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);
174 return 0; /* so the compiler shuts up. */
179 resolve_bind_address (struct sockaddr *sa)
181 struct address_list *al;
183 /* Make sure this is called only once. opt.bind_address doesn't
184 change during a Wget run. */
185 static int called, should_bind;
186 static ip_address ip;
190 sockaddr_set_data (sa, &ip, 0);
195 al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
198 /* #### We should be able to print the error message here. */
199 logprintf (LOG_NOTQUIET,
200 _("%s: unable to resolve bind address `%s'; disabling bind.\n"),
201 exec_name, opt.bind_address);
206 /* Pick the first address in the list and use it as bind address.
207 Perhaps we should try multiple addresses in succession, but I
208 don't think that's necessary in practice. */
209 ip = *address_list_address_at (al, 0);
210 address_list_release (al);
212 sockaddr_set_data (sa, &ip, 0);
219 const struct sockaddr *addr;
225 connect_with_timeout_callback (void *arg)
227 struct cwt_context *ctx = (struct cwt_context *)arg;
228 ctx->result = connect (ctx->fd, ctx->addr, ctx->addrlen);
231 /* Like connect, but specifies a timeout. If connecting takes longer
232 than TIMEOUT seconds, -1 is returned and errno is set to
236 connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
239 struct cwt_context ctx;
242 ctx.addrlen = addrlen;
244 if (run_with_timeout (timeout, connect_with_timeout_callback, &ctx))
249 if (ctx.result == -1 && errno == EINTR)
254 /* Connect via TCP to the specified address and port.
256 If PRINT is non-NULL, it is the host name to print that we're
260 connect_to_ip (const ip_address *ip, int port, const char *print)
262 struct sockaddr_storage ss;
263 struct sockaddr *sa = (struct sockaddr *)&ss;
266 /* If PRINT is non-NULL, print the "Connecting to..." line, with
267 PRINT being the host name we're connecting to. */
270 const char *txt_addr = pretty_print_address (ip);
271 if (print && 0 != strcmp (print, txt_addr))
272 logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
273 escnonprint (print), txt_addr, port);
275 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
278 /* Store the sockaddr info to SA. */
279 sockaddr_set_data (sa, ip, port);
281 /* Create the socket of the family appropriate for the address. */
282 sock = socket (sa->sa_family, SOCK_STREAM, 0);
286 #if defined(ENABLE_IPV6) && defined(IPV6_V6ONLY)
289 /* In case of error, we will go on anyway... */
290 int err = setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof (on));
293 DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
298 /* For very small rate limits, set the buffer size (and hence,
299 hopefully, the kernel's TCP window size) to the per-second limit.
300 That way we should never have to sleep for more than 1s between
302 if (opt.limit_rate && opt.limit_rate < 8192)
304 int bufsize = opt.limit_rate;
306 bufsize = 512; /* avoid pathologically small values */
308 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
309 (void *)&bufsize, (socklen_t)sizeof (bufsize));
311 /* When we add limit_rate support for writing, which is useful
312 for POST, we should also set SO_SNDBUF here. */
315 if (opt.bind_address)
317 /* Bind the client side of the socket to the requested
319 struct sockaddr_storage bind_ss;
320 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
321 if (resolve_bind_address (bind_sa))
323 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
328 /* Connect the socket to the remote endpoint. */
329 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
330 opt.connect_timeout) < 0)
336 logprintf (LOG_VERBOSE, _("connected.\n"));
337 DEBUGP (("Created socket %d.\n", sock));
342 /* Protect errno from possible modifications by close and
344 int save_errno = errno;
348 logprintf (LOG_VERBOSE, "failed: %s.\n", strerror (errno));
354 /* Connect via TCP to a remote host on the specified port.
356 HOST is resolved as an Internet host name. If HOST resolves to
357 more than one IP address, they are tried in the order returned by
358 DNS until connecting to one of them succeeds. */
361 connect_to_host (const char *host, int port)
366 struct address_list *al = lookup_host (host, 0);
372 address_list_get_bounds (al, &start, &end);
373 for (i = start; i < end; i++)
375 const ip_address *ip = address_list_address_at (al, i);
376 sock = connect_to_ip (ip, port, host);
380 address_list_set_connected (al);
381 address_list_release (al);
385 /* The attempt to connect has failed. Continue with the loop
386 and try next address. */
388 address_list_set_faulty (al, i);
391 /* Failed to connect to any of the addresses in AL. */
393 if (address_list_connected_p (al))
395 /* We connected to AL before, but cannot do so now. That might
396 indicate that our DNS cache entry for HOST has expired. */
397 address_list_release (al);
398 al = lookup_host (host, LH_REFRESH);
401 address_list_release (al);
406 /* Create a socket, bind it to local interface BIND_ADDRESS on port
407 *PORT, set up a listen backlog, and return the resulting socket, or
410 BIND_ADDRESS is the address of the interface to bind to. If it is
411 NULL, the socket is bound to the default address. PORT should
412 point to the port number that will be used for the binding. If
413 that number is 0, the system will choose a suitable port, and the
414 chosen value will be written to *PORT.
416 Calling accept() on such a socket waits for and accepts incoming
420 bind_local (const ip_address *bind_address, int *port)
423 int family = AF_INET;
424 struct sockaddr_storage ss;
425 struct sockaddr *sa = (struct sockaddr *)&ss;
427 /* For setting options with setsockopt. */
429 void *setopt_ptr = (void *)&setopt_val;
430 socklen_t setopt_size = sizeof (setopt_val);
433 if (bind_address->type == IPV6_ADDRESS)
437 sock = socket (family, SOCK_STREAM, 0);
442 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
446 sockaddr_set_data (sa, bind_address, *port);
447 if (bind (sock, sa, sockaddr_size (sa)) < 0)
452 DEBUGP (("Local socket fd %d bound.\n", sock));
454 /* If *PORT is 0, find out which port we've bound to. */
457 socklen_t addrlen = sockaddr_size (sa);
458 if (getsockname (sock, sa, &addrlen) < 0)
460 /* If we can't find out the socket's local address ("name"),
461 something is seriously wrong with the socket, and it's
462 unusable for us anyway because we must know the chosen
467 sockaddr_get_data (sa, NULL, port);
468 DEBUGP (("binding to address %s using port %i.\n",
469 pretty_print_address (bind_address), *port));
471 if (listen (sock, 1) < 0)
479 /* Like a call to accept(), but with the added check for timeout.
481 In other words, accept a client connection on LOCAL_SOCK, and
482 return the new socket used for communication with the client.
483 LOCAL_SOCK should have been bound, e.g. using bind_local().
485 The caller is blocked until a connection is established. If no
486 connection is established for opt.connect_timeout seconds, the
487 function exits with an error status. */
490 accept_connection (int local_sock)
494 /* We don't need the values provided by accept, but accept
495 apparently requires them to be present. */
496 struct sockaddr_storage ss;
497 struct sockaddr *sa = (struct sockaddr *)&ss;
498 socklen_t addrlen = sizeof (ss);
500 if (opt.connect_timeout)
502 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
508 sock = accept (local_sock, sa, &addrlen);
509 DEBUGP (("Accepted client at socket %d.\n", sock));
513 /* Get the IP address associated with the connection on FD and store
514 it to IP. Return 1 on success, 0 otherwise.
516 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
517 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
518 returns the address of the remote (peer's) side of the socket. */
521 socket_ip_address (int sock, ip_address *ip, int endpoint)
523 struct sockaddr_storage storage;
524 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
525 socklen_t addrlen = sizeof (storage);
528 if (endpoint == ENDPOINT_LOCAL)
529 ret = getsockname (sock, sockaddr, &addrlen);
530 else if (endpoint == ENDPOINT_PEER)
531 ret = getpeername (sock, sockaddr, &addrlen);
537 switch (sockaddr->sa_family)
542 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
543 ip->type = IPV6_ADDRESS;
544 ADDRESS_IPV6_IN6_ADDR (ip) = sa6->sin6_addr;
545 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
546 ADDRESS_IPV6_SCOPE (ip) = sa6->sin6_scope_id;
548 DEBUGP (("conaddr is: %s\n", pretty_print_address (ip)));
554 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
555 ip->type = IPV4_ADDRESS;
556 ADDRESS_IPV4_IN_ADDR (ip) = sa->sin_addr;
557 DEBUGP (("conaddr is: %s\n", pretty_print_address (ip)));
567 /* Return non-zero if the error from the connect code can be
568 considered retryable. Wget normally retries after errors, but the
569 exception are the "unsupported protocol" type errors (possible on
570 IPv4/IPv6 dual family systems) and "connection refused". */
573 retryable_socket_connect_error (int err)
575 /* Have to guard against some of these values not being defined.
576 Cannot use a switch statement because some of the values might be
580 || err == EAFNOSUPPORT
583 || err == EPFNOSUPPORT
585 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
586 || err == ESOCKTNOSUPPORT
588 #ifdef EPROTONOSUPPORT
589 || err == EPROTONOSUPPORT
592 || err == ENOPROTOOPT
594 /* Apparently, older versions of Linux and BSD used EINVAL
595 instead of EAFNOSUPPORT and such. */
600 if (!opt.retry_connrefused)
601 if (err == ECONNREFUSED
603 || err == ENETUNREACH /* network is unreachable */
606 || err == EHOSTUNREACH /* host is unreachable */
615 # ifndef HAVE_GETADDRINFO_AI_ADDRCONFIG
617 /* Return non-zero if the INET6 socket family is supported on the
620 This doesn't guarantee that we're able to connect to IPv6 hosts,
621 but it's better than nothing. It is only used on systems where
622 getaddrinfo doesn't support AI_ADDRCONFIG. (See lookup_host.) */
625 socket_has_inet6 (void)
627 static int supported = -1;
630 int sock = socket (AF_INET6, SOCK_STREAM, 0);
642 # endif/* not HAVE_GETADDRINFO_AI_ADDRCONFIG */
643 #endif /* ENABLE_IPV6 */
645 /* Wait for a single descriptor to become available, timing out after
646 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
647 -1 for error. The argument WAIT_FOR can be a combination of
648 WAIT_FOR_READ and WAIT_FOR_WRITE.
650 This is a mere convenience wrapper around the select call, and
651 should be taken as such (for example, it doesn't implement Wget's
652 0-timeout-means-no-timeout semantics.) */
655 select_fd (int fd, double maxtime, int wait_for)
659 fd_set *rd = NULL, *wr = NULL;
660 struct timeval tmout;
665 if (wait_for & WAIT_FOR_READ)
667 if (wait_for & WAIT_FOR_WRITE)
670 tmout.tv_sec = (long) maxtime;
671 tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
674 result = select (fd + 1, rd, wr, NULL, &tmout);
675 while (result < 0 && errno == EINTR);
679 #else /* not HAVE_SELECT */
681 /* If select() unavailable, just return 1. In most usages in Wget,
682 this is the appropriate response -- "if we can't poll, go ahead
683 with the blocking operation". If a specific part of code needs
684 different behavior, it can use #ifdef HAVE_SELECT to test whether
685 polling really occurs. */
688 #endif /* not HAVE_SELECT */
692 test_socket_open (int sock)
698 /* Check if we still have a valid (non-EOF) connection. From Andrew
699 * Maholski's code in the Unix Socket FAQ. */
701 FD_ZERO (&check_set);
702 FD_SET (sock, &check_set);
704 /* Wait one microsecond */
708 /* If we get a timeout, then that means still connected */
709 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
711 /* Connection is valid (not EOF), so continue */
717 /* Without select, it's hard to know for sure. */
722 /* Basic socket operations, mostly EINTR wrappers. */
725 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
726 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
727 # define close(fd) closesocket (fd)
731 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
732 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
736 sock_read (int fd, char *buf, int bufsize)
740 res = read (fd, buf, bufsize);
741 while (res == -1 && errno == EINTR);
746 sock_write (int fd, char *buf, int bufsize)
750 res = write (fd, buf, bufsize);
751 while (res == -1 && errno == EINTR);
756 sock_poll (int fd, double timeout, int wait_for)
758 return select_fd (fd, timeout, wait_for);
762 sock_peek (int fd, char *buf, int bufsize)
766 res = recv (fd, buf, bufsize, MSG_PEEK);
767 while (res == -1 && errno == EINTR);
775 DEBUGP (("Closed fd %d\n", fd));
781 /* Reading and writing from the network. We build around the socket
782 (file descriptor) API, but support "extended" operations for things
783 that are not mere file descriptors under the hood, such as SSL
786 That way the user code can call fd_read(fd, ...) and we'll run read
787 or SSL_read or whatever is necessary. */
789 static struct hash_table *transport_map;
790 static int transport_map_modified_tick;
792 struct transport_info {
801 /* Register the transport layer operations that will be used when
802 reading, writing, and polling FD.
804 This should be used for transport layers like SSL that piggyback on
805 sockets. FD should otherwise be a real socket, on which you can
806 call getpeername, etc. */
809 fd_register_transport (int fd, fd_reader_t reader, fd_writer_t writer,
810 fd_poller_t poller, fd_peeker_t peeker,
811 fd_closer_t closer, void *ctx)
813 struct transport_info *info;
815 /* The file descriptor must be non-negative to be registered.
816 Negative values are ignored by fd_close(), and -1 cannot be used as
820 info = xnew (struct transport_info);
821 info->reader = reader;
822 info->writer = writer;
823 info->poller = poller;
824 info->peeker = peeker;
825 info->closer = closer;
828 transport_map = hash_table_new (0, NULL, NULL);
829 hash_table_put (transport_map, (void *) fd, info);
830 ++transport_map_modified_tick;
833 /* When fd_read/fd_write are called multiple times in a loop, they should
834 remember the INFO pointer instead of fetching it every time. It is
835 not enough to compare FD to LAST_FD because FD might have been
836 closed and reopened. modified_tick ensures that changes to
837 transport_map will not be unnoticed.
839 This is a macro because we want the static storage variables to be
842 #define LAZY_RETRIEVE_INFO(info) do { \
843 static struct transport_info *last_info; \
844 static int last_fd = -1, last_tick; \
845 if (!transport_map) \
847 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
851 info = hash_table_get (transport_map, (void *) fd); \
854 last_tick = transport_map_modified_tick; \
859 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
862 timeout = opt.read_timeout;
866 if (info && info->poller)
867 test = info->poller (fd, timeout, wf, info->ctx);
869 test = sock_poll (fd, timeout, wf);
878 /* Read no more than BUFSIZE bytes of data from FD, storing them to
879 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
880 received after that many seconds. If TIMEOUT is -1, the value of
881 opt.timeout is used for TIMEOUT. */
884 fd_read (int fd, char *buf, int bufsize, double timeout)
886 struct transport_info *info;
887 LAZY_RETRIEVE_INFO (info);
888 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
890 if (info && info->reader)
891 return info->reader (fd, buf, bufsize, info->ctx);
893 return sock_read (fd, buf, bufsize);
896 /* Like fd_read, except it provides a "preview" of the data that will
897 be read by subsequent calls to fd_read. Specifically, it copies no
898 more than BUFSIZE bytes of the currently available data to BUF and
899 returns the number of bytes copied. Return values and timeout
900 semantics are the same as those of fd_read.
902 CAVEAT: Do not assume that the first subsequent call to fd_read
903 will retrieve the same amount of data. Reading can return more or
904 less data, depending on the TCP implementation and other
905 circumstances. However, barring an error, it can be expected that
906 all the peeked data will eventually be read by fd_read. */
909 fd_peek (int fd, char *buf, int bufsize, double timeout)
911 struct transport_info *info;
912 LAZY_RETRIEVE_INFO (info);
913 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
915 if (info && info->peeker)
916 return info->peeker (fd, buf, bufsize, info->ctx);
918 return sock_peek (fd, buf, bufsize);
921 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
922 the operation aborts if no data is received after that many
923 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
927 fd_write (int fd, char *buf, int bufsize, double timeout)
930 struct transport_info *info;
931 LAZY_RETRIEVE_INFO (info);
933 /* `write' may write less than LEN bytes, thus the loop keeps trying
934 it until all was written, or an error occurred. */
938 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
940 if (info && info->writer)
941 res = info->writer (fd, buf, bufsize, info->ctx);
943 res = sock_write (fd, buf, bufsize);
952 /* Close the file descriptor FD. */
957 struct transport_info *info;
961 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
962 per socket, so that particular optimization wouldn't work. */
965 info = hash_table_get (transport_map, (void *) fd);
967 if (info && info->closer)
968 info->closer (fd, info->ctx);
974 hash_table_remove (transport_map, (void *) fd);
976 ++transport_map_modified_tick;