FWD: [ECOS] Only return EWOULDBLOCK if no data moved

FWD: [ECOS] Only return EWOULDBLOCK if no data moved

[Andrew Lunn]

Hi Folks

What do people think about this patch? 

Should be apply it, or try to fix the network stack?

       Thanks
          Andrew


----- Forwarded message from Roland Ca?ebohm <roland.cassebohm@visionsystems.de> -----

From: Roland Ca?ebohm <roland.cassebohm@visionsystems.de>
To: ecos-discuss@ecos.sourceware.org
Date: Tue, 27 Jan 2004 18:03:26 +0100
Subject: [ECOS] Only return EWOULDBLOCK if no data moved
X-Spam-Status: No, hits=-2.5 required=5.0 tests=AWL,BAYES_00 autolearn=ham 
	version=2.63

Hello,

both TCP/IP stacks could return EWOULDBLOCK even if there are data moved
in some cases.
My first solution was the attached patch, but maybe it is better to fix
the problem in the stacks it self?

I have seen there was a semilar problem in the serial driver:

2003-03-20  Gary Thomas  <gary@mlbassoc.com>

        * src/common/serial.c: Only return -EAGAIN if no data moved.

The attached patch would fix it for all possible places where something
like this could happen.

Any suggestions?

Roland

Index: io/fileio/current/src/io.cxx
===================================================================
RCS file: /home/cassebohm/net/USERS/CVSROOT/VSprojects/ecos/packages/io/fileio/current/src/io.cxx,v
retrieving revision 1.1.1.1.2.1
diff -u -5 -p -r1.1.1.1.2.1 io.cxx
--- io/fileio/current/src/io.cxx	12 Nov 2003 10:57:18 -0000	1.1.1.1.2.1
+++ io/fileio/current/src/io.cxx	27 Jan 2004 16:50:50 -0000
@@ -144,11 +144,16 @@ readwritev( int fd, const cyg_iovec *_io
     cyg_fp_free( fp );
 
     CYG_CANCELLATION_POINT;
 
     if( ret != 0 )
-        FILEIO_RETURN(ret);
+    {
+        if ((ret == EWOULDBLOCK || ret == EAGAIN) && cnt)
+            FILEIO_RETURN_VALUE(cnt);
+        else
+            FILEIO_RETURN(ret);
+    }
  
     FILEIO_RETURN_VALUE(cnt);
 }
 
 //==========================================================================


-- 
Before posting, please read the FAQ: http://sources.redhat.com/fom/ecos
and search the list archive: http://sources.redhat.com/ml/ecos-discuss

----- End forwarded message -----

Re: FWD: [ECOS] Only return EWOULDBLOCK if no data moved

[Jonathan Larmour]

Andrew Lunn wrote:
> Hi Folks
> 
> What do people think about this patch? 
> 
> Should be apply it, or try to fix the network stack?

The patch is correct. POSIX says for read() in relation to return values:

[EAGAIN] or [EWOULDBLOCK]: The file descriptor is for a connection-made 
socket, is marked O_NONBLOCK, and no data is waiting to be received.

In the case considered, some data is waiting to be received, so 
EAGAIN/EWOULDBLOCK is the wrong thing to return.

Jifl


> ----- Forwarded message from Roland Ca?ebohm <roland.cassebohm@visionsystems.de> -----
> 
> From: Roland Ca?ebohm <roland.cassebohm@visionsystems.de>
> To: ecos-discuss@ecos.sourceware.org
> Date: Tue, 27 Jan 2004 18:03:26 +0100
> Subject: [ECOS] Only return EWOULDBLOCK if no data moved
> X-Spam-Status: No, hits=-2.5 required=5.0 tests=AWL,BAYES_00 autolearn=ham 
> 	version=2.63
> 
> Hello,
> 
> both TCP/IP stacks could return EWOULDBLOCK even if there are data moved
> in some cases.
> My first solution was the attached patch, but maybe it is better to fix
> the problem in the stacks it self?
> 
> I have seen there was a semilar problem in the serial driver:
> 
> 2003-03-20  Gary Thomas  <gary@mlbassoc.com>
> 
>         * src/common/serial.c: Only return -EAGAIN if no data moved.
> 
> The attached patch would fix it for all possible places where something
> like this could happen.



> Any suggestions?
> 
> Roland
> 
> Index: io/fileio/current/src/io.cxx
> ===================================================================
> RCS file: /home/cassebohm/net/USERS/CVSROOT/VSprojects/ecos/packages/io/fileio/current/src/io.cxx,v
> retrieving revision 1.1.1.1.2.1
> diff -u -5 -p -r1.1.1.1.2.1 io.cxx
> --- io/fileio/current/src/io.cxx	12 Nov 2003 10:57:18 -0000	1.1.1.1.2.1
> +++ io/fileio/current/src/io.cxx	27 Jan 2004 16:50:50 -0000
> @@ -144,11 +144,16 @@ readwritev( int fd, const cyg_iovec *_io
>      cyg_fp_free( fp );
>  
>      CYG_CANCELLATION_POINT;
>  
>      if( ret != 0 )
> -        FILEIO_RETURN(ret);
> +    {
> +        if ((ret == EWOULDBLOCK || ret == EAGAIN) && cnt)
> +            FILEIO_RETURN_VALUE(cnt);
> +        else
> +            FILEIO_RETURN(ret);
> +    }
>   
>      FILEIO_RETURN_VALUE(cnt);
>  }
>  
>  //==========================================================================
> 
> 


-- 
eCosCentric    http://www.eCosCentric.com/    The eCos and RedBoot experts
Visit us at Embedded World 2004, Nürnberg, Germany, 17-19 Feb, Stand 12-449
--["No sense being pessimistic, it wouldn't work anyway"]-- Opinions==mine

Re: FWD: [ECOS] Only return EWOULDBLOCK if no data moved

[Nick Garnett]

Andrew Lunn <andrew@lunn.ch> writes:

> Hi Folks
> 
> What do people think about this patch? 
> 
> Should be apply it, or try to fix the network stack?

I think it is OK to apply this patch as it stands. In theory it makes
it slightly easier to write filesystems/devices/netstacks. However,
since EWOULDBLOCK and EAGAIN are actually synonyms, there is no need
to check for both.

-- 
Nick Garnett                    eCos Kernel Architect
http://www.ecoscentric.com      The eCos and RedBoot experts