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);
359 if (sock < 0 && address_list_cached_p (al))
361 /* We were unable to connect to any address in a list we've
362 obtained from cache. There is a possibility that the host is
363 under dynamic DNS and has changed its address. Resolve it
365 forget_host_lookup (host);
373 test_socket_open (int sock)
379 /* Check if we still have a valid (non-EOF) connection. From Andrew
380 * Maholski's code in the Unix Socket FAQ. */
382 FD_ZERO (&check_set);
383 FD_SET (sock, &check_set);
385 /* Wait one microsecond */
389 /* If we get a timeout, then that means still connected */
390 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
392 /* Connection is valid (not EOF), so continue */
398 /* Without select, it's hard to know for sure. */
403 /* Create a socket and bind it to PORT locally. Calling accept() on
404 such a socket waits for and accepts incoming TCP connections. The
405 resulting socket is stored to LOCAL_SOCK. */
408 bindport (const ip_address *bind_address, int *port, int *local_sock)
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 if ((sock = socket (family, SOCK_STREAM, 0)) < 0)
429 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
433 # ifdef HAVE_IPV6_V6ONLY
434 if (family == AF_INET6)
435 setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, setopt_ptr, setopt_size);
440 sockaddr_set_data (sa, bind_address, *port);
441 if (bind (sock, sa, sockaddr_size (sa)) < 0)
446 DEBUGP (("Local socket fd %d bound.\n", sock));
449 socklen_t sa_len = sockaddr_size (sa);
450 if (getsockname (sock, sa, &sa_len) < 0)
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)
469 /* Wait for file descriptor FD to be readable or writable or both,
470 timing out after MAXTIME seconds. Returns 1 if FD is available, 0
471 for timeout and -1 for error. The argument WAIT_FOR can be a
472 combination of WAIT_READ and WAIT_WRITE.
474 This is a mere convenience wrapper around the select call, and
475 should be taken as such. */
478 select_fd (int fd, double maxtime, int wait_for)
481 fd_set *rd = NULL, *wr = NULL;
482 struct timeval tmout;
487 if (wait_for & WAIT_FOR_READ)
489 if (wait_for & WAIT_FOR_WRITE)
492 tmout.tv_sec = (long) maxtime;
493 tmout.tv_usec = 1000000L * (maxtime - (long) maxtime);
496 result = select (fd + 1, rd, wr, NULL, &tmout);
497 while (result < 0 && errno == EINTR);
501 #endif /* HAVE_SELECT */
503 /* Accept a connection on LOCAL_SOCK, and store the new socket to
504 *SOCK. It blocks the caller until a connection is established. If
505 no connection is established for opt.connect_timeout seconds, the
506 function exits with an error status. */
509 acceptport (int local_sock, int *sock)
511 struct sockaddr_storage ss;
512 struct sockaddr *sa = (struct sockaddr *)&ss;
513 socklen_t addrlen = sizeof (ss);
516 if (opt.connect_timeout)
517 if (select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ) <= 0)
520 if ((*sock = accept (local_sock, sa, &addrlen)) < 0)
522 DEBUGP (("Created socket fd %d.\n", *sock));
526 /* Get the IP address associated with the connection on FD and store
527 it to IP. Return 1 on success, 0 otherwise.
529 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
530 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
531 returns the address of the remote (peer's) side of the socket. */
534 socket_ip_address (int sock, ip_address *ip, int endpoint)
536 struct sockaddr_storage storage;
537 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
538 socklen_t addrlen = sizeof (storage);
541 if (endpoint == ENDPOINT_LOCAL)
542 ret = getsockname (sock, sockaddr, &addrlen);
543 else if (endpoint == ENDPOINT_PEER)
544 ret = getpeername (sock, sockaddr, &addrlen);
550 switch (sockaddr->sa_family)
555 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
556 ip->type = IPV6_ADDRESS;
557 ADDRESS_IPV6_IN6_ADDR (ip) = sa6->sin6_addr;
558 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
559 ADDRESS_IPV6_SCOPE (ip) = sa6->sin6_scope_id;
561 DEBUGP (("conaddr is: %s\n", pretty_print_address (ip)));
567 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
568 ip->type = IPV4_ADDRESS;
569 ADDRESS_IPV4_IN_ADDR (ip) = sa->sin_addr;
570 DEBUGP (("conaddr is: %s\n", pretty_print_address (ip)));
580 /* Basic socket operations, mostly EINTR wrappers. */
583 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
584 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
585 # define close(fd) closesocket (fd)
589 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
590 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
594 sock_read (int fd, char *buf, int bufsize)
598 res = read (fd, buf, bufsize);
599 while (res == -1 && errno == EINTR);
604 sock_write (int fd, char *buf, int bufsize)
608 res = write (fd, buf, bufsize);
609 while (res == -1 && errno == EINTR);
614 sock_poll (int fd, double timeout, int wait_for)
617 return select_fd (fd, timeout, wait_for);
627 DEBUGP (("Closed fd %d\n", fd));
633 /* Reading and writing from the network. We build around the socket
634 (file descriptor) API, but support "extended" operations for things
635 that are not mere file descriptors under the hood, such as SSL
638 That way the user code can call xread(fd, ...) and we'll run read
639 or SSL_read or whatever is necessary. */
641 static struct hash_table *extended_map;
642 static int extended_map_modified_tick;
644 struct extended_info {
653 register_extended (int fd, xreader_t reader, xwriter_t writer,
654 xpoller_t poller, xcloser_t closer, void *ctx)
656 struct extended_info *info;
658 /* The file descriptor must be non-negative to be registered.
659 Negative values are ignored by xclose(), and -1 cannot be used as
663 info = xnew (struct extended_info);
664 info->reader = reader;
665 info->writer = writer;
666 info->poller = poller;
667 info->closer = closer;
670 extended_map = hash_table_new (0, NULL, NULL);
671 hash_table_put (extended_map, (void *) fd, info);
672 ++extended_map_modified_tick;
675 /* When xread/xwrite are called multiple times in a loop, they should
676 remember the INFO pointer instead of fetching it every time. It is
677 not enough to compare FD to LAST_FD because FD might have been
678 closed and reopened. modified_tick ensures that changes to
679 extended_map will not be unnoticed.
681 This is a macro because we want the static storage variables to be
684 #define LAZY_RETRIEVE_INFO(info) do { \
685 static struct extended_info *last_info; \
686 static int last_fd = -1, last_tick; \
689 else if (last_fd == fd && last_tick == extended_map_modified_tick) \
693 info = hash_table_get (extended_map, (void *) fd); \
696 last_tick = extended_map_modified_tick; \
700 /* Read no more than BUFSIZE bytes of data from FD, storing them to
701 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
702 received after that many seconds. If TIMEOUT is -1, the value of
703 opt.timeout is used for TIMEOUT. */
706 xread (int fd, char *buf, int bufsize, double timeout)
708 struct extended_info *info;
709 LAZY_RETRIEVE_INFO (info);
711 timeout = opt.read_timeout;
715 if (info && info->poller)
716 test = info->poller (fd, timeout, WAIT_FOR_READ, info->ctx);
718 test = sock_poll (fd, timeout, WAIT_FOR_READ);
724 if (info && info->reader)
725 return info->reader (fd, buf, bufsize, info->ctx);
727 return sock_read (fd, buf, bufsize);
730 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
731 the operation aborts if no data is received after that many
732 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
736 xwrite (int fd, char *buf, int bufsize, double timeout)
739 struct extended_info *info;
740 LAZY_RETRIEVE_INFO (info);
742 timeout = opt.read_timeout;
744 /* `write' may write less than LEN bytes, thus the loop keeps trying
745 it until all was written, or an error occurred. */
752 if (info && info->poller)
753 test = info->poller (fd, timeout, WAIT_FOR_WRITE, info->ctx);
755 test = sock_poll (fd, timeout, WAIT_FOR_WRITE);
761 if (info && info->writer)
762 res = info->writer (fd, buf, bufsize, info->ctx);
764 res = sock_write (fd, buf, bufsize);
773 /* Close the file descriptor FD. */
778 struct extended_info *info;
782 /* We don't need to be extra-fast here, so save some code by
783 avoiding LAZY_RETRIEVE_INFO. */
786 info = hash_table_get (extended_map, (void *) fd);
788 if (info && info->closer)
789 info->closer (fd, info->ctx);
795 hash_table_remove (extended_map, (void *) fd);
797 ++extended_map_modified_tick;