AllowedBaseDomain will stay non-empty in certain error conditions. None of these conditions should be hit anyways, but this provides an extra safety check.
AllowedBaseDomain is only zero-ed out if the domain is not found in the (new) AllowedDomains configuration setting. If the domain is found, AllowedBaseDomain is not emptied and this code will be run every single time.
//untested
This fixes an issue where, due to clock skew, one system can get a cert
and try to use it before it thinks it's actually valid. The tolerance of
30 seconds should be high enough for pretty much any set of systems
using NTP.
Fixes#1035
1) Use the new LeaseExtend
2) Use default values controlled by mount tuning/system defaults instead
of a random hard coded value
3) Remove grace periods
This ensures that we can safely rotate and modify configuration
parameters with multiple requests in flight.
As a side effect we also get a cache, which should provide a nice
speedup since we don't need to decrypt/deserialize constantly, which
would happen even with the physical LRU.
This can be seen via System(). In the PKI backend, if the CA is
reconfigured but not fully (e.g. an intermediate CSR is generated but no
corresponding cert set) and there are already leases (issued certs), the
CRL is unable to be built. As a result revocation fails. But in this
case we don't actually need revocation to be successful since the CRL is
useless after unmounting. By checking taint status we know if we can
simply fast-path out of revocation with a success in this case.
Fixes#946
The secretAccessKeysRevoke revoke function now asserts that it is
not dealing with STS keys by checking a new internal data flag. Defaults
to IAM when the flag is not found.
Factored out genUsername into its own function to share between STS and
IAM secret creation functions.
Fixed bad call to "WriteOperation" instead of "UpdateOperation" in
aws/backend_test
