Skip to content

Latest commit

 

History

History
238 lines (190 loc) · 5.72 KB

README.md

File metadata and controls

238 lines (190 loc) · 5.72 KB
Raw

Stolen Credentials Pipeline

A first attempt at a pipeline for processing stolen credentials for DIVD CSIRT

Overview

The pipeline has the followign flow:

flowchart TB
    in[Stolen or\nleaked\ncredentials]
    normalize([Normalize notebook])
    ingest([Ingest notebook])
    leak_db[(Leak\nspecific\nDatabase)]
    nml_db[(NML Database)]
    nml_import[(NML importer)]
    nml_hashes[NML hashes]
    sum[(Summarize\nnotebook)]
    apex[Apex lists]
    scatsec((Scattered Secrets))
    divd_db[(DIVD leak \ndatabase)]
    query_nb([Query notebook])
    query_svc((Query service))
    result[Result set]
    register([DIVD register\nnotebook])
    certs[Certs\ncsirts\nsecurity teams]
    nml_exp([NML exporter])
    nml[NML]

    nml_db --> ingest
    nml_hashes --> nml_import
    nml_import <--> nml_db
    scatsec --> ingest
    in --> normalize 
    normalize ---> ingest 
    ingest <--> leak_db 
    divd_db ---> ingest 
    register --> divd_db
    leak_db --> register 
    leak_db --> sum
    sum --> apex
    leak_db --> query_nb ---> result
    leak_db --> query_svc ---> result
    apex --> certs
    result ---> certs
    leak_db --> nml_exp
    nml_exp ---> nml
Loading

Stolen or leaked credentials

This is a set of stolen and/or leaked credentials, obtained by DIVD in line with the credentials policy.

Normalize notebook

The normalize notebook is a notebook that attempts to normalize the data to one or more tsv files with the following fields:

  • username
  • passwd
  • url (Url that the credentials belong to)
  • ts_found (Timestamp when the data was "found")
  • ts_leaked (Timestamp when the data was stolen/leaked)
  • has_name (0/1 if the record has a name)
  • has_dob (0/1 if the record has a date of birth)
  • has_addr (0/1 if the record has a address)
  • has_phone (0/1 if the record has a )
  • has_cc (0/1 if the record has creditcard data)
  • has bankacc (0/1 if the record has a bank account)
  • has_ssn (0/1 if the record has a ssn)
  • has ip (0/1 if the record has an ip address)
  • extra_data (json object with extra data)

There is no single Jupyter notebook that works for all formats, but the generic normalize notebook in this repo contains helper functions that are usefull.

Ingest notebook

The ingest notebook takes care of the following tasks:

  • normalizing the username to lowercase utf8 NFC normalisation
  • masking the password
  • calculating a nml_hash
  • determine the apex of username
  • determine the apex of url
  • Ingesting the data into a leak specific database and storing results
  • checking the nml_hash agaist the divd database of earlier breaches
  • checking the nml_hash against the nml_database (future)
  • checking the nml_hash against the Scattered Secrets service (future)

Leak Specific Database looks as follows

erDiagram
    ENTITY {
        str username "indexed"
        str masked_passwd
        str nml_hash "indexed"
        str url
        str email_apex "indexed"
        str url_apex "indexed"
        date ts_found
        date ts_leaked
        bool has_name
        bool has_dob
        bool has_addr
        bool has_phone
        bool has_cc
        bool has_bankacc
        bool has_ssn
        bool has_ip 
        str extra_data
    }

    DIVD {
        str nml_hash PK "indexed"
        str case_id
        str sub_id
        str description
        date date
    }

    NML {
        str nml_hash PK "indexed"
        date first
        date last
    }

    SCATTERED {
        str nml_hash PK "indexed"
        str extra_data
    }

    ENTITY ||--0+ DIVD : "on nml_hash"
    ENTITY ||--0+ NML : "on nml_hash"
    ENTITY ||--0+ SCATTERED : "on nml_hash"
Loading

All indexes are created collate nocase this should be used in queries as well to get the best speed results.

DIVD register notebook

The DIVD register notebook takes the data from the leak specific database and registers it in the DIVD leak database of leaks. WHich is stored in a (TODO) central location.

This database only contains records for which a nml_hash could be calculated

DIVD leak database

The DIVD leak database looks as follows:

erDiagram
    LEAK {
        str case_id PK 
        str sub_id PK "Optional sub identifier"
        str description
        date date
    }

    DIVD {
        str nml_hash PK "indexed"
        str case FK
        str sub_id FK
    }

    LEAK 1+--0+ DIVD : "on case and sub_id"
Loading

Summarize notebook

This notebook turns a leak_db into two csv files.

email_apex_summary.csv

Contains:

  • email_apex (apex of the username (if it is an email)
  • count_of_unique_username_password entries (nml_hashes actually)

url_apex_summary.csv

Contains:

  • url_apex (apex of the url that the credentials belong to)
  • cont_of_unique_usernames

Query service (future)

TBD

Query notebook

Takes a list of apexes and exports two cvs per apex

<apex>-email-hits.csv

Records where email_apex matches <apex>. It has the following records:

  • username
  • masked_passwd
  • url
  • ts_found
  • ts_leaked
  • has_name
  • has_dob
  • has_addr
  • has_phone
  • has_cc
  • has bankacc
  • has_ssn
  • has ip
  • extra_data
  • divd_previous (list of earlier DIVD breaches)
  • in_nml (0/1 if data is in nml)
  • nml_first (first date seen in nml by us)
  • nml_last (last seen in nml by us)

<apex>-url-hits.csv

Records where url_apex matches <apex>. Same fields as above.

NML downloader

The No More Leaks download notebook downloads the data from no more leaks

NML Importer

The No More Leaks import cookbook imports the data into the shared no more leaks database

NML database

This database looks like this

erDiagram
    HASHES {
        text nml_hash PK "indexed unique" 
        date first
        date lsst
    }

    FILES {
        text name
        text md5
    }
Loading