Re: [PATCH-tip v4] sched: Fix NULL user_cpus_ptr check in dup_user_cpus_ptr()

[Will Deacon <will@kernel.org>]

On Tue, Nov 29, 2022 at 11:03:35AM -0500, Waiman Long wrote:
> On 11/29/22 10:57, Will Deacon wrote:
> > On Tue, Nov 29, 2022 at 10:32:49AM -0500, Waiman Long wrote:
> > > On 11/29/22 09:07, Will Deacon wrote:
> > > > On Mon, Nov 28, 2022 at 10:11:52AM -0500, Waiman Long wrote:
> > > > > On 11/28/22 07:00, Will Deacon wrote:
> > > > > > On Sun, Nov 27, 2022 at 08:43:27PM -0500, Waiman Long wrote:
> > > > > > > On 11/24/22 21:39, Waiman Long wrote:
> > > > > > > > In general, a non-null user_cpus_ptr will remain set until the task dies.
> > > > > > > > A possible exception to this is the fact that do_set_cpus_allowed()
> > > > > > > > will clear a non-null user_cpus_ptr. To allow this possible racing
> > > > > > > > condition, we need to check for NULL user_cpus_ptr under the pi_lock
> > > > > > > > before duping the user mask.
> > > > > > > > 
> > > > > > > > Fixes: 851a723e45d1 ("sched: Always clear user_cpus_ptr in do_set_cpus_allowed()")
> > > > > > > > Signed-off-by: Waiman Long <longman@redhat.com>
> > > > > > > This is actually a pre-existing use-after-free bug since commit 07ec77a1d4e8
> > > > > > > ("sched: Allow task CPU affinity to be restricted on asymmetric systems").
> > > > > > > So it needs to be fixed in the stable release as well. Will resend the patch
> > > > > > > with an additional fixes tag and updated commit log.
> > > > > > Please can you elaborate on the use-after-free here? Looking at
> > > > > > 07ec77a1d4e8, the mask is only freed in free_task() when the usage refcount
> > > > > > has dropped to zero and I can't see how that can race with fork().
> > > > > > 
> > > > > > What am I missing?
> > > > > I missed that at first. The current task cloning process copies the content
> > > > > of the task structure over to the newly cloned/forked task. IOW, if
> > > > > user_cpus_ptr had been set up previously, it will be copied over to the
> > > > > cloned task. Now if user_cpus_ptr of the source task is cleared right after
> > > > > that and before dup_user_cpus_ptr() is called. The obsolete user_cpus_ptr
> > > > > value in the cloned task will remain and get used even if it has been freed.
> > > > > That is what I call as use-after-free and double-free.
> > > > If the parent task can be modified concurrently with dup_task_struct() then
> > > > surely we'd have bigger issues because that's not going to be atomic? At the
> > > > very least we'd have a data race, but it also feels like we could end up
> > > > with inconsistent task state in the child. In fact, couldn't the normal
> > > > 'cpus_mask' be corrupted by a concurrent set_cpus_allowed_common()?
> > > > 
> > > > Or am I still failing to understand the race?
> > > > 
> > > A major difference between cpus_mask and user_cpus_ptr is that for
> > > cpus_mask, the bitmap is embedded into task_struct whereas user_cpus_ptr is
> > > a pointer to an external bitmap. So there is no issue of use-after-free wrt
> > > cpus_mask. That is not the case where the memory of the user_cpus_ptr of the
> > > parent task is freed, but then a reference to that memory is still available
> > > in the child's task struct and may be used.
> > Sure, I'm not saying there's a UAF on cpus_mask, but I'm concerned that we
> > could corrupt the data and end up with an affinity mask that doesn't correspond
> > to anything meaningful. Do you agree that's possible?
> That is certainly possible. So we have to be careful about it.

Hmm, but we're not being particularly careful, are we? I hacked memcpy()
to be byte-to-byte to make things a bit easier to reproduce, and sure enough
I can race a bog-standard sched_setaffinity() call w/ fork():

[ 1663.935258] BUG: KCSAN: data-race in arch_dup_task_struct+0x4c/0x224
[ 1663.936872]
[ 1663.937292] race at unknown origin, with read to 0xffff06a44b8880a9 of 1 bytes by task 351 on cpu 0:
[ 1663.938770]  arch_dup_task_struct+0x4c/0x224
[ 1663.939621]  dup_task_struct+0x68/0x2a8
[ 1663.940381]  copy_process+0x208/0x1404
[ 1663.941109]  kernel_clone+0xdc/0x2c8
[ 1663.941814]  __arm64_sys_clone+0x9c/0xd4
[ 1663.942909]  invoke_syscall+0x54/0x170
[ 1663.943816]  el0_svc_common+0x100/0x148
[ 1663.944607]  do_el0_svc+0x40/0x10c
[ 1663.945333]  el0_svc+0x2c/0x7c
[ 1663.946006]  el0t_64_sync_handler+0x84/0xf0
[ 1663.946804]  el0t_64_sync+0x18c/0x190

I then managed to get the child process to run with an affinity mask (i.e.
'task_struct::cpus_mask') of *zero*, which triggers the select_fallback_rq()
logic:

 | process 14622 (waiman) no longer affine to cpu0

So sure, it's not a UAF, but I still think it's an issue that should be
fixed.

Will