Compliance, Mapped to Your Cloud

Use corporate login credentials instead of personal accounts, such as Gmail accounts. It is recommended fully-managed corporate Google accounts be used for increased visibility, auditing, and controlling access to Cloud Platform resources. Email accounts based outside of the user's organization, such as personal accounts, should not be used for business purposes.

User managed service accounts should not have user-managed keys. Anyone who has access to the keys will be able to access resources through the service account. GCP-managed keys are used by Cloud Platform services such as App Engine and Compute Engine. These keys cannot be downloaded. Google will keep the keys and automatically rotate them on an approximately weekly basis. User-managed keys are created, downloadable, and managed by users. They expire 10 years from creation. For user-managed keys, the user has to take ownership of key management activities which include: * Key storage * Key distribution * Key revocation * Key rotation * Protecting the keys from unauthorized users * Key recovery Even with key owner precautions, keys can be easily leaked by common development malpractices like checking keys into the source code or leaving them in the Downloads directory, or accidentally leaving them on support blogs/channels. It is recommended to prevent user-managed service account keys.

A service account is a special Google account that belongs to an application or a VM, instead of to an individual end-user. The application uses the service account to call the service's Google API so that users aren't directly involved. It's recommended not to use admin access for ServiceAccount. Service accounts represent service-level security of the Resources (application or a VM) which can be determined by the roles assigned to it. Enrolling ServiceAccount with Admin rights gives full access to an assigned application or a VM. A ServiceAccount Access holder can perform critical actions like delete, update change settings, etc. without user intervention. For this reason, it's recommended that service accounts not have Admin rights.

It is recommended to assign the `Service Account User (iam.serviceAccountUser)` and `Service Account Token Creator (iam.serviceAccountTokenCreator)` roles to a user for a specific service account rather than assigning the role to a user at project level. A service account is a special Google account that belongs to an application or a virtual machine (VM), instead of to an individual end-user. Application/VM-Instance uses the service account to call the service's Google API so that users aren't directly involved. In addition to being an identity, a service account is a resource that has IAM policies attached to it. These policies determine who can use the service account. Users with IAM roles to update the App Engine and Compute Engine instances (such as App Engine Deployer or Compute Instance Admin) can effectively run code as the service accounts used to run these instances, and indirectly gain access to all the resources for which the service accounts have access. Similarly, SSH access to a Compute Engine instance may also provide the ability to execute code as that instance/Service account. Based on business needs, there could be multiple user-managed service accounts configured for a project. Granting the `iam.serviceAccountUser` or `iam.serviceAserviceAccountTokenCreatorccountUser` roles to a user for a project gives the user access to all service accounts in the project, including service accounts that may be created in the future. This can result in elevation of privileges by using service accounts and corresponding `Compute Engine instances`. In order to implement `least privileges` best practices, IAM users should not be assigned the Service Account User or Service Account Token Creator roles at the project level. Instead, these roles should be assigned to a user for a specific service account, giving that user access to the service account. The `Service Account User` allows a user to bind a service account to a long-running job service, whereas the `Service Account Token Creator role` allows a user to directly impersonate (or assert) the identity of a service account.

Service Account keys consist of a key ID (Private\_key\_Id) and Private key, which are used to sign programmatic requests users make to Google cloud services accessible to that particular service account. It is recommended that all Service Account keys are regularly rotated. Rotating Service Account keys will reduce the window of opportunity for an access key that is associated with a compromised or terminated account to be used. Service Account keys should be rotated to ensure that data cannot be accessed with an old key that might have been lost, cracked, or stolen. Each service account is associated with a key pair managed by Google Cloud Platform (GCP). It is used for service-to-service authentication within GCP. Google rotates the keys daily.

It is recommended that the principle of 'Separation of Duties' is enforced while assigning service-account related roles to users. The built-in/predefined IAM role `Service Account admin` allows the user/identity to create, delete, and manage service account(s). The built-in/predefined IAM role `Service Account User` allows the user/identity (with adequate privileges on Compute and App Engine) to assign service account(s) to Apps/Compute Instances. Separation of duties is the concept of ensuring that one individual does not have all necessary permissions to be able to complete a malicious action. In Cloud IAM - service accounts, this could be an action such as using a service account to access resources that user should not normally have access to. Separation of duties is a business control typically used in larger organizations, meant to help avoid security or privacy incidents and errors. It is considered best practice. No user should have `Service Account Admin` and `Service Account User` roles assigned at the same time.

It is recommended that the IAM policy on Cloud KMS `cryptokeys` should restrict anonymous and/or public access. Granting permissions to `allUsers` or `allAuthenticatedUsers` allows anyone to access the dataset. Such access might not be desirable if sensitive data is stored at the location. In this case, ensure that anonymous and/or public access to a Cloud KMS `cryptokey` is not allowed.

Google Cloud Key Management Service stores cryptographic keys in a hierarchical structure designed for useful and elegant access control management. The format for the rotation schedule depends on the client library that is used. For the gcloud command-line tool, the next rotation time must be in `ISO` or `RFC3339` format, and the rotation period must be in the form `INTEGER[UNIT]`, where units can be one of seconds (s), minutes (m), hours (h) or days (d). Set a key rotation period and starting time. A key can be created with a specified `rotation period`, which is the time between when new key versions are generated automatically. A key can also be created with a specified next rotation time. A key is a named object representing a `cryptographic` key used for a specific purpose. The key material, the actual bits used for `encryption`, can change over time as new key versions are created. A key is used to protect some `corpus of data`. A collection of files could be encrypted with the same key and people with `decrypt` permissions on that key would be able to decrypt those files. Therefore, it's necessary to make sure the `rotation period` is set to a specific time.

It is recommended that the principle of 'Separation of Duties' is enforced while assigning KMS related roles to users. The built-in/predefined IAM role `Cloud KMS Admin` allows the user/identity to create, delete, and manage service account(s). The built-in/predefined IAM role `Cloud KMS CryptoKey Encrypter/Decrypter` allows the user/identity (with adequate privileges on concerned resources) to encrypt and decrypt data at rest using an encryption key(s). The built-in/predefined IAM role `Cloud KMS CryptoKey Encrypter` allows the user/identity (with adequate privileges on concerned resources) to encrypt data at rest using an encryption key(s). The built-in/predefined IAM role `Cloud KMS CryptoKey Decrypter` allows the user/identity (with adequate privileges on concerned resources) to decrypt data at rest using an encryption key(s). Separation of duties is the concept of ensuring that one individual does not have all necessary permissions to be able to complete a malicious action. In Cloud KMS, this could be an action such as using a key to access and decrypt data a user should not normally have access to. Separation of duties is a business control typically used in larger organizations, meant to help avoid security or privacy incidents and errors. It is considered best practice. No user(s) should have `Cloud KMS Admin` and any of the `Cloud KMS CryptoKey Encrypter/Decrypter`, `Cloud KMS CryptoKey Encrypter`, `Cloud KMS CryptoKey Decrypter` roles assigned at the same time.

API Keys should only be used for services in cases where other authentication methods are unavailable. In this case, unrestricted keys are insecure because they can be viewed publicly, such as from within a browser, or they can be accessed on a device where the key resides. It is recommended to restrict API key usage to trusted hosts, HTTP referrers and apps. It is recommended to use the more secure standard authentication flow instead. Security risks involved in using API-Keys appear below: * API keys are simple encrypted strings * API keys do not identify the user or the application making the API request * API keys are typically accessible to clients, making it easy to discover and steal an API key In light of these potential risks, Google recommends using the standard authentication flow instead of API keys. However, there are limited cases where API keys are more appropriate. For example, if there is a mobile application that needs to use the Google Cloud Translation API, but doesn't otherwise need a backend server, API keys are the simplest way to authenticate to that API. In order to reduce attack vectors, API-Keys can be restricted only to trusted hosts, HTTP referrers and applications.

API Keys should only be used for services in cases where other authentication methods are unavailable. API keys are always at risk because they can be viewed publicly, such as from within a browser, or they can be accessed on a device where the key resides. It is recommended to restrict API keys to use(call) only APIs required by an application. Security risks involved in using API-Keys appear below: * API keys are simple encrypted strings * API keys do not identify the user or the application making the API request * API keys are typically accessible to clients, making it easy to discover and steal an API key In light of these potential risks, Google recommends using the standard authentication flow instead of API keys. However, there are limited cases where API keys are more appropriate. For example, if there is a mobile application that needs to use the Google Cloud Translation API, but doesn't otherwise need a backend server, API keys are the simplest way to authenticate to that API. In order to reduce attack surfaces by providing `least privileges`, API-Keys can be restricted to use (call) only APIs required by an application.

API Keys should only be used for services in cases where other authentication methods are unavailable. If they are in use it is recommended to rotate API keys every 90 days. Security risks involved in using API-Keys appear below: * API keys are simple encrypted strings * API keys do not identify the user or the application making the API request * API keys are typically accessible to clients, making it easy to discover and steal an API key Because of these potential risks, Google recommends using the standard authentication flow instead of API Keys. However, there are limited cases where API keys are more appropriate. For example, if there is a mobile application that needs to use the Google Cloud Translation API, but doesn't otherwise need a backend server, API keys are the simplest way to authenticate to that API. Once a key is stolen, it has no expiration, meaning it may be used indefinitely unless the project owner revokes or regenerates the key. Rotating API keys will reduce the window of opportunity for an access key that is associated with a compromised or terminated account to be used. API keys should be rotated to ensure that data cannot be accessed with an old key that might have been lost, cracked, or stolen.

When you use Dataproc, cluster and job data is stored on Persistent Disks (PDs) associated with the Compute Engine VMs in your cluster and in a Cloud Storage staging bucket. This PD and bucket data is encrypted using a Google-generated data encryption key (DEK) and key encryption key (KEK). The CMEK feature allows you to create, use, and revoke the key encryption key (KEK). Google still controls the data encryption key (DEK). "Cloud services offer the ability to protect data related to those services using encryption keys managed by the customer within Cloud KMS. These encryption keys are called customer-managed encryption keys (CMEK). When you protect data in Google Cloud services with CMEK, the CMEK key is within your control.

It is recommended that Cloud Audit Logging is configured to track all admin activities and read, write access to user data. Cloud Audit Logging maintains two audit logs for each project, folder, and organization: Admin Activity and Data Access. 1. Admin Activity logs contain log entries for API calls or other administrative actions that modify the configuration or metadata of resources. Admin Activity audit logs are enabled for all services and cannot be configured. 2. Data Access audit logs record API calls that create, modify, or read user-provided data. These are disabled by default and should be enabled. There are three kinds of Data Access audit log information: * Admin read: Records operations that read metadata or configuration information. Admin activity audit logs record writes of metadata and configuration information that cannot be disabled. * Data read: Records operations that read user-provided data. * Data write: Records operations that write user-provided data. It is recommended to have an effective default audit config configured in such a way that: 1. Logtype is set to DATA\_READ (to log user activity tracking) and DATA\_WRITES (to log changes/tampering to user data). 2. Audit config is enabled for all the services supported by the Data Access audit logs feature. 3. Logs should be captured for all users, i.e., there are no exempted users in any of the audit config sections. This will ensure overriding the audit config will not contradict the requirement.

It is recommended to create a sink that will export copies of all the log entries. This can help aggregate logs from multiple projects and export them to a Security Information and Event Management (SIEM). Log entries are held in Cloud Logging. To aggregate logs, export them to a SIEM. To keep them longer, it is recommended to set up a log sink. Exporting involves writing a filter that selects the log entries to export, and choosing a destination in Cloud Storage, BigQuery, or Cloud Pub/Sub. The filter and destination are held in an object called a sink. To ensure all log entries are exported to sinks, ensure that there is no filter configured for a sink. Sinks can be created in projects, organizations, folders, and billing accounts.

Enabling retention policies on log buckets will protect logs stored in cloud storage buckets from being overwritten or accidentally deleted. It is recommended to set up retention policies and configure Bucket Lock on all storage buckets that are used as log sinks. Logs can be exported by creating one or more sinks that include a log filter and a destination. As cloud logging receives new log entries, they are compared against each sink. If a log entry matches a sink's filter, then a copy of the log entry is written to the destination. Sinks can be configured to export logs in storage buckets. It is recommended to configure a data retention policy for these cloud storage buckets and to lock the data retention policy; thus permanently preventing the policy from being removed. This way, if the system is ever compromised by an attacker or a malicious insider who wants to cover their tracks, the activity logs are definitely preserved for forensics and security investigations. Locking a bucket is an irreversible action. Once you lock a bucket, you cannot remove the retention policy from the bucket or decrease the retention period for the policy. You will then have to wait for the retention period for all items within the bucket before you can delete them, and then the bucket.

In order to prevent unnecessary project ownership assignments to users/service-accounts and further misuses of projects and resources, all `roles/Owner` assignments should be monitored. Members (users/Service-Accounts) with a role assignment to primitive role `roles/Owner` are project owners. The project owner has all the privileges on the project the role belongs to. These are summarized below: * All viewer permissions on all GCP Services within the project * Permissions for actions that modify the state of all GCP services within the project * Manage roles and permissions for a project and all resources within the project * Set up billing for a project Granting the owner role to a member (user/Service-Account) will allow that member to modify the Identity and Access Management (IAM) policy. Therefore,grant the owner role only if the member has a legitimate purpose to manage the IAM policy. This is because the project IAM policy contains sensitive access control data. Having a minimal set of users allowed to manage IAM policy will simplify any auditing that may be necessary.

Google Cloud Platform (GCP) services write audit log entries to the Admin Activity and Data Access logs to help answer the questions of, "who did what, where, and when?" within GCP projects. Cloud audit logging records information includes the identity of the API caller, the time of the API call, the source IP address of the API caller, the request parameters, and the response elements returned by GCP services. Admin activity and data access logs produced by cloud audit logging enable security analysis, resource change tracking, and compliance auditing. Configuring the metric filter and alerts for audit configuration changes ensures the recommended state of audit configuration is maintained so that all activities in the project are audit-able at any point in time.

It is recommended that a metric filter and alarm be established for changes to Identity and Access Management (IAM) role creation, deletion and updating activities. Google Cloud IAM provides predefined roles that give granular access to specific Google Cloud Platform resources and prevent unwanted access to other resources. However, to cater to organization-specific needs, IAM also provides the ability to create custom roles. Project owners and administrators with the Organization Role Administrator role or the IAM Role Administrator role can create custom roles. Monitoring role creation, deletion and updating activities will help in identifying any over-privileged role at early stages.

It is recommended that a metric filter and alarm be established for Virtual Private Cloud(VPC) Network Firewall rule changes. Monitoring for create or update firewall rule events gives insight to network access changes and may reduce the time it takes to detect suspicious activity.

It is recommended that a metric filter and alarm be established for Virtual Private Cloud (VPC) network route changes. Google Cloud Platform (GCP) routes define the paths network traffic takes from a VM instance to another destination. The other destination can be inside the organization VPC network (such as another VM) or outside of it. Every route consists of a destination and a next hop. Traffic whose destination IP is within the destination range is sent to the next hop for delivery. Monitoring changes to route tables will help ensure that all VPC traffic flows through an expected path.

It is recommended that a metric filter and alarm be established for Virtual Private Cloud (VPC) changes. It is possible to have more than one VPC within a project. In addition, it is also possible to create a peer connection between two VPCs enabling network traffic to route between VPCs. Monitoring changes to a VPC will help ensure VPC traffic flow is not getting impacted.

It is recommended that a metric filter and alarm be established for Cloud Storage Bucket IAM changes. Monitoring changes to cloud storage bucket permissions may reduce the time needed to detect and correct permissions on sensitive cloud storage buckets and objects inside the bucket.

It is recommended that a metric filter and alarm be established for SQL instance configuration changes. Monitoring changes to SQL instance configuration changes may reduce the time needed to detect and correct misconfigurations done on the SQL server. Below are a few of the configurable options which may the impact security posture of an SQL instance: * Enable auto backups and high availability: Misconfiguration may adversely impact business continuity, disaster recovery, and high availability * Authorize networks: Misconfiguration may increase exposure to untrusted networks

Cloud DNS logging records the queries from the name servers within your VPC to Stackdriver. Logged queries can come from Compute Engine VMs, GKE containers, or other GCP resources provisioned within the VPC. Security monitoring and forensics cannot depend solely on IP addresses from VPC flow logs, especially when considering the dynamic IP usage of cloud resources, HTTP virtual host routing, and other technology that can obscure the DNS name used by a client from the IP address. Monitoring of Cloud DNS logs provides visibility to DNS names requested by the clients within the VPC. These logs can be monitored for anomalous domain names, evaluated against threat intelligence , and Note: For full capture of DNS, firewall must block egress UDP/53 (DNS) and TCP/443 (DNS over HTTPS) to prevent client from using external DNS name server for resolution.

Logging enabled on a HTTPS Load Balancer will show all network traffic and its destination. Logging will allow you to view HTTPS network traffic to your web applications.

To prevent use of `default` network, a project should not have a `default` network. The `default` network has a preconfigured network configuration and automatically generates the following insecure firewall rules: * default-allow-internal: Allows ingress connections for all protocols and ports among instances in the network. * default-allow-ssh: Allows ingress connections on TCP port 22(SSH) from any source to any instance in the network. * default-allow-rdp: Allows ingress connections on TCP port 3389(RDP) from any source to any instance in the network. * default-allow-icmp: Allows ingress ICMP traffic from any source to any instance in the network. These automatically created firewall rules do not get audit logged and cannot be configured to enable firewall rule logging. Furthermore, the default network is an auto mode network, which means that its subnets use the same predefined range of IP addresses, and as a result, it's not possible to use Cloud VPN or VPC Network Peering with the default network. Based on organization security and networking requirements, the organization should create a new network and delete the `default` network.

Cloud Domain Name System (DNS) is a fast, reliable and cost-effective domain name system that powers millions of domains on the internet. Domain Name System Security Extensions (DNSSEC) in Cloud DNS enables domain owners to take easy steps to protect their domains against DNS hijacking and man-in-the-middle and other attacks. Domain Name System Security Extensions (DNSSEC) adds security to the DNS protocol by enabling DNS responses to be validated. Having a trustworthy DNS that translates a domain name like [www.example.com](https://hub.powerpipe.io/mods/[org]/[name]/controls/www.example.com) into its associated IP address is an increasingly important building block of today’s web-based applications. Attackers can hijack this process of domain/IP lookup and redirect users to a malicious site through DNS hijacking and man-in-the-middle attacks. DNSSEC helps mitigate the risk of such attacks by cryptographically signing DNS records. As a result, it prevents attackers from issuing fake DNS responses that may misdirect browsers to nefarious websites.

**NOTE**: Currently, the SHA1 algorithm has been removed from general use by Google, and, if being used, needs to be whitelisted on a project basis by Google and will also, therefore, require a Google Cloud support contract. DNSSEC algorithm numbers in this registry may be used in CERT RRs. Zone signing (DNSSEC) and transaction security mechanisms (SIG(0) and TSIG) make use of particular subsets of these algorithms. The algorithm used for key signing should be a recommended one and it should be strong. The algorithm used for key signing should be a recommended one and it should be strong. When enabling DNSSEC for a managed zone, or creating a managed zone with DNSSEC, the user can select the DNSSEC signing algorithms and the denial-of-existence type. Changing the DNSSEC settings is only effective for a managed zone if DNSSEC is not already enabled. If there is a need to change the settings for a managed zone where it has been enabled, turn DNSSEC off and then re-enable it with different settings.

**NOTE**: Currently, the SHA1 algorithm has been removed from general use by Google, and, if being used, needs to be whitelisted on a project basis by Google and will also, therefore, require a Google Cloud support contract. DNSSEC algorithm numbers in this registry may be used in CERT RRs. Zone signing (DNSSEC) and transaction security mechanisms (SIG(0) and TSIG) make use of particular subsets of these algorithms. The algorithm used for key signing should be a recommended one and it should be strong. The algorithm used for key signing should be a recommended one and it should be strong. When enabling DNSSEC for a managed zone, or creating a managed zone with DNSSEC, the user can select the DNSSEC signing algorithms and the denial-of-existence type. Changing the DNSSEC settings is only effective for a managed zone if DNSSEC is not already enabled. If there is a need to change the settings for a managed zone where it has been enabled, turn DNSSEC off and then re-enable it with different settings.

GCP `Firewall Rules` are specific to a `VPC Network`. Each rule either `allows` or `denies` traffic when its conditions are met. Its conditions allow the user to specify the type of traffic, such as ports and protocols, and the source or destination of the traffic, including IP addresses, subnets, and instances. Firewall rules are defined at the VPC network level and are specific to the network in which they are defined. The rules themselves cannot be shared among networks. Firewall rules only support IPv4 traffic. When specifying a source for an ingress rule or a destination for an egress rule by address, only an `IPv4` address or `IPv4 block in CIDR` notation can be used. Generic `(0.0.0.0/0)` incoming traffic from the internet to VPC or VM instance using `SSH` on `Port 22` can be avoided. GCP `Firewall Rules` within a `VPC Network` apply to outgoing (egress) traffic from instances and incoming (ingress) traffic to instances in the network. Egress and ingress traffic flows are controlled even if the traffic stays within the network (for example, instance-to-instance communication). For an instance to have outgoing Internet access, the network must have a valid Internet gateway route or custom route whose destination IP is specified. This route simply defines the path to the Internet, to avoid the most general `(0.0.0.0/0)` destination `IP Range` specified from the Internet through `SSH` with the default `Port 22`. Generic access from the Internet to a specific IP Range needs to be restricted.

GCP `Firewall Rules` are specific to a `VPC Network`. Each rule either `allows` or `denies` traffic when its conditions are met. Its conditions allow the user to specify the type of traffic, such as ports and protocols, and the source or destination of the traffic, including IP addresses, subnets, and instances. Firewall rules are defined at the VPC network level and are specific to the network in which they are defined. The rules themselves cannot be shared among networks. Firewall rules only support IPv4 traffic. When specifying a source for an ingress rule or a destination for an egress rule by address, only an `IPv4` address or `IPv4 block in CIDR` notation can be used. Generic `(0.0.0.0/0)` incoming traffic from the internet to VPC or VM instance using `RDP` on `Port 3389` can be avoided. GCP `Firewall Rules` within a `VPC Network` apply to outgoing (egress) traffic from instances and incoming (ingress) traffic to instances in the network. Egress and ingress traffic flows are controlled even if the traffic stays within the network (for example, instance-to-instance communication). For an instance to have outgoing Internet access, the network must have a valid Internet gateway route or custom route whose destination IP is specified. This route simply defines the path to the Internet, to avoid the most general `(0.0.0.0/0)` destination `IP Range` specified from the Internet through `RDP` with the default Port `3389`. Generic access from the Internet to a specific IP Range needs to be restricted.

Flow Logs is a feature that enables users to capture information about the IP traffic going to and from network interfaces in the organization's VPC Subnets. Once a flow log is created, the user can view and retrieve its data in Stackdriver Logging. It is recommended that Flow Logs be enabled for every business-critical VPC subnet. VPC networks and subnetworks not reserved for internal HTTP(S) load balancing provide logically isolated and secure network partitions where GCP resources can be launched. When Flow Logs are enabled for a subnet, VMs within that subnet start reporting on all Transmission Control Protocol (TCP) and User Datagram Protocol (UDP) flows. Each VM samples the TCP and UDP flows it sees, inbound and outbound, whether the flow is to or from another VM, a host in the on-premises datacenter, a Google service, or a host on the Internet. If two GCP VMs are communicating, and both are in subnets that have VPC Flow Logs enabled, both VMs report the flows. Flow Logs supports the following use cases: * Network monitoring * Understanding network usage and optimizing network traffic expenses * Network forensics * Real-time security analysis Flow Logs provide visibility into network traffic for each VM inside the subnet and can be used to detect anomalous traffic or provide insight during security workflows. The Flow Logs must be configured such that all network traffic is logged, the interval of logging is granular to provide detailed information on the connections, no logs are filtered, and metadata to facilitate investigations are included. **Note**: Subnets reserved for use by internal HTTP(S) load balancers do not support VPC flow logs.

It is recommended to configure your instance to not use the default Compute Engine service account because it has the Editor role on the project. The default Compute Engine service account has the Editor role on the project, which allows read and write access to most Google Cloud Services. To defend against privilege escalations if your VM is compromised and prevent an attacker from gaining access to all of your project, it is recommended to not use the default Compute Engine service account. Instead, you should create a new service account and assigning only the permissions needed by your instance. The default Compute Engine service account is named `[PROJECT_NUMBER]- compute@developer.gserviceaccount.com`.

To support principle of least privileges and prevent potential privilege escalation it is recommended that instances are not assigned to default service account `Compute Engine default service account` with Scope `Allow full access to all Cloud APIs`. Along with ability to optionally create, manage and use user managed custom service accounts, Google Compute Engine provides default service account `Compute Engine default service account` for an instances to access necessary cloud services. `Project Editor role` is assigned to `Compute Engine default service account` hence, This service account has almost all capabilities over all cloud services except billing. However, when `Compute Engine default service account` assigned to an instance it can operate in 3 scopes. 1. Allow default access: Allows only minimum access required to run an Instance (Least Privileges) 2. Allow full access to all Cloud APIs: Allow full access to all the cloud APIs/Services (Too much access) 3. Set access for each API: Allows Instance administrator to choose only those APIs that are needed to perform specific business functionality expected by instance When an instance is configured with `Compute Engine default service account` with Scope `Allow full access to all Cloud APIs`, based on IAM roles assigned to the user(s) accessing Instance, it may allow user to perform cloud operations/API calls that user is not supposed to perform leading to successful privilege escalation.

It is recommended to use Instance specific SSH key(s) instead of using common/shared project-wide SSH key(s) to access Instances. Project-wide SSH keys are stored in Compute/Project-meta-data. Project wide SSH keys can be used to login into all the instances within project. Using project-wide SSH keys eases the SSH key management but if compromised, poses the security risk which can impact all the instances within project. It is recommended to use Instance specific SSH keys which can limit the attack surface if the SSH keys are compromised.

Enabling OS login binds SSH certificates to IAM users and facilitates effective SSH certificate management. Enabling osLogin ensures that SSH keys used to connect to instances are mapped with IAM users. Revoking access to IAM user will revoke all the SSH keys associated with that particular user. It facilitates centralized and automated SSH key pair management which is useful in handling cases like response to compromised SSH key pairs and/or revocation of external/third-party/Vendor users.

Interacting with a serial port is often referred to as the serial console, which is similar to using a terminal window, in that input and output is entirely in text mode and there is no graphical interface or mouse support. If you enable the interactive serial console on an instance, clients can attempt to connect to that instance from any IP address. Therefore interactive serial console support should be disabled. A virtual machine instance has four virtual serial ports. Interacting with a serial port is similar to using a terminal window, in that input and output is entirely in text mode and there is no graphical interface or mouse support. The interactive serial console does not support IP-based access restrictions such as IP whitelists. If you enable the interactive serial console on an instance, clients can attempt to connect to that instance from any IP address. This allows anybody to connect to that instance if they know the correct SSH key, username, project ID, zone, and instance name.Therefore interactive serial console support should be disabled.

Compute Engine instance cannot forward a packet unless the source IP address of the packet matches the IP address of the instance. Similarly, GCP won't deliver a packet whose destination IP address is different than the IP address of the instance receiving the packet. However, both capabilities are required if you want to use instances to help route packets. Forwarding of data packets should be disabled to prevent data loss or information disclosure. Compute Engine instance cannot forward a packet unless the source IP address of the packet matches the IP address of the instance. Similarly, GCP won't deliver a packet whose destination IP address is different than the IP address of the instance receiving the packet. However, both capabilities are required if you want to use instances to help route packets. To enable this source and destination IP check, disable the canIpForward field, which allows an instance to send and receive packets with non-matching destination or source IPs.

Customer-Supplied Encryption Keys (CSEK) are a feature in Google Cloud Storage and Google Compute Engine. If you supply your own encryption keys, Google uses your key to protect the Google-generated keys used to encrypt and decrypt your data. By default, Google Compute Engine encrypts all data at rest. Compute Engine handles and manages this encryption for you without any additional actions on your part. However, if you wanted to control and manage this encryption yourself, you can provide your own encryption keys. If you provide your own encryption keys, Compute Engine uses your key to protect the Google-generated keys used to encrypt and decrypt your data. Only users who can provide the correct key can use resources protected by a customer-supplied encryption key. Google does not store your keys on its servers and cannot access your protected data unless you provide the key. This also means that if you forget or lose your key, there is no way for Google to recover the key or to recover any data encrypted with the lost key. At least business critical VMs should have VM disks encrypted with CSEK.

To defend against advanced threats and ensure that the boot loader and firmware on your VMs are signed and untampered, it is recommended that Compute instances are launched with Shielded VM enabled. Shielded VMs are virtual machines (VMs) on Google Cloud Platform hardened by a set of security controls that help defend against rootkits and bootkits. Integrity monitoring helps you understand and make decisions about the state of your VM instances and the Shielded VM vTPM enables Measured Boot by performing the measurements needed to create a known good boot baseline, called the integrity policy baseline. The integrity policy baseline is used for comparison with measurements from subsequent VM boots to determine if anything has changed. Secure Boot helps ensure that the system only runs authentic software by verifying the digital signature of all boot components, and halting the boot process if signature verification fails.

Compute instances should not be configured to have external IP addresses. To reduce your attack surface, Compute instances should not have public IP addresses. Instead, instances should be configured behind load balancers, to minimize the instance's exposure to the internet.

Google Cloud encrypts data at-rest and in-transit, but customer data must be decrypted for processing. Confidential Computing is a breakthrough technology which encrypts data in-use while it is being processed. Confidential Computing environments keep data encrypted in memory and elsewhere outside the central processing unit (CPU). Confidential VMs leverage the Secure Encrypted Virtualization (SEV) feature of AMD EPYC™ CPUs. Customer data will stay encrypted while it is used, indexed, queried, or trained on. Encryption keys are generated in hardware, per VM, and not exportable. Thanks to built-in hardware optimizations of both performance and security, there is no significant performance penalty to Confidential Computing workloads. Confidential Computing enables customers' sensitive code and other data encrypted in memory during processing. Google does not have access to the encryption keys. Confidential VM can help alleviate concerns about risk related to either dependency on Google infrastructure or Google insiders' access to customer data in the clear.

It is recommended that IAM policy on Cloud Storage bucket does not allows anonymous or public access. Allowing anonymous or public access grants permissions to anyone to access bucket content. Such access might not be desired if you are storing any sensitive data. Hence, ensure that anonymous or public access to a bucket is not allowed.

It is recommended that uniform bucket-level access is enabled on Cloud Storage buckets. It is recommended to use uniform bucket-level access to unify and simplify how you grant access to your Cloud Storage resources. Cloud Storage offers two systems for granting users permission to access your buckets and objects: Cloud Identity and Access Management (Cloud IAM) and Access Control Lists (ACLs). These systems act in parallel - in order for a user to access a Cloud Storage resource, only one of the systems needs to grant the user permission. Cloud IAM is used throughout Google Cloud and allows you to grant a variety of permissions at the bucket and project levels. ACLs are used only by Cloud Storage and have limited permission options, but they allow you to grant permissions on a per-object basis. In order to support a uniform permissioning system, cloud storage has uniform bucket- level access. Using this feature disables ACLs for all Cloud Storage resources: access to Cloud Storage resources then is granted exclusively through Cloud IAM. Enabling uniform bucket-level access guarantees that if a Storage bucket is not publicly accessible, no object in the bucket is publicly accessible either.

It is recommended to enforce all incoming connections to SQL database instance to use `SSL`. SQL database connections if successfully trapped (MITM); can reveal sensitive data like credentials, database queries, query outputs etc. For security, it is recommended to always use SSL encryption when connecting to your instance. This recommendation is applicable for Postgresql, MySql generation 1, MySql generation 2 and SQL Server 2017 instances.

Database Server should accept connections only from trusted Network(s)/IP(s) and restrict access from the world. To minimize attack surface on a Database server instance, only trusted/known and required IP(s) should be white-listed to connect to it. An authorized network should not have IPs/networks configured to `0.0.0.0/0` which will allow access to the instance from anywhere in the world. Note that authorized networks apply only to instances with public IPs.

It is recommended to configure Second Generation SQL instance to use private IPs instead of public IPs. To lower the organization's attack surface, Cloud SQL databases should not have public IPs. Private IPs provide improved network security and lower latency for your application.

It is recommended to have all SQL database instances set to enable automated backups. Backups provide a way to restore a Cloud SQL instance to recover lost data or recover from a problem with that instance. Automated backups need to be set for any instance that contains data that should be protected from loss or damage.This recommendation is applicable for SQL Server, PostgreSql, MySql generation 1 and MySql generation 2 instances.

It is recommended to set `skip_show_database` database flag for Cloud SQL MySQL instance to `on`. 'skip\_show\_database' database flag prevents people from using the SHOW DATABASES statement if they do not have the SHOW DATABASES privilege. This can improve security if you have concerns about users being able to see databases belonging to other users. Its effect depends on the SHOW DATABASES privilege: If the variable value is ON, the SHOW DATABASES statement is permitted only to users who have the SHOW DATABASES privilege, and the statement displays all database names. If the value is OFF, SHOW DATABASES is permitted to all users, but displays the names of only those databases for which the user has the SHOW DATABASES or other privilege. This recommendation is applicable to Mysql database instances.

It is recommended to set the `local_infile` database flag for a Cloud SQL MySQL instance to `off`. The `local_infile` flag controls the server-side LOCAL capability for LOAD DATA statements. Depending on the `local_infile` setting, the server refuses or permits local data loading by clients that have LOCAL enabled on the client side. To explicitly cause the server to refuse LOAD DATA LOCAL statements (regardless of how client programs and libraries are configured at build time or runtime), start mysqld with local\_infile disabled. local\_infile can also be set at runtime. Due to security issues associated with the `local_infile` flag, it is recommended to disable it. This recommendation is applicable to MySQL database instances.

The `log_error_verbosity` flag controls the verbosity/details of messages logged. Valid values are: * `TERSE` * `DEFAULT` * `VERBOSE` `TERSE` excludes the logging of `DETAIL`, `HINT`, `QUERY`, and `CONTEXT` error information. `VERBOSE` output includes the `SQLSTATE` error code, source code file name, function name, and line number that generated the error. Ensure an appropriate value is set to 'DEFAULT' or stricter. Auditing helps in troubleshooting operational problems and also permits forensic analysis. If `log_error_verbosity` is not set to the correct value, too many details or too few details may be logged. This flag should be configured with a value of 'DEFAULT' or stricter. This recommendation is applicable to PostgreSQL database instances.

Enabling the `log_connections` setting causes each attempted connection to the server to be logged, along with successful completion of client authentication. This parameter cannot be changed after the session starts. PostgreSQL does not log attempted connections by default. Enabling the `log_connections` setting will create log entries for each attempted connection as well as successful completion of client authentication which can be useful in troubleshooting issues and to determine any unusual connection attempts to the server. This recommendation is applicable to PostgreSQL database instances.

Enabling the `log_disconnections` setting logs the end of each session, including the session duration. PostgreSQL does not log session details such as duration and session end by default. Enabling the `log_disconnections` setting will create log entries at the end of each session which can be useful in troubleshooting issues and determine any unusual activity across a time period. The `log_disconnections` and `log_connections` work hand in hand and generally, the pair would be enabled/disabled together. This recommendation is applicable to PostgreSQL database instances.

The value of `log_statement` flag determined the SQL statements that are logged. Valid values are: * none * ddl * mod * all The value `ddl` logs all data definition statements. The value `mod` logs all ddl statements, plus data-modifying statements. The statements are logged after a basic parsing is done and statement type is determined, thus this does not logs statements with errors. When using extended query protocol, logging occurs after an Execute message is received and values of the Bind parameters are included. A value of 'ddl' is recommended unless otherwise directed by your organization's logging policy. Auditing helps in forensic analysis. If log\_statement is not set to the correct value, too many statements may be logged leading to issues in finding the relevant information from the logs, or too few statements may be logged with relevant information missing from the logs. Setting log\_statement to align with your organization's security and logging policies facilitates later auditing and review of database activities. This recommendation is applicable to PostgreSQL database instances.

The `log_min_messages` flag defines the minimum message severity level that is considered as an error statement. Messages for error statements are logged with the SQL statement. Valid values include `DEBUG5`, `DEBUG4`, `DEBUG3`, `DEBUG2`, `DEBUG1`, `INFO`, `NOTICE`, `WARNING`, `ERROR`, `LOG`, `FATAL`, and `PANIC`. Each severity level includes the subsequent levels mentioned above. ERROR is considered the best practice setting. Changes should only be made in accordance with the organization's logging policy. Auditing helps in troubleshooting operational problems and also permits forensic analysis. If `log_min_messages` is not set to the correct value, messages may not be classified as error messages appropriately. An organization will need to decide their own threshold for logging `log_min_messages` flag. This recommendation is applicable to PostgreSQL database instances.

The `log_min_error_statement` flag defines the minimum message severity level that are considered as an error statement. Messages for error statements are logged with the SQL statement. Valid values include `DEBUG5`, `DEBUG4`, `DEBUG3`, `DEBUG2`, `DEBUG1`, `INFO`, `NOTICE`, `WARNING`, `ERROR`, `LOG`, `FATAL`, and `PANIC`. Each severity level includes the subsequent levels mentioned above. Ensure a value of `ERROR` or stricter is set. Auditing helps in troubleshooting operational problems and also permits forensic analysis. If `log_min_error_statement` is not set to the correct value, messages may not be classified as error messages appropriately. Considering general log messages as error messages would make is difficult to find actual errors and considering only stricter severity levels as error messages may skip actual errors to log their SQL statements. The `log_min_error_statement` flag should be set to `ERROR` or stricter. This recommendation is applicable to PostgreSQL database instances.

The `log_min_duration_statement` flag defines the minimum amount of execution time of a statement in milliseconds where the total duration of the statement is logged. Ensure that `log_min_duration_statement` is disabled, i.e., a value of `-1` is set. Logging SQL statements may include sensitive information that should not be recorded in logs. This recommendation is applicable to PostgreSQL database instances.

Ensure `cloudsql.enable_pgaudit` database flag for Cloud SQL PostgreSQL instance is set to on to allow for centralized logging. As numerous other recommendations in this section consist of turning on flags for logging purposes, your organization will need a way to manage these logs. You may have a solution already in place. If you do not, consider installing and enabling the open source pgaudit extension within PostgreSQL and enabling its corresponding flag of `cloudsql.enable_pgaudit`. This flag and installing the extension enables database auditing in PostgreSQL through the open-source pgAudit extension. This extension provides detailed session and object logging to comply with government, financial, & ISO standards and provides auditing capabilities to mitigate threats by monitoring security events on the instance. Enabling the flag and settings later in this recommendation will send these logs to Google Logs Explorer so that you can access them in a central location. to This recommendation is applicable only to PostgreSQL database instances.

Instance addresses can be public IP or private IP. Public IP means that the instance is accessible through the public internet. In contrast, instances using only private IP are not accessible through the public internet, but are accessible through a Virtual Private Cloud (VPC). Limiting network access to your database will limit potential attacks. Setting databases access only to private will reduce attack surface.

It is recommended to `set external scripts` enabled database flag for Cloud SQL SQL Server instance to `off` `external scripts enabled` enable the execution of scripts with certain remote language extensions. This property is OFF by default. When Advanced Analytics Services is installed, setup can optionally set this property to true. As the External Scripts Enabled feature allows scripts external to SQL such as files located in an R library to be executed, which could adversely affect the security of the system, hence this should be disabled.This recommendation is applicable to SQL Server database instances.

It is recommended to set `cross db ownership chaining` database flag for Cloud SQL SQL Server instance to `off`. Use the `cross db ownership` for chaining option to configure cross-database ownership chaining for an instance of Microsoft SQL Server. This server option allows you to control cross-database ownership chaining at the database level or to allow cross-database ownership chaining for all databases.Enabling `cross db ownership` is not recommended unless all of the databases hosted by the instance of SQL Server must participate in crossdatabase ownership chaining and you are aware of the security implications of this setting.This recommendation is applicable to SQL Server database instances.

It is recommended to set `user connections` database flag for Cloud SQL SQL Server instance according organization-defined value. The `user connections` option specifies the maximum number of simultaneous user connections that are allowed on an instance of SQL Server. The actual number of user connections allowed also depends on the version of SQL Server that you are using, and also the limits of your application or applications and hardware. SQL Server allows a maximum of 32,767 user connections. Because user connections is by default a self- configuring value, with SQL Server adjusting the maximum number of user connections automatically as needed, up to the maximum value allowable. For example, if only 10 users are logged in, 10 user connection objects are allocated. In most cases, you do not have to change the value for this option. The default is 0, which means that the maximum (32,767) user connections are allowed. However if there is a number defined here that limits connections, SQL Server will not allow anymore above this limit. If the connections are at the limit, any new requests will be dropped potentially causing lost data or outages for those using the database.

It is recommended that, `user options` database flag for Cloud SQL SQL Server instance should not be configured. The `user options` option specifies global defaults for all users. A list of default query processing options is established for the duration of a user's work session. The user options option allows you to change the default values of the SET options (if the server's default settings are not appropriate). A user can override these defaults by using the SET statement. You can configure user options dynamically for new logins. After you change the setting of user options, new login sessions use the new setting; current login sessions are not affected. This recommendation is applicable to SQL Server database instances.

It is recommended to set `remote access` database flag for Cloud SQL SQL Server instance to `off`. The `remote access` option controls the execution of stored procedures from local or remote servers on which instances of SQL Server are running. This default value for this option is 1. This grants permission to run local stored procedures from remote servers or remote stored procedures from the local server.To prevent local stored procedures from being run from a remote server or remote stored procedures from being run on the local server, this must be disabled. The Remote Access option controls the execution of local stored procedures on remote servers or remote stored procedures on local server. 'Remote access' functionality can be abused to launch a Denial-of-Service (DoS) attack on remote servers by off-loading query processing to a target, hence this should be disabled. This recommendation is applicable to SQL Server database instances.

It is recommended to set `3625 (trace flag)` database flag for Cloud SQL SQL Server instance to `on`. Microsoft SQL Trace Flags are frequently used to diagnose performance issues or to debug stored procedures or complex computer systems, but they may also be recommended by Microsoft Support to address behavior that is negatively impacting a specific workload. All documented trace flags and those recommended by Microsoft Support are fully supported in a production environment when used as directed. `3625(trace log)` Limits the amount of information returned to users who are not members of the sysadmin fixed server role, by masking the parameters of some error messages using '**\*\***'. Setting this in a Google Cloud flag for the instance allows for security through obscurity and prevents the disclosure of sensitive information, hence this is recommended to set this flag globally to on to prevent the flag having been left off, or changed by bad actors. This recommendation is applicable to SQL Server database instances.

It is recommended to set `contained database authentication` database flag for Cloud SQL on the SQL Server instance is set to `off`. A contained database includes all database settings and metadata required to define the database and has no configuration dependencies on the instance of the Database Engine where the database is installed. Users can connect to the database without authenticating a login at the Database Engine level. Isolating the database from the Database Engine makes it possible to easily move the database to another instance of SQL Server.Contained databases have some unique threats that should be understood and mitigated by SQL Server Database Engine administrators. Most of the threats are related to the USER WITH PASSWORD authentication process, which moves the authentication boundary from the Database Engine level to the database level, hence this is recommended to disable this flag.This recommendation is applicable to SQL Server database instances.

It is recommended that the IAM policy on BigQuery datasets does not allow anonymous and/or public access. Granting permissions to `allUsers` or `allAuthenticatedUsers` allows anyone to access the dataset. Such access might not be desirable if sensitive data is being stored in the dataset. Therefore, ensure that anonymous and/or public access to a dataset is not allowed.

BigQuery by default encrypts the data as rest by employing `Envelope Encryption` using Google managed cryptographic keys. The data is encrypted using the `data encryption keys` and data encryption keys themselves are further encrypted using `key encryption keys`. This is seamless and do not require any additional input from the user. However, if you want to have greater control, Customer-managed encryption keys (CMEK) can be used as encryption key management solution for BigQuery Data Sets. If CMEK is used, the CMEK is used to encrypt the data encryption keys instead of using google-managed encryption keys. BigQuery stores the table and CMEK association and the encryption/decryption is done automatically. Applying the Default Customer-managed keys on BigQuery data sets ensures that all the new tables created in the future will be encrypted using CMEK but existing tables need to be updated to use CMEK individually. **Note: Google does not store your keys on its servers and cannot access your protected data unless you provide the key. This also means that if you forget or lose your key, there is no way for Google to recover the key or to recover any data encrypted with the lost key.**

BigQuery by default encrypts the data as rest by employing `Envelope Encryption` using Google managed cryptographic keys. The data is encrypted using the data encryption keys and `data encryption keys` themselves are further encrypted using `key encryption keys`. This is seamless and do not require any additional input from the user.However, if you want to have greater control, Customer-managed encryption keys (CMEK) can be used as encryption key management solution for BigQuery Data Sets. BigQuery by default encrypts the data as rest by employing `Envelope Encryption` using Google managed cryptographic keys. This is seamless and does not require any additional input from the user. For greater control over the encryption, customer-managed encryption keys (CMEK) can be used as encryption key management solution for BigQuery Data Sets. Setting a Default Customer-managed encryption key (CMEK) for a data set ensure any tables created in future will use the specified CMEK if none other is provided. **Note: Google does not store your keys on its servers and cannot access your protected data unless you provide the key. This also means that if you forget or lose your key, there is no way for Google to recover the key or to recover any data encrypted with the lost key.**