On Tue, 2023-11-21 at 13:55 -0800, Andrii Nakryiko wrote: [...] > > So, assuming that this tests (and a few others) are sane, Andrii's suggestion > > of calling grow_stack_state()/update_stack_depth() in > > check_stack_access_within_bounds() does not immediately work: doing so > > would change > > the behavior in check_stack_range_initialized() and allow the access. > > > > On the other hand, perhaps the test is not sane and the access should be > > permitted, in the spirit of allowing reads of uninitialized stack? Perhaps the > > different treatment of slots beyond state->allocated_stack and STACK_INVALID > > yes, I think this divergence is not intentional, but maybe Eduard > remembers some other quirks and why it is what it is, let's see. Yes, this is probably an overlook from my side. I should have allowed reads beyond allocated stack in this case when doing changes for STACK_INVALID. Sorry for delayed response. [...]
