LimaCharlie Query Language
  • 15 Oct 2024
  • 4 Minutes to read
  • Contributors
  • Dark
    Light

LimaCharlie Query Language

  • Dark
    Light

Article summary

Beta Feature

LCQL is currently in Beta, and features may change in the future.

LimaCharlie Query Language (LCQL) provides a flexible, intuitive and interactive way to explore your data in LimaCharlie. Telemetry ingested via EDR sensors or adapters are searchable via LCQL, and can be searched en masse. Sample use cases for LCQL include:

  • Analyze your entire, multi-platform fleet for network connections of interest.

  • Search across all Windows Event Logs for unique user activity.

  • Look at all Linux systems for specific package installation events.

  • Analyze all volume mounts and unmounts on macOS devices

  • And many more!!!

Querying the past 30 days within LimaCharlie is free.

Queries can also be saved by clicking the Save button. To save, view, or edit saved queries, users will need to be granted appropriate permissions:

  • query.del - Delete saved queries

  • query.get - Get saved queries

  • query.get.mtd - Get saved queries metadata

  • query.set - Set saved queries

  • query.set.mtd- Set saved queries metadata

The steps below walk you through creating your own LCQL queries. If you're looking for samples or LCQL inspiration, check out our LCQL Examples page.

Building LimaCharlie Queries

LCQL queries contain 4 components with a 5th optional one, each component is separated by a pipe (|):

  1. Timeframe: the time range the query applies to. This can be either a single offset in the past like -1h or -30m. Or it can be a date time range like 2022-01-22 10:00:00 to 2022-01-25 14:00:00.

  2. Sensors: the set of sensors to query. This can be either * for all sensors, a list of space separated SIDs like 111-... 222-... 333-..., or it can be a sensor selector like plat == windows.

  3. Events: the list of events to include in the query, space separated like NEW_PROCESS DNS_REQUEST, or a * to go over all event types.

  4. Filter: the actual query filter. The filters are a series of statements combined with " and " and " or " that can be associated with parenthesis (()). String literals, when used, can be double-quoted to be case insensitive or single-quoted to be case sensitive. Selectors behave like D&R rules, for example: event/FILE_PATH.

The currently supported filter operators are listed below:

operator

example

is or ==

event/FILE_PATH is "c:\windows\calc.exe"

is not or !=

event/FILE_IS_SIGNED != 0

contains

event/FILE_PATH contains 'evil'

not contains

event/FILE_PATH not contains 'system32'

matches

event/FILE_PATH matches ".*system[0-9a-z].*"

not matches

starts with

event/FILE_PATH starts with "c:\windows"

not starts with

ends with

event/FILE_PATH ends with '.eXe'

not ends with

cidr

event/NETWORK_CONNECTIONS/IP_ADDRESS cidr "10.1.0.0/16"

is lower than

event/NETWORK_CONNECTIONS/PORT is lower than 1024

is greater than

is platform

is platform "windows"

is not platform

is not platform "linux"

is tagged

is tagged "vip"

is not tagged

is public address

event/NETWORK_CONNECTIONS/IP_ADDRESS is public address

is private address

scope

event/NETWORK_CONNECTIONS scope (event/IP_ADDRESS is public address and event/PORT is 443)

with child / with descendant / with events

event/FILE_PATH contains "evil" with child (event/COMMAND_LINE contains "powershell")

  1. Projection (optional): a list of fields you would like to extract from the results with a possible alias, like: event/FILE_PATH as path event/USER_NAME AS user_name event/COMMAND_LINE. The Projection can also support a grouping functionality by adding GROUP BY(field1 field2 ...) at the end of the projection statement.

When grouping, all fields being projected must either be in the GROUP BY statement, or have an aggregator modifier. An aggregator modifer is, for example, COUNT( host ) or COUNT_UNIQUE( host ) instead of just host.

A full example with grouping is:

-1h | * | DNS_REQUEST | event/DOMAIN_NAME contains "apple" | event/DOMAIN_NAME as dns COUNT_UNIQUE(routing/hostname) as hostcount GROUP BY(dns host)

which would give you the number of hosts having resolved a domain containing apple, grouped by domain.

All of this can result in a query like:

-30m | plat == windows | NEW_PROCESS | event/COMMAND_LINE contains "powershell" and event/FILE_PATH not contains "powershell" | event/COMMAND_LINE as cli event/FILE_PATH as path routing/hostname as host

OR

-30m | plat == windows | * | event/COMMAND_LINE contains "powershell" and event/FILE_PATH not contains "powershell"

Projection Syntax

Note: There is no space between BY and the ( opening of the parentheses in a projection.

Example: GROUP BY(dns host) or COUNT_UNIQUE(routing/hostname)

Using the CLI

The command line interface found in the Python CLI/SDK can be invoked like limacharlie query once installed (pip install limacharlie).

Context

To streamline day to day usage, the first 3 components of the query are set seperatly and remain between queries.
These 3 component can be set through the following commands:

  1. set_time to set the timeframe of the query, like set_time -3h based on the ParseDuration() strings.

  2. set_sensors to set the sensors who's data is queried, like set_sensors plat == windows, based on the sensor selector grammar.

  3. set_events to set the events that should be queried, space separated like NEW_PROCESS DNS_REQUEST. This command supports tab completion.

Once set, you can specify the last component(s): the Filter, and the Projection.

Several other commands are avaible to make your job easier:

  • set_limit_event to set a maximum number of events to scan during the query.

  • set_output to mirror the queries and their results to a file.

  • set_format to display results either in json or table.

  • stats to display the total costs incurred from the queries during this session.

Querying

Paged Mode

The main method of running a query as described above (in paged mode) is to use the q (for "query") command.

Paged mode means that an initial subset of the results will be returned (usually in the 1000s of elements) and if you want to fetch more of the results, you can use the n (for "next") command to fetch the next page.

Some queries cannot be done in paged mode, like queries that do aggregation or queries that use a stateful filter (like with child). In those cases, all results over the entire timeline are computed.

For example:
q event/DOMAIN_NAME contains 'google' | event/DOMAIN_NAME as domain COUNT_UNIQUE(routing/sid) as count GROUP BY(domain)

This command supports tab completion for elements of the query, like event/DO + "tab" will suggest event/DOMAIN_NAME or other relevant elements that exist as part of the schema.

Non Paged Mode

You can also force a full query over all the data (no paging) by using the "query all" (qa) command like:

qa event/DOMAIN_NAME contains 'google' | event/DOMAIN_NAME as domain COUNT_UNIQUE(routing/sid) as count GROUP BY(domain)

Dry Run

To simulate running a query, use the dryrun command. This will query the LimaCharlie API and return to you an aproximate worst case cost for the query (assuming you fetch all pages over its entire time range).

For example:
dryrun event/COMMAND_LINE contains "powershell" and event/FILE_PATH not contains "powershell"


Was this article helpful?


What's Next