On 05/29/2009 06:10 PM, Herbert Xu wrote:
> On Fri, May 29, 2009 at 11:32:54AM -0400, Jarod Wilson wrote:
>> At present, its entirely possible to add a test vector to testmgr with
>> an input longer than a page in length w/o specifying a .np option, and
>> overflow the page of memory allocated to {a,}xbuf[0], silently
>> corrupting memory. I know, because I've accidentally done it. :)
>>
>> While this doesn't currently happen in practice w/the existing code,
>> due to all !np vectors being less than a 4k page in length (and the
>> page allocation loop often returns contiguous pages anyway), explicit
>> checks or a way to remove the 4k limit would be a good idea.
>>
>> A few ways to fix and/or work around this:
>>
>> 1) allocate some larger guaranteed contiguous buffers using
>> __get_free_pages() or kmalloc and use them in the !np case
>>
>> 2) catch the > PAGE_SIZE && !np case and then do things similar to how
>> they are done in the np case
>>
>> 3) catch the > PAGE_SIZE && !np case and simply exit with an error
>>
>> Since there currently aren't any test vectors that are actually larger
>> than a page and not tagged np, option 1 seems like a waste of memory
>> and option 2 sounds like unnecessary complexity, so I'd offer up
>> option 3 as the most viable alternative right now.
>>
>> Signed-off-by: Jarod Wilson <jarod@xxxxxxxxxx>
>
> I just posted exactly the same thing yesterday :)
Oh, haha, serves me right for not looking first... Your variant seems to
be a bit more complete too, as I didn't look at any of the possible cases
where there might be overflows when using scatterlists. Cool, worksforme!
Thanks much,
--
Jarod Wilson
jarod@xxxxxxxxxx
--
To unsubscribe from this list: send the line "unsubscribe linux-crypto" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
