1 /* Establishing and handling network connections.
2 Copyright (C) 1996-2006 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 3 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, see <http://www.gnu.org/licenses/>.
19 In addition, as a special exception, the Free Software Foundation
20 gives permission to link the code of its release of Wget with the
21 OpenSSL project's "OpenSSL" library (or with modified versions of it
22 that use the same license as the "OpenSSL" library), and distribute
23 the linked executables. You must obey the GNU General Public License
24 in all respects for all of the code used other than "OpenSSL". If you
25 modify this file, you may extend this exception to your version of the
26 file, but you are not obligated to do so. If you do not wish to do
27 so, delete this exception statement from your version. */
39 # include <sys/socket.h>
41 # include <netinet/in.h>
43 # include <arpa/inet.h>
45 #endif /* not WINDOWS */
49 #ifdef HAVE_SYS_SELECT_H
50 # include <sys/select.h>
51 #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);
366 address_list_get_bounds (al, &start, &end);
367 for (i = start; i < end; i++)
369 const ip_address *ip = address_list_address_at (al, i);
370 sock = connect_to_ip (ip, port, host);
374 address_list_set_connected (al);
375 address_list_release (al);
379 /* The attempt to connect has failed. Continue with the loop
380 and try next address. */
382 address_list_set_faulty (al, i);
385 /* Failed to connect to any of the addresses in AL. */
387 if (address_list_connected_p (al))
389 /* We connected to AL before, but cannot do so now. That might
390 indicate that our DNS cache entry for HOST has expired. */
391 address_list_release (al);
392 al = lookup_host (host, LH_REFRESH);
395 address_list_release (al);
400 /* Create a socket, bind it to local interface BIND_ADDRESS on port
401 *PORT, set up a listen backlog, and return the resulting socket, or
404 BIND_ADDRESS is the address of the interface to bind to. If it is
405 NULL, the socket is bound to the default address. PORT should
406 point to the port number that will be used for the binding. If
407 that number is 0, the system will choose a suitable port, and the
408 chosen value will be written to *PORT.
410 Calling accept() on such a socket waits for and accepts incoming
414 bind_local (const ip_address *bind_address, int *port)
417 struct sockaddr_storage ss;
418 struct sockaddr *sa = (struct sockaddr *)&ss;
420 /* For setting options with setsockopt. */
422 void *setopt_ptr = (void *)&setopt_val;
423 socklen_t setopt_size = sizeof (setopt_val);
425 sock = socket (bind_address->family, SOCK_STREAM, 0);
430 setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
434 sockaddr_set_data (sa, bind_address, *port);
435 if (bind (sock, sa, sockaddr_size (sa)) < 0)
440 DEBUGP (("Local socket fd %d bound.\n", sock));
442 /* If *PORT is 0, find out which port we've bound to. */
445 socklen_t addrlen = sockaddr_size (sa);
446 if (getsockname (sock, sa, &addrlen) < 0)
448 /* If we can't find out the socket's local address ("name"),
449 something is seriously wrong with the socket, and it's
450 unusable for us anyway because we must know the chosen
455 sockaddr_get_data (sa, NULL, port);
456 DEBUGP (("binding to address %s using port %i.\n",
457 print_address (bind_address), *port));
459 if (listen (sock, 1) < 0)
467 /* Like a call to accept(), but with the added check for timeout.
469 In other words, accept a client connection on LOCAL_SOCK, and
470 return the new socket used for communication with the client.
471 LOCAL_SOCK should have been bound, e.g. using bind_local().
473 The caller is blocked until a connection is established. If no
474 connection is established for opt.connect_timeout seconds, the
475 function exits with an error status. */
478 accept_connection (int local_sock)
482 /* We don't need the values provided by accept, but accept
483 apparently requires them to be present. */
484 struct sockaddr_storage ss;
485 struct sockaddr *sa = (struct sockaddr *)&ss;
486 socklen_t addrlen = sizeof (ss);
488 if (opt.connect_timeout)
490 int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
496 sock = accept (local_sock, sa, &addrlen);
497 DEBUGP (("Accepted client at socket %d.\n", sock));
501 /* Get the IP address associated with the connection on FD and store
502 it to IP. Return true on success, false otherwise.
504 If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
505 (client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
506 returns the address of the remote (peer's) side of the socket. */
509 socket_ip_address (int sock, ip_address *ip, int endpoint)
511 struct sockaddr_storage storage;
512 struct sockaddr *sockaddr = (struct sockaddr *)&storage;
513 socklen_t addrlen = sizeof (storage);
516 if (endpoint == ENDPOINT_LOCAL)
517 ret = getsockname (sock, sockaddr, &addrlen);
518 else if (endpoint == ENDPOINT_PEER)
519 ret = getpeername (sock, sockaddr, &addrlen);
525 ip->family = sockaddr->sa_family;
526 switch (sockaddr->sa_family)
531 struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
532 ip->data.d6 = sa6->sin6_addr;
533 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
534 ip->ipv6_scope = sa6->sin6_scope_id;
536 DEBUGP (("conaddr is: %s\n", print_address (ip)));
542 struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
543 ip->data.d4 = sa->sin_addr;
544 DEBUGP (("conaddr is: %s\n", print_address (ip)));
552 /* Return true if the error from the connect code can be considered
553 retryable. Wget normally retries after errors, but the exception
554 are the "unsupported protocol" type errors (possible on IPv4/IPv6
555 dual family systems) and "connection refused". */
558 retryable_socket_connect_error (int err)
560 /* Have to guard against some of these values not being defined.
561 Cannot use a switch statement because some of the values might be
565 || err == EAFNOSUPPORT
568 || err == EPFNOSUPPORT
570 #ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
571 || err == ESOCKTNOSUPPORT
573 #ifdef EPROTONOSUPPORT
574 || err == EPROTONOSUPPORT
577 || err == ENOPROTOOPT
579 /* Apparently, older versions of Linux and BSD used EINVAL
580 instead of EAFNOSUPPORT and such. */
585 if (!opt.retry_connrefused)
586 if (err == ECONNREFUSED
588 || err == ENETUNREACH /* network is unreachable */
591 || err == EHOSTUNREACH /* host is unreachable */
599 /* Wait for a single descriptor to become available, timing out after
600 MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
601 -1 for error. The argument WAIT_FOR can be a combination of
602 WAIT_FOR_READ and WAIT_FOR_WRITE.
604 This is a mere convenience wrapper around the select call, and
605 should be taken as such (for example, it doesn't implement Wget's
606 0-timeout-means-no-timeout semantics.) */
609 select_fd (int fd, double maxtime, int wait_for)
612 fd_set *rd = NULL, *wr = NULL;
613 struct timeval tmout;
618 if (wait_for & WAIT_FOR_READ)
620 if (wait_for & WAIT_FOR_WRITE)
623 tmout.tv_sec = (long) maxtime;
624 tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
627 result = select (fd + 1, rd, wr, NULL, &tmout);
628 while (result < 0 && errno == EINTR);
633 /* Return true iff the connection to the remote site established
634 through SOCK is still open.
636 Specifically, this function returns true if SOCK is not ready for
637 reading. This is because, when the connection closes, the socket
638 is ready for reading because EOF is about to be delivered. A side
639 effect of this method is that sockets that have pending data are
640 considered non-open. This is actually a good thing for callers of
641 this function, where such pending data can only be unwanted
642 leftover from a previous request. */
645 test_socket_open (int sock)
650 /* Check if we still have a valid (non-EOF) connection. From Andrew
651 * Maholski's code in the Unix Socket FAQ. */
653 FD_ZERO (&check_set);
654 FD_SET (sock, &check_set);
656 /* Wait one microsecond */
660 if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
661 /* We got a timeout, it means we're still connected. */
664 /* Read now would not wait, it means we have either pending data
669 /* Basic socket operations, mostly EINTR wrappers. */
672 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
673 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
674 # define close(fd) closesocket (fd)
678 # define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
679 # define write(fd, buf, cnt) send (fd, buf, cnt, 0)
683 sock_read (int fd, char *buf, int bufsize)
687 res = read (fd, buf, bufsize);
688 while (res == -1 && errno == EINTR);
693 sock_write (int fd, char *buf, int bufsize)
697 res = write (fd, buf, bufsize);
698 while (res == -1 && errno == EINTR);
703 sock_poll (int fd, double timeout, int wait_for)
705 return select_fd (fd, timeout, wait_for);
709 sock_peek (int fd, char *buf, int bufsize)
713 res = recv (fd, buf, bufsize, MSG_PEEK);
714 while (res == -1 && errno == EINTR);
722 DEBUGP (("Closed fd %d\n", fd));
728 /* Reading and writing from the network. We build around the socket
729 (file descriptor) API, but support "extended" operations for things
730 that are not mere file descriptors under the hood, such as SSL
733 That way the user code can call fd_read(fd, ...) and we'll run read
734 or SSL_read or whatever is necessary. */
736 static struct hash_table *transport_map;
737 static unsigned int transport_map_modified_tick;
739 struct transport_info {
740 struct transport_implementation *imp;
744 /* Register the transport layer operations that will be used when
745 reading, writing, and polling FD.
747 This should be used for transport layers like SSL that piggyback on
748 sockets. FD should otherwise be a real socket, on which you can
749 call getpeername, etc. */
752 fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
754 struct transport_info *info;
756 /* The file descriptor must be non-negative to be registered.
757 Negative values are ignored by fd_close(), and -1 cannot be used as
761 info = xnew (struct transport_info);
765 transport_map = hash_table_new (0, NULL, NULL);
766 hash_table_put (transport_map, (void *)(intptr_t) fd, info);
767 ++transport_map_modified_tick;
770 /* Return context of the transport registered with
771 fd_register_transport. This assumes fd_register_transport was
772 previously called on FD. */
775 fd_transport_context (int fd)
777 struct transport_info *info = hash_table_get (transport_map, (void *)(intptr_t) fd);
781 /* When fd_read/fd_write are called multiple times in a loop, they should
782 remember the INFO pointer instead of fetching it every time. It is
783 not enough to compare FD to LAST_FD because FD might have been
784 closed and reopened. modified_tick ensures that changes to
785 transport_map will not be unnoticed.
787 This is a macro because we want the static storage variables to be
790 #define LAZY_RETRIEVE_INFO(info) do { \
791 static struct transport_info *last_info; \
792 static int last_fd = -1; \
793 static unsigned int last_tick; \
794 if (!transport_map) \
796 else if (last_fd == fd && last_tick == transport_map_modified_tick) \
800 info = hash_table_get (transport_map, (void *)(intptr_t) fd); \
803 last_tick = transport_map_modified_tick; \
808 poll_internal (int fd, struct transport_info *info, int wf, double timeout)
811 timeout = opt.read_timeout;
815 if (info && info->imp->poller)
816 test = info->imp->poller (fd, timeout, wf, info->ctx);
818 test = sock_poll (fd, timeout, wf);
827 /* Read no more than BUFSIZE bytes of data from FD, storing them to
828 BUF. If TIMEOUT is non-zero, the operation aborts if no data is
829 received after that many seconds. If TIMEOUT is -1, the value of
830 opt.timeout is used for TIMEOUT. */
833 fd_read (int fd, char *buf, int bufsize, double timeout)
835 struct transport_info *info;
836 LAZY_RETRIEVE_INFO (info);
837 if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
839 if (info && info->imp->reader)
840 return info->imp->reader (fd, buf, bufsize, info->ctx);
842 return sock_read (fd, buf, bufsize);
845 /* Like fd_read, except it provides a "preview" of the data that will
846 be read by subsequent calls to fd_read. Specifically, it copies no
847 more than BUFSIZE bytes of the currently available data to BUF and
848 returns the number of bytes copied. Return values and timeout
849 semantics are the same as those of fd_read.
851 CAVEAT: Do not assume that the first subsequent call to fd_read
852 will retrieve the same amount of data. Reading can return more or
853 less data, depending on the TCP implementation and other
854 circumstances. However, barring an error, it can be expected that
855 all the peeked data will eventually be read by fd_read. */
858 fd_peek (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->peeker)
865 return info->imp->peeker (fd, buf, bufsize, info->ctx);
867 return sock_peek (fd, buf, bufsize);
870 /* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
871 the operation aborts if no data is received after that many
872 seconds. If TIMEOUT is -1, the value of opt.timeout is used for
876 fd_write (int fd, char *buf, int bufsize, double timeout)
879 struct transport_info *info;
880 LAZY_RETRIEVE_INFO (info);
882 /* `write' may write less than LEN bytes, thus the loop keeps trying
883 it until all was written, or an error occurred. */
887 if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
889 if (info && info->imp->writer)
890 res = info->imp->writer (fd, buf, bufsize, info->ctx);
892 res = sock_write (fd, buf, bufsize);
901 /* Report the most recent error(s) on FD. This should only be called
902 after fd_* functions, such as fd_read and fd_write, and only if
903 they return a negative result. For errors coming from other calls
904 such as setsockopt or fopen, strerror should continue to be
907 If the transport doesn't support error messages or doesn't supply
908 one, strerror(errno) is returned. The returned error message
909 should not be used after fd_close has been called. */
914 /* Don't bother with LAZY_RETRIEVE_INFO, as this will only be called
915 in case of error, never in a tight loop. */
916 struct transport_info *info = NULL;
918 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
920 if (info && info->imp->errstr)
922 const char *err = info->imp->errstr (fd, info->ctx);
925 /* else, fall through and print the system error. */
927 return strerror (errno);
930 /* Close the file descriptor FD. */
935 struct transport_info *info;
939 /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
940 per socket, so that particular optimization wouldn't work. */
943 info = hash_table_get (transport_map, (void *)(intptr_t) fd);
945 if (info && info->imp->closer)
946 info->imp->closer (fd, info->ctx);
952 hash_table_remove (transport_map, (void *)(intptr_t) fd);
954 ++transport_map_modified_tick;