[Discuss-gnuradio] Re: memory leakage in usrp2::eth_buffer

[Gaetano Mendola]

This is the patch:

--- gnuradio-3.3.0/usrp2/host/lib/eth_buffer.cc 2010-06-01
07:04:09.000000000 +0200
+++ gnuradio-3.3.0-fix/usrp2/host/lib/eth_buffer.cc     2011-03-10
11:05:52.652353106 +0100
@@ -156,8 +156,14 @@
   {
     // if we have background thread, stop it here

-    if (!d_using_tpring && d_buf)
+    if (!d_using_tpring && d_buf) {
        free(d_buf);
+    }
+    else {
+        if(d_buf) {
+            munmap(d_buf, d_buflen);
+        }
+    }
        
     return d_ethernet->close();
   }


On Wed, Mar 9, 2011 at 7:35 PM, Gaetano Mendola <address@hidden> wrote:
> Hi,
> in the method usrp2::eth_buffer::open in case the attempt to use the
> socket option PACKET_RX_RING
> fails an mmap is performed instead of a malloc.
>
> The method usrp2::eth_buffer::close however has to perform a munmap in
> order to let the
> kernel release the mapped memory, indeed only closing the file
> descriptor isn't enough.
>
> I suggest to change (inside the close method):
>
>    if (!d_using_tpring && d_buf)
>           free(d_buf);
>
> to:
>
>
>  if (!d_using_tpring) {
>    free(d_buf);
>  } else {
>    if (d_buf) munmap(d_buf, d_buflen);
>  }
>
> note that there is no need to test for d_buf in case of free call
> indeed freeing a pointer to 0 is perfect valid code, the standard
> say that in case of free(NULL) no action is taken, to the other side
> munmap a NULL pointer is not permitted (hence the test).
>
> Regards
> Gaetano Mendola
>
> --
> cpp-today.blogspot.com
>



-- 
cpp-today.blogspot.com