[PATCH 19/31] mm/mremap: retry if either pte_offset_map_*lock() fails

From: Hugh Dickins
Date: Mon May 22 2023 - 01:13:45 EST


move_ptes() return -EAGAIN if pte_offset_map_lock() of old fails, or if
pte_offset_map_nolock() of new fails: move_page_tables() retry if so.

But that does need a pmd_none() check inside, to stop endless loop when
huge shmem is truncated (thank you to syzbot); and move_huge_pmd() must
tolerate that a page table might have been allocated there just before
(of course it would be more satisfying to remove the empty page table,
but this is not a path worth optimizing).

Signed-off-by: Hugh Dickins <hughd@xxxxxxxxxx>
---
mm/huge_memory.c | 5 +++--
mm/mremap.c | 28 ++++++++++++++++++++--------
2 files changed, 23 insertions(+), 10 deletions(-)

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 624671aaa60d..d4bd5fa7c823 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -1760,9 +1760,10 @@ bool move_huge_pmd(struct vm_area_struct *vma, unsigned long old_addr,

/*
* The destination pmd shouldn't be established, free_pgtables()
- * should have release it.
+ * should have released it; but move_page_tables() might have already
+ * inserted a page table, if racing against shmem/file collapse.
*/
- if (WARN_ON(!pmd_none(*new_pmd))) {
+ if (!pmd_none(*new_pmd)) {
VM_BUG_ON(pmd_trans_huge(*new_pmd));
return false;
}
diff --git a/mm/mremap.c b/mm/mremap.c
index b11ce6c92099..1fc47b4f38d7 100644
--- a/mm/mremap.c
+++ b/mm/mremap.c
@@ -133,7 +133,7 @@ static pte_t move_soft_dirty_pte(pte_t pte)
return pte;
}

-static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
+static int move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
unsigned long old_addr, unsigned long old_end,
struct vm_area_struct *new_vma, pmd_t *new_pmd,
unsigned long new_addr, bool need_rmap_locks)
@@ -143,6 +143,7 @@ static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
spinlock_t *old_ptl, *new_ptl;
bool force_flush = false;
unsigned long len = old_end - old_addr;
+ int err = 0;

/*
* When need_rmap_locks is true, we take the i_mmap_rwsem and anon_vma
@@ -170,8 +171,16 @@ static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
* pte locks because exclusive mmap_lock prevents deadlock.
*/
old_pte = pte_offset_map_lock(mm, old_pmd, old_addr, &old_ptl);
- new_pte = pte_offset_map(new_pmd, new_addr);
- new_ptl = pte_lockptr(mm, new_pmd);
+ if (!old_pte) {
+ err = -EAGAIN;
+ goto out;
+ }
+ new_pte = pte_offset_map_nolock(mm, new_pmd, new_addr, &new_ptl);
+ if (!new_pte) {
+ pte_unmap_unlock(old_pte, old_ptl);
+ err = -EAGAIN;
+ goto out;
+ }
if (new_ptl != old_ptl)
spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
flush_tlb_batched_pending(vma->vm_mm);
@@ -208,8 +217,10 @@ static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
spin_unlock(new_ptl);
pte_unmap(new_pte - 1);
pte_unmap_unlock(old_pte - 1, old_ptl);
+out:
if (need_rmap_locks)
drop_rmap_locks(vma);
+ return err;
}

#ifndef arch_supports_page_table_move
@@ -537,6 +548,7 @@ unsigned long move_page_tables(struct vm_area_struct *vma,
new_pmd = alloc_new_pmd(vma->vm_mm, vma, new_addr);
if (!new_pmd)
break;
+again:
if (is_swap_pmd(*old_pmd) || pmd_trans_huge(*old_pmd) ||
pmd_devmap(*old_pmd)) {
if (extent == HPAGE_PMD_SIZE &&
@@ -544,8 +556,6 @@ unsigned long move_page_tables(struct vm_area_struct *vma,
old_pmd, new_pmd, need_rmap_locks))
continue;
split_huge_pmd(vma, old_pmd, old_addr);
- if (pmd_trans_unstable(old_pmd))
- continue;
} else if (IS_ENABLED(CONFIG_HAVE_MOVE_PMD) &&
extent == PMD_SIZE) {
/*
@@ -556,11 +566,13 @@ unsigned long move_page_tables(struct vm_area_struct *vma,
old_pmd, new_pmd, true))
continue;
}
-
+ if (pmd_none(*old_pmd))
+ continue;
if (pte_alloc(new_vma->vm_mm, new_pmd))
break;
- move_ptes(vma, old_pmd, old_addr, old_addr + extent, new_vma,
- new_pmd, new_addr, need_rmap_locks);
+ if (move_ptes(vma, old_pmd, old_addr, old_addr + extent,
+ new_vma, new_pmd, new_addr, need_rmap_locks) < 0)
+ goto again;
}

mmu_notifier_invalidate_range_end(&range);
--
2.35.3