Overview of the Unified Data Model

Supported in:

This document provides an overview of the Unified Data Model (UDM). For more detail about UDM fields, including a description of each, see the UDM field list. For more information about parser mapping, see Important UDM fields for parser mapping.

The UDM is a Google Security Operations standard data structure that stores information about data received from sources. It is also called the schema. Google Security Operations stores the original data it receives in two formats, as the original raw log and as a structured UDM record. The UDM record is a structured representation of the original log.

If a parser exists for the specified log type, the raw log is used to create a UDM record. Customers can also transform raw logs to structured UDM format before sending the data to Google Security Operations using the Ingestion API.

Some of the benefits of UDM include:

  • Stores the same type of record from different vendors using the same semantics.
  • Easier to identify relationships between users, hosts, and IP addresses because the data is normalized into the standard UDM schema.
  • Easier to write rules since the rules can be platform-independent.
  • Easier to support log types from new devices.

Although you can search for events with a raw log search, a UDM search works faster and with more precision because of its specificity.

Google Security Operations uses the UDM schema for all of the events it collects. UDM includes thousands of fields that are used to describe and categorize events. For example, UDM can handle endpoint process events as easily as network communication events.

UDM structure

UDM events are made up of multiple sections. The first section found in every UDM event is the metadata section. It provides a basic description of the event, including the timestamp when the event occurred and the timestamp when it was ingested into Google Security Operations. It also includes the product information, version, and description. The ingestion parser classifies each event based on a predefined event type, independently of the specific product log. With the metadata fields alone, you can quickly start searching the data.

In addition to the metadata section, other sections describe additional aspects of the event. If a section is unnecessary, it isn't included, saving memory.

  • principal: Entity which originates the activity in the event. Sections that reference the source (src) and destination (target) are also included.
  • intermediary: Systems that events pass through, like a proxy server or an SMTP relay.
  • observer: Systems such as packet sniffers that passively monitor through traffic.

Example UDM searches

This section provides examples of UDM searches that demonstrate some of the basic syntax, features, and capabilities of UDM search.

Example: search for successful Microsoft Windows 4624 logins

The following search lists the Microsoft Windows 4624 successful login events, along with when the events were generated, based on just two UDM fields:

metadata.event_type = "USER_LOGIN" AND metadata.product_event_type = "4624"

Example: search for all successful logins

The following search lists all successful login events, regardless of vendor or application:

metadata.event_type = "USER_LOGIN" AND security_result.action = "ALLOW" AND target.user.userid != "SYSTEM" AND target.user.userid != /.*$/

Example: search for successful user logins

The following example illustrates how to search for userid "fkolzig" and determine when the user with this user ID successfully logged in. You can complete this search by using the target section. The target section includes subsections and fields describing the target. For example, the target in this case is a user and has a number of associated attributes, but the target could also be a file, a registry setting, or an asset. This example searches for "fkolzig" using the target.user.userid field.

metadata.event_type = "USER_LOGIN" AND metadata.product_event_type = "4624" AND target.user.userid = "fkolzig"

Example: search your network data

The following example searches network data for RDP events with a target.port

of 3389 and a principal.ip of 35.235.240.5. It also includes a UDM field from the network section, the direction of the data (network.direction).

metadata.product_event_type = "3" AND target.port = 3389 AND network.direction = "OUTBOUND" and principal.ip = "35.235.240.5"

Example: search for a specific process

To examine the processes created on your servers, search for instances of the net.exe command and search for this specific file in its expected path. The field you are searching for is target.process.file.full_path. For this search, you include the specific command issued in the target.process.command_line

field. You can also add a field in the about section which is the description of Microsoft Sysmon event code 1 (ProcessCreate).

Here is the UDM search:

metadata.product_event_type = "1" AND target.process.file.full_path = "C:\Windows\System32\net.exe"

Optionally, you could add the following UDM search fields:

  • principal.user.userid: Identify the user issuing the command.
  • principal.process.file.md5: Identify the MD5 hash.
  • principal.process.command_line: Command line.

Example: search for successful user logins associated with a specific department

The following example searches for logins by users (metadata.event_type is USER_LOGIN) associated with the marketing department (target.user.department is marketing) of your enterprise. Although target.user.department is not directly connected with user login events, it is still present in the LDAP data ingested about your users.

metadata.event_type = "USER_LOGIN" AND target.user.department = "Marketing"

Logical objects: Event and Entity

The UDM schema describes all available attributes that store data. Each UDM record identifies whether it describes an Event or Entity. Data is stored in different fields depending on whether the record describes an Event versus an Entity and also which value is set in the metadata.event_type or metadata.entity_type field.

  • UDM Event: Stores data for an action that occurred in the environment. The original event log describes the action as it was recorded by the device, such as firewall and web proxy. This is the UDM Event data model.
  • UDM Entity: Contextual representation of elements such as assets, users, and resources in your environment. It is obtained from a source of truth data source. This is the UDM Entity data model.

Here are two high level visual representations of the Event data model and the Entity data model.

Event data model

Figure: Event data model

Entity data model

Figure: Entity data model

Structure of a UDM Event

The UDM Event contains multiple sections that each store a subset of the data for a single record. The sections are:

  • metadata
  • principal
  • target
  • src
  • observer
  • intermediary
  • about
  • network
  • security_result
  • extensions

    Event data
model

    Figure: Event data model

The metadata section stores the timestamp, defines the event_type, and describes the device.

The principal, target, src, observer, and intermediary sections store information about the objects involved in the event. An object could be a device, user, or process. Most of the time, only a subset of these sections are used. The fields that store data are determined by the type of event and the role that each object plays in the event.

The network section stores information related to network activity, such as email and network related communication.

  • Email data: Information in the to, from, cc, bcc, and other email fields.
  • HTTP data: Such as method, referral_url, and useragent.

The security_result section stores an action or classification recorded by a security product, such as an anti-virus product.

The about and extensions sections store additional vendor-specific event information not captured by the other sections. The extensions section is a free-form set of key-value pairs.

Each UDM event stores values from one original raw log event. Depending on the type of event, certain attributes are required while others are optional. The required versus optional attributes are determined by the metadata.event_type value. The Google Security Operations reads metadata.event_type and performs field validation specific to that event type after the logs are received.

If no data is stored in a section of the UDM record, for example the extensions section, then that section does not appear in the UDM record.

The metadata fields

This section describes fields required in a UDM event.

The event_timestamp field

UDM events must include data for the metadata.event_timestamp which is the GMT timestamp when the event occurred. The value must be encoded using one of the following standards: RFC 3339 or Proto3 timestamp.

The following examples illustrate how to specify the timestamp using RFC 3339 format, yyyy-mm-ddThh:mm:ss+hh:mm (year, month, day, hour, minute, second, and the offset from UTC time). The offset from UTC is minus 8 hours, indicating PST.

metadata {
  "event_timestamp": "2019-09-10T20:32:31-08:00"
}

metadata {
  event_timestamp: "2021-02-23T04:00:00.000Z"
}

You can also specify the value using the epoch format.

metadata {
event_timestamp: {
  "seconds": 1588180305
 }
}

The event_type field

The most important field in the UDM event is metadata.event_type. This value identifies the type of action performed and is independent of vendor, product, or platform. Examples values include PROCESS_OPEN, FILE_CREATION, USER_CREATION, and NETWORK_DNS. For the complete list, see the UDM field list document.

The metadata.event_type value determines which additional required and optional fields must be included in the UDM record. For information about which fields to include for each event type, see UDM usage guide.

The principal, target, src, intermediary, observer, and about attributes

The principal, target, src, intermediary, and observer attributes describe assets that are involved in the event. Each store information about objects involved in the activity, as recorded by the original raw log. This could be the device or user that performed the activity, the device or user that is the target of the activity. It might also describe a security device that observed the activity, such as an email proxy or network router.

The most commonly used attributes are:

  • principal: Object that performed the activity.
  • src: Object that initiates the activity, if different than the principal.
  • target: Object that is acted upon.

Every event type requires that at least one of these fields contains data.

The auxiliary fields are:

  • intermediary: Any object that acted as an intermediary in the event. This could include objects such as proxy servers and mail servers.
  • observer: Any object that does not directly interact with the traffic in question. This might be a vulnerability scanner or a packet sniffer device.
  • about: Any other objects that played a role in the event and is optional.

The principal attributes

Represents the acting entity or the device that originated the activity. The principal must include at least one machine detail (hostname, MAC address, IP address, product-specific identifiers like a CrowdStrike machine GUID) or user detail (for example, user name), and optionally include process details. It must not include any of the following fields: email, files, registry keys or values.

If the event takes place on a single machine, that machine is described in the principal attribute only. The machine does not need to be described in the target or src attributes.

The following JSON snippet illustrates how the principal attribute might be populated.

"principal": {
  "hostname": "jane_win10",
  "asset_id" : "Sophos.AV:C070123456-ABCDE",
    "ip" : "10.10.2.10",
    "port" : 60671,
    "user": {  "userid" : "john.smith" }
}

This attribute describes everything known about the device and user that was the principal actor in the event. This example includes the device's IP address, port number, and hostname. It also includes a vendor-specific asset identifier, from Sophos, which is a unique identifier generated by the third-party security product.

The target attributes

Represents a target device being referenced by the event, or an object on the target device. For example, in a firewall connection from device A to device B, device A is captured as the principal and device B is captured as the target.

For a process injection by process C into target process D, process C is the principal and process D is the target.

principal versus
target

Figure: Principal versus target

The following example illustrates how the target field could be populated.

target {
   ip: "192.0.2.31"
   port: 80
}

If more information is available in the original raw log, such as hostname, additional IP addresses, MAC addresses, and proprietary asset identifiers, it should also be included in the target and principal fields.

Both principal and target can represent actors on the same machine. For example, process A (principal) running on machine X could act on process B (target) also on machine X.

The src attribute

Represents a source object being acted upon by the participant along with the device or process context for the source object (the machine where the source object resides). For example, if user U copies file A on machine X to file B on machine Y, both file A and machine X would be specified in the src portion of the UDM event.

The intermediary attribute

Represents details about one or more intermediate devices processing activity described in the event. This could include device details about devices such as proxy servers and SMTP relay servers.

The observer attribute

Represents an observer device which is not a direct intermediary, but which observes and reports on the event in question. This could include a packet sniffer or network-based vulnerability scanner.

The about attribute

This store details about an object referenced by the event which is not described in the principal, src, target, intermediary or observer fields. For example, it could capture the following:

  • Email file attachments.
  • Domains, URLs, or IP addressed embedded within an email body.
  • DLLs that are loaded during a PROCESS_LAUNCH event.

The security_result attribute

This section contains information about security risks and threats that are found by a security system and the actions taken to mitigate those risks and threats.

Here are types of information that would be stored in the security_result attribute:

  • An email security proxy detected a phishing attempt (security_result.category = MAIL_PHISHING) and blocked (security_result.action = BLOCK) the email.
  • An email security proxy firewall detected two infected attachments (security_result.category = SOFTWARE_MALICIOUS) and quarantined and disinfected (security_result.action = QUARANTINE or security_result.action = ALLOW_WITH_MODIFICATION) these attachments and then forwarded the disinfected email.
  • An SSO system allows a login (security_result.category = AUTH_VIOLATION) which was blocked (security_result.action = BLOCK).
  • A malware sandbox detected spyware (security_result.category = SOFTWARE_MALICIOUS) in a file attachment five minutes after the file was delivered (security_result.action = ALLOW) to the user in their inbox.

The network attribute

Network attributes store data about network-related events and details about protocols within sub-messages. This includes activity, such as emails sent and received, and HTTP requests.

The extensions attribute

Fields under this attribute store additional metadata about the event captured in the original raw log. It can contain information about vulnerabilities or additional authentication-related information.

Structure of a UDM Entity

A UDM entity record stores information about any entity within an organization. If the metadata.entity_type is USER, the record stores information about the user under the entity.user attribute. If the metadata.entity_type is ASSET, the record stores information about an asset, such as workstation, laptop, phone, and virtual machine.

Entity data
model

Figure: Event data model

The metadata fields

This section contains fields required in a UDM Entity, such as:

  • collection_timestamp: the date & time the record was collected.
  • entity_type: the type of entity, such as asset, user, and resource.

The entity attribute

The fields under the entity attribute store information about the specific entity, such as hostname and IP address if it is an asset, or windows_sid and email address if it is a user. Notice that the field name is 'entity', but the field type is a Noun. A Noun is a commonly used data structure that stores information in both entities and events.

  • If the metadata.entity_type is USER, then data is stored under the entity.user attribute.
  • If the metadata.entity_type is ASSET, then data is stored under the entity.asset attribute.

The relation attribute

Fields under the relation attribute store information about other entities that the primary entity is related to. For example, if the primary entity is a User and the user has been issued a laptop. The laptop is a related entity. Information about the laptop is stored as an 'entity' record with a metadata.entity_type = ASSET. Information about the user is stored as an 'entity' record with the metadata.entity_type = USER.

The user entity record also captures the relationship between the user and the laptop, using fields under the "relation" attribute. The relation.relationship field stores the relationship that the user has to the laptop, specifically that the user owns the laptop. The relation.entity_type field stores the value ASSET, because the laptop is a device.

Fields under the relations.entity attribute store information about the laptop, such as the hostname and MAC address. Notice again that the field name is 'entity' and the field type is a Noun. A Noun is a commonly used data structure. Fields under the relation.entity attribute store information about the laptop.

The relation.direction field stores the directionality of the relationship between user and the laptop, specifically whether the relationship is bidirectional versus unidirectional.