From: Andrea Arcangeli <aarcange@redhat.com>
As far as the rmap code is concerned, UFFDIO_REMAP only alters the
page->mapping and page->index. It does it while holding the page
lock. However folio_referenced() is doing rmap walks without taking the
folio lock first, so folio_lock_anon_vma_read() must be updated to
re-check that the folio->mapping didn't change after we obtained the
anon_vma read lock.
UFFDIO_REMAP takes the anon_vma lock for writing before altering the
folio->mapping, so if the folio->mapping is still the same after
obtaining the anon_vma read lock (without the folio lock), the rmap
walks can go ahead safely (and UFFDIO_REMAP will wait the rmap walk to
complete before proceeding).
UFFDIO_REMAP serializes against itself with the folio lock.
All other places taking the anon_vma lock while holding the mmap_lock
for writing, don't need to check if the folio->mapping has changed
after taking the anon_vma lock, regardless of the folio lock, because
UFFDIO_REMAP holds the mmap_lock for reading.
There's one constraint enforced to allow this simplification: the
source pages passed to UFFDIO_REMAP must be mapped only in one vma,
but this constraint is an acceptable tradeoff for UFFDIO_REMAP
users.
The source addresses passed to UFFDIO_REMAP can be set as
VM_DONTCOPY with MADV_DONTFORK to avoid any risk of the mapcount of
the pages increasing if some thread of the process forks() before
UFFDIO_REMAP run.
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Suren Baghdasaryan <surenb@google.com>
---
mm/rmap.c | 13 +++++++++++++
1 file changed, 13 insertions(+)
diff --git a/mm/rmap.c b/mm/rmap.c
index ec7f8e6c9e48..c1ebbd23fa61 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -542,6 +542,7 @@ struct anon_vma *folio_lock_anon_vma_read(struct folio *folio,
struct anon_vma *root_anon_vma;
unsigned long anon_mapping;
+repeat:
rcu_read_lock();
anon_mapping = (unsigned long)READ_ONCE(folio->mapping);
if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
@@ -586,6 +587,18 @@ struct anon_vma *folio_lock_anon_vma_read(struct folio *folio,
rcu_read_unlock();
anon_vma_lock_read(anon_vma);
+ /*
+ * Check if UFFDIO_REMAP changed the anon_vma. This is needed
+ * because we don't assume the folio was locked.
+ */
+ if (unlikely((unsigned long) READ_ONCE(folio->mapping) !=
+ anon_mapping)) {
+ anon_vma_unlock_read(anon_vma);
+ put_anon_vma(anon_vma);
+ anon_vma = NULL;
+ goto repeat;
+ }
+
if (atomic_dec_and_test(&anon_vma->refcount)) {
/*
* Oops, we held the last refcount, release the lock
--
2.42.0.515.g380fc7ccd1-googOn 23.09.23 03:31, Suren Baghdasaryan wrote: > From: Andrea Arcangeli <aarcange@redhat.com> > > As far as the rmap code is concerned, UFFDIO_REMAP only alters the > page->mapping and page->index. It does it while holding the page > lock. However folio_referenced() is doing rmap walks without taking the > folio lock first, so folio_lock_anon_vma_read() must be updated to > re-check that the folio->mapping didn't change after we obtained the > anon_vma read lock. I'm curious: why don't we need this for existing users of page_move_anon_rmap()? What's special about UFFDIO_REMAP? -- Cheers, David / dhildenb
On Mon, Oct 02, 2023 at 04:42:50PM +0200, David Hildenbrand wrote: > On 23.09.23 03:31, Suren Baghdasaryan wrote: > > From: Andrea Arcangeli <aarcange@redhat.com> > > > > As far as the rmap code is concerned, UFFDIO_REMAP only alters the > > page->mapping and page->index. It does it while holding the page > > lock. However folio_referenced() is doing rmap walks without taking the > > folio lock first, so folio_lock_anon_vma_read() must be updated to > > re-check that the folio->mapping didn't change after we obtained the > > anon_vma read lock. > > I'm curious: why don't we need this for existing users of > page_move_anon_rmap()? What's special about UFFDIO_REMAP? Totally no expert on anon vma so I'm prone to errors, but IIUC the difference here is root anon vma cannot change in page_move_anon_rmap(), while UFFDIO_REMAP can. Thanks, -- Peter Xu
On 02.10.23 17:23, Peter Xu wrote: > On Mon, Oct 02, 2023 at 04:42:50PM +0200, David Hildenbrand wrote: >> On 23.09.23 03:31, Suren Baghdasaryan wrote: >>> From: Andrea Arcangeli <aarcange@redhat.com> >>> >>> As far as the rmap code is concerned, UFFDIO_REMAP only alters the >>> page->mapping and page->index. It does it while holding the page >>> lock. However folio_referenced() is doing rmap walks without taking the >>> folio lock first, so folio_lock_anon_vma_read() must be updated to >>> re-check that the folio->mapping didn't change after we obtained the >>> anon_vma read lock. >> >> I'm curious: why don't we need this for existing users of >> page_move_anon_rmap()? What's special about UFFDIO_REMAP? > > Totally no expert on anon vma so I'm prone to errors, but IIUC the > difference here is root anon vma cannot change in page_move_anon_rmap(), > while UFFDIO_REMAP can. That does sound reasonable, thanks. Probably we can do better with the patch description (once [1] is used to move the folio to the other anon_vma). "mm/rmap: support move to different root anon_vma in folio_move_anon_rmap() For now, folio_move_anon_rmap() was only used to move a folio to a different anon_vma after fork(), whereby the root anon_vma stayed unchanged. For that, it was sufficient to hold the page lock when calling folio_move_anon_rmap(). However, we want to make use of folio_move_anon_rmap() to move folios between VMAs that have a different root anon_vma. As folio_referenced() performs an RMAP walk without holding the page lock but only holding the anon_vma in read mode, holding the page lock is insufficient. When moving to an anon_vma with a different root anon_vma, we'll have to hold both, the page lock and the anon_vma lock in write mode. Consequently, whenever we succeeded in folio_lock_anon_vma_read() to read-lock the anon_vma, we have to re-check if the mapping was changed in the meantime. If that was the case, we have to retry. Note that folio_move_anon_rmap() must only be called if the anon page is exclusive to a process, and must not be called on KSM folios. This is a preparation for UFFDIO_REMAP, which will hold the page lock, the anon_vma lock in write mode, and the mmap_lock in read mode. " In addition, we should document these locking details for folio_move_anon_rmap() and probably not mention UFFDIO_REMAP in the comment in folio_lock_anon_vma_read(), but instead say "folio_move_anon_rmap() might have changed the anon_vma as we might not hold the page lock here." [1] https://lkml.kernel.org/r/20231002142949.235104-3-david@redhat.com -- Cheers, David / dhildenb
On Mon, Oct 2, 2023 at 10:30 AM David Hildenbrand <david@redhat.com> wrote: > > On 02.10.23 17:23, Peter Xu wrote: > > On Mon, Oct 02, 2023 at 04:42:50PM +0200, David Hildenbrand wrote: > >> On 23.09.23 03:31, Suren Baghdasaryan wrote: > >>> From: Andrea Arcangeli <aarcange@redhat.com> > >>> > >>> As far as the rmap code is concerned, UFFDIO_REMAP only alters the > >>> page->mapping and page->index. It does it while holding the page > >>> lock. However folio_referenced() is doing rmap walks without taking the > >>> folio lock first, so folio_lock_anon_vma_read() must be updated to > >>> re-check that the folio->mapping didn't change after we obtained the > >>> anon_vma read lock. > >> > >> I'm curious: why don't we need this for existing users of > >> page_move_anon_rmap()? What's special about UFFDIO_REMAP? > > > > Totally no expert on anon vma so I'm prone to errors, but IIUC the > > difference here is root anon vma cannot change in page_move_anon_rmap(), > > while UFFDIO_REMAP can. > > That does sound reasonable, thanks. > > Probably we can do better with the patch description (once [1] is used > to move the folio to the other anon_vma). I'll develop the next version with your patches in the baseline. Hopefully by the time of my posting your patches will be in the mm-unstable. > > "mm/rmap: support move to different root anon_vma in folio_move_anon_rmap() > > For now, folio_move_anon_rmap() was only used to move a folio to a > different anon_vma after fork(), whereby the root anon_vma stayed > unchanged. For that, it was sufficient to hold the page lock when > calling folio_move_anon_rmap(). > > However, we want to make use of folio_move_anon_rmap() to move folios > between VMAs that have a different root anon_vma. As folio_referenced() > performs an RMAP walk without holding the page lock but only holding the > anon_vma in read mode, holding the page lock is insufficient. > > When moving to an anon_vma with a different root anon_vma, we'll have to > hold both, the page lock and the anon_vma lock in write mode. > Consequently, whenever we succeeded in folio_lock_anon_vma_read() to > read-lock the anon_vma, we have to re-check if the mapping was changed > in the meantime. If that was the case, we have to retry. > > Note that folio_move_anon_rmap() must only be called if the anon page is > exclusive to a process, and must not be called on KSM folios. > > This is a preparation for UFFDIO_REMAP, which will hold the page lock, > the anon_vma lock in write mode, and the mmap_lock in read mode. > " Thanks for taking time to write this up! Looks really clear to me. I'll reuse it. > > In addition, we should document these locking details for > folio_move_anon_rmap() and probably not mention UFFDIO_REMAP in the > comment in folio_lock_anon_vma_read(), but instead say > "folio_move_anon_rmap() might have changed the anon_vma as we might not > hold the page lock here." Sounds good. Will add. > > > [1] https://lkml.kernel.org/r/20231002142949.235104-3-david@redhat.com > > -- > Cheers, > > David / dhildenb >
Suren,
Sorry to review so late.
On Fri, Sep 22, 2023 at 06:31:44PM -0700, Suren Baghdasaryan wrote:
> diff --git a/mm/rmap.c b/mm/rmap.c
> index ec7f8e6c9e48..c1ebbd23fa61 100644
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -542,6 +542,7 @@ struct anon_vma *folio_lock_anon_vma_read(struct folio *folio,
> struct anon_vma *root_anon_vma;
> unsigned long anon_mapping;
>
> +repeat:
> rcu_read_lock();
> anon_mapping = (unsigned long)READ_ONCE(folio->mapping);
> if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
> @@ -586,6 +587,18 @@ struct anon_vma *folio_lock_anon_vma_read(struct folio *folio,
> rcu_read_unlock();
> anon_vma_lock_read(anon_vma);
>
> + /*
> + * Check if UFFDIO_REMAP changed the anon_vma. This is needed
> + * because we don't assume the folio was locked.
> + */
> + if (unlikely((unsigned long) READ_ONCE(folio->mapping) !=
> + anon_mapping)) {
> + anon_vma_unlock_read(anon_vma);
> + put_anon_vma(anon_vma);
> + anon_vma = NULL;
> + goto repeat;
> + }
We have an open-coded fast path above this:
if (down_read_trylock(&root_anon_vma->rwsem)) {
/*
* If the folio is still mapped, then this anon_vma is still
* its anon_vma, and holding the mutex ensures that it will
* not go away, see anon_vma_free().
*/
if (!folio_mapped(folio)) {
up_read(&root_anon_vma->rwsem);
anon_vma = NULL;
}
goto out;
}
Would that also need such check?
> +
> if (atomic_dec_and_test(&anon_vma->refcount)) {
> /*
> * Oops, we held the last refcount, release the lock
> --
> 2.42.0.515.g380fc7ccd1-goog
>
--
Peter Xu
On Thu, Sep 28, 2023 at 9:23 AM Peter Xu <peterx@redhat.com> wrote:
>
> Suren,
>
> Sorry to review so late.
>
> On Fri, Sep 22, 2023 at 06:31:44PM -0700, Suren Baghdasaryan wrote:
> > diff --git a/mm/rmap.c b/mm/rmap.c
> > index ec7f8e6c9e48..c1ebbd23fa61 100644
> > --- a/mm/rmap.c
> > +++ b/mm/rmap.c
> > @@ -542,6 +542,7 @@ struct anon_vma *folio_lock_anon_vma_read(struct folio *folio,
> > struct anon_vma *root_anon_vma;
> > unsigned long anon_mapping;
> >
> > +repeat:
> > rcu_read_lock();
> > anon_mapping = (unsigned long)READ_ONCE(folio->mapping);
> > if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
> > @@ -586,6 +587,18 @@ struct anon_vma *folio_lock_anon_vma_read(struct folio *folio,
> > rcu_read_unlock();
> > anon_vma_lock_read(anon_vma);
> >
> > + /*
> > + * Check if UFFDIO_REMAP changed the anon_vma. This is needed
> > + * because we don't assume the folio was locked.
> > + */
> > + if (unlikely((unsigned long) READ_ONCE(folio->mapping) !=
> > + anon_mapping)) {
> > + anon_vma_unlock_read(anon_vma);
> > + put_anon_vma(anon_vma);
> > + anon_vma = NULL;
> > + goto repeat;
> > + }
>
> We have an open-coded fast path above this:
>
> if (down_read_trylock(&root_anon_vma->rwsem)) {
> /*
> * If the folio is still mapped, then this anon_vma is still
> * its anon_vma, and holding the mutex ensures that it will
> * not go away, see anon_vma_free().
> */
> if (!folio_mapped(folio)) {
> up_read(&root_anon_vma->rwsem);
> anon_vma = NULL;
> }
> goto out;
> }
>
> Would that also need such check?
Yes, I think they should be handled the same way. Will fix. Thanks!
>
> > +
> > if (atomic_dec_and_test(&anon_vma->refcount)) {
> > /*
> > * Oops, we held the last refcount, release the lock
> > --
> > 2.42.0.515.g380fc7ccd1-goog
> >
>
> --
> Peter Xu
>
© 2016 - 2026 Red Hat, Inc.