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. */
31 #define USE_GNULIB_ALLOC
43 # include <sys/socket.h>
45 # include <netinet/in.h>
47 # include <arpa/inet.h>
49 #endif /* not WINDOWS */
53 #ifdef HAVE_SYS_SELECT_H
54 # include <sys/select.h>
55 #endif /* HAVE_SYS_SELECT_H */
56 #ifdef HAVE_SYS_TIME_H
57 # include <sys/time.h>
64 /* Define sockaddr_storage where unavailable (presumably on IPv4-only
68 # ifndef HAVE_STRUCT_SOCKADDR_STORAGE
69 # define sockaddr_storage sockaddr_in
71 #endif /* ENABLE_IPV6 */
73 /* Fill SA as per the data in IP and PORT. SA shoult point to struct
74 sockaddr_storage if ENABLE_IPV6 is defined, to struct sockaddr_in
78 sockaddr_set_data (struct sockaddr *sa, const ip_address *ip, int port)
84 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
86 sin->sin_family = AF_INET;
87 sin->sin_port = htons (port);
88 sin->sin_addr = ip->data.d4;
94 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
96 sin6->sin6_family = AF_INET6;
97 sin6->sin6_port = htons (port);
98 sin6->sin6_addr = ip->data.d6;
99 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
100 sin6->sin6_scope_id = ip->ipv6_scope;
104 #endif /* ENABLE_IPV6 */
110 /* Get the data of SA, specifically the IP address and the port. If
111 you're not interested in one or the other information, pass NULL as
115 sockaddr_get_data (const struct sockaddr *sa, ip_address *ip, int *port)
117 switch (sa->sa_family)
121 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
124 ip->family = AF_INET;
125 ip->data.d4 = sin->sin_addr;
128 *port = ntohs (sin->sin_port);
134 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
137 ip->family = AF_INET6;
138 ip->data.d6 = sin6->sin6_addr;
139 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
140 ip->ipv6_scope = sin6->sin6_scope_id;
144 *port = ntohs (sin6->sin6_port);
153 /* Return the size of the sockaddr structure depending on its
157 sockaddr_size (const struct sockaddr *sa)
159 switch (sa->sa_family)
162 return sizeof (struct sockaddr_in);
165 return sizeof (struct sockaddr_in6);
172 /* Resolve the bind address specified via --bind-address and store it
173 to SA. The resolved value is stored in a static variable and
174 reused after the first invocation of this function.
176 Returns true on success, false on failure. */
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 bool 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, quote (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 = 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)));
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);
370 logprintf (LOG_NOTQUIET,
371 _("%s: unable to resolve host address %s\n"),
372 exec_name, quote (host));
376 address_list_get_bounds (al, &start, &end);
377 for (i = start; i < end; i++)
379 const ip_address *ip = address_list_address_at (al, i);
380 sock = connect_to_ip (ip, port, host);
384 address_list_set_connected (al);
385 address_list_release (al);
389 /* The attempt to connect has failed. Continue with the loop
390 and try next address. */
392 address_list_set_faulty (al, i);
395 /* Failed to connect to any of the addresses in AL. */
397 if (address_list_connected_p (al))
399 /* We connected to AL before, but cannot do so now. That might
400 indicate that our DNS cache entry for HOST has expired. */
401 address_list_release (al);
402 al = lookup_host (host, LH_REFRESH);
405 address_list_release (al);
410 /* Create a socket, bind it to local interface BIND_ADDRESS on port
411 *PORT, set up a listen backlog, and return the resulting socket, or
414 BIND_ADDRESS is the address of the interface to bind to. If it is
415 NULL, the socket is bound to the default address. PORT should
416 point to the port number that will be used for the binding. If
417 that number is 0, the system will choose a suitable port, and the
418 chosen value will be written to *PORT.
420 Calling accept() on such a socket waits for and accepts incoming
424 bind_local (const ip_address *bind_address, int *port)
427 struct sockaddr_storage ss;
428 struct sockaddr *sa = (struct sockaddr *)&ss;
430 /* For setting options with setsockopt. */
432 void *setopt_ptr = (void *)&setopt_val;
433 socklen_t setopt_size = sizeof (setopt_val);
435 sock = socket (bind_address->family, SOCK_STREAM, 0);
440 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
444 sockaddr_set_data (sa, bind_address, *port);
445 if (bind (sock, sa, sockaddr_size (sa)) < 0)
450 DEBUGP (("Local socket fd %d bound.\n", sock));
452 /* If *PORT is 0, find out which port we've bound to. */
455 socklen_t addrlen = sockaddr_size (sa);
456 if (getsockname (sock, sa, &addrlen) < 0)
458 /* If we can't find out the socket's local address ("name"),
459 something is seriously wrong with the socket, and it's
460 unusable for us anyway because we must know the chosen
465 sockaddr_get_data (sa, NULL, port);
466 DEBUGP (("binding to address %s using port %i.\n",
467 print_address (bind_address), *port));
469 if (listen (sock, 1) < 0)
477 /* Like a call to accept(), but with the added check for timeout.
479 In other words, accept a client connection on LOCAL_SOCK, and
480 return the new socket used for communication with the client.
481 LOCAL_SOCK should have been bound, e.g. using bind_local().
483 The caller is blocked until a connection is established. If no
484 connection is established for opt.connect_timeout seconds, the
485 function exits with an error status. */
488 accept_connection (int local_sock)
492 /* We don't need the values provided by accept, but accept
493 apparently requires them to be present. */
494 struct sockaddr_storage ss;
495 struct sockaddr *sa = (struct sockaddr *)&ss;
496 socklen_t addrlen = sizeof (ss);
498 if (opt.connect_timeout)
500 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
506 sock = accept (local_sock, sa, &addrlen);
507 DEBUGP (("Accepted client at socket %d.\n", sock));
511 /* Get the IP address associated with the connection on FD and store
512 it to IP. Return true on success, false otherwise.
514 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
515 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
516 returns the address of the remote (peer's) side of the socket. */
519 socket_ip_address (int sock, ip_address *ip, int endpoint)
521 struct sockaddr_storage storage;
522 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
523 socklen_t addrlen = sizeof (storage);
526 if (endpoint == ENDPOINT_LOCAL)
527 ret = getsockname (sock, sockaddr, &addrlen);
528 else if (endpoint == ENDPOINT_PEER)
529 ret = getpeername (sock, sockaddr, &addrlen);
535 ip->family = sockaddr->sa_family;
536 switch (sockaddr->sa_family)
541 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
542 ip->data.d6 = sa6->sin6_addr;
543 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
544 ip->ipv6_scope = sa6->sin6_scope_id;
546 DEBUGP (("conaddr is: %s\n", print_address (ip)));
552 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
553 ip->data.d4 = sa->sin_addr;
554 DEBUGP (("conaddr is: %s\n", print_address (ip)));
562 /* Return true if the error from the connect code can be considered
563 retryable. Wget normally retries after errors, but the exception
564 are the "unsupported protocol" type errors (possible on IPv4/IPv6
565 dual family systems) and "connection refused". */
568 retryable_socket_connect_error (int err)
570 /* Have to guard against some of these values not being defined.
571 Cannot use a switch statement because some of the values might be
575 || err == EAFNOSUPPORT
578 || err == EPFNOSUPPORT
580 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
581 || err == ESOCKTNOSUPPORT
583 #ifdef EPROTONOSUPPORT
584 || err == EPROTONOSUPPORT
587 || err == ENOPROTOOPT
589 /* Apparently, older versions of Linux and BSD used EINVAL
590 instead of EAFNOSUPPORT and such. */
595 if (!opt.retry_connrefused)
596 if (err == ECONNREFUSED
598 || err == ENETUNREACH /* network is unreachable */
601 || err == EHOSTUNREACH /* host is unreachable */
609 /* Wait for a single descriptor to become available, timing out after
610 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
611 -1 for error. The argument WAIT_FOR can be a combination of
612 WAIT_FOR_READ and WAIT_FOR_WRITE.
614 This is a mere convenience wrapper around the select call, and
615 should be taken as such (for example, it doesn't implement Wget's
616 0-timeout-means-no-timeout semantics.) */
619 select_fd (int fd, double maxtime, int wait_for)
622 fd_set *rd = NULL, *wr = NULL;
623 struct timeval tmout;
628 if (wait_for & WAIT_FOR_READ)
630 if (wait_for & WAIT_FOR_WRITE)
633 tmout.tv_sec = (long) maxtime;
634 tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
637 result = select (fd + 1, rd, wr, NULL, &tmout);
638 while (result < 0 && errno == EINTR);
643 /* Return true iff the connection to the remote site established
644 through SOCK is still open.
646 Specifically, this function returns true if SOCK is not ready for
647 reading. This is because, when the connection closes, the socket
648 is ready for reading because EOF is about to be delivered. A side
649 effect of this method is that sockets that have pending data are
650 considered non-open. This is actually a good thing for callers of
651 this function, where such pending data can only be unwanted
652 leftover from a previous request. */
655 test_socket_open (int sock)
660 /* Check if we still have a valid (non-EOF) connection. From Andrew
661 * Maholski's code in the Unix Socket FAQ. */
663 FD_ZERO (&check_set);
664 FD_SET (sock, &check_set);
666 /* Wait one microsecond */
670 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
671 /* We got a timeout, it means we're still connected. */
674 /* Read now would not wait, it means we have either pending data
679 /* Basic socket operations, mostly EINTR wrappers. */
681 #if defined(WINDOWS) || defined(MSDOS)
682 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
683 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
684 # define close(fd) closesocket (fd)
688 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
689 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
693 sock_read (int fd, char *buf, int bufsize)
697 res = read (fd, buf, bufsize);
698 while (res == -1 && errno == EINTR);
703 sock_write (int fd, char *buf, int bufsize)
707 res = write (fd, buf, bufsize);
708 while (res == -1 && errno == EINTR);
713 sock_poll (int fd, double timeout, int wait_for)
715 return select_fd (fd, timeout, wait_for);
719 sock_peek (int fd, char *buf, int bufsize)
723 res = recv (fd, buf, bufsize, MSG_PEEK);
724 while (res == -1 && errno == EINTR);
732 DEBUGP (("Closed fd %d\n", fd));
738 /* Reading and writing from the network. We build around the socket
739 (file descriptor) API, but support "extended" operations for things
740 that are not mere file descriptors under the hood, such as SSL
743 That way the user code can call fd_read(fd, ...) and we'll run read
744 or SSL_read or whatever is necessary. */
746 static struct hash_table *transport_map;
747 static unsigned int transport_map_modified_tick;
749 struct transport_info {
750 struct transport_implementation *imp;
754 /* Register the transport layer operations that will be used when
755 reading, writing, and polling FD.
757 This should be used for transport layers like SSL that piggyback on
758 sockets. FD should otherwise be a real socket, on which you can
759 call getpeername, etc. */
762 fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
764 struct transport_info *info;
766 /* The file descriptor must be non-negative to be registered.
767 Negative values are ignored by fd_close(), and -1 cannot be used as
771 info = xnew (struct transport_info);
775 transport_map = hash_table_new (0, NULL, NULL);
776 hash_table_put (transport_map, (void *)(intptr_t) fd, info);
777 ++transport_map_modified_tick;
780 /* Return context of the transport registered with
781 fd_register_transport. This assumes fd_register_transport was
782 previously called on FD. */
785 fd_transport_context (int fd)
787 struct transport_info *info = hash_table_get (transport_map, (void *)(intptr_t) fd);
791 /* When fd_read/fd_write are called multiple times in a loop, they should
792 remember the INFO pointer instead of fetching it every time. It is
793 not enough to compare FD to LAST_FD because FD might have been
794 closed and reopened. modified_tick ensures that changes to
795 transport_map will not be unnoticed.
797 This is a macro because we want the static storage variables to be
800 #define LAZY_RETRIEVE_INFO(info) do { \
801 static struct transport_info *last_info; \
802 static int last_fd = -1; \
803 static unsigned int last_tick; \
804 if (!transport_map) \
806 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
810 info = hash_table_get (transport_map, (void *)(intptr_t) fd); \
813 last_tick = transport_map_modified_tick; \
818 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
821 timeout = opt.read_timeout;
825 if (info && info->imp->poller)
826 test = info->imp->poller (fd, timeout, wf, info->ctx);
828 test = sock_poll (fd, timeout, wf);
837 /* Read no more than BUFSIZE bytes of data from FD, storing them to
838 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
839 received after that many seconds. If TIMEOUT is -1, the value of
840 opt.timeout is used for TIMEOUT. */
843 fd_read (int fd, char *buf, int bufsize, double timeout)
845 struct transport_info *info;
846 LAZY_RETRIEVE_INFO (info);
847 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
849 if (info && info->imp->reader)
850 return info->imp->reader (fd, buf, bufsize, info->ctx);
852 return sock_read (fd, buf, bufsize);
855 /* Like fd_read, except it provides a "preview" of the data that will
856 be read by subsequent calls to fd_read. Specifically, it copies no
857 more than BUFSIZE bytes of the currently available data to BUF and
858 returns the number of bytes copied. Return values and timeout
859 semantics are the same as those of fd_read.
861 CAVEAT: Do not assume that the first subsequent call to fd_read
862 will retrieve the same amount of data. Reading can return more or
863 less data, depending on the TCP implementation and other
864 circumstances. However, barring an error, it can be expected that
865 all the peeked data will eventually be read by fd_read. */
868 fd_peek (int fd, char *buf, int bufsize, double timeout)
870 struct transport_info *info;
871 LAZY_RETRIEVE_INFO (info);
872 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
874 if (info && info->imp->peeker)
875 return info->imp->peeker (fd, buf, bufsize, info->ctx);
877 return sock_peek (fd, buf, bufsize);
880 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
881 the operation aborts if no data is received after that many
882 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
886 fd_write (int fd, char *buf, int bufsize, double timeout)
889 struct transport_info *info;
890 LAZY_RETRIEVE_INFO (info);
892 /* `write' may write less than LEN bytes, thus the loop keeps trying
893 it until all was written, or an error occurred. */
897 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
899 if (info && info->imp->writer)
900 res = info->imp->writer (fd, buf, bufsize, info->ctx);
902 res = sock_write (fd, buf, bufsize);
911 /* Report the most recent error(s) on FD. This should only be called
912 after fd_* functions, such as fd_read and fd_write, and only if
913 they return a negative result. For errors coming from other calls
914 such as setsockopt or fopen, strerror should continue to be
917 If the transport doesn't support error messages or doesn't supply
918 one, strerror(errno) is returned. The returned error message
919 should not be used after fd_close has been called. */
924 /* Don't bother with LAZY_RETRIEVE_INFO, as this will only be called
925 in case of error, never in a tight loop. */
926 struct transport_info *info = NULL;
928 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
930 if (info && info->imp->errstr)
932 const char *err = info->imp->errstr (fd, info->ctx);
935 /* else, fall through and print the system error. */
937 return strerror (errno);
940 /* Close the file descriptor FD. */
945 struct transport_info *info;
949 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
950 per socket, so that particular optimization wouldn't work. */
953 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
955 if (info && info->imp->closer)
956 info->imp->closer (fd, info->ctx);
962 hash_table_remove (transport_map, (void *)(intptr_t) fd);
964 ++transport_map_modified_tick;