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 (0 != strcmp (print, txt_addr))
276 char *str = NULL, *name;
278 if (opt.enable_iri && (name = idn_decode ((char *) print)) != NULL)
280 int len = strlen (print) + strlen (name) + 4;
282 snprintf (str, len, "%s (%s)", name, print);
287 logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
288 str ? str : escnonprint_uri (print), txt_addr, port);
294 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
297 /* Store the sockaddr info to SA. */
298 sockaddr_set_data (sa, ip, port);
300 /* Create the socket of the family appropriate for the address. */
301 sock = socket (sa->sa_family, SOCK_STREAM, 0);
305 #if defined(ENABLE_IPV6) && defined(IPV6_V6ONLY)
308 /* In case of error, we will go on anyway... */
309 int err = setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof (on));
312 DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
316 /* For very small rate limits, set the buffer size (and hence,
317 hopefully, the kernel's TCP window size) to the per-second limit.
318 That way we should never have to sleep for more than 1s between
320 if (opt.limit_rate && opt.limit_rate < 8192)
322 int bufsize = opt.limit_rate;
324 bufsize = 512; /* avoid pathologically small values */
326 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
327 (void *)&bufsize, (socklen_t)sizeof (bufsize));
329 /* When we add limit_rate support for writing, which is useful
330 for POST, we should also set SO_SNDBUF here. */
333 if (opt.bind_address)
335 /* Bind the client side of the socket to the requested
337 struct sockaddr_storage bind_ss;
338 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
339 if (resolve_bind_address (bind_sa))
341 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
346 /* Connect the socket to the remote endpoint. */
347 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
348 opt.connect_timeout) < 0)
354 logprintf (LOG_VERBOSE, _("connected.\n"));
355 DEBUGP (("Created socket %d.\n", sock));
360 /* Protect errno from possible modifications by close and
362 int save_errno = errno;
366 logprintf (LOG_VERBOSE, _("failed: %s.\n"), strerror (errno));
372 /* Connect via TCP to a remote host on the specified port.
374 HOST is resolved as an Internet host name. If HOST resolves to
375 more than one IP address, they are tried in the order returned by
376 DNS until connecting to one of them succeeds. */
379 connect_to_host (const char *host, int port)
384 struct address_list *al = lookup_host (host, 0);
389 logprintf (LOG_NOTQUIET,
390 _("%s: unable to resolve host address %s\n"),
391 exec_name, quote (host));
395 address_list_get_bounds (al, &start, &end);
396 for (i = start; i < end; i++)
398 const ip_address *ip = address_list_address_at (al, i);
399 sock = connect_to_ip (ip, port, host);
403 address_list_set_connected (al);
404 address_list_release (al);
408 /* The attempt to connect has failed. Continue with the loop
409 and try next address. */
411 address_list_set_faulty (al, i);
414 /* Failed to connect to any of the addresses in AL. */
416 if (address_list_connected_p (al))
418 /* We connected to AL before, but cannot do so now. That might
419 indicate that our DNS cache entry for HOST has expired. */
420 address_list_release (al);
421 al = lookup_host (host, LH_REFRESH);
424 address_list_release (al);
429 /* Create a socket, bind it to local interface BIND_ADDRESS on port
430 *PORT, set up a listen backlog, and return the resulting socket, or
433 BIND_ADDRESS is the address of the interface to bind to. If it is
434 NULL, the socket is bound to the default address. PORT should
435 point to the port number that will be used for the binding. If
436 that number is 0, the system will choose a suitable port, and the
437 chosen value will be written to *PORT.
439 Calling accept() on such a socket waits for and accepts incoming
443 bind_local (const ip_address *bind_address, int *port)
446 struct sockaddr_storage ss;
447 struct sockaddr *sa = (struct sockaddr *)&ss;
449 /* For setting options with setsockopt. */
451 void *setopt_ptr = (void *)&setopt_val;
452 socklen_t setopt_size = sizeof (setopt_val);
454 sock = socket (bind_address->family, SOCK_STREAM, 0);
459 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
463 sockaddr_set_data (sa, bind_address, *port);
464 if (bind (sock, sa, sockaddr_size (sa)) < 0)
469 DEBUGP (("Local socket fd %d bound.\n", sock));
471 /* If *PORT is 0, find out which port we've bound to. */
474 socklen_t addrlen = sockaddr_size (sa);
475 if (getsockname (sock, sa, &addrlen) < 0)
477 /* If we can't find out the socket's local address ("name"),
478 something is seriously wrong with the socket, and it's
479 unusable for us anyway because we must know the chosen
484 sockaddr_get_data (sa, NULL, port);
485 DEBUGP (("binding to address %s using port %i.\n",
486 print_address (bind_address), *port));
488 if (listen (sock, 1) < 0)
496 /* Like a call to accept(), but with the added check for timeout.
498 In other words, accept a client connection on LOCAL_SOCK, and
499 return the new socket used for communication with the client.
500 LOCAL_SOCK should have been bound, e.g. using bind_local().
502 The caller is blocked until a connection is established. If no
503 connection is established for opt.connect_timeout seconds, the
504 function exits with an error status. */
507 accept_connection (int local_sock)
511 /* We don't need the values provided by accept, but accept
512 apparently requires them to be present. */
513 struct sockaddr_storage ss;
514 struct sockaddr *sa = (struct sockaddr *)&ss;
515 socklen_t addrlen = sizeof (ss);
517 if (opt.connect_timeout)
519 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
525 sock = accept (local_sock, sa, &addrlen);
526 DEBUGP (("Accepted client at socket %d.\n", sock));
530 /* Get the IP address associated with the connection on FD and store
531 it to IP. Return true on success, false otherwise.
533 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
534 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
535 returns the address of the remote (peer's) side of the socket. */
538 socket_ip_address (int sock, ip_address *ip, int endpoint)
540 struct sockaddr_storage storage;
541 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
542 socklen_t addrlen = sizeof (storage);
545 if (endpoint == ENDPOINT_LOCAL)
546 ret = getsockname (sock, sockaddr, &addrlen);
547 else if (endpoint == ENDPOINT_PEER)
548 ret = getpeername (sock, sockaddr, &addrlen);
554 ip->family = sockaddr->sa_family;
555 switch (sockaddr->sa_family)
560 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
561 ip->data.d6 = sa6->sin6_addr;
562 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
563 ip->ipv6_scope = sa6->sin6_scope_id;
565 DEBUGP (("conaddr is: %s\n", print_address (ip)));
571 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
572 ip->data.d4 = sa->sin_addr;
573 DEBUGP (("conaddr is: %s\n", print_address (ip)));
581 /* Return true if the error from the connect code can be considered
582 retryable. Wget normally retries after errors, but the exception
583 are the "unsupported protocol" type errors (possible on IPv4/IPv6
584 dual family systems) and "connection refused". */
587 retryable_socket_connect_error (int err)
589 /* Have to guard against some of these values not being defined.
590 Cannot use a switch statement because some of the values might be
594 || err == EAFNOSUPPORT
597 || err == EPFNOSUPPORT
599 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
600 || err == ESOCKTNOSUPPORT
602 #ifdef EPROTONOSUPPORT
603 || err == EPROTONOSUPPORT
606 || err == ENOPROTOOPT
608 /* Apparently, older versions of Linux and BSD used EINVAL
609 instead of EAFNOSUPPORT and such. */
614 if (!opt.retry_connrefused)
615 if (err == ECONNREFUSED
617 || err == ENETUNREACH /* network is unreachable */
620 || err == EHOSTUNREACH /* host is unreachable */
628 /* Wait for a single descriptor to become available, timing out after
629 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
630 -1 for error. The argument WAIT_FOR can be a combination of
631 WAIT_FOR_READ and WAIT_FOR_WRITE.
633 This is a mere convenience wrapper around the select call, and
634 should be taken as such (for example, it doesn't implement Wget's
635 0-timeout-means-no-timeout semantics.) */
638 select_fd (int fd, double maxtime, int wait_for)
641 fd_set *rd = NULL, *wr = NULL;
642 struct timeval tmout;
647 if (wait_for & WAIT_FOR_READ)
649 if (wait_for & WAIT_FOR_WRITE)
652 tmout.tv_sec = (long) maxtime;
653 tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
656 result = select (fd + 1, rd, wr, NULL, &tmout);
657 while (result < 0 && errno == EINTR);
662 /* Return true iff the connection to the remote site established
663 through SOCK is still open.
665 Specifically, this function returns true if SOCK is not ready for
666 reading. This is because, when the connection closes, the socket
667 is ready for reading because EOF is about to be delivered. A side
668 effect of this method is that sockets that have pending data are
669 considered non-open. This is actually a good thing for callers of
670 this function, where such pending data can only be unwanted
671 leftover from a previous request. */
674 test_socket_open (int sock)
679 /* Check if we still have a valid (non-EOF) connection. From Andrew
680 * Maholski's code in the Unix Socket FAQ. */
682 FD_ZERO (&check_set);
683 FD_SET (sock, &check_set);
685 /* Wait one microsecond */
689 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
690 /* We got a timeout, it means we're still connected. */
693 /* Read now would not wait, it means we have either pending data
698 /* Basic socket operations, mostly EINTR wrappers. */
700 #if defined(WINDOWS) || defined(MSDOS)
701 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
702 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
703 # define close(fd) closesocket (fd)
707 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
708 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
712 sock_read (int fd, char *buf, int bufsize)
716 res = read (fd, buf, bufsize);
717 while (res == -1 && errno == EINTR);
722 sock_write (int fd, char *buf, int bufsize)
726 res = write (fd, buf, bufsize);
727 while (res == -1 && errno == EINTR);
732 sock_poll (int fd, double timeout, int wait_for)
734 return select_fd (fd, timeout, wait_for);
738 sock_peek (int fd, char *buf, int bufsize)
742 res = recv (fd, buf, bufsize, MSG_PEEK);
743 while (res == -1 && errno == EINTR);
751 DEBUGP (("Closed fd %d\n", fd));
757 /* Reading and writing from the network. We build around the socket
758 (file descriptor) API, but support "extended" operations for things
759 that are not mere file descriptors under the hood, such as SSL
762 That way the user code can call fd_read(fd, ...) and we'll run read
763 or SSL_read or whatever is necessary. */
765 static struct hash_table *transport_map;
766 static unsigned int transport_map_modified_tick;
768 struct transport_info {
769 struct transport_implementation *imp;
773 /* Register the transport layer operations that will be used when
774 reading, writing, and polling FD.
776 This should be used for transport layers like SSL that piggyback on
777 sockets. FD should otherwise be a real socket, on which you can
778 call getpeername, etc. */
781 fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
783 struct transport_info *info;
785 /* The file descriptor must be non-negative to be registered.
786 Negative values are ignored by fd_close(), and -1 cannot be used as
790 info = xnew (struct transport_info);
794 transport_map = hash_table_new (0, NULL, NULL);
795 hash_table_put (transport_map, (void *)(intptr_t) fd, info);
796 ++transport_map_modified_tick;
799 /* Return context of the transport registered with
800 fd_register_transport. This assumes fd_register_transport was
801 previously called on FD. */
804 fd_transport_context (int fd)
806 struct transport_info *info = hash_table_get (transport_map, (void *)(intptr_t) fd);
810 /* When fd_read/fd_write are called multiple times in a loop, they should
811 remember the INFO pointer instead of fetching it every time. It is
812 not enough to compare FD to LAST_FD because FD might have been
813 closed and reopened. modified_tick ensures that changes to
814 transport_map will not be unnoticed.
816 This is a macro because we want the static storage variables to be
819 #define LAZY_RETRIEVE_INFO(info) do { \
820 static struct transport_info *last_info; \
821 static int last_fd = -1; \
822 static unsigned int last_tick; \
823 if (!transport_map) \
825 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
829 info = hash_table_get (transport_map, (void *)(intptr_t) fd); \
832 last_tick = transport_map_modified_tick; \
837 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
840 timeout = opt.read_timeout;
844 if (info && info->imp->poller)
845 test = info->imp->poller (fd, timeout, wf, info->ctx);
847 test = sock_poll (fd, timeout, wf);
856 /* Read no more than BUFSIZE bytes of data from FD, storing them to
857 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
858 received after that many seconds. If TIMEOUT is -1, the value of
859 opt.timeout is used for TIMEOUT. */
862 fd_read (int fd, char *buf, int bufsize, double timeout)
864 struct transport_info *info;
865 LAZY_RETRIEVE_INFO (info);
866 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
868 if (info && info->imp->reader)
869 return info->imp->reader (fd, buf, bufsize, info->ctx);
871 return sock_read (fd, buf, bufsize);
874 /* Like fd_read, except it provides a "preview" of the data that will
875 be read by subsequent calls to fd_read. Specifically, it copies no
876 more than BUFSIZE bytes of the currently available data to BUF and
877 returns the number of bytes copied. Return values and timeout
878 semantics are the same as those of fd_read.
880 CAVEAT: Do not assume that the first subsequent call to fd_read
881 will retrieve the same amount of data. Reading can return more or
882 less data, depending on the TCP implementation and other
883 circumstances. However, barring an error, it can be expected that
884 all the peeked data will eventually be read by fd_read. */
887 fd_peek (int fd, char *buf, int bufsize, double timeout)
889 struct transport_info *info;
890 LAZY_RETRIEVE_INFO (info);
891 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
893 if (info && info->imp->peeker)
894 return info->imp->peeker (fd, buf, bufsize, info->ctx);
896 return sock_peek (fd, buf, bufsize);
899 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
900 the operation aborts if no data is received after that many
901 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
905 fd_write (int fd, char *buf, int bufsize, double timeout)
908 struct transport_info *info;
909 LAZY_RETRIEVE_INFO (info);
911 /* `write' may write less than LEN bytes, thus the loop keeps trying
912 it until all was written, or an error occurred. */
916 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
918 if (info && info->imp->writer)
919 res = info->imp->writer (fd, buf, bufsize, info->ctx);
921 res = sock_write (fd, buf, bufsize);
930 /* Report the most recent error(s) on FD. This should only be called
931 after fd_* functions, such as fd_read and fd_write, and only if
932 they return a negative result. For errors coming from other calls
933 such as setsockopt or fopen, strerror should continue to be
936 If the transport doesn't support error messages or doesn't supply
937 one, strerror(errno) is returned. The returned error message
938 should not be used after fd_close has been called. */
943 /* Don't bother with LAZY_RETRIEVE_INFO, as this will only be called
944 in case of error, never in a tight loop. */
945 struct transport_info *info = NULL;
947 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
949 if (info && info->imp->errstr)
951 const char *err = info->imp->errstr (fd, info->ctx);
954 /* else, fall through and print the system error. */
956 return strerror (errno);
959 /* Close the file descriptor FD. */
964 struct transport_info *info;
968 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
969 per socket, so that particular optimization wouldn't work. */
972 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
974 if (info && info->imp->closer)
975 info->imp->closer (fd, info->ctx);
981 hash_table_remove (transport_map, (void *)(intptr_t) fd);
983 ++transport_map_modified_tick;