/* Establishing and handling network connections.
- Copyright (C) 1995, 1996, 1997, 2001, 2002 Free Software Foundation, Inc.
+ Copyright (C) 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
+ 2004, 2005, 2006, 2007 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
-(at your option) any later version.
+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,
but WITHOUT ANY WARRANTY; without even the implied warranty of
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
file, but you are not obligated to do so. If you do not wish to do
so, delete this exception statement from your version. */
-#include <config.h>
+#include "wget.h"
#include <stdio.h>
#include <stdlib.h>
-#include <sys/types.h>
#ifdef HAVE_UNISTD_H
# include <unistd.h>
#endif
#include <assert.h>
-#ifdef WINDOWS
-# include <winsock.h>
-#else
+#ifndef WINDOWS
# include <sys/socket.h>
# include <netdb.h>
# include <netinet/in.h>
-#ifndef __BEOS__
-# include <arpa/inet.h>
-#endif
-#endif /* WINDOWS */
+# ifndef __BEOS__
+# 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 "wget.h"
#include "utils.h"
#include "host.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 */
-/* Variables shared by bindport and acceptport: */
-static int msock = -1;
-static struct sockaddr *addr;
+/* 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. */
+
+static void
+sockaddr_set_data (struct sockaddr *sa, const ip_address *ip, int port)
+{
+ switch (ip->family)
+ {
+ 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 = ip->data.d4;
+ break;
+ }
+#ifdef ENABLE_IPV6
+ 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 = ip->data.d6;
+#ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
+ sin6->sin6_scope_id = ip->ipv6_scope;
+#endif
+ break;
+ }
+#endif /* ENABLE_IPV6 */
+ default:
+ abort ();
+ }
+}
-static ip_address bind_address;
-static int bind_address_resolved;
+/* Get the data of SA, specifically the IP address and the port. If
+ you're not interested in one or the other information, pass NULL as
+ the pointer. */
static void
-resolve_bind_address (void)
+sockaddr_get_data (const struct sockaddr *sa, ip_address *ip, int *port)
+{
+ switch (sa->sa_family)
+ {
+ case AF_INET:
+ {
+ struct sockaddr_in *sin = (struct sockaddr_in *)sa;
+ if (ip)
+ {
+ ip->family = AF_INET;
+ ip->data.d4 = sin->sin_addr;
+ }
+ if (port)
+ *port = ntohs (sin->sin_port);
+ break;
+ }
+#ifdef ENABLE_IPV6
+ case AF_INET6:
+ {
+ struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
+ if (ip)
+ {
+ ip->family = AF_INET6;
+ ip->data.d6 = sin6->sin6_addr;
+#ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
+ ip->ipv6_scope = sin6->sin6_scope_id;
+#endif
+ }
+ if (port)
+ *port = ntohs (sin6->sin6_port);
+ break;
+ }
+#endif
+ default:
+ abort ();
+ }
+}
+
+/* Return the size of the sockaddr structure depending on its
+ family. */
+
+static socklen_t
+sockaddr_size (const struct sockaddr *sa)
+{
+ switch (sa->sa_family)
+ {
+ case AF_INET:
+ return sizeof (struct sockaddr_in);
+#ifdef ENABLE_IPV6
+ case AF_INET6:
+ return sizeof (struct sockaddr_in6);
+#endif
+ default:
+ abort ();
+ }
+}
+\f
+/* 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;
- if (bind_address_resolved || opt.bind_address == NULL)
- /* Nothing to do. */
- return;
+ /* Make sure this is called only once. opt.bind_address doesn't
+ change during a Wget run. */
+ static bool called, should_bind;
+ static ip_address ip;
+ if (called)
+ {
+ if (should_bind)
+ sockaddr_set_data (sa, &ip, 0);
+ return should_bind;
+ }
+ called = true;
- al = lookup_host (opt.bind_address, 1);
+ al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
if (!al)
{
+ /* #### We should be able to print the error message here. */
logprintf (LOG_NOTQUIET,
- _("Unable to convert `%s' to a bind address. Reverting to ANY.\n"),
- opt.bind_address);
- return;
+ _("%s: unable to resolve bind address `%s'; disabling bind.\n"),
+ exec_name, opt.bind_address);
+ should_bind = false;
+ return false;
}
- address_list_copy_one (al, 0, &bind_address);
+ /* Pick the first address in the list and use it as bind address.
+ Perhaps we should try multiple addresses in succession, but I
+ don't think that's necessary in practice. */
+ ip = *address_list_address_at (al, 0);
address_list_release (al);
- bind_address_resolved = 1;
+
+ sockaddr_set_data (sa, &ip, 0);
+ should_bind = true;
+ return true;
}
\f
struct cwt_context {
int fd;
const struct sockaddr *addr;
- int addrlen;
+ socklen_t addrlen;
int result;
};
ETIMEDOUT. */
static int
-connect_with_timeout (int fd, const struct sockaddr *addr, int addrlen,
- double timeout)
+connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
+ double timeout)
{
struct cwt_context ctx;
ctx.fd = fd;
return ctx.result;
}
\f
-/* A kludge, but still better than passing the host name all the way
- to connect_to_one. */
-static const char *connection_host_name;
+/* Connect via TCP to the specified address and port.
-void
-set_connection_host_name (const char *host)
-{
- if (host)
- assert (connection_host_name == NULL);
- else
- assert (connection_host_name != NULL);
-
- connection_host_name = host;
-}
+ If PRINT is non-NULL, it is the host name to print that we're
+ connecting to. */
-/* Connect to a remote host whose address has been resolved. */
int
-connect_to_one (ip_address *addr, unsigned short port, int silent)
+connect_to_ip (const ip_address *ip, int port, const char *print)
{
- wget_sockaddr sa;
- int sock, save_errno;
-
- /* Set port and protocol */
- wget_sockaddr_set_address (&sa, ip_default_family, port, addr);
+ struct sockaddr_storage ss;
+ struct sockaddr *sa = (struct sockaddr *)&ss;
+ int sock;
- if (!silent)
+ /* If PRINT is non-NULL, print the "Connecting to..." line, with
+ PRINT being the host name we're connecting to. */
+ if (print)
{
- char *pretty_addr = pretty_print_address (addr);
- if (connection_host_name
- && 0 != strcmp (connection_host_name, pretty_addr))
- logprintf (LOG_VERBOSE, _("Connecting to %s[%s]:%hu... "),
- connection_host_name, pretty_addr, port);
+ const char *txt_addr = print_address (ip);
+ if (print && 0 != strcmp (print, txt_addr))
+ logprintf (LOG_VERBOSE, _("Connecting to %s|%s|:%d... "),
+ escnonprint (print), txt_addr, port);
else
- logprintf (LOG_VERBOSE, _("Connecting to %s:%hu... "),
- pretty_addr, port);
+ logprintf (LOG_VERBOSE, _("Connecting to %s:%d... "), txt_addr, port);
}
- /* Make an internet socket, stream type. */
- sock = socket (ip_default_family, SOCK_STREAM, 0);
+ /* Store the sockaddr info to SA. */
+ sockaddr_set_data (sa, ip, port);
+
+ /* Create the socket of the family appropriate for the address. */
+ sock = socket (sa->sa_family, SOCK_STREAM, 0);
if (sock < 0)
- goto out;
+ 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 size of the kernel window) to the size of the
- limit. That way we don't sleep for more than 1s between network
- reads. */
+ 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
+ network reads. */
if (opt.limit_rate && opt.limit_rate < 8192)
{
int bufsize = opt.limit_rate;
if (bufsize < 512)
- bufsize = 512;
+ bufsize = 512; /* avoid pathologically small values */
#ifdef SO_RCVBUF
setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
- (char *)&bufsize, sizeof (bufsize));
+ (void *)&bufsize, (socklen_t)sizeof (bufsize));
#endif
- /* When we add opt.limit_rate support for writing, as with
- `--post-file', also set SO_SNDBUF here. */
+ /* When we add limit_rate support for writing, which is useful
+ for POST, we should also set SO_SNDBUF here. */
}
- resolve_bind_address ();
- if (bind_address_resolved)
+ if (opt.bind_address)
{
- /* Bind the client side to the requested address. */
- wget_sockaddr bsa;
- wget_sockaddr_set_address (&bsa, ip_default_family, 0, &bind_address);
- if (bind (sock, &bsa.sa, sockaddr_len ()))
- {
- close (sock);
- sock = -1;
- goto out;
- }
+ /* Bind the client side of the socket to the requested
+ address. */
+ struct sockaddr_storage bind_ss;
+ struct sockaddr *bind_sa = (struct sockaddr *)&bind_ss;
+ if (resolve_bind_address (bind_sa))
+ {
+ if (bind (sock, bind_sa, sockaddr_size (bind_sa)) < 0)
+ goto err;
+ }
}
- /* Connect the socket to the remote host. */
- if (connect_with_timeout (sock, &sa.sa, sockaddr_len (),
- opt.connect_timeout) < 0)
- {
- close (sock);
- sock = -1;
- goto out;
- }
-
- out:
- if (sock >= 0)
- {
- /* Success. */
- if (!silent)
- logprintf (LOG_VERBOSE, _("connected.\n"));
- DEBUGP (("Created socket %d.\n", sock));
- }
- else
- {
- save_errno = errno;
- if (!silent)
- logprintf (LOG_VERBOSE, "failed: %s.\n", strerror (errno));
- errno = save_errno;
- }
+ /* Connect the socket to the remote endpoint. */
+ if (connect_with_timeout (sock, sa, sockaddr_size (sa),
+ opt.connect_timeout) < 0)
+ goto err;
+ /* Success. */
+ assert (sock >= 0);
+ if (print)
+ logprintf (LOG_VERBOSE, _("connected.\n"));
+ DEBUGP (("Created socket %d.\n", sock));
return sock;
+
+ err:
+ {
+ /* Protect errno from possible modifications by close and
+ logprintf. */
+ int save_errno = errno;
+ if (sock >= 0)
+ fd_close (sock);
+ if (print)
+ logprintf (LOG_VERBOSE, _("failed: %s.\n"), strerror (errno));
+ errno = save_errno;
+ return -1;
+ }
}
-/* Connect to a remote host whose address has been resolved. */
+/* 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_many (struct address_list *al, unsigned short port, int silent)
+connect_to_host (const char *host, int port)
{
int i, start, end;
+ int sock;
+
+ struct address_list *al = lookup_host (host, 0);
+
+ retry:
+ if (!al)
+ {
+ logprintf (LOG_NOTQUIET,
+ _("%s: unable to resolve host address `%s'\n"),
+ exec_name, host);
+ return E_HOST;
+ }
address_list_get_bounds (al, &start, &end);
for (i = start; i < end; i++)
{
- ip_address addr;
- int sock;
- address_list_copy_one (al, i, &addr);
-
- sock = connect_to_one (&addr, port, silent);
+ const ip_address *ip = address_list_address_at (al, i);
+ sock = connect_to_ip (ip, port, host);
if (sock >= 0)
- /* Success. */
- return sock;
+ {
+ /* 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_set_faulty (al, i);
+ }
- /* The attempt to connect has failed. Continue with the loop
- and try next address. */
+ /* Failed to connect to any of the addresses in AL. */
+
+ if (address_list_connected_p (al))
+ {
+ /* 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;
}
+ 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.
+
+ BIND_ADDRESS is the address of the interface to bind to. If it is
+ NULL, the socket is bound to the default address. PORT should
+ point to the port number that will be used for the binding. If
+ that number is 0, the system will choose a suitable port, and the
+ chosen value will be written to *PORT.
+
+ Calling accept() on such a socket waits for and accepts incoming
+ TCP connections. */
int
-test_socket_open (int sock)
+bind_local (const ip_address *bind_address, int *port)
{
-#ifdef HAVE_SELECT
- fd_set check_set;
- struct timeval to;
+ int sock;
+ struct sockaddr_storage ss;
+ struct sockaddr *sa = (struct sockaddr *)&ss;
- /* Check if we still have a valid (non-EOF) connection. From Andrew
- * Maholski's code in the Unix Socket FAQ. */
+ /* For setting options with setsockopt. */
+ int setopt_val = 1;
+ void *setopt_ptr = (void *)&setopt_val;
+ socklen_t setopt_size = sizeof (setopt_val);
- FD_ZERO (&check_set);
- FD_SET (sock, &check_set);
+ sock = socket (bind_address->family, SOCK_STREAM, 0);
+ if (sock < 0)
+ return -1;
- /* Wait one microsecond */
- to.tv_sec = 0;
- to.tv_usec = 1;
+#ifdef SO_REUSEADDR
+ setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
+#endif
- /* If we get a timeout, then that means still connected */
- if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
+ xzero (ss);
+ sockaddr_set_data (sa, bind_address, *port);
+ if (bind (sock, sa, sockaddr_size (sa)) < 0)
{
- /* Connection is valid (not EOF), so continue */
- return 1;
+ fd_close (sock);
+ return -1;
}
- else
- return 0;
-#else
- /* Without select, it's hard to know for sure. */
- return 1;
-#endif
+ 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 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. */
+ fd_close (sock);
+ return -1;
+ }
+ sockaddr_get_data (sa, NULL, port);
+ DEBUGP (("binding to address %s using port %i.\n",
+ print_address (bind_address), *port));
+ }
+ if (listen (sock, 1) < 0)
+ {
+ fd_close (sock);
+ return -1;
+ }
+ return sock;
}
-/* Bind the local port PORT. This does all the necessary work, which
- is creating a socket, setting SO_REUSEADDR option on it, then
- calling bind() and listen(). If *PORT is 0, a random port is
- chosen by the system, and its value is stored to *PORT. The
- internal variable MPORT is set to the value of the ensuing master
- socket. Call acceptport() to block for and accept a connection. */
-uerr_t
-bindport (unsigned short *port, int family)
-{
- int optval = 1;
- wget_sockaddr srv;
- memset (&srv, 0, sizeof (wget_sockaddr));
+/* Like a call to accept(), but with the added check for timeout.
- msock = -1;
+ In other words, accept a client connection on LOCAL_SOCK, and
+ return the new socket used for communication with the client.
+ LOCAL_SOCK should have been bound, e.g. using bind_local().
- if ((msock = socket (family, SOCK_STREAM, 0)) < 0)
- return CONSOCKERR;
+ The caller is blocked until a connection is established. If no
+ connection is established for opt.connect_timeout seconds, the
+ function exits with an error status. */
-#ifdef SO_REUSEADDR
- if (setsockopt (msock, SOL_SOCKET, SO_REUSEADDR,
- (char *)&optval, sizeof (optval)) < 0)
- return CONSOCKERR;
-#endif
+int
+accept_connection (int local_sock)
+{
+ int sock;
- resolve_bind_address ();
- wget_sockaddr_set_address (&srv, ip_default_family, htons (*port),
- bind_address_resolved ? &bind_address : NULL);
- if (bind (msock, &srv.sa, sockaddr_len ()) < 0)
- {
- CLOSE (msock);
- msock = -1;
- return BINDERR;
- }
- DEBUGP (("Master socket fd %d bound.\n", msock));
- if (!*port)
+ /* We don't need the values provided by accept, but accept
+ apparently requires them to be present. */
+ struct sockaddr_storage ss;
+ struct sockaddr *sa = (struct sockaddr *)&ss;
+ socklen_t addrlen = sizeof (ss);
+
+ if (opt.connect_timeout)
{
- /* #### addrlen should be a 32-bit type, which int is not
- guaranteed to be. Oh, and don't try to make it a size_t,
- because that can be 64-bit. */
- int sa_len = sockaddr_len ();
- if (getsockname (msock, &srv.sa, &sa_len) < 0)
- {
- CLOSE (msock);
- msock = -1;
- return CONPORTERR;
- }
- *port = wget_sockaddr_get_port (&srv);
- DEBUGP (("using port %i.\n", *port));
+ int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
+ if (test == 0)
+ errno = ETIMEDOUT;
+ if (test <= 0)
+ return -1;
}
- if (listen (msock, 1) < 0)
+ sock = accept (local_sock, sa, &addrlen);
+ DEBUGP (("Accepted client at socket %d.\n", sock));
+ return sock;
+}
+
+/* Get the IP address associated with the connection on FD and store
+ 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. */
+
+bool
+socket_ip_address (int sock, ip_address *ip, int endpoint)
+{
+ struct sockaddr_storage storage;
+ struct sockaddr *sockaddr = (struct sockaddr *)&storage;
+ socklen_t addrlen = sizeof (storage);
+ int ret;
+
+ if (endpoint == ENDPOINT_LOCAL)
+ ret = getsockname (sock, sockaddr, &addrlen);
+ else if (endpoint == ENDPOINT_PEER)
+ ret = getpeername (sock, sockaddr, &addrlen);
+ else
+ abort ();
+ if (ret < 0)
+ return false;
+
+ ip->family = sockaddr->sa_family;
+ switch (sockaddr->sa_family)
{
- CLOSE (msock);
- msock = -1;
- return LISTENERR;
+#ifdef ENABLE_IPV6
+ case AF_INET6:
+ {
+ struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)&storage;
+ ip->data.d6 = sa6->sin6_addr;
+#ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
+ ip->ipv6_scope = sa6->sin6_scope_id;
+#endif
+ DEBUGP (("conaddr is: %s\n", print_address (ip)));
+ return true;
+ }
+#endif
+ case AF_INET:
+ {
+ struct sockaddr_in *sa = (struct sockaddr_in *)&storage;
+ ip->data.d4 = sa->sin_addr;
+ DEBUGP (("conaddr is: %s\n", print_address (ip)));
+ return true;
+ }
+ default:
+ abort ();
}
- return BINDOK;
}
-#ifdef HAVE_SELECT
-/* Wait for file descriptor FD to be available, timing out after
- MAXTIME seconds. "Available" means readable if writep is 0,
- writeable otherwise.
+/* 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". */
+
+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 (false
+#ifdef EAFNOSUPPORT
+ || err == EAFNOSUPPORT
+#endif
+#ifdef EPFNOSUPPORT
+ || err == EPFNOSUPPORT
+#endif
+#ifdef ESOCKTNOSUPPORT /* no, "sockt" is not a typo! */
+ || err == ESOCKTNOSUPPORT
+#endif
+#ifdef EPROTONOSUPPORT
+ || err == EPROTONOSUPPORT
+#endif
+#ifdef ENOPROTOOPT
+ || err == ENOPROTOOPT
+#endif
+ /* Apparently, older versions of Linux and BSD used EINVAL
+ instead of EAFNOSUPPORT and such. */
+ || err == EINVAL
+ )
+ return false;
+
+ 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;
+
+ return true;
+}
+
+/* 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.
- Returns 1 if FD is available, 0 for timeout and -1 for error. */
+ This is a mere convenience wrapper around the select call, and
+ 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 writep)
+select_fd (int fd, double maxtime, int wait_for)
{
- fd_set fds;
- fd_set *rd = NULL, *wrt = NULL;
+ fd_set fdset;
+ fd_set *rd = NULL, *wr = NULL;
struct timeval tmout;
int result;
- FD_ZERO (&fds);
- FD_SET (fd, &fds);
- *(writep ? &wrt : &rd) = &fds;
+ FD_ZERO (&fdset);
+ FD_SET (fd, &fdset);
+ if (wait_for & WAIT_FOR_READ)
+ rd = &fdset;
+ if (wait_for & WAIT_FOR_WRITE)
+ wr = &fdset;
- tmout.tv_sec = (long)maxtime;
- tmout.tv_usec = 1000000L * (maxtime - (long)maxtime);
+ tmout.tv_sec = (long) maxtime;
+ tmout.tv_usec = 1000000 * (maxtime - (long) maxtime);
do
- result = select (fd + 1, rd, wrt, NULL, &tmout);
+ result = select (fd + 1, rd, wr, NULL, &tmout);
while (result < 0 && errno == EINTR);
- /* When we've timed out, set errno to ETIMEDOUT for the convenience
- of the caller. */
- if (result == 0)
- errno = ETIMEDOUT;
-
return result;
}
-#endif /* HAVE_SELECT */
-/* Call accept() on MSOCK and store the result to *SOCK. This assumes
- that bindport() has been used to initialize MSOCK to a correct
- value. It blocks the caller until a connection is established. If
- no connection is established for OPT.CONNECT_TIMEOUT seconds, the
- function exits with an error status. */
-uerr_t
-acceptport (int *sock)
+/* Return true iff the connection to the remote site established
+ through SOCK is still open.
+
+ 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. */
+
+bool
+test_socket_open (int sock)
{
- int addrlen = sockaddr_len ();
+ 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);
-#ifdef HAVE_SELECT
- if (select_fd (msock, opt.connect_timeout, 0) <= 0)
- return ACCEPTERR;
+ /* Wait one microsecond */
+ to.tv_sec = 0;
+ to.tv_usec = 1;
+
+ if (select (sock + 1, &check_set, NULL, NULL, &to) == 0)
+ /* We got a timeout, it means we're still connected. */
+ return true;
+ else
+ /* Read now would not wait, it means we have either pending data
+ or EOF/error. */
+ return false;
+}
+\f
+/* Basic socket operations, mostly EINTR wrappers. */
+
+#if defined(WINDOWS) || defined(MSDOS)
+# define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
+# define write(fd, buf, cnt) send (fd, buf, cnt, 0)
+# define close(fd) closesocket (fd)
+#endif
+
+#ifdef __BEOS__
+# define read(fd, buf, cnt) recv (fd, buf, cnt, 0)
+# define write(fd, buf, cnt) send (fd, buf, cnt, 0)
#endif
- if ((*sock = accept (msock, addr, &addrlen)) < 0)
- return ACCEPTERR;
- DEBUGP (("Created socket fd %d.\n", *sock));
- return ACCEPTOK;
+
+static int
+sock_read (int fd, char *buf, int bufsize)
+{
+ int res;
+ do
+ res = read (fd, buf, bufsize);
+ while (res == -1 && errno == EINTR);
+ return res;
}
-/* Close SOCK, as well as the most recently remembered MSOCK, created
- via bindport(). If SOCK is -1, close MSOCK only. */
-void
-closeport (int sock)
+static int
+sock_write (int fd, char *buf, int bufsize)
{
- /*shutdown (sock, 2);*/
- if (sock != -1)
- CLOSE (sock);
- if (msock != -1)
- CLOSE (msock);
- msock = -1;
+ int res;
+ do
+ res = write (fd, buf, bufsize);
+ while (res == -1 && errno == EINTR);
+ return res;
}
-/* Return the local IP address associated with the connection on FD. */
+static int
+sock_poll (int fd, double timeout, int wait_for)
+{
+ return select_fd (fd, timeout, wait_for);
+}
-int
-conaddr (int fd, ip_address *ip)
+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
+sock_close (int fd)
+{
+ close (fd);
+ DEBUGP (("Closed fd %d\n", fd));
+}
+#undef read
+#undef write
+#undef close
+\f
+/* Reading and writing from the network. We build around the socket
+ (file descriptor) API, but support "extended" operations for things
+ that are not mere file descriptors under the hood, such as SSL
+ sockets.
+
+ 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 unsigned int transport_map_modified_tick;
+
+struct transport_info {
+ struct transport_implementation *imp;
+ void *ctx;
+};
+
+/* Register the transport layer operations that will be used when
+ reading, writing, and polling FD.
+
+ This should be used for transport layers like SSL that piggyback on
+ sockets. FD should otherwise be a real socket, on which you can
+ call getpeername, etc. */
+
+void
+fd_register_transport (int fd, struct transport_implementation *imp, void *ctx)
{
- wget_sockaddr mysrv;
+ struct transport_info *info;
+
+ /* The file descriptor must be non-negative to be registered.
+ Negative values are ignored by fd_close(), and -1 cannot be used as
+ hash key. */
+ assert (fd >= 0);
+
+ info = xnew (struct transport_info);
+ info->imp = imp;
+ info->ctx = ctx;
+ if (!transport_map)
+ transport_map = hash_table_new (0, NULL, NULL);
+ hash_table_put (transport_map, (void *)(intptr_t) fd, info);
+ ++transport_map_modified_tick;
+}
- /* see bindport() for discussion of using `int' here. */
- int addrlen = sizeof (mysrv);
+/* Return context of the transport registered with
+ fd_register_transport. This assumes fd_register_transport was
+ previously called on FD. */
- if (getsockname (fd, &mysrv.sa, (int *)&addrlen) < 0)
- return 0;
+void *
+fd_transport_context (int fd)
+{
+ struct transport_info *info = hash_table_get (transport_map, (void *)(intptr_t) fd);
+ return info->ctx;
+}
- switch (mysrv.sa.sa_family)
+/* 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
+ transport_map will not be unnoticed.
+
+ This is a macro because we want the static storage variables to be
+ per-function. */
+
+#define LAZY_RETRIEVE_INFO(info) do { \
+ static struct transport_info *last_info; \
+ 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 *)(intptr_t) fd); \
+ last_fd = fd; \
+ last_info = info; \
+ last_tick = transport_map_modified_tick; \
+ } \
+} while (0)
+
+static bool
+poll_internal (int fd, struct transport_info *info, int wf, double timeout)
+{
+ if (timeout == -1)
+ timeout = opt.read_timeout;
+ if (timeout)
{
-#ifdef ENABLE_IPV6
- case AF_INET6:
- memcpy (ip, &mysrv.sin6.sin6_addr, 16);
- return 1;
-#endif
- case AF_INET:
- map_ipv4_to_ip ((ip4_address *)&mysrv.sin.sin_addr, ip);
- return 1;
- default:
- abort ();
+ int test;
+ 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 false;
}
- return 0;
+ return true;
}
-/* Read at most LEN bytes from FD, storing them to BUF. This is
- virtually the same as read(), but takes care of EINTR braindamage
- and uses select() to timeout the stale connections (a connection is
- stale if more than OPT.READ_TIMEOUT time is spent in select() or
- read()). */
+/* 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
-iread (int fd, char *buf, int len)
+fd_read (int fd, char *buf, int bufsize, double timeout)
{
- int res;
+ struct transport_info *info;
+ LAZY_RETRIEVE_INFO (info);
+ if (!poll_internal (fd, info, WAIT_FOR_READ, timeout))
+ return -1;
+ if (info && info->imp->reader)
+ return info->imp->reader (fd, buf, bufsize, info->ctx);
+ else
+ return sock_read (fd, buf, bufsize);
+}
-#ifdef HAVE_SELECT
- if (opt.read_timeout)
- if (select_fd (fd, opt.read_timeout, 0) <= 0)
- return -1;
-#endif
- do
- res = READ (fd, buf, len);
- while (res == -1 && errno == EINTR);
+/* 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.
- return res;
+ 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->imp->peeker)
+ return info->imp->peeker (fd, buf, bufsize, info->ctx);
+ else
+ return sock_peek (fd, buf, bufsize);
}
-/* Write LEN bytes from BUF to FD. This is similar to iread(), but
- unlike iread(), it makes sure that all of BUF is actually written
- to FD, so callers needn't bother with checking that the return
- value equals to LEN. Instead, you should simply check for -1. */
+/* 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
-iwrite (int fd, char *buf, int len)
+fd_write (int fd, char *buf, int bufsize, double timeout)
{
- int res = 0;
+ int res;
+ struct transport_info *info;
+ LAZY_RETRIEVE_INFO (info);
- /* `write' may write less than LEN bytes, thus the outward loop
- keeps trying it until all was written, or an error occurred. The
- inner loop is reserved for the usual EINTR f*kage, and the
- innermost loop deals with the same during select(). */
- while (len > 0)
+ /* `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)
{
-#ifdef HAVE_SELECT
- if (opt.read_timeout)
- if (select_fd (fd, opt.read_timeout, 1) <= 0)
- return -1;
-#endif
- do
- res = WRITE (fd, buf, len);
- while (res == -1 && errno == EINTR);
+ if (!poll_internal (fd, info, WAIT_FOR_WRITE, timeout))
+ return -1;
+ if (info && info->imp->writer)
+ res = info->imp->writer (fd, buf, bufsize, info->ctx);
+ else
+ res = sock_write (fd, buf, bufsize);
if (res <= 0)
- break;
+ break;
buf += res;
- len -= res;
+ bufsize -= res;
}
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
+fd_close (int fd)
+{
+ struct transport_info *info;
+ if (fd < 0)
+ return;
+
+ /* 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)
+ info = hash_table_get (transport_map, (void *)(intptr_t) fd);
+
+ if (info && info->imp->closer)
+ info->imp->closer (fd, info->ctx);
+ else
+ sock_close (fd);
+
+ if (info)
+ {
+ hash_table_remove (transport_map, (void *)(intptr_t) fd);
+ xfree (info);
+ ++transport_map_modified_tick;
+ }
+}
projects / wget / blobdiff