that our DNS cache entry for HOST has expired. */
385 address_list_release (al);
386 al = lookup_host (host, LH_REFRESH);
389 address_list_release (al);
394 /* Create a socket, bind it to local interface BIND_ADDRESS on port
395 *PORT, set up a listen backlog, and return the resulting socket, or
398 BIND_ADDRESS is the address of the interface to bind to. If it is
399 NULL, the socket is bound to the default address. PORT should
400 point to the port number that will be used for the binding. If
401 that number is 0, the system will choose a suitable port, and the
402 chosen value will be written to *PORT.
404 Calling accept() on such a socket waits for and accepts incoming
408 bind_local (const ip_address *bind_address, int *port)
411 int family = AF_INET;
412 struct sockaddr_storage ss;
413 struct sockaddr *sa = (struct sockaddr *)&ss;
415 /* For setting options with setsockopt. */
417 void *setopt_ptr = (void *)&setopt_val;
418 socklen_t setopt_size = sizeof (setopt_val);
421 if (bind_address->type == IPV6_ADDRESS)
425 sock = socket (family, SOCK_STREAM, 0);
430 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
434 sockaddr_set_data (sa, bind_address, *port);
435 if (bind (sock, sa, sockaddr_size (sa)) < 0)
440 DEBUGP (("Local socket fd %d bound.\n", sock));
442 /* If *PORT is 0, find out which port we've bound to. */
445 socklen_t addrlen = sockaddr_size (sa);
446 if (getsockname (sock, sa, &addrlen) < 0)
448 /* If we can't find out the socket's local address ("name"),
449 something is seriously wrong with the socket, and it's
450 unusable for us anyway because we must know the chosen
455 sockaddr_get_data (sa, NULL, port);
456 DEBUGP (("binding to address %s using port %i.\n",
457 pretty_print_address (bind_address), *port));
459 if (listen (sock, 1) < 0)
467 /* Like a call to accept(), but with the added check for timeout.
469 In other words, accept a client connection on LOCAL_SOCK, and
470 return the new socket used for communication with the client.
471 LOCAL_SOCK should have been bound, e.g. using bind_local().
473 The caller is blocked until a connection is established. If no
474 connection is established for opt.connect_timeout seconds, the
475 function exits with an error status. */
478 accept_connection (int local_sock)
482 /* We don't need the values provided by accept, but accept
483 apparently requires them to be present. */
484 struct sockaddr_storage ss;
485 struct sockaddr *sa = (struct sockaddr *)&ss;
486 socklen_t addrlen = sizeof (ss);
488 if (opt.connect_timeout)
490 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
496 sock = accept (local_sock, sa, &addrlen);
497 DEBUGP (("Accepted client at socket %d.\n", sock));
501 /* Get the IP address associated with the connection on FD and store
502 it to IP. Return 1 on success, 0 otherwise.
504 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
505 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
506 returns the address of the remote (peer's) side of the socket. */
509 socket_ip_address (int sock, ip_address *ip, int endpoint)
511 struct sockaddr_storage storage;
512 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
513 socklen_t addrlen = sizeof (storage);
516 if (endpoint == ENDPOINT_LOCAL)
517 ret = getsockname (sock, sockaddr, &addrlen);
518 else if (endpoint == ENDPOINT_PEER)
519 ret = getpeername (sock, sockaddr, &addrlen);
525 switch (sockaddr->sa_family)
530 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
531 ip->type = IPV6_ADDRESS;
532 ADDRESS_IPV6_IN6_ADDR (ip) = sa6->sin6_addr;
533 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
534 ADDRESS_IPV6_SCOPE (ip) = sa6->sin6_scope_id;
536 DEBUGP (("conaddr is: %s\n", pretty_print_address (ip)));
542 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
543 ip->type = IPV4_ADDRESS;
544 ADDRESS_IPV4_IN_ADDR (ip) = sa->sin_addr;
545 DEBUGP (("conaddr is: %s\n", pretty_print_address (ip)));
555 /* Return non-zero if the error from the connect code can be
556 considered retryable. Wget normally retries after errors, but the
557 exception are the "unsupported protocol" type errors (possible on
558 IPv4/IPv6 dual family systems) and "connection refused". */
561 retryable_socket_connect_error (int err)
563 /* Have to guard against some of these values not being defined.
564 Cannot use a switch statement because some of the values might be
568 || err == EAFNOSUPPORT
571 || err == EPFNOSUPPORT
573 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
574 || err == ESOCKTNOSUPPORT
576 #ifdef EPROTONOSUPPORT
577 || err == EPROTONOSUPPORT
580 || err == ENOPROTOOPT
582 /* Apparently, older versions of Linux and BSD used EINVAL
583 instead of EAFNOSUPPORT and such. */
588 if (err == ECONNREFUSED && !opt.retry_connrefused)
595 # ifndef HAVE_GETADDRINFO_AI_ADDRCONFIG
597 /* Return non-zero if the INET6 socket family is supported on the
600 This doesn't guarantee that we're able to connect to IPv6 hosts,
601 but it's better than nothing. It is only used on systems where
602 getaddrinfo doesn't support AI_ADDRCONFIG. (See lookup_host.) */
605 socket_has_inet6 (void)
607 static int supported = -1;
610 int sock = socket (AF_INET6, SOCK_STREAM, 0);
622 # endif/* not HAVE_GETADDRINFO_AI_ADDRCONFIG */
623 #endif /* ENABLE_IPV6 */
625 /* Wait for a single descriptor to become available, timing out after
626 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
627 -1 for error. The argument WAIT_FOR can be a combination of
628 WAIT_FOR_READ and WAIT_FOR_WRITE.
630 This is a mere convenience wrapper around the select call, and
631 should be taken as such (for example, it doesn't implement Wget's
632 0-timeout-means-no-timeout semantics.) */
635 select_fd (int fd, double maxtime, int wait_for)
639 fd_set *rd = NULL, *wr = NULL;
640 struct timeval tmout;
645 if (wait_for & WAIT_FOR_READ)
647 if (wait_for & WAIT_FOR_WRITE)
650 tmout.tv_sec = (long) maxtime;
651 tmout.tv_usec = 1000000L * (maxtime - (long) maxtime);
654 result = select (fd + 1, rd, wr, NULL, &tmout);
655 while (result < 0 && errno == EINTR);
659 #else /* not HAVE_SELECT */
661 /* If select() unavailable, just return 1. In most usages in Wget,
662 this is the appropriate response -- "if we can't poll, go ahead
663 with the blocking operation". If a specific part of code needs
664 different behavior, it can use #ifdef HAVE_SELECT to test whether
665 polling really occurs. */
668 #endif /* not HAVE_SELECT */
672 test_socket_open (int sock)
678 /* Check if we still have a valid (non-EOF) connection. From Andrew
679 * Maholski's code in the Unix Socket FAQ. */
681 FD_ZERO (&check_set);
682 FD_SET (sock, &check_set);
684 /* Wait one microsecond */
688 /* If we get a timeout, then that means still connected */
689 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
691 /* Connection is valid (not EOF), so continue */
697 /* Without select, it's hard to know for sure. */
702 /* Basic socket operations, mostly EINTR wrappers. */
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 int transport_map_modified_tick;
772 struct transport_info {
781 /* Register the transport layer operations that will be used when
782 reading, writing, and polling FD.
784 This should be used for transport layers like SSL that piggyback on
785 sockets. FD should otherwise be a real socket, on which you can
786 call getpeername, etc. */
789 fd_register_transport (int fd, fd_reader_t reader, fd_writer_t writer,
790 fd_poller_t poller, fd_peeker_t peeker,
791 fd_closer_t closer, void *ctx)
793 struct transport_info *info;
795 /* The file descriptor must be non-negative to be registered.
796 Negative values are ignored by fd_close(), and -1 cannot be used as
800 info = xnew (struct transport_info);
801 info->reader = reader;
802 info->writer = writer;
803 info->poller = poller;
804 info->peeker = peeker;
805 info->closer = closer;
808 transport_map = hash_table_new (0, NULL, NULL);
809 hash_table_put (transport_map, (void *) fd, info);
810 ++transport_map_modified_tick;
813 /* When fd_read/fd_write are called multiple times in a loop, they should
814 remember the INFO pointer instead of fetching it every time. It is
815 not enough to compare FD to LAST_FD because FD might have been
816 closed and reopened. modified_tick ensures that changes to
817 transport_map will not be unnoticed.
819 This is a macro because we want the static storage variables to be
822 #define LAZY_RETRIEVE_INFO(info) do { \
823 static struct transport_info *last_info; \
824 static int last_fd = -1, last_tick; \
825 if (!transport_map) \
827 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
831 info = hash_table_get (transport_map, (void *) fd); \
834 last_tick = transport_map_modified_tick; \
839 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
842 timeout = opt.read_timeout;
846 if (info && info->poller)
847 test = info->poller (fd, timeout, wf, info->ctx);
849 test = sock_poll (fd, timeout, wf);
858 /* Read no more than BUFSIZE bytes of data from FD, storing them to
859 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
860 received after that many seconds. If TIMEOUT is -1, the value of
861 opt.timeout is used for TIMEOUT. */
864 fd_read (int fd, char *buf, int bufsize, double timeout)
866 struct transport_info *info;
867 LAZY_RETRIEVE_INFO (info);
868 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
870 if (info && info->reader)
871 return info->reader (fd, buf, bufsize, info->ctx);
873 return sock_read (fd, buf, bufsize);
876 /* Like fd_read, except it provides a "preview" of the data that will
877 be read by subsequent calls to fd_read. Specifically, it copies no
878 more than BUFSIZE bytes of the currently available data to BUF and
879 returns the number of bytes copied. Return values and timeout
880 semantics are the same as those of fd_read.
882 CAVEAT: Do not assume that the first subsequent call to fd_read
883 will retrieve the same amount of data. Reading can return more or
884 less data, depending on the TCP implementation and other
885 circumstances. However, barring an error, it can be expected that
886 all the peeked data will eventually be read by fd_read. */
889 fd_peek (int fd, char *buf, int bufsize, double timeout)
891 struct transport_info *info;
892 LAZY_RETRIEVE_INFO (info);
893 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
895 if (info && info->peeker)
896 return info->peeker (fd, buf, bufsize, info->ctx);
898 return sock_peek (fd, buf, bufsize);
901 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
902 the operation aborts if no data is received after that many
903 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
907 fd_write (int fd, char *buf, int bufsize, double timeout)
910 struct transport_info *info;
911 LAZY_RETRIEVE_INFO (info);
913 /* `write' may write less than LEN bytes, thus the loop keeps trying
914 it until all was written, or an error occurred. */
918 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
920 if (info && info->writer)
921 res = info->writer (fd, buf, bufsize, info->ctx);
923 res = sock_write (fd, buf, bufsize);
932 /* Close the file descriptor FD. */
937 struct transport_info *info;
941 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
942 per socket, so that particular optimization wouldn't work. */
945 info = hash_table_get (transport_map, (void *) fd);
947 if (info && info->closer)
948 info->closer (fd, info->ctx);
954 hash_table_remove (transport_map, (void *) fd);
956 ++transport_map_modified_tick;