[4/4] mm: Protect kernel pgtables in apply_to_pte_range()

Commit Message

Alexander Gordeev March 28, 2025, 9:13 a.m. UTC

The lazy MMU mode can only be entered and left under the protection
of the page table locks for all page tables which may be modified.
Yet, when it comes to kernel mappings apply_to_pte_range() does not
take any locks. That does not conform arch_enter|leave_lazy_mmu_mode()
semantics and could potentially lead to re-schedulling a process while
in lazy MMU mode or racing on a kernel page table updates.

Signed-off-by: Alexander Gordeev <[e-mail]>
---
 mm/kasan/shadow.c | 7 ++-----
 mm/memory.c       | 5 ++++-
 2 files changed, 6 insertions(+), 6 deletions(-)

Patch

diff --git a/mm/kasan/shadow.c b/mm/kasan/shadow.c
index edfa77959474..6531a7aa8562 100644
--- a/mm/kasan/shadow.c
+++ b/mm/kasan/shadow.c
@@ -308,14 +308,14 @@  static int kasan_populate_vmalloc_pte(pte_t *ptep, unsigned long addr,
 	__memset((void *)page, KASAN_VMALLOC_INVALID, PAGE_SIZE);
 	pte = pfn_pte(PFN_DOWN(__pa(page)), PAGE_KERNEL);
 
-	spin_lock(&init_mm.page_table_lock);
 	if (likely(pte_none(ptep_get(ptep)))) {
 		set_pte_at(&init_mm, addr, ptep, pte);
 		page = 0;
 	}
-	spin_unlock(&init_mm.page_table_lock);
+
 	if (page)
 		free_page(page);
+
 	return 0;
 }
 
@@ -401,13 +401,10 @@  static int kasan_depopulate_vmalloc_pte(pte_t *ptep, unsigned long addr,
 
 	page = (unsigned long)__va(pte_pfn(ptep_get(ptep)) << PAGE_SHIFT);
 
-	spin_lock(&init_mm.page_table_lock);
-
 	if (likely(!pte_none(ptep_get(ptep)))) {
 		pte_clear(&init_mm, addr, ptep);
 		free_page(page);
 	}
-	spin_unlock(&init_mm.page_table_lock);
 
 	return 0;
 }
diff --git a/mm/memory.c b/mm/memory.c
index 00f253404db5..c000377cad0c 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -2895,6 +2895,7 @@  static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
 			pte = pte_offset_kernel(pmd, addr);
 		if (!pte)
 			return err;
+		spin_lock(&init_mm.page_table_lock);
 	} else {
 		if (create)
 			pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
@@ -2920,7 +2921,9 @@  static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
 
 	arch_leave_lazy_mmu_mode();
 
-	if (mm != &init_mm)
+	if (mm == &init_mm)
+		spin_unlock(&init_mm.page_table_lock);
+	else
 		pte_unmap_unlock(mapped_pte, ptl);
 
 	*mask |= PGTBL_PTE_MODIFIED;