[PATCH] doc: update 'unique identifiers'

[PATCH] doc: update 'unique identifiers'

[Tobin C. Harding]

Advice about what to use as a unique identifier is no longer valid since
patch series was merged to hash pointers printed with %p. We can use
this as a unique identifier now.

Signed-off-by: Tobin C. Harding <me@tobin.cc>
---
 Documentation/security/self-protection.rst | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/Documentation/security/self-protection.rst b/Documentation/security/self-protection.rst
index 60c8bd8b77bf..f10f47cad825 100644
--- a/Documentation/security/self-protection.rst
+++ b/Documentation/security/self-protection.rst
@@ -274,8 +274,8 @@ Unique identifiers
 ------------------
 
 Kernel memory addresses must never be used as identifiers exposed to
-userspace. Instead, use an atomic counter, an idr, or similar unique
-identifier.
+userspace. Printk specifier %p hashes addresses by default now and can be
+used as a unique identifier.
 
 Memory initialization
 ---------------------
-- 
2.7.4

Re: [PATCH] doc: update 'unique identifiers'

[Kees Cook]

On Mon, Dec 4, 2017 at 1:22 PM, Tobin C. Harding <me@tobin.cc> wrote:
> Advice about what to use as a unique identifier is no longer valid since
> patch series was merged to hash pointers printed with %p. We can use
> this as a unique identifier now.
>
> Signed-off-by: Tobin C. Harding <me@tobin.cc>

I don't agree: %p should still not be encouraged. Exposing an
identifier to userspace needs careful consideration, and atomics,
idrs, etc, continue to be a good recommendation here, as far as I'm
concerned.

-Kees

> ---
>  Documentation/security/self-protection.rst | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/Documentation/security/self-protection.rst b/Documentation/security/self-protection.rst
> index 60c8bd8b77bf..f10f47cad825 100644
> --- a/Documentation/security/self-protection.rst
> +++ b/Documentation/security/self-protection.rst
> @@ -274,8 +274,8 @@ Unique identifiers
>  ------------------
>
>  Kernel memory addresses must never be used as identifiers exposed to
> -userspace. Instead, use an atomic counter, an idr, or similar unique
> -identifier.
> +userspace. Printk specifier %p hashes addresses by default now and can be
> +used as a unique identifier.
>
>  Memory initialization
>  ---------------------
> --
> 2.7.4
>



-- 
Kees Cook
Pixel Security

Re: [PATCH] doc: update 'unique identifiers'

[Tobin C. Harding]

On Mon, Dec 04, 2017 at 01:28:45PM -0800, Kees Cook wrote:
> On Mon, Dec 4, 2017 at 1:22 PM, Tobin C. Harding <me@tobin.cc> wrote:
> > Advice about what to use as a unique identifier is no longer valid since
> > patch series was merged to hash pointers printed with %p. We can use
> > this as a unique identifier now.
> >
> > Signed-off-by: Tobin C. Harding <me@tobin.cc>
> 
> I don't agree: %p should still not be encouraged. Exposing an
> identifier to userspace needs careful consideration, and atomics,
> idrs, etc, continue to be a good recommendation here, as far as I'm
> concerned.

Ok no worries, so these docs are valid and current as is? I have no
agenda with this patch, just attempting to keep the docs in line with
the code :)

thanks,
Tobin.

Re: [PATCH] doc: update 'unique identifiers'

[Kees Cook]

On Mon, Dec 4, 2017 at 1:44 PM, Tobin C. Harding <me@tobin.cc> wrote:
> On Mon, Dec 04, 2017 at 01:28:45PM -0800, Kees Cook wrote:
>> On Mon, Dec 4, 2017 at 1:22 PM, Tobin C. Harding <me@tobin.cc> wrote:
>> > Advice about what to use as a unique identifier is no longer valid since
>> > patch series was merged to hash pointers printed with %p. We can use
>> > this as a unique identifier now.
>> >
>> > Signed-off-by: Tobin C. Harding <me@tobin.cc>
>>
>> I don't agree: %p should still not be encouraged. Exposing an
>> identifier to userspace needs careful consideration, and atomics,
>> idrs, etc, continue to be a good recommendation here, as far as I'm
>> concerned.
>
> Ok no worries, so these docs are valid and current as is? I have no
> agenda with this patch, just attempting to keep the docs in line with
> the code :)

I think a section could be added/updated discussing leaks and %p (in
that it is hashing now), that would be quite welcome!

I do, probably need to go through this document and update a few things.

-Kees

-- 
Kees Cook
Pixel Security

Re: [PATCH] doc: update 'unique identifiers'

[Tobin C. Harding]

On Mon, Dec 04, 2017 at 01:51:42PM -0800, Kees Cook wrote:
> On Mon, Dec 4, 2017 at 1:44 PM, Tobin C. Harding <me@tobin.cc> wrote:
> > On Mon, Dec 04, 2017 at 01:28:45PM -0800, Kees Cook wrote:
> >> On Mon, Dec 4, 2017 at 1:22 PM, Tobin C. Harding <me@tobin.cc> wrote:
> >> > Advice about what to use as a unique identifier is no longer valid since
> >> > patch series was merged to hash pointers printed with %p. We can use
> >> > this as a unique identifier now.
> >> >
> >> > Signed-off-by: Tobin C. Harding <me@tobin.cc>
> >>
> >> I don't agree: %p should still not be encouraged. Exposing an
> >> identifier to userspace needs careful consideration, and atomics,
> >> idrs, etc, continue to be a good recommendation here, as far as I'm
> >> concerned.
> >
> > Ok no worries, so these docs are valid and current as is? I have no
> > agenda with this patch, just attempting to keep the docs in line with
> > the code :)
> 
> I think a section could be added/updated discussing leaks and %p (in
> that it is hashing now), that would be quite welcome!
> 
> I do, probably need to go through this document and update a few things.

How about I do whatever generates the least amount of work for you. Is
it easier if I add the %p stuff for you to review or is it easier to
just leave it for you to do in your own time?

thanks,
Tobin.

Re: [PATCH] doc: update 'unique identifiers'

[Kees Cook]

On Mon, Dec 4, 2017 at 3:39 PM, Tobin C. Harding <me@tobin.cc> wrote:
> On Mon, Dec 04, 2017 at 01:51:42PM -0800, Kees Cook wrote:
>> On Mon, Dec 4, 2017 at 1:44 PM, Tobin C. Harding <me@tobin.cc> wrote:
>> > On Mon, Dec 04, 2017 at 01:28:45PM -0800, Kees Cook wrote:
>> >> On Mon, Dec 4, 2017 at 1:22 PM, Tobin C. Harding <me@tobin.cc> wrote:
>> >> > Advice about what to use as a unique identifier is no longer valid since
>> >> > patch series was merged to hash pointers printed with %p. We can use
>> >> > this as a unique identifier now.
>> >> >
>> >> > Signed-off-by: Tobin C. Harding <me@tobin.cc>
>> >>
>> >> I don't agree: %p should still not be encouraged. Exposing an
>> >> identifier to userspace needs careful consideration, and atomics,
>> >> idrs, etc, continue to be a good recommendation here, as far as I'm
>> >> concerned.
>> >
>> > Ok no worries, so these docs are valid and current as is? I have no
>> > agenda with this patch, just attempting to keep the docs in line with
>> > the code :)
>>
>> I think a section could be added/updated discussing leaks and %p (in
>> that it is hashing now), that would be quite welcome!
>>
>> I do, probably need to go through this document and update a few things.
>
> How about I do whatever generates the least amount of work for you. Is
> it easier if I add the %p stuff for you to review or is it easier to
> just leave it for you to do in your own time?

If you can write a section on %p leaks, that would be great!

I can clean up other things as work on top of that.

Thanks!

-Kees


>
> thanks,
> Tobin.



-- 
Kees Cook
Pixel Security