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>
63 /* Define sockaddr_storage where unavailable (presumably on IPv4-only
67 # ifndef HAVE_STRUCT_SOCKADDR_STORAGE
68 # define sockaddr_storage sockaddr_in
70 #endif /* ENABLE_IPV6 */
72 /* Fill SA as per the data in IP and PORT. SA shoult point to struct
73 sockaddr_storage if ENABLE_IPV6 is defined, to struct sockaddr_in
77 sockaddr_set_data (struct sockaddr *sa, const ip_address *ip, int port)
83 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
85 sin->sin_family = AF_INET;
86 sin->sin_port = htons (port);
87 sin->sin_addr = ip->data.d4;
93 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
95 sin6->sin6_family = AF_INET6;
96 sin6->sin6_port = htons (port);
97 sin6->sin6_addr = ip->data.d6;
98 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
99 sin6->sin6_scope_id = ip->ipv6_scope;
103 #endif /* ENABLE_IPV6 */
109 /* Get the data of SA, specifically the IP address and the port. If
110 you're not interested in one or the other information, pass NULL as
114 sockaddr_get_data (const struct sockaddr *sa, ip_address *ip, int *port)
116 switch (sa->sa_family)
120 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
123 ip->family = AF_INET;
124 ip->data.d4 = sin->sin_addr;
127 *port = ntohs (sin->sin_port);
133 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
136 ip->family = AF_INET6;
137 ip->data.d6 = sin6->sin6_addr;
138 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
139 ip->ipv6_scope = sin6->sin6_scope_id;
143 *port = ntohs (sin6->sin6_port);
152 /* Return the size of the sockaddr structure depending on its
156 sockaddr_size (const struct sockaddr *sa)
158 switch (sa->sa_family)
161 return sizeof (struct sockaddr_in);
164 return sizeof (struct sockaddr_in6);
171 /* Resolve the bind address specified via --bind-address and store it
172 to SA. The resolved value is stored in a static variable and
173 reused after the first invocation of this function.
175 Returns true on success, false on failure. */
178 resolve_bind_address (struct sockaddr *sa)
180 struct address_list *al;
182 /* Make sure this is called only once. opt.bind_address doesn't
183 change during a Wget run. */
184 static bool called, should_bind;
185 static ip_address ip;
189 sockaddr_set_data (sa, &ip, 0);
194 al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
197 /* #### We should be able to print the error message here. */
198 logprintf (LOG_NOTQUIET,
199 _("%s: unable to resolve bind address %s; disabling bind.\n"),
200 exec_name, quote (opt.bind_address));
205 /* Pick the first address in the list and use it as bind address.
206 Perhaps we should try multiple addresses in succession, but I
207 don't think that's necessary in practice. */
208 ip = *address_list_address_at (al, 0);
209 address_list_release (al);
211 sockaddr_set_data (sa, &ip, 0);
218 const struct sockaddr *addr;
224 connect_with_timeout_callback (void *arg)
226 struct cwt_context *ctx = (struct cwt_context *)arg;
227 ctx->result = connect (ctx->fd, ctx->addr, ctx->addrlen);
230 /* Like connect, but specifies a timeout. If connecting takes longer
231 than TIMEOUT seconds, -1 is returned and errno is set to
235 connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
238 struct cwt_context ctx;
241 ctx.addrlen = addrlen;
243 if (run_with_timeout (timeout, connect_with_timeout_callback, &ctx))
248 if (ctx.result == -1 && errno == EINTR)
253 /* Connect via TCP to the specified address and port.
255 If PRINT is non-NULL, it is the host name to print that we're
259 connect_to_ip (const ip_address *ip, int port, const char *print)
261 struct sockaddr_storage ss;
262 struct sockaddr *sa = (struct sockaddr *)&ss;
265 /* If PRINT is non-NULL, print the "Connecting to..." line, with
266 PRINT being the host name we're connecting to. */
269 const char *txt_addr = print_address (ip);
270 if (0 != strcmp (print, txt_addr))
272 char *str = NULL, *name;
274 if (opt.enable_iri && (name = idn_decode ((char *) print)) != NULL)
276 int len = strlen (print) + strlen (name) + 4;
278 snprintf (str, len, "%s (%s)", name, print);
283 logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
284 str ? str : escnonprint_uri (print), txt_addr, port);
290 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
293 /* Store the sockaddr info to SA. */
294 sockaddr_set_data (sa, ip, port);
296 /* Create the socket of the family appropriate for the address. */
297 sock = socket (sa->sa_family, SOCK_STREAM, 0);
301 #if defined(ENABLE_IPV6) && defined(IPV6_V6ONLY)
304 /* In case of error, we will go on anyway... */
305 int err = setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof (on));
308 DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
312 /* For very small rate limits, set the buffer size (and hence,
313 hopefully, the kernel's TCP window size) to the per-second limit.
314 That way we should never have to sleep for more than 1s between
316 if (opt.limit_rate && opt.limit_rate < 8192)
318 int bufsize = opt.limit_rate;
320 bufsize = 512; /* avoid pathologically small values */
322 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
323 (void *)&bufsize, (socklen_t)sizeof (bufsize));
325 /* When we add limit_rate support for writing, which is useful
326 for POST, we should also set SO_SNDBUF here. */
329 if (opt.bind_address)
331 /* Bind the client side of the socket to the requested
333 struct sockaddr_storage bind_ss;
334 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
335 if (resolve_bind_address (bind_sa))
337 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
342 /* Connect the socket to the remote endpoint. */
343 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
344 opt.connect_timeout) < 0)
350 logprintf (LOG_VERBOSE, _("connected.\n"));
351 DEBUGP (("Created socket %d.\n", sock));
356 /* Protect errno from possible modifications by close and
358 int save_errno = errno;
362 logprintf (LOG_VERBOSE, _("failed: %s.\n"), strerror (errno));
368 /* Connect via TCP to a remote host on the specified port.
370 HOST is resolved as an Internet host name. If HOST resolves to
371 more than one IP address, they are tried in the order returned by
372 DNS until connecting to one of them succeeds. */
375 connect_to_host (const char *host, int port)
380 struct address_list *al = lookup_host (host, 0);
385 logprintf (LOG_NOTQUIET,
386 _("%s: unable to resolve host address %s\n"),
387 exec_name, quote (host));
391 address_list_get_bounds (al, &start, &end);
392 for (i = start; i < end; i++)
394 const ip_address *ip = address_list_address_at (al, i);
395 sock = connect_to_ip (ip, port, host);
399 address_list_set_connected (al);
400 address_list_release (al);
404 /* The attempt to connect has failed. Continue with the loop
405 and try next address. */
407 address_list_set_faulty (al, i);
410 /* Failed to connect to any of the addresses in AL. */
412 if (address_list_connected_p (al))
414 /* We connected to AL before, but cannot do so now. That might
415 indicate that our DNS cache entry for HOST has expired. */
416 address_list_release (al);
417 al = lookup_host (host, LH_REFRESH);
420 address_list_release (al);
425 /* Create a socket, bind it to local interface BIND_ADDRESS on port
426 *PORT, set up a listen backlog, and return the resulting socket, or
429 BIND_ADDRESS is the address of the interface to bind to. If it is
430 NULL, the socket is bound to the default address. PORT should
431 point to the port number that will be used for the binding. If
432 that number is 0, the system will choose a suitable port, and the
433 chosen value will be written to *PORT.
435 Calling accept() on such a socket waits for and accepts incoming
439 bind_local (const ip_address *bind_address, int *port)
442 struct sockaddr_storage ss;
443 struct sockaddr *sa = (struct sockaddr *)&ss;
445 /* For setting options with setsockopt. */
447 void *setopt_ptr = (void *)&setopt_val;
448 socklen_t setopt_size = sizeof (setopt_val);
450 sock = socket (bind_address->family, SOCK_STREAM, 0);
455 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
459 sockaddr_set_data (sa, bind_address, *port);
460 if (bind (sock, sa, sockaddr_size (sa)) < 0)
465 DEBUGP (("Local socket fd %d bound.\n", sock));
467 /* If *PORT is 0, find out which port we've bound to. */
470 socklen_t addrlen = sockaddr_size (sa);
471 if (getsockname (sock, sa, &addrlen) < 0)
473 /* If we can't find out the socket's local address ("name"),
474 something is seriously wrong with the socket, and it's
475 unusable for us anyway because we must know the chosen
480 sockaddr_get_data (sa, NULL, port);
481 DEBUGP (("binding to address %s using port %i.\n",
482 print_address (bind_address), *port));
484 if (listen (sock, 1) < 0)
492 /* Like a call to accept(), but with the added check for timeout.
494 In other words, accept a client connection on LOCAL_SOCK, and
495 return the new socket used for communication with the client.
496 LOCAL_SOCK should have been bound, e.g. using bind_local().
498 The caller is blocked until a connection is established. If no
499 connection is established for opt.connect_timeout seconds, the
500 function exits with an error status. */
503 accept_connection (int local_sock)
507 /* We don't need the values provided by accept, but accept
508 apparently requires them to be present. */
509 struct sockaddr_storage ss;
510 struct sockaddr *sa = (struct sockaddr *)&ss;
511 socklen_t addrlen = sizeof (ss);
513 if (opt.connect_timeout)
515 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
521 sock = accept (local_sock, sa, &addrlen);
522 DEBUGP (("Accepted client at socket %d.\n", sock));
526 /* Get the IP address associated with the connection on FD and store
527 it to IP. Return true on success, false otherwise.
529 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
530 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
531 returns the address of the remote (peer's) side of the socket. */
534 socket_ip_address (int sock, ip_address *ip, int endpoint)
536 struct sockaddr_storage storage;
537 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
538 socklen_t addrlen = sizeof (storage);
541 if (endpoint == ENDPOINT_LOCAL)
542 ret = getsockname (sock, sockaddr, &addrlen);
543 else if (endpoint == ENDPOINT_PEER)
544 ret = getpeername (sock, sockaddr, &addrlen);
550 ip->family = sockaddr->sa_family;
551 switch (sockaddr->sa_family)
556 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
557 ip->data.d6 = sa6->sin6_addr;
558 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
559 ip->ipv6_scope = sa6->sin6_scope_id;
561 DEBUGP (("conaddr is: %s\n", print_address (ip)));
567 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
568 ip->data.d4 = sa->sin_addr;
569 DEBUGP (("conaddr is: %s\n", print_address (ip)));
577 /* Return true if the error from the connect code can be considered
578 retryable. Wget normally retries after errors, but the exception
579 are the "unsupported protocol" type errors (possible on IPv4/IPv6
580 dual family systems) and "connection refused". */
583 retryable_socket_connect_error (int err)
585 /* Have to guard against some of these values not being defined.
586 Cannot use a switch statement because some of the values might be
590 || err == EAFNOSUPPORT
593 || err == EPFNOSUPPORT
595 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
596 || err == ESOCKTNOSUPPORT
598 #ifdef EPROTONOSUPPORT
599 || err == EPROTONOSUPPORT
602 || err == ENOPROTOOPT
604 /* Apparently, older versions of Linux and BSD used EINVAL
605 instead of EAFNOSUPPORT and such. */
610 if (!opt.retry_connrefused)
611 if (err == ECONNREFUSED
613 || err == ENETUNREACH /* network is unreachable */
616 || err == EHOSTUNREACH /* host is unreachable */
624 /* Wait for a single descriptor to become available, timing out after
625 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
626 -1 for error. The argument WAIT_FOR can be a combination of
627 WAIT_FOR_READ and WAIT_FOR_WRITE.
629 This is a mere convenience wrapper around the select call, and
630 should be taken as such (for example, it doesn't implement Wget's
631 0-timeout-means-no-timeout semantics.) */
634 select_fd (int fd, double maxtime, int wait_for)
637 fd_set *rd = NULL, *wr = NULL;
638 struct timeval tmout;
643 if (wait_for & WAIT_FOR_READ)
645 if (wait_for & WAIT_FOR_WRITE)
648 tmout.tv_sec = (long) maxtime;
649 tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
652 result = select (fd + 1, rd, wr, NULL, &tmout);
653 while (result < 0 && errno == EINTR);
658 /* Return true iff the connection to the remote site established
659 through SOCK is still open.
661 Specifically, this function returns true if SOCK is not ready for
662 reading. This is because, when the connection closes, the socket
663 is ready for reading because EOF is about to be delivered. A side
664 effect of this method is that sockets that have pending data are
665 considered non-open. This is actually a good thing for callers of
666 this function, where such pending data can only be unwanted
667 leftover from a previous request. */
670 test_socket_open (int sock)
675 /* Check if we still have a valid (non-EOF) connection. From Andrew
676 * Maholski's code in the Unix Socket FAQ. */
678 FD_ZERO (&check_set);
679 FD_SET (sock, &check_set);
681 /* Wait one microsecond */
685 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
686 /* We got a timeout, it means we're still connected. */
689 /* Read now would not wait, it means we have either pending data
694 /* Basic socket operations, mostly EINTR wrappers. */
696 #if defined(WINDOWS) || defined(MSDOS)
697 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
698 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
699 # define close(fd) closesocket (fd)
703 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
704 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
708 sock_read (int fd, char *buf, int bufsize)
712 res = read (fd, buf, bufsize);
713 while (res == -1 && errno == EINTR);
718 sock_write (int fd, char *buf, int bufsize)
722 res = write (fd, buf, bufsize);
723 while (res == -1 && errno == EINTR);
728 sock_poll (int fd, double timeout, int wait_for)
730 return select_fd (fd, timeout, wait_for);
734 sock_peek (int fd, char *buf, int bufsize)
738 res = recv (fd, buf, bufsize, MSG_PEEK);
739 while (res == -1 && errno == EINTR);
747 DEBUGP (("Closed fd %d\n", fd));
753 /* Reading and writing from the network. We build around the socket
754 (file descriptor) API, but support "extended" operations for things
755 that are not mere file descriptors under the hood, such as SSL
758 That way the user code can call fd_read(fd, ...) and we'll run read
759 or SSL_read or whatever is necessary. */
761 static struct hash_table *transport_map;
762 static unsigned int transport_map_modified_tick;
764 struct transport_info {
765 struct transport_implementation *imp;
769 /* Register the transport layer operations that will be used when
770 reading, writing, and polling FD.
772 This should be used for transport layers like SSL that piggyback on
773 sockets. FD should otherwise be a real socket, on which you can
774 call getpeername, etc. */
777 fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
779 struct transport_info *info;
781 /* The file descriptor must be non-negative to be registered.
782 Negative values are ignored by fd_close(), and -1 cannot be used as
786 info = xnew (struct transport_info);
790 transport_map = hash_table_new (0, NULL, NULL);
791 hash_table_put (transport_map, (void *)(intptr_t) fd, info);
792 ++transport_map_modified_tick;
795 /* Return context of the transport registered with
796 fd_register_transport. This assumes fd_register_transport was
797 previously called on FD. */
800 fd_transport_context (int fd)
802 struct transport_info *info = hash_table_get (transport_map, (void *)(intptr_t) fd);
806 /* When fd_read/fd_write are called multiple times in a loop, they should
807 remember the INFO pointer instead of fetching it every time. It is
808 not enough to compare FD to LAST_FD because FD might have been
809 closed and reopened. modified_tick ensures that changes to
810 transport_map will not be unnoticed.
812 This is a macro because we want the static storage variables to be
815 #define LAZY_RETRIEVE_INFO(info) do { \
816 static struct transport_info *last_info; \
817 static int last_fd = -1; \
818 static unsigned int last_tick; \
819 if (!transport_map) \
821 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
825 info = hash_table_get (transport_map, (void *)(intptr_t) fd); \
828 last_tick = transport_map_modified_tick; \
833 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
836 timeout = opt.read_timeout;
840 if (info && info->imp->poller)
841 test = info->imp->poller (fd, timeout, wf, info->ctx);
843 test = sock_poll (fd, timeout, wf);
852 /* Read no more than BUFSIZE bytes of data from FD, storing them to
853 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
854 received after that many seconds. If TIMEOUT is -1, the value of
855 opt.timeout is used for TIMEOUT. */
858 fd_read (int fd, char *buf, int bufsize, double timeout)
860 struct transport_info *info;
861 LAZY_RETRIEVE_INFO (info);
862 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
864 if (info && info->imp->reader)
865 return info->imp->reader (fd, buf, bufsize, info->ctx);
867 return sock_read (fd, buf, bufsize);
870 /* Like fd_read, except it provides a "preview" of the data that will
871 be read by subsequent calls to fd_read. Specifically, it copies no
872 more than BUFSIZE bytes of the currently available data to BUF and
873 returns the number of bytes copied. Return values and timeout
874 semantics are the same as those of fd_read.
876 CAVEAT: Do not assume that the first subsequent call to fd_read
877 will retrieve the same amount of data. Reading can return more or
878 less data, depending on the TCP implementation and other
879 circumstances. However, barring an error, it can be expected that
880 all the peeked data will eventually be read by fd_read. */
883 fd_peek (int fd, char *buf, int bufsize, double timeout)
885 struct transport_info *info;
886 LAZY_RETRIEVE_INFO (info);
887 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
889 if (info && info->imp->peeker)
890 return info->imp->peeker (fd, buf, bufsize, info->ctx);
892 return sock_peek (fd, buf, bufsize);
895 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
896 the operation aborts if no data is received after that many
897 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
901 fd_write (int fd, char *buf, int bufsize, double timeout)
904 struct transport_info *info;
905 LAZY_RETRIEVE_INFO (info);
907 /* `write' may write less than LEN bytes, thus the loop keeps trying
908 it until all was written, or an error occurred. */
912 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
914 if (info && info->imp->writer)
915 res = info->imp->writer (fd, buf, bufsize, info->ctx);
917 res = sock_write (fd, buf, bufsize);
926 /* Report the most recent error(s) on FD. This should only be called
927 after fd_* functions, such as fd_read and fd_write, and only if
928 they return a negative result. For errors coming from other calls
929 such as setsockopt or fopen, strerror should continue to be
932 If the transport doesn't support error messages or doesn't supply
933 one, strerror(errno) is returned. The returned error message
934 should not be used after fd_close has been called. */
939 /* Don't bother with LAZY_RETRIEVE_INFO, as this will only be called
940 in case of error, never in a tight loop. */
941 struct transport_info *info = NULL;
943 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
945 if (info && info->imp->errstr)
947 const char *err = info->imp->errstr (fd, info->ctx);
950 /* else, fall through and print the system error. */
952 return strerror (errno);
955 /* Close the file descriptor FD. */
960 struct transport_info *info;
964 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
965 per socket, so that particular optimization wouldn't work. */
968 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
970 if (info && info->imp->closer)
971 info->imp->closer (fd, info->ctx);
977 hash_table_remove (transport_map, (void *)(intptr_t) fd);
979 ++transport_map_modified_tick;