On Thu, Apr 05, 2018 at 12:31:11AM -0700, Christoph Hellwig wrote: > On Thu, Apr 05, 2018 at 09:19:28AM +0200, Lars-Peter Clausen wrote: > > On 04/05/2018 08:31 AM, Kees Cook wrote: > > > On Wed, Apr 4, 2018 at 3:31 AM, Greg KH <gre...@linuxfoundation.org> > > > wrote: > > >> Lars-Peter Clausen (2): > > >> usb: gadget: ffs: Execute copy_to_user() with USER_DS set > > > > > > https://git.kernel.org/linus/4058ebf33cb0be88ca516f968eda24ab7b6b93e4 > > > > > > Isn't there a better way to do this without the set_fs() usage? We've > > > been try to eliminate it in the kernel. I thought there was a safer > > > way to use iters now? > > > > The problem is use_mm(). It needs to be accompanied with set_fs(DS_USER) to > > work reliably. This has simply been missing for this particular instance of > > use_mm(). > > To me it seems like use_mm() should do set_fs(USER_DS) and unuse_mm() > should do set_fs(KERNEL_DS) to get drivers outo of this mess. I'll see > if I can come up with patches for the next merge window.
Yes on the former, not quite on the latter (we need to go back to the state before use_mm()). Said that, drm users of use_mm() look rather fishy and might be worth a good look...
