[PATCH 2.5.73-mm1] Make sure truncate fix has no race

[Dave McCracken <dmccr@us.ibm.com>]

[-- Attachment #1: Type: text/plain, Size: 489 bytes --]


Paul McKenney pointed out that reading the truncate sequence number in
do_no_page might not be entirely safe if the ->nopage callout takes no
locks.  The simple solution is to move the read before the unlock of
page_table_lock.  Here's a patch that does it.

Dave McCracken

======================================================================
Dave McCracken          IBM Linux Base Kernel Team      1-512-838-3059
dmccr@us.ibm.com                                        T/L   678-3059

[-- Attachment #2: trunc-2.5.73-mm1-1.diff --]
[-- Type: text/plain, Size: 897 bytes --]

--- 2.5.73-mm1/mm/memory.c	2003-06-27 10:40:48.000000000 -0500
+++ 2.5.73-mm1-trunc/mm/memory.c	2003-06-27 10:47:10.000000000 -0500
@@ -1402,11 +1402,11 @@ do_no_page(struct mm_struct *mm, struct 
 		return do_anonymous_page(mm, vma, page_table,
 					pmd, write_access, address);
 	pte_unmap(page_table);
-	spin_unlock(&mm->page_table_lock);
 
 	mapping = vma->vm_file->f_dentry->d_inode->i_mapping;
-retry:
 	sequence = atomic_read(&mapping->truncate_count);
+	spin_unlock(&mm->page_table_lock);
+retry:
 	new_page = vma->vm_ops->nopage(vma, address & PAGE_MASK, 0);
 
 	/* no page was available -- either SIGBUS or OOM */
@@ -1441,6 +1441,7 @@ retry:
 	 * retry getting the page.
 	 */
 	if (unlikely(sequence != atomic_read(&mapping->truncate_count))) {
+		sequence = atomic_read(&mapping->truncate_count);
 		spin_unlock(&mm->page_table_lock);
 		page_cache_release(new_page);
 		goto retry;