[Fixed Cc: address for stable, Cc'ed Juergen]
On Tue, 22 Aug 2017 14:04:42 -0700
Jacob Keller <jacob.e.keller@xxxxxxxxx> wrote:
> When responding to an affinity hint we directly copied a cpumask value,
> intsead of using cpumask_copy. According to cpumask.h this is not
> correct because cpumask_t is only guaranteed to have enough space for
> the number of CPUs in the system, and may not be as big as we expect.
> Thus a direct copy results in an out-of-bound read and potentially
> a crash if the pages are aligned just right. This will be easily
> detected on a kernel with KASAN enabled:
I still think commit message of my patch
(ae9c9586f61e914dc1c6fe2e6ac1fb2bf07283bc.1502792828.git.sbrivio@xxxxxxxxxx)
was perhaps a bit clearer, but okay, this is also clear, fair enough.
> KASAN reports:
> [ 25.242312] BUG: KASAN: slab-out-of-bounds in i40e_irq_affinity_notify+0x30/0x50 [i40e] at addr ffff880462eea960
[...]
> [ 25.242597] ==================================================================
This is also taken from my message, not terribly happy about it
(and still happier with it than without). Fair enough, whatever it
takes to get this applied as soon as possible...
> Fixes: 96db776a3682 ("i40e/i40evf: fix interrupt affinity bug", 2016-09-14)
> Signed-off-by: Jacob Keller <jacob.e.keller@xxxxxxxxx>
> Cc: stable@xxxxxxxxxxxxxxx # 4.10+
FWIW,
Acked-by: Stefano Brivio <sbrivio@xxxxxxxxxx>
--
Stefano