[PATCH 2/3] hugetlb: take hugetlb vma_lock when clearing vma_lock->vma pointer

[Mike Kravetz]

hugetlb file truncation/hole punch code may need to back out and take
locks in order in the routine hugetlb_unmap_file_folio().  This code
could race with vma freeing as pointed out in [1] and result in
accessing a stale vma pointer.  To address this, take the vma_lock when
clearing the vma_lock->vma pointer.

[1] https://lore.kernel.org/linux-mm/01f10195-7088-4462-6def-909549c75ef4@xxxxxxxxxx/

Fixes: "hugetlb: use new vma_lock for pmd sharing synchronization"
Signed-off-by: Mike Kravetz <mike.kravetz@xxxxxxxxxx>
---
 mm/hugetlb.c | 42 ++++++++++++++++++++++++++++++++----------
 1 file changed, 32 insertions(+), 10 deletions(-)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 38c45aafe00f..e9194ca6d44b 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -93,6 +93,7 @@ struct mutex *hugetlb_fault_mutex_table ____cacheline_aligned_in_smp;
 static int hugetlb_acct_memory(struct hstate *h, long delta);
 static void hugetlb_vma_lock_free(struct vm_area_struct *vma);
 static void hugetlb_vma_lock_alloc(struct vm_area_struct *vma);
+static void __hugetlb_vma_unlock_write_free(struct vm_area_struct *vma);
 
 static inline bool subpool_is_free(struct hugepage_subpool *spool)
 {
@@ -5192,8 +5193,7 @@ void __unmap_hugepage_range_final(struct mmu_gather *tlb,
 	 * be asynchrously deleted.  If the page tables are shared, there
 	 * will be issues when accessed by someone else.
 	 */
-	hugetlb_vma_unlock_write(vma);
-	hugetlb_vma_lock_free(vma);
+	__hugetlb_vma_unlock_write_free(vma);
 
 	i_mmap_unlock_write(vma->vm_file->f_mapping);
 }
@@ -6942,6 +6942,30 @@ void hugetlb_vma_lock_release(struct kref *kref)
 	kfree(vma_lock);
 }
 
+void __hugetlb_vma_unlock_write_put(struct hugetlb_vma_lock *vma_lock)
+{
+	struct vm_area_struct *vma = vma_lock->vma;
+
+	/*
+	 * vma_lock structure may or not be released as a result of put,
+	 * it certainly will no longer be attached to vma so clear pointer.
+	 * Semaphore synchronizes access to vma_lock->vma field.
+	 */
+	vma_lock->vma = NULL;
+	vma->vm_private_data = NULL;
+	up_write(&vma_lock->rw_sema);
+	kref_put(&vma_lock->refs, hugetlb_vma_lock_release);
+}
+
+static void __hugetlb_vma_unlock_write_free(struct vm_area_struct *vma)
+{
+	if (__vma_shareable_flags_pmd(vma)) {
+		struct hugetlb_vma_lock *vma_lock = vma->vm_private_data;
+
+		__hugetlb_vma_unlock_write_put(vma_lock);
+	}
+}
+
 static void hugetlb_vma_lock_free(struct vm_area_struct *vma)
 {
 	/*
@@ -6953,14 +6977,8 @@ static void hugetlb_vma_lock_free(struct vm_area_struct *vma)
 	if (vma->vm_private_data) {
 		struct hugetlb_vma_lock *vma_lock = vma->vm_private_data;
 
-		/*
-		 * vma_lock structure may or not be released, but it
-		 * certainly will no longer be attached to vma so clear
-		 * pointer.
-		 */
-		vma_lock->vma = NULL;
-		kref_put(&vma_lock->refs, hugetlb_vma_lock_release);
-		vma->vm_private_data = NULL;
+		down_write(&vma_lock->rw_sema);
+		__hugetlb_vma_unlock_write_put(vma_lock);
 	}
 }
 
@@ -7111,6 +7129,10 @@ void hugetlb_vma_lock_release(struct kref *kref)
 {
 }
 
+static void __hugetlb_vma_unlock_write_free(struct vm_area_struct *vma)
+{
+}
+
 static void hugetlb_vma_lock_free(struct vm_area_struct *vma)
 {
 }
-- 
2.37.3