[PATCH 0/3 V3] libxfs: better cache allocation error handling

[Eric Sandeen <sandeen@redhat.com>]

OK, 3rd try.

This is to handle an underlying problem that an actual ENOMEM cache
node allocation failure is indistinguishable from "current cache full"
and so ENOMEM may send us into a pointless cache-expansion loop which
ultimately overflows the cache size variable, rather than just reporting
the bad news up the chain to callers.

Darrick suggested "I think I'd almost rather we change cache_node_allocate
to return int and the cache node as an outparam since we do that elsewhere"

on the last review cycle, and so I've done that here, I agree that it
is cleaner.

The first patch is maybe too granular but for some reason this stuff
hurts my brain a little and I thought it helped with clarity; it's a
no-op change to cache_node_allocate's signature, returning an int
and passing the cache node as a param.

Second patch should actually catch and bubble up any low-level error.

3rd patch is defensive against a case which in theory should almost never
happen after the other fixes.