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 /* Apparently needed for Interix: */
67 /* Define sockaddr_storage where unavailable (presumably on IPv4-only
71 # ifndef HAVE_STRUCT_SOCKADDR_STORAGE
72 # define sockaddr_storage sockaddr_in
74 #endif /* ENABLE_IPV6 */
76 /* Fill SA as per the data in IP and PORT. SA shoult point to struct
77 sockaddr_storage if ENABLE_IPV6 is defined, to struct sockaddr_in
81 sockaddr_set_data (struct sockaddr *sa, const ip_address *ip, int port)
87 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
89 sin->sin_family = AF_INET;
90 sin->sin_port = htons (port);
91 sin->sin_addr = ip->data.d4;
97 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
99 sin6->sin6_family = AF_INET6;
100 sin6->sin6_port = htons (port);
101 sin6->sin6_addr = ip->data.d6;
102 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
103 sin6->sin6_scope_id = ip->ipv6_scope;
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->family = AF_INET;
128 ip->data.d4 = sin->sin_addr;
131 *port = ntohs (sin->sin_port);
137 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
140 ip->family = AF_INET6;
141 ip->data.d6 = sin6->sin6_addr;
142 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
143 ip->ipv6_scope = 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);
175 /* Resolve the bind address specified via --bind-address and store it
176 to SA. The resolved value is stored in a static variable and
177 reused after the first invocation of this function.
179 Returns true on success, false on failure. */
182 resolve_bind_address (struct sockaddr *sa)
184 struct address_list *al;
186 /* Make sure this is called only once. opt.bind_address doesn't
187 change during a Wget run. */
188 static bool called, should_bind;
189 static ip_address ip;
193 sockaddr_set_data (sa, &ip, 0);
198 al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
201 /* #### We should be able to print the error message here. */
202 logprintf (LOG_NOTQUIET,
203 _("%s: unable to resolve bind address %s; disabling bind.\n"),
204 exec_name, quote (opt.bind_address));
209 /* Pick the first address in the list and use it as bind address.
210 Perhaps we should try multiple addresses in succession, but I
211 don't think that's necessary in practice. */
212 ip = *address_list_address_at (al, 0);
213 address_list_release (al);
215 sockaddr_set_data (sa, &ip, 0);
222 const struct sockaddr *addr;
228 connect_with_timeout_callback (void *arg)
230 struct cwt_context *ctx = (struct cwt_context *)arg;
231 ctx->result = connect (ctx->fd, ctx->addr, ctx->addrlen);
234 /* Like connect, but specifies a timeout. If connecting takes longer
235 than TIMEOUT seconds, -1 is returned and errno is set to
239 connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
242 struct cwt_context ctx;
245 ctx.addrlen = addrlen;
247 if (run_with_timeout (timeout, connect_with_timeout_callback, &ctx))
252 if (ctx.result == -1 && errno == EINTR)
257 /* Connect via TCP to the specified address and port.
259 If PRINT is non-NULL, it is the host name to print that we're
263 connect_to_ip (const ip_address *ip, int port, const char *print)
265 struct sockaddr_storage ss;
266 struct sockaddr *sa = (struct sockaddr *)&ss;
269 /* If PRINT is non-NULL, print the "Connecting to..." line, with
270 PRINT being the host name we're connecting to. */
273 const char *txt_addr = print_address (ip);
274 if (print && 0 != strcmp (print, txt_addr))
275 logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
276 escnonprint_uri (print), txt_addr, port);
278 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
281 /* Store the sockaddr info to SA. */
282 sockaddr_set_data (sa, ip, port);
284 /* Create the socket of the family appropriate for the address. */
285 sock = socket (sa->sa_family, SOCK_STREAM, 0);
289 #if defined(ENABLE_IPV6) && defined(IPV6_V6ONLY)
292 /* In case of error, we will go on anyway... */
293 int err = setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof (on));
296 DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
300 /* For very small rate limits, set the buffer size (and hence,
301 hopefully, the kernel's TCP window size) to the per-second limit.
302 That way we should never have to sleep for more than 1s between
304 if (opt.limit_rate && opt.limit_rate < 8192)
306 int bufsize = opt.limit_rate;
308 bufsize = 512; /* avoid pathologically small values */
310 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
311 (void *)&bufsize, (socklen_t)sizeof (bufsize));
313 /* When we add limit_rate support for writing, which is useful
314 for POST, we should also set SO_SNDBUF here. */
317 if (opt.bind_address)
319 /* Bind the client side of the socket to the requested
321 struct sockaddr_storage bind_ss;
322 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
323 if (resolve_bind_address (bind_sa))
325 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
330 /* Connect the socket to the remote endpoint. */
331 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
332 opt.connect_timeout) < 0)
338 logprintf (LOG_VERBOSE, _("connected.\n"));
339 DEBUGP (("Created socket %d.\n", sock));
344 /* Protect errno from possible modifications by close and
346 int save_errno = errno;
350 logprintf (LOG_VERBOSE, _("failed: %s.\n"), strerror (errno));
356 /* Connect via TCP to a remote host on the specified port.
358 HOST is resolved as an Internet host name. If HOST resolves to
359 more than one IP address, they are tried in the order returned by
360 DNS until connecting to one of them succeeds. */
363 connect_to_host (const char *host, int port)
368 struct address_list *al = lookup_host (host, 0);
373 logprintf (LOG_NOTQUIET,
374 _("%s: unable to resolve host address %s\n"),
375 exec_name, quote (host));
379 address_list_get_bounds (al, &start, &end);
380 for (i = start; i < end; i++)
382 const ip_address *ip = address_list_address_at (al, i);
383 sock = connect_to_ip (ip, port, host);
387 address_list_set_connected (al);
388 address_list_release (al);
392 /* The attempt to connect has failed. Continue with the loop
393 and try next address. */
395 address_list_set_faulty (al, i);
398 /* Failed to connect to any of the addresses in AL. */
400 if (address_list_connected_p (al))
402 /* We connected to AL before, but cannot do so now. That might
403 indicate that our DNS cache entry for HOST has expired. */
404 address_list_release (al);
405 al = lookup_host (host, LH_REFRESH);
408 address_list_release (al);
413 /* Create a socket, bind it to local interface BIND_ADDRESS on port
414 *PORT, set up a listen backlog, and return the resulting socket, or
417 BIND_ADDRESS is the address of the interface to bind to. If it is
418 NULL, the socket is bound to the default address. PORT should
419 point to the port number that will be used for the binding. If
420 that number is 0, the system will choose a suitable port, and the
421 chosen value will be written to *PORT.
423 Calling accept() on such a socket waits for and accepts incoming
427 bind_local (const ip_address *bind_address, int *port)
430 struct sockaddr_storage ss;
431 struct sockaddr *sa = (struct sockaddr *)&ss;
433 /* For setting options with setsockopt. */
435 void *setopt_ptr = (void *)&setopt_val;
436 socklen_t setopt_size = sizeof (setopt_val);
438 sock = socket (bind_address->family, SOCK_STREAM, 0);
443 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
447 sockaddr_set_data (sa, bind_address, *port);
448 if (bind (sock, sa, sockaddr_size (sa)) < 0)
453 DEBUGP (("Local socket fd %d bound.\n", sock));
455 /* If *PORT is 0, find out which port we've bound to. */
458 socklen_t addrlen = sockaddr_size (sa);
459 if (getsockname (sock, sa, &addrlen) < 0)
461 /* If we can't find out the socket's local address ("name"),
462 something is seriously wrong with the socket, and it's
463 unusable for us anyway because we must know the chosen
468 sockaddr_get_data (sa, NULL, port);
469 DEBUGP (("binding to address %s using port %i.\n",
470 print_address (bind_address), *port));
472 if (listen (sock, 1) < 0)
480 /* Like a call to accept(), but with the added check for timeout.
482 In other words, accept a client connection on LOCAL_SOCK, and
483 return the new socket used for communication with the client.
484 LOCAL_SOCK should have been bound, e.g. using bind_local().
486 The caller is blocked until a connection is established. If no
487 connection is established for opt.connect_timeout seconds, the
488 function exits with an error status. */
491 accept_connection (int local_sock)
495 /* We don't need the values provided by accept, but accept
496 apparently requires them to be present. */
497 struct sockaddr_storage ss;
498 struct sockaddr *sa = (struct sockaddr *)&ss;
499 socklen_t addrlen = sizeof (ss);
501 if (opt.connect_timeout)
503 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
509 sock = accept (local_sock, sa, &addrlen);
510 DEBUGP (("Accepted client at socket %d.\n", sock));
514 /* Get the IP address associated with the connection on FD and store
515 it to IP. Return true on success, false otherwise.
517 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
518 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
519 returns the address of the remote (peer's) side of the socket. */
522 socket_ip_address (int sock, ip_address *ip, int endpoint)
524 struct sockaddr_storage storage;
525 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
526 socklen_t addrlen = sizeof (storage);
529 if (endpoint == ENDPOINT_LOCAL)
530 ret = getsockname (sock, sockaddr, &addrlen);
531 else if (endpoint == ENDPOINT_PEER)
532 ret = getpeername (sock, sockaddr, &addrlen);
538 ip->family = sockaddr->sa_family;
539 switch (sockaddr->sa_family)
544 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
545 ip->data.d6 = sa6->sin6_addr;
546 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
547 ip->ipv6_scope = sa6->sin6_scope_id;
549 DEBUGP (("conaddr is: %s\n", print_address (ip)));
555 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
556 ip->data.d4 = sa->sin_addr;
557 DEBUGP (("conaddr is: %s\n", print_address (ip)));
565 /* Return true if the error from the connect code can be considered
566 retryable. Wget normally retries after errors, but the exception
567 are the "unsupported protocol" type errors (possible on IPv4/IPv6
568 dual family systems) and "connection refused". */
571 retryable_socket_connect_error (int err)
573 /* Have to guard against some of these values not being defined.
574 Cannot use a switch statement because some of the values might be
578 || err == EAFNOSUPPORT
581 || err == EPFNOSUPPORT
583 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
584 || err == ESOCKTNOSUPPORT
586 #ifdef EPROTONOSUPPORT
587 || err == EPROTONOSUPPORT
590 || err == ENOPROTOOPT
592 /* Apparently, older versions of Linux and BSD used EINVAL
593 instead of EAFNOSUPPORT and such. */
598 if (!opt.retry_connrefused)
599 if (err == ECONNREFUSED
601 || err == ENETUNREACH /* network is unreachable */
604 || err == EHOSTUNREACH /* host is unreachable */
612 /* Wait for a single descriptor to become available, timing out after
613 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
614 -1 for error. The argument WAIT_FOR can be a combination of
615 WAIT_FOR_READ and WAIT_FOR_WRITE.
617 This is a mere convenience wrapper around the select call, and
618 should be taken as such (for example, it doesn't implement Wget's
619 0-timeout-means-no-timeout semantics.) */
622 select_fd (int fd, double maxtime, int wait_for)
625 fd_set *rd = NULL, *wr = NULL;
626 struct timeval tmout;
631 if (wait_for & WAIT_FOR_READ)
633 if (wait_for & WAIT_FOR_WRITE)
636 tmout.tv_sec = (long) maxtime;
637 tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
640 result = select (fd + 1, rd, wr, NULL, &tmout);
641 while (result < 0 && errno == EINTR);
646 /* Return true iff the connection to the remote site established
647 through SOCK is still open.
649 Specifically, this function returns true if SOCK is not ready for
650 reading. This is because, when the connection closes, the socket
651 is ready for reading because EOF is about to be delivered. A side
652 effect of this method is that sockets that have pending data are
653 considered non-open. This is actually a good thing for callers of
654 this function, where such pending data can only be unwanted
655 leftover from a previous request. */
658 test_socket_open (int sock)
663 /* Check if we still have a valid (non-EOF) connection. From Andrew
664 * Maholski's code in the Unix Socket FAQ. */
666 FD_ZERO (&check_set);
667 FD_SET (sock, &check_set);
669 /* Wait one microsecond */
673 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
674 /* We got a timeout, it means we're still connected. */
677 /* Read now would not wait, it means we have either pending data
682 /* Basic socket operations, mostly EINTR wrappers. */
684 #if defined(WINDOWS) || defined(MSDOS)
685 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
686 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
687 # define close(fd) closesocket (fd)
691 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
692 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
696 sock_read (int fd, char *buf, int bufsize)
700 res = read (fd, buf, bufsize);
701 while (res == -1 && errno == EINTR);
706 sock_write (int fd, char *buf, int bufsize)
710 res = write (fd, buf, bufsize);
711 while (res == -1 && errno == EINTR);
716 sock_poll (int fd, double timeout, int wait_for)
718 return select_fd (fd, timeout, wait_for);
722 sock_peek (int fd, char *buf, int bufsize)
726 res = recv (fd, buf, bufsize, MSG_PEEK);
727 while (res == -1 && errno == EINTR);
735 DEBUGP (("Closed fd %d\n", fd));
741 /* Reading and writing from the network. We build around the socket
742 (file descriptor) API, but support "extended" operations for things
743 that are not mere file descriptors under the hood, such as SSL
746 That way the user code can call fd_read(fd, ...) and we'll run read
747 or SSL_read or whatever is necessary. */
749 static struct hash_table *transport_map;
750 static unsigned int transport_map_modified_tick;
752 struct transport_info {
753 struct transport_implementation *imp;
757 /* Register the transport layer operations that will be used when
758 reading, writing, and polling FD.
760 This should be used for transport layers like SSL that piggyback on
761 sockets. FD should otherwise be a real socket, on which you can
762 call getpeername, etc. */
765 fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
767 struct transport_info *info;
769 /* The file descriptor must be non-negative to be registered.
770 Negative values are ignored by fd_close(), and -1 cannot be used as
774 info = xnew (struct transport_info);
778 transport_map = hash_table_new (0, NULL, NULL);
779 hash_table_put (transport_map, (void *)(intptr_t) fd, info);
780 ++transport_map_modified_tick;
783 /* Return context of the transport registered with
784 fd_register_transport. This assumes fd_register_transport was
785 previously called on FD. */
788 fd_transport_context (int fd)
790 struct transport_info *info = hash_table_get (transport_map, (void *)(intptr_t) fd);
794 /* When fd_read/fd_write are called multiple times in a loop, they should
795 remember the INFO pointer instead of fetching it every time. It is
796 not enough to compare FD to LAST_FD because FD might have been
797 closed and reopened. modified_tick ensures that changes to
798 transport_map will not be unnoticed.
800 This is a macro because we want the static storage variables to be
803 #define LAZY_RETRIEVE_INFO(info) do { \
804 static struct transport_info *last_info; \
805 static int last_fd = -1; \
806 static unsigned int last_tick; \
807 if (!transport_map) \
809 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
813 info = hash_table_get (transport_map, (void *)(intptr_t) fd); \
816 last_tick = transport_map_modified_tick; \
821 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
824 timeout = opt.read_timeout;
828 if (info && info->imp->poller)
829 test = info->imp->poller (fd, timeout, wf, info->ctx);
831 test = sock_poll (fd, timeout, wf);
840 /* Read no more than BUFSIZE bytes of data from FD, storing them to
841 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
842 received after that many seconds. If TIMEOUT is -1, the value of
843 opt.timeout is used for TIMEOUT. */
846 fd_read (int fd, char *buf, int bufsize, double timeout)
848 struct transport_info *info;
849 LAZY_RETRIEVE_INFO (info);
850 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
852 if (info && info->imp->reader)
853 return info->imp->reader (fd, buf, bufsize, info->ctx);
855 return sock_read (fd, buf, bufsize);
858 /* Like fd_read, except it provides a "preview" of the data that will
859 be read by subsequent calls to fd_read. Specifically, it copies no
860 more than BUFSIZE bytes of the currently available data to BUF and
861 returns the number of bytes copied. Return values and timeout
862 semantics are the same as those of fd_read.
864 CAVEAT: Do not assume that the first subsequent call to fd_read
865 will retrieve the same amount of data. Reading can return more or
866 less data, depending on the TCP implementation and other
867 circumstances. However, barring an error, it can be expected that
868 all the peeked data will eventually be read by fd_read. */
871 fd_peek (int fd, char *buf, int bufsize, double timeout)
873 struct transport_info *info;
874 LAZY_RETRIEVE_INFO (info);
875 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
877 if (info && info->imp->peeker)
878 return info->imp->peeker (fd, buf, bufsize, info->ctx);
880 return sock_peek (fd, buf, bufsize);
883 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
884 the operation aborts if no data is received after that many
885 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
889 fd_write (int fd, char *buf, int bufsize, double timeout)
892 struct transport_info *info;
893 LAZY_RETRIEVE_INFO (info);
895 /* `write' may write less than LEN bytes, thus the loop keeps trying
896 it until all was written, or an error occurred. */
900 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
902 if (info && info->imp->writer)
903 res = info->imp->writer (fd, buf, bufsize, info->ctx);
905 res = sock_write (fd, buf, bufsize);
914 /* Report the most recent error(s) on FD. This should only be called
915 after fd_* functions, such as fd_read and fd_write, and only if
916 they return a negative result. For errors coming from other calls
917 such as setsockopt or fopen, strerror should continue to be
920 If the transport doesn't support error messages or doesn't supply
921 one, strerror(errno) is returned. The returned error message
922 should not be used after fd_close has been called. */
927 /* Don't bother with LAZY_RETRIEVE_INFO, as this will only be called
928 in case of error, never in a tight loop. */
929 struct transport_info *info = NULL;
931 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
933 if (info && info->imp->errstr)
935 const char *err = info->imp->errstr (fd, info->ctx);
938 /* else, fall through and print the system error. */
940 return strerror (errno);
943 /* Close the file descriptor FD. */
948 struct transport_info *info;
952 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
953 per socket, so that particular optimization wouldn't work. */
956 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
958 if (info && info->imp->closer)
959 info->imp->closer (fd, info->ctx);
965 hash_table_remove (transport_map, (void *)(intptr_t) fd);
967 ++transport_map_modified_tick;