On Sun, 24 Feb 2019, Mike Kravetz wrote: > > User can change a node specific hugetlb count. i.e. > > /sys/devices/system/node/node1/hugepages/hugepages-2048kB/nr_hugepages > > the calculated value of count is a total number of huge pages. It could > > be overflow when a user entering a crazy high value. If so, the total > > number of huge pages could be a small value which is not user expect. > > We can simply fix it by setting count to ULONG_MAX, then it goes on. This > > may be more in line with user's intention of allocating as many huge pages > > as possible. > > > > Signed-off-by: Jing Xiangfeng <jingxiangf...@huawei.com> > > Thank you. > > Acked-by: Mike Kravetz <mike.krav...@oracle.com> > > > --- > > mm/hugetlb.c | 7 +++++++ > > 1 file changed, 7 insertions(+) > > > > diff --git a/mm/hugetlb.c b/mm/hugetlb.c > > index afef616..6688894 100644 > > --- a/mm/hugetlb.c > > +++ b/mm/hugetlb.c > > @@ -2423,7 +2423,14 @@ static ssize_t __nr_hugepages_store_common(bool > > obey_mempolicy, > > * per node hstate attribute: adjust count to global, > > * but restrict alloc/free to the specified node. > > */ > > + unsigned long old_count = count; > > count += h->nr_huge_pages - h->nr_huge_pages_node[nid]; > > + /* > > + * If user specified count causes overflow, set to > > + * largest possible value. > > + */ > > + if (count < old_count) > > + count = ULONG_MAX; > > init_nodemask_of_node(nodes_allowed, nid); > > } else > > nodes_allowed = &node_states[N_MEMORY]; > >
Looks like this fixes the overflow issue, but isn't there already a possible underflow since we don't hold hugetlb_lock? Even if count == 0, what prevents h->nr_huge_pages_node[nid] being greater than h->nr_huge_pages here? I think the per hstate values need to be read with READ_ONCE() and stored on the stack to do any sane bounds checking.
