Floating point exception in process status report

S

sonyk

Hi everybody, looking forward to receive a help regarding the problem below.
Recently 'ps' utility has crashed on one of servers that I work with. We are using custom BSD-based operating system, I'm not sure, but I think it's FreeBSD3 that our system is derived from.

here is a part of code from ps/print.c:
Code: 
double
getpmem(k)
	KINFO *k;
{
	static int failure;
	struct proc *p;
	struct eproc *e;
	double fracmem;
	int szptudot;

	if (!nlistread)
		failure = donlist();
	if (failure)
		return (0.0);

	p = KI_PROC(k);
	e = KI_EPROC(k);
	if ((p->p_flag & P_INMEM) == 0)
		return (0.0);
#ifndef NEWVM
	szptudot = UPAGES + clrnd(ctopt(p->p_dsize + p->p_ssize + e->e_xsize));
	fracmem = ((float)p->p_rssize + szptudot)/CLSIZE/mempages;
	if (p->p_textp && e->e_xccount)
		fracmem += ((float)e->e_xrssize)/CLSIZE/e->e_xccount/mempages;
#else
	/* XXX want pmap ptpages, segtab, etc. (per architecture) */
	szptudot = UPAGES;
	/* XXX don't have info about shared */
	fracmem = ((float)e->e_vm.vm_rssize + szptudot)/CLSIZE/mempages;
#endif
	return (100.0 * fracmem);
}

return (100.0 * fracmem); raised a floating point exception as I see in a crash dump, but I'm unable to see the value of fracmem and also I'm not a very proficient C programmer and I can't even suggest why floating point exception could have place here...

getpmem(k) is called from pmem(k, ve):

Code: 
void
pmem(k, ve)
	KINFO *k;
	VARENT *ve;
{
	VAR *v;

	v = ve->var;
	(void)printf("%*.1f", v->width, getpmem(k));
}

but pmem is called from ps/ps.c as a handler like it's described in ps/keyword.c:

Code: 
VAR var[] = {
...
{"%mem", "%MEM", NULL, 0, pmem, 4},
...
}

I hope everything is described very precisely. Waiting for your comments, thanks in advance.
 
This code is so old, it's not even funny ;)

This is the same code block in FreeBSD 8

Code: 
static double
getpmem(KINFO *k)
static double
getpmem(KINFO *k)
{
        static int failure;
        double fracmem;

        if (!nlistread)
                failure = donlist();
        if (failure)
        if (failure)
                return (0.0);

        if ((k->ki_p->ki_flag & P_INMEM) == 0)
                return (0.0);
        /* XXX want pmap ptpages, segtab, etc. (per architecture) */
        /* XXX don't have info about shared */
        fracmem = ((float)k->ki_p->ki_rssize) / mempages;
        return (100.0 * fracmem);
}

void
pmem(KINFO *k, VARENT *ve)
{
        VAR *v;

        v = ve->var;
        (void)printf("%*.1f", v->width, getpmem(k));
}
 
Yes, I have already compared it with new one. But I still need answers. Why could floating point exception occur there? Is there a warranty that it will not occur in a newer version? It was a single occurrence, but in my case it is very undesirable to have a possibility of one occurrence. And if we are talking about how old my code is... Could you please tell about differences and difficulties that I may face if I upgrade only 'ps' utility? Since I'm not very familiar with all BSD changes from 3-8 and having in mind that it's fully impossible to upgrade whole system version.
 
FreeBSD (which is an integral OS) can not be upgraded in parts due to severe library changes, and certainly not over 5 major versions. FreeBSD 3 is severely EOL, so hop over to at least FreeBSD 7 or live with the consequences, I'm afraid.
 
Let's say I have no idea why your ps utility started crashing 'recently' on only one of your servers, because you make it sound like this configuration has been static since the FreeBSD 3 days. I've never experienced FPE errors with ps, ever. Must be something introduced on that system, and recently.
 
But talking about C. What could be there to cause FPE? I can't find anywhere on internet the reasons for FPE in C.
 
finally, thanks to my team-mate, we were able to reproduce the bug. "ps -u /var", "ps u /sbin" made utility to crash with message:

Code: 
USER       PID %CPU %MEM   VSZ  RSS  TT  STAT STARTED       TIME COMMAND
ps: nlist: can't find following symbols: _fscale _ccpu _avail_start _avail_end
Floating point exception (core dumped)

maybe someone knows in what ps utility version this was fixed and what is the root of the issue ?
 
I'm by no means a programmer so I could very well be wrong..

Reading this: http://www.freebsd.org/cgi/man.cgi?...ion=3&manpath=FreeBSD+3.0-RELEASE&format=html

There are several options an FPE could occur.
Code: 
#define fp_except_t int
#define FP_X_INV	 0x01	   /* invalid */
#define FP_X_OFL	 0x08	   /* overflow */
#define FP_X_UFL	 0x10	   /* underflow */
#define FP_X_DZ	 0x04	   /* divide-by-zero */
#define FP_X_IMP	 0x20	   /* loss of precision */
#define FP_X_DNML	 0x02	   /* denormal */

Looking at where the error happened it could be an overflow.
 
What is the purpose of [cmd=]ps -u /var[/cmd], or am I missing something here?
 
Don't do it then. Seriously, both the -v and -u option don't have any options.
 
I know that i can live without doing it, but end user - can't. So I need to find and remove a breakable place.
 
I don't know why anyone would need
Code: 
$ ps -u /var
ps: illegal argument: /var
It's a nonsensical command.
 
I now have looked through open/net/free BSD repositories for a "ps: nlist: can't find following symbols:" phrase, and looks like the problem is actual for NetBSD. I will look for version differences now.
 
You must log in or register to reply here.
Back
Top