mm: fix migration hangs on anon_vma lock
authorHugh Dickins <hughd@google.com>
Thu, 13 Jan 2011 23:47:30 +0000 (15:47 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 14 Jan 2011 01:32:49 +0000 (17:32 -0800)
Increased usage of page migration in mmotm reveals that the anon_vma
locking in unmap_and_move() has been deficient since 2.6.36 (or even
earlier).  Review at the time of f18194275c39835cb84563500995e0d503a32d9a
("mm: fix hang on anon_vma->root->lock") missed the issue here: the
anon_vma to which we get a reference may already have been freed back to
its slab (it is in use when we check page_mapped, but that can change),
and so its anon_vma->root may be switched at any moment by reuse in
anon_vma_prepare.

Perhaps we could fix that with a get_anon_vma_unless_zero(), but let's
not: just rely on page_lock_anon_vma() to do all the hard thinking for us,
then we don't need any rcu read locking over here.

In removing the rcu_unlock label: since PageAnon is a bit in
page->mapping, it's impossible for a !page->mapping page to be anon; but
insert VM_BUG_ON in case the implementation ever changes.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Hugh Dickins <hughd@google.com>
Reviewed-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: "Jun'ichi Nomura" <j-nomura@ce.jp.nec.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: <stable@kernel.org> [2.6.37, 2.6.36]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/migrate.c

index 89a6bc8cd3072df5df12e9a52b952e33b5231db2..a20cf12edededf14c4a47a437d2e46252b9504f4 100644 (file)
@@ -622,7 +622,6 @@ static int unmap_and_move(new_page_t get_new_page, unsigned long private,
        int *result = NULL;
        struct page *newpage = get_new_page(page, private, &result);
        int remap_swapcache = 1;
-       int rcu_locked = 0;
        int charge = 0;
        struct mem_cgroup *mem = NULL;
        struct anon_vma *anon_vma = NULL;
@@ -694,20 +693,26 @@ static int unmap_and_move(new_page_t get_new_page, unsigned long private,
        /*
         * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
         * we cannot notice that anon_vma is freed while we migrates a page.
-        * This rcu_read_lock() delays freeing anon_vma pointer until the end
+        * This get_anon_vma() delays freeing anon_vma pointer until the end
         * of migration. File cache pages are no problem because of page_lock()
         * File Caches may use write_page() or lock_page() in migration, then,
         * just care Anon page here.
         */
        if (PageAnon(page)) {
-               rcu_read_lock();
-               rcu_locked = 1;
-
-               /* Determine how to safely use anon_vma */
-               if (!page_mapped(page)) {
-                       if (!PageSwapCache(page))
-                               goto rcu_unlock;
-
+               /*
+                * Only page_lock_anon_vma() understands the subtleties of
+                * getting a hold on an anon_vma from outside one of its mms.
+                */
+               anon_vma = page_lock_anon_vma(page);
+               if (anon_vma) {
+                       /*
+                        * Take a reference count on the anon_vma if the
+                        * page is mapped so that it is guaranteed to
+                        * exist when the page is remapped later
+                        */
+                       get_anon_vma(anon_vma);
+                       page_unlock_anon_vma(anon_vma);
+               } else if (PageSwapCache(page)) {
                        /*
                         * We cannot be sure that the anon_vma of an unmapped
                         * swapcache page is safe to use because we don't
@@ -722,13 +727,7 @@ static int unmap_and_move(new_page_t get_new_page, unsigned long private,
                         */
                        remap_swapcache = 0;
                } else {
-                       /*
-                        * Take a reference count on the anon_vma if the
-                        * page is mapped so that it is guaranteed to
-                        * exist when the page is remapped later
-                        */
-                       anon_vma = page_anon_vma(page);
-                       get_anon_vma(anon_vma);
+                       goto uncharge;
                }
        }
 
@@ -745,16 +744,10 @@ static int unmap_and_move(new_page_t get_new_page, unsigned long private,
         * free the metadata, so the page can be freed.
         */
        if (!page->mapping) {
-               if (!PageAnon(page) && page_has_private(page)) {
-                       /*
-                        * Go direct to try_to_free_buffers() here because
-                        * a) that's what try_to_release_page() would do anyway
-                        * b) we may be under rcu_read_lock() here, so we can't
-                        *    use GFP_KERNEL which is what try_to_release_page()
-                        *    needs to be effective.
-                        */
+               VM_BUG_ON(PageAnon(page));
+               if (page_has_private(page)) {
                        try_to_free_buffers(page);
-                       goto rcu_unlock;
+                       goto uncharge;
                }
                goto skip_unmap;
        }
@@ -768,14 +761,11 @@ skip_unmap:
 
        if (rc && remap_swapcache)
                remove_migration_ptes(page, page);
-rcu_unlock:
 
        /* Drop an anon_vma reference if we took one */
        if (anon_vma)
                drop_anon_vma(anon_vma);
 
-       if (rcu_locked)
-               rcu_read_unlock();
 uncharge:
        if (!charge)
                mem_cgroup_end_migration(mem, page, newpage);