[PATCH] smaps: shared fix

The point of the smaps "shared" is to count the number of pages that are
mapped by more than one process, according to Mauricio Lin.  However, smaps
uses page_count for this, so it will return a false positive for every page
that is mapped by just that one process, which is also in pagecache or
swapcache.  There are false positive situations for anonymous pages not in
swapcache as well: - page reclaim, migration - get_user_pages (eg.
direct-io, ptrace)

Use page_mapcount instead, to count the number of mappings to the page.

Use vm_normal_page so that weird things like /dev/mem aren't counted either.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
index 35787f9..91b7c15 100644 (file)
--- a/fs/proc/task_mmu.c
+++ b/fs/proc/task_mmu.c
@@ -204,7 +204,6 @@ static void smaps_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
 {
        pte_t *pte, ptent;
        spinlock_t *ptl;
-       unsigned long pfn;
        struct page *page;
 
        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
@@ -214,12 +213,12 @@ static void smaps_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
                        continue;
 
                mss->resident += PAGE_SIZE;
-               pfn = pte_pfn(ptent);
-               if (!pfn_valid(pfn))
+
+               page = vm_normal_page(vma, addr, ptent);
+               if (!page)
                        continue;
 
-               page = pfn_to_page(pfn);
-               if (page_count(page) >= 2) {
+               if (page_mapcount(page) >= 2) {
                        if (pte_dirty(ptent))
                                mss->shared_dirty += PAGE_SIZE;
                        else