Re: [PATCH v2 8/8] binder: use per-vma lock in page installation

[Carlos Llamas <cmllamas@google.com>]

On Thu, Nov 07, 2024 at 07:33:24PM +0000, Carlos Llamas wrote:
> On Thu, Nov 07, 2024 at 10:52:30AM -0800, Suren Baghdasaryan wrote:
> > On Thu, Nov 7, 2024 at 10:27 AM Suren Baghdasaryan <surenb@google.com> wrote:
> > >
> > > On Thu, Nov 7, 2024 at 10:19 AM Carlos Llamas <cmllamas@google.com> wrote:
> > > >
> > > > On Thu, Nov 07, 2024 at 10:04:23AM -0800, Suren Baghdasaryan wrote:
> > > > > On Thu, Nov 7, 2024 at 9:55 AM Carlos Llamas <cmllamas@google.com> wrote:
> > > > > > On Thu, Nov 07, 2024 at 08:16:39AM -0800, Suren Baghdasaryan wrote:
> > > > > > > On Wed, Nov 6, 2024 at 8:03 PM Carlos Llamas <cmllamas@google.com> wrote:
> > > > > > > > +static int binder_page_insert(struct binder_alloc *alloc,
> > > > > > > > +                             unsigned long addr,
> > > > > > > > +                             struct page *page)
> > > > > > > > +{
> > > > > > > > +       struct mm_struct *mm = alloc->mm;
> > > > > > > > +       struct vm_area_struct *vma;
> > > > > > > > +       int ret = -ESRCH;
> > > > > > > > +
> > > > > > > > +       if (!mmget_not_zero(mm))
> > > > > > > > +               return -ESRCH;
> > > > > > > > +
> > > > > > > > +       /* attempt per-vma lock first */
> > > > > > > > +       vma = lock_vma_under_rcu(mm, addr);
> > > > > > > > +       if (!vma)
> > > > > > > > +               goto lock_mmap;
> > > > > > > > +
> > > > > > > > +       if (binder_alloc_is_mapped(alloc))
> > > > > > >
> > > > > > > I don't think you need this check here. lock_vma_under_rcu() ensures
> > > > > > > that the VMA was not detached from the tree after locking the VMA, so
> > > > > > > if you got a VMA it's in the tree and it can't be removed (because
> > > > > > > it's locked). remove_vma()->vma_close()->vma->vm_ops->close() is
> > > > > > > called after VMA gets detached from the tree and that won't happen
> > > > > > > while VMA is locked. So, if lock_vma_under_rcu() returns a VMA,
> > > > > > > binder_alloc_is_mapped() has to always return true. A WARN_ON() check
> > > > > > > here to ensure that might be a better option.
> > > > > >
> > > > > > Yes we are guaranteed to have _a_ non-isolated vma. However, the check
> > > > > > validates that it's the _expected_ vma. IIUC, our vma could have been
> > > > > > unmapped (clearing alloc->mapped) and a _new_ unrelated vma could have
> > > > > > gotten the same address space assigned?
> > > > >
> > > > > No, this should never happen. lock_vma_under_rcu() specifically checks
> > > > > the address range *after* it locks the VMA:
> > > > > https://elixir.bootlin.com/linux/v6.11.6/source/mm/memory.c#L6026
> > > >
> > > > The scenario I'm describing is the following:
> > > >
> > > > Proc A                          Proc B
> > > >                                 mmap(addr, binder_fd)
> > > > binder_page_insert()
> > > > mmget_not_zero()
> > > >                                 munmap(addr)
> > > >                                 alloc->mapped = false;
> > > >                                 [...]
> > > >                                 // mmap other vma but same addr
> > > >                                 mmap(addr, other_fd)
> > > >
> > > > vma = lock_vma_under_rcu()
> > > >
> > > > Isn't there a chance for the vma that Proc A receives is an unrelated
> > > > vma that was placed in the same address range?
> > >
> > > Ah, I see now. The VMA is a valid one and at the address we specified
> > > but it does not belong to the binder. Yes, then you do need this
> > > check.
> > 
> > Is this scenario possible?:
> > 
> >  Proc A                          Proc B
> >                                  mmap(addr, binder_fd)
> >  binder_page_insert()
> >  mmget_not_zero()
> >                                  munmap(addr)
> >                                  alloc->mapped = false;
> >                                  [...]
> >                                  // mmap other vma but same addr
> >                                  mmap(addr, other_fd)
> >                                  mmap(other_addr, binder_fd)
> >  vma = lock_vma_under_rcu(addr)
> > 
> > If so, I think your binder_alloc_is_mapped() check will return true
> > but the binder area is mapped at a different other_addr. To avoid that
> > I think you can check that "addr" still belongs to [alloc->vm_start,
> > alloc->buffer_size] after you obtained and locked the VMA.
> 
> Wait, I thought that vm_ops->close() was called with the mmap_lock in
> exclusive mode. This is where binder clears the alloc->mapped. If this
> is not the case (was it ever?), then I'd definitely need to fix this.

On a second though, we don't need the mmap_sem in exclusive mode. When
we acquire the vma through lock_vma_under_rcu() we guarantee it's not
isolated and if our alloc->mapped is set, that means it has not been
unmapped either. So we are good to proceed.

--
Carlos Llamas