ongoworks:security
A Meteor package that provides a simple, logical, plain language API for defining write security on your MongoDB collections. Wraps the core allow/deny security.
SUMMARY OF CHANGES IN 2.0
- BREAKING: This package no longer automatically sets up
allow
anddeny
functions for you. This allows people to use it for theSecurity.can
feature but not necessarily enable client-side write operations. To keep the old behavior when updating to 2.0, simply change all of yourapply()
toallowInClientCode()
. Otherwise if you do not want it to set upallow
anddeny
functions for you, remove the.apply()
part of your chains. - You can track simple read security now using
Security.can(userId).read(doc).for(myCollection)
. For publications, though, it is usually a better idea to filter your queries rather than check security. - When using
defineMethod
, we now recommend passing anallow
function rather than adeny
function because it is easier to reason about. However, the internal logic has not changed; client-side writes are still secured usingdeny
functions if turned on. Also,deny
functions indefineMethod
continue to be supported for now. - When using
Security.can
, you can now pass theid
toupdate
,remove
, orread
rather than thedoc
. It had been documented that you could pass thatid
, but that did not actually work. Now you can pass either theid
or thedoc
and if you pass theid
, thedoc
will be retrieved, fetching only the properties required byfetch
.
Installation
$ meteor add ongoworks:security
How It Works
Call permit
to begin a new rule chain. Then optionally call one or more restriction methods. Here are some examples:
/server/security.js:
// Anyone may insert, update, or remove a post without restriction
Posts.permit(['insert', 'update', 'remove']);
// No one may insert, update, or remove posts
Posts.permit(['insert', 'update', 'remove']).never();
// Users may insert posts only if they are logged in
Posts.permit('insert').ifLoggedIn();
// Users may remove posts only if they are logged in with an "admin" role
Posts.permit('remove').ifHasRole('admin');
// Admin users may update any properties of any post, but regular users may
// update posts only if they don't try to change the `author` or `date` properties
Posts.permit('update').ifHasRole('admin');
Posts.permit('update').ifLoggedIn().exceptProps(['author', 'date']);
Built-In Rule Chain Methods
- never() - Prevents the database operations from untrusted code. Should be the same as not defining any rules, but it never hurts to be extra careful.
- ifLoggedIn() - Allows the database operations from untrusted code only when there is a logged in user.
- ifHasUserId(userId) - Allows the database operations from untrusted code only for the given user.
- ifHasRole(role) - Allows the database operations from untrusted code only for users with the given role. Using this method requires adding the
alanning:roles
package to your app. If you use role groups, an alternative syntax isifHasRole({role: role, group: group})
- onlyProps(props) - Allows the database operations from untrusted code for the given top-level doc properties only.
props
can be a string or an array of strings. - exceptProps(props) - Allows the database operations from untrusted code for all top-level doc properties except those specified.
props
can be a string or an array of strings.
Checking Your Rules in a Method
Security.can
allows you to check your rules in any server code.
Insert syntax:
if (Security.can(userId).insert(doc).for(MyCollection).check()) {}
// OR
Security.can(userId).insert(doc).for(MyCollection).throw();
If you pass additional arguments to insert
, they will be in the arguments passed to the allow
functions of the defined methods.
Update syntax:
if (Security.can(userId).update(id || currentDoc, modifier).for(MyCollection).check()) {}
// OR
Security.can(userId).update(id || currentDoc, modifier).for(MyCollection).throw();
If the first argument is an ID, the doc will be retrieved for you. If you already have the current document, pass that.
If you pass additional arguments to insert
, they will be in the arguments passed to the allow
functions of the defined methods.
Remove syntax:
if (Security.can(userId).remove(id || currentDoc).for(MyCollection).check()) {}
// OR
Security.can(userId).remove(id || currentDoc).for(MyCollection).throw();
If the first argument is an ID, the doc will be retrieved for you. If you already have the current document, pass that.
If you pass additional arguments to insert
, they will be in the arguments passed to the allow
functions of the defined methods.
Read syntax:
if (Security.can(userId).read(id || currentDoc).for(MyCollection).check()) {}
// OR
Security.can(userId).read(id || currentDoc).for(MyCollection).throw();
If the first argument is an ID, the doc will be retrieved for you. If you already have the current document, pass that.
If you pass additional arguments to insert
, they will be in the arguments passed to the allow
functions of the defined methods.
For example, say you have a method that will insert a post and then increment a counter in the user's document. You do not want to do either unless both are allowed by your security rules.
Meteor.methods({
insertPost: function (post) {
check(post, Schemas.Post);
Security.can(this.userId).insert(post).for(Post).throw();
var userModifier = {
$inc: {
postsCount: 1
}
};
Security.can(this.userId).update(this.userId, userModifier).for(Meteor.users).throw();
var postId = Post.insert(post);
Meteor.users.update(this.userId, userModifier);
return postId;
}
});
Applying Your Rules to Client-Side Writes
In Meteor, turning on client-side writes usually involves defining allow
and/or deny
functions. Instead, you can use your rules defined by this package as your allow
and deny
functions. To do so, add allowInClientCode()
to the end of your chain:
Posts.permit('update').ifLoggedIn().exceptProps(['author', 'date']).allowInClientCode();
This will automatically define the proper allow/deny rules for you.
API
Note: This entire API and all rule methods are available only in server code. As a security best practice, you should not define your security rules in client code or in server code that is sent to clients. Meteor allow/deny functions are documented as server-only functions, although they are currently available in client code, too.
Security.permit(types)
If you want to apply the same rule to multiple collections at once, you can do
Security.permit(['insert', 'update']).collections([Collection1, Collection2])...ruleChainMethods();
which is equivalent to
Collection1.permit(['insert', 'update'])...ruleChainMethods();
Collection2.permit(['insert', 'update'])...ruleChainMethods();
Security.defineMethod(name, definition)
Call Security.defineMethod
to define a method that may be used in the rule chain to restrict the current rule. Pass a definition
argument, which must contain a deny
property set to a deny
function for that rule. The deny
function is the same as the standard Meteor one, except that it receives a type
string as its first argument and the second argument is whatever the user passes to your method when calling it. The full function signature for inserts and removes is (type, arg, userId, doc)
and for updates is (type, arg, userId, doc, fields, modifier)
.
As an example, here is the definition for the built-in ifHasUserId
method:
Security.defineMethod('ifHasUserId', {
fetch: [],
transform: null,
allow(type, arg, userId) {
return userId === arg;
},
});
And here's an example of using the doc
property to create a method that can be used with Meteor.users
to check whether it's the current user's document:
Security.defineMethod('ifIsCurrentUser', {
fetch: [],
transform: null,
allow(type, arg, userId, doc) {
return userId === doc._id;
},
});
Transformations
If a rule is applied to a collection and that collection has a transform
function, the doc
received by your rule's deny function will be transformed. In most cases, you will want to prevent this by adding transform: null
to your rule definition. Alternatively, you can set transform
to a function in your rule definition, and that transformation will be run before calling the deny function.
Fetch
It's good practice to include fetch: []
in your rule definition, listing any fields you need for your deny logic. However, the fetch
option is not yet implemented. Currently all fields are fetched.
Security.Rule
An object of this type is returned throughout the rule chain.
Details
-
Simply adding this package to your app does not affect your app security in any way. Only calling
apply
on a rule chain for a collection will affect your app security. -
If you have not defined any rules for a collection, nothing is allowed (assuming you have removed the
insecure
package). -
It is fine and often necessary to apply more than one rule to the same collection. Each rule is evaluated separately, and at least one must pass.
-
You can mix 'n' match
allowInClientCode
rules with normalallow/deny
functions, but keep in mind that yourallow
functions may have no effect if you've called Securityapply
for the same collection. -
If you want to use this with the Meteor.users collections, use the Security.permit() syntax. Working example:
Security.permit(['insert', 'update', 'remove']).collections([ Meteor.users ]).never().allowInClientCode();
Logic
Rules within the same chain combine with AND. Multiple chains for the same collection combine with OR. In other words, at least one chain of rules must pass for the collection-operation combination. They are evaluated in the order they are defined. As soon as one passes for the collection-operation, no more are tested.
For example:
// You can remove a post if you have admin role
Posts.permit('remove').ifHasRole('admin');
// OR You can remove a post if you are logged in AND you created it AND it is not a Friday
Posts.permit('remove').ifLoggedIn().ifCreated().ifNotFriday();
// If neither of the above are true, the default behavior is to deny removal
Examples
Example 1
Here's how you might make your own rule that ensures the ownerId
property on a document is set to the current user.
Security.defineMethod('ownsDocument', {
fetch: [],
allow(type, field, userId, doc) {
if (!field) field = 'userId';
return userId === doc[field];
}
});
And then you can use it like this:
Posts.permit(['insert', 'update']).ownsDocument('ownerId');
Which means:
- "A user can insert a post from a client if they set themselves as the author"
- "A user can update a post from a client if they are currently set as the author."
Using with CollectionFS
This package supports the special "download" allow/deny for the CollectionFS packages, but you must use the Security.permit
syntax rather than myFSCollection.permit
. For example:
Security.permit(['insert', 'update', 'remove']).collections([Photos]).ifHasRole('admin').allowInClientCode();
Security.permit(['download']).collections([Photos]).ifLoggedIn().allowInClientCode();
Client/Common Code
You can call this package API in common code if you need to. When running on the client, the functions are simple stubs that do not actually do anything.
Contributing
You are welcome to submit pull requests if you have ideas for fixing or improving the API. If you come up with generally useful security rules, you should publish your own package that depends on this one and document the rules it provides so that others can use them. You may then submit a pull request to add a link to your package documentation in this readme.