Re: sctp: kernel memory overwrite attempt detected in sctp_getsockopt_assoc_stats

[Dmitry Vyukov]

On Mon, Jan 16, 2017 at 2:57 PM, Neil Horman <nhorman@xxxxxxxxxxxxx> wrote:
> On Mon, Jan 16, 2017 at 08:11:40AM +0100, Dmitry Vyukov wrote:
>> On Sun, Jan 15, 2017 at 9:35 PM, Neil Horman <nhorman@xxxxxxxxxxxxx> wrote:
>> > On Sun, Jan 15, 2017 at 06:29:59PM +0100, Dmitry Vyukov wrote:
>> >> Hello,
>> >>
>> >> I've enabled CONFIG_HARDENED_USERCOPY_PAGESPAN on syzkaller fuzzer and
>> >> now I am seeing lots of:
>> >>
>> > If I'm not mistaken, its because thats specifically what that option does.  From
>> > the Kconfig:
>> > onfig HARDENED_USERCOPY_PAGESPAN
>> >         bool "Refuse to copy allocations that span multiple pages"
>> >         depends on HARDENED_USERCOPY
>> >         depends on EXPERT
>> >         help
>> >           When a multi-page allocation is done without __GFP_COMP,
>> >           hardened usercopy will reject attempts to copy it. There are,
>> >           however, several cases of this in the kernel that have not all
>> >           been removed. This config is intended to be used only while
>> >           trying to find such users.
>> >
>> > So, if the fuzzer does a setsockopt and the data it passes crosses a page
>> > boundary, it seems like this will get triggered.  Based on the fact that its
>> > only used to find users that do this, it seems like not the sort of thing that
>> > you want enabled while running a fuzzer that might do it arbitrarily.
>>
>>
>> The code also takes into account compound pages. As far as I
>> understand the intention of the check is to effectively find
>> out-of-bounds copies (e.g. goes beyond the current heap allocation). I
>> would expect that stacks are allocated as compound pages and don't
>> trigger this check. I don't see it is firing in other similar places.
>>
> Honestly, I'm not overly familiar with stack page allocation, at least not so
> far as compound vs. single page allocation is concerned.  I suppose the question
> your really asking here is: Have you found a case in which the syscall fuzzer
> has forced the allocation of an insecure non-compound page on the stack, or is
> this a false positive warning.  I can't provide the answer to that.


Yes. I added Kees, author of CONFIG_HARDENED_USERCOPY_PAGESPAN, to To line.
Kees, is this a false positive?