move_pages_pte() fixes

Lokesh recently raised an issue about UFFDIO_MOVE getting into a deadlock state when it goes into split_folio() with raised folio refcount. split_folio() expects the reference count to be exactly mapcount + num_pages_in_folio + 1 (see can_split_folio()) and fails with EAGAIN otherwise. If multiple processes are trying to move the same large folio, they raise the refcount (all tasks succeed in that) then one of them succeeds in locking the folio, while others will block in folio_lock() while keeping the refcount raised. The winner of this race will proceed with calling split_folio() and will fail returning EAGAIN to the caller and unlocking the folio. The next competing process will get the folio locked and will go through the same flow. In the meantime the original winner will be retried and will block in folio_lock(), getting into the queue of waiting processes only to repeat the same path. All this results in a livelock.
An easy fix would be to avoid waiting for the folio lock while holding folio refcount, similar to madvise_free_huge_pmd() where folio lock is acquired before raising the folio refcount. Since we lock and take a refcount of the folio while holding the PTE lock, changing the order of these operations should not break anything.
Modify move_pages_pte() to try locking the folio first and if that fails and the folio is large then return EAGAIN without touching the folio refcount. If the folio is single-page then split_folio() is not called, so we don't have this issue.
Lokesh has a reproducer [1] and I verified that this change fixes the issue.

[1] https://github.com/lokeshgidra/uffd_move_ioctl_deadlock

Fixes adef440691ba ("userfaultfd: UFFDIO_MOVE uABI")

Reported-by Lokesh Gidra <[email protected]>

Signed-off-by Suren Baghdasaryan <[email protected]>

Reviewed-by Peter Xu <[email protected]>

Cc [email protected]

---
Note this patch is v2 of [2] but I did not bump up the version because now it's part of the patchset which is at its v1. Hopefully that's not too confusing.

Changes since v1 [2]:
- Rebased over mm-hotfixes-unstable to avoid merge conflicts with [3]
- Added Reviewed-by, per Peter Xu
- Added a note about PTL lock in the changelog, per Liam R. Howlett
- CC'ed stable

[2] https://lore.kernel.org/all/[email protected]/
[3] https://lore.kernel.org/all/[email protected]/

 mm/userfaultfd.c | 17 ++++++++++++++++-
 1 file changed, 16 insertions(+), 1 deletion(-)

mm/userfaultfd.c +16 -1

index 8eae4ea3cafd..e0f1e38ac5d8 100644
@@ -1250,6 +1250,7 @@ static int move_pages_pte(struct mm_struct *mm, pmd_t *dst_pmd, pmd_t *src_pmd,
 		 */
 		if (!src_folio) {
 			struct folio *folio;
+			bool locked;
 
 			/*
 			 * Pin the page while holding the lock to be sure the

@@ -1269,12 +1270,26 @@ static int move_pages_pte(struct mm_struct *mm, pmd_t *dst_pmd, pmd_t *src_pmd,
 				goto out;
 			}
 
+			locked = folio_trylock(folio);
+			/*
+			 * We avoid waiting for folio lock with a raised refcount
+			 * for large folios because extra refcounts will result in
+			 * split_folio() failing later and retrying. If multiple
+			 * tasks are trying to move a large folio we can end
+			 * livelocking.
+			 */
+			if (!locked && folio_test_large(folio)) {
+				spin_unlock(src_ptl);
+				err = -EAGAIN;
+				goto out;
+			}
+
 			folio_get(folio);
 			src_folio = folio;
 			src_folio_pte = orig_src_pte;
 			spin_unlock(src_ptl);
 
-			if (!folio_trylock(src_folio)) {
+			if (!locked) {
 				pte_unmap(&orig_src_pte);
 				pte_unmap(&orig_dst_pte);
 				src_pte = dst_pte = NULL;

Signature · 1 line

2.48.1.658.g4767266eb4-goog

Current implementation of move_pages_pte() copies source and destination PTEs in order to detect concurrent changes to PTEs involved in the move. However these copies are also used to unmap the PTEs, which will fail if CONFIG_HIGHPTE is enabled because the copies are allocated on the stack. Fix this by using the actual PTEs which were kmap()ed.

Fixes adef440691ba ("userfaultfd: UFFDIO_MOVE uABI")

Reported-by Peter Xu <[email protected]>

Signed-off-by Suren Baghdasaryan <[email protected]>

Cc [email protected]