[PATCH] unpaged: anon in VM_UNPAGED

copy_one_pte needs to copy the anonymous COWed pages in a VM_UNPAGED area,
zap_pte_range needs to free them, do_wp_page needs to COW them: just like
ordinary pages, not like the unpaged.

But recognizing them is a little subtle: because PageReserved is no longer a
condition for remap_pfn_range, we can now mmap all of /dev/mem (whether the
distro permits, and whether it's advisable on this or that architecture, is
another matter).  So if we can see a PageAnon, it may not be ours to mess with
(or may be ours from elsewhere in the address space).  I suspect there's an
entertaining insoluble self-referential problem here, but the page_is_anon
function does a good practical job, and MAP_PRIVATE PROT_WRITE VM_UNPAGED will
always be an odd choice.

In updating the comment on page_address_in_vma, noticed a potential NULL
dereference, in a path we don't actually take, but fixed it.
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

mm/memory.c

@@ -349,6 +349,22 @@ void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
 	dump_stack();
 }
 
+/*
+ * page_is_anon applies strict checks for an anonymous page belonging to
+ * this vma at this address.  It is used on VM_UNPAGED vmas, which are
+ * usually populated with shared originals (which must not be counted),
+ * but occasionally contain private COWed copies (when !VM_SHARED, or
+ * perhaps via ptrace when VM_SHARED).  An mmap of /dev/mem might window
+ * free pages, pages from other processes, or from other parts of this:
+ * it's tricky, but try not to be deceived by foreign anonymous pages.
+ */
+static inline int page_is_anon(struct page *page,
+			struct vm_area_struct *vma, unsigned long addr)
+{
+	return page && PageAnon(page) && page_mapped(page) &&
+		page_address_in_vma(page, vma) == addr;
+}
+
 /*
  * copy one vm_area from one task to the other. Assumes the page tables
  * already present in the new task to be cleared in the whole range
@@ -381,23 +397,22 @@ copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
 		goto out_set_pte;
 	}
 
-	/* If the region is VM_UNPAGED, the mapping is not
-	 * mapped via rmap - duplicate the pte as is.
-	 */
-	if (vm_flags & VM_UNPAGED)
+	pfn = pte_pfn(pte);
+	page = pfn_valid(pfn)? pfn_to_page(pfn): NULL;
+
+	if (unlikely(vm_flags & VM_UNPAGED))
+		if (!page_is_anon(page, vma, addr))
 			goto out_set_pte;
 
-	pfn = pte_pfn(pte);
-	/* If the pte points outside of valid memory but
+	/*
+	 * If the pte points outside of valid memory but
 	 * the region is not VM_UNPAGED, we have a problem.
 	 */
-	if (unlikely(!pfn_valid(pfn))) {
+	if (unlikely(!page)) {
 		print_bad_pte(vma, pte, addr);
 		goto out_set_pte; /* try to do something sane */
 	}
 
-	page = pfn_to_page(pfn);
-
 	/*
 	 * If it's a COW mapping, write protect it both
 	 * in the parent and the child
@@ -568,17 +583,20 @@ static unsigned long zap_pte_range(struct mmu_gather *tlb,
 			continue;
 		}
 		if (pte_present(ptent)) {
-			struct page *page = NULL;
+			struct page *page;
+			unsigned long pfn;
 
 			(*zap_work) -= PAGE_SIZE;
 
-			if (!(vma->vm_flags & VM_UNPAGED)) {
-				unsigned long pfn = pte_pfn(ptent);
-				if (unlikely(!pfn_valid(pfn)))
+			pfn = pte_pfn(ptent);
+			page = pfn_valid(pfn)? pfn_to_page(pfn): NULL;
+
+			if (unlikely(vma->vm_flags & VM_UNPAGED)) {
+				if (!page_is_anon(page, vma, addr))
+					page = NULL;
+			} else if (unlikely(!page))
 				print_bad_pte(vma, ptent, addr);
-				else
-					page = pfn_to_page(pfn);
-			}
+
 			if (unlikely(details) && page) {
 				/*
 				 * unmap_shared_mapping_pages() wants to
@@ -1295,7 +1313,8 @@ static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
 	old_page = pfn_to_page(pfn);
 	src_page = old_page;
 
-	if (unlikely(vma->vm_flags & VM_UNPAGED)) {
+	if (unlikely(vma->vm_flags & VM_UNPAGED))
+		if (!page_is_anon(old_page, vma, address)) {
 			old_page = NULL;
 			goto gotten;
 		}

mm/rmap.c

@@ -225,7 +225,9 @@ vma_address(struct page *page, struct vm_area_struct *vma)
 
 /*
  * At what user virtual address is page expected in vma? checking that the
- * page matches the vma: currently only used by unuse_process, on anon pages.
+ * page matches the vma: currently only used on anon pages, by unuse_vma;
+ * and by extraordinary checks on anon pages in VM_UNPAGED vmas, taking
+ * care that an mmap of /dev/mem might window free and foreign pages.
  */
 unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
 {
@@ -234,7 +236,8 @@ unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
 		    (void *)page->mapping - PAGE_MAPPING_ANON)
 			return -EFAULT;
 	} else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) {
-		if (vma->vm_file->f_mapping != page->mapping)
+		if (!vma->vm_file ||
+		    vma->vm_file->f_mapping != page->mapping)
 			return -EFAULT;
 	} else
 		return -EFAULT;