Re: [PATCH 5/6] x86: kaslr: select memory region from e820 maps

[Kees Cook]

On Fri, Apr 26, 2013 at 2:51 PM, Yinghai Lu <yinghai@xxxxxxxxxx> wrote:
> On Fri, Apr 26, 2013 at 12:03 PM, Kees Cook <keescook@xxxxxxxxxxxx> wrote:
>> This chooses the largest contiguous RAM region for the KASLR offset
>> to live in.
>>
>> Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
>> ---
>> v2:
>>  - make sure to exclude e820 regions outside the 32-bit memory range.
>
> Do you need to execlude range that is used for initrd and possible
> command_line and boot_param ?

Yeah, and while doing a stress test here, I realized there's another
problem. In the original version of this, the stack and heap are set
up after relocation. In the C port, they're set up before, so there's
even more to avoid. To illustrate... here's a CONFIG_RELOCATABLE=n
boot:

LOAD_PHYS:0x0000000001000000
input:    0x0000000001dfe24d-0x00000000023db865
output:   0x0000000001000000-0x00000000023c98c0
heap:     0x00000000023e0740-0x00000000023e8740
stack:    0x00000000023ec698
chosen:   0x0000000001000000

(stack is just cheating and reporting sp in decompress_kernel)

And a CONFIG_RELOCATABLE=y and "noaslr" boot:

LOAD_PHYS:0x0000000001000000
input:    0x000000000108b25e-0x00000000016b3e96
output:   0x0000000000200000-0x00000000016a1db8
heap:     0x00000000016b9600-0x00000000016c1600
stack:    0x00000000016c5558
chosen:   0x0000000000200000

In that case, it's just so far under LOAD_PHYSICAL_START that it's
safe. But if KASLR picks an area overlapping input, heap, or stack
it's hosed. :)

-Kees

--
Kees Cook
Chrome OS Security
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/