mm: fix clear_page_dirty_for_io vs fault race

Fix msync data loss and (less importantly) dirty page accounting
inaccuracies due to the race remaining in clear_page_dirty_for_io().

The deleted comment explains what the race was, and the added comments
explain how it is fixed.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

2 files changed, 21 insertions, 5 deletions

diff --git a/mm/memory.c b/mm/memory.c
index 61d51da7e17c..50dd3d1f4d18 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -1765,6 +1765,15 @@ gotten:
 unlock:
         pte_unmap_unlock(page_table, ptl);
         if (dirty_page) {
+                /*
+                 * Yes, Virginia, this is actually required to prevent a race
+                 * with clear_page_dirty_for_io() from clearing the page dirty
+                 * bit after it clear all dirty ptes, but before a racing
+                 * do_wp_page installs a dirty pte.
+                 *
+                 * do_no_page is protected similarly.
+                 */
+                wait_on_page_locked(dirty_page);
                 set_page_dirty_balance(dirty_page);
                 put_page(dirty_page);
         }
diff --git a/mm/page-writeback.c b/mm/page-writeback.c
index 886ea0d5a136..e62482718012 100644
--- a/mm/page-writeback.c
+++ b/mm/page-writeback.c
@@ -918,6 +918,8 @@ int clear_page_dirty_for_io(struct page *page)
 {
         struct address_space *mapping = page_mapping(page);
 
+        BUG_ON(!PageLocked(page));
+
         if (mapping && mapping_cap_account_dirty(mapping)) {
                 /*
                  * Yes, Virginia, this is indeed insane.
@@ -943,14 +945,19 @@ int clear_page_dirty_for_io(struct page *page)
                  * We basically use the page "master dirty bit"
                  * as a serialization point for all the different
                  * threads doing their things.
-                 *
-                 * FIXME! We still have a race here: if somebody
-                 * adds the page back to the page tables in
-                 * between the "page_mkclean()" and the "TestClearPageDirty()",
-                 * we might have it mapped without the dirty bit set.
                  */
                 if (page_mkclean(page))
                         set_page_dirty(page);
+                /*
+                 * We carefully synchronise fault handlers against
+                 * installing a dirty pte and marking the page dirty
+                 * at this point. We do this by having them hold the
+                 * page lock at some point after installing their
+                 * pte, but before marking the page dirty.
+                 * Pages are always locked coming in here, so we get
+                 * the desired exclusion. See mm/memory.c:do_wp_page()
+                 * for more comments.
+                 */
                 if (TestClearPageDirty(page)) {
                         dec_zone_page_state(page, NR_FILE_DIRTY);
                         return 1;