[PATCH] mm: __delete_from_page_cache WARN_ON(page_mapped)

From: Hugh Dickins
Date: Sun Feb 28 2016 - 23:49:24 EST

Commit e1534ae95004 ("mm: differentiate page_mapped() from page_mapcount()
for compound pages") changed the famous BUG_ON(page_mapped(page)) in
__delete_from_page_cache() to VM_BUG_ON_PAGE(page_mapped(page)): which
gives us more info when CONFIG_DEBUG_VM=y, but nothing at all when not.

Although it has not usually been very helpul, being hit long after the
error in question, we do need to know if it actually happens on users'
systems; but reinstating a crash there is likely to be opposed :)

In the non-debug case, use WARN_ON() plus dump_page() and add_taint() -
I don't really believe LOCKDEP_NOW_UNRELIABLE, but that seems to be the
standard procedure now. Move that, or the VM_BUG_ON_PAGE(), up before
the deletion from tree: so that the unNULLified page->mapping gives a
little more information.

If the inode is being evicted (rather than truncated), it won't have
any vmas left, so it's safe(ish) to assume that the raised mapcount is
erroneous, and we can discount it from page_count to avoid leaking the
page (I'm less worried by leaking the occasional 4kB, than losing a
potential 2MB page with each 4kB page leaked).

Signed-off-by: Hugh Dickins <hughd@xxxxxxxxxx>
I think this should go into v4.5, so I've written it with an atomic_sub
on page->_count; but Joonsoo will probably want some page_ref thingy.

mm/filemap.c | 22 +++++++++++++++++++++-
1 file changed, 21 insertions(+), 1 deletion(-)

--- 4.5-rc6/mm/filemap.c 2016-02-28 09:04:38.816707844 -0800
+++ linux/mm/filemap.c 2016-02-28 19:45:23.406263928 -0800
@@ -195,6 +195,27 @@ void __delete_from_page_cache(struct pag
cleancache_invalidate_page(mapping, page);

+ VM_BUG_ON_PAGE(page_mapped(page), page);
+ if (!IS_ENABLED(CONFIG_DEBUG_VM) && WARN_ON(page_mapped(page))) {
+ int mapcount;
+ dump_page(page, "still mapped when deleted");
+ mapcount = page_mapcount(page);
+ if (mapping_exiting(mapping) &&
+ page_count(page) >= mapcount + 2) {
+ /*
+ * All vmas have already been torn down, so it's
+ * a good bet that actually the page is unmapped,
+ * and we'd prefer not to leak it: if we're wrong,
+ * some other bad page check should catch it later.
+ */
+ page_mapcount_reset(page);
+ atomic_sub(mapcount, &page->_count);
+ }
+ }
page_cache_tree_delete(mapping, page, shadow);

page->mapping = NULL;
@@ -205,7 +226,6 @@ void __delete_from_page_cache(struct pag
__dec_zone_page_state(page, NR_FILE_PAGES);
if (PageSwapBacked(page))
__dec_zone_page_state(page, NR_SHMEM);
- VM_BUG_ON_PAGE(page_mapped(page), page);

* At this point page must be either written or cleaned by truncate.