Re: [PATCH v6 7/9] kfence, Documentation: add KFENCE documentation

[Alexander Potapenko <glider@google.com>]

On Fri, Oct 30, 2020 at 3:50 AM Jann Horn <jannh@google.com> wrote:
>
> On Thu, Oct 29, 2020 at 2:17 PM Marco Elver <elver@google.com> wrote:
> > Add KFENCE documentation in dev-tools/kfence.rst, and add to index.
> [...]
> > +The KFENCE memory pool is of fixed size, and if the pool is exhausted, no
> > +further KFENCE allocations occur. With ``CONFIG_KFENCE_NUM_OBJECTS`` (default
> > +255), the number of available guarded objects can be controlled. Each object
> > +requires 2 pages, one for the object itself and the other one used as a guard
> > +page; object pages are interleaved with guard pages, and every object page is
> > +therefore surrounded by two guard pages.
> > +
> > +The total memory dedicated to the KFENCE memory pool can be computed as::
> > +
> > +    ( #objects + 1 ) * 2 * PAGE_SIZE
>
> Plus memory overhead from shattered hugepages. With the default object
> count, on x86, we allocate 2MiB of memory pool, but if we have to
> shatter a 2MiB hugepage for that, we may cause the allocation of one
> extra page table, or 4KiB. Of course that's pretty much negligible.
> But on arm64 it's worse, because there we have to disable hugepages in
> the linear map completely. So on a device with 4GiB memory, we might
> end up with something on the order of 4GiB/2MiB * 0x1000 bytes = 8MiB
> of extra L1 page tables that wouldn't have been needed otherwise -
> significantly more than the default memory pool size.

Note that with CONFIG_RODATA_FULL_DEFAULT_ENABLED (which is on by
default now) these hugepages are already disabled (see patch 3/9)

> If the memory overhead is documented, this detail should probably be
> documented, too.

But, yes, documenting that also makes sense.

> > +Using the default config, and assuming a page size of 4 KiB, results in
> > +dedicating 2 MiB to the KFENCE memory pool.
> [...]
> > +For such errors, the address where the corruption as well as the invalidly
>
> nit: "the address where the corruption occurred" or "the address of
> the corruption"
>
> > +written bytes (offset from the address) are shown; in this representation, '.'
> > +denote untouched bytes. In the example above ``0xac`` is the value written to
> > +the invalid address at offset 0, and the remaining '.' denote that no following
> > +bytes have been touched. Note that, real values are only shown for
> > +``CONFIG_DEBUG_KERNEL=y`` builds; to avoid information disclosure for non-debug
> > +builds, '!' is used instead to denote invalidly written bytes.
> [...]
> > +KFENCE objects each reside on a dedicated page, at either the left or right
> > +page boundaries selected at random. The pages to the left and right of the
> > +object page are "guard pages", whose attributes are changed to a protected
> > +state, and cause page faults on any attempted access. Such page faults are then
> > +intercepted by KFENCE, which handles the fault gracefully by reporting an
> > +out-of-bounds access.
>
> ... and marking the page as accessible so that the faulting code can
> continue (wrongly) executing.
>
>
> [...]
> > +Interface
> > +---------
> > +
> > +The following describes the functions which are used by allocators as well page
>
> nit: "as well as"?
>
>
>
> > +handling code to set up and deal with KFENCE allocations.



-- 
Alexander Potapenko
Software Engineer

Google Germany GmbH
Erika-Mann-Straße, 33
80636 München

Geschäftsführer: Paul Manicle, Halimah DeLaine Prado
Registergericht und -nummer: Hamburg, HRB 86891
Sitz der Gesellschaft: Hamburg