GNU bug report logs - #78507
[Security] Heap Buffer Overflow in GNU Coreutils sort (CWE-122)

Previous Next

Full log

View this message in rfc822 format

From: Pádraig Brady <P <at> draigBrady.com>
To: Med Maatallah <hotelsmaatallahrecemail <at> gmail.com>, 78507 <at> debbugs.gnu.org
Subject: bug#78507: [Security] Heap Buffer Overflow in GNU Coreutils sort (CWE-122)
Date: Tue, 20 May 2025 16:15:26 +0100

On 20/05/2025 10:31, Med Maatallah wrote:
> Dear GNU Coreutils Maintainers,
> 
> I am reporting a heap buffer overflow vulnerability (CWE-122) I've
> discovered in the GNU Coreutils sort utility. This issue affects the
> traditional key specification syntax processing and leads to an
> out-of-bounds read.
> Vulnerability Details
> 
> The vulnerability occurs when the traditional key specification syntax (
> +POS1[.C1][OPTS]) is used with UINTMAX_MAX as the character position value.
> The begfield() function in src/sort.c performs unsafe pointer arithmetic
> that leads to integer wraparound, resulting in a pointer that points one
> byte before the start of an allocated heap buffer.
> 
> The vulnerable code is in the begfield() function in src/sort.c:
> 
> 
> static char *begfield (struct line const *line, struct keyfield const *key){
>    char *ptr = line->text, *lim = ptr + line->length - 1;
>    size_t sword = key->sword;
>    size_t schar = key->schar;
> 
>    /* The leading field separator itself is included in a field when -t
>      is absent.  */
> 
>    if (tab != TAB_DEFAULT)
>      while (ptr < lim && sword--)
>        {
>          while (ptr < lim && *ptr != tab)
>            ++ptr;
>          if (ptr < lim)
>            ++ptr;
>        }
>    else
>      while (ptr < lim && sword--)
>        {
>          while (ptr < lim && blanks[to_uchar (*ptr)])
>            ++ptr;
>          while (ptr < lim && !blanks[to_uchar (*ptr)])
>            ++ptr;
>        }
> 
>    /* If we're ignoring leading blanks when computing the Start     of
> the field, skip past them here.  */
>    if (key->skipsblanks)
>      while (ptr < lim && blanks[to_uchar (*ptr)])
>        ++ptr;
> 
>    /* Advance PTR by SCHAR (if possible), but no further than LIM. */
>    ptr = MIN (lim, ptr + schar);
> 
>    return ptr;}
> 
> The issue lies in the expression ptr + schar when schar is set to
> UINTMAX_MAX (18446744073709551615 on 64-bit systems). This triggers integer
> wraparound due to size_t arithmetic, causing the calculation to effectively
> become ptr - 1. As a result, the function returns a pointer that's one byte
> before the start of the allocated buffer.
> 
> The vulnerability is exploitable when:
> 
>     1. A user passes the key specification in traditional format (
>     +0.18446744073709551615R)
>     2. During command-line parsing in main(), this sets key->schar to
>     UINTMAX_MAX
>     3. In fillbuf(), the begfield() function is called to precompute key
>     positions
>     4. The underflow occurs during the line key pointer calculation
>     5. The function returns a pointer before the buffer start
>     6. This invalid pointer is later passed through the call chain:
>        - keycompare() function assigns the pointer to texta
>        - When using -R (random sort), it calls compare_random()
>        - compare_random() calls xstrxfrm() with the invalid pointer
>        - xstrxfrm() calls strxfrm() on the out-of-bounds address
>        - strxfrm() attempts to read the byte before the buffer, triggering
>        the overflow
> 
> Technical Impact
> 
> This is a heap buffer overflow (read) that accesses memory one byte before
> an allocated buffer. The vulnerability could lead to program crashes and
> potentially information disclosure depending on the memory layout.
> Proof of Concept
> 
> The vulnerability can be reliably reproduced with this simple test case:
> 
> bash
> 
> # Create a test file with any contentecho -e "aa\nbb" > poc_input.txt
> # Execute vulnerable command (traditional key format + random sort option)
> ./sort +0.18446744073709551615R poc_input.txt
> 
> When compiled with AddressSanitizer, this command produces the following
> error:
> 
> [image: image.png]
> 
> The ASan output clearly shows that the issue is a READ one byte before a
> 672-byte heap-allocated region. The call stack confirms the path from
> begfield() through keycompare() and compare_random() to strxfrm().
> Proposed Fix
> 
> A proper fix would involve checking for integer overflow before performing
> the pointer arithmetic in begfield(). Here's a suggested fix:
> 
> c
> 
> /* Inside begfield() *//* Advance PTR by SCHAR (if possible), but no
> further than LIM. */if (schar > 0) {
>    /* Check if adding schar would overflow or wrap negatively */
>    if (SIZE_MAX - (uintptr_t)ptr < schar) {
>      /* If it would overflow, safest is to set to end of current segment */
>      ptr = lim;
>    } else {
>      ptr = MIN(lim, ptr + schar);
>    }} else {
>    /* Original behavior for schar == 0 */
>    ptr = MIN(lim, ptr + schar);}
> 
> This fix guards against the integer overflow by checking if ptr + schar
> would exceed the maximum representable size_t value, which indicates a
> wraparound would occur.
> Affected Versions
> 
> This vulnerability affects GNU Coreutils through at least version, and
> potentially earlier versions. I've confirmed the issue in the current
> development code.
> CVE Request
> 
> I would like to request a CVE for this vulnerability.
> 
> Thank you for your attention to this matter.
> 
> Sincerely, Mohamed Maatallah (@Zephkek)

Indeed. I introduced this in coreutils 7.2 (2009).
One can repro on Fedora for e.g. with:

_POSIX2_VERSION=200809 LC_ALL=C valgrind sort +0.18446744073709551615R poc_input.txt
==984625== Memcheck, a memory error detector
==984625== Using Valgrind-3.24.0 and LibVEX; rerun with -h for copyright info
==984625== Command: sort +0.18446744073709551615R poc_input.txt
==984625==
==984625== Invalid read of size 1

Going back to the more verbose code from coreutils 7.1 avoids the issue.
I'll test a bit more here and post a full patch in a while.

thanks!
Pádraig

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.