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);
178 resolve_bind_address (struct sockaddr *sa)
180 struct address_list *al;
182 /* Make sure this is called only once. opt.bind_address doesn't
183 change during a Wget run. */
184 static int called, should_bind;
185 static ip_address ip;
189 sockaddr_set_data (sa, &ip, 0);
194 al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
197 /* #### We should be able to print the error message here. */
198 logprintf (LOG_NOTQUIET,
199 _("%s: unable to resolve bind address `%s'; disabling bind.\n"),
200 exec_name, opt.bind_address);
205 /* Pick the first address in the list and use it as bind address.
206 Perhaps we should try multiple addresses in succession, but I
207 don't think that's necessary in practice. */
208 ip = *address_list_address_at (al, 0);
209 address_list_release (al);
211 sockaddr_set_data (sa, &ip, 0);
218 const struct sockaddr *addr;
224 connect_with_timeout_callback (void *arg)
226 struct cwt_context *ctx = (struct cwt_context *)arg;
227 ctx->result = connect (ctx->fd, ctx->addr, ctx->addrlen);
230 /* Like connect, but specifies a timeout. If connecting takes longer
231 than TIMEOUT seconds, -1 is returned and errno is set to
235 connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
238 struct cwt_context ctx;
241 ctx.addrlen = addrlen;
243 if (run_with_timeout (timeout, connect_with_timeout_callback, &ctx))
248 if (ctx.result == -1 && errno == EINTR)
253 /* Connect via TCP to the specified address and port.
255 If PRINT is non-NULL, it is the host name to print that we're
259 connect_to_ip (const ip_address *ip, int port, const char *print)
261 struct sockaddr_storage ss;
262 struct sockaddr *sa = (struct sockaddr *)&ss;
265 /* If PRINT is non-NULL, print the "Connecting to..." line, with
266 PRINT being the host name we're connecting to. */
269 const char *txt_addr = pretty_print_address (ip);
270 if (print && 0 != strcmp (print, txt_addr))
271 logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
272 escnonprint (print), txt_addr, port);
274 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
277 /* Store the sockaddr info to SA. */
278 sockaddr_set_data (sa, ip, port);
280 /* Create the socket of the family appropriate for the address. */
281 sock = socket (sa->sa_family, SOCK_STREAM, 0);
285 #if defined(ENABLE_IPV6) && defined(IPV6_V6ONLY)
288 /* In case of error, we will go on anyway... */
289 int err = setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof (on));
292 DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
297 /* For very small rate limits, set the buffer size (and hence,
298 hopefully, the kernel's TCP window size) to the per-second limit.
299 That way we should never have to sleep for more than 1s between
301 if (opt.limit_rate && opt.limit_rate < 8192)
303 int bufsize = opt.limit_rate;
305 bufsize = 512; /* avoid pathologically small values */
307 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
308 (void *)&bufsize, (socklen_t)sizeof (bufsize));
310 /* When we add limit_rate support for writing, which is useful
311 for POST, we should also set SO_SNDBUF here. */
314 if (opt.bind_address)
316 /* Bind the client side of the socket to the requested
318 struct sockaddr_storage bind_ss;
319 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
320 if (resolve_bind_address (bind_sa))
322 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
327 /* Connect the socket to the remote endpoint. */
328 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
329 opt.connect_timeout) < 0)
335 logprintf (LOG_VERBOSE, _("connected.\n"));
336 DEBUGP (("Created socket %d.\n", sock));
341 /* Protect errno from possible modifications by close and
343 int save_errno = errno;
347 logprintf (LOG_VERBOSE, "failed: %s.\n", strerror (errno));
353 /* Connect via TCP to a remote host on the specified port.
355 HOST is resolved as an Internet host name. If HOST resolves to
356 more than one IP address, they are tried in the order returned by
357 DNS until connecting to one of them succeeds. */
360 connect_to_host (const char *host, int port)
365 struct address_list *al = lookup_host (host, 0);
371 address_list_get_bounds (al, &start, &end);
372 for (i = start; i < end; i++)
374 const ip_address *ip = address_list_address_at (al, i);
375 sock = connect_to_ip (ip, port, host);
379 address_list_set_connected (al);
380 address_list_release (al);
384 /* The attempt to connect has failed. Continue with the loop
385 and try next address. */
387 address_list_set_faulty (al, i);
390 /* Failed to connect to any of the addresses in AL. */
392 if (address_list_connected_p (al))
394 /* We connected to AL before, but cannot do so now. That might
395 indicate that our DNS cache entry for HOST has expired. */
396 address_list_release (al);
397 al = lookup_host (host, LH_REFRESH);
400 address_list_release (al);
405 /* Create a socket, bind it to local interface BIND_ADDRESS on port
406 *PORT, set up a listen backlog, and return the resulting socket, or
409 BIND_ADDRESS is the address of the interface to bind to. If it is
410 NULL, the socket is bound to the default address. PORT should
411 point to the port number that will be used for the binding. If
412 that number is 0, the system will choose a suitable port, and the
413 chosen value will be written to *PORT.
415 Calling accept() on such a socket waits for and accepts incoming
419 bind_local (const ip_address *bind_address, int *port)
422 int family = AF_INET;
423 struct sockaddr_storage ss;
424 struct sockaddr *sa = (struct sockaddr *)&ss;
426 /* For setting options with setsockopt. */
428 void *setopt_ptr = (void *)&setopt_val;
429 socklen_t setopt_size = sizeof (setopt_val);
432 if (bind_address->type == IPV6_ADDRESS)
436 sock = socket (family, SOCK_STREAM, 0);
441 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
445 sockaddr_set_data (sa, bind_address, *port);
446 if (bind (sock, sa, sockaddr_size (sa)) < 0)
451 DEBUGP (("Local socket fd %d bound.\n", sock));
453 /* If *PORT is 0, find out which port we've bound to. */
456 socklen_t addrlen = sockaddr_size (sa);
457 if (getsockname (sock, sa, &addrlen) < 0)
459 /* If we can't find out the socket's local address ("name"),
460 something is seriously wrong with the socket, and it's
461 unusable for us anyway because we must know the chosen
466 sockaddr_get_data (sa, NULL, port);
467 DEBUGP (("binding to address %s using port %i.\n",
468 pretty_print_address (bind_address), *port));
470 if (listen (sock, 1) < 0)
478 /* Like a call to accept(), but with the added check for timeout.
480 In other words, accept a client connection on LOCAL_SOCK, and
481 return the new socket used for communication with the client.
482 LOCAL_SOCK should have been bound, e.g. using bind_local().
484 The caller is blocked until a connection is established. If no
485 connection is established for opt.connect_timeout seconds, the
486 function exits with an error status. */
489 accept_connection (int local_sock)
493 /* We don't need the values provided by accept, but accept
494 apparently requires them to be present. */
495 struct sockaddr_storage ss;
496 struct sockaddr *sa = (struct sockaddr *)&ss;
497 socklen_t addrlen = sizeof (ss);
499 if (opt.connect_timeout)
501 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
507 sock = accept (local_sock, sa, &addrlen);
508 DEBUGP (("Accepted client at socket %d.\n", sock));
512 /* Get the IP address associated with the connection on FD and store
513 it to IP. Return 1 on success, 0 otherwise.
515 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
516 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
517 returns the address of the remote (peer's) side of the socket. */
520 socket_ip_address (int sock, ip_address *ip, int endpoint)
522 struct sockaddr_storage storage;
523 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
524 socklen_t addrlen = sizeof (storage);
527 if (endpoint == ENDPOINT_LOCAL)
528 ret = getsockname (sock, sockaddr, &addrlen);
529 else if (endpoint == ENDPOINT_PEER)
530 ret = getpeername (sock, sockaddr, &addrlen);
536 switch (sockaddr->sa_family)
541 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
542 ip->type = IPV6_ADDRESS;
543 ADDRESS_IPV6_IN6_ADDR (ip) = sa6->sin6_addr;
544 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
545 ADDRESS_IPV6_SCOPE (ip) = sa6->sin6_scope_id;
547 DEBUGP (("conaddr is: %s\n", pretty_print_address (ip)));
553 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
554 ip->type = IPV4_ADDRESS;
555 ADDRESS_IPV4_IN_ADDR (ip) = sa->sin_addr;
556 DEBUGP (("conaddr is: %s\n", pretty_print_address (ip)));
564 /* Return non-zero if the error from the connect code can be
565 considered retryable. Wget normally retries after errors, but the
566 exception are the "unsupported protocol" type errors (possible on
567 IPv4/IPv6 dual family systems) and "connection refused". */
570 retryable_socket_connect_error (int err)
572 /* Have to guard against some of these values not being defined.
573 Cannot use a switch statement because some of the values might be
577 || err == EAFNOSUPPORT
580 || err == EPFNOSUPPORT
582 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
583 || err == ESOCKTNOSUPPORT
585 #ifdef EPROTONOSUPPORT
586 || err == EPROTONOSUPPORT
589 || err == ENOPROTOOPT
591 /* Apparently, older versions of Linux and BSD used EINVAL
592 instead of EAFNOSUPPORT and such. */
597 if (!opt.retry_connrefused)
598 if (err == ECONNREFUSED
600 || err == ENETUNREACH /* network is unreachable */
603 || err == EHOSTUNREACH /* host is unreachable */
612 # ifndef HAVE_GETADDRINFO_AI_ADDRCONFIG
614 /* Return non-zero if the INET6 socket family is supported on the
617 This doesn't guarantee that we're able to connect to IPv6 hosts,
618 but it's better than nothing. It is only used on systems where
619 getaddrinfo doesn't support AI_ADDRCONFIG. (See lookup_host.) */
622 socket_has_inet6 (void)
624 static int supported = -1;
627 int sock = socket (AF_INET6, SOCK_STREAM, 0);
639 # endif/* not HAVE_GETADDRINFO_AI_ADDRCONFIG */
640 #endif /* ENABLE_IPV6 */
642 /* Wait for a single descriptor to become available, timing out after
643 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
644 -1 for error. The argument WAIT_FOR can be a combination of
645 WAIT_FOR_READ and WAIT_FOR_WRITE.
647 This is a mere convenience wrapper around the select call, and
648 should be taken as such (for example, it doesn't implement Wget's
649 0-timeout-means-no-timeout semantics.) */
652 select_fd (int fd, double maxtime, int wait_for)
656 fd_set *rd = NULL, *wr = NULL;
657 struct timeval tmout;
662 if (wait_for & WAIT_FOR_READ)
664 if (wait_for & WAIT_FOR_WRITE)
667 tmout.tv_sec = (long) maxtime;
668 tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
671 result = select (fd + 1, rd, wr, NULL, &tmout);
672 while (result < 0 && errno == EINTR);
676 #else /* not HAVE_SELECT */
678 /* If select() unavailable, just return 1. In most usages in Wget,
679 this is the appropriate response -- "if we can't poll, go ahead
680 with the blocking operation". If a specific part of code needs
681 different behavior, it can use #ifdef HAVE_SELECT to test whether
682 polling really occurs. */
685 #endif /* not HAVE_SELECT */
689 test_socket_open (int sock)
695 /* Check if we still have a valid (non-EOF) connection. From Andrew
696 * Maholski's code in the Unix Socket FAQ. */
698 FD_ZERO (&check_set);
699 FD_SET (sock, &check_set);
701 /* Wait one microsecond */
705 /* If we get a timeout, then that means still connected */
706 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
708 /* Connection is valid (not EOF), so continue */
714 /* Without select, it's hard to know for sure. */
719 /* Basic socket operations, mostly EINTR wrappers. */
722 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
723 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
724 # define close(fd) closesocket (fd)
728 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
729 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
733 sock_read (int fd, char *buf, int bufsize)
737 res = read (fd, buf, bufsize);
738 while (res == -1 && errno == EINTR);
743 sock_write (int fd, char *buf, int bufsize)
747 res = write (fd, buf, bufsize);
748 while (res == -1 && errno == EINTR);
753 sock_poll (int fd, double timeout, int wait_for)
755 return select_fd (fd, timeout, wait_for);
759 sock_peek (int fd, char *buf, int bufsize)
763 res = recv (fd, buf, bufsize, MSG_PEEK);
764 while (res == -1 && errno == EINTR);
772 DEBUGP (("Closed fd %d\n", fd));
778 /* Reading and writing from the network. We build around the socket
779 (file descriptor) API, but support "extended" operations for things
780 that are not mere file descriptors under the hood, such as SSL
783 That way the user code can call fd_read(fd, ...) and we'll run read
784 or SSL_read or whatever is necessary. */
786 static struct hash_table *transport_map;
787 static int transport_map_modified_tick;
789 struct transport_info {
798 /* Register the transport layer operations that will be used when
799 reading, writing, and polling FD.
801 This should be used for transport layers like SSL that piggyback on
802 sockets. FD should otherwise be a real socket, on which you can
803 call getpeername, etc. */
806 fd_register_transport (int fd, fd_reader_t reader, fd_writer_t writer,
807 fd_poller_t poller, fd_peeker_t peeker,
808 fd_closer_t closer, void *ctx)
810 struct transport_info *info;
812 /* The file descriptor must be non-negative to be registered.
813 Negative values are ignored by fd_close(), and -1 cannot be used as
817 info = xnew (struct transport_info);
818 info->reader = reader;
819 info->writer = writer;
820 info->poller = poller;
821 info->peeker = peeker;
822 info->closer = closer;
825 transport_map = hash_table_new (0, NULL, NULL);
826 hash_table_put (transport_map, (void *) fd, info);
827 ++transport_map_modified_tick;
830 /* When fd_read/fd_write are called multiple times in a loop, they should
831 remember the INFO pointer instead of fetching it every time. It is
832 not enough to compare FD to LAST_FD because FD might have been
833 closed and reopened. modified_tick ensures that changes to
834 transport_map will not be unnoticed.
836 This is a macro because we want the static storage variables to be
839 #define LAZY_RETRIEVE_INFO(info) do { \
840 static struct transport_info *last_info; \
841 static int last_fd = -1, last_tick; \
842 if (!transport_map) \
844 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
848 info = hash_table_get (transport_map, (void *) fd); \
851 last_tick = transport_map_modified_tick; \
856 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
859 timeout = opt.read_timeout;
863 if (info && info->poller)
864 test = info->poller (fd, timeout, wf, info->ctx);
866 test = sock_poll (fd, timeout, wf);
875 /* Read no more than BUFSIZE bytes of data from FD, storing them to
876 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
877 received after that many seconds. If TIMEOUT is -1, the value of
878 opt.timeout is used for TIMEOUT. */
881 fd_read (int fd, char *buf, int bufsize, double timeout)
883 struct transport_info *info;
884 LAZY_RETRIEVE_INFO (info);
885 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
887 if (info && info->reader)
888 return info->reader (fd, buf, bufsize, info->ctx);
890 return sock_read (fd, buf, bufsize);
893 /* Like fd_read, except it provides a "preview" of the data that will
894 be read by subsequent calls to fd_read. Specifically, it copies no
895 more than BUFSIZE bytes of the currently available data to BUF and
896 returns the number of bytes copied. Return values and timeout
897 semantics are the same as those of fd_read.
899 CAVEAT: Do not assume that the first subsequent call to fd_read
900 will retrieve the same amount of data. Reading can return more or
901 less data, depending on the TCP implementation and other
902 circumstances. However, barring an error, it can be expected that
903 all the peeked data will eventually be read by fd_read. */
906 fd_peek (int fd, char *buf, int bufsize, double timeout)
908 struct transport_info *info;
909 LAZY_RETRIEVE_INFO (info);
910 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
912 if (info && info->peeker)
913 return info->peeker (fd, buf, bufsize, info->ctx);
915 return sock_peek (fd, buf, bufsize);
918 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
919 the operation aborts if no data is received after that many
920 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
924 fd_write (int fd, char *buf, int bufsize, double timeout)
927 struct transport_info *info;
928 LAZY_RETRIEVE_INFO (info);
930 /* `write' may write less than LEN bytes, thus the loop keeps trying
931 it until all was written, or an error occurred. */
935 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
937 if (info && info->writer)
938 res = info->writer (fd, buf, bufsize, info->ctx);
940 res = sock_write (fd, buf, bufsize);
949 /* Close the file descriptor FD. */
954 struct transport_info *info;
958 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
959 per socket, so that particular optimization wouldn't work. */
962 info = hash_table_get (transport_map, (void *) fd);
964 if (info && info->closer)
965 info->closer (fd, info->ctx);
971 hash_table_remove (transport_map, (void *) fd);
973 ++transport_map_modified_tick;