[PATCH] Fix buffer underrun in i386-dis.c.

[PATCH] Fix buffer underrun in i386-dis.c.

[Roland McGrath]

When disassembling any instruction without a REX prefix, the print_insn
function touches all_prefixes[-1].  This is usually harmless in most
builds, because the word preceding all_prefixes will probably be the
last_seg_prefix variable and it was usually zero already.  But in some
kinds of builds, all buffer underruns are caught and cause a crash.

AFAICT the obvious local workaround is in fact the proper fix.  In the
similar cases nearby, there is a PREFIX_FOO bit in the "prefixes" bitmask
that guards use of last_foo_prefix.  But there is no such bit for the REX
prefixes.  We could test "rex != 0" instead, I suppose.

OK for trunk and binutils-2.24 branch and gdb-7.7 branch?


Thanks,
Roland


opcodes/
2014-01-09  Bradley Nelson  <bradnelson@google.com>
	    Roland McGrath  <mcgrathr@google.com>

	* i386-dis.c (print_insn): Do not touch all_prefixes[-1] when
	last_rex_prefix is -1.

--- a/opcodes/i386-dis.c
+++ b/opcodes/i386-dis.c
@@ -12645,7 +12645,7 @@ print_insn (bfd_vma pc, disassemble_info *info)
     }

   /* Check if the REX prefix is used.  */
-  if (rex_ignored == 0 && (rex ^ rex_used) == 0)
+  if (rex_ignored == 0 && (rex ^ rex_used) == 0 && last_rex_prefix >= 0)
     all_prefixes[last_rex_prefix] = 0;

   /* Check if the SEG prefix is used.  */

Re: [PATCH] Fix buffer underrun in i386-dis.c.

[H.J. Lu]

On Thu, Jan 9, 2014 at 10:26 AM, Roland McGrath <mcgrathr@google.com> wrote:
> When disassembling any instruction without a REX prefix, the print_insn
> function touches all_prefixes[-1].  This is usually harmless in most
> builds, because the word preceding all_prefixes will probably be the
> last_seg_prefix variable and it was usually zero already.  But in some
> kinds of builds, all buffer underruns are caught and cause a crash.
>
> AFAICT the obvious local workaround is in fact the proper fix.  In the
> similar cases nearby, there is a PREFIX_FOO bit in the "prefixes" bitmask
> that guards use of last_foo_prefix.  But there is no such bit for the REX
> prefixes.  We could test "rex != 0" instead, I suppose.
>
> OK for trunk and binutils-2.24 branch and gdb-7.7 branch?

OK for trunk and binutils-2.24 branch.

Thanks.

>
> Thanks,
> Roland
>
>
> opcodes/
> 2014-01-09  Bradley Nelson  <bradnelson@google.com>
>             Roland McGrath  <mcgrathr@google.com>
>
>         * i386-dis.c (print_insn): Do not touch all_prefixes[-1] when
>         last_rex_prefix is -1.
>
> --- a/opcodes/i386-dis.c
> +++ b/opcodes/i386-dis.c
> @@ -12645,7 +12645,7 @@ print_insn (bfd_vma pc, disassemble_info *info)
>      }
>
>    /* Check if the REX prefix is used.  */
> -  if (rex_ignored == 0 && (rex ^ rex_used) == 0)
> +  if (rex_ignored == 0 && (rex ^ rex_used) == 0 && last_rex_prefix >= 0)
>      all_prefixes[last_rex_prefix] = 0;
>
>    /* Check if the SEG prefix is used.  */



-- 
H.J.

Re: [PATCH] Fix buffer underrun in i386-dis.c.

[Pedro Alves]

On 01/09/2014 06:31 PM, H.J. Lu wrote:
> On Thu, Jan 9, 2014 at 10:26 AM, Roland McGrath <mcgrathr@google.com> wrote:
>> When disassembling any instruction without a REX prefix, the print_insn
>> function touches all_prefixes[-1].  This is usually harmless in most
>> builds, because the word preceding all_prefixes will probably be the
>> last_seg_prefix variable and it was usually zero already.  But in some
>> kinds of builds, all buffer underruns are caught and cause a crash.
>>
>> AFAICT the obvious local workaround is in fact the proper fix.  In the
>> similar cases nearby, there is a PREFIX_FOO bit in the "prefixes" bitmask
>> that guards use of last_foo_prefix.  But there is no such bit for the REX
>> prefixes.  We could test "rex != 0" instead, I suppose.
>>
>> OK for trunk and binutils-2.24 branch and gdb-7.7 branch?
> 
> OK for trunk and binutils-2.24 branch.

In that case, OK for gdb-7.7 too.

-- 
Pedro Alves

Re: [PATCH] Fix buffer underrun in i386-dis.c.

[Roland McGrath]

Committed all around.

Thanks,
Roland