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 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. */
40 #include <sys/socket.h>
41 #include <sys/select.h>
46 # else /* def __VMS */
48 # endif /* def __VMS [else] */
49 # include <netinet/in.h>
51 # include <arpa/inet.h>
53 #endif /* not WINDOWS */
57 #ifdef HAVE_SYS_TIME_H
58 # include <sys/time.h>
65 /* Apparently needed for Interix: */
70 /* Define sockaddr_storage where unavailable (presumably on IPv4-only
74 # ifndef HAVE_STRUCT_SOCKADDR_STORAGE
75 # define sockaddr_storage sockaddr_in
77 #endif /* ENABLE_IPV6 */
79 /* Fill SA as per the data in IP and PORT. SA shoult point to struct
80 sockaddr_storage if ENABLE_IPV6 is defined, to struct sockaddr_in
84 sockaddr_set_data (struct sockaddr *sa, const ip_address *ip, int port)
90 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
92 sin->sin_family = AF_INET;
93 sin->sin_port = htons (port);
94 sin->sin_addr = ip->data.d4;
100 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
102 sin6->sin6_family = AF_INET6;
103 sin6->sin6_port = htons (port);
104 sin6->sin6_addr = ip->data.d6;
105 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
106 sin6->sin6_scope_id = ip->ipv6_scope;
110 #endif /* ENABLE_IPV6 */
116 /* Get the data of SA, specifically the IP address and the port. If
117 you're not interested in one or the other information, pass NULL as
121 sockaddr_get_data (const struct sockaddr *sa, ip_address *ip, int *port)
123 switch (sa->sa_family)
127 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
130 ip->family = AF_INET;
131 ip->data.d4 = sin->sin_addr;
134 *port = ntohs (sin->sin_port);
140 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
143 ip->family = AF_INET6;
144 ip->data.d6 = sin6->sin6_addr;
145 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
146 ip->ipv6_scope = sin6->sin6_scope_id;
150 *port = ntohs (sin6->sin6_port);
159 /* Return the size of the sockaddr structure depending on its
163 sockaddr_size (const struct sockaddr *sa)
165 switch (sa->sa_family)
168 return sizeof (struct sockaddr_in);
171 return sizeof (struct sockaddr_in6);
178 /* Resolve the bind address specified via --bind-address and store it
179 to SA. The resolved value is stored in a static variable and
180 reused after the first invocation of this function.
182 Returns true on success, false on failure. */
185 resolve_bind_address (struct sockaddr *sa)
187 struct address_list *al;
189 /* Make sure this is called only once. opt.bind_address doesn't
190 change during a Wget run. */
191 static bool called, should_bind;
192 static ip_address ip;
196 sockaddr_set_data (sa, &ip, 0);
201 al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
204 /* #### We should be able to print the error message here. */
205 logprintf (LOG_NOTQUIET,
206 _("%s: unable to resolve bind address %s; disabling bind.\n"),
207 exec_name, quote (opt.bind_address));
212 /* Pick the first address in the list and use it as bind address.
213 Perhaps we should try multiple addresses in succession, but I
214 don't think that's necessary in practice. */
215 ip = *address_list_address_at (al, 0);
216 address_list_release (al);
218 sockaddr_set_data (sa, &ip, 0);
225 const struct sockaddr *addr;
231 connect_with_timeout_callback (void *arg)
233 struct cwt_context *ctx = (struct cwt_context *)arg;
234 ctx->result = connect (ctx->fd, ctx->addr, ctx->addrlen);
237 /* Like connect, but specifies a timeout. If connecting takes longer
238 than TIMEOUT seconds, -1 is returned and errno is set to
242 connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
245 struct cwt_context ctx;
248 ctx.addrlen = addrlen;
250 if (run_with_timeout (timeout, connect_with_timeout_callback, &ctx))
255 if (ctx.result == -1 && errno == EINTR)
260 /* Connect via TCP to the specified address and port.
262 If PRINT is non-NULL, it is the host name to print that we're
266 connect_to_ip (const ip_address *ip, int port, const char *print)
268 struct sockaddr_storage ss;
269 struct sockaddr *sa = (struct sockaddr *)&ss;
272 /* If PRINT is non-NULL, print the "Connecting to..." line, with
273 PRINT being the host name we're connecting to. */
276 const char *txt_addr = print_address (ip);
277 if (0 != strcmp (print, txt_addr))
279 char *str = NULL, *name;
281 if (opt.enable_iri && (name = idn_decode ((char *) print)) != NULL)
283 int len = strlen (print) + strlen (name) + 4;
285 snprintf (str, len, "%s (%s)", name, print);
290 logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
291 str ? str : escnonprint_uri (print), txt_addr, port);
297 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
300 /* Store the sockaddr info to SA. */
301 sockaddr_set_data (sa, ip, port);
303 /* Create the socket of the family appropriate for the address. */
304 sock = socket (sa->sa_family, SOCK_STREAM, 0);
308 #if defined(ENABLE_IPV6) && defined(IPV6_V6ONLY)
311 /* In case of error, we will go on anyway... */
312 int err = setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof (on));
315 DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
319 /* For very small rate limits, set the buffer size (and hence,
320 hopefully, the kernel's TCP window size) to the per-second limit.
321 That way we should never have to sleep for more than 1s between
323 if (opt.limit_rate && opt.limit_rate < 8192)
325 int bufsize = opt.limit_rate;
327 bufsize = 512; /* avoid pathologically small values */
329 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
330 (void *)&bufsize, (socklen_t)sizeof (bufsize));
332 /* When we add limit_rate support for writing, which is useful
333 for POST, we should also set SO_SNDBUF here. */
336 if (opt.bind_address)
338 /* Bind the client side of the socket to the requested
340 struct sockaddr_storage bind_ss;
341 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
342 if (resolve_bind_address (bind_sa))
344 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
349 /* Connect the socket to the remote endpoint. */
350 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
351 opt.connect_timeout) < 0)
357 logprintf (LOG_VERBOSE, _("connected.\n"));
358 DEBUGP (("Created socket %d.\n", sock));
363 /* Protect errno from possible modifications by close and
365 int save_errno = errno;
369 logprintf (LOG_VERBOSE, _("failed: %s.\n"), strerror (errno));
375 /* Connect via TCP to a remote host on the specified port.
377 HOST is resolved as an Internet host name. If HOST resolves to
378 more than one IP address, they are tried in the order returned by
379 DNS until connecting to one of them succeeds. */
382 connect_to_host (const char *host, int port)
387 struct address_list *al = lookup_host (host, 0);
392 logprintf (LOG_NOTQUIET,
393 _("%s: unable to resolve host address %s\n"),
394 exec_name, quote (host));
398 address_list_get_bounds (al, &start, &end);
399 for (i = start; i < end; i++)
401 const ip_address *ip = address_list_address_at (al, i);
402 sock = connect_to_ip (ip, port, host);
406 address_list_set_connected (al);
407 address_list_release (al);
411 /* The attempt to connect has failed. Continue with the loop
412 and try next address. */
414 address_list_set_faulty (al, i);
417 /* Failed to connect to any of the addresses in AL. */
419 if (address_list_connected_p (al))
421 /* We connected to AL before, but cannot do so now. That might
422 indicate that our DNS cache entry for HOST has expired. */
423 address_list_release (al);
424 al = lookup_host (host, LH_REFRESH);
427 address_list_release (al);
432 /* Create a socket, bind it to local interface BIND_ADDRESS on port
433 *PORT, set up a listen backlog, and return the resulting socket, or
436 BIND_ADDRESS is the address of the interface to bind to. If it is
437 NULL, the socket is bound to the default address. PORT should
438 point to the port number that will be used for the binding. If
439 that number is 0, the system will choose a suitable port, and the
440 chosen value will be written to *PORT.
442 Calling accept() on such a socket waits for and accepts incoming
446 bind_local (const ip_address *bind_address, int *port)
449 struct sockaddr_storage ss;
450 struct sockaddr *sa = (struct sockaddr *)&ss;
452 /* For setting options with setsockopt. */
454 void *setopt_ptr = (void *)&setopt_val;
455 socklen_t setopt_size = sizeof (setopt_val);
457 sock = socket (bind_address->family, SOCK_STREAM, 0);
462 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
466 sockaddr_set_data (sa, bind_address, *port);
467 if (bind (sock, sa, sockaddr_size (sa)) < 0)
472 DEBUGP (("Local socket fd %d bound.\n", sock));
474 /* If *PORT is 0, find out which port we've bound to. */
477 socklen_t addrlen = sockaddr_size (sa);
478 if (getsockname (sock, sa, &addrlen) < 0)
480 /* If we can't find out the socket's local address ("name"),
481 something is seriously wrong with the socket, and it's
482 unusable for us anyway because we must know the chosen
487 sockaddr_get_data (sa, NULL, port);
488 DEBUGP (("binding to address %s using port %i.\n",
489 print_address (bind_address), *port));
491 if (listen (sock, 1) < 0)
499 /* Like a call to accept(), but with the added check for timeout.
501 In other words, accept a client connection on LOCAL_SOCK, and
502 return the new socket used for communication with the client.
503 LOCAL_SOCK should have been bound, e.g. using bind_local().
505 The caller is blocked until a connection is established. If no
506 connection is established for opt.connect_timeout seconds, the
507 function exits with an error status. */
510 accept_connection (int local_sock)
514 /* We don't need the values provided by accept, but accept
515 apparently requires them to be present. */
516 struct sockaddr_storage ss;
517 struct sockaddr *sa = (struct sockaddr *)&ss;
518 socklen_t addrlen = sizeof (ss);
520 if (opt.connect_timeout)
522 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
528 sock = accept (local_sock, sa, &addrlen);
529 DEBUGP (("Accepted client at socket %d.\n", sock));
533 /* Get the IP address associated with the connection on FD and store
534 it to IP. Return true on success, false otherwise.
536 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
537 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
538 returns the address of the remote (peer's) side of the socket. */
541 socket_ip_address (int sock, ip_address *ip, int endpoint)
543 struct sockaddr_storage storage;
544 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
545 socklen_t addrlen = sizeof (storage);
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);
659 result = select (fd + 1, rd, wr, NULL, &tmout);
660 while (result < 0 && errno == EINTR);
665 /* Return true iff the connection to the remote site established
666 through SOCK is still open.
668 Specifically, this function returns true if SOCK is not ready for
669 reading. This is because, when the connection closes, the socket
670 is ready for reading because EOF is about to be delivered. A side
671 effect of this method is that sockets that have pending data are
672 considered non-open. This is actually a good thing for callers of
673 this function, where such pending data can only be unwanted
674 leftover from a previous request. */
677 test_socket_open (int sock)
682 /* Check if we still have a valid (non-EOF) connection. From Andrew
683 * Maholski's code in the Unix Socket FAQ. */
685 FD_ZERO (&check_set);
686 FD_SET (sock, &check_set);
688 /* Wait one microsecond */
692 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
693 /* We got a timeout, it means we're still connected. */
696 /* Read now would not wait, it means we have either pending data
701 /* Basic socket operations, mostly EINTR wrappers. */
704 sock_read (int fd, char *buf, int bufsize)
708 res = read (fd, buf, bufsize);
709 while (res == -1 && errno == EINTR);
714 sock_write (int fd, char *buf, int bufsize)
718 res = write (fd, buf, bufsize);
719 while (res == -1 && errno == EINTR);
724 sock_poll (int fd, double timeout, int wait_for)
726 return select_fd (fd, timeout, wait_for);
730 sock_peek (int fd, char *buf, int bufsize)
734 res = recv (fd, buf, bufsize, MSG_PEEK);
735 while (res == -1 && errno == EINTR);
743 DEBUGP (("Closed fd %d\n", fd));
749 /* Reading and writing from the network. We build around the socket
750 (file descriptor) API, but support "extended" operations for things
751 that are not mere file descriptors under the hood, such as SSL
754 That way the user code can call fd_read(fd, ...) and we'll run read
755 or SSL_read or whatever is necessary. */
757 static struct hash_table *transport_map;
758 static unsigned int transport_map_modified_tick;
760 struct transport_info {
761 struct transport_implementation *imp;
765 /* Register the transport layer operations that will be used when
766 reading, writing, and polling FD.
768 This should be used for transport layers like SSL that piggyback on
769 sockets. FD should otherwise be a real socket, on which you can
770 call getpeername, etc. */
773 fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
775 struct transport_info *info;
777 /* The file descriptor must be non-negative to be registered.
778 Negative values are ignored by fd_close(), and -1 cannot be used as
782 info = xnew (struct transport_info);
786 transport_map = hash_table_new (0, NULL, NULL);
787 hash_table_put (transport_map, (void *)(intptr_t) fd, info);
788 ++transport_map_modified_tick;
791 /* Return context of the transport registered with
792 fd_register_transport. This assumes fd_register_transport was
793 previously called on FD. */
796 fd_transport_context (int fd)
798 struct transport_info *info = hash_table_get (transport_map, (void *)(intptr_t) fd);
802 /* When fd_read/fd_write are called multiple times in a loop, they should
803 remember the INFO pointer instead of fetching it every time. It is
804 not enough to compare FD to LAST_FD because FD might have been
805 closed and reopened. modified_tick ensures that changes to
806 transport_map will not be unnoticed.
808 This is a macro because we want the static storage variables to be
811 #define LAZY_RETRIEVE_INFO(info) do { \
812 static struct transport_info *last_info; \
813 static int last_fd = -1; \
814 static unsigned int last_tick; \
815 if (!transport_map) \
817 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
821 info = hash_table_get (transport_map, (void *)(intptr_t) fd); \
824 last_tick = transport_map_modified_tick; \
829 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
832 timeout = opt.read_timeout;
836 if (info && info->imp->poller)
837 test = info->imp->poller (fd, timeout, wf, info->ctx);
839 test = sock_poll (fd, timeout, wf);
848 /* Read no more than BUFSIZE bytes of data from FD, storing them to
849 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
850 received after that many seconds. If TIMEOUT is -1, the value of
851 opt.timeout is used for TIMEOUT. */
854 fd_read (int fd, char *buf, int bufsize, double timeout)
856 struct transport_info *info;
857 LAZY_RETRIEVE_INFO (info);
858 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
860 if (info && info->imp->reader)
861 return info->imp->reader (fd, buf, bufsize, info->ctx);
863 return sock_read (fd, buf, bufsize);
866 /* Like fd_read, except it provides a "preview" of the data that will
867 be read by subsequent calls to fd_read. Specifically, it copies no
868 more than BUFSIZE bytes of the currently available data to BUF and
869 returns the number of bytes copied. Return values and timeout
870 semantics are the same as those of fd_read.
872 CAVEAT: Do not assume that the first subsequent call to fd_read
873 will retrieve the same amount of data. Reading can return more or
874 less data, depending on the TCP implementation and other
875 circumstances. However, barring an error, it can be expected that
876 all the peeked data will eventually be read by fd_read. */
879 fd_peek (int fd, char *buf, int bufsize, double timeout)
881 struct transport_info *info;
882 LAZY_RETRIEVE_INFO (info);
883 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
885 if (info && info->imp->peeker)
886 return info->imp->peeker (fd, buf, bufsize, info->ctx);
888 return sock_peek (fd, buf, bufsize);
891 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
892 the operation aborts if no data is received after that many
893 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
897 fd_write (int fd, char *buf, int bufsize, double timeout)
900 struct transport_info *info;
901 LAZY_RETRIEVE_INFO (info);
903 /* `write' may write less than LEN bytes, thus the loop keeps trying
904 it until all was written, or an error occurred. */
908 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
910 if (info && info->imp->writer)
911 res = info->imp->writer (fd, buf, bufsize, info->ctx);
913 res = sock_write (fd, buf, bufsize);
922 /* Report the most recent error(s) on FD. This should only be called
923 after fd_* functions, such as fd_read and fd_write, and only if
924 they return a negative result. For errors coming from other calls
925 such as setsockopt or fopen, strerror should continue to be
928 If the transport doesn't support error messages or doesn't supply
929 one, strerror(errno) is returned. The returned error message
930 should not be used after fd_close has been called. */
935 /* Don't bother with LAZY_RETRIEVE_INFO, as this will only be called
936 in case of error, never in a tight loop. */
937 struct transport_info *info = NULL;
939 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
941 if (info && info->imp->errstr)
943 const char *err = info->imp->errstr (fd, info->ctx);
946 /* else, fall through and print the system error. */
948 return strerror (errno);
951 /* Close the file descriptor FD. */
956 struct transport_info *info;
960 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
961 per socket, so that particular optimization wouldn't work. */
964 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
966 if (info && info->imp->closer)
967 info->imp->closer (fd, info->ctx);
973 hash_table_remove (transport_map, (void *)(intptr_t) fd);
975 ++transport_map_modified_tick;