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 */
70 /* Fill SA as per the data in IP and PORT. SA shoult point to struct
71 sockaddr_storage if ENABLE_IPV6 is defined, to struct sockaddr_in
75 sockaddr_set_data (struct sockaddr *sa, const ip_address *ip, int port)
81 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
82 sin->sin_family = AF_INET;
83 sin->sin_port = htons (port);
84 sin->sin_addr = ADDRESS_IPV4_IN_ADDR (ip);
90 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
91 sin6->sin6_family = AF_INET6;
92 sin6->sin6_port = htons (port);
93 sin6->sin6_addr = ADDRESS_IPV6_IN6_ADDR (ip);
94 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
95 sin6->sin6_scope_id = ADDRESS_IPV6_SCOPE (ip);
99 #endif /* ENABLE_IPV6 */
105 /* Get the data of SA, specifically the IP address and the port. If
106 you're not interested in one or the other information, pass NULL as
110 sockaddr_get_data (const struct sockaddr *sa, ip_address *ip, int *port)
112 switch (sa->sa_family)
116 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
119 ip->type = IPV4_ADDRESS;
120 ADDRESS_IPV4_IN_ADDR (ip) = sin->sin_addr;
123 *port = ntohs (sin->sin_port);
129 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
132 ip->type = IPV6_ADDRESS;
133 ADDRESS_IPV6_IN6_ADDR (ip) = sin6->sin6_addr;
134 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
135 ADDRESS_IPV6_SCOPE (ip) = sin6->sin6_scope_id;
139 *port = ntohs (sin6->sin6_port);
148 /* Return the size of the sockaddr structure depending on its
152 sockaddr_size (const struct sockaddr *sa)
154 switch (sa->sa_family)
157 return sizeof (struct sockaddr_in);
160 return sizeof (struct sockaddr_in6);
164 return 0; /* so the compiler shuts up. */
169 resolve_bind_address (struct sockaddr *sa)
171 struct address_list *al;
173 /* Make sure this is called only once. opt.bind_address doesn't
174 change during a Wget run. */
175 static int called, should_bind;
176 static ip_address ip;
180 sockaddr_set_data (sa, &ip, 0);
185 al = lookup_host_passive (opt.bind_address);
188 /* #### We should be able to print the error message here. */
189 logprintf (LOG_NOTQUIET,
190 _("%s: unable to resolve bind address `%s'; disabling bind.\n"),
191 exec_name, opt.bind_address);
196 /* Pick the first address in the list and use it as bind address.
197 Perhaps we should try multiple addresses in succession, but I
198 don't think that's necessary in practice. */
199 ip = *address_list_address_at (al, 0);
200 address_list_release (al);
202 sockaddr_set_data (sa, &ip, 0);
209 const struct sockaddr *addr;
215 connect_with_timeout_callback (void *arg)
217 struct cwt_context *ctx = (struct cwt_context *)arg;
218 ctx->result = connect (ctx->fd, ctx->addr, ctx->addrlen);
221 /* Like connect, but specifies a timeout. If connecting takes longer
222 than TIMEOUT seconds, -1 is returned and errno is set to
226 connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
229 struct cwt_context ctx;
232 ctx.addrlen = addrlen;
234 if (run_with_timeout (timeout, connect_with_timeout_callback, &ctx))
239 if (ctx.result == -1 && errno == EINTR)
244 /* Connect to a remote endpoint whose IP address is known. */
247 connect_to_ip (const ip_address *ip, int port, const char *print)
249 struct sockaddr_storage ss;
250 struct sockaddr *sa = (struct sockaddr *)&ss;
253 /* If PRINT is non-NULL, print the "Connecting to..." line, with
254 PRINT being the host name we're connecting to. */
257 const char *txt_addr = pretty_print_address (ip);
258 if (print && 0 != strcmp (print, txt_addr))
259 logprintf (LOG_VERBOSE,
260 _("Connecting to %s|%s|:%d... "), print, txt_addr, port);
262 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
265 /* Store the sockaddr info to SA. */
266 sockaddr_set_data (sa, ip, port);
268 /* Create the socket of the family appropriate for the address. */
269 sock = socket (sa->sa_family, SOCK_STREAM, 0);
273 /* For very small rate limits, set the buffer size (and hence,
274 hopefully, the kernel's TCP window size) to the per-second limit.
275 That way we should never have to sleep for more than 1s between
277 if (opt.limit_rate && opt.limit_rate < 8192)
279 int bufsize = opt.limit_rate;
281 bufsize = 512; /* avoid pathologically small values */
283 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
284 (void *)&bufsize, (socklen_t)sizeof (bufsize));
286 /* When we add limit_rate support for writing, which is useful
287 for POST, we should also set SO_SNDBUF here. */
290 if (opt.bind_address)
292 /* Bind the client side of the socket to the requested
294 struct sockaddr_storage bind_ss;
295 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
296 if (resolve_bind_address (bind_sa))
298 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
303 /* Connect the socket to the remote endpoint. */
304 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
305 opt.connect_timeout) < 0)
311 logprintf (LOG_VERBOSE, _("connected.\n"));
312 DEBUGP (("Created socket %d.\n", sock));
317 /* Protect errno from possible modifications by close and
319 int save_errno = errno;
323 logprintf (LOG_VERBOSE, "failed: %s.\n", strerror (errno));
329 /* Connect to a remote endpoint specified by host name. */
332 connect_to_host (const char *host, int port)
335 struct address_list *al;
339 al = lookup_host (host, 0);
343 address_list_get_bounds (al, &start, &end);
344 for (i = start; i < end; i++)
346 const ip_address *ip = address_list_address_at (al, i);
347 sock = connect_to_ip (ip, port, host);
352 address_list_set_faulty (al, i);
354 /* The attempt to connect has failed. Continue with the loop
355 and try next address. */
357 address_list_release (al);
360 /* Mark a successful connection to one of the addresses. */
361 address_list_set_connected (al);
363 if (sock < 0 && address_list_connected_p (al))
365 /* We are unable to connect to any of HOST's addresses, although
366 we were previously able to connect to HOST. That might
367 indicate that HOST is under dynamic DNS and the addresses
368 we're connecting to have expired. Resolve it again. */
369 forget_host_lookup (host);
377 test_socket_open (int sock)
383 /* Check if we still have a valid (non-EOF) connection. From Andrew
384 * Maholski's code in the Unix Socket FAQ. */
386 FD_ZERO (&check_set);
387 FD_SET (sock, &check_set);
389 /* Wait one microsecond */
393 /* If we get a timeout, then that means still connected */
394 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
396 /* Connection is valid (not EOF), so continue */
402 /* Without select, it's hard to know for sure. */
407 /* Create a socket, bind it to local interface BIND_ADDRESS on port
408 *PORT, set up a listen backlog, and return the resulting socket, or
411 BIND_ADDRESS is the address of the interface to bind to. If it is
412 NULL, the socket is bound to the default address. PORT should
413 point to the port number that will be used for the binding. If
414 that number is 0, the system will choose a suitable port, and the
415 chosen value will be written to *PORT.
417 Calling accept() on such a socket waits for and accepts incoming
421 bind_local (const ip_address *bind_address, int *port)
424 int family = AF_INET;
425 struct sockaddr_storage ss;
426 struct sockaddr *sa = (struct sockaddr *)&ss;
428 /* For setting options with setsockopt. */
430 void *setopt_ptr = (void *)&setopt_val;
431 socklen_t setopt_size = sizeof (setopt_val);
434 if (bind_address->type == IPV6_ADDRESS)
438 sock = socket (family, SOCK_STREAM, 0);
443 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
447 # ifdef HAVE_IPV6_V6ONLY
448 if (family == AF_INET6)
449 setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, setopt_ptr, setopt_size);
454 sockaddr_set_data (sa, bind_address, *port);
455 if (bind (sock, sa, sockaddr_size (sa)) < 0)
460 DEBUGP (("Local socket fd %d bound.\n", sock));
462 /* If *PORT is 0, find out which port we've bound to. */
465 socklen_t sa_len = sockaddr_size (sa);
466 if (getsockname (sock, sa, &sa_len) < 0)
468 /* If we can't find out the socket's local address ("name"),
469 something is seriously wrong with the socket, and it's
470 unusable for us anyway because we must know the chosen
475 sockaddr_get_data (sa, NULL, port);
476 DEBUGP (("binding to address %s using port %i.\n",
477 pretty_print_address (bind_address), *port));
479 if (listen (sock, 1) < 0)
487 /* Like a call to accept(), but with the added check for timeout.
489 In other words, accept a client connection on LOCAL_SOCK, and
490 return the new socket used for communication with the client.
491 LOCAL_SOCK should have been bound, e.g. using bind_local().
493 The caller is blocked until a connection is established. If no
494 connection is established for opt.connect_timeout seconds, the
495 function exits with an error status. */
498 accept_connection (int local_sock)
502 /* We don't need the values provided by accept, but accept
503 apparently requires them to be present. */
504 struct sockaddr_storage ss;
505 struct sockaddr *sa = (struct sockaddr *)&ss;
506 socklen_t addrlen = sizeof (ss);
509 if (opt.connect_timeout)
511 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
518 sock = accept (local_sock, sa, &addrlen);
519 DEBUGP (("Accepted client at socket %d.\n", sock));
523 /* Get the IP address associated with the connection on FD and store
524 it to IP. Return 1 on success, 0 otherwise.
526 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
527 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
528 returns the address of the remote (peer's) side of the socket. */
531 socket_ip_address (int sock, ip_address *ip, int endpoint)
533 struct sockaddr_storage storage;
534 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
535 socklen_t addrlen = sizeof (storage);
538 if (endpoint == ENDPOINT_LOCAL)
539 ret = getsockname (sock, sockaddr, &addrlen);
540 else if (endpoint == ENDPOINT_PEER)
541 ret = getpeername (sock, sockaddr, &addrlen);
547 switch (sockaddr->sa_family)
552 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
553 ip->type = IPV6_ADDRESS;
554 ADDRESS_IPV6_IN6_ADDR (ip) = sa6->sin6_addr;
555 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
556 ADDRESS_IPV6_SCOPE (ip) = sa6->sin6_scope_id;
558 DEBUGP (("conaddr is: %s\n", pretty_print_address (ip)));
564 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
565 ip->type = IPV4_ADDRESS;
566 ADDRESS_IPV4_IN_ADDR (ip) = sa->sin_addr;
567 DEBUGP (("conaddr is: %s\n", pretty_print_address (ip)));
577 /* Return non-zero of the errno code passed to the function is a
578 result of an attempt to create a socket of unsupported family. */
581 unsupported_socket_family_error (int err)
583 /* Have to guard against some of these values not being defined.
584 Cannot use switch because some of the values might be equal. */
586 if (err == EAFNOSUPPORT) return 1;
589 if (err == EPFNOSUPPORT) return 1;
591 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
592 if (err == ESOCKTNOSUPPORT) return 1;
594 #ifdef EPROTONOSUPPORT
595 if (err == EPROTONOSUPPORT) return 1;
598 if (err == ENOPROTOOPT) return 1;
600 /* Apparently, older versions of Linux used EINVAL instead of
602 if (err == EINVAL) return 1;
609 /* Wait for file descriptor FD to be readable or writable or both,
610 timing out after MAXTIME seconds. Returns 1 if FD is available, 0
611 for timeout and -1 for error. The argument WAIT_FOR can be a
612 combination of WAIT_READ and WAIT_WRITE.
614 This is a mere convenience wrapper around the select call, and
615 should be taken as such. */
618 select_fd (int fd, double maxtime, int wait_for)
621 fd_set *rd = NULL, *wr = NULL;
622 struct timeval tmout;
627 if (wait_for & WAIT_FOR_READ)
629 if (wait_for & WAIT_FOR_WRITE)
632 tmout.tv_sec = (long) maxtime;
633 tmout.tv_usec = 1000000L * (maxtime - (long) maxtime);
636 result = select (fd + 1, rd, wr, NULL, &tmout);
637 while (result < 0 && errno == EINTR);
642 #endif /* HAVE_SELECT */
644 /* Basic socket operations, mostly EINTR wrappers. */
647 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
648 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
649 # define close(fd) closesocket (fd)
653 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
654 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
658 sock_read (int fd, char *buf, int bufsize)
662 res = read (fd, buf, bufsize);
663 while (res == -1 && errno == EINTR);
668 sock_write (int fd, char *buf, int bufsize)
672 res = write (fd, buf, bufsize);
673 while (res == -1 && errno == EINTR);
678 sock_poll (int fd, double timeout, int wait_for)
681 return select_fd (fd, timeout, wait_for);
691 DEBUGP (("Closed fd %d\n", fd));
697 /* Reading and writing from the network. We build around the socket
698 (file descriptor) API, but support "extended" operations for things
699 that are not mere file descriptors under the hood, such as SSL
702 That way the user code can call xread(fd, ...) and we'll run read
703 or SSL_read or whatever is necessary. */
705 static struct hash_table *extended_map;
706 static int extended_map_modified_tick;
708 struct extended_info {
716 /* Register the handlers for operations on FD. This is meant
717 primarily for transport layers like SSL that piggyback on sockets,
718 but with their own readers, writers, etc. */
721 register_extended (int fd, xreader_t reader, xwriter_t writer,
722 xpoller_t poller, xcloser_t closer, void *ctx)
724 struct extended_info *info;
726 /* The file descriptor must be non-negative to be registered.
727 Negative values are ignored by xclose(), and -1 cannot be used as
731 info = xnew (struct extended_info);
732 info->reader = reader;
733 info->writer = writer;
734 info->poller = poller;
735 info->closer = closer;
738 extended_map = hash_table_new (0, NULL, NULL);
739 hash_table_put (extended_map, (void *) fd, info);
740 ++extended_map_modified_tick;
743 /* When xread/xwrite are called multiple times in a loop, they should
744 remember the INFO pointer instead of fetching it every time. It is
745 not enough to compare FD to LAST_FD because FD might have been
746 closed and reopened. modified_tick ensures that changes to
747 extended_map will not be unnoticed.
749 This is a macro because we want the static storage variables to be
752 #define LAZY_RETRIEVE_INFO(info) do { \
753 static struct extended_info *last_info; \
754 static int last_fd = -1, last_tick; \
757 else if (last_fd == fd && last_tick == extended_map_modified_tick) \
761 info = hash_table_get (extended_map, (void *) fd); \
764 last_tick = extended_map_modified_tick; \
768 /* Read no more than BUFSIZE bytes of data from FD, storing them to
769 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
770 received after that many seconds. If TIMEOUT is -1, the value of
771 opt.timeout is used for TIMEOUT. */
774 xread (int fd, char *buf, int bufsize, double timeout)
776 struct extended_info *info;
777 LAZY_RETRIEVE_INFO (info);
779 timeout = opt.read_timeout;
783 if (info && info->poller)
784 test = info->poller (fd, timeout, WAIT_FOR_READ, info->ctx);
786 test = sock_poll (fd, timeout, WAIT_FOR_READ);
792 if (info && info->reader)
793 return info->reader (fd, buf, bufsize, info->ctx);
795 return sock_read (fd, buf, bufsize);
798 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
799 the operation aborts if no data is received after that many
800 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
804 xwrite (int fd, char *buf, int bufsize, double timeout)
807 struct extended_info *info;
808 LAZY_RETRIEVE_INFO (info);
810 timeout = opt.read_timeout;
812 /* `write' may write less than LEN bytes, thus the loop keeps trying
813 it until all was written, or an error occurred. */
820 if (info && info->poller)
821 test = info->poller (fd, timeout, WAIT_FOR_WRITE, info->ctx);
823 test = sock_poll (fd, timeout, WAIT_FOR_WRITE);
829 if (info && info->writer)
830 res = info->writer (fd, buf, bufsize, info->ctx);
832 res = sock_write (fd, buf, bufsize);
841 /* Close the file descriptor FD. */
846 struct extended_info *info;
850 /* We don't need to be extra-fast here, so save some code by
851 avoiding LAZY_RETRIEVE_INFO. */
854 info = hash_table_get (extended_map, (void *) fd);
856 if (info && info->closer)
857 info->closer (fd, info->ctx);
863 hash_table_remove (extended_map, (void *) fd);
865 ++extended_map_modified_tick;