Skip to content

IDMEFv2 : Alert Class Comments

Gilles Lehmann edited this page Jan 17, 2023 · 17 revisions

The Alert class contains high level information about the event that triggered the alert.

                                                  +---------------------------+  
                                                  |           Alert           |  
                                                  +---------------------------+  
                                                  | STRING      Version       |  
                                                  | UUID        ID            |  
                                                  | STRING      Entity        |  
                                                  | ENUM[]      Category      |  
                                                  | ENUM        Cause         |  
                                                  | STRING      Description   |  
                                                  | ENUM        Status        |  
                                                  | ENUM        Severity      |  
                                                  | FLOAT       Confidence    |  
                                                  | STRING      Note          |  
                                                  | TIMESTAMP   CreateTime    |  
                                                  | TIMESTAMP   StartTime     |  
                                                  | TIMESTAMP   CeaseTime     |  
                                                  | TIMESTAMP   DeleteTime    |  
                                                  | STRING[]    AltNames      |  
                                                  | STRING[]    AltCategory   |  
                                                  | URI[]       Ref           |  
                                                  | UUID[]      CorrelID      |  
                                                  | CONDITION[] AggrCondition |  
                                                  | UUID[]      PredID        |  
                                                  | UUID[]      RelID         |  
                                                  +---------------------------+  

Version

The IDMEF version, here 2.0

TBM : During the definition/standardisation process we might use a version number including the version of the draft so 2.00 for IDMEFv2 Draft V00.

ID

Mandatory. Unique identifier for the alert.

Comments : The alert has to be unique in the organisation but possibly also out of the organisation if the alert is sent to some centralised security central for deeper analysis, forensic or storage. This explains the use of UUIDs.

Entity

Optional. Tenant ID to support multi-tenancy (e.g. decentralized infrastructure, local agency, subsidiary company, etc.). Should be used when there are multiple sites/locations or multiple tenants (e.g. by Managed Security Services Providers).

Comments : The list of possible entities has to be defined by the organisation so it's a closed list. It can be a list of entities/sites/subsidiaries in the same organisation or a list of client for a service provider.

Category

Optional. The incident's category & subcategory as listed in [[ENISA-RIST] using the format "category.subcategory" (e.g. "Attempt.Exploit").

Comments : Category is a multi-purpose attributes. It helps automate computation of the alert (correlation, aggregation, etc.) and it quickly gives an information to the operator about what kind of incident is going on, thus helping him to decide if an action or a further analysis is needed. More details about the category attribute [here](https://github.com/IDMEFv2/IDMEFv2-Specification/wiki/IDMEFv2-:-Format-comments#category).

TBM : The category list is "based" on ENISA-RIST but not "copied" as "Natural Hazards" have been added for example.

Cause

Optional. Event cause, if known at the time of detection.If unknown, this key SHOULD NOT be defined by the analyser and may be filled later on by a manager or a human operator.

Comments : Cause can possibly be added/modified by another analyser or later by the operator or the analyst. If this information is known it's important for the operator as well as for the analyst. In general it's important to know if the organisation is under a voluntary attack or if it's "only" a malfunction or an error. Although, human errors can sometime lead to more damages than malicious attacks.

TBM : The cause description should be modified : "Event cause. The cause can be modified by any analyser on the way of the alert and later by the operator and/or the analyst if new investigation reveals and confirms a different cause of the event."

Description

Optional. Short free text human-readable description.

Comments : The description can add detail to the alert classification for easiest/faster comprehension by the operator.

Example :

  • Cryptoware WannaCry blocked on pegasus server
  • Unknown person entering through east doorway

Severity

Optional. Severity of the alert.

Enum = Unknown, Info, Low, Medium, High

Comments :

Closed list indicating the severity of the potential incident. Ideally the severity of the alert might depend on multiple criteria : classification, target, source, cause, etc. The severity of alerts should also be defined in the organisation security policy. Indeed, two similar alerts may have different severity depending on many factors.

Examples :

  • Multiple failed authentication detected by a SIEM may not have the same severity if the target is a "public" file server or a very sensitive financial server.
  • A server scan attempt detected by an IDS is problematic in the local network but "normal" on a DMZ
  • For the same situation, an MSSP may have different "Guaranteed fault repair time" depending on the clients contracts
  • High temperature in a server detected by a network monitoring console is more severe is it happens in the middle of summer during a heatwave
  • An unknown person entering a building detected by a CCTV may be more problematic if the building is the cafeteria or the ammunition depot

Notes : Severity could be a number, between 0 an 1 for example. But it is difficult to define a precise way of calculating the severity, so severity is very often an approximation. A number is also less intuitive for the operator who sometime needs to filter the alerts depending on their severity. The severity values are often associated with colours on the monitoring console : Uknonwn (Gray), Info (Blue), Low (Yellow), Medium (Orange), High (Red)

See more comments about severity here.

Confidence

Optional. A floating-point value between 0 and 1 indicating the analyzer's confidence in its own reliability of this particular detection, where 0 means that the detection is surely incorrect while 1 means there is no doubt about the detection made.

Comments : The confidence of detection of the "same" event might depends on the detection method and the detection data.

Examples :

  • A log manager detects with no doubt an "authentication failed" string using a regex in a log file, the confidence will be 1
  • A log manager detects abnormal behaviour using machine learning algorithm on logs files, the confidence might be of 0,75
  • A camera "recognize" an employee using face detection on a very good picture of the employee taken with full luminosity in the middle of the day, the confidence might be of 0,95
  • The same employee is entering the building the day after, looking down at his smart-phone (or wearing a covid mask) while passing in front of the camera and it's very late in the day with bad luminosity. The camera algorithm still recognise the employee but with a lower confidence.
  • A correlator identifies a series of events that might correspond to a known procedure for attacking a SAP server but this hypothesis needs to be confirmed manually by the operator. The confidence will be average until the operator has verified the hypothesis.

TBM :

  • Should confidence be an enum (like Severity) : unknown, low, medium, high ?
  • Maybe it's easier to come to a number with confidence as it's correlated to "algorithm" efficacy ?
  • Otherwise add "If key is not presented, analyser does not know (or has no capability to estimate the confidence)." ?

Note

Optional. Free text human-readable additional note, possibly a longer description of the incident if is not already obvious.

Comments : The Note attribute can be used to store any additional information. It can be additional information about the event and/or about the incident resolution, although the incident resolution information should in principle be stored elsewhere (with a link with the external tool in AltNames)

CreateTime

Mandatory. Timestamp indicating when the message was created. May point out delay between detection and processing of the events.

TBM : Is this passage necessary ? "May point out delay between detection and processing of the events."

StartTime

Optional. Timestamp indicating the deduced start of the event. In case the event is not part of a series, this attribute MAY instead be set to the timestamp initially present in the event (if any).

Comments: StartTime can be in the future.

CeaseTime

Optional. Timestamp indicating the deduced end of the event.

Comments: CeaseTime can be in the future.

TBC: No CeaseTime means we know the event is still going on OR we don't know if it's finish

DeleteTime

Optional. Timestamp indicating when the message must be deleted. This attribute MUST be specified if the message has to be deleted after this date, e.g. for technical, organizational or ethical reasons.

Use cases :

  • Fake events generated to test a correlation rules
  • Confidential events that should not be archived more than necessary

TBM | TBR :

  • Add "legal" reasons.
  • Interesting concept but probably rare. Maybe we shouldn't keep it ?

AltNames

Optional. Alternative identifiers; strings which help pair the event to internal systems' information (for example ticket IDs inside a request tracking systems).

AltCategory

Optional. Alternate categories from a reference other than [[ENISA-RIST] (e.g. MISP, MITRE ATT@CK or another proprietary/internal reference).

Ref

Optional. References to sources of information related to the alert and/or vulnerability, and specific to this alert. This MAY be a URL to additional info, or a URN in a registered or unregistered ad-hoc namespace bearing reasonable information value and uniqueness, such as "urn:cve:CVE-2013-2266".

CorrelID

Optional. Identifiers for the messages which were used as information sources to create this message, in case the message has been created based on correlation/analysis/deduction from other messages.

AggrCondition

Optional. A list of IDMEF fields used to aggregate events. The values for these fields will be the same in all aggregated events. This attribute should mostly be set by intermediary nodes, which detect duplicates, or aggregate events, spanning multiple detection windows, into a longer one.The "StartTime" and "CeaseTime" attributes are used in conjunction with this attribute to describe the aggregation window.

Comments: The use of a "condition" instead of a plain list is a try to avoid possible huge lists of hundreds or thousands IDs.

TBD:

  • We need EXAMPLES
  • This attribute is not intuitive and maybe too complicated in certain cases, should we also keep an AggrId

PredID

Optional. A list containing the identifiers of previous messages which are obsoleted by this message. The obsoleted alerts SHOULD NOT be used anymore. This field can be used to "update" an alert.

RelID

Optional. A list containing the identifiers of other messages related to this message.

Comments:

  • We distinguish events that are the result of a correlation and events that are related
  • TBQ : What does it means exactly to be related (without being correlated ?)
  • TBR : Not sure we should keep this one ...

Clone this wiki locally