diff --git a/configure.in b/configure.in
index 8f9e7d0..67f5a54 100644
--- a/configure.in
+++ b/configure.in
@@ -1144,7 +1144,7 @@ AC_CHECK_HEADERS(limits.h sys/file.h sys/ioctl.h sys/syscall.h\
 		 syscall.h pwd.h grp.h a.out.h utime.h direct.h sys/resource.h \
 		 sys/mkdev.h sys/utime.h xti.h netinet/in_systm.h float.h ieeefp.h \
 		 ucontext.h intrinsics.h langinfo.h locale.h sys/sendfile.h time.h \
-		 net/socket.h sys/socket.h process.h sys/prctl.h)
+		 net/socket.h sys/socket.h process.h sys/prctl.h sys/uio.h)
 
 AC_TYPE_SIZE_T
 RUBY_CHECK_SIZEOF(size_t, [int long void*], [], [@%:@include <sys/types.h>])
@@ -1424,7 +1424,7 @@ AC_CHECK_FUNCS(fmod killpg wait4 waitpid fork spawnv syscall __syscall chroot ge
 	      setuid setgid daemon select_large_fdset setenv unsetenv\
               mktime timegm gmtime_r clock_gettime gettimeofday poll ppoll\
               pread sendfile shutdown sigaltstack dl_iterate_phdr\
-              dup3 pipe2 posix_memalign memalign)
+              dup3 pipe2 posix_memalign memalign writev)
 
 AC_CACHE_CHECK(for unsetenv returns a value, rb_cv_unsetenv_return_value,
   [AC_TRY_COMPILE([
diff --git a/io.c b/io.c
index e7116cd..3173d85 100644
--- a/io.c
+++ b/io.c
@@ -78,6 +78,10 @@
 #include <sys/syscall.h>
 #endif
 
+#ifdef HAVE_SYS_UIO_H
+#include <sys/uio.h>
+#endif
+
 #if defined(__BEOS__) || defined(__HAIKU__)
 # ifndef NOFILE
 #  define NOFILE (OPEN_MAX)
@@ -840,6 +844,14 @@ struct io_internal_write_struct {
     size_t capa;
 };
 
+#ifdef HAVE_WRITEV
+struct io_internal_writev_struct {
+    int fd;
+    const struct iovec *iov;
+    int iovcnt;
+};
+#endif
+
 static VALUE
 internal_read_func(void *ptr)
 {
@@ -854,6 +866,15 @@ internal_write_func(void *ptr)
     return write(iis->fd, iis->buf, iis->capa);
 }
 
+#ifdef HAVE_WRITEV
+static VALUE
+internal_writev_func(void *ptr)
+{
+    struct io_internal_writev_struct *iis = ptr;
+    return writev(iis->fd, iis->iov, iis->iovcnt);
+}
+#endif
+
 static ssize_t
 rb_read_internal(int fd, void *buf, size_t count)
 {
@@ -876,6 +897,19 @@ rb_write_internal(int fd, const void *buf, size_t count)
     return (ssize_t)rb_thread_io_blocking_region(internal_write_func, &iis, fd);
 }
 
+#ifdef HAVE_WRITEV
+static ssize_t
+rb_writev_internal(int fd, const struct iovec *iov, int iovcnt)
+{
+    struct io_internal_writev_struct iis;
+    iis.fd = fd;
+    iis.iov = iov;
+    iis.iovcnt = iovcnt;
+
+    return (ssize_t)rb_thread_io_blocking_region(internal_writev_func, &iis, fd);
+}
+#endif
+
 static long
 io_writable_length(rb_io_t *fptr, long l)
 {
@@ -1063,6 +1097,43 @@ struct write_arg {
     int nosync;
 };
 
+#ifdef HAVE_WRITEV
+static VALUE
+io_binwritev_string(VALUE arg)
+{
+    struct binwrite_arg *p = (struct binwrite_arg *)arg;
+    rb_io_t *fptr = p->fptr;
+    long r;
+
+    if (fptr->wbuf.len) {
+	struct iovec iov[2];
+
+	iov[0].iov_base = fptr->wbuf.ptr+fptr->wbuf.off;
+	iov[0].iov_len = fptr->wbuf.len;
+	iov[1].iov_base = (char *)p->ptr;
+	iov[1].iov_len = p->length;
+
+	r = rb_writev_internal(fptr->fd, iov, 2);
+
+	if (fptr->wbuf.len <= r) {
+	    r -= fptr->wbuf.len;
+	    fptr->wbuf.off = 0;
+	    fptr->wbuf.len = 0;
+	}
+	else {
+	    fptr->wbuf.off += (int)r;
+	    fptr->wbuf.len -= (int)r;
+	    r = 0L;
+	}
+    }
+    else {
+	long l = io_writable_length(fptr, p->length);
+	r = rb_write_internal(fptr->fd, p->ptr, l);
+    }
+
+    return r;
+}
+#else
 static VALUE
 io_binwrite_string(VALUE arg)
 {
@@ -1070,6 +1141,7 @@ io_binwrite_string(VALUE arg)
     long l = io_writable_length(p->fptr, p->length);
     return rb_write_internal(p->fptr->fd, p->ptr, l);
 }
+#endif
 
 static long
 io_binwrite(VALUE str, const char *ptr, long len, rb_io_t *fptr, int nosync)
@@ -1088,7 +1160,19 @@ io_binwrite(VALUE str, const char *ptr, long len, rb_io_t *fptr, int nosync)
         (fptr->wbuf.ptr && fptr->wbuf.capa <= fptr->wbuf.len + len)) {
 	struct binwrite_arg arg;
 
-        /* xxx: use writev to avoid double write if available */
+	arg.fptr = fptr;
+	arg.str = str;
+#ifdef HAVE_WRITEV
+      retry:
+	arg.ptr = ptr + offset;
+	arg.length = n;
+	if (fptr->write_lock) {
+	    r = rb_mutex_synchronize(fptr->write_lock, io_binwritev_string, (VALUE)&arg);
+	}
+	else {
+	    r = io_binwritev_string((VALUE)&arg);
+	}
+#else
         if (fptr->wbuf.len && fptr->wbuf.len+len <= fptr->wbuf.capa) {
             if (fptr->wbuf.capa < fptr->wbuf.off+fptr->wbuf.len+len) {
                 MEMMOVE(fptr->wbuf.ptr, fptr->wbuf.ptr+fptr->wbuf.off, char, fptr->wbuf.len);
@@ -1107,8 +1191,6 @@ io_binwrite(VALUE str, const char *ptr, long len, rb_io_t *fptr, int nosync)
 	if (fptr->stdio_file != stderr && !rb_thread_fd_writable(fptr->fd)) {
 	    rb_io_check_closed(fptr);
 	}
-	arg.fptr = fptr;
-	arg.str = str;
       retry:
 	arg.ptr = ptr + offset;
 	arg.length = n;
@@ -1119,6 +1201,7 @@ io_binwrite(VALUE str, const char *ptr, long len, rb_io_t *fptr, int nosync)
 	    long l = io_writable_length(fptr, n);
 	    r = rb_write_internal(fptr->fd, ptr+offset, l);
 	}
+#endif
 	/* xxx: other threads may modify given string. */
         if (r == n) return len;
         if (0 <= r) {
@@ -9373,10 +9456,6 @@ simple_sendfile(int out_fd, int in_fd, off_t *offset, off_t count)
  */
 #  define USE_SENDFILE
 
-#  ifdef HAVE_SYS_UIO_H
-#   include <sys/uio.h>
-#  endif
-
 static ssize_t
 simple_sendfile(int out_fd, int in_fd, off_t *offset, off_t count)
 {