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 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)
337 struct address_list *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);
351 address_list_set_connected (al);
352 address_list_release (al);
356 /* The attempt to connect has failed. Continue with the loop
357 and try next address. */
359 address_list_set_faulty (al, i);
362 /* Failed to connect to any of the addresses in AL. */
364 if (address_list_connected_p (al))
366 /* We connected to AL before, but cannot do so now. That might
367 indicate that our DNS cache entry for HOST has expired. */
368 address_list_release (al);
369 al = lookup_host (host, LH_REFRESH);
372 address_list_release (al);
378 test_socket_open (int sock)
384 /* Check if we still have a valid (non-EOF) connection. From Andrew
385 * Maholski's code in the Unix Socket FAQ. */
387 FD_ZERO (&check_set);
388 FD_SET (sock, &check_set);
390 /* Wait one microsecond */
394 /* If we get a timeout, then that means still connected */
395 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
397 /* Connection is valid (not EOF), so continue */
403 /* Without select, it's hard to know for sure. */
408 /* Create a socket, bind it to local interface BIND_ADDRESS on port
409 *PORT, set up a listen backlog, and return the resulting socket, or
412 BIND_ADDRESS is the address of the interface to bind to. If it is
413 NULL, the socket is bound to the default address. PORT should
414 point to the port number that will be used for the binding. If
415 that number is 0, the system will choose a suitable port, and the
416 chosen value will be written to *PORT.
418 Calling accept() on such a socket waits for and accepts incoming
422 bind_local (const ip_address *bind_address, int *port)
425 int family = AF_INET;
426 struct sockaddr_storage ss;
427 struct sockaddr *sa = (struct sockaddr *)&ss;
429 /* For setting options with setsockopt. */
431 void *setopt_ptr = (void *)&setopt_val;
432 socklen_t setopt_size = sizeof (setopt_val);
435 if (bind_address->type == IPV6_ADDRESS)
439 sock = socket (family, SOCK_STREAM, 0);
444 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
448 # ifdef HAVE_IPV6_V6ONLY
449 if (family == AF_INET6)
450 setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, setopt_ptr, setopt_size);
455 sockaddr_set_data (sa, bind_address, *port);
456 if (bind (sock, sa, sockaddr_size (sa)) < 0)
461 DEBUGP (("Local socket fd %d bound.\n", sock));
463 /* If *PORT is 0, find out which port we've bound to. */
466 socklen_t sa_len = sockaddr_size (sa);
467 if (getsockname (sock, sa, &sa_len) < 0)
469 /* If we can't find out the socket's local address ("name"),
470 something is seriously wrong with the socket, and it's
471 unusable for us anyway because we must know the chosen
476 sockaddr_get_data (sa, NULL, port);
477 DEBUGP (("binding to address %s using port %i.\n",
478 pretty_print_address (bind_address), *port));
480 if (listen (sock, 1) < 0)
488 /* Like a call to accept(), but with the added check for timeout.
490 In other words, accept a client connection on LOCAL_SOCK, and
491 return the new socket used for communication with the client.
492 LOCAL_SOCK should have been bound, e.g. using bind_local().
494 The caller is blocked until a connection is established. If no
495 connection is established for opt.connect_timeout seconds, the
496 function exits with an error status. */
499 accept_connection (int local_sock)
503 /* We don't need the values provided by accept, but accept
504 apparently requires them to be present. */
505 struct sockaddr_storage ss;
506 struct sockaddr *sa = (struct sockaddr *)&ss;
507 socklen_t addrlen = sizeof (ss);
509 if (opt.connect_timeout)
511 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
517 sock = accept (local_sock, sa, &addrlen);
518 DEBUGP (("Accepted client at socket %d.\n", sock));
522 /* Get the IP address associated with the connection on FD and store
523 it to IP. Return 1 on success, 0 otherwise.
525 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
526 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
527 returns the address of the remote (peer's) side of the socket. */
530 socket_ip_address (int sock, ip_address *ip, int endpoint)
532 struct sockaddr_storage storage;
533 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
534 socklen_t addrlen = sizeof (storage);
537 if (endpoint == ENDPOINT_LOCAL)
538 ret = getsockname (sock, sockaddr, &addrlen);
539 else if (endpoint == ENDPOINT_PEER)
540 ret = getpeername (sock, sockaddr, &addrlen);
546 switch (sockaddr->sa_family)
551 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
552 ip->type = IPV6_ADDRESS;
553 ADDRESS_IPV6_IN6_ADDR (ip) = sa6->sin6_addr;
554 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
555 ADDRESS_IPV6_SCOPE (ip) = sa6->sin6_scope_id;
557 DEBUGP (("conaddr is: %s\n", pretty_print_address (ip)));
563 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
564 ip->type = IPV4_ADDRESS;
565 ADDRESS_IPV4_IN_ADDR (ip) = sa->sin_addr;
566 DEBUGP (("conaddr is: %s\n", pretty_print_address (ip)));
576 /* Return non-zero if the error from the connect code can be
577 considered retryable. Wget normally retries after errors, but the
578 exception are the "unsupported protocol" type errors (possible on
579 IPv4/IPv6 dual family systems) and "connection refused". */
582 retryable_socket_connect_error (int err)
584 /* Have to guard against some of these values not being defined.
585 Cannot use a switch statement because some of the values might be
589 || err == EAFNOSUPPORT
592 || err == EPFNOSUPPORT
594 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
595 || err == ESOCKTNOSUPPORT
597 #ifdef EPROTONOSUPPORT
598 || err == EPROTONOSUPPORT
601 || err == ENOPROTOOPT
603 /* Apparently, older versions of Linux and BSD used EINVAL
604 instead of EAFNOSUPPORT and such. */
609 if (err == ECONNREFUSED && !opt.retry_connrefused)
615 /* Return non-zero if the INET6 socket family is supported on the
618 This doesn't guarantee that we're able to connect to IPv6 hosts,
619 but it's better than nothing. It is only used on systems where
620 getaddrinfo doesn't support AI_ADDRCONFIG. (See lookup_host.) */
623 socket_has_inet6 (void)
625 static int supported = -1;
628 int sock = socket (AF_INET6, SOCK_STREAM, 0);
640 /* Wait for a single descriptor to become available, timing out after
641 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
642 -1 for error. The argument WAIT_FOR can be a combination of
643 WAIT_FOR_READ and WAIT_FOR_WRITE.
645 This is a mere convenience wrapper around the select call, and
646 should be taken as such (for example, it doesn't implement Wget's
647 0-timeout-means-no-timeout semantics.) */
650 select_fd (int fd, double maxtime, int wait_for)
654 fd_set *rd = NULL, *wr = NULL;
655 struct timeval tmout;
660 if (wait_for & WAIT_FOR_READ)
662 if (wait_for & WAIT_FOR_WRITE)
665 tmout.tv_sec = (long) maxtime;
666 tmout.tv_usec = 1000000L * (maxtime - (long) maxtime);
669 result = select (fd + 1, rd, wr, NULL, &tmout);
670 while (result < 0 && errno == EINTR);
674 #else /* not HAVE_SELECT */
676 /* If select() unavailable, just return 1. In most usages in Wget,
677 this is the appropriate response -- "if we can't poll, go ahead
678 with the blocking operation". If a specific part of code needs
679 different behavior, it can use #ifdef HAVE_SELECT to test whether
680 polling really occurs. */
683 #endif /* not HAVE_SELECT */
686 /* Basic socket operations, mostly EINTR wrappers. */
689 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
690 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
691 # define close(fd) closesocket (fd)
695 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
696 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
700 sock_read (int fd, char *buf, int bufsize)
704 res = read (fd, buf, bufsize);
705 while (res == -1 && errno == EINTR);
710 sock_write (int fd, char *buf, int bufsize)
714 res = write (fd, buf, bufsize);
715 while (res == -1 && errno == EINTR);
720 sock_poll (int fd, double timeout, int wait_for)
722 return select_fd (fd, timeout, wait_for);
729 DEBUGP (("Closed fd %d\n", fd));
735 /* Reading and writing from the network. We build around the socket
736 (file descriptor) API, but support "extended" operations for things
737 that are not mere file descriptors under the hood, such as SSL
740 That way the user code can call xread(fd, ...) and we'll run read
741 or SSL_read or whatever is necessary. */
743 static struct hash_table *transport_map;
744 static int transport_map_modified_tick;
746 struct transport_info {
754 /* Register the transport layer operations that will be used when
755 reading, writing, and polling FD.
757 This should be used for transport layers like SSL that piggyback on
758 sockets. FD should otherwise be a real socket, on which you can
759 call getpeername, etc. */
762 register_transport (int fd, xreader_t reader, xwriter_t writer,
763 xpoller_t poller, xcloser_t closer, void *ctx)
765 struct transport_info *info;
767 /* The file descriptor must be non-negative to be registered.
768 Negative values are ignored by xclose(), and -1 cannot be used as
772 info = xnew (struct transport_info);
773 info->reader = reader;
774 info->writer = writer;
775 info->poller = poller;
776 info->closer = closer;
779 transport_map = hash_table_new (0, NULL, NULL);
780 hash_table_put (transport_map, (void *) fd, info);
781 ++transport_map_modified_tick;
784 /* When xread/xwrite are called multiple times in a loop, they should
785 remember the INFO pointer instead of fetching it every time. It is
786 not enough to compare FD to LAST_FD because FD might have been
787 closed and reopened. modified_tick ensures that changes to
788 transport_map will not be unnoticed.
790 This is a macro because we want the static storage variables to be
793 #define LAZY_RETRIEVE_INFO(info) do { \
794 static struct transport_info *last_info; \
795 static int last_fd = -1, last_tick; \
796 if (!transport_map) \
798 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
802 info = hash_table_get (transport_map, (void *) fd); \
805 last_tick = transport_map_modified_tick; \
809 /* Read no more than BUFSIZE bytes of data from FD, storing them to
810 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
811 received after that many seconds. If TIMEOUT is -1, the value of
812 opt.timeout is used for TIMEOUT. */
815 xread (int fd, char *buf, int bufsize, double timeout)
817 struct transport_info *info;
818 LAZY_RETRIEVE_INFO (info);
820 timeout = opt.read_timeout;
824 if (info && info->poller)
825 test = info->poller (fd, timeout, WAIT_FOR_READ, info->ctx);
827 test = sock_poll (fd, timeout, WAIT_FOR_READ);
833 if (info && info->reader)
834 return info->reader (fd, buf, bufsize, info->ctx);
836 return sock_read (fd, buf, bufsize);
839 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
840 the operation aborts if no data is received after that many
841 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
845 xwrite (int fd, char *buf, int bufsize, double timeout)
848 struct transport_info *info;
849 LAZY_RETRIEVE_INFO (info);
851 timeout = opt.read_timeout;
853 /* `write' may write less than LEN bytes, thus the loop keeps trying
854 it until all was written, or an error occurred. */
861 if (info && info->poller)
862 test = info->poller (fd, timeout, WAIT_FOR_WRITE, info->ctx);
864 test = sock_poll (fd, timeout, WAIT_FOR_WRITE);
870 if (info && info->writer)
871 res = info->writer (fd, buf, bufsize, info->ctx);
873 res = sock_write (fd, buf, bufsize);
882 /* Close the file descriptor FD. */
887 struct transport_info *info;
891 /* Don't use LAZY_RETRIEVE_INFO because xclose() is only called once
892 per socket, so that particular optimization wouldn't work. */
895 info = hash_table_get (transport_map, (void *) fd);
897 if (info && info->closer)
898 info->closer (fd, info->ctx);
904 hash_table_remove (transport_map, (void *) fd);
906 ++transport_map_modified_tick;