lkml.org 
[lkml]   [2017]   [Jul]   [18]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [PATCH v2 2/8] exec: Move security_bprm_secureexec() earlier
On Mon, Jul 10, 2017 at 7:07 PM, Kees Cook <keescook@chromium.org> wrote:
> On Mon, Jul 10, 2017 at 10:18 AM, Eric W. Biederman
> <ebiederm@xmission.com> wrote:
>> Kees Cook <keescook@chromium.org> writes:
>>
>>> On Mon, Jul 10, 2017 at 1:57 AM, Eric W. Biederman
>>> <ebiederm@xmission.com> wrote:
>>>> Kees Cook <keescook@chromium.org> writes:
>>>>
>>>>> There are several places where exec needs to know if a privilege-gain has
>>>>> happened. These should be using the results of security_bprm_secureexec()
>>>>> but it is getting (needlessly) called very late.
>>>>
>>>> It is hard to tell at a glance but I believe this introduces a
>>>> regression.
>>>>
>>>> cap_bprm_set_creds is currently called before cap_bprm_secureexec and
>>>> it has a number of cases such as no_new_privs and ptrace that can result
>>>> in some of the precomputed credential changes not happening.
>>>>
>>>> Without accounting for that I believe your cap_bprm_securexec now
>>>> returns a postive value too early.
>>>
>>> It's still before cap_bprm_secureexec. cap_brpm_set_creds() is in
>>> prepare_binprm(), which is well before exec_binprm() and it's eventual
>>> call to setup_new_exec().
>>
>> Good point. I didn't double check and the set in the name had me
>> thinking it was setting the creds on current.
>>
>> Is there any reason we need a second security hook? It feels like we
>> should be able to just fold the secureexec hook into the set_creds hook.
>>
>> The two are so interrelated I fear that having them separate only
>> encourages them to diverge in trivial ways as it is easy to forget about
>> some detail or other.
>>
>> Certainly having them called from different functions seems wrong. If
>> we know enough in prepare_binprm we know enough.
>
> Hmmm, yes. That would let us have the secureexec-ness knowledge before
> copy_strings(), in case we ever need to make that logic
> secureexec-aware.
>
> I'll dig through the LSMs to examine the set_creds hooks to see if
> this could be possible.

So, yes, after digging, this is very possible. In fact, it's highly
desirable. Both commoncaps and AppArmor save state into the bprm
struct between bprm_set_creds and bprm_secureexec explicitly to return
a sane value from bprm_secureexec. (And Smack and SELinux both have
trivial tests that just repeat from bprm_set_creds.)

I've reworked the series to just remove bprm_secureexec entirely. It
comes out nicely, removing more than it adds:

14 files changed, 54 insertions(+), 144 deletions(-)

I'll send the patches in the morning (perhaps to go through -mm since
it touches fs/exec.c, binfmt_elf*.c, and security/).

-Kees

--
Kees Cook
Pixel Security

\
 
 \ /
  Last update: 2017-07-18 08:45    [W:0.065 / U:0.396 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site