Re: [PATCH] Add support for mkostemp, mkstemps and mkostemps

["Anthony G. Basile" <basile@opensource.dyc.edu>]

Okay let me try those fixes.  Some points:

1. I forgot to remove a group of headers when I refactorized.  So 
mkostemp.c mkostemps.c mkstemp.c and mkstemps.c only need

#include <fcntl.h>
#include "libc.h"

or

#define _GNU_SOURCE
#include "libc.h"


2. This is from uclibc.  Clearly, static is critical here, but still 
they never initialize a value of 'value' on first entry into the 
function, so that memory is dirty to start.  Mine is worse, but I wonder 
if this is still a bug there.

static void brain_damaged_fillrand(unsigned char *buf, unsigned int len)
{
	...
         static uint64_t value;
         gettimeofday(&tv, NULL);
         value += ((uint64_t) tv.tv_usec << 16) ^ tv.tv_sec ^ getpid();
	...


3. I retested your address approach.  I like it but it only maps to 
upper and lower case letters, no numbers which uclibc and glibc do.

clock_gettime(CLOCK_REALTIME, &ts);
r = ts.tv_nsec*65537 ^ (uintptr_t)&ts / 16 + (uintptr_t)template;
for (i=0; i<6; i++, r>>=5)
	template[i] = 'A'+(r&15)+(r&16)*2;




On 02/02/2013 03:14 PM, Szabolcs Nagy wrote:
> * Anthony G. Basile <basile@opensource.dyc.edu> [2013-02-02 13:45:31 -0500]:
>> +	/* Null terminate the template before the suffix,
>> +	   and save the char for adding back the suffix */
>> +	char suffix = template[l];
>> +	template[l] = '\0';
>
> if you set only the XXXXXX part in __randname, then the \0 is unnecessary
>
>> +	int fd, retries = 100;
>> +	while (retries--) {
>> +		if (!*__randname(template)) return -1;
>
> __randname cannot fail, so the check is unnecessary
>
>> +/* This assumes that a check for the
>> +   template size has alrady been made */
>> +char *__randname(char *template)
>> +{
>> +	struct timespec ts;
>> +	size_t i, l = strlen(template);
>> +
>> +	/* r is intentially uninialized and 'dirty' */
>> +	unsigned long r;
>> +
>
> it's undefined behaviour so the compiler is allowed to
> completely remove the code of this function
>
> if you seen this kind of code somewhere, that's a critical
> bug that should be reported
>
> the original address based entropy source was ok
>
>> +	clock_gettime(CLOCK_REALTIME, &ts);
>> +	r += ((uint64_t) ts.tv_nsec << 16) ^ ts.tv_sec;
>> +	for (i=1; i<=6; i++, r>>=6)
>> +		template[l-i] = __map_letter(r);
>> +
>> +	return template;
>> +}


-- 
Anthony G. Basile, Ph. D.
Chair of Information Technology
D'Youville College
Buffalo, NY 14201
(716) 829-8197