[PATCH] mm: unlink_file_vma, remove_vma

Divide remove_vm_struct into two parts: first anon_vma_unlink plus
unlink_file_vma, to unlink the vma from the list and tree by which rmap or
vmtruncate might find it; then remove_vma to close, fput and free.

The intention here is to do the anon_vma_unlink and unlink_file_vma earlier,
in free_pgtables before freeing any page tables: so we can be sure that any
page tables traversed by rmap and vmtruncate are stable (and other, ordinary
cases are stabilized by holding mmap_sem).

This will be crucial to traversing pgd,pud,pmd without page_table_lock.  But
testing the split-out patch showed that lifting the page_table_lock is
symbiotically necessary to make this change - the lock ordering is wrong to
move those unlinks into free_pgtables while it's under ptlock.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

2 files changed, 28 insertions, 14 deletions

diff --git a/include/linux/mm.h b/include/linux/mm.h
index 376a466743bc..0c64484d8ae0 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -834,6 +834,7 @@ extern int split_vma(struct mm_struct *,
 extern int insert_vm_struct(struct mm_struct *, struct vm_area_struct *);
 extern void __vma_link_rb(struct mm_struct *, struct vm_area_struct *,
         struct rb_node **, struct rb_node *);
+extern void unlink_file_vma(struct vm_area_struct *);
 extern struct vm_area_struct *copy_vma(struct vm_area_struct **,
         unsigned long addr, unsigned long len, pgoff_t pgoff);
 extern void exit_mmap(struct mm_struct *);
diff --git a/mm/mmap.c b/mm/mmap.c
index eeefe19a0fac..a3984fad3fc2 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -181,26 +181,44 @@ static void __remove_shared_vm_struct(struct vm_area_struct *vma,
 }
 
 /*
- * Remove one vm structure and free it.
+ * Unlink a file-based vm structure from its prio_tree, to hide
+ * vma from rmap and vmtruncate before freeing its page tables.
  */
-static void remove_vm_struct(struct vm_area_struct *vma)
+void unlink_file_vma(struct vm_area_struct *vma)
 {
         struct file *file = vma->vm_file;
 
-        might_sleep();
         if (file) {
                 struct address_space *mapping = file->f_mapping;
                 spin_lock(&mapping->i_mmap_lock);
                 __remove_shared_vm_struct(vma, file, mapping);
                 spin_unlock(&mapping->i_mmap_lock);
         }
+}
+
+/*
+ * Close a vm structure and free it, returning the next.
+ */
+static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
+{
+        struct vm_area_struct *next = vma->vm_next;
+
+        /*
+         * Hide vma from rmap and vmtruncate before freeing page tables:
+         * to be moved into free_pgtables once page_table_lock is lifted
+         * from it, but until then lock ordering forbids that move.
+         */
+        anon_vma_unlink(vma);
+        unlink_file_vma(vma);
+
+        might_sleep();
         if (vma->vm_ops && vma->vm_ops->close)
                 vma->vm_ops->close(vma);
-        if (file)
-                fput(file);
-        anon_vma_unlink(vma);
+        if (vma->vm_file)
+                fput(vma->vm_file);
         mpol_free(vma_policy(vma));
         kmem_cache_free(vm_area_cachep, vma);
+        return next;
 }
 
 asmlinkage unsigned long sys_brk(unsigned long brk)
@@ -1612,15 +1630,13 @@ find_extend_vma(struct mm_struct * mm, unsigned long addr)
 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
 {
         do {
-                struct vm_area_struct *next = vma->vm_next;
                 long nrpages = vma_pages(vma);
 
                 mm->total_vm -= nrpages;
                 if (vma->vm_flags & VM_LOCKED)
                         mm->locked_vm -= nrpages;
                 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
-                remove_vm_struct(vma);
-                vma = next;
+                vma = remove_vma(vma);
         } while (vma);
         validate_mm(mm);
 }
@@ -1944,11 +1960,8 @@ void exit_mmap(struct mm_struct *mm)
          * Walk the list again, actually closing and freeing it
          * without holding any MM locks.
          */
-        while (vma) {
-                struct vm_area_struct *next = vma->vm_next;
-                remove_vm_struct(vma);
-                vma = next;
-        }
+        while (vma)
+                vma = remove_vma(vma);
 
         BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
 }