On Thu, Oct 30, 2025 at 07:47:34PM +0100, Vlastimil Babka wrote:
> On 10/30/25 19:31, Vlastimil Babka wrote:
> > On 10/30/25 17:43, Lorenzo Stoakes wrote:
> >> On Thu, Oct 30, 2025 at 04:31:56PM +0000, Pedro Falcato wrote:
> >>> On Thu, Oct 30, 2025 at 04:23:58PM +0000, Lorenzo Stoakes wrote:
> >>> > On Thu, Oct 30, 2025 at 04:16:20PM +0000, Pedro Falcato wrote:
> >>> > > On Wed, Oct 29, 2025 at 04:50:31PM +0000, Lorenzo Stoakes wrote:
> >>> > > > Currently, if a user needs to determine if guard regions are
> >>> > > > present in a
> >>> > > > range, they have to scan all VMAs (or have knowledge of which ones
> >>> > > > might
> >>> > > > have guard regions).
> >>> > > >
> >>> > > > Since commit 8e2f2aeb8b48 ("fs/proc/task_mmu: add guard region bit
> >>> > > > to
> >>> > > > pagemap") and the related commit a516403787e0 ("fs/proc: extend the
> >>> > > > PAGEMAP_SCAN ioctl to report guard regions"), users can use either
> >>> > > > /proc/$pid/pagemap or the PAGEMAP_SCAN functionality to perform this
> >>> > > > operation at a virtual address level.
> >>> > > >
> >>> > > > This is not ideal, and it gives no visibility at a /proc/$pid/smaps
> >>> > > > level
> >>> > > > that guard regions exist in ranges.
> >>> > > >
> >>> > > > This patch remedies the situation by establishing a new VMA flag,
> >>> > > > VM_MAYBE_GUARD, to indicate that a VMA may contain guard regions
> >>> > > > (it is
> >>> > > > uncertain because we cannot reasonably determine whether a
> >>> > > > MADV_GUARD_REMOVE call has removed all of the guard regions in a
> >>> > > > VMA, and
> >>> > > > additionally VMAs may change across merge/split).
> >>> > > >
> >>> > > > We utilise 0x800 for this flag which makes it available to 32-bit
> >>> > > > architectures also, a flag that was previously used by
> >>> > > > VM_DENYWRITE, which
> >>> > > > was removed in commit 8d0920bde5eb ("mm: remove VM_DENYWRITE") and
> >>> > > > hasn't
> >>> > > > bee reused yet.
> >>> > > >
> >>> > > > The MADV_GUARD_INSTALL madvise() operation now must take an mmap
> >>> > > > write
> >>> > > > lock (and also VMA write lock) whereas previously it did not, but
> >>> > > > this
> >>> > > > seems a reasonable overhead.
> >>> > >
> >>> > > Do you though? Could it be possible to simply atomically set the flag
> >>> > > with
> >>> > > the read lock held? This would make it so we can't split the VMA (and
> >>> > > tightly
> >>> >
> >>> > VMA flags are not accessed atomically so no I don't think we can do
> >>> > that in any
> >>> > workable way.
> >>> >
> >>>
> >>> FWIW I think you could work it as an atomic flag and treat those races as
> >>> benign
> >>> (this one, at least).
> >>
> >> It's not benign as we need to ensure that page tables are correctly
> >> propagated
> >> on fork.
> >
> > Could we use MADVISE_VMA_READ_LOCK mode (would be actually an improvement
> > over the current MADVISE_MMAP_READ_LOCK), together with the atomic flag
> > setting? I think the places that could race with us to cause RMW use vma
> > write lock so that would be excluded. Fork AFAICS unfortunately doesn't (for
> > the oldmm) and it probably would't make sense to start doing it. Maybe we
> > could think of something to deal with this special case...
>
> During discussion with Pedro off-list I realized fork takes mmap lock for
> write on the old mm, so if we kept taking mmap sem for read, then vma lock
> for read in addition (which should be cheap enough, also we'd only need it
> in case VM_MAYBE_GUARD is not yet set), and set the flag atomicaly, perhaps
> that would cover all non-bening races?
>
>
We take VMA write lock in dup_mmap() on each mpnt (old VMA).
We take the VMA write lock (vma_start_write()) for each mpnt.
We then vm_area_dup() the mpnt to the new VMA before calling:
copy_page_range()
-> vma_needs_copy()
Which is where the check is done.
So we are holding the VMA write lock, so a VMA read lock should suffice no?
For belts + braces we could atomically read the flag in vma_needs_copy(),
though note it's intended VM_COPY_ON_FORK could have more than one flag.
We could drop that for now and be explicit.
