[PATCH 1/1] mm/hugetlb: fix possible deadlock in __hugetlb_zap_begin

[Lance Yang <lance.yang@linux.dev>]

From: Lance Yang <lance.yang@linux.dev>

The hugetlb VMA unmap path contains a potential deadlock, as reported
by syzbot. In __hugetlb_zap_begin(), vma_lock is acquired before
i_mmap_lock. This lock ordering conflicts with the page fault path in
hugetlb_fault(), which acquires i_mmap_lock first, establishing the
correct dependency as i_mmap_lock -> vma_lock.

Chain exists of:
  &hugetlbfs_i_mmap_rwsem_key
    --> &hugetlb_fault_mutex_table[i]
      --> &vma_lock->rw_sema

Possible unsafe locking scenario:

CPU0                                 CPU1
----                                 ----
lock(&vma_lock->rw_sema);
                                     lock(&hugetlb_fault_mutex_table[i]);
                                     lock(&vma_lock->rw_sema);
lock(&hugetlbfs_i_mmap_rwsem_key);

Resolve the deadlock by reordering the locks in __hugetlb_zap_begin() to
follow the established i_mmap_lock -> vma_lock order.

Reported-by: syzbot+3f5f9a0d292454409ca6@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/linux-mm/69113a97.a70a0220.22f260.00ca.GAE@google.com/
Signed-off-by: Lance Yang <lance.yang@linux.dev>
---
 mm/hugetlb.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index b1f47b87ae65..2719995af18e 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -5327,9 +5327,9 @@ void __hugetlb_zap_begin(struct vm_area_struct *vma,
 		return;
 
 	adjust_range_if_pmd_sharing_possible(vma, start, end);
-	hugetlb_vma_lock_write(vma);
 	if (vma->vm_file)
 		i_mmap_lock_write(vma->vm_file->f_mapping);
+	hugetlb_vma_lock_write(vma);
 }
 
 void __hugetlb_zap_end(struct vm_area_struct *vma,
-- 
2.49.0