sdc-cloudapi
This repository is part of the Triton DataCenter project. See the contribution guidelines and general documentation at the main Triton project page.
CloudAPI is the HTTP API that customers use to interact with SmartDataCenter.
Adding CloudAPI to SDC
cloudapi
is not created by default during SDC setup. You can create it
by running in the root global zone (either inside COAL or on an SDC headnode
server):
sdcadm post-setup cloudapi
Development
A CloudAPI server should be running in a cloudapi zone after running the sdcadm command above. Alternatively, a more manual approach is:
git clone git@github.com:TritonDataCenter/sdc-cloudapi.git
cd sdc-cloudapi
git submodule update --init
make all
node main.js -f ./etc/cloudapi.config.json
Configuration file.
The configuration file ./etc/cloudapi.cfg
needs to be created before
the CloudAPI server can run. Consequently, this file is also required in
order to run the test suite.
There is an example cloudapi.coal.cfg
file in the repository, with the
default values every required variable should take if we were running the
tests on our development machine, which has access to a COAL setup.
Please remember that if you're trying to modify this file within an actual
cloudapi zone, the config file is created - and automatically updated - by the
config-agent
service using the template
file also in this repo
(sapi_manifests/cloudapi/template
) and the SAPI configuration values.
Testing
Before testing, you need to create an environment the tests can operate in, on the headnode you're using for tests. Assuming that you'll be testing using COAL's headnode, and that you've already created a cloudapi zone on that headnode, the easiest way to prepare the headnode for CloudAPI testing will be running the following from the global zone:
/zones/`vmadm lookup -1 alias=cloudapi0`/root/opt/smartdc/cloudapi/tools/coal-setup.sh
This script will hack DAPI for headnode provisioning, update imgapi to allow local custom images, and install some services, images packages required for thorough testing.
Once you've completed this process, run the following from within the cloudapi zone:
./test/runtests
The runtests
script does a safety check for a canary file before attempting
to run any tests, to prevent unwanted writes. If the canary is not found,
runtests
will let you know; create (e.g. using touch) the file and rerun
runtests
.
To run a specific test file, and not the entire test suite, use the -f flag with runtests. For example:
./test/runtests -f nics.test.js
This will run test/nics.test.js. If you want to run multiple test files, -f effectively globs too:
./test/runtests -f machines
This will run all the machines* test files in test/.
It is possible to run the test suite outside of a cloudapi zone, but this is
an increasingly unbeaten path. If you are so inclined, then ensure that
./etc/cloudapi.cfg
is set up appropriately and execute:
make test
But your life will be simpler if you stick to a cloudapi zone and use the
runtests
script; various config settings and environmental flags are set by
runtests
automatically.
Image management
If you want to test image management using COAL, the faster approach is to run the aforementioned coal-setup.sh script from the global zone. Amongst other things, local image management setup will be completed.
Testing RBAC
This section assumes your setup includes a reasonably recent version of UFDS and Mahi. If you're not sure, please update both to latest.
There's an utility script intented to speed up ENV setup for RBAC testing in your local setup. Assuming you want to test RBAC in COAL, you'll need to:
-
Setup CloudAPI zone (see above).
-
Add the account
account
and the useruser
, both with passwordjoypass123
, and both of them using the SSH key~/.ssh/id_rsa.pub
:./tools/create-account.sh headnode
-
Clone v7.3 or later branch of node-smartdc from https://github.com/TritonDataCenter/node-smartdc
-
Assuming you want to test in COAL, you should have the following ENV vars setup to operate as the account owner:
SDC_URL=https://<IP of cloudapi zone> SDC_TESTING=true SDC_ACCOUNT=account SDC_KEY_ID=`ssh-keygen -l -E md5 -f ~/.ssh/id_rsa.pub| awk '{print $2}' | tr -d '\n'|cut -c 5-`
And, in order to operate as the account user instead, you just need to add the ENV var:
SDC_USER=user
given we already created both with the same SSH key/fingerprint.
If you want to also test machines creation and the associated actions, you'll need to hack the setup the same way we do for testing:
/zones/`vmadm lookup -1 \
alias=cloudapi0`/root/opt/smartdc/cloudapi/tools/coal-setup.sh
For more information on RBAC you can check CloudAPI docs and the Access Control User Guide.
How CloudAPI Auth works using RBAC
Roles and Policies are used in CloudAPI to provide access control for accounts'
sub users. Authorization for account sub users is always made using HTTP
Signature. The following is a brief description of CloudAPI access control
process for sub users (all of this assuming account_mgmt
feature is enabled
and req.version >= 7.2.0
):
Identify request resource
CloudAPI identifies the name
of the resource
for the request. This can
be either a collection of resources or an individual one. While this usually
matches the request path, it's not always true. For example:
a. ListFirewallRules
: Firewal Rules Resource Collection: /:account/fwrules
.
b. GetFirewallRule
: Individual Firewall Rule Resource: /:account/fwrules/:fwruleid
c. EnableFirewallRule
: The same individual firewall Rule resource than for
GetFirewallRule
, identified by /:account/fwrules/:fwruleid
even when the
path for this request would be /:account/fwrules/:fwruleid/enable
.
It's to say, for a given individual resource, all actions happening over this
resource will share the name
which is the path for the main GetResource
request. For example, every action listed under the Machines
epigraph in
CloudAPI docs related to an individual machine will have the same resource,
"the machine", identified by /:account/machines/:machineid
, even when these
actions could be rename machine, enable firewall, add tags, create snapshot,
audit ...
Load resource role-tags when exist
Once the name
of the resource
for the current request has been identified,
CloudAPI checks if there are any role-tag
associated with the current resource.
(role-tag
s are just a set of one or more roles associated with the current
resource. CloudAPI customers can associate role-tag
to resources using the
names of the roles they want to provide some kind of access to the resource.)
role-tag
loading is done differently depending if the current resource is an
individual machine (given machines store role-tag
by themselves) or something
else. Everything but machines uses UFDS' sdcAccountResource
objectclass,
which has the following attributes:
dn: resource-uuid=:resource_uuid, uuid=:account_uuid, ou=users, o=smartdc
account: account_uuid
memberrole: [aRoleDN, anotherRoleDN, ...]
name: :resource_name
objectclass: sdcaccountresource
uuid: :resource_uuid
Behind the scences, CloudAPI "translates" the role DNs into their respective role objects.
For machines, given each machine may have a role_tag
member in VMAPI, which
is an array of roles' UUIDs, CloudAPI does exactly the same regarding role
translation from UUID into the collection of role objects.
(Please, note that, in order to be able to use machine role-tag
to handle
sub-user auth, we need to preload machine loading for all the machine related
requests).
In both cases, our request object will have the following properties:
req.resourcename = :resource_name
req.resource = {
name: req.resourcename,
account: req.account.uuid,
roles: [[ {role_object}[ , {role_object}, ... ] ]]
};
Ask MAHI to authorize/deny user access
When CloudAPI detects that the current request is being performed by an account
sub-user, it will load the sub-user active roles (i.e. user.default_roles
), and
will pass those, together with the current resource roles collected into the
previous step, to aperture
for user authorization. Additionally, the current
request path
, method
and route name
are also given to aperture.
What needs to happen for the user to get access to the current resource then?:
a. The user must have at least one of the roles assigned to the resource.
b. For these roles, at least one of the policies associated with them must have
a rule which allows the current request method for the given route name, for
example: CAN get AND head IF route::string = listusers