Re: [PATCH v2 2/3] binder: do not initialize locals passed to copy_from_user()

[Rasmus Villemoes <linux@rasmusvillemoes.dk>]

On 02/03/2020 19.31, Jann Horn wrote:
> On Mon, Mar 2, 2020 at 7:17 PM Alexander Potapenko <glider@google.com> wrote:
>> On Mon, Mar 2, 2020 at 3:00 PM Joe Perches <joe@perches.com> wrote:
>>>
>>> So?  CONFIG_INIT_STACK_ALL by design slows down code.
>> Correct.
>>
>>> This marking would likely need to be done for nearly all
>>> 3000+ copy_from_user entries.
>> Unfortunately, yes. I was just hoping to do so for a handful of hot
>> cases that we encounter, but in the long-term a compiler solution must
>> supersede them.
>>
>>> Why not try to get something done on the compiler side
>>> to mark the function itself rather than the uses?
>> This is being worked on in the meantime as well (see
>> http://lists.llvm.org/pipermail/cfe-dev/2020-February/064633.html)
>> Do you have any particular requisitions about how this should look on
>> the source level?
> 
> Just thinking out loud: Should this be a function attribute, or should
> it be a builtin - something like __builtin_assume_initialized(ptr,
> len)? That would make it also work for macros,

But with macros (and static inlines), the compiler sees all the
initialization being done, no?

and it might simplify
> the handling of inlining in the compiler. And you wouldn't need such a
> complicated attribute that refers to function arguments by index and
> such.

Does copy_from_user guarantee to zero-initialize the remaining buffer if
copying fails partway through? Otherwise it will be hard for the
compiler to make use of an annotation such as __assume_initialized(buf,
size - ret_from_cfu) - it will have to say "ok, the caller is bailing
out unless ret_from_cfu is 0, and in that case, yes, the whole local
struct variable is indeed initialized". And we can't make the annotation
unconditionally __assume_initialized(buf, size) [unless c_f_u comes with
that guarantee] because we don't know that all callers of c_f_u() bail
out on non-zero.


Somewhat related: I've long wanted a bunch of function attributes

  __may_read(ptr, bytes)
  __may_write(ptr, bytes)
  __will_write(ptr, bytes)

The first could be used to warn about passing an uninitialized or
too-small buffer (e.g.

  struct pollfd fds[4];
  poll(fds, sizeof(fds), ...) // whoops, should have been ARRAY_SIZE)

the second also for warning about a too-small buffer, and the third
would essentially be the same as __assume_initializes. Perhaps with some
sanitization option the compiler could also instrument the function
definition to not read/write beyond the area declared via those attributes.

But the attribute syntax doesn't currently allow complex expressions in
terms of the parameter names; I'd want to annotate poll as

  int poll(struct pollfd *fds, nfds_t nfds, int to) __may_rw(fds, nfds *
sizeof(*fds))

Rasmus