Acx is an access control library that can do whatever shit Casbin can and much more...
def deps do
[
{:acx, git: "https://github.com/ngoclinhng/acx.git"}
]
endLet's say we have just built a wonderful blogging system, and now we want
to add the access control feature to it to control who can do what with
the resource named blog_post. Our system requirements would look something
like this:
| blog_post.create | blog_post.read | blog_post.modify | blog_post.delete | |
|---|---|---|---|---|
| alice | yes | yes | yes | yes |
| bob | no | yes | no | yes |
| peter | yes | yes | yes | no |
Based on this requirements, our first step is to choose an appropriate access control model. Let's say we choose to go with the ACL model. Similar to Casbin, in Acx, an access control model is abstracted into a config file based on the PERM Meta-Model. The content of the config file for our system would look like so:
# blog_ac.conf
# We want each request to be a tuple of three items, in which first item
# associated with the attribute named `sub`, second `obj` and third `act`.
# An example of a valid request based on this definition is
# `["alice, "blog_post", "read"]` (can `alice` `read` `blog_post`?).
[request_definition]
r = sub, obj, act
# Each policy definition should have a key and a list of attributes separated by
# an equal `=` sign. In Acx all policy rules have in common the `eft` attribute
# and it can only take value of either `"allow"` or `"deny"`, so you can ommit
# it in your policy definition.
[policy_definition]
p = sub, obj, act
# Policy effect defines whether the access should be approved or denied
# if multiple policy rules match the request.
# We use the following policy effect for our blog system to mean that:
# if there's any matched policy rule of type `allow` (i.e `eft` == "allow"),
# the final effect is `allow`. Which also means if there's no match or all
# matches are of type `deny`, the final effect is `deny`.
[policy_effect]
e = some(where (p.eft == allow))
# matchers is just a boolean expression used to determine whether a request
# matches the given policy rule.
[matchers]
m = r.sub == p.sub && r.obj == p.obj && r.act == p.act
Done with the model. Our next step is to define policy rules based on the
system requirements and the policy definition. We can choose to put these
rules in a database or in our case a *.csv file named blog_ac.csv:
p, alice, blog_post, create
p, alice, blog_post, read
p, alice, blog_post, modify
p, alice, blog_post, delete
p, bob, blog_post, read
p, peter, blog_post, create
p, peter, blog_post, read
p, peter, blog_post, modify
Note that, first of all , since we don't specify the value for the eft
attribute for any of the above rules, all of our rules are of type allow
(a.k.a yes) by default. Second, we don't have to define any deny
(a.k.a no) rules for our system.
The final step is to combine the model, the policy rules and Acx to construct our access control system.
alias Acx.{EnforcerSupervisor, EnforcerServer}
# Give our system a name so that we can reference it by its name
# rather than the process ID (a.k.a `pid`).
ename = "blog_ac"
# Starts a new enforcer process and supervises it.
EnforcerSupervisor.start_enforcer(ename, blog_ac.conf)
# Load policy rules.
EnforcerServer.load_policies(ename, blog_ac.csv)
new_req = ["alice", "blog_post", "read"]
case EnforcerServer.allow?(ename, new_req) do
true ->
# Yes, this `new_req` is allowed
false ->
# Nope, `new_req` is denied (not allowed)
end
If you are not a fan of supervision tree or stateful server, read on to figure our how to use Acx without any of those.
Our ACL access control system is working just fine for initial purpose, but now our bussiness is expanding like nuts, so we need a more flexible access control model to meet new bussiness requirements. We went back to the drawing-board and came up with this design for our new system:
We assign different roles to different users, bob has the role reader,
peter has the role author and alice has the role admin, and so on...
We then define mappings from role to permission (instead of asking
who can do what like in the ACL model, now it's time to ask which role can
do what?). We also define mappings from role to role to represent
inheritance. In the above diagram, we have admin inherits from author,
which in turn inherits from reader.
Note that has role or inherits from relation is transitive.
Based on this design, the config file for our new model would look like so:
# blog_ac.conf
[request_definition]
r = sub, obj, act
[policy_definition]
p = sub, obj, act
# This is the name of the mapping we mentioned above, I call it `g`
# to make it compatible with Casbin (which for some reason only allows name
# like `g, g2, ...`) but you can name it whatever shit you like so long as
# you're consistent.
[role_definition]
g = _, _
[policy_effect]
e = some(where (p.eft == allow))
# We change this part `r.sub == p.sub` of our initial matcher expression to
# `g(r.sub, p.sub)` to mean that: if `r.sub` has role (or inherits from)
# `p.sub` and ... and ...
[matchers]
m = g(r.sub, p.sub) && r.obj == p.obj && r.act == p.act
And the content of the file blog_ac.csv now become:
p, reader, blog_post, read
p, author, blog_post, modify
p, author, blog_post, create
p, admin, blog_post, delete
g, bob, reader
g, peter, author
g, alice, admin
g, author, reader
g, admin, author
Finally:
alias Acx.{EnforcerSupervisor, EnforcerServer}
ename = "blog_ac"
EnforcerSupervisor.start_enforcer(ename, blog_ac.conf)
EnforcerServer.load_policies(ename, blog_ac.csv)
# You only have to add this new line to load mapping rules. Unlike Casbin
# Acx distinguishes from `normal` policy rules and `mapping` rules.
# We've just happended to put the two types of rules in the same `*.csv` file.
EnforcerServer.load_mapping_policies(ename, blog_ac.csv)
new_req = ["alice", "blog_post", "read"]
case EnforcerServer.allow?(ename, new_req) do
true ->
# Yes, this `new_req` is allowed
false ->
# Nope, `new_req` is denied (not allowed)
end
As you can see, the cost of swithching or upgrading to another access control mechanism is just as cheap as modifying the configuration.
TODO
The config file:
# restful_ac.conf
[request_definition]
r = sub, obj, act
[policy_definition]
p = sub, obj, act
[policy_effect]
e = some(where (p.eft == allow))
# The function named `match?` will be defined later in code.
[matchers]
m = r.sub == p.sub && match?(r.obj, p.obj) && match?(r.act, p.act)Policy rules restful_ac.csv:
p, alice, /alice_data/.*, GET
p, alice, /alice_data/resource1, POST
p, bob, /alice_data/resource2, GET
p, bob, /bob_data/.*, POST
p, peter, /peter_data, (GET)|(POST)
Code:
alias Acx.{EnforcerSupervisor, EnforcerServer}
ename = "restful_ac"
EnforcerSupervisor.start_enforcer(ename, restful_ac.conf)
EnforcerServer.load_policies(ename, restful_ac.csv)
# We have to define the `match?/2` and add it to our enforcer system.
# This anonymous function is idential to the built-in function
# `regex_match?/2`, but I redifine it here to illustrate the idea of
# how you can customize the system to meet your need.
fun = fn str, pattern ->
case Regex.compile("^#{pattern}$") do
{:error, _} ->
false
{:ok, r} ->
Regex.match?(r, str)
end
end
# add `fun` to our system. Note that the name `:match?` is an atom not
# a string.
EnforcerServer.add_fun(ename, {:match?, fun})
new_req = ["alice", "/alice_data/foo", "GET"]
case EnforcerServer.allow?(ename, new_req) do
true ->
# Yes, this `new_req` is allowed
false ->
# Nope, `new_req` is denied (not allowed)
end