[RFC/PATCH RESEND -next 20/21] fs: dcache: manually unpoison dname after allocation to shut up kasan's reports

[Dmitry Vyukov]

On Tue, Jul 15, 2014 at 1:34 PM, Andrey Ryabinin <a.ryabinin at samsung.com> wrote:
> On 07/15/14 10:12, Joonsoo Kim wrote:
>> On Wed, Jul 09, 2014 at 03:30:14PM +0400, Andrey Ryabinin wrote:
>>> We need to manually unpoison rounded up allocation size for dname
>>> to avoid kasan's reports in __d_lookup_rcu.
>>> __d_lookup_rcu may validly read a little beyound allocated size.
>>
>> If it read a little beyond allocated size, IMHO, it is better to
>> allocate correct size.
>>
>> kmalloc(name->len + 1, GFP_KERNEL); -->
>> kmalloc(roundup(name->len + 1, sizeof(unsigned long ), GFP_KERNEL);
>>
>> Isn't it?
>>
>
> It's not needed here because kmalloc always roundup allocation size.
>
> This out of bound access happens in dentry_string_cmp() if CONFIG_DCACHE_WORD_ACCESS=y.
> dentry_string_cmp() relays on fact that kmalloc always round up allocation size,
> in other words it's by design.
>
> That was discussed some time ago here - https://lkml.org/lkml/2013/10/3/493.
> Since filesystem's maintainer don't want to add needless round up here, I'm not going to do it.
>
> I think this patch needs only more detailed description why we not simply allocate more.
> Also I think it would be better to rename unpoisoin_shadow to something like kasan_mark_allocated().


Note that this poison/unpoison functionality can be used in other
contexts. E.g. when you allocate a bunch of pages, then at some point
poison a part of it to ensure that nobody touches it, then unpoison it
back. Allocated/unallocated looks like a bad fit here, because it has
nothing to do with allocation state. Poison/unpoison is also what we
use in user-space.