X-Git-Url: http://sjero.net/git/?a=blobdiff_plain;f=src%2Fconnect.c;h=eeb4eb288f9a9d7b5782d8a93b14c28a0486d445;hb=d9fea91a0a319e348adb504bd3edff148ff3d8a0;hp=90729aec262970d41724bcc45683d01c66148d57;hpb=02f90322e8fae14dedd7865394bbd3e499ef506d;p=wget

diff --git a/src/connect.c b/src/connect.c
index 90729aec..eeb4eb28 100644
--- a/src/connect.c
+++ b/src/connect.c
@@ -182,7 +182,7 @@ resolve_bind_address (struct sockaddr *sa)
     }
   called = 1;
 
-  al = lookup_host_passive (opt.bind_address);
+  al = lookup_host (opt.bind_address, LH_BIND | LH_SILENT);
   if (!al)
     {
       /* #### We should be able to print the error message here. */
@@ -241,7 +241,10 @@ connect_with_timeout (int fd, const struct sockaddr *addr, socklen_t addrlen,
   return ctx.result;
 }
 
-/* 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)
@@ -318,7 +321,7 @@ connect_to_ip (const ip_address *ip, int port, const char *print)
        logprintf.  */
     int save_errno = errno;
     if (sock >= 0)
-      xclose (sock);
+      fd_close (sock);
     if (print)
       logprintf (LOG_VERBOSE, "failed: %s.\n", strerror (errno));
     errno = save_errno;
@@ -326,17 +329,21 @@ connect_to_ip (const ip_address *ip, int port, const char *print)
   }
 }
 
-/* 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 sock = -1;
+  int sock;
+
+  struct address_list *al = lookup_host (host, 0);
 
- again:
-  al = lookup_host (host, 0);
+ retry:
   if (!al)
     return E_HOST;
 
@@ -346,70 +353,49 @@ connect_to_host (const char *host, int port)
       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_set_faulty (al, i);
     }
-  address_list_release (al);
 
-  if (sock >= 0)
-    /* Mark a successful connection to one of the addresses. */
-    address_list_set_connected (al);
+  /* Failed to connect to any of the addresses in AL. */
 
-  if (sock < 0 && address_list_connected_p (al))
+  if (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.  */
-      forget_host_lookup (host);
-      goto again;
+      /* 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 sock;
+  return -1;
 }
+
+/* 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.
 
-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
-}
+   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.
 
-/* Create a socket and bind it to PORT locally.  Calling accept() on
-   such a socket waits for and accepts incoming TCP connections.  The
-   resulting socket is stored to LOCAL_SOCK.  */
+   Calling accept() on such a socket waits for and accepts incoming
+   TCP connections.  */
 
-uerr_t
-bindport (const ip_address *bind_address, int *port, int *local_sock)
+int
+bind_local (const ip_address *bind_address, int *port)
 {
   int sock;
   int family = AF_INET;
@@ -426,8 +412,9 @@ bindport (const ip_address *bind_address, int *port, int *local_sock)
     family = AF_INET6;
 #endif
 
-  if ((sock = socket (family, SOCK_STREAM, 0)) < 0)
-    return CONSOCKERR;
+  sock = socket (family, SOCK_STREAM, 0);
+  if (sock < 0)
+    return -1;
 
 #ifdef SO_REUSEADDR
   setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, setopt_ptr, setopt_size);
@@ -444,17 +431,23 @@ bindport (const ip_address *bind_address, int *port, int *local_sock)
   sockaddr_set_data (sa, bind_address, *port);
   if (bind (sock, sa, sockaddr_size (sa)) < 0)
     {
-      xclose (sock);
-      return BINDERR;
+      fd_close (sock);
+      return -1;
     }
   DEBUGP (("Local socket fd %d bound.\n", sock));
-  if (!*port)
+
+  /* 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)
 	{
-	  xclose (sock);
-	  return CONPORTERR;
+	  /* 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", 
@@ -462,69 +455,44 @@ bindport (const ip_address *bind_address, int *port, int *local_sock)
     }
   if (listen (sock, 1) < 0)
     {
-      xclose (sock);
-      return LISTENERR;
+      fd_close (sock);
+      return -1;
     }
-  *local_sock = sock;
-  return BINDOK;
+  return sock;
 }
 
-#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.
-
-   This is a mere convenience wrapper around the select call, and
-   should be taken as such.  */
-
-int
-select_fd (int fd, double maxtime, int wait_for)
-{
-  fd_set fdset;
-  fd_set *rd = NULL, *wr = NULL;
-  struct timeval tmout;
-  int result;
-
-  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);
-
-  do
-    result = select (fd + 1, rd, wr, NULL, &tmout);
-  while (result < 0 && errno == EINTR);
+/* Like a call to accept(), but with the added check for timeout.
 
-  return result;
-}
-#endif /* HAVE_SELECT */
+   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().
 
-/* Accept a connection on LOCAL_SOCK, and store the new socket to
-   *SOCK.  It blocks the caller until a connection is established.  If
-   no connection is established for opt.connect_timeout seconds, the
+   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.  */
 
-uerr_t
-acceptport (int local_sock, int *sock)
+int
+accept_connection (int local_sock)
 {
+  int sock;
+
+  /* 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);
 
-#ifdef HAVE_SELECT
   if (opt.connect_timeout)
-    if (select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ) <= 0)
-      return ACCEPTERR;
-#endif
-  if ((*sock = accept (local_sock, sa, &addrlen)) < 0)
-    return ACCEPTERR;
-  DEBUGP (("Created socket fd %d.\n", *sock));
-  return ACCEPTOK;
+    {
+      int test = select_fd (local_sock, opt.connect_timeout, WAIT_FOR_READ);
+      if (test == 0)
+	errno = ETIMEDOUT;
+      if (test <= 0)
+	return -1;
+    }
+  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
@@ -580,6 +548,153 @@ socket_ip_address (int sock, ip_address *ip, int endpoint)
 
   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".  */
+
+int
+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
+#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 0;
+
+  if (err == ECONNREFUSED && !opt.retry_connrefused)
+    return 0;
+
+  return 1;
+}
+
+#ifdef ENABLE_IPV6
+# ifndef HAVE_GETADDRINFO_AI_ADDRCONFIG
+
+/* 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.)  */
+
+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;
+}
+
+# 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
+   WAIT_FOR_READ and WAIT_FOR_WRITE.
+
+   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 wait_for)
+{
+#ifdef HAVE_SELECT
+  fd_set fdset;
+  fd_set *rd = NULL, *wr = NULL;
+  struct timeval tmout;
+  int result;
+
+  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);
+
+  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 */
+}
+
+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
+}
 
 /* Basic socket operations, mostly EINTR wrappers.  */
 
@@ -607,7 +722,7 @@ sock_read (int fd, char *buf, int bufsize)
 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);
@@ -617,11 +732,17 @@ sock_write (int fd, char *buf, int bufsize)
 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
@@ -639,133 +760,151 @@ sock_close (int fd)
    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 *extended_map;
-static int extended_map_modified_tick;
+static struct hash_table *transport_map;
+static int transport_map_modified_tick;
 
-struct extended_info {
-  xreader_t reader;
-  xwriter_t writer;
-  xpoller_t poller;
-  xcloser_t closer;
+struct transport_info {
+  fd_reader_t reader;
+  fd_writer_t writer;
+  fd_poller_t poller;
+  fd_peeker_t peeker;
+  fd_closer_t closer;
   void *ctx;
 };
 
-/* Register the handlers for operations on FD.  This is meant
-   primarily for transport layers like SSL that piggyback on sockets,
-   but with their own readers, writers, etc.  */
+/* 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
-register_extended (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 extended_info *info;
+  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 = xnew (struct extended_info);
+  info = xnew (struct transport_info);
   info->reader = reader;
   info->writer = writer;
   info->poller = poller;
+  info->peeker = peeker;
   info->closer = closer;
   info->ctx = ctx;
-  if (!extended_map)
-    extended_map = hash_table_new (0, NULL, NULL);
-  hash_table_put (extended_map, (void *) fd, info);
-  ++extended_map_modified_tick;
+  if (!transport_map)
+    transport_map = hash_table_new (0, NULL, NULL);
+  hash_table_put (transport_map, (void *) fd, info);
+  ++transport_map_modified_tick;
 }
 
-/* When xread/xwrite are called multiple times in a loop, they should
+/* 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
-   extended_map will not be unnoticed.
+   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 extended_info *last_info;				\
+  static struct transport_info *last_info;				\
   static int last_fd = -1, last_tick;					\
-  if (!extended_map)							\
+  if (!transport_map)							\
     info = NULL;							\
-  else if (last_fd == fd && last_tick == extended_map_modified_tick)	\
+  else if (last_fd == fd && last_tick == transport_map_modified_tick)	\
     info = last_info;							\
   else									\
     {									\
-      info = hash_table_get (extended_map, (void *) fd);		\
+      info = hash_table_get (transport_map, (void *) fd);		\
       last_fd = fd;							\
       last_info = info;							\
-      last_tick = extended_map_modified_tick;				\
+      last_tick = transport_map_modified_tick;				\
     }									\
 } 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 extended_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);
 }
 
+/* The same as xread, but don't actually read the data, just copy it
+   instead.  */
+
+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 extended_info *info;
+  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
@@ -781,17 +920,17 @@ xwrite (int fd, char *buf, int bufsize, double timeout)
 /* Close the file descriptor FD.  */
 
 void
-xclose (int fd)
+fd_close (int fd)
 {
-  struct extended_info *info;
+  struct transport_info *info;
   if (fd < 0)
     return;
 
-  /* We don't need to be extra-fast here, so save some code by
-     avoiding LAZY_RETRIEVE_INFO. */
+  /* 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 (extended_map)
-    info = hash_table_get (extended_map, (void *) fd);
+  if (transport_map)
+    info = hash_table_get (transport_map, (void *) fd);
 
   if (info && info->closer)
     info->closer (fd, info->ctx);
@@ -800,8 +939,8 @@ xclose (int fd)
 
   if (info)
     {
-      hash_table_remove (extended_map, (void *) fd);
+      hash_table_remove (transport_map, (void *) fd);
       xfree (info);
-      ++extended_map_modified_tick;
+      ++transport_map_modified_tick;
     }
 }