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;
90 sin->sin_family = AF_INET;
91 sin->sin_port = htons (port);
92 sin->sin_addr = ADDRESS_IPV4_IN_ADDR (ip);
98 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
99 sin6->sin6_family = AF_INET6;
100 sin6->sin6_port = htons (port);
101 sin6->sin6_addr = ADDRESS_IPV6_IN6_ADDR (ip);
102 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
103 sin6->sin6_scope_id = ADDRESS_IPV6_SCOPE (ip);
107 #endif /* ENABLE_IPV6 */
113 /* Get the data of SA, specifically the IP address and the port. If
114 you're not interested in one or the other information, pass NULL as
118 sockaddr_get_data (const struct sockaddr *sa, ip_address *ip, int *port)
120 switch (sa->sa_family)
124 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
127 ip->type = IPV4_ADDRESS;
128 ADDRESS_IPV4_IN_ADDR (ip) = sin->sin_addr;
131 *port = ntohs (sin->sin_port);
137 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
140 ip->type = IPV6_ADDRESS;
141 ADDRESS_IPV6_IN6_ADDR (ip) = sin6->sin6_addr;
142 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
143 ADDRESS_IPV6_SCOPE (ip) = sin6->sin6_scope_id;
147 *port = ntohs (sin6->sin6_port);
156 /* Return the size of the sockaddr structure depending on its
160 sockaddr_size (const struct sockaddr *sa)
162 switch (sa->sa_family)
165 return sizeof (struct sockaddr_in);
168 return sizeof (struct sockaddr_in6);
172 return 0; /* so the compiler shuts up. */
177 resolve_bind_address (struct sockaddr *sa)
179 struct address_list *al;
181 /* Make sure this is called only once. opt.bind_address doesn't
182 change during a Wget run. */
183 static int called, should_bind;
184 static ip_address ip;
188 sockaddr_set_data (sa, &ip, 0);
193 al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
196 /* #### We should be able to print the error message here. */
197 logprintf (LOG_NOTQUIET,
198 _("%s: unable to resolve bind address `%s'; disabling bind.\n"),
199 exec_name, opt.bind_address);
204 /* Pick the first address in the list and use it as bind address.
205 Perhaps we should try multiple addresses in succession, but I
206 don't think that's necessary in practice. */
207 ip = *address_list_address_at (al, 0);
208 address_list_release (al);
210 sockaddr_set_data (sa, &ip, 0);
217 const struct sockaddr *addr;
223 connect_with_timeout_callback (void *arg)
225 struct cwt_context *ctx = (struct cwt_context *)arg;
226 ctx->result = connect (ctx->fd, ctx->addr, ctx->addrlen);
229 /* Like connect, but specifies a timeout. If connecting takes longer
230 than TIMEOUT seconds, -1 is returned and errno is set to
234 connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
237 struct cwt_context ctx;
240 ctx.addrlen = addrlen;
242 if (run_with_timeout (timeout, connect_with_timeout_callback, &ctx))
247 if (ctx.result == -1 && errno == EINTR)
252 /* Connect via TCP to the specified address and port.
254 If PRINT is non-NULL, it is the host name to print that we're
258 connect_to_ip (const ip_address *ip, int port, const char *print)
260 struct sockaddr_storage ss;
261 struct sockaddr *sa = (struct sockaddr *)&ss;
264 /* If PRINT is non-NULL, print the "Connecting to..." line, with
265 PRINT being the host name we're connecting to. */
268 const char *txt_addr = pretty_print_address (ip);
269 if (print && 0 != strcmp (print, txt_addr))
270 logprintf (LOG_VERBOSE,
271 _("Connecting to %s|%s|:%d... "), print, txt_addr, port);
273 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
276 /* Store the sockaddr info to SA. */
277 sockaddr_set_data (sa, ip, port);
279 /* Create the socket of the family appropriate for the address. */
280 sock = socket (sa->sa_family, SOCK_STREAM, 0);
284 /* For very small rate limits, set the buffer size (and hence,
285 hopefully, the kernel's TCP window size) to the per-second limit.
286 That way we should never have to sleep for more than 1s between
288 if (opt.limit_rate && opt.limit_rate < 8192)
290 int bufsize = opt.limit_rate;
292 bufsize = 512; /* avoid pathologically small values */
294 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
295 (void *)&bufsize, (socklen_t)sizeof (bufsize));
297 /* When we add limit_rate support for writing, which is useful
298 for POST, we should also set SO_SNDBUF here. */
301 if (opt.bind_address)
303 /* Bind the client side of the socket to the requested
305 struct sockaddr_storage bind_ss;
306 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
307 if (resolve_bind_address (bind_sa))
309 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
314 /* Connect the socket to the remote endpoint. */
315 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
316 opt.connect_timeout) < 0)
322 logprintf (LOG_VERBOSE, _("connected.\n"));
323 DEBUGP (("Created socket %d.\n", sock));
328 /* Protect errno from possible modifications by close and
330 int save_errno = errno;
334 logprintf (LOG_VERBOSE, "failed: %s.\n", strerror (errno));
340 /* Connect via TCP to a remote host on the specified port.
342 HOST is resolved as an Internet host name. If HOST resolves to
343 more than one IP address, they are tried in the order returned by
344 DNS until connecting to one of them succeeds. */
347 connect_to_host (const char *host, int port)
352 struct address_list *al = lookup_host (host, 0);
358 address_list_get_bounds (al, &start, &end);
359 for (i = start; i < end; i++)
361 const ip_address *ip = address_list_address_at (al, i);
362 sock = connect_to_ip (ip, port, host);
366 address_list_set_connected (al);
367 address_list_release (al);
371 /* The attempt to connect has failed. Continue with the loop
372 and try next address. */
374 address_list_set_faulty (al, i);
377 /* Failed to connect to any of the addresses in AL. */
379 if (address_list_connected_p (al))
381 /* We connected to AL before, but cannot do so now. That might
382 indicate that our DNS cache entry for HOST has expired. */
383 address_list_release (al);
384 al = lookup_host (host, LH_REFRESH);
387 address_list_release (al);
392 /* Create a socket, bind it to local interface BIND_ADDRESS on port
393 *PORT, set up a listen backlog, and return the resulting socket, or
396 BIND_ADDRESS is the address of the interface to bind to. If it is
397 NULL, the socket is bound to the default address. PORT should
398 point to the port number that will be used for the binding. If
399 that number is 0, the system will choose a suitable port, and the
400 chosen value will be written to *PORT.
402 Calling accept() on such a socket waits for and accepts incoming
406 bind_local (const ip_address *bind_address, int *port)
409 int family = AF_INET;
410 struct sockaddr_storage ss;
411 struct sockaddr *sa = (struct sockaddr *)&ss;
413 /* For setting options with setsockopt. */
415 void *setopt_ptr = (void *)&setopt_val;
416 socklen_t setopt_size = sizeof (setopt_val);
419 if (bind_address->type == IPV6_ADDRESS)
423 sock = socket (family, SOCK_STREAM, 0);
428 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
432 # ifdef HAVE_IPV6_V6ONLY
433 if (family == AF_INET6)
434 setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, setopt_ptr, setopt_size);
439 sockaddr_set_data (sa, bind_address, *port);
440 if (bind (sock, sa, sockaddr_size (sa)) < 0)
445 DEBUGP (("Local socket fd %d bound.\n", sock));
447 /* If *PORT is 0, find out which port we've bound to. */
450 socklen_t sa_len = sockaddr_size (sa);
451 if (getsockname (sock, sa, &sa_len) < 0)
453 /* If we can't find out the socket's local address ("name"),
454 something is seriously wrong with the socket, and it's
455 unusable for us anyway because we must know the chosen
460 sockaddr_get_data (sa, NULL, port);
461 DEBUGP (("binding to address %s using port %i.\n",
462 pretty_print_address (bind_address), *port));
464 if (listen (sock, 1) < 0)
472 /* Like a call to accept(), but with the added check for timeout.
474 In other words, accept a client connection on LOCAL_SOCK, and
475 return the new socket used for communication with the client.
476 LOCAL_SOCK should have been bound, e.g. using bind_local().
478 The caller is blocked until a connection is established. If no
479 connection is established for opt.connect_timeout seconds, the
480 function exits with an error status. */
483 accept_connection (int local_sock)
487 /* We don't need the values provided by accept, but accept
488 apparently requires them to be present. */
489 struct sockaddr_storage ss;
490 struct sockaddr *sa = (struct sockaddr *)&ss;
491 socklen_t addrlen = sizeof (ss);
493 if (opt.connect_timeout)
495 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
501 sock = accept (local_sock, sa, &addrlen);
502 DEBUGP (("Accepted client at socket %d.\n", sock));
506 /* Get the IP address associated with the connection on FD and store
507 it to IP. Return 1 on success, 0 otherwise.
509 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
510 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
511 returns the address of the remote (peer's) side of the socket. */
514 socket_ip_address (int sock, ip_address *ip, int endpoint)
516 struct sockaddr_storage storage;
517 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
518 socklen_t addrlen = sizeof (storage);
521 if (endpoint == ENDPOINT_LOCAL)
522 ret = getsockname (sock, sockaddr, &addrlen);
523 else if (endpoint == ENDPOINT_PEER)
524 ret = getpeername (sock, sockaddr, &addrlen);
530 switch (sockaddr->sa_family)
535 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
536 ip->type = IPV6_ADDRESS;
537 ADDRESS_IPV6_IN6_ADDR (ip) = sa6->sin6_addr;
538 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
539 ADDRESS_IPV6_SCOPE (ip) = sa6->sin6_scope_id;
541 DEBUGP (("conaddr is: %s\n", pretty_print_address (ip)));
547 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
548 ip->type = IPV4_ADDRESS;
549 ADDRESS_IPV4_IN_ADDR (ip) = sa->sin_addr;
550 DEBUGP (("conaddr is: %s\n", pretty_print_address (ip)));
560 /* Return non-zero if the error from the connect code can be
561 considered retryable. Wget normally retries after errors, but the
562 exception are the "unsupported protocol" type errors (possible on
563 IPv4/IPv6 dual family systems) and "connection refused". */
566 retryable_socket_connect_error (int err)
568 /* Have to guard against some of these values not being defined.
569 Cannot use a switch statement because some of the values might be
573 || err == EAFNOSUPPORT
576 || err == EPFNOSUPPORT
578 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
579 || err == ESOCKTNOSUPPORT
581 #ifdef EPROTONOSUPPORT
582 || err == EPROTONOSUPPORT
585 || err == ENOPROTOOPT
587 /* Apparently, older versions of Linux and BSD used EINVAL
588 instead of EAFNOSUPPORT and such. */
593 if (err == ECONNREFUSED && !opt.retry_connrefused)
600 # ifndef HAVE_GETADDRINFO_AI_ADDRCONFIG
602 /* Return non-zero if the INET6 socket family is supported on the
605 This doesn't guarantee that we're able to connect to IPv6 hosts,
606 but it's better than nothing. It is only used on systems where
607 getaddrinfo doesn't support AI_ADDRCONFIG. (See lookup_host.) */
610 socket_has_inet6 (void)
612 static int supported = -1;
615 int sock = socket (AF_INET6, SOCK_STREAM, 0);
627 # endif/* not HAVE_GETADDRINFO_AI_ADDRCONFIG */
628 #endif /* ENABLE_IPV6 */
630 /* Wait for a single descriptor to become available, timing out after
631 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
632 -1 for error. The argument WAIT_FOR can be a combination of
633 WAIT_FOR_READ and WAIT_FOR_WRITE.
635 This is a mere convenience wrapper around the select call, and
636 should be taken as such (for example, it doesn't implement Wget's
637 0-timeout-means-no-timeout semantics.) */
640 select_fd (int fd, double maxtime, int wait_for)
644 fd_set *rd = NULL, *wr = NULL;
645 struct timeval tmout;
650 if (wait_for & WAIT_FOR_READ)
652 if (wait_for & WAIT_FOR_WRITE)
655 tmout.tv_sec = (long) maxtime;
656 tmout.tv_usec = 1000000L * (maxtime - (long) maxtime);
659 result = select (fd + 1, rd, wr, NULL, &tmout);
660 while (result < 0 && errno == EINTR);
664 #else /* not HAVE_SELECT */
666 /* If select() unavailable, just return 1. In most usages in Wget,
667 this is the appropriate response -- "if we can't poll, go ahead
668 with the blocking operation". If a specific part of code needs
669 different behavior, it can use #ifdef HAVE_SELECT to test whether
670 polling really occurs. */
673 #endif /* not HAVE_SELECT */
677 test_socket_open (int sock)
683 /* Check if we still have a valid (non-EOF) connection. From Andrew
684 * Maholski's code in the Unix Socket FAQ. */
686 FD_ZERO (&check_set);
687 FD_SET (sock, &check_set);
689 /* Wait one microsecond */
693 /* If we get a timeout, then that means still connected */
694 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
696 /* Connection is valid (not EOF), so continue */
702 /* Without select, it's hard to know for sure. */
707 /* Basic socket operations, mostly EINTR wrappers. */
710 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
711 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
712 # define close(fd) closesocket (fd)
716 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
717 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
721 sock_read (int fd, char *buf, int bufsize)
725 res = read (fd, buf, bufsize);
726 while (res == -1 && errno == EINTR);
731 sock_write (int fd, char *buf, int bufsize)
735 res = write (fd, buf, bufsize);
736 while (res == -1 && errno == EINTR);
741 sock_poll (int fd, double timeout, int wait_for)
743 return select_fd (fd, timeout, wait_for);
747 sock_peek (int fd, char *buf, int bufsize)
751 res = recv (fd, buf, bufsize, MSG_PEEK);
752 while (res == -1 && errno == EINTR);
760 DEBUGP (("Closed fd %d\n", fd));
766 /* Reading and writing from the network. We build around the socket
767 (file descriptor) API, but support "extended" operations for things
768 that are not mere file descriptors under the hood, such as SSL
771 That way the user code can call fd_read(fd, ...) and we'll run read
772 or SSL_read or whatever is necessary. */
774 static struct hash_table *transport_map;
775 static int transport_map_modified_tick;
777 struct transport_info {
786 /* Register the transport layer operations that will be used when
787 reading, writing, and polling FD.
789 This should be used for transport layers like SSL that piggyback on
790 sockets. FD should otherwise be a real socket, on which you can
791 call getpeername, etc. */
794 fd_register_transport (int fd, fd_reader_t reader, fd_writer_t writer,
795 fd_poller_t poller, fd_peeker_t peeker,
796 fd_closer_t closer, void *ctx)
798 struct transport_info *info;
800 /* The file descriptor must be non-negative to be registered.
801 Negative values are ignored by fd_close(), and -1 cannot be used as
805 info = xnew (struct transport_info);
806 info->reader = reader;
807 info->writer = writer;
808 info->poller = poller;
809 info->peeker = peeker;
810 info->closer = closer;
813 transport_map = hash_table_new (0, NULL, NULL);
814 hash_table_put (transport_map, (void *) fd, info);
815 ++transport_map_modified_tick;
818 /* When fd_read/fd_write are called multiple times in a loop, they should
819 remember the INFO pointer instead of fetching it every time. It is
820 not enough to compare FD to LAST_FD because FD might have been
821 closed and reopened. modified_tick ensures that changes to
822 transport_map will not be unnoticed.
824 This is a macro because we want the static storage variables to be
827 #define LAZY_RETRIEVE_INFO(info) do { \
828 static struct transport_info *last_info; \
829 static int last_fd = -1, last_tick; \
830 if (!transport_map) \
832 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
836 info = hash_table_get (transport_map, (void *) fd); \
839 last_tick = transport_map_modified_tick; \
844 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
847 timeout = opt.read_timeout;
851 if (info && info->poller)
852 test = info->poller (fd, timeout, wf, info->ctx);
854 test = sock_poll (fd, timeout, wf);
863 /* Read no more than BUFSIZE bytes of data from FD, storing them to
864 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
865 received after that many seconds. If TIMEOUT is -1, the value of
866 opt.timeout is used for TIMEOUT. */
869 fd_read (int fd, char *buf, int bufsize, double timeout)
871 struct transport_info *info;
872 LAZY_RETRIEVE_INFO (info);
873 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
875 if (info && info->reader)
876 return info->reader (fd, buf, bufsize, info->ctx);
878 return sock_read (fd, buf, bufsize);
881 /* The same as xread, but don't actually read the data, just copy it
885 fd_peek (int fd, char *buf, int bufsize, double timeout)
887 struct transport_info *info;
888 LAZY_RETRIEVE_INFO (info);
889 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
891 if (info && info->peeker)
892 return info->peeker (fd, buf, bufsize, info->ctx);
894 return sock_peek (fd, buf, bufsize);
897 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
898 the operation aborts if no data is received after that many
899 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
903 fd_write (int fd, char *buf, int bufsize, double timeout)
906 struct transport_info *info;
907 LAZY_RETRIEVE_INFO (info);
909 /* `write' may write less than LEN bytes, thus the loop keeps trying
910 it until all was written, or an error occurred. */
914 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
916 if (info && info->writer)
917 res = info->writer (fd, buf, bufsize, info->ctx);
919 res = sock_write (fd, buf, bufsize);
928 /* Close the file descriptor FD. */
933 struct transport_info *info;
937 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
938 per socket, so that particular optimization wouldn't work. */
941 info = hash_table_get (transport_map, (void *) fd);
943 if (info && info->closer)
944 info->closer (fd, info->ctx);
950 hash_table_remove (transport_map, (void *) fd);
952 ++transport_map_modified_tick;