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 (0 != strcmp (print, txt_addr))
271 char *str = NULL, *name;
273 if (opt.enable_iri && (name = idn_decode ((char *) print)) != NULL)
275 int len = strlen (print) + strlen (name) + 4;
277 snprintf (str, len, "%s (%s)", name, print);
282 logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
283 str ? str : escnonprint_uri (print), txt_addr, port);
289 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
292 /* Store the sockaddr info to SA. */
293 sockaddr_set_data (sa, ip, port);
295 /* Create the socket of the family appropriate for the address. */
296 sock = socket (sa->sa_family, SOCK_STREAM, 0);
300 #if defined(ENABLE_IPV6) && defined(IPV6_V6ONLY)
303 /* In case of error, we will go on anyway... */
304 int err = setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof (on));
307 DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
311 /* For very small rate limits, set the buffer size (and hence,
312 hopefully, the kernel's TCP window size) to the per-second limit.
313 That way we should never have to sleep for more than 1s between
315 if (opt.limit_rate && opt.limit_rate < 8192)
317 int bufsize = opt.limit_rate;
319 bufsize = 512; /* avoid pathologically small values */
321 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
322 (void *)&bufsize, (socklen_t)sizeof (bufsize));
324 /* When we add limit_rate support for writing, which is useful
325 for POST, we should also set SO_SNDBUF here. */
328 if (opt.bind_address)
330 /* Bind the client side of the socket to the requested
332 struct sockaddr_storage bind_ss;
333 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
334 if (resolve_bind_address (bind_sa))
336 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
341 /* Connect the socket to the remote endpoint. */
342 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
343 opt.connect_timeout) < 0)
349 logprintf (LOG_VERBOSE, _("connected.\n"));
350 DEBUGP (("Created socket %d.\n", sock));
355 /* Protect errno from possible modifications by close and
357 int save_errno = errno;
361 logprintf (LOG_VERBOSE, _("failed: %s.\n"), strerror (errno));
367 /* Connect via TCP to a remote host on the specified port.
369 HOST is resolved as an Internet host name. If HOST resolves to
370 more than one IP address, they are tried in the order returned by
371 DNS until connecting to one of them succeeds. */
374 connect_to_host (const char *host, int port)
379 struct address_list *al = lookup_host (host, 0);
384 logprintf (LOG_NOTQUIET,
385 _("%s: unable to resolve host address %s\n"),
386 exec_name, quote (host));
390 address_list_get_bounds (al, &start, &end);
391 for (i = start; i < end; i++)
393 const ip_address *ip = address_list_address_at (al, i);
394 sock = connect_to_ip (ip, port, host);
398 address_list_set_connected (al);
399 address_list_release (al);
403 /* The attempt to connect has failed. Continue with the loop
404 and try next address. */
406 address_list_set_faulty (al, i);
409 /* Failed to connect to any of the addresses in AL. */
411 if (address_list_connected_p (al))
413 /* We connected to AL before, but cannot do so now. That might
414 indicate that our DNS cache entry for HOST has expired. */
415 address_list_release (al);
416 al = lookup_host (host, LH_REFRESH);
419 address_list_release (al);
424 /* Create a socket, bind it to local interface BIND_ADDRESS on port
425 *PORT, set up a listen backlog, and return the resulting socket, or
428 BIND_ADDRESS is the address of the interface to bind to. If it is
429 NULL, the socket is bound to the default address. PORT should
430 point to the port number that will be used for the binding. If
431 that number is 0, the system will choose a suitable port, and the
432 chosen value will be written to *PORT.
434 Calling accept() on such a socket waits for and accepts incoming
438 bind_local (const ip_address *bind_address, int *port)
441 struct sockaddr_storage ss;
442 struct sockaddr *sa = (struct sockaddr *)&ss;
444 /* For setting options with setsockopt. */
446 void *setopt_ptr = (void *)&setopt_val;
447 socklen_t setopt_size = sizeof (setopt_val);
449 sock = socket (bind_address->family, SOCK_STREAM, 0);
454 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
458 sockaddr_set_data (sa, bind_address, *port);
459 if (bind (sock, sa, sockaddr_size (sa)) < 0)
464 DEBUGP (("Local socket fd %d bound.\n", sock));
466 /* If *PORT is 0, find out which port we've bound to. */
469 socklen_t addrlen = sockaddr_size (sa);
470 if (getsockname (sock, sa, &addrlen) < 0)
472 /* If we can't find out the socket's local address ("name"),
473 something is seriously wrong with the socket, and it's
474 unusable for us anyway because we must know the chosen
479 sockaddr_get_data (sa, NULL, port);
480 DEBUGP (("binding to address %s using port %i.\n",
481 print_address (bind_address), *port));
483 if (listen (sock, 1) < 0)
491 /* Like a call to accept(), but with the added check for timeout.
493 In other words, accept a client connection on LOCAL_SOCK, and
494 return the new socket used for communication with the client.
495 LOCAL_SOCK should have been bound, e.g. using bind_local().
497 The caller is blocked until a connection is established. If no
498 connection is established for opt.connect_timeout seconds, the
499 function exits with an error status. */
502 accept_connection (int local_sock)
506 /* We don't need the values provided by accept, but accept
507 apparently requires them to be present. */
508 struct sockaddr_storage ss;
509 struct sockaddr *sa = (struct sockaddr *)&ss;
510 socklen_t addrlen = sizeof (ss);
512 if (opt.connect_timeout)
514 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
520 sock = accept (local_sock, sa, &addrlen);
521 DEBUGP (("Accepted client at socket %d.\n", sock));
525 /* Get the IP address associated with the connection on FD and store
526 it to IP. Return true on success, false otherwise.
528 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
529 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
530 returns the address of the remote (peer's) side of the socket. */
533 socket_ip_address (int sock, ip_address *ip, int endpoint)
535 struct sockaddr_storage storage;
536 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
537 socklen_t addrlen = sizeof (storage);
540 if (endpoint == ENDPOINT_LOCAL)
541 ret = getsockname (sock, sockaddr, &addrlen);
542 else if (endpoint == ENDPOINT_PEER)
543 ret = getpeername (sock, sockaddr, &addrlen);
549 ip->family = sockaddr->sa_family;
550 switch (sockaddr->sa_family)
555 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
556 ip->data.d6 = sa6->sin6_addr;
557 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
558 ip->ipv6_scope = sa6->sin6_scope_id;
560 DEBUGP (("conaddr is: %s\n", print_address (ip)));
566 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
567 ip->data.d4 = sa->sin_addr;
568 DEBUGP (("conaddr is: %s\n", print_address (ip)));
576 /* Return true if the error from the connect code can be considered
577 retryable. Wget normally retries after errors, but the exception
578 are the "unsupported protocol" type errors (possible on IPv4/IPv6
579 dual family systems) and "connection refused". */
582 retryable_socket_connect_error (int err)
584 /* Have to guard against some of these values not being defined.
585 Cannot use a switch statement because some of the values might be
589 || err == EAFNOSUPPORT
592 || err == EPFNOSUPPORT
594 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
595 || err == ESOCKTNOSUPPORT
597 #ifdef EPROTONOSUPPORT
598 || err == EPROTONOSUPPORT
601 || err == ENOPROTOOPT
603 /* Apparently, older versions of Linux and BSD used EINVAL
604 instead of EAFNOSUPPORT and such. */
609 if (!opt.retry_connrefused)
610 if (err == ECONNREFUSED
612 || err == ENETUNREACH /* network is unreachable */
615 || err == EHOSTUNREACH /* host is unreachable */
623 /* Wait for a single descriptor to become available, timing out after
624 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
625 -1 for error. The argument WAIT_FOR can be a combination of
626 WAIT_FOR_READ and WAIT_FOR_WRITE.
628 This is a mere convenience wrapper around the select call, and
629 should be taken as such (for example, it doesn't implement Wget's
630 0-timeout-means-no-timeout semantics.) */
633 select_fd (int fd, double maxtime, int wait_for)
636 fd_set *rd = NULL, *wr = NULL;
637 struct timeval tmout;
642 if (wait_for & WAIT_FOR_READ)
644 if (wait_for & WAIT_FOR_WRITE)
647 tmout.tv_sec = (long) maxtime;
648 tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
651 result = select (fd + 1, rd, wr, NULL, &tmout);
652 while (result < 0 && errno == EINTR);
657 /* Return true iff the connection to the remote site established
658 through SOCK is still open.
660 Specifically, this function returns true if SOCK is not ready for
661 reading. This is because, when the connection closes, the socket
662 is ready for reading because EOF is about to be delivered. A side
663 effect of this method is that sockets that have pending data are
664 considered non-open. This is actually a good thing for callers of
665 this function, where such pending data can only be unwanted
666 leftover from a previous request. */
669 test_socket_open (int sock)
674 /* Check if we still have a valid (non-EOF) connection. From Andrew
675 * Maholski's code in the Unix Socket FAQ. */
677 FD_ZERO (&check_set);
678 FD_SET (sock, &check_set);
680 /* Wait one microsecond */
684 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
685 /* We got a timeout, it means we're still connected. */
688 /* Read now would not wait, it means we have either pending data
693 /* Basic socket operations, mostly EINTR wrappers. */
695 #if defined(WINDOWS) || defined(MSDOS)
696 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
697 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
698 # define close(fd) closesocket (fd)
702 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
703 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
707 sock_read (int fd, char *buf, int bufsize)
711 res = read (fd, buf, bufsize);
712 while (res == -1 && errno == EINTR);
717 sock_write (int fd, char *buf, int bufsize)
721 res = write (fd, buf, bufsize);
722 while (res == -1 && errno == EINTR);
727 sock_poll (int fd, double timeout, int wait_for)
729 return select_fd (fd, timeout, wait_for);
733 sock_peek (int fd, char *buf, int bufsize)
737 res = recv (fd, buf, bufsize, MSG_PEEK);
738 while (res == -1 && errno == EINTR);
746 DEBUGP (("Closed fd %d\n", fd));
752 /* Reading and writing from the network. We build around the socket
753 (file descriptor) API, but support "extended" operations for things
754 that are not mere file descriptors under the hood, such as SSL
757 That way the user code can call fd_read(fd, ...) and we'll run read
758 or SSL_read or whatever is necessary. */
760 static struct hash_table *transport_map;
761 static unsigned int transport_map_modified_tick;
763 struct transport_info {
764 struct transport_implementation *imp;
768 /* Register the transport layer operations that will be used when
769 reading, writing, and polling FD.
771 This should be used for transport layers like SSL that piggyback on
772 sockets. FD should otherwise be a real socket, on which you can
773 call getpeername, etc. */
776 fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
778 struct transport_info *info;
780 /* The file descriptor must be non-negative to be registered.
781 Negative values are ignored by fd_close(), and -1 cannot be used as
785 info = xnew (struct transport_info);
789 transport_map = hash_table_new (0, NULL, NULL);
790 hash_table_put (transport_map, (void *)(intptr_t) fd, info);
791 ++transport_map_modified_tick;
794 /* Return context of the transport registered with
795 fd_register_transport. This assumes fd_register_transport was
796 previously called on FD. */
799 fd_transport_context (int fd)
801 struct transport_info *info = hash_table_get (transport_map, (void *)(intptr_t) fd);
805 /* When fd_read/fd_write are called multiple times in a loop, they should
806 remember the INFO pointer instead of fetching it every time. It is
807 not enough to compare FD to LAST_FD because FD might have been
808 closed and reopened. modified_tick ensures that changes to
809 transport_map will not be unnoticed.
811 This is a macro because we want the static storage variables to be
814 #define LAZY_RETRIEVE_INFO(info) do { \
815 static struct transport_info *last_info; \
816 static int last_fd = -1; \
817 static unsigned int last_tick; \
818 if (!transport_map) \
820 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
824 info = hash_table_get (transport_map, (void *)(intptr_t) fd); \
827 last_tick = transport_map_modified_tick; \
832 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
835 timeout = opt.read_timeout;
839 if (info && info->imp->poller)
840 test = info->imp->poller (fd, timeout, wf, info->ctx);
842 test = sock_poll (fd, timeout, wf);
851 /* Read no more than BUFSIZE bytes of data from FD, storing them to
852 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
853 received after that many seconds. If TIMEOUT is -1, the value of
854 opt.timeout is used for TIMEOUT. */
857 fd_read (int fd, char *buf, int bufsize, double timeout)
859 struct transport_info *info;
860 LAZY_RETRIEVE_INFO (info);
861 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
863 if (info && info->imp->reader)
864 return info->imp->reader (fd, buf, bufsize, info->ctx);
866 return sock_read (fd, buf, bufsize);
869 /* Like fd_read, except it provides a "preview" of the data that will
870 be read by subsequent calls to fd_read. Specifically, it copies no
871 more than BUFSIZE bytes of the currently available data to BUF and
872 returns the number of bytes copied. Return values and timeout
873 semantics are the same as those of fd_read.
875 CAVEAT: Do not assume that the first subsequent call to fd_read
876 will retrieve the same amount of data. Reading can return more or
877 less data, depending on the TCP implementation and other
878 circumstances. However, barring an error, it can be expected that
879 all the peeked data will eventually be read by fd_read. */
882 fd_peek (int fd, char *buf, int bufsize, double timeout)
884 struct transport_info *info;
885 LAZY_RETRIEVE_INFO (info);
886 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
888 if (info && info->imp->peeker)
889 return info->imp->peeker (fd, buf, bufsize, info->ctx);
891 return sock_peek (fd, buf, bufsize);
894 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
895 the operation aborts if no data is received after that many
896 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
900 fd_write (int fd, char *buf, int bufsize, double timeout)
903 struct transport_info *info;
904 LAZY_RETRIEVE_INFO (info);
906 /* `write' may write less than LEN bytes, thus the loop keeps trying
907 it until all was written, or an error occurred. */
911 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
913 if (info && info->imp->writer)
914 res = info->imp->writer (fd, buf, bufsize, info->ctx);
916 res = sock_write (fd, buf, bufsize);
925 /* Report the most recent error(s) on FD. This should only be called
926 after fd_* functions, such as fd_read and fd_write, and only if
927 they return a negative result. For errors coming from other calls
928 such as setsockopt or fopen, strerror should continue to be
931 If the transport doesn't support error messages or doesn't supply
932 one, strerror(errno) is returned. The returned error message
933 should not be used after fd_close has been called. */
938 /* Don't bother with LAZY_RETRIEVE_INFO, as this will only be called
939 in case of error, never in a tight loop. */
940 struct transport_info *info = NULL;
942 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
944 if (info && info->imp->errstr)
946 const char *err = info->imp->errstr (fd, info->ctx);
949 /* else, fall through and print the system error. */
951 return strerror (errno);
954 /* Close the file descriptor FD. */
959 struct transport_info *info;
963 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
964 per socket, so that particular optimization wouldn't work. */
967 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
969 if (info && info->imp->closer)
970 info->imp->closer (fd, info->ctx);
976 hash_table_remove (transport_map, (void *)(intptr_t) fd);
978 ++transport_map_modified_tick;