Re: [PATCH] security: CONFIG_HARDENED_USERCOPY does not need to select BUG

[Kamal Mostafa]

On Fri, Jun 29, 2018 at 01:27:08PM -0700, Kees Cook wrote:
> On Fri, Jun 29, 2018 at 1:04 PM, Kamal Mostafa <kamal@xxxxxxxxxxxxx> wrote:
> > Allows for CONFIG_HARDENED_USERCOPY without CONFIG_BUG.
> >
> > Signed-off-by: Kamal Mostafa <kamal@xxxxxxxxxxxxx>
> > ---
> >  security/Kconfig | 1 -
> >  1 file changed, 1 deletion(-)
> >
> > diff --git a/security/Kconfig b/security/Kconfig
> > index c430206..7667774 100644
> > --- a/security/Kconfig
> > +++ b/security/Kconfig
> > @@ -153,7 +153,6 @@ config HAVE_HARDENED_USERCOPY_ALLOCATOR
> >  config HARDENED_USERCOPY
> >         bool "Harden memory copies between kernel and userspace"
> >         depends on HAVE_HARDENED_USERCOPY_ALLOCATOR
> > -       select BUG
> >         imply STRICT_DEVMEM
> >         help
> >           This option checks for obviously wrong memory regions when
> 
> Do the lkdtm tests for usercopy correctly halt the kernel thread if
> CONFIG_BUG is removed?
> 

Yes, they do...

With this config (specifically disabling 'FALLBACK'):

    CONFIG_HARDENED_USERCOPY=y
    # CONFIG_HARDENED_USERCOPY_FALLBACK is not set
    # CONFIG_BUG is not set

I ran the usercopy tests as follows:

    modprobe lkdtm
    cd /sys/kernel/debug/provoke-crash
    cat DIRECT | grep USERCOPY | while read x ; do echo $x | tee DIRECT || echo $?; done

Resulting in this command line output:

    USERCOPY_HEAP_SIZE_TO
    139
    USERCOPY_HEAP_SIZE_FROM
    139
    USERCOPY_HEAP_WHITELIST_TO
    139
    USERCOPY_HEAP_WHITELIST_FROM
    139
    USERCOPY_STACK_FRAME_TO
    139
    USERCOPY_STACK_FRAME_FROM
    139
    USERCOPY_STACK_BEYOND
    139
    USERCOPY_KERNEL
    139

Each test case yields kernel log output like:

    lkdtm: Performing direct entry USERCOPY_HEAP_SIZE_TO
    lkdtm: attempting good copy_to_user of correct size
    lkdtm: attempting bad copy_to_user of too large size
    usercopy: Kernel memory exposure attempt detected from SLUB object 'kmalloc-1024' (offset 16, size 1024)!
    invalid opcode: 0000 [#17] SMP PTI
    ... {panic dump} ...

Each 'tee' gets terminated with SIGSEGV and no instances appear of the
lkdtm/usercopy.c warning "copy_{to/from}_user failed, but lacked Oops".

If I leave CONFIG_HARDENED_USERCOPY_FALLBACK=y then the pair of
WHITELIST tests don't trigger a panic or SIGSEGV, as expected.

 -Kamal