[ECOS] TCP close(...) action

[ECOS] TCP close(...) action

[Matt Jerdonek]

Hi Folks,

I have a simple program with two threads that send /
recv from a single TCP socket.  One thread blocks on a
recv call while the other thread sends data on the
same socket.

I put a close call in the send thread.  The expected
behavior was for the recv thread to wake (with 0 bytes
of data) and a FIN to be sent on the ethernet.  The
actual behavior was that the recv thread never woke
and the FIN was not sent.  (Is this intended or a
bug?).  I found the soclose function would not be
invoked because the recv was still using the file
handle.  Once the recv released the file handle, the
FIN flowed on the ethernet.

I worked around this issue by calling
cyg_thread_release from my application, which woke up
the recv thread.  But I wonder if there is better
solution?  Could (or should) the close call be made to
wake up blocked threads?  If so, any suggestions?

Thanks,
-- Matt

__________________________________
Do you Yahoo!?
The New Yahoo! Shopping - with improved product search
http://shopping.yahoo.com

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

Re: [ECOS] TCP close(...) action

[Gary Thomas]

On Sat, 2003-10-18 at 12:24, Matt Jerdonek wrote:
> Hi Folks,
> 
> I have a simple program with two threads that send /
> recv from a single TCP socket.  One thread blocks on a
> recv call while the other thread sends data on the
> same socket.
> 
> I put a close call in the send thread.  The expected
> behavior was for the recv thread to wake (with 0 bytes
> of data) and a FIN to be sent on the ethernet.  The
> actual behavior was that the recv thread never woke
> and the FIN was not sent.  (Is this intended or a
> bug?).  I found the soclose function would not be
> invoked because the recv was still using the file
> handle.  Once the recv released the file handle, the
> FIN flowed on the ethernet.
> 
> I worked around this issue by calling
> cyg_thread_release from my application, which woke up
> the recv thread.  But I wonder if there is better
> solution?  Could (or should) the close call be made to
> wake up blocked threads?  If so, any suggestions?
> 

How did you create this socket?  How did you set up
the send and receive threads?  Maybe a code snippet would
help us understand your problem.

-- 
Gary Thomas <gary@mlbassoc.com>
MLB Associates


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

Re: [ECOS] TCP close(...) action

[Matt Jerdonek]

Gary,

I can't send the actual code, so I put together this
program.  I verified this program exhibits the same
behavior as I initially described.

In addition, I put some comments in the code regarding
the f_ucount.  This is a member of the cyg_file
structure which is associated with each socket.  Once
the f_ucount reaches 0, the FIN will be sent on the
ethernet.


#include <cyg/kernel/kapi.h>
#include <assert.h>
#include <stdio.h>
#include <math.h>
#include <stdlib.h>
#include <network.h>
#include <cyg/posix/pthread.h>
#include <unistd.h>

#define STACK_SIZE 0x4000
unsigned char Stack[STACK_SIZE];
pthread_t ThreadHandle;
int sockfd;
void RecvThread(void);

int main(void)
{
	pthread_attr_t ThreadAttrib;
	
	init_all_network_interfaces();
	
	// Create thread to make TCP connection
	pthread_attr_init(&ThreadAttrib);
	pthread_attr_setstackaddr(&ThreadAttrib, 
                                  &Stack[STACK_SIZE]);
	pthread_attr_setstacksize(&ThreadAttrib,
                                   STACK_SIZE);
	pthread_create(&ThreadHandle, &ThreadAttrib,
                      (void *)RecvThread, 
                      (void *)NULL);
	
	// Give the thread plenty of time to connect
	sleep(15);
	
	// Workaround ... 
        //  ... call cyg_thread_release() here ...
        //  ... on the RecvThread
	
	// Close the socket
	printf("Closing socket\n");
	close(sockfd);
	
	// The close will decrement the f_ucount ... 
        // ... from 2 to 1, but the FIN will not ...
	// ... be sent until the recv is released. 
        // ... See function fp_ucount_dec( ) ...
	// ... in io/fileio/ver/src/fd.cxx
	
	sleep(10);
	return(0);
}

void RecvThread(void)
{
	// Create socket and connect to host
	sockfd = socket(AF_INET, SOCK_STREAM, 0);
	if (sockfd > 0)
	{
		struct sockaddr RmtHost;
		struct sockaddr_in *pRmtHost = 
                   (struct sockaddr_in *)&RmtHost;
		
		// At this point the cyg_file ..
                // ... structure that describes ... 
                // ... the sockfd ...
		// ... as a f_ucount = 1
			
		// Setup connect
		memset( &RmtHost, 0, 
                        sizeof(struct sockaddr) );
		pRmtHost->sin_len = 
                        sizeof( struct sockaddr_in );	
		pRmtHost->sin_family = AF_INET;
		pRmtHost->sin_port = htons(4096);
		pRmtHost->sin_addr.s_addr = 
                        htonl(0xc8640196);
			
		// Connected to the host
		if (connect(sockfd, &RmtHost, 
                      sizeof(struct sockaddr)) >= 0)
		{
			unsigned char RxBuffer[100];
			printf("Socket connected\n");
			
			// Will block indefinitely ...
                        // ... since no data will ...
                        // ... be received 
			recv(sockfd, &RxBuffer[0], 
                                          1, 0);
			
			// Inside the recv, ... 
                        // ... the f_ucount is now 2
			
			printf("Recv woke\n");
		}
	}	
}



--- Gary Thomas <gary@mlbassoc.com> wrote:
> On Sat, 2003-10-18 at 12:24, Matt Jerdonek wrote:
> > Hi Folks,
> > 
> > I have a simple program with two threads that send
> /
> > recv from a single TCP socket.  One thread blocks
> on a
> > recv call while the other thread sends data on the
> > same socket.
> > 
> > I put a close call in the send thread.  The
> expected
> > behavior was for the recv thread to wake (with 0
> bytes
> > of data) and a FIN to be sent on the ethernet. 
> The
> > actual behavior was that the recv thread never
> woke
> > and the FIN was not sent.  (Is this intended or a
> > bug?).  I found the soclose function would not be
> > invoked because the recv was still using the file
> > handle.  Once the recv released the file handle,
> the
> > FIN flowed on the ethernet.
> > 
> > I worked around this issue by calling
> > cyg_thread_release from my application, which woke
> up
> > the recv thread.  But I wonder if there is better
> > solution?  Could (or should) the close call be
> made to
> > wake up blocked threads?  If so, any suggestions?
> > 
> 
> How did you create this socket?  How did you set up
> the send and receive threads?  Maybe a code snippet
> would
> help us understand your problem.
> 
> -- 
> Gary Thomas <gary@mlbassoc.com>
> MLB Associates
> 
> 
> -- 
> Before posting, please read the FAQ:
> http://sources.redhat.com/fom/ecos
> and search the list archive:
> http://sources.redhat.com/ml/ecos-discuss
> 


__________________________________
Do you Yahoo!?
The New Yahoo! Shopping - with improved product search
http://shopping.yahoo.com

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

Re: [ECOS] TCP close(...) action

[Christoph Csebits]

Hi,

try to set the SO_LINGER option  with setsockopt, e.g:

   struct linger   ling = { 1, 0 };
   if ( setsockopt( sockfd, SOL_SOCKET, SO_LINGER, &ling, sizeof ( ling ) ) < 0)
   {
      // something went wrong, clean up
   }

regards, Christoph


On Mon, Oct 20, 2003 at 08:25:43AM -0700, Matt Jerdonek wrote:
> > > Hi Folks,
> > > 
> > > I have a simple program with two threads that send
> > /
> > > recv from a single TCP socket.  One thread blocks
> > on a
> > > recv call while the other thread sends data on the
> > > same socket.
> > > 
> > > I put a close call in the send thread.  The
> > expected
> > > behavior was for the recv thread to wake (with 0
> > bytes
> > > of data) and a FIN to be sent on the ethernet. 
> > The
> > > actual behavior was that the recv thread never
> > woke
> > > and the FIN was not sent.  (Is this intended or a
> > > bug?).  I found the soclose function would not be
> > > invoked because the recv was still using the file
> > > handle.  Once the recv released the file handle,
> > the
> > > FIN flowed on the ethernet.
> > > 
> > > I worked around this issue by calling
> > > cyg_thread_release from my application, which woke
> > up
> > > the recv thread.  But I wonder if there is better
> > > solution?  Could (or should) the close call be
> > made to
> > > wake up blocked threads?  If so, any suggestions?
> > > 

-- 

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

Re: [ECOS] TCP close(...) action

[Andrew Lunn]

On Sat, Oct 18, 2003 at 11:24:40AM -0700, Matt Jerdonek wrote:
> Hi Folks,
> 
> I have a simple program with two threads that send /
> recv from a single TCP socket.  One thread blocks on a
> recv call while the other thread sends data on the
> same socket.
> 
> I put a close call in the send thread.  The expected
> behavior was for the recv thread to wake (with 0 bytes
> of data) and a FIN to be sent on the ethernet.  The
> actual behavior was that the recv thread never woke
> and the FIN was not sent.  (Is this intended or a
> bug?).  I found the soclose function would not be
> invoked because the recv was still using the file
> handle.  Once the recv released the file handle, the
> FIN flowed on the ethernet.
> 
> I worked around this issue by calling
> cyg_thread_release from my application, which woke up
> the recv thread.  But I wonder if there is better
> solution?  Could (or should) the close call be made to
> wake up blocked threads?  If so, any suggestions?

I think what its currently doing is at least sensible. Should close
cause other calls using the socket to fail and return an error? What
error should it return? To answer these questions i think you need to
look at the standards and see what other multithreaded OS's do in the
same situation. If you re-write your test program to use POSIX
threads, it should then be possible to run it on linux, SunOS, *BSD
etc. We can then decide if this is right or wrong.

     Andrew

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

Re: [ECOS] TCP close(...) action

[Matt Jerdonek]

> If you re-write your test program to use POSIX
> threads, it should then be possible to run it on
> linux, SunOS, *BSD etc. We can then decide if this 
> is right or wrong.
> 
>      Andrew
> 
> -- 


I ran the program on linux and got the same result
(the close did not wake the recv).  So, I guess that
eCos is behaving correctly.

As I mentioned before, I have a workaround using
cyg_thread_release(..).  Does anyone know of a POSIX
method for waking the recv other than ending and
restarting the thread? 

Thanks,
-- Matt

__________________________________
Do you Yahoo!?
The New Yahoo! Shopping - with improved product search
http://shopping.yahoo.com

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

Re: [ECOS] TCP close(...) action

[Nick Garnett]

Matt Jerdonek <maj1224@yahoo.com> writes:

> > If you re-write your test program to use POSIX
> > threads, it should then be possible to run it on
> > linux, SunOS, *BSD etc. We can then decide if this 
> > is right or wrong.
> > 
> >      Andrew
> > 
> > -- 
> 
> 
> I ran the program on linux and got the same result
> (the close did not wake the recv).  So, I guess that
> eCos is behaving correctly.

Given that the eCos file handling model is almost identical to that of
Linux, BSD and any other Unix variant, I would expect that they would
all behave in much the same way.

> 
> As I mentioned before, I have a workaround using
> cyg_thread_release(..).  Does anyone know of a POSIX
> method for waking the recv other than ending and
> restarting the thread? 

You might be able to do something fancy using select() in place of the
recv(). Signalling a select()ing thread should work -- although that
is essentially just the same as the cyg_thread_release() you are
already doing.

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


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

RE: [ECOS] TCP close(...) action

[Jay Foster]

You could try sending the process a signal with kill().  SIGUSR1 might work.

-----Original Message-----
From: Matt Jerdonek [mailto:maj1224@yahoo.com]
Sent: Wednesday, October 22, 2003 9:06 AM
To: Andrew Lunn
Cc: Discussion eCos
Subject: Re: [ECOS] TCP close(...) action




> If you re-write your test program to use POSIX
> threads, it should then be possible to run it on
> linux, SunOS, *BSD etc. We can then decide if this 
> is right or wrong.
> 
>      Andrew
> 
> -- 


I ran the program on linux and got the same result
(the close did not wake the recv).  So, I guess that
eCos is behaving correctly.

As I mentioned before, I have a workaround using
cyg_thread_release(..).  Does anyone know of a POSIX
method for waking the recv other than ending and
restarting the thread? 

Thanks,
-- Matt

__________________________________
Do you Yahoo!?
The New Yahoo! Shopping - with improved product search
http://shopping.yahoo.com

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

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