[PATCH v3 10/15] mm/pte_ref: add support for page fault path

[Qi Zheng]

In the page fault path, we need to take a reference of
the PTE page table page if the pmd entry is not none,
which ensures the PTE page table page will not be released
by other threads.

And the mmap_lock may be unlocked in advance in some cases
in handle_pte_fault(), then the pmd entry will no longer
be stable:

	thread A		thread B
	page fault		collapse_huge_page
	==========		==================

	mmap_read_unlock()
				mmap_write_lock()
				pgtable_trans_huge_deposit()
				set_pmd_at()
	/* pmd entry is changed! */
	pte_put()

So we should call pte_put() before dropping the mmap_lock.

Signed-off-by: Qi Zheng <zhengqi.arch@xxxxxxxxxxxxx>
---
 fs/userfaultfd.c |  1 +
 mm/filemap.c     |  2 ++
 mm/internal.h    |  1 +
 mm/khugepaged.c  |  8 +++++++-
 mm/memory.c      | 33 ++++++++++++++++++++++++---------
 5 files changed, 35 insertions(+), 10 deletions(-)

diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
index 22bf14ab2d16..ddbcefa7e0a6 100644
--- a/fs/userfaultfd.c
+++ b/fs/userfaultfd.c
@@ -509,6 +509,7 @@ vm_fault_t handle_userfault(struct vm_fault *vmf, unsigned long reason)
 		must_wait = userfaultfd_huge_must_wait(ctx, vmf->vma,
 						       vmf->address,
 						       vmf->flags, reason);
+	pte_put_vmf(vmf);
 	mmap_read_unlock(mm);
 
 	if (likely(must_wait && !READ_ONCE(ctx->released))) {
diff --git a/mm/filemap.c b/mm/filemap.c
index aa47ee11a3d8..4fdc74dc6736 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -1708,6 +1708,7 @@ bool __folio_lock_or_retry(struct folio *folio, struct vm_fault *vmf)
 		if (flags & FAULT_FLAG_RETRY_NOWAIT)
 			return false;
 
+		pte_put_vmf(vmf);
 		mmap_read_unlock(mm);
 		if (flags & FAULT_FLAG_KILLABLE)
 			folio_wait_locked_killable(folio);
@@ -1720,6 +1721,7 @@ bool __folio_lock_or_retry(struct folio *folio, struct vm_fault *vmf)
 
 		ret = __folio_lock_killable(folio);
 		if (ret) {
+			pte_put_vmf(vmf);
 			mmap_read_unlock(mm);
 			return false;
 		}
diff --git a/mm/internal.h b/mm/internal.h
index 474d6e3443f8..460418828a76 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -488,6 +488,7 @@ static inline struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
 	if (fault_flag_allow_retry_first(flags) &&
 	    !(flags & FAULT_FLAG_RETRY_NOWAIT)) {
 		fpin = get_file(vmf->vma->vm_file);
+		pte_put_vmf(vmf);
 		mmap_read_unlock(vmf->vma->vm_mm);
 	}
 	return fpin;
diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index e99101162f1a..92b0494f4a00 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -1019,10 +1019,13 @@ static bool __collapse_huge_page_swapin(struct mm_struct *mm,
 			.pmd = pmd,
 		};
 
-		vmf.pte = pte_offset_map(pmd, address);
+		vmf.pte = pte_tryget_map(pmd, address);
+		if (!vmf.pte)
+			continue;
 		vmf.orig_pte = *vmf.pte;
 		if (!is_swap_pte(vmf.orig_pte)) {
 			pte_unmap(vmf.pte);
+			pte_put_vmf(&vmf);
 			continue;
 		}
 		swapped_in++;
@@ -1041,7 +1044,10 @@ static bool __collapse_huge_page_swapin(struct mm_struct *mm,
 				trace_mm_collapse_huge_page_swapin(mm, swapped_in, referenced, 0);
 				return false;
 			}
+		} else {
+			pte_put_vmf(&vmf);
 		}
+
 		if (ret & VM_FAULT_ERROR) {
 			trace_mm_collapse_huge_page_swapin(mm, swapped_in, referenced, 0);
 			return false;
diff --git a/mm/memory.c b/mm/memory.c
index ea4d651ac8c7..5cc4ce0af665 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -4571,8 +4571,10 @@ static vm_fault_t wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud)
 static vm_fault_t handle_pte_fault(struct vm_fault *vmf)
 {
 	pte_t entry;
+	vm_fault_t ret;
 
-	if (unlikely(pmd_none(*vmf->pmd))) {
+retry:
+	if (unlikely(pmd_none(READ_ONCE(*vmf->pmd)))) {
 		/*
 		 * Leave __pte_alloc() until later: because vm_ops->fault may
 		 * want to allocate huge page, and if we expose page table
@@ -4595,13 +4597,16 @@ static vm_fault_t handle_pte_fault(struct vm_fault *vmf)
 		 */
 		if (pmd_devmap_trans_unstable(vmf->pmd))
 			return 0;
+
 		/*
 		 * A regular pmd is established and it can't morph into a huge
 		 * pmd from under us anymore at this point because we hold the
 		 * mmap_lock read mode and khugepaged takes it in write mode.
 		 * So now it's safe to run pte_offset_map().
 		 */
-		vmf->pte = pte_offset_map(vmf->pmd, vmf->address);
+		vmf->pte = pte_tryget_map(vmf->pmd, vmf->address);
+		if (!vmf->pte)
+			goto retry;
 		vmf->orig_pte = *vmf->pte;
 
 		/*
@@ -4616,6 +4621,7 @@ static vm_fault_t handle_pte_fault(struct vm_fault *vmf)
 		if (pte_none(vmf->orig_pte)) {
 			pte_unmap(vmf->pte);
 			vmf->pte = NULL;
+			pte_put_vmf(vmf);
 		}
 	}
 
@@ -4626,11 +4632,15 @@ static vm_fault_t handle_pte_fault(struct vm_fault *vmf)
 			return do_fault(vmf);
 	}
 
-	if (!pte_present(vmf->orig_pte))
-		return do_swap_page(vmf);
+	if (!pte_present(vmf->orig_pte)) {
+		ret = do_swap_page(vmf);
+		goto put;
+	}
 
-	if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma))
-		return do_numa_page(vmf);
+	if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) {
+		ret = do_numa_page(vmf);
+		goto put;
+	}
 
 	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
 	spin_lock(vmf->ptl);
@@ -4640,8 +4650,10 @@ static vm_fault_t handle_pte_fault(struct vm_fault *vmf)
 		goto unlock;
 	}
 	if (vmf->flags & FAULT_FLAG_WRITE) {
-		if (!pte_write(entry))
-			return do_wp_page(vmf);
+		if (!pte_write(entry)) {
+			ret = do_wp_page(vmf);
+			goto put;
+		}
 		entry = pte_mkdirty(entry);
 	}
 	entry = pte_mkyoung(entry);
@@ -4663,7 +4675,10 @@ static vm_fault_t handle_pte_fault(struct vm_fault *vmf)
 	}
 unlock:
 	pte_unmap_unlock(vmf->pte, vmf->ptl);
-	return 0;
+	ret = 0;
+put:
+	pte_put_vmf(vmf);
+	return ret;
 }
 
 /*
-- 
2.11.0