4.14-stable review patch. If anyone has any objections, please let me know.
------------------
From: Roman Gushchin guro@fb.com
commit d79f7aa496fc94d763f67b833a1f36f4c171176f upstream.
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/20180313130041.8078-1-guro@fb.com Signed-off-by: Roman Gushchin guro@fb.com Reviewed-by: Andrew Morton akpm@linux-foundation.org Cc: Alexander Viro viro@zeniv.linux.org.uk Cc: Michal Hocko mhocko@suse.com Cc: Johannes Weiner hannes@cmpxchg.org Signed-off-by: Andrew Morton akpm@linux-foundation.org Signed-off-by: Linus Torvalds torvalds@linux-foundation.org Signed-off-by: Greg Kroah-Hartman gregkh@linuxfoundation.org
--- mm/util.c | 7 +++++++ 1 file changed, 7 insertions(+)
--- a/mm/util.c +++ b/mm/util.c @@ -636,6 +636,13 @@ int __vm_enough_memory(struct mm_struct 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. */ if (free <= totalreserve_pages)