From: Herbert Xu <herbert@gondor.apana.org.au>
Date: Tue, 7 Feb 2017 20:38:27 +0800

This series tackles the problem of table allocation failures during
insertion.  The issue is that we cannot vmalloc during insertion.
This series deals with this by introducing nested tables.

The first two patches removes manual hash table walks which cannot
work on a nested table.

The final patch introduces nested tables.

I've tested this with test_rhashtable and it appears to work.

rhashtable is indeed getting quite complex, but I understand your
motivation for doing this work.

I also agree that only true OOM situations should cause insertion
failures.

Given the delicate nature of change rhashtable this way, I'm going
to let these changes sit for a couple days before applying them.

Thanks Herbert.