In order to implement this efficiently, I have introduced the concept of
"singleton" backends -- currently, 'sys' and 'cubbyhole'. There isn't
much reason to allow sys to be mounted at multiple places, and there
isn't much reason you'd need multiple per-token storage areas. By
restricting it to just one, I can store that particular mount instead of
iterating through them in order to call the appropriate revoke function.
Additionally, because revocation on the backend needs to be triggered by
the token store, the token store's salt is kept in the router and
client tokens going to the cubbyhole backend are double-salted by the
router. This allows the token store to drive when revocation happens
using its salted tokens.
/cc @armon - This is a reasonably major refactor that I think cleans up
a lot of the logic with secrets in responses. The reason for the
refactor is that while implementing Renew/Revoke in logical/framework I
found the existing API to be really awkward to work with.
Primarily, we needed a way to send down internal data for Vault core to
store since not all the data you need to revoke a key is always sent
down to the user (for example the user than AWS key belongs to).
At first, I was doing this manually in logical/framework with
req.Storage, but this is going to be such a common event that I think
its something core should assist with. Additionally, I think the added
context for secrets will be useful in the future when we have a Vault
API for returning orphaned out keys: we can also return the internal
data that might help an operator.
So this leads me to this refactor. I've removed most of the fields in
`logical.Response` and replaced it with a single `*Secret` pointer. If
this is non-nil, then the response represents a secret. The Secret
struct encapsulates all the lease info and such.
It also has some fields on it that are only populated at _request_ time
for Revoke/Renew operations. There is precedent for this sort of
behavior in the Go stdlib where http.Request/http.Response have fields
that differ based on client/server. I copied this style.
All core unit tests pass. The APIs fail for obvious reasons but I'll fix
that up in the next commit.
Jeff Mitchell