On Fri, Nov 29, 2024 at 12:44:10PM +0100, Alice Ryhl wrote:
> On Wed, Nov 27, 2024 at 4:41 PM Jann Horn <jannh@xxxxxxxxxx> wrote:
> >
> > On Wed, Nov 27, 2024 at 1:01 PM Alice Ryhl <aliceryhl@xxxxxxxxxx> wrote:
> > > On Tue, Nov 26, 2024 at 11:10 PM Jann Horn <jannh@xxxxxxxxxx> wrote:
> > > >
> > > > On Fri, Nov 22, 2024 at 4:41 PM Alice Ryhl <aliceryhl@xxxxxxxxxx> wrote:
> > > > > This adds a type called VmAreaRef which is used when referencing a vma
> > > > > that you have read access to. Here, read access means that you hold
> > > > > either the mmap read lock or the vma read lock (or stronger).
> > > > >
> > > > > Additionally, a vma_lookup method is added to the mmap read guard, which
> > > > > enables you to obtain a &VmAreaRef in safe Rust code.
> > > > >
> > > > > This patch only provides a way to lock the mmap read lock, but a
> > > > > follow-up patch also provides a way to just lock the vma read lock.
> > > > >
> > > > > Acked-by: Lorenzo Stoakes <lorenzo.stoakes@xxxxxxxxxx> (for mm bits)
> > > > > Signed-off-by: Alice Ryhl <aliceryhl@xxxxxxxxxx>
> > > >
> > > > Reviewed-by: Jann Horn <jannh@xxxxxxxxxx>
> > >
> > > Thanks!
> > >
> > > > with one comment:
> > > >
> > > > > + /// Zap pages in the given page range.
> > > > > + ///
> > > > > + /// This clears page table mappings for the range at the leaf level, leaving all other page
> > > > > + /// tables intact, and freeing any memory referenced by the VMA in this range. That is,
> > > > > + /// anonymous memory is completely freed, file-backed memory has its reference count on page
> > > > > + /// cache folio's dropped, any dirty data will still be written back to disk as usual.
> > > > > + #[inline]
> > > > > + pub fn zap_page_range_single(&self, address: usize, size: usize) {
> > > > > + // SAFETY: By the type invariants, the caller has read access to this VMA, which is
> > > > > + // sufficient for this method call. This method has no requirements on the vma flags. Any
> > > > > + // value of `address` and `size` is allowed.
> > > >
> > > > If we really want to allow any address and size, we might want to add
> > > > an early bailout in zap_page_range_single(). The comment on top of
> > > > zap_page_range_single() currently says "The range must fit into one
> > > > VMA", and it looks like by the point we reach a bailout, we could have
> > > > gone through an interval tree walk via
> > > > mmu_notifier_invalidate_range_start()->__mmu_notifier_invalidate_range_start()->mn_itree_invalidate()
> > > > for a range that ends before it starts; I don't know how safe that is.
> > >
> > > I could change the comment on zap_page_range_single() to say:
> > >
> > > "The range should be contained within a single VMA. Otherwise an error
> > > is returned."
> > >
> > > And then I can add an overflow check at the top of
> > > zap_page_range_single(). Sounds ok?
> >
> > Yes, I think changing the comment like that and adding a check for
> > whether address+size wraps around there addresses this.
>
> Hmm. Looking at this again now ...
>
> For one, the function returns void so we can at best handle overflow
> by performing a no-op.
>
> Another question is, are we actually okay to call it with ranges
> outside the vma? Does that just no-op or what? Maybe the Rust side
> should just bail out early if the address range is not a subset of the
> vma?
In unmap_single_vma() invoked by zap_page_range_single() we check this:
unsigned long start = max(vma->vm_start, start_addr);
...
if (start >= vma->vm_end)
return;
end = min(vma->vm_end, end_addr);
if (end <= vma->vm_start)
return;
So you could specify start <= vma->vm_end and end > vma->vm_start and it'll get
clamped like:
0 MAX
start <-clamp->
<---------------------->xxxxxxxxxxxxxxxxxxxx
|--------| VMA
xxxxxxxxxxxxxxxx<-------------------------->
<-clamp->
However note that we will tell mmum_notifier_range_init() and
hugetlb_zap_begin() incorrect values in this case... not sure.
I'd prefer if rust would strictly only allowed ranges within the VMA.
>
> Alice
