1 /* Establishing and handling network connections.
2 Copyright (C) 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007, 2008 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget. If not, see <http://www.gnu.org/licenses/>.
20 Additional permission under GNU GPL version 3 section 7
22 If you modify this program, or any covered work, by linking or
23 combining it with the OpenSSL project's OpenSSL library (or a
24 modified version of that library), containing parts covered by the
25 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
26 grants you additional permission to convey the resulting work.
27 Corresponding Source for a non-source form of such a combination
28 shall include the source code for the parts of OpenSSL used as well
29 as that of the covered work. */
41 # include <sys/socket.h>
43 # include <netinet/in.h>
45 # include <arpa/inet.h>
47 #endif /* not WINDOWS */
51 #ifdef HAVE_SYS_SELECT_H
52 # include <sys/select.h>
53 #endif /* HAVE_SYS_SELECT_H */
54 #ifdef HAVE_SYS_TIME_H
55 # include <sys/time.h>
62 /* Define sockaddr_storage where unavailable (presumably on IPv4-only
66 # ifndef HAVE_STRUCT_SOCKADDR_STORAGE
67 # define sockaddr_storage sockaddr_in
69 #endif /* ENABLE_IPV6 */
71 /* Fill SA as per the data in IP and PORT. SA shoult point to struct
72 sockaddr_storage if ENABLE_IPV6 is defined, to struct sockaddr_in
76 sockaddr_set_data (struct sockaddr *sa, const ip_address *ip, int port)
82 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
84 sin->sin_family = AF_INET;
85 sin->sin_port = htons (port);
86 sin->sin_addr = ip->data.d4;
92 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
94 sin6->sin6_family = AF_INET6;
95 sin6->sin6_port = htons (port);
96 sin6->sin6_addr = ip->data.d6;
97 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
98 sin6->sin6_scope_id = ip->ipv6_scope;
102 #endif /* ENABLE_IPV6 */
108 /* Get the data of SA, specifically the IP address and the port. If
109 you're not interested in one or the other information, pass NULL as
113 sockaddr_get_data (const struct sockaddr *sa, ip_address *ip, int *port)
115 switch (sa->sa_family)
119 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
122 ip->family = AF_INET;
123 ip->data.d4 = sin->sin_addr;
126 *port = ntohs (sin->sin_port);
132 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
135 ip->family = AF_INET6;
136 ip->data.d6 = sin6->sin6_addr;
137 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
138 ip->ipv6_scope = sin6->sin6_scope_id;
142 *port = ntohs (sin6->sin6_port);
151 /* Return the size of the sockaddr structure depending on its
155 sockaddr_size (const struct sockaddr *sa)
157 switch (sa->sa_family)
160 return sizeof (struct sockaddr_in);
163 return sizeof (struct sockaddr_in6);
170 /* Resolve the bind address specified via --bind-address and store it
171 to SA. The resolved value is stored in a static variable and
172 reused after the first invocation of this function.
174 Returns true on success, false on failure. */
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 bool 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, quote (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 = print_address (ip);
269 if (print && 0 != strcmp (print, txt_addr))
270 logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
271 escnonprint_uri (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 #if defined(ENABLE_IPV6) && defined(IPV6_V6ONLY)
287 /* In case of error, we will go on anyway... */
288 int err = setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof (on));
291 DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
295 /* For very small rate limits, set the buffer size (and hence,
296 hopefully, the kernel's TCP window size) to the per-second limit.
297 That way we should never have to sleep for more than 1s between
299 if (opt.limit_rate && opt.limit_rate < 8192)
301 int bufsize = opt.limit_rate;
303 bufsize = 512; /* avoid pathologically small values */
305 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
306 (void *)&bufsize, (socklen_t)sizeof (bufsize));
308 /* When we add limit_rate support for writing, which is useful
309 for POST, we should also set SO_SNDBUF here. */
312 if (opt.bind_address)
314 /* Bind the client side of the socket to the requested
316 struct sockaddr_storage bind_ss;
317 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
318 if (resolve_bind_address (bind_sa))
320 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
325 /* Connect the socket to the remote endpoint. */
326 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
327 opt.connect_timeout) < 0)
333 logprintf (LOG_VERBOSE, _("connected.\n"));
334 DEBUGP (("Created socket %d.\n", sock));
339 /* Protect errno from possible modifications by close and
341 int save_errno = errno;
345 logprintf (LOG_VERBOSE, _("failed: %s.\n"), strerror (errno));
351 /* Connect via TCP to a remote host on the specified port.
353 HOST is resolved as an Internet host name. If HOST resolves to
354 more than one IP address, they are tried in the order returned by
355 DNS until connecting to one of them succeeds. */
358 connect_to_host (const char *host, int port)
363 struct address_list *al = lookup_host (host, 0);
368 logprintf (LOG_NOTQUIET,
369 _("%s: unable to resolve host address %s\n"),
370 exec_name, quote (host));
374 address_list_get_bounds (al, &start, &end);
375 for (i = start; i < end; i++)
377 const ip_address *ip = address_list_address_at (al, i);
378 sock = connect_to_ip (ip, port, host);
382 address_list_set_connected (al);
383 address_list_release (al);
387 /* The attempt to connect has failed. Continue with the loop
388 and try next address. */
390 address_list_set_faulty (al, i);
393 /* Failed to connect to any of the addresses in AL. */
395 if (address_list_connected_p (al))
397 /* We connected to AL before, but cannot do so now. That might
398 indicate that our DNS cache entry for HOST has expired. */
399 address_list_release (al);
400 al = lookup_host (host, LH_REFRESH);
403 address_list_release (al);
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 struct sockaddr_storage ss;
426 struct sockaddr *sa = (struct sockaddr *)&ss;
428 /* For setting options with setsockopt. */
430 void *setopt_ptr = (void *)&setopt_val;
431 socklen_t setopt_size = sizeof (setopt_val);
433 sock = socket (bind_address->family, SOCK_STREAM, 0);
438 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
442 sockaddr_set_data (sa, bind_address, *port);
443 if (bind (sock, sa, sockaddr_size (sa)) < 0)
448 DEBUGP (("Local socket fd %d bound.\n", sock));
450 /* If *PORT is 0, find out which port we've bound to. */
453 socklen_t addrlen = sockaddr_size (sa);
454 if (getsockname (sock, sa, &addrlen) < 0)
456 /* If we can't find out the socket's local address ("name"),
457 something is seriously wrong with the socket, and it's
458 unusable for us anyway because we must know the chosen
463 sockaddr_get_data (sa, NULL, port);
464 DEBUGP (("binding to address %s using port %i.\n",
465 print_address (bind_address), *port));
467 if (listen (sock, 1) < 0)
475 /* Like a call to accept(), but with the added check for timeout.
477 In other words, accept a client connection on LOCAL_SOCK, and
478 return the new socket used for communication with the client.
479 LOCAL_SOCK should have been bound, e.g. using bind_local().
481 The caller is blocked until a connection is established. If no
482 connection is established for opt.connect_timeout seconds, the
483 function exits with an error status. */
486 accept_connection (int local_sock)
490 /* We don't need the values provided by accept, but accept
491 apparently requires them to be present. */
492 struct sockaddr_storage ss;
493 struct sockaddr *sa = (struct sockaddr *)&ss;
494 socklen_t addrlen = sizeof (ss);
496 if (opt.connect_timeout)
498 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
504 sock = accept (local_sock, sa, &addrlen);
505 DEBUGP (("Accepted client at socket %d.\n", sock));
509 /* Get the IP address associated with the connection on FD and store
510 it to IP. Return true on success, false otherwise.
512 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
513 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
514 returns the address of the remote (peer's) side of the socket. */
517 socket_ip_address (int sock, ip_address *ip, int endpoint)
519 struct sockaddr_storage storage;
520 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
521 socklen_t addrlen = sizeof (storage);
524 if (endpoint == ENDPOINT_LOCAL)
525 ret = getsockname (sock, sockaddr, &addrlen);
526 else if (endpoint == ENDPOINT_PEER)
527 ret = getpeername (sock, sockaddr, &addrlen);
533 ip->family = sockaddr->sa_family;
534 switch (sockaddr->sa_family)
539 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
540 ip->data.d6 = sa6->sin6_addr;
541 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
542 ip->ipv6_scope = sa6->sin6_scope_id;
544 DEBUGP (("conaddr is: %s\n", print_address (ip)));
550 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
551 ip->data.d4 = sa->sin_addr;
552 DEBUGP (("conaddr is: %s\n", print_address (ip)));
560 /* Return true if the error from the connect code can be considered
561 retryable. Wget normally retries after errors, but the exception
562 are the "unsupported protocol" type errors (possible on IPv4/IPv6
563 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 (!opt.retry_connrefused)
594 if (err == ECONNREFUSED
596 || err == ENETUNREACH /* network is unreachable */
599 || err == EHOSTUNREACH /* host is unreachable */
607 /* Wait for a single descriptor to become available, timing out after
608 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
609 -1 for error. The argument WAIT_FOR can be a combination of
610 WAIT_FOR_READ and WAIT_FOR_WRITE.
612 This is a mere convenience wrapper around the select call, and
613 should be taken as such (for example, it doesn't implement Wget's
614 0-timeout-means-no-timeout semantics.) */
617 select_fd (int fd, double maxtime, int wait_for)
620 fd_set *rd = NULL, *wr = NULL;
621 struct timeval tmout;
626 if (wait_for & WAIT_FOR_READ)
628 if (wait_for & WAIT_FOR_WRITE)
631 tmout.tv_sec = (long) maxtime;
632 tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
635 result = select (fd + 1, rd, wr, NULL, &tmout);
636 while (result < 0 && errno == EINTR);
641 /* Return true iff the connection to the remote site established
642 through SOCK is still open.
644 Specifically, this function returns true if SOCK is not ready for
645 reading. This is because, when the connection closes, the socket
646 is ready for reading because EOF is about to be delivered. A side
647 effect of this method is that sockets that have pending data are
648 considered non-open. This is actually a good thing for callers of
649 this function, where such pending data can only be unwanted
650 leftover from a previous request. */
653 test_socket_open (int sock)
658 /* Check if we still have a valid (non-EOF) connection. From Andrew
659 * Maholski's code in the Unix Socket FAQ. */
661 FD_ZERO (&check_set);
662 FD_SET (sock, &check_set);
664 /* Wait one microsecond */
668 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
669 /* We got a timeout, it means we're still connected. */
672 /* Read now would not wait, it means we have either pending data
677 /* Basic socket operations, mostly EINTR wrappers. */
679 #if defined(WINDOWS) || defined(MSDOS)
680 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
681 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
682 # define close(fd) closesocket (fd)
686 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
687 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
691 sock_read (int fd, char *buf, int bufsize)
695 res = read (fd, buf, bufsize);
696 while (res == -1 && errno == EINTR);
701 sock_write (int fd, char *buf, int bufsize)
705 res = write (fd, buf, bufsize);
706 while (res == -1 && errno == EINTR);
711 sock_poll (int fd, double timeout, int wait_for)
713 return select_fd (fd, timeout, wait_for);
717 sock_peek (int fd, char *buf, int bufsize)
721 res = recv (fd, buf, bufsize, MSG_PEEK);
722 while (res == -1 && errno == EINTR);
730 DEBUGP (("Closed fd %d\n", fd));
736 /* Reading and writing from the network. We build around the socket
737 (file descriptor) API, but support "extended" operations for things
738 that are not mere file descriptors under the hood, such as SSL
741 That way the user code can call fd_read(fd, ...) and we'll run read
742 or SSL_read or whatever is necessary. */
744 static struct hash_table *transport_map;
745 static unsigned int transport_map_modified_tick;
747 struct transport_info {
748 struct transport_implementation *imp;
752 /* Register the transport layer operations that will be used when
753 reading, writing, and polling FD.
755 This should be used for transport layers like SSL that piggyback on
756 sockets. FD should otherwise be a real socket, on which you can
757 call getpeername, etc. */
760 fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
762 struct transport_info *info;
764 /* The file descriptor must be non-negative to be registered.
765 Negative values are ignored by fd_close(), and -1 cannot be used as
769 info = xnew (struct transport_info);
773 transport_map = hash_table_new (0, NULL, NULL);
774 hash_table_put (transport_map, (void *)(intptr_t) fd, info);
775 ++transport_map_modified_tick;
778 /* Return context of the transport registered with
779 fd_register_transport. This assumes fd_register_transport was
780 previously called on FD. */
783 fd_transport_context (int fd)
785 struct transport_info *info = hash_table_get (transport_map, (void *)(intptr_t) fd);
789 /* When fd_read/fd_write are called multiple times in a loop, they should
790 remember the INFO pointer instead of fetching it every time. It is
791 not enough to compare FD to LAST_FD because FD might have been
792 closed and reopened. modified_tick ensures that changes to
793 transport_map will not be unnoticed.
795 This is a macro because we want the static storage variables to be
798 #define LAZY_RETRIEVE_INFO(info) do { \
799 static struct transport_info *last_info; \
800 static int last_fd = -1; \
801 static unsigned int last_tick; \
802 if (!transport_map) \
804 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
808 info = hash_table_get (transport_map, (void *)(intptr_t) fd); \
811 last_tick = transport_map_modified_tick; \
816 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
819 timeout = opt.read_timeout;
823 if (info && info->imp->poller)
824 test = info->imp->poller (fd, timeout, wf, info->ctx);
826 test = sock_poll (fd, timeout, wf);
835 /* Read no more than BUFSIZE bytes of data from FD, storing them to
836 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
837 received after that many seconds. If TIMEOUT is -1, the value of
838 opt.timeout is used for TIMEOUT. */
841 fd_read (int fd, char *buf, int bufsize, double timeout)
843 struct transport_info *info;
844 LAZY_RETRIEVE_INFO (info);
845 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
847 if (info && info->imp->reader)
848 return info->imp->reader (fd, buf, bufsize, info->ctx);
850 return sock_read (fd, buf, bufsize);
853 /* Like fd_read, except it provides a "preview" of the data that will
854 be read by subsequent calls to fd_read. Specifically, it copies no
855 more than BUFSIZE bytes of the currently available data to BUF and
856 returns the number of bytes copied. Return values and timeout
857 semantics are the same as those of fd_read.
859 CAVEAT: Do not assume that the first subsequent call to fd_read
860 will retrieve the same amount of data. Reading can return more or
861 less data, depending on the TCP implementation and other
862 circumstances. However, barring an error, it can be expected that
863 all the peeked data will eventually be read by fd_read. */
866 fd_peek (int fd, char *buf, int bufsize, double timeout)
868 struct transport_info *info;
869 LAZY_RETRIEVE_INFO (info);
870 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
872 if (info && info->imp->peeker)
873 return info->imp->peeker (fd, buf, bufsize, info->ctx);
875 return sock_peek (fd, buf, bufsize);
878 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
879 the operation aborts if no data is received after that many
880 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
884 fd_write (int fd, char *buf, int bufsize, double timeout)
887 struct transport_info *info;
888 LAZY_RETRIEVE_INFO (info);
890 /* `write' may write less than LEN bytes, thus the loop keeps trying
891 it until all was written, or an error occurred. */
895 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
897 if (info && info->imp->writer)
898 res = info->imp->writer (fd, buf, bufsize, info->ctx);
900 res = sock_write (fd, buf, bufsize);
909 /* Report the most recent error(s) on FD. This should only be called
910 after fd_* functions, such as fd_read and fd_write, and only if
911 they return a negative result. For errors coming from other calls
912 such as setsockopt or fopen, strerror should continue to be
915 If the transport doesn't support error messages or doesn't supply
916 one, strerror(errno) is returned. The returned error message
917 should not be used after fd_close has been called. */
922 /* Don't bother with LAZY_RETRIEVE_INFO, as this will only be called
923 in case of error, never in a tight loop. */
924 struct transport_info *info = NULL;
926 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
928 if (info && info->imp->errstr)
930 const char *err = info->imp->errstr (fd, info->ctx);
933 /* else, fall through and print the system error. */
935 return strerror (errno);
938 /* Close the file descriptor FD. */
943 struct transport_info *info;
947 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
948 per socket, so that particular optimization wouldn't work. */
951 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
953 if (info && info->imp->closer)
954 info->imp->closer (fd, info->ctx);
960 hash_table_remove (transport_map, (void *)(intptr_t) fd);
962 ++transport_map_modified_tick;