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 */
59 /* Define sockaddr_storage where unavailable (presumably on IPv4-only
63 # ifndef HAVE_STRUCT_SOCKADDR_STORAGE
64 # define sockaddr_storage sockaddr_in
66 #endif /* ENABLE_IPV6 */
68 /* Fill SA as per the data in IP and PORT. SA shoult point to struct
69 sockaddr_storage if ENABLE_IPV6 is defined, to struct sockaddr_in
73 sockaddr_set_data (struct sockaddr *sa, const ip_address *ip, int port)
79 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
81 sin->sin_family = AF_INET;
82 sin->sin_port = htons (port);
83 sin->sin_addr = ip->data.d4;
89 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
91 sin6->sin6_family = AF_INET6;
92 sin6->sin6_port = htons (port);
93 sin6->sin6_addr = ip->data.d6;
94 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
95 sin6->sin6_scope_id = ip->ipv6_scope;
99 #endif /* ENABLE_IPV6 */
105 /* Get the data of SA, specifically the IP address and the port. If
106 you're not interested in one or the other information, pass NULL as
110 sockaddr_get_data (const struct sockaddr *sa, ip_address *ip, int *port)
112 switch (sa->sa_family)
116 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
119 ip->family = AF_INET;
120 ip->data.d4 = sin->sin_addr;
123 *port = ntohs (sin->sin_port);
129 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
132 ip->family = AF_INET6;
133 ip->data.d6 = sin6->sin6_addr;
134 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
135 ip->ipv6_scope = sin6->sin6_scope_id;
139 *port = ntohs (sin6->sin6_port);
148 /* Return the size of the sockaddr structure depending on its
152 sockaddr_size (const struct sockaddr *sa)
154 switch (sa->sa_family)
157 return sizeof (struct sockaddr_in);
160 return sizeof (struct sockaddr_in6);
167 /* Resolve the bind address specified via --bind-address and store it
168 to SA. The resolved value is stored in a static variable and
169 reused after the first invocation of this function.
171 Returns true on success, false on failure. */
174 resolve_bind_address (struct sockaddr *sa)
176 struct address_list *al;
178 /* Make sure this is called only once. opt.bind_address doesn't
179 change during a Wget run. */
180 static bool called, should_bind;
181 static ip_address ip;
185 sockaddr_set_data (sa, &ip, 0);
190 al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
193 /* #### We should be able to print the error message here. */
194 logprintf (LOG_NOTQUIET,
195 _("%s: unable to resolve bind address `%s'; disabling bind.\n"),
196 exec_name, opt.bind_address);
201 /* Pick the first address in the list and use it as bind address.
202 Perhaps we should try multiple addresses in succession, but I
203 don't think that's necessary in practice. */
204 ip = *address_list_address_at (al, 0);
205 address_list_release (al);
207 sockaddr_set_data (sa, &ip, 0);
214 const struct sockaddr *addr;
220 connect_with_timeout_callback (void *arg)
222 struct cwt_context *ctx = (struct cwt_context *)arg;
223 ctx->result = connect (ctx->fd, ctx->addr, ctx->addrlen);
226 /* Like connect, but specifies a timeout. If connecting takes longer
227 than TIMEOUT seconds, -1 is returned and errno is set to
231 connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
234 struct cwt_context ctx;
237 ctx.addrlen = addrlen;
239 if (run_with_timeout (timeout, connect_with_timeout_callback, &ctx))
244 if (ctx.result == -1 && errno == EINTR)
249 /* Connect via TCP to the specified address and port.
251 If PRINT is non-NULL, it is the host name to print that we're
255 connect_to_ip (const ip_address *ip, int port, const char *print)
257 struct sockaddr_storage ss;
258 struct sockaddr *sa = (struct sockaddr *)&ss;
261 /* If PRINT is non-NULL, print the "Connecting to..." line, with
262 PRINT being the host name we're connecting to. */
265 const char *txt_addr = print_address (ip);
266 if (print && 0 != strcmp (print, txt_addr))
267 logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
268 escnonprint (print), txt_addr, port);
270 logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
273 /* Store the sockaddr info to SA. */
274 sockaddr_set_data (sa, ip, port);
276 /* Create the socket of the family appropriate for the address. */
277 sock = socket (sa->sa_family, SOCK_STREAM, 0);
281 #if defined(ENABLE_IPV6) && defined(IPV6_V6ONLY)
284 /* In case of error, we will go on anyway... */
285 int err = setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof (on));
288 DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
292 /* For very small rate limits, set the buffer size (and hence,
293 hopefully, the kernel's TCP window size) to the per-second limit.
294 That way we should never have to sleep for more than 1s between
296 if (opt.limit_rate && opt.limit_rate < 8192)
298 int bufsize = opt.limit_rate;
300 bufsize = 512; /* avoid pathologically small values */
302 setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
303 (void *)&bufsize, (socklen_t)sizeof (bufsize));
305 /* When we add limit_rate support for writing, which is useful
306 for POST, we should also set SO_SNDBUF here. */
309 if (opt.bind_address)
311 /* Bind the client side of the socket to the requested
313 struct sockaddr_storage bind_ss;
314 struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
315 if (resolve_bind_address (bind_sa))
317 if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
322 /* Connect the socket to the remote endpoint. */
323 if (connect_with_timeout (sock, sa, sockaddr_size (sa),
324 opt.connect_timeout) < 0)
330 logprintf (LOG_VERBOSE, _("connected.\n"));
331 DEBUGP (("Created socket %d.\n", sock));
336 /* Protect errno from possible modifications by close and
338 int save_errno = errno;
342 logprintf (LOG_VERBOSE, _("failed: %s.\n"), strerror (errno));
348 /* Connect via TCP to a remote host on the specified port.
350 HOST is resolved as an Internet host name. If HOST resolves to
351 more than one IP address, they are tried in the order returned by
352 DNS until connecting to one of them succeeds. */
355 connect_to_host (const char *host, int port)
360 struct address_list *al = lookup_host (host, 0);
365 logprintf (LOG_NOTQUIET,
366 _("%s: unable to resolve host address `%s'\n"),
371 address_list_get_bounds (al, &start, &end);
372 for (i = start; i < end; i++)
374 const ip_address *ip = address_list_address_at (al, i);
375 sock = connect_to_ip (ip, port, host);
379 address_list_set_connected (al);
380 address_list_release (al);
384 /* The attempt to connect has failed. Continue with the loop
385 and try next address. */
387 address_list_set_faulty (al, i);
390 /* Failed to connect to any of the addresses in AL. */
392 if (address_list_connected_p (al))
394 /* We connected to AL before, but cannot do so now. That might
395 indicate that our DNS cache entry for HOST has expired. */
396 address_list_release (al);
397 al = lookup_host (host, LH_REFRESH);
400 address_list_release (al);
405 /* Create a socket, bind it to local interface BIND_ADDRESS on port
406 *PORT, set up a listen backlog, and return the resulting socket, or
409 BIND_ADDRESS is the address of the interface to bind to. If it is
410 NULL, the socket is bound to the default address. PORT should
411 point to the port number that will be used for the binding. If
412 that number is 0, the system will choose a suitable port, and the
413 chosen value will be written to *PORT.
415 Calling accept() on such a socket waits for and accepts incoming
419 bind_local (const ip_address *bind_address, int *port)
422 struct sockaddr_storage ss;
423 struct sockaddr *sa = (struct sockaddr *)&ss;
425 /* For setting options with setsockopt. */
427 void *setopt_ptr = (void *)&setopt_val;
428 socklen_t setopt_size = sizeof (setopt_val);
430 sock = socket (bind_address->family, SOCK_STREAM, 0);
435 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
439 sockaddr_set_data (sa, bind_address, *port);
440 if (bind (sock, sa, sockaddr_size (sa)) < 0)
445 DEBUGP (("Local socket fd %d bound.\n", sock));
447 /* If *PORT is 0, find out which port we've bound to. */
450 socklen_t addrlen = sockaddr_size (sa);
451 if (getsockname (sock, sa, &addrlen) < 0)
453 /* If we can't find out the socket's local address ("name"),
454 something is seriously wrong with the socket, and it's
455 unusable for us anyway because we must know the chosen
460 sockaddr_get_data (sa, NULL, port);
461 DEBUGP (("binding to address %s using port %i.\n",
462 print_address (bind_address), *port));
464 if (listen (sock, 1) < 0)
472 /* Like a call to accept(), but with the added check for timeout.
474 In other words, accept a client connection on LOCAL_SOCK, and
475 return the new socket used for communication with the client.
476 LOCAL_SOCK should have been bound, e.g. using bind_local().
478 The caller is blocked until a connection is established. If no
479 connection is established for opt.connect_timeout seconds, the
480 function exits with an error status. */
483 accept_connection (int local_sock)
487 /* We don't need the values provided by accept, but accept
488 apparently requires them to be present. */
489 struct sockaddr_storage ss;
490 struct sockaddr *sa = (struct sockaddr *)&ss;
491 socklen_t addrlen = sizeof (ss);
493 if (opt.connect_timeout)
495 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
501 sock = accept (local_sock, sa, &addrlen);
502 DEBUGP (("Accepted client at socket %d.\n", sock));
506 /* Get the IP address associated with the connection on FD and store
507 it to IP. Return true on success, false otherwise.
509 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
510 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
511 returns the address of the remote (peer's) side of the socket. */
514 socket_ip_address (int sock, ip_address *ip, int endpoint)
516 struct sockaddr_storage storage;
517 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
518 socklen_t addrlen = sizeof (storage);
521 if (endpoint == ENDPOINT_LOCAL)
522 ret = getsockname (sock, sockaddr, &addrlen);
523 else if (endpoint == ENDPOINT_PEER)
524 ret = getpeername (sock, sockaddr, &addrlen);
530 ip->family = sockaddr->sa_family;
531 switch (sockaddr->sa_family)
536 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
537 ip->data.d6 = sa6->sin6_addr;
538 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
539 ip->ipv6_scope = sa6->sin6_scope_id;
541 DEBUGP (("conaddr is: %s\n", print_address (ip)));
547 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
548 ip->data.d4 = sa->sin_addr;
549 DEBUGP (("conaddr is: %s\n", print_address (ip)));
557 /* Return true if the error from the connect code can be considered
558 retryable. Wget normally retries after errors, but the exception
559 are the "unsupported protocol" type errors (possible on IPv4/IPv6
560 dual family systems) and "connection refused". */
563 retryable_socket_connect_error (int err)
565 /* Have to guard against some of these values not being defined.
566 Cannot use a switch statement because some of the values might be
570 || err == EAFNOSUPPORT
573 || err == EPFNOSUPPORT
575 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
576 || err == ESOCKTNOSUPPORT
578 #ifdef EPROTONOSUPPORT
579 || err == EPROTONOSUPPORT
582 || err == ENOPROTOOPT
584 /* Apparently, older versions of Linux and BSD used EINVAL
585 instead of EAFNOSUPPORT and such. */
590 if (!opt.retry_connrefused)
591 if (err == ECONNREFUSED
593 || err == ENETUNREACH /* network is unreachable */
596 || err == EHOSTUNREACH /* host is unreachable */
604 /* Wait for a single descriptor to become available, timing out after
605 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
606 -1 for error. The argument WAIT_FOR can be a combination of
607 WAIT_FOR_READ and WAIT_FOR_WRITE.
609 This is a mere convenience wrapper around the select call, and
610 should be taken as such (for example, it doesn't implement Wget's
611 0-timeout-means-no-timeout semantics.) */
614 select_fd (int fd, double maxtime, int wait_for)
617 fd_set *rd = NULL, *wr = NULL;
618 struct timeval tmout;
623 if (wait_for & WAIT_FOR_READ)
625 if (wait_for & WAIT_FOR_WRITE)
628 tmout.tv_sec = (long) maxtime;
629 tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
632 result = select (fd + 1, rd, wr, NULL, &tmout);
633 while (result < 0 && errno == EINTR);
638 /* Return true iff the connection to the remote site established
639 through SOCK is still open.
641 Specifically, this function returns true if SOCK is not ready for
642 reading. This is because, when the connection closes, the socket
643 is ready for reading because EOF is about to be delivered. A side
644 effect of this method is that sockets that have pending data are
645 considered non-open. This is actually a good thing for callers of
646 this function, where such pending data can only be unwanted
647 leftover from a previous request. */
650 test_socket_open (int sock)
655 /* Check if we still have a valid (non-EOF) connection. From Andrew
656 * Maholski's code in the Unix Socket FAQ. */
658 FD_ZERO (&check_set);
659 FD_SET (sock, &check_set);
661 /* Wait one microsecond */
665 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
666 /* We got a timeout, it means we're still connected. */
669 /* Read now would not wait, it means we have either pending data
674 /* Basic socket operations, mostly EINTR wrappers. */
676 #if defined(WINDOWS) || defined(MSDOS)
677 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
678 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
679 # define close(fd) closesocket (fd)
683 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
684 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
688 sock_read (int fd, char *buf, int bufsize)
692 res = read (fd, buf, bufsize);
693 while (res == -1 && errno == EINTR);
698 sock_write (int fd, char *buf, int bufsize)
702 res = write (fd, buf, bufsize);
703 while (res == -1 && errno == EINTR);
708 sock_poll (int fd, double timeout, int wait_for)
710 return select_fd (fd, timeout, wait_for);
714 sock_peek (int fd, char *buf, int bufsize)
718 res = recv (fd, buf, bufsize, MSG_PEEK);
719 while (res == -1 && errno == EINTR);
727 DEBUGP (("Closed fd %d\n", fd));
733 /* Reading and writing from the network. We build around the socket
734 (file descriptor) API, but support "extended" operations for things
735 that are not mere file descriptors under the hood, such as SSL
738 That way the user code can call fd_read(fd, ...) and we'll run read
739 or SSL_read or whatever is necessary. */
741 static struct hash_table *transport_map;
742 static unsigned int transport_map_modified_tick;
744 struct transport_info {
745 struct transport_implementation *imp;
749 /* Register the transport layer operations that will be used when
750 reading, writing, and polling FD.
752 This should be used for transport layers like SSL that piggyback on
753 sockets. FD should otherwise be a real socket, on which you can
754 call getpeername, etc. */
757 fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
759 struct transport_info *info;
761 /* The file descriptor must be non-negative to be registered.
762 Negative values are ignored by fd_close(), and -1 cannot be used as
766 info = xnew (struct transport_info);
770 transport_map = hash_table_new (0, NULL, NULL);
771 hash_table_put (transport_map, (void *)(intptr_t) fd, info);
772 ++transport_map_modified_tick;
775 /* Return context of the transport registered with
776 fd_register_transport. This assumes fd_register_transport was
777 previously called on FD. */
780 fd_transport_context (int fd)
782 struct transport_info *info = hash_table_get (transport_map, (void *)(intptr_t) fd);
786 /* When fd_read/fd_write are called multiple times in a loop, they should
787 remember the INFO pointer instead of fetching it every time. It is
788 not enough to compare FD to LAST_FD because FD might have been
789 closed and reopened. modified_tick ensures that changes to
790 transport_map will not be unnoticed.
792 This is a macro because we want the static storage variables to be
795 #define LAZY_RETRIEVE_INFO(info) do { \
796 static struct transport_info *last_info; \
797 static int last_fd = -1; \
798 static unsigned int last_tick; \
799 if (!transport_map) \
801 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
805 info = hash_table_get (transport_map, (void *)(intptr_t) fd); \
808 last_tick = transport_map_modified_tick; \
813 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
816 timeout = opt.read_timeout;
820 if (info && info->imp->poller)
821 test = info->imp->poller (fd, timeout, wf, info->ctx);
823 test = sock_poll (fd, timeout, wf);
832 /* Read no more than BUFSIZE bytes of data from FD, storing them to
833 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
834 received after that many seconds. If TIMEOUT is -1, the value of
835 opt.timeout is used for TIMEOUT. */
838 fd_read (int fd, char *buf, int bufsize, double timeout)
840 struct transport_info *info;
841 LAZY_RETRIEVE_INFO (info);
842 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
844 if (info && info->imp->reader)
845 return info->imp->reader (fd, buf, bufsize, info->ctx);
847 return sock_read (fd, buf, bufsize);
850 /* Like fd_read, except it provides a "preview" of the data that will
851 be read by subsequent calls to fd_read. Specifically, it copies no
852 more than BUFSIZE bytes of the currently available data to BUF and
853 returns the number of bytes copied. Return values and timeout
854 semantics are the same as those of fd_read.
856 CAVEAT: Do not assume that the first subsequent call to fd_read
857 will retrieve the same amount of data. Reading can return more or
858 less data, depending on the TCP implementation and other
859 circumstances. However, barring an error, it can be expected that
860 all the peeked data will eventually be read by fd_read. */
863 fd_peek (int fd, char *buf, int bufsize, double timeout)
865 struct transport_info *info;
866 LAZY_RETRIEVE_INFO (info);
867 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
869 if (info && info->imp->peeker)
870 return info->imp->peeker (fd, buf, bufsize, info->ctx);
872 return sock_peek (fd, buf, bufsize);
875 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
876 the operation aborts if no data is received after that many
877 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
881 fd_write (int fd, char *buf, int bufsize, double timeout)
884 struct transport_info *info;
885 LAZY_RETRIEVE_INFO (info);
887 /* `write' may write less than LEN bytes, thus the loop keeps trying
888 it until all was written, or an error occurred. */
892 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
894 if (info && info->imp->writer)
895 res = info->imp->writer (fd, buf, bufsize, info->ctx);
897 res = sock_write (fd, buf, bufsize);
906 /* Report the most recent error(s) on FD. This should only be called
907 after fd_* functions, such as fd_read and fd_write, and only if
908 they return a negative result. For errors coming from other calls
909 such as setsockopt or fopen, strerror should continue to be
912 If the transport doesn't support error messages or doesn't supply
913 one, strerror(errno) is returned. The returned error message
914 should not be used after fd_close has been called. */
919 /* Don't bother with LAZY_RETRIEVE_INFO, as this will only be called
920 in case of error, never in a tight loop. */
921 struct transport_info *info = NULL;
923 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
925 if (info && info->imp->errstr)
927 const char *err = info->imp->errstr (fd, info->ctx);
930 /* else, fall through and print the system error. */
932 return strerror (errno);
935 /* Close the file descriptor FD. */
940 struct transport_info *info;
944 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
945 per socket, so that particular optimization wouldn't work. */
948 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
950 if (info && info->imp->closer)
951 info->imp->closer (fd, info->ctx);
957 hash_table_remove (transport_map, (void *)(intptr_t) fd);
959 ++transport_map_modified_tick;