1 /* Establishing and handling network connections.
2 Copyright (C) 1996-2005 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 GNU Wget is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with Wget; if not, write to the Free Software Foundation, Inc.,
18 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
20 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
40 # include <sys/socket.h>
42 # include <netinet/in.h>
44 # include <arpa/inet.h>
46 #endif /* not WINDOWS */
50 #ifdef HAVE_SYS_SELECT_H
51 # include <sys/select.h>
52 #endif /* HAVE_SYS_SELECT_H */
60 /* Define sockaddr_storage where unavailable (presumably on IPv4-only
64 # ifndef HAVE_STRUCT_SOCKADDR_STORAGE
65 # define sockaddr_storage sockaddr_in
67 #endif /* ENABLE_IPV6 */
69 /* Fill SA as per the data in IP and PORT. SA shoult point to struct
70 sockaddr_storage if ENABLE_IPV6 is defined, to struct sockaddr_in
74 sockaddr_set_data (struct sockaddr *sa, const ip_address *ip, int port)
80 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
82 sin->sin_family = AF_INET;
83 sin->sin_port = htons (port);
84 sin->sin_addr = ip->data.d4;
90 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
92 sin6->sin6_family = AF_INET6;
93 sin6->sin6_port = htons (port);
94 sin6->sin6_addr = ip->data.d6;
95 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
96 sin6->sin6_scope_id = ip->ipv6_scope;
100 #endif /* ENABLE_IPV6 */
106 /* Get the data of SA, specifically the IP address and the port. If
107 you're not interested in one or the other information, pass NULL as
111 sockaddr_get_data (const struct sockaddr *sa, ip_address *ip, int *port)
113 switch (sa->sa_family)
117 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
120 ip->family = AF_INET;
121 ip->data.d4 = sin->sin_addr;
124 *port = ntohs (sin->sin_port);
130 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
133 ip->family = AF_INET6;
134 ip->data.d6 = sin6->sin6_addr;
135 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
136 ip->ipv6_scope = sin6->sin6_scope_id;
140 *port = ntohs (sin6->sin6_port);
149 /* Return the size of the sockaddr structure depending on its
153 sockaddr_size (const struct sockaddr *sa)
155 switch (sa->sa_family)
158 return sizeof (struct sockaddr_in);
161 return sizeof (struct sockaddr_in6);
168 /* Resolve the bind address specified via --bind-address and store it
169 to SA. The resolved value is stored in a static variable and
170 reused after the first invocation of this function.
172 Returns true on success, false on failure. */
175 resolve_bind_address (struct sockaddr *sa)
177 struct address_list *al;
179 /* Make sure this is called only once. opt.bind_address doesn't
180 change during a Wget run. */
181 static bool called, should_bind;
182 static ip_address ip;
186 sockaddr_set_data (sa, &ip, 0);
191 al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
194 /* #### We should be able to print the error message here. */
195 logprintf (LOG_NOTQUIET,
196 _("%s: unable to resolve bind address `%s'; disabling bind.\n"),
197 exec_name, opt.bind_address);
202 /* Pick the first address in the list and use it as bind address.
203 Perhaps we should try multiple addresses in succession, but I
204 don't think that's necessary in practice. */
205 ip = *address_list_address_at (al, 0);
206 address_list_release (al);
208 sockaddr_set_data (sa, &ip, 0);
215 const struct sockaddr *addr;
221 connect_with_timeout_callback (void *arg)
223 struct cwt_context *ctx = (struct cwt_context *)arg;
224 ctx->result = connect (ctx->fd, ctx->addr, ctx->addrlen);
227 /* Like connect, but specifies a timeout. If connecting takes longer
228 than TIMEOUT seconds, -1 is returned and errno is set to
232 connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
235 struct cwt_context ctx;
238 ctx.addrlen = addrlen;
240 if (run_with_timeout (timeout, connect_with_timeout_callback, &ctx))
245 if (ctx.result == -1 && errno == EINTR)
250 /* Connect via TCP to the specified address and port.
252 If PRINT is non-NULL, it is the host name to print that we're
256 connect_to_ip (const ip_address *ip, int port, const char *print)
258 struct sockaddr_storage ss;
259 struct sockaddr *sa = (struct sockaddr *)&ss;
262 /* If PRINT is non-NULL, print the "Connecting to..." line, with
263 PRINT being the host name we're connecting to. */
266 const char *txt_addr = print_address (ip);
267 if (print && 0 != strcmp (print, txt_addr))
268 logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
269 escnonprint (print), txt_addr, port);
271 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
274 /* Store the sockaddr info to SA. */
275 sockaddr_set_data (sa, ip, port);
277 /* Create the socket of the family appropriate for the address. */
278 sock = socket (sa->sa_family, SOCK_STREAM, 0);
282 #if defined(ENABLE_IPV6) && defined(IPV6_V6ONLY)
285 /* In case of error, we will go on anyway... */
286 int err = setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof (on));
289 DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
293 /* For very small rate limits, set the buffer size (and hence,
294 hopefully, the kernel's TCP window size) to the per-second limit.
295 That way we should never have to sleep for more than 1s between
297 if (opt.limit_rate && opt.limit_rate < 8192)
299 int bufsize = opt.limit_rate;
301 bufsize = 512; /* avoid pathologically small values */
303 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
304 (void *)&bufsize, (socklen_t)sizeof (bufsize));
306 /* When we add limit_rate support for writing, which is useful
307 for POST, we should also set SO_SNDBUF here. */
310 if (opt.bind_address)
312 /* Bind the client side of the socket to the requested
314 struct sockaddr_storage bind_ss;
315 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
316 if (resolve_bind_address (bind_sa))
318 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
323 /* Connect the socket to the remote endpoint. */
324 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
325 opt.connect_timeout) < 0)
331 logprintf (LOG_VERBOSE, _("connected.\n"));
332 DEBUGP (("Created socket %d.\n", sock));
337 /* Protect errno from possible modifications by close and
339 int save_errno = errno;
343 logprintf (LOG_VERBOSE, _("failed: %s.\n"), strerror (errno));
349 /* Connect via TCP to a remote host on the specified port.
351 HOST is resolved as an Internet host name. If HOST resolves to
352 more than one IP address, they are tried in the order returned by
353 DNS until connecting to one of them succeeds. */
356 connect_to_host (const char *host, int port)
361 struct address_list *al = lookup_host (host, 0);
367 address_list_get_bounds (al, &start, &end);
368 for (i = start; i < end; i++)
370 const ip_address *ip = address_list_address_at (al, i);
371 sock = connect_to_ip (ip, port, host);
375 address_list_set_connected (al);
376 address_list_release (al);
380 /* The attempt to connect has failed. Continue with the loop
381 and try next address. */
383 address_list_set_faulty (al, i);
386 /* Failed to connect to any of the addresses in AL. */
388 if (address_list_connected_p (al))
390 /* We connected to AL before, but cannot do so now. That might
391 indicate that our DNS cache entry for HOST has expired. */
392 address_list_release (al);
393 al = lookup_host (host, LH_REFRESH);
396 address_list_release (al);
401 /* Create a socket, bind it to local interface BIND_ADDRESS on port
402 *PORT, set up a listen backlog, and return the resulting socket, or
405 BIND_ADDRESS is the address of the interface to bind to. If it is
406 NULL, the socket is bound to the default address. PORT should
407 point to the port number that will be used for the binding. If
408 that number is 0, the system will choose a suitable port, and the
409 chosen value will be written to *PORT.
411 Calling accept() on such a socket waits for and accepts incoming
415 bind_local (const ip_address *bind_address, int *port)
418 struct sockaddr_storage ss;
419 struct sockaddr *sa = (struct sockaddr *)&ss;
421 /* For setting options with setsockopt. */
423 void *setopt_ptr = (void *)&setopt_val;
424 socklen_t setopt_size = sizeof (setopt_val);
426 sock = socket (bind_address->family, SOCK_STREAM, 0);
431 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
435 sockaddr_set_data (sa, bind_address, *port);
436 if (bind (sock, sa, sockaddr_size (sa)) < 0)
441 DEBUGP (("Local socket fd %d bound.\n", sock));
443 /* If *PORT is 0, find out which port we've bound to. */
446 socklen_t addrlen = sockaddr_size (sa);
447 if (getsockname (sock, sa, &addrlen) < 0)
449 /* If we can't find out the socket's local address ("name"),
450 something is seriously wrong with the socket, and it's
451 unusable for us anyway because we must know the chosen
456 sockaddr_get_data (sa, NULL, port);
457 DEBUGP (("binding to address %s using port %i.\n",
458 print_address (bind_address), *port));
460 if (listen (sock, 1) < 0)
468 /* Like a call to accept(), but with the added check for timeout.
470 In other words, accept a client connection on LOCAL_SOCK, and
471 return the new socket used for communication with the client.
472 LOCAL_SOCK should have been bound, e.g. using bind_local().
474 The caller is blocked until a connection is established. If no
475 connection is established for opt.connect_timeout seconds, the
476 function exits with an error status. */
479 accept_connection (int local_sock)
483 /* We don't need the values provided by accept, but accept
484 apparently requires them to be present. */
485 struct sockaddr_storage ss;
486 struct sockaddr *sa = (struct sockaddr *)&ss;
487 socklen_t addrlen = sizeof (ss);
489 if (opt.connect_timeout)
491 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
497 sock = accept (local_sock, sa, &addrlen);
498 DEBUGP (("Accepted client at socket %d.\n", sock));
502 /* Get the IP address associated with the connection on FD and store
503 it to IP. Return true on success, false otherwise.
505 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
506 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
507 returns the address of the remote (peer's) side of the socket. */
510 socket_ip_address (int sock, ip_address *ip, int endpoint)
512 struct sockaddr_storage storage;
513 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
514 socklen_t addrlen = sizeof (storage);
517 if (endpoint == ENDPOINT_LOCAL)
518 ret = getsockname (sock, sockaddr, &addrlen);
519 else if (endpoint == ENDPOINT_PEER)
520 ret = getpeername (sock, sockaddr, &addrlen);
526 ip->family = sockaddr->sa_family;
527 switch (sockaddr->sa_family)
532 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
533 ip->data.d6 = sa6->sin6_addr;
534 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
535 ip->ipv6_scope = sa6->sin6_scope_id;
537 DEBUGP (("conaddr is: %s\n", print_address (ip)));
543 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
544 ip->data.d4 = sa->sin_addr;
545 DEBUGP (("conaddr is: %s\n", print_address (ip)));
553 /* Return true if the error from the connect code can be considered
554 retryable. Wget normally retries after errors, but the exception
555 are the "unsupported protocol" type errors (possible on IPv4/IPv6
556 dual family systems) and "connection refused". */
559 retryable_socket_connect_error (int err)
561 /* Have to guard against some of these values not being defined.
562 Cannot use a switch statement because some of the values might be
566 || err == EAFNOSUPPORT
569 || err == EPFNOSUPPORT
571 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
572 || err == ESOCKTNOSUPPORT
574 #ifdef EPROTONOSUPPORT
575 || err == EPROTONOSUPPORT
578 || err == ENOPROTOOPT
580 /* Apparently, older versions of Linux and BSD used EINVAL
581 instead of EAFNOSUPPORT and such. */
586 if (!opt.retry_connrefused)
587 if (err == ECONNREFUSED
589 || err == ENETUNREACH /* network is unreachable */
592 || err == EHOSTUNREACH /* host is unreachable */
600 /* Wait for a single descriptor to become available, timing out after
601 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
602 -1 for error. The argument WAIT_FOR can be a combination of
603 WAIT_FOR_READ and WAIT_FOR_WRITE.
605 This is a mere convenience wrapper around the select call, and
606 should be taken as such (for example, it doesn't implement Wget's
607 0-timeout-means-no-timeout semantics.) */
610 select_fd (int fd, double maxtime, int wait_for)
613 fd_set *rd = NULL, *wr = NULL;
614 struct timeval tmout;
619 if (wait_for & WAIT_FOR_READ)
621 if (wait_for & WAIT_FOR_WRITE)
624 tmout.tv_sec = (long) maxtime;
625 tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
628 result = select (fd + 1, rd, wr, NULL, &tmout);
629 while (result < 0 && errno == EINTR);
634 /* Return true iff the connection to the remote site established
635 through SOCK is still open.
637 Specifically, this function returns true if SOCK is not ready for
638 reading. This is because, when the connection closes, the socket
639 is ready for reading because EOF is about to be delivered. A side
640 effect of this method is that sockets that have pending data are
641 considered non-open. This is actually a good thing for callers of
642 this function, where such pending data can only be unwanted
643 leftover from a previous request. */
646 test_socket_open (int sock)
651 /* Check if we still have a valid (non-EOF) connection. From Andrew
652 * Maholski's code in the Unix Socket FAQ. */
654 FD_ZERO (&check_set);
655 FD_SET (sock, &check_set);
657 /* Wait one microsecond */
661 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
662 /* We got a timeout, it means we're still connected. */
665 /* Read now would not wait, it means we have either pending data
670 /* Basic socket operations, mostly EINTR wrappers. */
673 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
674 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
675 # define close(fd) closesocket (fd)
679 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
680 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
684 sock_read (int fd, char *buf, int bufsize)
688 res = read (fd, buf, bufsize);
689 while (res == -1 && errno == EINTR);
694 sock_write (int fd, char *buf, int bufsize)
698 res = write (fd, buf, bufsize);
699 while (res == -1 && errno == EINTR);
704 sock_poll (int fd, double timeout, int wait_for)
706 return select_fd (fd, timeout, wait_for);
710 sock_peek (int fd, char *buf, int bufsize)
714 res = recv (fd, buf, bufsize, MSG_PEEK);
715 while (res == -1 && errno == EINTR);
723 DEBUGP (("Closed fd %d\n", fd));
729 /* Reading and writing from the network. We build around the socket
730 (file descriptor) API, but support "extended" operations for things
731 that are not mere file descriptors under the hood, such as SSL
734 That way the user code can call fd_read(fd, ...) and we'll run read
735 or SSL_read or whatever is necessary. */
737 static struct hash_table *transport_map;
738 static unsigned int transport_map_modified_tick;
740 struct transport_info {
741 struct transport_implementation *imp;
745 /* Register the transport layer operations that will be used when
746 reading, writing, and polling FD.
748 This should be used for transport layers like SSL that piggyback on
749 sockets. FD should otherwise be a real socket, on which you can
750 call getpeername, etc. */
753 fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
755 struct transport_info *info;
757 /* The file descriptor must be non-negative to be registered.
758 Negative values are ignored by fd_close(), and -1 cannot be used as
762 info = xnew (struct transport_info);
766 transport_map = hash_table_new (0, NULL, NULL);
767 hash_table_put (transport_map, (void *) fd, info);
768 ++transport_map_modified_tick;
771 /* Return context of the transport registered with
772 fd_register_transport. This assumes fd_register_transport was
773 previously called on FD. */
776 fd_transport_context (int fd)
778 struct transport_info *info = hash_table_get (transport_map, (void *) fd);
782 /* When fd_read/fd_write are called multiple times in a loop, they should
783 remember the INFO pointer instead of fetching it every time. It is
784 not enough to compare FD to LAST_FD because FD might have been
785 closed and reopened. modified_tick ensures that changes to
786 transport_map will not be unnoticed.
788 This is a macro because we want the static storage variables to be
791 #define LAZY_RETRIEVE_INFO(info) do { \
792 static struct transport_info *last_info; \
793 static int last_fd = -1; \
794 static unsigned int last_tick; \
795 if (!transport_map) \
797 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
801 info = hash_table_get (transport_map, (void *) fd); \
804 last_tick = transport_map_modified_tick; \
809 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
812 timeout = opt.read_timeout;
816 if (info && info->imp->poller)
817 test = info->imp->poller (fd, timeout, wf, info->ctx);
819 test = sock_poll (fd, timeout, wf);
828 /* Read no more than BUFSIZE bytes of data from FD, storing them to
829 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
830 received after that many seconds. If TIMEOUT is -1, the value of
831 opt.timeout is used for TIMEOUT. */
834 fd_read (int fd, char *buf, int bufsize, double timeout)
836 struct transport_info *info;
837 LAZY_RETRIEVE_INFO (info);
838 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
840 if (info && info->imp->reader)
841 return info->imp->reader (fd, buf, bufsize, info->ctx);
843 return sock_read (fd, buf, bufsize);
846 /* Like fd_read, except it provides a "preview" of the data that will
847 be read by subsequent calls to fd_read. Specifically, it copies no
848 more than BUFSIZE bytes of the currently available data to BUF and
849 returns the number of bytes copied. Return values and timeout
850 semantics are the same as those of fd_read.
852 CAVEAT: Do not assume that the first subsequent call to fd_read
853 will retrieve the same amount of data. Reading can return more or
854 less data, depending on the TCP implementation and other
855 circumstances. However, barring an error, it can be expected that
856 all the peeked data will eventually be read by fd_read. */
859 fd_peek (int fd, char *buf, int bufsize, double timeout)
861 struct transport_info *info;
862 LAZY_RETRIEVE_INFO (info);
863 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
865 if (info && info->imp->peeker)
866 return info->imp->peeker (fd, buf, bufsize, info->ctx);
868 return sock_peek (fd, buf, bufsize);
871 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
872 the operation aborts if no data is received after that many
873 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
877 fd_write (int fd, char *buf, int bufsize, double timeout)
880 struct transport_info *info;
881 LAZY_RETRIEVE_INFO (info);
883 /* `write' may write less than LEN bytes, thus the loop keeps trying
884 it until all was written, or an error occurred. */
888 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
890 if (info && info->imp->writer)
891 res = info->imp->writer (fd, buf, bufsize, info->ctx);
893 res = sock_write (fd, buf, bufsize);
902 /* Report the most recent error(s) on FD. This should only be called
903 after fd_* functions, such as fd_read and fd_write, and only if
904 they return a negative result. For errors coming from other calls
905 such as setsockopt or fopen, strerror should continue to be
908 If the transport doesn't support error messages or doesn't supply
909 one, strerror(errno) is returned. The returned error message
910 should not be used after fd_close has been called. */
915 /* Don't bother with LAZY_RETRIEVE_INFO, as this will only be called
916 in case of error, never in a tight loop. */
917 struct transport_info *info = NULL;
919 info = hash_table_get (transport_map, (void *) fd);
921 if (info && info->imp->errstr)
923 const char *err = info->imp->errstr (fd, info->ctx);
926 /* else, fall through and print the system error. */
928 return strerror (errno);
931 /* Close the file descriptor FD. */
936 struct transport_info *info;
940 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
941 per socket, so that particular optimization wouldn't work. */
944 info = hash_table_get (transport_map, (void *) fd);
946 if (info && info->imp->closer)
947 info->imp->closer (fd, info->ctx);
953 hash_table_remove (transport_map, (void *) fd);
955 ++transport_map_modified_tick;