On Wed, May 05, 2021 at 05:10:33PM +0200, David Hildenbrand wrote: > On 05.05.21 15:24, Michal Hocko wrote: > > On Thu 29-04-21 14:25:17, David Hildenbrand wrote: > > > A driver might set a page logically offline -- PageOffline() -- and > > > turn the page inaccessible in the hypervisor; after that, access to page > > > content can be fatal. One example is virtio-mem; while unplugged memory > > > -- marked as PageOffline() can currently be read in the hypervisor, this > > > will no longer be the case in the future; for example, when having > > > a virtio-mem device backed by huge pages in the hypervisor. > > > > > > Some special PFN walkers -- i.e., /proc/kcore -- read content of random > > > pages after checking PageOffline(); however, these PFN walkers can race > > > with drivers that set PageOffline(). > > > > > > Let's introduce page_offline_(begin|end|freeze|unfreeze) for > > > synchronizing. > > > > > > page_offline_freeze()/page_offline_unfreeze() allows for a subsystem to > > > synchronize with such drivers, achieving that a page cannot be set > > > PageOffline() while frozen. > > > > > > page_offline_begin()/page_offline_end() is used by drivers that care about > > > such races when setting a page PageOffline(). > > > > > > For simplicity, use a rwsem for now; neither drivers nor users are > > > performance sensitive. > > > > Please add a note to the PageOffline documentation as well. While are > > adding the api close enough an explicit note there wouldn't hurt. > > Will do. > > > > > > Signed-off-by: David Hildenbrand <david@xxxxxxxxxx> > > > > As to the patch itself, I am slightly worried that other pfn walkers > > might be less tolerant to the locking than the proc ones. On the other > > hand most users shouldn't really care as they do not tend to touch the > > memory content and PageOffline check without any synchronization should > > be sufficient for those. Let's try this out and see where we get... > > My thinking. Users that actually read random page content (as discussed in > the cover letter) are > > 1. Hibernation > 2. Dumping (/proc/kcore, /proc/vmcore) > 3. Physical memory access bypassing the kernel via /dev/mem > 4. Live debug tools (kgdb) I think you can add 5. Very old drivers > Other PFN walkers really shouldn't (and don't) access random page content. > > Thanks! > > -- > Thanks, > > David / dhildenb > > -- Sincerely yours, Mike.
