/* Establishing and handling network connections.
- Copyright (C) 1995, 1996, 1997, 2001, 2002 Free Software Foundation, Inc.
+ Copyright (C) 1996-2006 Free Software Foundation, Inc.
This file is part of GNU Wget.
GNU Wget is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
+the Free Software Foundation; either version 3 of the License, or
(at your option) any later version.
GNU Wget is distributed in the hope that it will be useful,
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
-along with Wget; if not, write to the Free Software
-Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+along with Wget. If not, see <http://www.gnu.org/licenses/>.
In addition, as a special exception, the Free Software Foundation
gives permission to link the code of its release of Wget with the
#include <stdio.h>
#include <stdlib.h>
-#include <sys/types.h>
#ifdef HAVE_UNISTD_H
# include <unistd.h>
#endif
#endif /* not WINDOWS */
#include <errno.h>
-#ifdef HAVE_STRING_H
-# include <string.h>
-#else
-# include <strings.h>
-#endif /* HAVE_STRING_H */
+#include <string.h>
#ifdef HAVE_SYS_SELECT_H
# include <sys/select.h>
#endif /* HAVE_SYS_SELECT_H */
#include "connect.h"
#include "hash.h"
-#ifndef errno
-extern int errno;
-#endif
-
/* Define sockaddr_storage where unavailable (presumably on IPv4-only
hosts). */
static void
sockaddr_set_data (struct sockaddr *sa, const ip_address *ip, int port)
{
- switch (ip->type)
+ switch (ip->family)
{
- case IPV4_ADDRESS:
+ case AF_INET:
{
struct sockaddr_in *sin = (struct sockaddr_in *)sa;
xzero (*sin);
sin->sin_family = AF_INET;
sin->sin_port = htons (port);
- sin->sin_addr = ADDRESS_IPV4_IN_ADDR (ip);
+ sin->sin_addr = ip->data.d4;
break;
}
#ifdef ENABLE_IPV6
- case IPV6_ADDRESS:
+ case AF_INET6:
{
struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
xzero (*sin6);
sin6->sin6_family = AF_INET6;
sin6->sin6_port = htons (port);
- sin6->sin6_addr = ADDRESS_IPV6_IN6_ADDR (ip);
+ sin6->sin6_addr = ip->data.d6;
#ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
- sin6->sin6_scope_id = ADDRESS_IPV6_SCOPE (ip);
+ sin6->sin6_scope_id = ip->ipv6_scope;
#endif
break;
}
struct sockaddr_in *sin = (struct sockaddr_in *)sa;
if (ip)
{
- ip->type = IPV4_ADDRESS;
- ADDRESS_IPV4_IN_ADDR (ip) = sin->sin_addr;
+ ip->family = AF_INET;
+ ip->data.d4 = sin->sin_addr;
}
if (port)
*port = ntohs (sin->sin_port);
struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
if (ip)
{
- ip->type = IPV6_ADDRESS;
- ADDRESS_IPV6_IN6_ADDR (ip) = sin6->sin6_addr;
+ ip->family = AF_INET6;
+ ip->data.d6 = sin6->sin6_addr;
#ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
- ADDRESS_IPV6_SCOPE (ip) = sin6->sin6_scope_id;
+ ip->ipv6_scope = sin6->sin6_scope_id;
#endif
}
if (port)
#endif
default:
abort ();
- return 0; /* so the compiler shuts up. */
}
}
\f
-static int
+/* Resolve the bind address specified via --bind-address and store it
+ to SA. The resolved value is stored in a static variable and
+ reused after the first invocation of this function.
+
+ Returns true on success, false on failure. */
+
+static bool
resolve_bind_address (struct sockaddr *sa)
{
struct address_list *al;
/* Make sure this is called only once. opt.bind_address doesn't
change during a Wget run. */
- static int called, should_bind;
+ static bool called, should_bind;
static ip_address ip;
if (called)
{
sockaddr_set_data (sa, &ip, 0);
return should_bind;
}
- called = 1;
+ called = true;
al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
if (!al)
logprintf (LOG_NOTQUIET,
_("%s: unable to resolve bind address `%s'; disabling bind.\n"),
exec_name, opt.bind_address);
- should_bind = 0;
- return 0;
+ should_bind = false;
+ return false;
}
/* Pick the first address in the list and use it as bind address.
address_list_release (al);
sockaddr_set_data (sa, &ip, 0);
- should_bind = 1;
- return 1;
+ should_bind = true;
+ return true;
}
\f
struct cwt_context {
PRINT being the host name we're connecting to. */
if (print)
{
- const char *txt_addr = pretty_print_address (ip);
+ const char *txt_addr = print_address (ip);
if (print && 0 != strcmp (print, txt_addr))
- logprintf (LOG_VERBOSE,
- _("Connecting to %s|%s|:%d... "), print, txt_addr, port);
+ logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
+ escnonprint (print), txt_addr, port);
else
logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
}
if (sock < 0)
goto err;
+#if defined(ENABLE_IPV6) && defined(IPV6_V6ONLY)
+ if (opt.ipv6_only) {
+ int on = 1;
+ /* In case of error, we will go on anyway... */
+ int err = setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof (on));
+ IF_DEBUG
+ if (err < 0)
+ DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
+ }
+#endif
+
/* For very small rate limits, set the buffer size (and hence,
hopefully, the kernel's TCP window size) to the per-second limit.
That way we should never have to sleep for more than 1s between
if (sock >= 0)
fd_close (sock);
if (print)
- logprintf (LOG_VERBOSE, "failed: %s.\n", strerror (errno));
+ logprintf (LOG_VERBOSE, _("failed: %s.\n"), strerror (errno));
errno = save_errno;
return -1;
}
bind_local (const ip_address *bind_address, int *port)
{
int sock;
- int family = AF_INET;
struct sockaddr_storage ss;
struct sockaddr *sa = (struct sockaddr *)&ss;
void *setopt_ptr = (void *)&setopt_val;
socklen_t setopt_size = sizeof (setopt_val);
-#ifdef ENABLE_IPV6
- if (bind_address->type == IPV6_ADDRESS)
- family = AF_INET6;
-#endif
-
- sock = socket (family, SOCK_STREAM, 0);
+ sock = socket (bind_address->family, SOCK_STREAM, 0);
if (sock < 0)
return -1;
return -1;
}
sockaddr_get_data (sa, NULL, port);
- DEBUGP (("binding to address %s using port %i.\n",
- pretty_print_address (bind_address), *port));
+ DEBUGP (("binding to address %s using port %i.\n",
+ print_address (bind_address), *port));
}
if (listen (sock, 1) < 0)
{
}
/* Get the IP address associated with the connection on FD and store
- it to IP. Return 1 on success, 0 otherwise.
+ it to IP. Return true on success, false otherwise.
If ENDPOINT is ENDPOINT_LOCAL, it returns the address of the local
(client) side of the socket. Else if ENDPOINT is ENDPOINT_PEER, it
returns the address of the remote (peer's) side of the socket. */
-int
+bool
socket_ip_address (int sock, ip_address *ip, int endpoint)
{
struct sockaddr_storage storage;
else
abort ();
if (ret < 0)
- return 0;
+ return false;
+ ip->family = sockaddr->sa_family;
switch (sockaddr->sa_family)
{
#ifdef ENABLE_IPV6
case AF_INET6:
{
struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
- ip->type = IPV6_ADDRESS;
- ADDRESS_IPV6_IN6_ADDR (ip) = sa6->sin6_addr;
+ ip->data.d6 = sa6->sin6_addr;
#ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
- ADDRESS_IPV6_SCOPE (ip) = sa6->sin6_scope_id;
+ ip->ipv6_scope = sa6->sin6_scope_id;
#endif
- DEBUGP (("conaddr is: %s\n", pretty_print_address (ip)));
- return 1;
+ DEBUGP (("conaddr is: %s\n", print_address (ip)));
+ return true;
}
#endif
case AF_INET:
{
struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
- ip->type = IPV4_ADDRESS;
- ADDRESS_IPV4_IN_ADDR (ip) = sa->sin_addr;
- DEBUGP (("conaddr is: %s\n", pretty_print_address (ip)));
- return 1;
+ ip->data.d4 = sa->sin_addr;
+ DEBUGP (("conaddr is: %s\n", print_address (ip)));
+ return true;
}
default:
abort ();
}
-
- return 0;
}
-/* Return non-zero if the error from the connect code can be
- considered retryable. Wget normally retries after errors, but the
- exception are the "unsupported protocol" type errors (possible on
- IPv4/IPv6 dual family systems) and "connection refused". */
+/* Return true if the error from the connect code can be considered
+ retryable. Wget normally retries after errors, but the exception
+ are the "unsupported protocol" type errors (possible on IPv4/IPv6
+ dual family systems) and "connection refused". */
-int
+bool
retryable_socket_connect_error (int err)
{
/* Have to guard against some of these values not being defined.
Cannot use a switch statement because some of the values might be
equal. */
- if (0
+ if (false
#ifdef EAFNOSUPPORT
|| err == EAFNOSUPPORT
#endif
instead of EAFNOSUPPORT and such. */
|| err == EINVAL
)
- return 0;
-
- if (err == ECONNREFUSED && !opt.retry_connrefused)
- return 0;
-
- return 1;
-}
-
-#ifdef ENABLE_IPV6
-# ifndef HAVE_GETADDRINFO_AI_ADDRCONFIG
+ return false;
-/* Return non-zero if the INET6 socket family is supported on the
- system.
-
- This doesn't guarantee that we're able to connect to IPv6 hosts,
- but it's better than nothing. It is only used on systems where
- getaddrinfo doesn't support AI_ADDRCONFIG. (See lookup_host.) */
+ if (!opt.retry_connrefused)
+ if (err == ECONNREFUSED
+#ifdef ENETUNREACH
+ || err == ENETUNREACH /* network is unreachable */
+#endif
+#ifdef EHOSTUNREACH
+ || err == EHOSTUNREACH /* host is unreachable */
+#endif
+ )
+ return false;
-int
-socket_has_inet6 (void)
-{
- static int supported = -1;
- if (supported == -1)
- {
- int sock = socket (AF_INET6, SOCK_STREAM, 0);
- if (sock < 0)
- supported = 0;
- else
- {
- fd_close (sock);
- supported = 1;
- }
- }
- return supported;
+ return true;
}
-# endif/* not HAVE_GETADDRINFO_AI_ADDRCONFIG */
-#endif /* ENABLE_IPV6 */
-
/* Wait for a single descriptor to become available, timing out after
MAXTIME seconds. Returns 1 if FD is available, 0 for timeout and
-1 for error. The argument WAIT_FOR can be a combination of
int
select_fd (int fd, double maxtime, int wait_for)
{
-#ifdef HAVE_SELECT
fd_set fdset;
fd_set *rd = NULL, *wr = NULL;
struct timeval tmout;
wr = &fdset;
tmout.tv_sec = (long) maxtime;
- tmout.tv_usec = 1000000L * (maxtime - (long) maxtime);
+ tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
do
result = select (fd + 1, rd, wr, NULL, &tmout);
while (result < 0 && errno == EINTR);
return result;
+}
-#else /* not HAVE_SELECT */
-
- /* If select() unavailable, just return 1. In most usages in Wget,
- this is the appropriate response -- "if we can't poll, go ahead
- with the blocking operation". If a specific part of code needs
- different behavior, it can use #ifdef HAVE_SELECT to test whether
- polling really occurs. */
- return 1;
+/* Return true iff the connection to the remote site established
+ through SOCK is still open.
-#endif /* not HAVE_SELECT */
-}
+ Specifically, this function returns true if SOCK is not ready for
+ reading. This is because, when the connection closes, the socket
+ is ready for reading because EOF is about to be delivered. A side
+ effect of this method is that sockets that have pending data are
+ considered non-open. This is actually a good thing for callers of
+ this function, where such pending data can only be unwanted
+ leftover from a previous request. */
-int
+bool
test_socket_open (int sock)
{
-#ifdef HAVE_SELECT
fd_set check_set;
struct timeval to;
to.tv_sec = 0;
to.tv_usec = 1;
- /* If we get a timeout, then that means still connected */
if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
- {
- /* Connection is valid (not EOF), so continue */
- return 1;
- }
+ /* We got a timeout, it means we're still connected. */
+ return true;
else
- return 0;
-#else
- /* Without select, it's hard to know for sure. */
- return 1;
-#endif
+ /* Read now would not wait, it means we have either pending data
+ or EOF/error. */
+ return false;
}
\f
/* Basic socket operations, mostly EINTR wrappers. */
or SSL_read or whatever is necessary. */
static struct hash_table *transport_map;
-static int transport_map_modified_tick;
+static unsigned int transport_map_modified_tick;
struct transport_info {
- fd_reader_t reader;
- fd_writer_t writer;
- fd_poller_t poller;
- fd_peeker_t peeker;
- fd_closer_t closer;
+ struct transport_implementation *imp;
void *ctx;
};
call getpeername, etc. */
void
-fd_register_transport (int fd, fd_reader_t reader, fd_writer_t writer,
- fd_poller_t poller, fd_peeker_t peeker,
- fd_closer_t closer, void *ctx)
+fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
{
struct transport_info *info;
assert (fd >= 0);
info = xnew (struct transport_info);
- info->reader = reader;
- info->writer = writer;
- info->poller = poller;
- info->peeker = peeker;
- info->closer = closer;
+ info->imp = imp;
info->ctx = ctx;
if (!transport_map)
transport_map = hash_table_new (0, NULL, NULL);
- hash_table_put (transport_map, (void *) fd, info);
+ hash_table_put (transport_map, (void *)(intptr_t) fd, info);
++transport_map_modified_tick;
}
+/* Return context of the transport registered with
+ fd_register_transport. This assumes fd_register_transport was
+ previously called on FD. */
+
+void *
+fd_transport_context (int fd)
+{
+ struct transport_info *info = hash_table_get (transport_map, (void *)(intptr_t) fd);
+ return info->ctx;
+}
+
/* When fd_read/fd_write are called multiple times in a loop, they should
remember the INFO pointer instead of fetching it every time. It is
not enough to compare FD to LAST_FD because FD might have been
#define LAZY_RETRIEVE_INFO(info) do { \
static struct transport_info *last_info; \
- static int last_fd = -1, last_tick; \
+ static int last_fd = -1; \
+ static unsigned int last_tick; \
if (!transport_map) \
info = NULL; \
else if (last_fd == fd && last_tick == transport_map_modified_tick) \
info = last_info; \
else \
{ \
- info = hash_table_get (transport_map, (void *) fd); \
+ info = hash_table_get (transport_map, (void *)(intptr_t) fd); \
last_fd = fd; \
last_info = info; \
last_tick = transport_map_modified_tick; \
} \
} while (0)
-static int
+static bool
poll_internal (int fd, struct transport_info *info, int wf, double timeout)
{
if (timeout == -1)
if (timeout)
{
int test;
- if (info && info->poller)
- test = info->poller (fd, timeout, wf, info->ctx);
+ if (info && info->imp->poller)
+ test = info->imp->poller (fd, timeout, wf, info->ctx);
else
test = sock_poll (fd, timeout, wf);
if (test == 0)
errno = ETIMEDOUT;
if (test <= 0)
- return 0;
+ return false;
}
- return 1;
+ return true;
}
/* Read no more than BUFSIZE bytes of data from FD, storing them to
LAZY_RETRIEVE_INFO (info);
if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
return -1;
- if (info && info->reader)
- return info->reader (fd, buf, bufsize, info->ctx);
+ if (info && info->imp->reader)
+ return info->imp->reader (fd, buf, bufsize, info->ctx);
else
return sock_read (fd, buf, bufsize);
}
-/* The same as fd_read, but don't actually read the data, just find out
- what's the number of bytes available for reading. */
+/* Like fd_read, except it provides a "preview" of the data that will
+ be read by subsequent calls to fd_read. Specifically, it copies no
+ more than BUFSIZE bytes of the currently available data to BUF and
+ returns the number of bytes copied. Return values and timeout
+ semantics are the same as those of fd_read.
+
+ CAVEAT: Do not assume that the first subsequent call to fd_read
+ will retrieve the same amount of data. Reading can return more or
+ less data, depending on the TCP implementation and other
+ circumstances. However, barring an error, it can be expected that
+ all the peeked data will eventually be read by fd_read. */
int
fd_peek (int fd, char *buf, int bufsize, double timeout)
LAZY_RETRIEVE_INFO (info);
if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
return -1;
- if (info && info->peeker)
- return info->peeker (fd, buf, bufsize, info->ctx);
+ if (info && info->imp->peeker)
+ return info->imp->peeker (fd, buf, bufsize, info->ctx);
else
return sock_peek (fd, buf, bufsize);
}
{
if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
return -1;
- if (info && info->writer)
- res = info->writer (fd, buf, bufsize, info->ctx);
+ if (info && info->imp->writer)
+ res = info->imp->writer (fd, buf, bufsize, info->ctx);
else
res = sock_write (fd, buf, bufsize);
if (res <= 0)
return res;
}
+/* Report the most recent error(s) on FD. This should only be called
+ after fd_* functions, such as fd_read and fd_write, and only if
+ they return a negative result. For errors coming from other calls
+ such as setsockopt or fopen, strerror should continue to be
+ used.
+
+ If the transport doesn't support error messages or doesn't supply
+ one, strerror(errno) is returned. The returned error message
+ should not be used after fd_close has been called. */
+
+const char *
+fd_errstr (int fd)
+{
+ /* Don't bother with LAZY_RETRIEVE_INFO, as this will only be called
+ in case of error, never in a tight loop. */
+ struct transport_info *info = NULL;
+ if (transport_map)
+ info = hash_table_get (transport_map, (void *)(intptr_t) fd);
+
+ if (info && info->imp->errstr)
+ {
+ const char *err = info->imp->errstr (fd, info->ctx);
+ if (err)
+ return err;
+ /* else, fall through and print the system error. */
+ }
+ return strerror (errno);
+}
+
/* Close the file descriptor FD. */
void
per socket, so that particular optimization wouldn't work. */
info = NULL;
if (transport_map)
- info = hash_table_get (transport_map, (void *) fd);
+ info = hash_table_get (transport_map, (void *)(intptr_t) fd);
- if (info && info->closer)
- info->closer (fd, info->ctx);
+ if (info && info->imp->closer)
+ info->imp->closer (fd, info->ctx);
else
sock_close (fd);
if (info)
{
- hash_table_remove (transport_map, (void *) fd);
+ hash_table_remove (transport_map, (void *)(intptr_t) fd);
xfree (info);
++transport_map_modified_tick;
}