On Thu, 2019-08-15 at 10:49 +0200, Greg Kroah-Hartman wrote:
On Wed, Aug 14, 2019 at 07:17:07PM +0200, Knut Omang wrote:
I notice the discussion and your response here: http://linux-kernel.2935.n7.nabble.com/debugfs-and-module-unloading-td865175... I assume that means that protection against module unload while a debugfs file is open is now safe.
It should be, if you set the *owner field of your file_operations properly. Try it and see!
Might be a case for a KTF selftest to play with the timing to increase the chance :) Wasn't able to make it crash with these simple, short files.
I notice I had set the .owner field correctly myself in that driver code I referred to, so that's a "copy regression".
On older kernels, having this code in place is far better than an unprotected debugfs entry/exit - I have tested it extensively in the past :-)
Yes, it seems to work, but again, it really is racy and will fail. Please don't use it.
I perfectly agree with you that reducing the hole for a race condition is generally a bad idea, but from the above mail thread it seems that's the only available choice for older kernels?
I have no idea, but please, do not use that pattern of code as it is racy in all kernels, from all of time.
Ok, will remove it :-)
I tried in vain to find the commit from Al Viro that made the code safe, to identify which kernels that are safe from this issue, but he has a **lot** of commits, do you have a clue for what/where to look?
It will be good to have a mention/comment on this for future reference, like the earliest kernel version where this is safe.
Maybe we can even get rid of some more of the remaining of these too.. (I notice there's 65 cases of 'if (!try_module_get(THIS_MODULE))' right now)