mm: fix page_vma_mapped_walk() for ksm pages
authorHugh Dickins <hughd@google.com>
Fri, 7 Apr 2017 23:04:39 +0000 (16:04 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Sat, 8 Apr 2017 07:47:48 +0000 (00:47 -0700)
Doug Smythies reports oops with KSM in this backtrace, I've been seeing
the same:

  page_vma_mapped_walk+0xe6/0x5b0
  page_referenced_one+0x91/0x1a0
  rmap_walk_ksm+0x100/0x190
  rmap_walk+0x4f/0x60
  page_referenced+0x149/0x170
  shrink_active_list+0x1c2/0x430
  shrink_node_memcg+0x67a/0x7a0
  shrink_node+0xe1/0x320
  kswapd+0x34b/0x720

Just as observed in commit 4b0ece6fa016 ("mm: migrate: fix
remove_migration_pte() for ksm pages"), you cannot use page->index
calculations on ksm pages.

page_vma_mapped_walk() is relying on __vma_address(), where a ksm page
can lead it off the end of the page table, and into whatever nonsense is
in the next page, ending as an oops inside check_pte()'s pte_page().

KSM tells page_vma_mapped_walk() exactly where to look for the page, it
does not need any page->index calculation: and that's so also for all
the normal and file and anon pages - just not for THPs and their
subpages.  Get out early in most cases: instead of a PageKsm test, move
down the earlier not-THP-page test, as suggested by Kirill.

I'm also slightly worried that this loop can stray into other vmas, so
added a vm_end test to prevent surprises; though I have not imagined
anything worse than a very contrived case, in which a page mlocked in
the next vma might be reclaimed because it is not mlocked in this vma.

Fixes: ace71a19cec5 ("mm: introduce page_vma_mapped_walk()")
Link: http://lkml.kernel.org/r/alpine.LSU.2.11.1704031104400.1118@eggly.anvils
Signed-off-by: Hugh Dickins <hughd@google.com>
Reported-by: Doug Smythies <dsmythies@telus.net>
Tested-by: Doug Smythies <dsmythies@telus.net>
Reviewed-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/page_vma_mapped.c

index c4c9def8ffea47b4838fc3095221ee90e0c0fae3..de9c40d7304aa0e714bdd32abe79517ec3d73038 100644 (file)
@@ -111,12 +111,8 @@ bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw)
        if (pvmw->pmd && !pvmw->pte)
                return not_found(pvmw);
 
-       /* Only for THP, seek to next pte entry makes sense */
-       if (pvmw->pte) {
-               if (!PageTransHuge(pvmw->page) || PageHuge(pvmw->page))
-                       return not_found(pvmw);
+       if (pvmw->pte)
                goto next_pte;
-       }
 
        if (unlikely(PageHuge(pvmw->page))) {
                /* when pud is not present, pte will be NULL */
@@ -165,9 +161,14 @@ restart:
        while (1) {
                if (check_pte(pvmw))
                        return true;
-next_pte:      do {
+next_pte:
+               /* Seek to next pte only makes sense for THP */
+               if (!PageTransHuge(pvmw->page) || PageHuge(pvmw->page))
+                       return not_found(pvmw);
+               do {
                        pvmw->address += PAGE_SIZE;
-                       if (pvmw->address >=
+                       if (pvmw->address >= pvmw->vma->vm_end ||
+                           pvmw->address >=
                                        __vma_address(pvmw->page, pvmw->vma) +
                                        hpage_nr_pages(pvmw->page) * PAGE_SIZE)
                                return not_found(pvmw);