#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). */
+
+#ifndef ENABLE_IPV6
+# ifndef HAVE_STRUCT_SOCKADDR_STORAGE
+# define sockaddr_storage sockaddr_in
+# endif
+#endif /* ENABLE_IPV6 */
-\f
/* Fill SA as per the data in IP and PORT. SA shoult point to struct
sockaddr_storage if ENABLE_IPV6 is defined, to struct sockaddr_in
otherwise. */
case IPV4_ADDRESS:
{
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);
case IPV6_ADDRESS:
{
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);
#endif
default:
abort ();
- return 0; /* so the compiler shuts up. */
}
}
\f
return ctx.result;
}
\f
-/* Connect to a remote endpoint whose IP address is known. */
+/* Connect via TCP to the specified address and port.
+
+ If PRINT is non-NULL, it is the host name to print that we're
+ connecting to. */
int
connect_to_ip (const ip_address *ip, int port, const char *print)
{
struct sockaddr_storage ss;
struct sockaddr *sa = (struct sockaddr *)&ss;
- int sock = -1;
+ int sock;
/* If PRINT is non-NULL, print the "Connecting to..." line, with
PRINT being the host name we're connecting to. */
{
const char *txt_addr = pretty_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));
+#ifdef ENABLE_DEBUG
+ if (err < 0)
+ DEBUGP (("Failed setting IPV6_V6ONLY: %s", strerror (errno)));
+#endif
+ }
+#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
logprintf. */
int save_errno = errno;
if (sock >= 0)
- xclose (sock);
+ 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;
}
}
-/* Connect to a remote endpoint specified by host name. */
+/* Connect via TCP to a remote host on the specified port.
+
+ HOST is resolved as an Internet host name. If HOST resolves to
+ more than one IP address, they are tried in the order returned by
+ DNS until connecting to one of them succeeds. */
int
connect_to_host (const char *host, int port)
{
int i, start, end;
- struct address_list *al;
- int lh_flags = 0;
- int sock = -1;
+ int sock;
+
+ struct address_list *al = lookup_host (host, 0);
- again:
- al = lookup_host (host, lh_flags);
+ retry:
if (!al)
return E_HOST;
const ip_address *ip = address_list_address_at (al, i);
sock = connect_to_ip (ip, port, host);
if (sock >= 0)
- /* Success. */
- break;
-
- address_list_set_faulty (al, i);
+ {
+ /* Success. */
+ address_list_set_connected (al);
+ address_list_release (al);
+ return sock;
+ }
/* The attempt to connect has failed. Continue with the loop
and try next address. */
- }
- address_list_release (al);
-
- if (sock >= 0)
- /* Mark a successful connection to one of the addresses. */
- address_list_set_connected (al);
- if (sock < 0 && address_list_connected_p (al))
- {
- /* We are unable to connect to any of HOST's addresses, although
- we were previously able to connect to HOST. That might
- indicate that HOST is under dynamic DNS and the addresses
- we're connecting to have expired. Resolve it again. */
- lh_flags |= LH_REFRESH;
- goto again;
+ address_list_set_faulty (al, i);
}
- return sock;
-}
-
-int
-test_socket_open (int sock)
-{
-#ifdef HAVE_SELECT
- fd_set check_set;
- struct timeval to;
-
- /* Check if we still have a valid (non-EOF) connection. From Andrew
- * Maholski's code in the Unix Socket FAQ. */
-
- FD_ZERO (&check_set);
- FD_SET (sock, &check_set);
+ /* Failed to connect to any of the addresses in AL. */
- /* Wait one microsecond */
- 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)
+ if (address_list_connected_p (al))
{
- /* Connection is valid (not EOF), so continue */
- return 1;
+ /* We connected to AL before, but cannot do so now. That might
+ indicate that our DNS cache entry for HOST has expired. */
+ address_list_release (al);
+ al = lookup_host (host, LH_REFRESH);
+ goto retry;
}
- else
- return 0;
-#else
- /* Without select, it's hard to know for sure. */
- return 1;
-#endif
-}
+ address_list_release (al);
+ return -1;
+}
+\f
/* Create a socket, bind it to local interface BIND_ADDRESS on port
*PORT, set up a listen backlog, and return the resulting socket, or
-1 in case of error.
setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
#endif
-#ifdef ENABLE_IPV6
-# ifdef HAVE_IPV6_V6ONLY
- if (family == AF_INET6)
- setsockopt (sock, IPPROTO_IPV6, IPV6_V6ONLY, setopt_ptr, setopt_size);
-# endif
-#endif
-
xzero (ss);
sockaddr_set_data (sa, bind_address, *port);
if (bind (sock, sa, sockaddr_size (sa)) < 0)
{
- xclose (sock);
+ fd_close (sock);
return -1;
}
DEBUGP (("Local socket fd %d bound.\n", sock));
/* If *PORT is 0, find out which port we've bound to. */
if (*port == 0)
{
- socklen_t sa_len = sockaddr_size (sa);
- if (getsockname (sock, sa, &sa_len) < 0)
+ socklen_t addrlen = sockaddr_size (sa);
+ if (getsockname (sock, sa, &addrlen) < 0)
{
/* If we can't find out the socket's local address ("name"),
something is seriously wrong with the socket, and it's
unusable for us anyway because we must know the chosen
port. */
- xclose (sock);
+ fd_close (sock);
return -1;
}
sockaddr_get_data (sa, NULL, port);
}
if (listen (sock, 1) < 0)
{
- xclose (sock);
+ fd_close (sock);
return -1;
}
return sock;
struct sockaddr *sa = (struct sockaddr *)&ss;
socklen_t addrlen = sizeof (ss);
-#ifdef HAVE_SELECT
if (opt.connect_timeout)
{
int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
if (test <= 0)
return -1;
}
-#endif
sock = accept (local_sock, sa, &addrlen);
DEBUGP (("Accepted client at socket %d.\n", sock));
return sock;
default:
abort ();
}
-
- return 0;
}
/* Return non-zero if the error from the connect code can be
)
return 0;
- if (err == ECONNREFUSED && !opt.retry_connrefused)
- return 0;
+ if (!opt.retry_connrefused)
+ if (err == ECONNREFUSED
+#ifdef ENETUNREACH
+ || err == ENETUNREACH /* network is unreachable */
+#endif
+#ifdef EHOSTUNREACH
+ || err == EHOSTUNREACH /* host is unreachable */
+#endif
+ )
+ return 0;
return 1;
}
-#ifdef HAVE_SELECT
-
-/* Wait for file descriptor FD to be readable or writable or both,
- 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 WAIT_READ and WAIT_WRITE.
+/* 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
+ WAIT_FOR_READ and WAIT_FOR_WRITE.
This is a mere convenience wrapper around the select call, and
- should be taken as such. */
+ should be taken as such (for example, it doesn't implement Wget's
+ 0-timeout-means-no-timeout semantics.) */
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;
+
+#endif /* not HAVE_SELECT */
}
-#endif /* HAVE_SELECT */
+int
+test_socket_open (int sock)
+{
+#ifdef HAVE_SELECT
+ fd_set check_set;
+ struct timeval to;
+
+ /* Check if we still have a valid (non-EOF) connection. From Andrew
+ * Maholski's code in the Unix Socket FAQ. */
+
+ FD_ZERO (&check_set);
+ FD_SET (sock, &check_set);
+
+ /* Wait one microsecond */
+ 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;
+ }
+ else
+ return 0;
+#else
+ /* Without select, it's hard to know for sure. */
+ return 1;
+#endif
+}
\f
/* Basic socket operations, mostly EINTR wrappers. */
static int
sock_write (int fd, char *buf, int bufsize)
{
- int res = 0;
+ int res;
do
res = write (fd, buf, bufsize);
while (res == -1 && errno == EINTR);
static int
sock_poll (int fd, double timeout, int wait_for)
{
-#ifdef HAVE_SELECT
return select_fd (fd, timeout, wait_for);
-#else
- return 1;
-#endif
+}
+
+static int
+sock_peek (int fd, char *buf, int bufsize)
+{
+ int res;
+ do
+ res = recv (fd, buf, bufsize, MSG_PEEK);
+ while (res == -1 && errno == EINTR);
+ return res;
}
static void
that are not mere file descriptors under the hood, such as SSL
sockets.
- That way the user code can call xread(fd, ...) and we'll run read
+ That way the user code can call fd_read(fd, ...) and we'll run read
or SSL_read or whatever is necessary. */
static struct hash_table *transport_map;
static int transport_map_modified_tick;
struct transport_info {
- xreader_t reader;
- xwriter_t writer;
- xpoller_t poller;
- xcloser_t closer;
+ fd_reader_t reader;
+ fd_writer_t writer;
+ fd_poller_t poller;
+ fd_peeker_t peeker;
+ fd_closer_t closer;
void *ctx;
};
call getpeername, etc. */
void
-register_transport (int fd, xreader_t reader, xwriter_t writer,
- xpoller_t poller, xcloser_t closer, void *ctx)
+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)
{
struct transport_info *info;
/* The file descriptor must be non-negative to be registered.
- Negative values are ignored by xclose(), and -1 cannot be used as
+ Negative values are ignored by fd_close(), and -1 cannot be used as
hash key. */
assert (fd >= 0);
info->reader = reader;
info->writer = writer;
info->poller = poller;
+ info->peeker = peeker;
info->closer = closer;
info->ctx = ctx;
if (!transport_map)
++transport_map_modified_tick;
}
-/* When xread/xwrite are called multiple times in a loop, they should
+/* 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 *) 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
closed and reopened. modified_tick ensures that changes to
} \
} while (0)
-/* Read no more than BUFSIZE bytes of data from FD, storing them to
- BUF. If TIMEOUT is non-zero, the operation aborts if no data is
- received after that many seconds. If TIMEOUT is -1, the value of
- opt.timeout is used for TIMEOUT. */
-
-int
-xread (int fd, char *buf, int bufsize, double timeout)
+static int
+poll_internal (int fd, struct transport_info *info, int wf, double timeout)
{
- struct transport_info *info;
- LAZY_RETRIEVE_INFO (info);
if (timeout == -1)
timeout = opt.read_timeout;
if (timeout)
{
int test;
if (info && info->poller)
- test = info->poller (fd, timeout, WAIT_FOR_READ, info->ctx);
+ test = info->poller (fd, timeout, wf, info->ctx);
else
- test = sock_poll (fd, timeout, WAIT_FOR_READ);
+ test = sock_poll (fd, timeout, wf);
if (test == 0)
errno = ETIMEDOUT;
if (test <= 0)
- return -1;
+ return 0;
}
+ return 1;
+}
+
+/* Read no more than BUFSIZE bytes of data from FD, storing them to
+ BUF. If TIMEOUT is non-zero, the operation aborts if no data is
+ received after that many seconds. If TIMEOUT is -1, the value of
+ opt.timeout is used for TIMEOUT. */
+
+int
+fd_read (int fd, char *buf, int bufsize, double timeout)
+{
+ struct transport_info *info;
+ 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);
else
return sock_read (fd, buf, bufsize);
}
+/* 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)
+{
+ struct transport_info *info;
+ 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);
+ else
+ return sock_peek (fd, buf, bufsize);
+}
+
/* Write the entire contents of BUF to FD. If TIMEOUT is non-zero,
the operation aborts if no data is received after that many
seconds. If TIMEOUT is -1, the value of opt.timeout is used for
TIMEOUT. */
int
-xwrite (int fd, char *buf, int bufsize, double timeout)
+fd_write (int fd, char *buf, int bufsize, double timeout)
{
int res;
struct transport_info *info;
LAZY_RETRIEVE_INFO (info);
- if (timeout == -1)
- timeout = opt.read_timeout;
/* `write' may write less than LEN bytes, thus the loop keeps trying
it until all was written, or an error occurred. */
res = 0;
while (bufsize > 0)
{
- if (timeout)
- {
- int test;
- if (info && info->poller)
- test = info->poller (fd, timeout, WAIT_FOR_WRITE, info->ctx);
- else
- test = sock_poll (fd, timeout, WAIT_FOR_WRITE);
- if (test == 0)
- errno = ETIMEDOUT;
- if (test <= 0)
- return -1;
- }
+ if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
+ return -1;
if (info && info->writer)
res = info->writer (fd, buf, bufsize, info->ctx);
else
/* Close the file descriptor FD. */
void
-xclose (int fd)
+fd_close (int fd)
{
struct transport_info *info;
if (fd < 0)
return;
- /* Don't use LAZY_RETRIEVE_INFO because xclose() is only called once
+ /* Don't use LAZY_RETRIEVE_INFO because fd_close() is only called once
per socket, so that particular optimization wouldn't work. */
info = NULL;
if (transport_map)