Re: [PATCH 0/6] RFC: adding support to GCC for detecting trust boundaries

On Mon, 2021-12-06 at 13:40 -0600, Segher Boessenkool wrote:

On Mon, Dec 06, 2021 at 11:12:00AM -0700, Martin Sebor wrote:

quoted

On 11/13/21 1:37 PM, David Malcolm via Gcc-patches wrote:

quoted

Approach 1: Custom Address Spaces
=================================

GCC's C frontend supports target-specific address spaces; see:
  https://gcc.gnu.org/onlinedocs/gcc/Named-Address-Spaces.html
Quoting the N1275 draft of ISO/IEC DTR 18037:
  "Address space names are ordinary identifiers, sharing the same
name
  space as variables and typedef names.  Any such names follow the
same
  rules for scope as other ordinary identifiers (such as typedef
names).
  An implementation may provide an implementation-defined set of
  intrinsic address spaces that are, in effect, predefined at the
start
  of every translation unit.  The names of intrinsic address spaces
must
  be reserved identifiers (beginning with an underscore and an
uppercase
  letter or with two underscores).  An implementation may also
  optionally support a means for new address space names to be
defined
  within a translation unit."

Patch 1a in the following patch kit for GCC implements such a means
to
define new address spaces names in a translation unit, via a
pragma:
  #prgama GCC custom_address_space(NAME_OF_ADDRESS_SPACE)

For example, the Linux kernel could perhaps write:

  #define __kernel
  #pragma GCC custom_address_space(__user)
  #pragma GCC custom_address_space(__iomem)
  #pragma GCC custom_address_space(__percpu)
  #pragma GCC custom_address_space(__rcu)

and thus the C frontend can complain about code that mismatches
__user
and kernel pointers, e.g.:

custom-address-space-1.c: In function ‘test_argpass_to_p’:
custom-address-space-1.c:29:14: error: passing argument 1 of 
‘accepts_p’
from pointer to non-enclosed address space
   29 |   accepts_p (p_user);
      |              ^~~~~~
custom-address-space-1.c:21:24: note: expected ‘void *’ but
argument is
of type ‘__user void *’
   21 | extern void accepts_p (void *);
      |                        ^~~~~~
custom-address-space-1.c: In function ‘test_cast_k_to_u’:
custom-address-space-1.c:135:12: warning: cast to ‘__user’ address 
space
pointer from disjoint generic address space pointer
  135 |   p_user = (void __user *)p_kernel;
      |            ^

This seems like an excellent use of named address spaces :)

It has some big problems though.

Thanks for raising these points.

Named address spaces are completely target-specific.  Defining them
with
a pragma like this does not allow you to set the pointer mode or
anything related to a custom LEGITIMATE_ADDRESS_P.

My thinking was that each custom address space is based on an existing
address space, but is disjoint from it, where "based on" means "what it
looks like in terms of RTL generation" (clearly I'm handwaving here).  

In patch 1a, the custom address spaces are all based on the generic
address space (but disjoint from it); syntax could be added to base
them on one of the target-specific address spaces.

  It does not allow
you to sayy zero pointers are invalid in some address spaces and not in
others.

Syntax could be added for this, I suppose.

You cannot provide any of the DWARF address space stuff this
way.

True.  I confess that I haven't thought about the debugging experience,
and I'd need to think what would happen at the DWARF level.

But most importantly, there are only four bits for the address
space field internally, and they are used by however a backend wants to
use them.

One of the ideas of patch 1a is to divide up this 4-bit space between
the target-specific and the custom address spaces.  The backend code
would need to be tweaked to decode the 4-bit value to get at the
underlying target-specific address space value.  This is done by the
function ensure_builtin_addr_space in patch 1a, though I've likely
missed some places.

IIRC, the target that's currently using the most address spaces is avr,
which I believe has 8 target-specific address spaces, in addition to
the generic one, i.e. 9 builtin address spaces, which would leave room
for up to 6 user-defined address spaces.  The Linux kernel's smatch
annotations currently effectively introduce 4 custom address spaces,
__user, __iomem, __percpu, and __rcu (assuming that __kernel is the
generic address space), so it's something of a tight squeeze, but it
does fit.  This doesn't account for out-of-tree backends, of course.

None of this cannot be solved, but all of it will have to be solved.

(nods)

IMO it will be best to not mix this with address spaces in the user
interface (it is of course fine to *implement* it like that, or with
big overlap at least).

I was thinking the other way around, in that it should look like
address spaces in terms of the user's source code, but has some
implementation differences.

quoted

quoted

The patch doesn't yet maintain a good distinction between implicit
target-specific address spaces and user-defined address spaces,

And that will have to be fixed in the user code syntax at least.

quoted

quoted

has at
least one known major bug, and has only been lightly tested.  I can
fix these issues, but was hoping for feedback that this approach is
the
right direction from both the GCC and Linux development
communities.

Allowing the user to define new address spaces does not jibe well with
how targets do (validly!) use them.

I think from a user's perspective it's a nice approach - my feeling is
that it makes certain things easier for the user, whilst complicating
things from a backend implementation perspective.

Plus you've raised various technical issues which I'd have to resolve
if we went in this direction.

quoted

quoted

Approach 2: An "untrusted" attribute
====================================

Alternatively, patch 1b in the kit implements:

  __attribute__((untrusted))

which can be applied to types as a qualifier (similarly to const,
volatile, etc) to mark a trust boundary, hence the kernel could
have:

  #define __user __attribute__((untrusted))

where my patched GCC treats
  T *
vs
  T __attribute__((untrusted)) *
as being different types and thus the C frontend can complain (even
without
-fanalyzer) about e.g.:

extern void accepts_p(void *);

void test_argpass_to_p(void __user *p_user)
{
  accepts_p(p_user);
}

untrusted-pointer-1.c: In function ‘test_argpass_to_p’:
untrusted-pointer-1.c:22:13: error: passing argument 1 of
‘accepts_p’
from pointer with different trust level
   22 |   accepts_p(p_user);
      |              ^~~~~~
untrusted-pointer-1.c:14:23: note: expected ‘void *’ but argument
is of
type ‘__attribute__((untrusted)) void *’
   14 | extern void accepts_p(void *);
      |                        ^~~~~~

So you'd get enforcement of __user vs non-__user pointers as part
of
GCC's regular type-checking.  (You need an explicit cast to convert
between the untrusted vs trusted types).

As with the named address space idea, this approach also looks
reasonable to me.  If you anticipate using the attribute only
in the analyzer I would suggest to consider introducing it in
the analyzer's namespace (e.g., analyzer::untrusted, or even
gnu::analyzer::untrusted).

I don't see any fundamental problems with this approach.  It also is
very much in line with how Perl handles this (and some copycat
languages
do as well), the "tainted" flag on data.

(nods)

quoted

quoted

This approach is much less expressive that the custom addres space
approach; it would only cover the trust boundary aspect; it
wouldn't
cover any differences between generic pointers and __user, vs
__iomem,
__percpu, and __rcu which I admit I only dimly understand.

Yes, it does not have any of the big problems that come with those
address spaces either!  :-)

Indeed - I think this "untrusted attribute" approach is much simpler
implementation-wise than the "custom address space" approach, which is
also in its favor.

I'm wondering if anyone from the kernel development community has
strong opinions here, since the custom address space approach is
potentially much more expressive.

Otherwise I think we're both preferring the "untrusted attribute"
approach (patch 1b).

quoted

quoted

Other attributes
================

Patch 2 in the kit adds:
  __attribute__((returns_zero_on_success))
and
  __attribute__((returns_nonzero_on_success))
as hints to the analyzer that it's worth bifurcating the analysis
of
such functions (to explore failure vs success, and thus to better
explore error-handling paths).  It's also a hint to the human
reader of
the source code.

I thing being able to express something along these lines would
be useful even outside the analyzer, both for warnings and, when
done right, perhaps also for optimization.  So I'm in favor of
something like this.  I'll just reiterate here the comment on
this attribute I sent you privately some time ago.

What is "success" though?  You probably want it so some checker can
make
sure you do handle failure some way, but how do you see what is
handling
failure and what is handling the successful case?

"success" and "failure" in this case are purely in terms of how we
label events for the user in the analyzer, such as in event (3) in the
following:

infoleak-antipatterns-1.c: In function ‘infoleak_stack_unchecked_err’:
infoleak-antipatterns-1.c:118:10: warning: potential exposure of
sensitive information by copying uninitialized data from stack across
trust boundary [CWE-200] [-Wanalyzer-exposure-through-uninit-copy]
  118 |   err |= copy_to_user (dst, &st, sizeof(st));
      |          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  ‘infoleak_stack_unchecked_err’: events 1-4
    |
    |  110 |   struct infoleak_buf st;
    |      |                       ^~
    |      |                       |
    |      |                       (1) source region created on stack here
    |      |                       (2) capacity: 256 bytes
    |......
    |  117 |   int err = copy_from_user (&st, src, sizeof(st));
    |      |             ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    |      |             |
    |      |             (3) when ‘copy_from_user’ fails, returning non-zero
    |  118 |   err |= copy_to_user (dst, &st, sizeof(st));
    |      |          ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    |      |          |
    |      |          (4) uninitialized data copied from stack here
    |

i.e. it's allows the analyzer to provide a hint to the reader of the
analyzer output.  The attribute is also a hint to the human reader of
the source code.

Thanks
Dave