"bind"ing sockets, -ObjC, and memset....A puzzle

Thread View: comp.sys.next.programmer

4 messages

[T] Back to Threads [V] Tree View [G] Group Articles

4 total messages Started by John Jay Feiler Wed, 26 Aug 1992 04:32

#802

Author: John Jay Feiler
Date: Wed, 26 Aug 1992 04:32

86 lines
2750 bytes

I had a strange problem today, and I thought I'd throw it out here to see if  
anyone had an explanation for it.  (at least one better than mine)

Background:

I'm doing some heterogenous ipc using internet sockets, so I wrote a small  
program to test my socket implementation, basically copied out of one of the  
Nutshell books.  Worked great.  Took the same code, and put in inside a  
method of an ObjC object.  No dice.  The call to bind failed, regardless of  
what port I gave it.  Hmmmm..., says I, what could be the problem?  Somehow,  
I got it down to the compiler flag -ObjC making the differnece.  The  
following (seriously incomplete) code, without the call to memset(&sin,...),  
will fail to bind correctly if compiled with -ObjC, but works fine without  
the -ObjC flag.

-------foo.m------------
#import <sys/types.h>
#import <sys/socket.h>
#import <netinet/in.h>
#import <netdb.h>
#import <stdio.h>
#import <libc.h>
extern int errno;

int main()
{
	char hostname[64];
	register int s;
	struct hostent *hp;
	struct sockaddr_in sin;
	int	i;
		
	gethostname(hostname,sizeof(hostname));

	if((hp = gethostbyname(hostname)) == NULL)
	{
		fprintf(stderr,"%s: unknown host.\n",hostname);
		exit (1);
	}

	if((s = socket(AF_INET,SOCK_STREAM,0))<0)
	{
		perror("server: socket");
		exit(1);
	}
	
	/********************
		if this memset is not performed, bind will likely fail
		with errno 49.
		could not bind to address.  Don't ask me why, It just
		does......
	********************/
	memset(&sin,0,sizeof(sin));
	
	
	sin.sin_family = AF_INET;
	bcopy(hp->h_addr, &sin.sin_addr, hp->h_length);
	
	/***********
		bind the address to the socket

		let's try all the ports we can think of, until we find one

		we should get #1024, because 1-1023 are reserved
	***********/
	
	for(i=1;i<30000;i++)
	{
		sin.sin_port = htons(i);
		if(bind(s, (struct sockaddr *)&sin, sizeof(sin)) == 0)
			printf("bound to port %d\n",i);
	}
	close(s);
	exit(0);
}
---------------------------

Now, it seems to me that bind shouldn't be looking at anything else in the  
sockaddr structure, other than the fields I'm setting, but somehow the memset  
call solves the problem.  Any ideas as to what might be causing things to  
fail without the memset and with the -ObjC flag?

	Puzzled,

		John

Re: "bind"ing sockets, -ObjC, and memset....A puzzle

#880

Author: simsong Simson L. Garfinkel
Date: Fri, 28 Aug 1992 19:02

44 lines
1236 bytes

Apparently, when you include the -ObjC flag, all stack variables are  
initialized to 0.

Here's a simple test program:

int     foo(n)
{
        int     a,b;
        char    f[2342];
        int     c;

        printf("a=%d  b=%d  c=%d\n",a,b,c);
        a = n;
        c = n;
}

main()
{
        int a,b,c;

        foo(0);
        foo(2);
}



And here is the output:

altos> cc x.c
altos> a.out
a=84032826  b=212  c=0
a=212  b=57628  c=0
altos> cc -ObjC x.c
altos> a.out
a=0  b=0  c=0
a=0  b=0  c=0
altos> %


(ps: Other variables in the sin structure are used; it's always a good idea to  
initialize things to 0.)
-- 

................................................................simson

Re: "bind"ing sockets, -ObjC, and memset....A puzzle

#935

Author: Daryll Strauss
Date: Mon, 31 Aug 1992 16:47

12 lines
1041 bytes

In article <1992Aug28.190230.2670@next.cambridge.ma.us>, simsong (Simson L. Garfinkel) writes:
|> Apparently, when you include the -ObjC flag, all stack variables are  
|> initialized to 0.
|> 

It would be a really bad idea to rely on that "feature." You must clear
the sockaddr structure to use it. The C spec says that stack variables
have unknown values. I've need heard anything different in for ObjC. So
relying on the compiler to initialize variables for you is a disaster
waiting to happen.

						- |Daryll

Re: "bind"ing sockets, -ObjC, and memset....A puzzle

#947

Author: Moises Oliveira
Date: Mon, 31 Aug 1992 19:22

50 lines
2014 bytes

In article <17tieoINNsr5@news.aero.org> strauss@Aero.org (Daryll Strauss) writes:
>In article <1992Aug28.190230.2670@next.cambridge.ma.us>, simsong (Simson L. Garfinkel) writes:
>|> Apparently, when you include the -ObjC flag, all stack variables are  
>|> initialized to 0.
>|> 
>
>It would be a really bad idea to rely on that "feature." You must clear
>the sockaddr structure to use it. The C spec says that stack variables
>have unknown values. I've need heard anything different in for ObjC. So
>relying on the compiler to initialize variables for you is a disaster
>waiting to happen.

It's especially a bad idea seeing as how it's plain out
untrue!  The ObjC flag to the linker DOES NOT clear stack
variables to 0.  When you pass ObjC to the linker, it pulls
in a bunch of support for ObjC.  Because of this, in an
itsy-bitsy tiny example program like the one posted, a
large portion of the stack will end up filled with 0
values.   

But this will only remain true for the first function
call.  To see what I mean, change the program a bit: 

int     foo(n)
{
        int     a,b;
        int     c;

        printf("a=%d  b=%d  c=%d\n",a,b,c);
        a = n;
        c = n;
}

main()
{
        int a,b,c;

	foo(17);
	foo(13);
	foo(11);
	foo(7);
	foo(5);
	foo(3);
	foo(2);
	foo(1);
}

The sample program and output that Simson Garfinkel
posted works by mere coincidence.

Thread Navigation

This is a paginated view of messages in the thread with full content displayed inline.

Messages are displayed in chronological order, with the original post highlighted in green.

Use pagination controls to navigate through all messages in large threads.

Back to All Threads