mm: fix swapops.h:131 bug if remap_file_pages raced migration

Add remove_linear_migration_ptes_from_nonlinear(), to fix an interesting
little include/linux/swapops.h:131 BUG_ON(!PageLocked) found by trinity:
indicating that remove_migration_ptes() failed to find one of the
migration entries that was temporarily inserted.

The problem comes from remap_file_pages()'s switch from vma_interval_tree
(good for inserting the migration entry) to i_mmap_nonlinear list (no good
for locating it again); but can only be a problem if the remap_file_pages()
range does not cover the whole of the vma (zap_pte() clears the range).

remove_migration_ptes() needs a file_nonlinear method to go down the
i_mmap_nonlinear list, applying linear location to look for migration
entries in those vmas too, just in case there was this race.

The file_nonlinear method does need rmap_walk_control.arg to do this;
but it never needed vma passed in - vma comes from its own iteration.

Reported-and-tested-by: Dave Jones <davej@redhat.com>
Reported-and-tested-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

3 files changed, 36 insertions, 4 deletions

diff --git a/include/linux/rmap.h b/include/linux/rmap.h
index 1da693d51255..b66c2110cb1f 100644
--- a/include/linux/rmap.h
+++ b/include/linux/rmap.h
@@ -250,8 +250,7 @@ struct rmap_walk_control {
         int (*rmap_one)(struct page *page, struct vm_area_struct *vma,
                                         unsigned long addr, void *arg);
         int (*done)(struct page *page);
-        int (*file_nonlinear)(struct page *, struct address_space *,
-                                        struct vm_area_struct *vma);
+        int (*file_nonlinear)(struct page *, struct address_space *, void *arg);
         struct anon_vma *(*anon_lock)(struct page *page);
         bool (*invalid_vma)(struct vm_area_struct *vma, void *arg);
 };
diff --git a/mm/migrate.c b/mm/migrate.c
index b494fdb9a636..bed48809e5d0 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -178,6 +178,37 @@ out:
 }
 
 /*
+ * Congratulations to trinity for discovering this bug.
+ * mm/fremap.c's remap_file_pages() accepts any range within a single vma to
+ * convert that vma to VM_NONLINEAR; and generic_file_remap_pages() will then
+ * replace the specified range by file ptes throughout (maybe populated after).
+ * If page migration finds a page within that range, while it's still located
+ * by vma_interval_tree rather than lost to i_mmap_nonlinear list, no problem:
+ * zap_pte() clears the temporary migration entry before mmap_sem is dropped.
+ * But if the migrating page is in a part of the vma outside the range to be
+ * remapped, then it will not be cleared, and remove_migration_ptes() needs to
+ * deal with it.  Fortunately, this part of the vma is of course still linear,
+ * so we just need to use linear location on the nonlinear list.
+ */
+static int remove_linear_migration_ptes_from_nonlinear(struct page *page,
+                struct address_space *mapping, void *arg)
+{
+        struct vm_area_struct *vma;
+        /* hugetlbfs does not support remap_pages, so no huge pgoff worries */
+        pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
+        unsigned long addr;
+
+        list_for_each_entry(vma,
+                &mapping->i_mmap_nonlinear, shared.nonlinear) {
+
+                addr = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
+                if (addr >= vma->vm_start && addr < vma->vm_end)
+                        remove_migration_pte(page, vma, addr, arg);
+        }
+        return SWAP_AGAIN;
+}
+
+/*
  * Get rid of all migration entries and replace them by
  * references to the indicated page.
  */
@@ -186,6 +217,7 @@ static void remove_migration_ptes(struct page *old, struct page *new)
         struct rmap_walk_control rwc = {
                 .rmap_one = remove_migration_pte,
                 .arg = old,
+                .file_nonlinear = remove_linear_migration_ptes_from_nonlinear,
         };
 
         rmap_walk(new, &rwc);
diff --git a/mm/rmap.c b/mm/rmap.c
index d9d42316a99a..8fc049f9a5a6 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1360,8 +1360,9 @@ static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount,
 }
 
 static int try_to_unmap_nonlinear(struct page *page,
-                struct address_space *mapping, struct vm_area_struct *vma)
+                struct address_space *mapping, void *arg)
 {
+        struct vm_area_struct *vma;
         int ret = SWAP_AGAIN;
         unsigned long cursor;
         unsigned long max_nl_cursor = 0;
@@ -1663,7 +1664,7 @@ static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc)
         if (list_empty(&mapping->i_mmap_nonlinear))
                 goto done;
 
-        ret = rwc->file_nonlinear(page, mapping, vma);
+        ret = rwc->file_nonlinear(page, mapping, rwc->arg);
 
 done:
         mutex_unlock(&mapping->i_mmap_mutex);