On Mon, Jul 10, 2023 at 07:43:01AM +0200, Michal Hocko wrote: > On Fri 07-07-23 11:56:49, Luis Chamberlain wrote: > > On Mon, Jul 03, 2023 at 03:47:22PM +0200, Michal Hocko wrote: > > > On Fri 30-06-23 16:05:33, Luis Chamberlain wrote: > > > [...] > > > > What prevents code from racing the free with a random module_put() > > > > called by some other piece of code? > > > > > > Wouldn't be ref count a garbage already? How can you race when freeing > > > if module_put fail? > > > > It could yes, ie, so this risks at all being junk. > > Could you be more specific please? I still do not see a scenario where > module string name would be junk while refcount itself would be a valid > memory. That is true, but if refcount is invalid so will the memory for the string. > > So best IMHO is > > to tidy up all the get / puts and add respective tests to fix all > > this mess with proper messages as needed. My cursory review of the > > refcnt stuf is I see some races possible. > > It would likely be better to use refcount_t instead of atomic_t. Patches welcomed. > > While I'd be happy to help debugging aids, adding accesses to random > > memory for a string seems more risk prone. > > If there is really a scenario when module could be unloaded leaving > dangling struct module behind then we have a real problem as this is > exported to userspace IIRC. Not to mention module_get/put calls > modifying memory (UAF). That doesn't mean issues could not exist, given its all protected under privileged execution. All I'm suggesting is I look at this code and don't trust it, and think it could use some love. The selftests for kmod could be used to stress test but also stress-ng now also has module load and unloading so if there are races we can likely exploit them with either the kmod selftest or stress-ng module loading. Luis
