linux-kernel - Re: [PATCH v1 09/10] lkdtm: Fix lkdtm_EXEC

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <c87e4e91-ee66-77b8-863a-27a34f41623e@csgroup.eu>
Date:   Wed, 13 Oct 2021 14:45:11 +0200
From:   Christophe Leroy <christophe.leroy@...roup.eu>
To:     Kees Cook <keescook@...omium.org>
Cc:     Benjamin Herrenschmidt <benh@...nel.crashing.org>,
        Paul Mackerras <paulus@...ba.org>,
        Michael Ellerman <mpe@...erman.id.au>,
        Andrew Morton <akpm@...ux-foundation.org>,
        "James E.J. Bottomley" <James.Bottomley@...senpartnership.com>,
        Helge Deller <deller@....de>, Arnd Bergmann <arnd@...db.de>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        linux-kernel@...r.kernel.org, linuxppc-dev@...ts.ozlabs.org,
        linux-ia64@...r.kernel.org, linux-parisc@...r.kernel.org,
        linux-arch@...r.kernel.org, linux-mm@...ck.org
Subject: Re: [PATCH v1 09/10] lkdtm: Fix lkdtm_EXEC_RODATA()



Le 13/10/2021 à 09:48, Christophe Leroy a écrit :
> 
> 
> Le 13/10/2021 à 09:39, Christophe Leroy a écrit :
>>
>>
>> Le 13/10/2021 à 09:23, Kees Cook a écrit :
>>> On Mon, Oct 11, 2021 at 05:25:36PM +0200, Christophe Leroy wrote:
>>>> Behind a location, lkdtm_EXEC_RODATA() executes a real function,
>>>> not a copy of do_nothing().
>>>>
>>>> So do it directly instead of using execute_location().
>>>>
>>>> And fix displayed addresses by dereferencing the function descriptors.
>>>>
>>>> Signed-off-by: Christophe Leroy <christophe.leroy@...roup.eu>
>>>> ---
>>>>   drivers/misc/lkdtm/perms.c | 9 ++++++++-
>>>>   1 file changed, 8 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/misc/lkdtm/perms.c b/drivers/misc/lkdtm/perms.c
>>>> index 442d60ed25ef..da16564e1ecd 100644
>>>> --- a/drivers/misc/lkdtm/perms.c
>>>> +++ b/drivers/misc/lkdtm/perms.c
>>>> @@ -153,7 +153,14 @@ void lkdtm_EXEC_VMALLOC(void)
>>>>
>>>>   void lkdtm_EXEC_RODATA(void)
>>>>   {
>>>> -    execute_location(lkdtm_rodata_do_nothing, CODE_AS_IS);
>>>> +    pr_info("attempting ok execution at %px\n",
>>>> +        dereference_symbol_descriptor(do_nothing));
>>>> +    do_nothing();
>>>> +
>>>> +    pr_info("attempting bad execution at %px\n",
>>>> +        dereference_symbol_descriptor(lkdtm_rodata_do_nothing));
>>>> +    lkdtm_rodata_do_nothing();
>>>> +    pr_err("FAIL: func returned\n");
>>>>   }
>>>
>>> (In re-reading this more carefully, I see now why kallsyms.h is used
>>> earlier: _function_ vs _symbol_ descriptor.)
>>>
>>> In the next patch:
>>>
>>> static noinline void execute_location(void *dst, bool write)
>>> {
>>> ...
>>>         func = setup_function_descriptor(&fdesc, dst);
>>>         if (IS_ERR(func))
>>>                 return;
>>>
>>>         pr_info("attempting bad execution at %px\n", dst);
>>>         func();
>>>         pr_err("FAIL: func returned\n");
>>> }
>>>
>>> What are the conditions for which dereference_symbol_descriptor works
>>> but dereference _function_descriptor doesn't?
>>>
>>
>> When LKDTM is built as a module I guess ?
>>
> 
> To be more precise, dereference_symbol_descriptor() calls either 
> dereference_kernel_function_descriptor() or 
> dereference_module_function_descriptor()
> 
> Both functions call dereference_function_descriptor() after checking 
> that we want to dereference something that is in the OPD section.
> 
> If we call dereference_function_descriptor() directly instead of 
> dereference_symbol_descriptor() we skip the range verification and may 
> dereference something that is not a function descriptor.
> 
> Should we do that ?
> 

Indeed we are using it only for well known functions so using 
dereference_function_descriptor() is good enough. I'll use that in v2.