On Thu, Feb 4, 2021 at 5:13 PM Jason Gunthorpe jgg@ziepe.ca wrote:
On Wed, Feb 03, 2021 at 10:19:48PM +0100, Daniel Vetter wrote:
tldr; DMA buffers aren't normal memory, expecting that you can use them like that (like calling get_user_pages works, or that they're accounting like any other normal memory) cannot be guaranteed.
Since some userspace only runs on integrated devices, where all buffers are actually all resident system memory, there's a huge temptation to assume that a struct page is always present and useable like for any more pagecache backed mmap. This has the potential to result in a uapi nightmare.
To stop this gap require that DMA buffer mmaps are VM_SPECIAL, which blocks get_user_pages and all the other struct page based infrastructure for everyone. In spirit this is the uapi counterpart to the kernel-internal CONFIG_DMABUF_DEBUG.
Fast gup needs the special flag set on the PTE as well.. Feels weird to have a special VMA without also having special PTEs?
There's kinda no convenient & cheap way to check for the pte_special flag. This here should at least catch accidental misuse, people building their own ptes we can't stop. Maybe we should exclude VM_MIXEDMAP to catch vm_insert_page in one of these.
Hm looking at code I think we need to require VM_PFNMAP here to stop vm_insert_page. And looking at the various functions, that seems to be required (and I guess VM_IO is more for really funky architectures where io-space is somewhere else?). I guess I should check for VM_PFNMAP instead of VM_SPECIAL? -Daniel