mm: treat indirectly reclaimable memory as free in overcommit logic

Indirectly reclaimable memory can consume a significant part of total
memory and it's actually reclaimable (it will be released under actual
memory pressure).

So, the overcommit logic should treat it as free.

Otherwise, it's possible to cause random system-wide memory allocation
failures by consuming a significant amount of memory by indirectly
reclaimable memory, e.g.  dentry external names.

If overcommit policy GUESS is used, it might be used for denial of
service attack under some conditions.

The following program illustrates the approach.  It causes the kernel to
allocate an unreclaimable kmalloc-256 chunk for each stat() call, so
that at some point the overcommit logic may start blocking large
allocation system-wide.

  int main()
  {
  	char buf[256];
  	unsigned long i;
  	struct stat statbuf;

  	buf[0] = '/';
  	for (i = 1; i < sizeof(buf); i++)
  		buf[i] = '_';

  	for (i = 0; 1; i++) {
  		sprintf(&buf[248], "%8lu", i);
  		stat(buf, &statbuf);
  	}

  	return 0;
  }

This patch in combination with related indirectly reclaimable memory
patches closes this issue.

Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Roman Gushchin <[email protected]>
Reviewed-by: Andrew Morton <[email protected]>
Cc: Alexander Viro <[email protected]>
Cc: Michal Hocko <[email protected]>
Cc: Johannes Weiner <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mm/util.c

@@ -667,6 +667,13 @@ int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
 		 */
 		free += global_node_page_state(NR_SLAB_RECLAIMABLE);
 
+		/*
+		 * Part of the kernel memory, which can be released
+		 * under memory pressure.
+		 */
+		free += global_node_page_state(
+			NR_INDIRECTLY_RECLAIMABLE_BYTES) >> PAGE_SHIFT;
+
 		/*
 		 * Leave reserved pages. The pages are not for anonymous pages.
 		 */