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 (opt.bind_address, LH_BIND | LH_SILENT);
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 via TCP to the specified address and port.
246 If PRINT is non-NULL, it is the host name to print that we're
250 connect_to_ip (const ip_address *ip, int port, const char *print)
252 struct sockaddr_storage ss;
253 struct sockaddr *sa = (struct sockaddr *)&ss;
256 /* If PRINT is non-NULL, print the "Connecting to..." line, with
257 PRINT being the host name we're connecting to. */
260 const char *txt_addr = pretty_print_address (ip);
261 if (print && 0 != strcmp (print, txt_addr))
262 logprintf (LOG_VERBOSE,
263 _("Connecting to %s|%s|:%d... "), print, txt_addr, port);
265 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
268 /* Store the sockaddr info to SA. */
269 sockaddr_set_data (sa, ip, port);
271 /* Create the socket of the family appropriate for the address. */
272 sock = socket (sa->sa_family, SOCK_STREAM, 0);
276 /* For very small rate limits, set the buffer size (and hence,
277 hopefully, the kernel's TCP window size) to the per-second limit.
278 That way we should never have to sleep for more than 1s between
280 if (opt.limit_rate && opt.limit_rate < 8192)
282 int bufsize = opt.limit_rate;
284 bufsize = 512; /* avoid pathologically small values */
286 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
287 (void *)&bufsize, (socklen_t)sizeof (bufsize));
289 /* When we add limit_rate support for writing, which is useful
290 for POST, we should also set SO_SNDBUF here. */
293 if (opt.bind_address)
295 /* Bind the client side of the socket to the requested
297 struct sockaddr_storage bind_ss;
298 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
299 if (resolve_bind_address (bind_sa))
301 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
306 /* Connect the socket to the remote endpoint. */
307 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
308 opt.connect_timeout) < 0)
314 logprintf (LOG_VERBOSE, _("connected.\n"));
315 DEBUGP (("Created socket %d.\n", sock));
320 /* Protect errno from possible modifications by close and
322 int save_errno = errno;
326 logprintf (LOG_VERBOSE, "failed: %s.\n", strerror (errno));
332 /* Connect via TCP to a remote host on the specified port.
334 HOST is resolved as an Internet host name. If HOST resolves to
335 more than one IP address, they are tried in the order returned by
336 DNS until connecting to one of them succeeds. */
339 connect_to_host (const char *host, int port)
344 struct address_list *al = lookup_host (host, 0);
350 address_list_get_bounds (al, &start, &end);
351 for (i = start; i < end; i++)
353 const ip_address *ip = address_list_address_at (al, i);
354 sock = connect_to_ip (ip, port, host);
358 address_list_set_connected (al);
359 address_list_release (al);
363 /* The attempt to connect has failed. Continue with the loop
364 and try next address. */
366 address_list_set_faulty (al, i);
369 /* Failed to connect to any of the addresses in AL. */
371 if (address_list_connected_p (al))
373 /* We connected to AL before, but cannot do so now. That might
374 indicate that our DNS cache entry for HOST has expired. */
375 address_list_release (al);
376 al = lookup_host (host, LH_REFRESH);
379 address_list_release (al);
384 /* Create a socket, bind it to local interface BIND_ADDRESS on port
385 *PORT, set up a listen backlog, and return the resulting socket, or
388 BIND_ADDRESS is the address of the interface to bind to. If it is
389 NULL, the socket is bound to the default address. PORT should
390 point to the port number that will be used for the binding. If
391 that number is 0, the system will choose a suitable port, and the
392 chosen value will be written to *PORT.
394 Calling accept() on such a socket waits for and accepts incoming
398 bind_local (const ip_address *bind_address, int *port)
401 int family = AF_INET;
402 struct sockaddr_storage ss;
403 struct sockaddr *sa = (struct sockaddr *)&ss;
405 /* For setting options with setsockopt. */
407 void *setopt_ptr = (void *)&setopt_val;
408 socklen_t setopt_size = sizeof (setopt_val);
411 if (bind_address->type == IPV6_ADDRESS)
415 sock = socket (family, SOCK_STREAM, 0);
420 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
424 # ifdef HAVE_IPV6_V6ONLY
425 if (family == AF_INET6)
426 setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, setopt_ptr, setopt_size);
431 sockaddr_set_data (sa, bind_address, *port);
432 if (bind (sock, sa, sockaddr_size (sa)) < 0)
437 DEBUGP (("Local socket fd %d bound.\n", sock));
439 /* If *PORT is 0, find out which port we've bound to. */
442 socklen_t sa_len = sockaddr_size (sa);
443 if (getsockname (sock, sa, &sa_len) < 0)
445 /* If we can't find out the socket's local address ("name"),
446 something is seriously wrong with the socket, and it's
447 unusable for us anyway because we must know the chosen
452 sockaddr_get_data (sa, NULL, port);
453 DEBUGP (("binding to address %s using port %i.\n",
454 pretty_print_address (bind_address), *port));
456 if (listen (sock, 1) < 0)
464 /* Like a call to accept(), but with the added check for timeout.
466 In other words, accept a client connection on LOCAL_SOCK, and
467 return the new socket used for communication with the client.
468 LOCAL_SOCK should have been bound, e.g. using bind_local().
470 The caller is blocked until a connection is established. If no
471 connection is established for opt.connect_timeout seconds, the
472 function exits with an error status. */
475 accept_connection (int local_sock)
479 /* We don't need the values provided by accept, but accept
480 apparently requires them to be present. */
481 struct sockaddr_storage ss;
482 struct sockaddr *sa = (struct sockaddr *)&ss;
483 socklen_t addrlen = sizeof (ss);
485 if (opt.connect_timeout)
487 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
493 sock = accept (local_sock, sa, &addrlen);
494 DEBUGP (("Accepted client at socket %d.\n", sock));
498 /* Get the IP address associated with the connection on FD and store
499 it to IP. Return 1 on success, 0 otherwise.
501 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
502 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
503 returns the address of the remote (peer's) side of the socket. */
506 socket_ip_address (int sock, ip_address *ip, int endpoint)
508 struct sockaddr_storage storage;
509 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
510 socklen_t addrlen = sizeof (storage);
513 if (endpoint == ENDPOINT_LOCAL)
514 ret = getsockname (sock, sockaddr, &addrlen);
515 else if (endpoint == ENDPOINT_PEER)
516 ret = getpeername (sock, sockaddr, &addrlen);
522 switch (sockaddr->sa_family)
527 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
528 ip->type = IPV6_ADDRESS;
529 ADDRESS_IPV6_IN6_ADDR (ip) = sa6->sin6_addr;
530 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
531 ADDRESS_IPV6_SCOPE (ip) = sa6->sin6_scope_id;
533 DEBUGP (("conaddr is: %s\n", pretty_print_address (ip)));
539 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
540 ip->type = IPV4_ADDRESS;
541 ADDRESS_IPV4_IN_ADDR (ip) = sa->sin_addr;
542 DEBUGP (("conaddr is: %s\n", pretty_print_address (ip)));
552 /* Return non-zero if the error from the connect code can be
553 considered retryable. Wget normally retries after errors, but the
554 exception are the "unsupported protocol" type errors (possible on
555 IPv4/IPv6 dual family systems) and "connection refused". */
558 retryable_socket_connect_error (int err)
560 /* Have to guard against some of these values not being defined.
561 Cannot use a switch statement because some of the values might be
565 || err == EAFNOSUPPORT
568 || err == EPFNOSUPPORT
570 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
571 || err == ESOCKTNOSUPPORT
573 #ifdef EPROTONOSUPPORT
574 || err == EPROTONOSUPPORT
577 || err == ENOPROTOOPT
579 /* Apparently, older versions of Linux and BSD used EINVAL
580 instead of EAFNOSUPPORT and such. */
585 if (err == ECONNREFUSED && !opt.retry_connrefused)
592 # ifndef HAVE_GETADDRINFO_AI_ADDRCONFIG
594 /* Return non-zero if the INET6 socket family is supported on the
597 This doesn't guarantee that we're able to connect to IPv6 hosts,
598 but it's better than nothing. It is only used on systems where
599 getaddrinfo doesn't support AI_ADDRCONFIG. (See lookup_host.) */
602 socket_has_inet6 (void)
604 static int supported = -1;
607 int sock = socket (AF_INET6, SOCK_STREAM, 0);
619 # endif/* not HAVE_GETADDRINFO_AI_ADDRCONFIG */
620 #endif /* ENABLE_IPV6 */
622 /* Wait for a single descriptor to become available, timing out after
623 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
624 -1 for error. The argument WAIT_FOR can be a combination of
625 WAIT_FOR_READ and WAIT_FOR_WRITE.
627 This is a mere convenience wrapper around the select call, and
628 should be taken as such (for example, it doesn't implement Wget's
629 0-timeout-means-no-timeout semantics.) */
632 select_fd (int fd, double maxtime, int wait_for)
636 fd_set *rd = NULL, *wr = NULL;
637 struct timeval tmout;
642 if (wait_for & WAIT_FOR_READ)
644 if (wait_for & WAIT_FOR_WRITE)
647 tmout.tv_sec = (long) maxtime;
648 tmout.tv_usec = 1000000L * (maxtime - (long) maxtime);
651 result = select (fd + 1, rd, wr, NULL, &tmout);
652 while (result < 0 && errno == EINTR);
656 #else /* not HAVE_SELECT */
658 /* If select() unavailable, just return 1. In most usages in Wget,
659 this is the appropriate response -- "if we can't poll, go ahead
660 with the blocking operation". If a specific part of code needs
661 different behavior, it can use #ifdef HAVE_SELECT to test whether
662 polling really occurs. */
665 #endif /* not HAVE_SELECT */
669 test_socket_open (int sock)
675 /* Check if we still have a valid (non-EOF) connection. From Andrew
676 * Maholski's code in the Unix Socket FAQ. */
678 FD_ZERO (&check_set);
679 FD_SET (sock, &check_set);
681 /* Wait one microsecond */
685 /* If we get a timeout, then that means still connected */
686 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
688 /* Connection is valid (not EOF), so continue */
694 /* Without select, it's hard to know for sure. */
699 /* Basic socket operations, mostly EINTR wrappers. */
702 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
703 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
704 # define close(fd) closesocket (fd)
708 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
709 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
713 sock_read (int fd, char *buf, int bufsize)
717 res = read (fd, buf, bufsize);
718 while (res == -1 && errno == EINTR);
723 sock_write (int fd, char *buf, int bufsize)
727 res = write (fd, buf, bufsize);
728 while (res == -1 && errno == EINTR);
733 sock_poll (int fd, double timeout, int wait_for)
735 return select_fd (fd, timeout, wait_for);
739 sock_peek (int fd, char *buf, int bufsize)
743 res = recv (fd, buf, bufsize, MSG_PEEK);
744 while (res == -1 && errno == EINTR);
752 DEBUGP (("Closed fd %d\n", fd));
758 /* Reading and writing from the network. We build around the socket
759 (file descriptor) API, but support "extended" operations for things
760 that are not mere file descriptors under the hood, such as SSL
763 That way the user code can call fd_read(fd, ...) and we'll run read
764 or SSL_read or whatever is necessary. */
766 static struct hash_table *transport_map;
767 static int transport_map_modified_tick;
769 struct transport_info {
778 /* Register the transport layer operations that will be used when
779 reading, writing, and polling FD.
781 This should be used for transport layers like SSL that piggyback on
782 sockets. FD should otherwise be a real socket, on which you can
783 call getpeername, etc. */
786 fd_register_transport (int fd, fd_reader_t reader, fd_writer_t writer,
787 fd_poller_t poller, fd_peeker_t peeker,
788 fd_closer_t closer, void *ctx)
790 struct transport_info *info;
792 /* The file descriptor must be non-negative to be registered.
793 Negative values are ignored by fd_close(), and -1 cannot be used as
797 info = xnew (struct transport_info);
798 info->reader = reader;
799 info->writer = writer;
800 info->poller = poller;
801 info->peeker = peeker;
802 info->closer = closer;
805 transport_map = hash_table_new (0, NULL, NULL);
806 hash_table_put (transport_map, (void *) fd, info);
807 ++transport_map_modified_tick;
810 /* When fd_read/fd_write are called multiple times in a loop, they should
811 remember the INFO pointer instead of fetching it every time. It is
812 not enough to compare FD to LAST_FD because FD might have been
813 closed and reopened. modified_tick ensures that changes to
814 transport_map will not be unnoticed.
816 This is a macro because we want the static storage variables to be
819 #define LAZY_RETRIEVE_INFO(info) do { \
820 static struct transport_info *last_info; \
821 static int last_fd = -1, last_tick; \
822 if (!transport_map) \
824 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
828 info = hash_table_get (transport_map, (void *) fd); \
831 last_tick = transport_map_modified_tick; \
836 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
839 timeout = opt.read_timeout;
843 if (info && info->poller)
844 test = info->poller (fd, timeout, wf, info->ctx);
846 test = sock_poll (fd, timeout, wf);
855 /* Read no more than BUFSIZE bytes of data from FD, storing them to
856 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
857 received after that many seconds. If TIMEOUT is -1, the value of
858 opt.timeout is used for TIMEOUT. */
861 fd_read (int fd, char *buf, int bufsize, double timeout)
863 struct transport_info *info;
864 LAZY_RETRIEVE_INFO (info);
865 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
867 if (info && info->reader)
868 return info->reader (fd, buf, bufsize, info->ctx);
870 return sock_read (fd, buf, bufsize);
873 /* The same as xread, but don't actually read the data, just copy it
877 fd_peek (int fd, char *buf, int bufsize, double timeout)
879 struct transport_info *info;
880 LAZY_RETRIEVE_INFO (info);
881 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
883 if (info && info->peeker)
884 return info->peeker (fd, buf, bufsize, info->ctx);
886 return sock_peek (fd, buf, bufsize);
889 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
890 the operation aborts if no data is received after that many
891 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
895 fd_write (int fd, char *buf, int bufsize, double timeout)
898 struct transport_info *info;
899 LAZY_RETRIEVE_INFO (info);
901 /* `write' may write less than LEN bytes, thus the loop keeps trying
902 it until all was written, or an error occurred. */
906 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
908 if (info && info->writer)
909 res = info->writer (fd, buf, bufsize, info->ctx);
911 res = sock_write (fd, buf, bufsize);
920 /* Close the file descriptor FD. */
925 struct transport_info *info;
929 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
930 per socket, so that particular optimization wouldn't work. */
933 info = hash_table_get (transport_map, (void *) fd);
935 if (info && info->closer)
936 info->closer (fd, info->ctx);
942 hash_table_remove (transport_map, (void *) fd);
944 ++transport_map_modified_tick;