Security

In the Memgraph Community Edition you can create users and assign them passwords, but all the other features regarding the security (such as assigning user roles, privileges, and label-based access control) are Enterprise features.

Most databases used in production have multiple users accessing and modifying data within the database. This might pose a serious security concern for the system administrators that wish to grant only certain privileges to certain users. A typical example would be an internal database of some company which tracks data about their employees. Naturally, only certain users of the database should be able to perform queries which modify that data.

At Memgraph, we provide the administrators with the option of granting, denying or revoking a certain set of privileges to some users or groups of users (i.e. users that are assigned a specific user role), thereby eliminating such security concerns.

Creating users

Creating a user can be done by executing the following command:

CREATE USER user_name [IDENTIFIED BY 'password'];

If the username is an email address, you need to enclose it in backticks (`):

CREATE USER `alice@memgraph.com` IDENTIFIED BY '0042'; 

If you want to create a username that includes a space character, you must set the --auth-user-or-role-regex flag to include spaces. For example, --auth-user-or-role-name-regex=[a-zA-Z0-9_.+\-@ ]+.

If the user should authenticate themselves on each session, i.e. provide their password on each session, the part within the brackets is mandatory. Otherwise, the password is set to null and the user will be allowed to log-in using any password, provided that they use the correct username.

To set or alter a user's password run the following command:

SET PASSWORD FOR user_name TO 'new_password';

Setting the users's password to null removes the password, allowing the user to log-in using any password:

SET PASSWORD FOR user_name TO null;

To delete a user run the following command:

DROP USER user_name;

Password encryption algorithm

Memgraph offers multiple password encryption algorithms:

  • BCrypt
  • SHA256
  • SHA256 with multiple iterations (currently set to 1024 iterations)

The above algorithms can be specified at runtime using the flag --password-encryption-algorithm with the appropriate values of bcrypt, sha256 or sha256-multiple.

BCrypt This algorithm is the default algorithm for password encryption. It's the most secure algorithm and has the best protection against brute-force attack. However, if you're running connecting multiple concurrent enterprise users with passwords at the same time, it may not be the best choice for you as you might experience slower performance. The performance is slower only during authentication of the users, and should not degrade once the connection has been established.

SHA256 and SHA256 with multiple iterations SHA256 is an encryption algorithm that's usually not used for password encryption but is fast and secure enough to offer optimal performance when running a lot of concurrent opening connections to Memgraph.

Authentication

memgraph-platform image

If you are using Docker and memgraph-platform image, you should pass the username and password environment variables when starting Memgraph:

docker run -p 7687:7687 -p 7444:7444 -p 3000:3000 -e MGCONSOLE="--username <username> --password <password>" memgraph/memgraph-platform

Example:

docker run -p 7687:7687 -p 7444:7444 -p 3000:3000 -e MGCONSOLE="--username vlasta --password vp" memgraph/memgraph-platform

Upon connecting with Memgraph Lab you should select Connect Manually and enter username (and password).

memgraph and memgraph-mage images

If you are using Docker and memgraph or memgraph-mage image enter username and password when connecting manually to Memgraph Lab.

If you are connecting with mgconsole you should add the username and password flags to the docker run command:

docker run -it --entrypoint=mgconsole memgraph/memgraph --host CONTAINER_IP --username=<username> --password=<password>

Example:

docker run -it --entrypoint=mgconsole memgraph/memgraph --host 172.17.0.2 --username=vlasta --password=vp

Owners

The privileges of the owners of streams and triggers are propagated to the corresponding query executions:

  • in case of streams for the queries returned by the transformations
  • in case of triggers for trigger statements.

This means the execution of the queries will fail if the owner doesn't have the required privileges. There are a few important details:

  • If there are no existing users, no privilege check is performed similarly to regular queries.
  • If a stream or trigger is created without using a logged-in user session, the owner will be Null. From the point when the first user is created such streams and triggers will fail because the lack of owner is treated as a user without any privileges, so no queries are allowed to be executed.
  • Currently, there is no way of changing the owner. The only workaround for this is to delete the stream or trigger and then create it again with another user.

The user who executes the CREATE STREAM query is going to be the owner of the stream. Authentication and authorization are not supported in Memgraph Community, thus the owner will always be Null, and the privileges are not checked in Memgraph Community. In Memgraph Enterprise the privileges of the owner are used when executing the queries returned from a transformation, in other words, the execution of the queries will fail if the owner doesn't have the required privileges.

Role-based access control (Enterprise)

Role-Based Access Control (RBAC) simplifies data security by grouping users into roles based on their tasks. Instead of assigning permissions to each user, RBAC assigns privileges to roles. Users, when linked to roles, gain the necessary access for their responsibilities. For example, in a company, a manager's role might have different access levels than an employee's role. Through RBAC, organizations efficiently ensure that users only access data relevant to their role, enhancing security and minimizing risks.

User roles

Each user can be assigned at most one user role. User roles are abstractions that capture the privilege levels of a set of users.

For example, suppose that Dominik and Marko belong to upper management of a certain company. It makes sense to grant them a set of privileges that other users are not entitled to so, instead of granting those privileges to each of them, we can create a role with those privileges called manager which we assign to Dominik and Marko.

In other words, each privilege that is granted to a user role is automatically granted to all the users with that role (unless it has been explicitly denied to that user). Similarly, each privilege that is denied to a user role is automatically denied to all users with that role (even if it has been explicitly granted to that user).

To creat a user role, run the following query:

CREATE ROLE role_name;

To assigning a user with a certain user role, run the following query:

SET ROLE FOR user_name TO role_name;

To remove the role from the user, run the following query:

CLEAR ROLE FOR user_name;

To show all users with a certain role:

SHOW USERS FOR role_name;

To show what role a user has, run the following query:

SHOW ROLE FOR user_name;

To list all defined user roles run:

SHOW ROLES;

Privileges

At the moment, privileges are confined to users' abilities to perform certain OpenCypher queries. Namely users can be given permission to execute a subset of the following commands:

Privilege descritpionClause
Privilege to access data.MATCH
Privilege to modify data.MERGE, SET
Privilege to create and [delete]](/querying/read-and-modify-data) data.CREATE, DELETE, REMOVE
Privilege to index data.INDEX
Privilege to obtain statistics and information from Memgraph.STATS
Privilege to view and alter users, roles and privileges.AUTH
Privilege to enforce constraints.CONSTRAINT
Privilege to dump the database.DUMP
Privilege to use replication queries.REPLICATION
Privilege to access files in queries, for example, when using LOAD CSV clause.READ_FILE
Privilege to manage durability files.DURABILITY
Privilege to try and free memory.FREE_MEMORY
Privilege to use trigger queries.TRIGGER
Privilege to configure Memgraph during runtime and to attain the configuration of the given Memgraph instance.CONFIG
Privilege to use stream queries.STREAM
Privilege to read the content of Python query module files.MODULE_READ
Privilege to modify the content of Python query modules files.MODULE_WRITE
Privilege to connect to Memgraph monitoring server.WEBSOCKET
Privilege to show and terminate transactions.TRANSACTION_MANAGEMENT
Privilege to change storage mode.STORAGE_MODE
Privilege to manage multi-tenant databases.MULTI_DATABASE_EDIT
Privilege to use a database within the multi-tenant architecture.MULTI_DATABASE_USE
Privileges to specific labels.ALL LABELS
Privileges to specific relationships types.ALL EDGE TYPES

After the first user is created, Memgraph will execute a query if and only if either a user or its role is granted that privilege and neither the user nor its role are denied that privilege. Otherwise, Memgraph will not execute that specific query. Note that DENY is a stronger operation than GRANT. This is also notable from the fact that if neither the user nor its role are explicitly granted or denied a certain privilege, that user will not be able to perform that specific query. This effect also is known as a silent deny. The information above is neatly condensed in the following table:

User StatusRole StatusEffective Status
GRANTGRANTGRANT
GRANTDENYDENY
GRANTNULLGRANT
DENYGRANTDENY
DENYDENYDENY
DENYNULLDENY
NULLGRANTGRANT
NULLDENYDENY
NULLNULLDENY

Once the privilages are changed, they take full effect once the user reconnects to the database.

Grant privileges

To grant a certain set of privileges to a specific user or user role, use the following query:

GRANT privilege_list TO user_or_role;

For example, to grant AUTH and INDEX privileges to users with the moderator role, run:

GRANT AUTH, INDEX TO moderator:

Deny privileges

Similarly, denying privileges is done using the DENY keyword instead of GRANT.

For example, to deny AUTH and INDEX privileges to users with the moderator role, run:

DENY AUTH, INDEX TO moderator:

Revoke privileges

Both denied and granted privileges can be revoked, meaning that their status is not defined for that user or role. Revoking is done using the REVOKE keyword.

For example, to revoke AUTH and INDEX privileges to users with the moderator role, run:

REVOKE AUTH, INDEX TO moderator:

Although semantically unintuitive, the level of a certain privilege can be raised by using REVOKE. For instance, if a user has been denied the INDEX privilege, but the role it belongs to is granted that privilege, the user is unable to use indexing features.

If the user's INDEX privilege is revoked, they will be able to do use indexing features because his role is granted that privilege.

Manage all privileges at once

To grant, deny or revoke all privileges use the ALL PRIVILEGES construct:

GRANT ALL PRIVILEGES FROM <user>;
DENY ALL PRIVILEGES FROM <user>;
REVOKE ALL PRIVILEGES FROM <user>;

Show privileges

To check privilege for a certain user or role, run the following query:

SHOW PRIVILEGES FOR user_or_role;

Label-based access control (Enterprise)

Sometimes, authorizing the database by granting and denying clause privileges is not enough to make the database fully secure. Certain nodes and relationships can be confidential and must be restricted from viewing and manipulating by multiple users. Also, disabling users from executing certain commands is sometimes too restrictive.

In response to the need for such authorization, Memgraph has advanced its authorization features to enable authorization on node labels and relationship edge types. By applying authorization to graph's first class citizens, a database administrator can now keep all the data in one database while keeping any private data secure from those who don't have adequate permission.

Label-based permissions are divided into 4 hierarchical parts or levels:

  • NOTHING - denies user visibility and manipulation over nodes and relationships
  • READ - grants the user visibility over nodes and relationships
  • UPDATE - grants the user visibility and the ability to edit nodes and relationships
  • CREATE_DELETE - grants the user visibility, editing, creation, and deletion of a node or a relationship

Node permissions

Granting a certain set of node permissions can be done similarly to the clause privileges using the following command:

GRANT permission_level ON LABELS label_list TO user_or_role;

with the legend:

  • permission_level is either NOTHING, READ, UPDATE or CREATE_DELETE
  • label_list is a set of node labels, separated with a comma and with a colon in front of each label (e.g. :L1), or * for specifying all labels in the graph
  • user_or_role is the already created user or role in Memgraph

For example, granting a READ permission on labels L1 and L2 would be written as:

GRANT READ ON LABELS :L1, :L2 TO charlie;

while granting both READ and EDIT permissions for all labels in the graph, would be written as:

GRANT UPDATE ON LABELS * TO charlie;

For denying visibility to a node, the command would be written as:

GRANT NOTHING ON LABELS :L1 TO charlie;

Relationship permissions

Relationship permission queries are in essence the same as node permission queries, with the one difference that the name of the relationship type is EDGE_TYPE and not LABEL.

Granting a certain set of edge type permissions can be done similarly to the clause privileges by issuing the following command:

GRANT permission_level ON EDGE_TYPES edge_type_list TO user_or_role;

with the same legend as the node permissions.

For example, granting a READ permission on relationship type :CONNECTS would be written as:

GRANT READ ON EDGE_TYPES :CONNECTS TO charlie;

Revoke label-based permissions

To revoke any of the label-based permissions, users can use one of the following commands:

REVOKE (LABELS | EDGE_TYPES) label_or_edge_type_list FROM user_or_role

where:

  • label_or_edge_type_list is a list of labels or edge types with a colon in front of each label or edge type (or * for specifying all labels or edge types)
  • user_or_role is the existing user or role in Memgraph

Show privileges for label-based access control

To check which privileges an existing user or role has in Memgraph, it is enough to write

SHOW PRIVILEGES FOR user_or_role;

and all the values of clause privileges, as well as label-based permissions will be displayed.

Templates for granting privileges

To grant all privileges to a superuser (admin):

GRANT ALL PRIVILEGES TO admin;
GRANT CREATE_DELETE ON LABELS * TO admin;
GRANT CREATE_DELETE ON EDGE_TYPES * TO admin;

To grant all read and write privileges:

DENY ALL PRIVILEGES TO readWrite;
GRANT CREATE, DELETE, MERGE, SET, REMOVE, INDEX, MATCH, STATS TO readWrite;
GRANT CREATE_DELETE ON LABELS * TO readWrite;
GRANT CREATE_DELETE ON EDGE_TYPES * TO readWrite;

To grant read only privileges:

DENY ALL PRIVILEGES TO readonly;
GRANT MATCH, STATS TO readonly;
GRANT READ ON LABELS * TO readonly;
GRANT READ ON EDGE_TYPES * TO readonly;

Examples

Below are several examples of using the Enterprise security features.

Example of granting read permissions

Bob is a data analyst for the company. He is making sure he can extract any useful insights from the data imported into the database. For now, all the data is labeled with the DataPoint label. Alice has already created a data analyst role as well as Bob's account in Memgraph with:

CREATE ROLE analyst;
CREATE USER Bob IDENTIFIED BY 'test';
SET ROLE FOR Bob TO analyst;

Unfortunately, when he writes:

MATCH (n:DataPoint) RETURN n;

he gets an error that he can not execute the query. Why is that? The first problem that we encounter is that Bob can not perform MATCH queries, which we must explicitly grant.

The database administrator grants him and all the data analysts the MATCH query to traverse the graph with:

GRANT MATCH TO analyst;

Now Bob is able to perform a match. However, by executing the same query again, he is not able to get any results.

Since Bob is not an administrator, he was not able to see any data points in the graph. In other words, he does not have READ permission on the DataPoint label.

Memgraph's label-based access control is hierarchically constructed, and the first permission one can be given on node labels or relationship edge types is READ.

Alice now updates Bob's permissions by executing:

GRANT READ ON LABELS :DataPoint TO analyst;

Bob is now executing his queries normally and is able to get insights from the database with respect to all the data points in the graph!

Additionally, in the company, it was decided that all the data points would be connected in a time series fashion, depending on when they were ingested into the database. One DataPoint should therefore be connected to the previously inserted one. The relationship type is called :NEXT.

Bob now again has problems, because when he executes:

MATCH (n:DataPoint)-[e:NEXT]->(m:DataPoint);

he is not able to see the patterns. Although Bob can see all the data points, he doesn't have permission to view the relationships. The database administrator executes the following command to solve the problem:

GRANT READ ON EDGE_TYPES :NEXT TO analyst;

Since the users are initially constructed without any permission, they would need an explicit grant for every new label that appears in the database. This approach is called whitelisting, and is more secure for adding new entities in the database since confidential nodes and relationships are not leaked into the database before securing them.

Granting update permissions

Charlie is a tester and customer care specialist. He is in charge of reporting bugs and fixing issues in the database. A common problem that he is facing is updating the classes of the data points if they are labeled incorrectly. For example, the class of one DataPoint might be 'dog', while in fact it is an 'elephant', but it was wrongly selected in the rush of labeling many data points. Charlie needs to update the wrongly labeled data points, and he already has the IDs of all the nodes he must update.

The administrator has already set up his account with the following commands:

CREATE ROLE tester;
CREATE USER Charlie IDENTIFIED BY 'test';
SET ROLE FOR Charlie TO tester;
 
GRANT MATCH, SET TO tester;
 
GRANT READ ON LABELS :DataPoint TO tester;
GRANT READ ON EDGE_TYPES :NEXT TO tester;

He now has read privileges just like all the data analysts, but when he gets an authorization error while executing:

MATCH (n:DataPoint {id:505}) SET n.labelY = 'elephant';

The error occurs because Charlie does not have permission to update the existing nodes in the graph. The database administrator needs to update Charlie's permissions and grant him access to update the node properties with:

GRANT UPDATE ON LABELS :DataPoint TO tester;

Charlie is now able to update the labeled categories of any data point in the graph! The same permission applies if he needs to update a relationship property in the graph.

Granting full access permissions

David is the data engineer for the company. He is very skilled in database systems, and he has been assigned the task of deleting every data point in the system that's older than one year. Alice has his account set up with the following commands:

CREATE ROLE dataEngineer;
CREATE USER David IDENTIFIED BY 'test';
SET ROLE FOR David TO dataEngineer;
 
GRANT MATCH, DELETE TO dataEngineer;
 
GRANT UPDATE ON LABELS :DataPoint TO dataEngineer;
GRANT UPDATE ON EDGE_TYPES :NEXT TO dataEngineer;

However, UPDATE privilege capabilities only grant manipulation of properties, not the nodes and relationships themselves. Therefore, the query:

MATCH (n:DataPoint) WHERE localDateTime() - n.date > Duration({day:365}) DETACH DELETE n;

results in an error. The permission that grants read, update, create, and delete rights over the nodes and relationships in the graph is CREATE_DELETE. By executing the following commands:

GRANT CREATE_DELETE ON LABELS :DataPoint TO dataEngineer;
GRANT CREATE_DELETE ON EDGE_TYPES :NEXT TO dataEngineer;

The permission is executed on relationships as well, since David needs to detach the nodes prior to deleting them. David is now able to successfully delete the deprecated nodes.

Denying visibility

Eve is the new senior engineer, and she is making excellent progress in the company. The management therefore decided to grant her visibility and manipulation over all the nodes. However, there are certain confidential nodes that are only for the management people to see.

Since there could be a lot of different node labels or relationship types in the database, a shortcut can be made by granting NOTHING to the entity. The database administrator therefore sets Eve's role as:

CREATE ROLE seniorEngineer;
CREATE USER Eve IDENTIFIED BY 'test';
SET ROLE FOR Eve TO seniorEngineer;
 
GRANT MATCH, DELETE TO seniorEngineer;
 
GRANT CREATE_DELETE ON LABELS * TO seniorEngineer;
GRANT NOTHING ON LABELS :SecretLabel TO seniorEngineer;

When granting NOTHING, the user is denied both visibility and manipulation of the entity. Eve is now able to see all the domain data while the management is happy since they have not leaked any confidential data.