1 /* Establishing and handling network connections.
2 Copyright (C) 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011 Free Software
6 This file is part of GNU Wget.
8 GNU Wget is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 3 of the License, or
11 (at your option) any later version.
13 GNU Wget is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with Wget. If not, see <http://www.gnu.org/licenses/>.
21 Additional permission under GNU GPL version 3 section 7
23 If you modify this program, or any covered work, by linking or
24 combining it with the OpenSSL project's OpenSSL library (or a
25 modified version of that library), containing parts covered by the
26 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
27 grants you additional permission to convey the resulting work.
28 Corresponding Source for a non-source form of such a combination
29 shall include the source code for the parts of OpenSSL used as well
30 as that of the covered work. */
39 #include <sys/socket.h>
40 #include <sys/select.h>
45 # else /* def __VMS */
47 # endif /* def __VMS [else] */
48 # include <netinet/in.h>
50 # include <arpa/inet.h>
52 #endif /* not WINDOWS */
56 #ifdef HAVE_SYS_TIME_H
57 # include <sys/time.h>
64 /* Apparently needed for Interix: */
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 = ip->data.d4;
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 = ip->data.d6;
104 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
105 sin6->sin6_scope_id = ip->ipv6_scope;
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->family = AF_INET;
130 ip->data.d4 = sin->sin_addr;
133 *port = ntohs (sin->sin_port);
139 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
142 ip->family = AF_INET6;
143 ip->data.d6 = sin6->sin6_addr;
144 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
145 ip->ipv6_scope = 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);
177 /* Resolve the bind address specified via --bind-address and store it
178 to SA. The resolved value is stored in a static variable and
179 reused after the first invocation of this function.
181 Returns true on success, false on failure. */
184 resolve_bind_address (struct sockaddr *sa)
186 struct address_list *al;
188 /* Make sure this is called only once. opt.bind_address doesn't
189 change during a Wget run. */
190 static bool called, should_bind;
191 static ip_address ip;
195 sockaddr_set_data (sa, &ip, 0);
200 al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
203 /* #### We should be able to print the error message here. */
204 logprintf (LOG_NOTQUIET,
205 _("%s: unable to resolve bind address %s; disabling bind.\n"),
206 exec_name, quote (opt.bind_address));
211 /* Pick the first address in the list and use it as bind address.
212 Perhaps we should try multiple addresses in succession, but I
213 don't think that's necessary in practice. */
214 ip = *address_list_address_at (al, 0);
215 address_list_release (al);
217 sockaddr_set_data (sa, &ip, 0);
224 const struct sockaddr *addr;
230 connect_with_timeout_callback (void *arg)
232 struct cwt_context *ctx = (struct cwt_context *)arg;
233 ctx->result = connect (ctx->fd, ctx->addr, ctx->addrlen);
236 /* Like connect, but specifies a timeout. If connecting takes longer
237 than TIMEOUT seconds, -1 is returned and errno is set to
241 connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
244 struct cwt_context ctx;
247 ctx.addrlen = addrlen;
249 if (run_with_timeout (timeout, connect_with_timeout_callback, &ctx))
254 if (ctx.result == -1 && errno == EINTR)
259 /* Connect via TCP to the specified address and port.
261 If PRINT is non-NULL, it is the host name to print that we're
265 connect_to_ip (const ip_address *ip, int port, const char *print)
267 struct sockaddr_storage ss;
268 struct sockaddr *sa = (struct sockaddr *)&ss;
271 /* If PRINT is non-NULL, print the "Connecting to..." line, with
272 PRINT being the host name we're connecting to. */
275 const char *txt_addr = print_address (ip);
276 if (0 != strcmp (print, txt_addr))
278 char *str = NULL, *name;
280 if (opt.enable_iri && (name = idn_decode ((char *) print)) != NULL)
282 int len = strlen (print) + strlen (name) + 4;
284 snprintf (str, len, "%s (%s)", name, print);
289 logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
290 str ? str : escnonprint_uri (print), txt_addr, port);
296 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
299 /* Store the sockaddr info to SA. */
300 sockaddr_set_data (sa, ip, port);
302 /* Create the socket of the family appropriate for the address. */
303 sock = socket (sa->sa_family, SOCK_STREAM, 0);
307 #if defined(ENABLE_IPV6) && defined(IPV6_V6ONLY)
310 /* In case of error, we will go on anyway... */
311 int err = setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof (on));
314 DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
318 /* For very small rate limits, set the buffer size (and hence,
319 hopefully, the kernel's TCP window size) to the per-second limit.
320 That way we should never have to sleep for more than 1s between
322 if (opt.limit_rate && opt.limit_rate < 8192)
324 int bufsize = opt.limit_rate;
326 bufsize = 512; /* avoid pathologically small values */
328 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
329 (void *)&bufsize, (socklen_t)sizeof (bufsize));
331 /* When we add limit_rate support for writing, which is useful
332 for POST, we should also set SO_SNDBUF here. */
335 if (opt.bind_address)
337 /* Bind the client side of the socket to the requested
339 struct sockaddr_storage bind_ss;
340 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
341 if (resolve_bind_address (bind_sa))
343 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
348 /* Connect the socket to the remote endpoint. */
349 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
350 opt.connect_timeout) < 0)
356 logprintf (LOG_VERBOSE, _("connected.\n"));
357 DEBUGP (("Created socket %d.\n", sock));
362 /* Protect errno from possible modifications by close and
364 int save_errno = errno;
368 logprintf (LOG_VERBOSE, _("failed: %s.\n"), strerror (errno));
374 /* Connect via TCP to a remote host on the specified port.
376 HOST is resolved as an Internet host name. If HOST resolves to
377 more than one IP address, they are tried in the order returned by
378 DNS until connecting to one of them succeeds. */
381 connect_to_host (const char *host, int port)
386 struct address_list *al = lookup_host (host, 0);
391 logprintf (LOG_NOTQUIET,
392 _("%s: unable to resolve host address %s\n"),
393 exec_name, quote (host));
397 address_list_get_bounds (al, &start, &end);
398 for (i = start; i < end; i++)
400 const ip_address *ip = address_list_address_at (al, i);
401 sock = connect_to_ip (ip, port, host);
405 address_list_set_connected (al);
406 address_list_release (al);
410 /* The attempt to connect has failed. Continue with the loop
411 and try next address. */
413 address_list_set_faulty (al, i);
416 /* Failed to connect to any of the addresses in AL. */
418 if (address_list_connected_p (al))
420 /* We connected to AL before, but cannot do so now. That might
421 indicate that our DNS cache entry for HOST has expired. */
422 address_list_release (al);
423 al = lookup_host (host, LH_REFRESH);
426 address_list_release (al);
431 /* Create a socket, bind it to local interface BIND_ADDRESS on port
432 *PORT, set up a listen backlog, and return the resulting socket, or
435 BIND_ADDRESS is the address of the interface to bind to. If it is
436 NULL, the socket is bound to the default address. PORT should
437 point to the port number that will be used for the binding. If
438 that number is 0, the system will choose a suitable port, and the
439 chosen value will be written to *PORT.
441 Calling accept() on such a socket waits for and accepts incoming
445 bind_local (const ip_address *bind_address, int *port)
448 struct sockaddr_storage ss;
449 struct sockaddr *sa = (struct sockaddr *)&ss;
451 /* For setting options with setsockopt. */
453 void *setopt_ptr = (void *)&setopt_val;
454 socklen_t setopt_size = sizeof (setopt_val);
456 sock = socket (bind_address->family, SOCK_STREAM, 0);
461 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
465 sockaddr_set_data (sa, bind_address, *port);
466 if (bind (sock, sa, sockaddr_size (sa)) < 0)
471 DEBUGP (("Local socket fd %d bound.\n", sock));
473 /* If *PORT is 0, find out which port we've bound to. */
476 socklen_t addrlen = sockaddr_size (sa);
477 if (getsockname (sock, sa, &addrlen) < 0)
479 /* If we can't find out the socket's local address ("name"),
480 something is seriously wrong with the socket, and it's
481 unusable for us anyway because we must know the chosen
486 sockaddr_get_data (sa, NULL, port);
487 DEBUGP (("binding to address %s using port %i.\n",
488 print_address (bind_address), *port));
490 if (listen (sock, 1) < 0)
498 /* Like a call to accept(), but with the added check for timeout.
500 In other words, accept a client connection on LOCAL_SOCK, and
501 return the new socket used for communication with the client.
502 LOCAL_SOCK should have been bound, e.g. using bind_local().
504 The caller is blocked until a connection is established. If no
505 connection is established for opt.connect_timeout seconds, the
506 function exits with an error status. */
509 accept_connection (int local_sock)
513 /* We don't need the values provided by accept, but accept
514 apparently requires them to be present. */
515 struct sockaddr_storage ss;
516 struct sockaddr *sa = (struct sockaddr *)&ss;
517 socklen_t addrlen = sizeof (ss);
519 if (opt.connect_timeout)
521 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
527 sock = accept (local_sock, sa, &addrlen);
528 DEBUGP (("Accepted client at socket %d.\n", sock));
532 /* Get the IP address associated with the connection on FD and store
533 it to IP. Return true on success, false otherwise.
535 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
536 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
537 returns the address of the remote (peer's) side of the socket. */
540 socket_ip_address (int sock, ip_address *ip, int endpoint)
542 struct sockaddr_storage storage;
543 struct sockaddr *sockaddr = (struct sockaddr *) &storage;
544 socklen_t addrlen = sizeof (storage);
547 memset (sockaddr, 0, addrlen);
548 if (endpoint == ENDPOINT_LOCAL)
549 ret = getsockname (sock, sockaddr, &addrlen);
550 else if (endpoint == ENDPOINT_PEER)
551 ret = getpeername (sock, sockaddr, &addrlen);
557 ip->family = sockaddr->sa_family;
558 switch (sockaddr->sa_family)
563 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
564 ip->data.d6 = sa6->sin6_addr;
565 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
566 ip->ipv6_scope = sa6->sin6_scope_id;
568 DEBUGP (("conaddr is: %s\n", print_address (ip)));
574 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
575 ip->data.d4 = sa->sin_addr;
576 DEBUGP (("conaddr is: %s\n", print_address (ip)));
584 /* Return true if the error from the connect code can be considered
585 retryable. Wget normally retries after errors, but the exception
586 are the "unsupported protocol" type errors (possible on IPv4/IPv6
587 dual family systems) and "connection refused". */
590 retryable_socket_connect_error (int err)
592 /* Have to guard against some of these values not being defined.
593 Cannot use a switch statement because some of the values might be
597 || err == EAFNOSUPPORT
600 || err == EPFNOSUPPORT
602 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
603 || err == ESOCKTNOSUPPORT
605 #ifdef EPROTONOSUPPORT
606 || err == EPROTONOSUPPORT
609 || err == ENOPROTOOPT
611 /* Apparently, older versions of Linux and BSD used EINVAL
612 instead of EAFNOSUPPORT and such. */
617 if (!opt.retry_connrefused)
618 if (err == ECONNREFUSED
620 || err == ENETUNREACH /* network is unreachable */
623 || err == EHOSTUNREACH /* host is unreachable */
631 /* Wait for a single descriptor to become available, timing out after
632 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
633 -1 for error. The argument WAIT_FOR can be a combination of
634 WAIT_FOR_READ and WAIT_FOR_WRITE.
636 This is a mere convenience wrapper around the select call, and
637 should be taken as such (for example, it doesn't implement Wget's
638 0-timeout-means-no-timeout semantics.) */
641 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 = 1000000 * (maxtime - (long) maxtime);
660 result = select (fd + 1, rd, wr, NULL, &tmout);
662 /* gnulib select() converts blocking sockets to nonblocking in windows.
663 wget uses blocking sockets so we must convert them back to blocking. */
664 set_windows_fd_as_blocking_socket (fd);
667 while (result < 0 && errno == EINTR);
672 /* Return true iff the connection to the remote site established
673 through SOCK is still open.
675 Specifically, this function returns true if SOCK is not ready for
676 reading. This is because, when the connection closes, the socket
677 is ready for reading because EOF is about to be delivered. A side
678 effect of this method is that sockets that have pending data are
679 considered non-open. This is actually a good thing for callers of
680 this function, where such pending data can only be unwanted
681 leftover from a previous request. */
684 test_socket_open (int sock)
690 /* Check if we still have a valid (non-EOF) connection. From Andrew
691 * Maholski's code in the Unix Socket FAQ. */
693 FD_ZERO (&check_set);
694 FD_SET (sock, &check_set);
696 /* Wait one microsecond */
700 ret = select (sock + 1, &check_set, NULL, NULL, &to);
702 /* gnulib select() converts blocking sockets to nonblocking in windows.
703 wget uses blocking sockets so we must convert them back to blocking
705 set_windows_fd_as_blocking_socket ( sock );
709 /* We got a timeout, it means we're still connected. */
712 /* Read now would not wait, it means we have either pending data
717 /* Basic socket operations, mostly EINTR wrappers. */
720 sock_read (int fd, char *buf, int bufsize)
724 res = read (fd, buf, bufsize);
725 while (res == -1 && errno == EINTR);
730 sock_write (int fd, char *buf, int bufsize)
734 res = write (fd, buf, bufsize);
735 while (res == -1 && errno == EINTR);
740 sock_poll (int fd, double timeout, int wait_for)
742 return select_fd (fd, timeout, wait_for);
746 sock_peek (int fd, char *buf, int bufsize)
750 res = recv (fd, buf, bufsize, MSG_PEEK);
751 while (res == -1 && errno == EINTR);
759 DEBUGP (("Closed fd %d\n", fd));
765 /* Reading and writing from the network. We build around the socket
766 (file descriptor) API, but support "extended" operations for things
767 that are not mere file descriptors under the hood, such as SSL
770 That way the user code can call fd_read(fd, ...) and we'll run read
771 or SSL_read or whatever is necessary. */
773 static struct hash_table *transport_map;
774 static unsigned int transport_map_modified_tick;
776 struct transport_info {
777 struct transport_implementation *imp;
781 /* Register the transport layer operations that will be used when
782 reading, writing, and polling FD.
784 This should be used for transport layers like SSL that piggyback on
785 sockets. FD should otherwise be a real socket, on which you can
786 call getpeername, etc. */
789 fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
791 struct transport_info *info;
793 /* The file descriptor must be non-negative to be registered.
794 Negative values are ignored by fd_close(), and -1 cannot be used as
798 info = xnew (struct transport_info);
802 transport_map = hash_table_new (0, NULL, NULL);
803 hash_table_put (transport_map, (void *)(intptr_t) fd, info);
804 ++transport_map_modified_tick;
807 /* Return context of the transport registered with
808 fd_register_transport. This assumes fd_register_transport was
809 previously called on FD. */
812 fd_transport_context (int fd)
814 struct transport_info *info = hash_table_get (transport_map, (void *)(intptr_t) fd);
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; \
830 static unsigned int last_tick; \
831 if (!transport_map) \
833 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
837 info = hash_table_get (transport_map, (void *)(intptr_t) fd); \
840 last_tick = transport_map_modified_tick; \
845 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
848 timeout = opt.read_timeout;
852 if (info && info->imp->poller)
853 test = info->imp->poller (fd, timeout, wf, info->ctx);
855 test = sock_poll (fd, timeout, wf);
864 /* Read no more than BUFSIZE bytes of data from FD, storing them to
865 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
866 received after that many seconds. If TIMEOUT is -1, the value of
867 opt.timeout is used for TIMEOUT. */
870 fd_read (int fd, char *buf, int bufsize, double timeout)
872 struct transport_info *info;
873 LAZY_RETRIEVE_INFO (info);
874 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
876 if (info && info->imp->reader)
877 return info->imp->reader (fd, buf, bufsize, info->ctx);
879 return sock_read (fd, buf, bufsize);
882 /* Like fd_read, except it provides a "preview" of the data that will
883 be read by subsequent calls to fd_read. Specifically, it copies no
884 more than BUFSIZE bytes of the currently available data to BUF and
885 returns the number of bytes copied. Return values and timeout
886 semantics are the same as those of fd_read.
888 CAVEAT: Do not assume that the first subsequent call to fd_read
889 will retrieve the same amount of data. Reading can return more or
890 less data, depending on the TCP implementation and other
891 circumstances. However, barring an error, it can be expected that
892 all the peeked data will eventually be read by fd_read. */
895 fd_peek (int fd, char *buf, int bufsize, double timeout)
897 struct transport_info *info;
898 LAZY_RETRIEVE_INFO (info);
899 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
901 if (info && info->imp->peeker)
902 return info->imp->peeker (fd, buf, bufsize, info->ctx);
904 return sock_peek (fd, buf, bufsize);
907 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
908 the operation aborts if no data is received after that many
909 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
913 fd_write (int fd, char *buf, int bufsize, double timeout)
916 struct transport_info *info;
917 LAZY_RETRIEVE_INFO (info);
919 /* `write' may write less than LEN bytes, thus the loop keeps trying
920 it until all was written, or an error occurred. */
924 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
926 if (info && info->imp->writer)
927 res = info->imp->writer (fd, buf, bufsize, info->ctx);
929 res = sock_write (fd, buf, bufsize);
938 /* Report the most recent error(s) on FD. This should only be called
939 after fd_* functions, such as fd_read and fd_write, and only if
940 they return a negative result. For errors coming from other calls
941 such as setsockopt or fopen, strerror should continue to be
944 If the transport doesn't support error messages or doesn't supply
945 one, strerror(errno) is returned. The returned error message
946 should not be used after fd_close has been called. */
951 /* Don't bother with LAZY_RETRIEVE_INFO, as this will only be called
952 in case of error, never in a tight loop. */
953 struct transport_info *info = NULL;
955 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
957 if (info && info->imp->errstr)
959 const char *err = info->imp->errstr (fd, info->ctx);
962 /* else, fall through and print the system error. */
964 return strerror (errno);
967 /* Close the file descriptor FD. */
972 struct transport_info *info;
976 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
977 per socket, so that particular optimization wouldn't work. */
980 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
982 if (info && info->imp->closer)
983 info->imp->closer (fd, info->ctx);
989 hash_table_remove (transport_map, (void *)(intptr_t) fd);
991 ++transport_map_modified_tick;