diff --git a/CHANGES b/CHANGES
index 089679b..552282e 100644
--- a/CHANGES
+++ b/CHANGES
@@ -3,6 +3,7 @@
   move headers to <libowfat/> upon install
   fix fmt_ip6 (Erwin Hoffmann)
   add MSG_ZEROCOPY support (only used for buffers >8k)
+  use write in buffer_put for a slight perf improvement
 
 0.31:
   special case buffer_get_token with token length 1 through memccpy (almost 4x speedup)
diff --git a/buffer/buffer_put.c b/buffer/buffer_put.c
index f38c66d..9ee9a0d 100644
--- a/buffer/buffer_put.c
+++ b/buffer/buffer_put.c
@@ -1,4 +1,8 @@
 #include <string.h>
+#include <unistd.h>
+#ifndef __MINGW32__
+#include <sys/uio.h>
+#endif
 #include "buffer.h"
 
 extern int buffer_stubborn(ssize_t (*op)(),int fd,const char* buf, size_t len,void* cookie);
@@ -13,12 +17,32 @@ extern int buffer_stubborn(ssize_t (*op)(),int fd,const char* buf, size_t len,vo
 
 int buffer_put(buffer* b,const char* buf,size_t len) {
   if (__unlikely(len>b->a-b->p)) {	/* doesn't fit */
+#ifndef __MINGW32__
+    if (b->op==write) {
+      /* if it's write, we can substitute writev */
+      struct iovec v[2];
+      ssize_t r;
+      v[0].iov_base=b->x; v[0].iov_len=b->p;
+      v[1].iov_base=(char*)buf; v[1].iov_len=len;
+      r=writev(b->fd,v,2);
+      if (r<0) return -1;
+      if ((size_t)r>=b->p) {
+	r-=b->p;
+	b->p=0;
+	buf+=r;
+	len-=r;
+	if (len) goto do_memcpy;
+	return 0;
+      } /* else fall through */
+    }
+#endif
     if (buffer_flush(b)==-1) return -1;
     if (len>b->a) {
       if (buffer_stubborn(b->op,b->fd,buf,len,b)<0) return -1;
       return 0;
     }
   }
+do_memcpy:
   memcpy(b->x+b->p, buf, len);
   b->p+=len;
   return 0;
diff --git a/dns/dns_rcrw.c b/dns/dns_rcrw.c
index 68616b6..ee7318e 100644
--- a/dns/dns_rcrw.c
+++ b/dns/dns_rcrw.c
@@ -7,6 +7,7 @@
 #include "dns.h"
 #ifdef __MINGW32__
 #include <winsock2.h>
+#include <windows.h>
 #endif
 
 static stralloc data;
diff --git a/dns/dns_transmit.c b/dns/dns_transmit.c
index 9a8c015..fb3c335 100644
--- a/dns/dns_transmit.c
+++ b/dns/dns_transmit.c
@@ -1,6 +1,7 @@
 #include <sys/types.h>
 #ifdef WIN32
 #include <winsock2.h>
+#include <windows.h>
 #else
 #include <sys/socket.h>
 #endif
diff --git a/io/io_block.c b/io/io_block.c
index 3ccfb11..0b9a324 100644
--- a/io/io_block.c
+++ b/io/io_block.c
@@ -1,13 +1,14 @@
-#include <unistd.h>
-#include <fcntl.h>
-#include <errno.h>
-#include "io_internal.h"
-
 #ifdef __MINGW32__
 #include <winsock2.h>
+#include <windows.h>
 #include "windoze.h"
 #endif
 
+#include <unistd.h>
+#include <fcntl.h>
+#include <errno.h>
+#include "io_internal.h"
+
 #ifndef O_NDELAY
 #define O_NDELAY O_NONBLOCK
 #endif
diff --git a/io/io_nonblock.c b/io/io_nonblock.c
index fa41a72..9a3b14c 100644
--- a/io/io_nonblock.c
+++ b/io/io_nonblock.c
@@ -5,6 +5,7 @@
 
 #ifdef __MINGW32__
 #include <winsock2.h>
+#include <windows.h>
 #include "windoze.h"
 #endif
 
diff --git a/mmap/mmap_read.3 b/mmap/mmap_read.3
index 2379285..51d85df 100644
--- a/mmap/mmap_read.3
+++ b/mmap/mmap_read.3
@@ -4,7 +4,7 @@ mmap_read \- memory map a file for reading
 .SH SYNTAX
 .B #include <libowfat/mmap.h>
 
-char* \fBmmap_read\fP(const char* \fIfilename\fR,size_t* \fIfilesize\fR);
+const char* \fBmmap_read\fP(const char* \fIfilename\fR,size_t* \fIfilesize\fR);
 .SH DESCRIPTION
 mmap_read opens \fIfilename\fR for reading, maps the whole file into
 memory, closes the file, writes the length of the file to \fIfilesize\fR
diff --git a/mmap/mmap_readat.3 b/mmap/mmap_readat.3
index 384fc1b..d0011be 100644
--- a/mmap/mmap_readat.3
+++ b/mmap/mmap_readat.3
@@ -4,7 +4,7 @@ mmap_readat \- memory map a file for reading
 .SH SYNTAX
 .B #include <libowfat/mmap.h>
 
-char* \fBmmap_readat\fP(const char* \fIfilename\fR,size_t* \fIfilesize\fR,int \fIdirfd\fR);
+const char* \fBmmap_readat\fP(const char* \fIfilename\fR,size_t* \fIfilesize\fR,int \fIdirfd\fR);
 .SH DESCRIPTION
 mmap_readat opens \fIfilename\fR for reading, maps the whole file into
 memory, closes the file, writes the length of the file to \fIfilesize\fR
diff --git a/test/iob.c b/test/iob.c
index a2c6b1c..760401e 100644
--- a/test/iob.c
+++ b/test/iob.c
@@ -18,7 +18,7 @@ int main() {
   assert(iob_addbuf(b," fnord\n",7));
   assert(iob_addfile(b,fd,10,10));
   iob_send(1,b);
-#if 0
+#if 1
   do {
     r=iob_write(1,b,write_cb);
   } while (r>0);
diff --git a/test/uudecode.c b/test/uudecode.c
index 4b8d968..c5efdbb 100644
--- a/test/uudecode.c
+++ b/test/uudecode.c
@@ -8,6 +8,7 @@
 #include <stdint.h>
 #include <stdlib.h>
 #include <string.h>
+#include <sys/stat.h>
 #include <errno.h>
 #include "textcode.h"
 #include "str.h"
diff --git a/unix/iopause.c b/unix/iopause.c
index 316ea5b..3e4cd82 100644
--- a/unix/iopause.c
+++ b/unix/iopause.c
@@ -2,6 +2,7 @@
 #include "iopause.h"
 #ifdef __MINGW32__
 #include <winsock2.h>
+#include <windows.h>
 #else
 #include "select.h"
 #endif
diff --git a/unix/ndelay_off.c b/unix/ndelay_off.c
index b26c89b..4c4312b 100644
--- a/unix/ndelay_off.c
+++ b/unix/ndelay_off.c
@@ -5,6 +5,7 @@
 
 #ifdef __MINGW32__
 #include <winsock2.h>
+#include <windows.h>
 #include "windoze.h"
 #endif
 
diff --git a/unix/ndelay_on.c b/unix/ndelay_on.c
index 64e9036..c9e2f14 100644
--- a/unix/ndelay_on.c
+++ b/unix/ndelay_on.c
@@ -5,6 +5,7 @@
 
 #ifdef __MINGW32__
 #include <winsock2.h>
+#include <windows.h>
 #include "windoze.h"
 #endif
 
diff --git a/unix/winsock2errno.c b/unix/winsock2errno.c
index c0cf71f..3a9d49c 100644
--- a/unix/winsock2errno.c
+++ b/unix/winsock2errno.c
@@ -1,5 +1,6 @@
 #ifdef __MINGW32__
 #include <winsock2.h>
+#include <windows.h>
 #include <errno.h>
 #include <stdio.h>
 #include "socket.h"