Jump to: navigation, search

Session Recording

Name: Session Recording
Ticket: #TODO
Author: Nikolai_Kondrashov
Incomplete.png Pending review
Last updated: 2016-11-25 by Nikolai Kondrashov

Overview

Rationale

Many companies in highly regulated industries like financial or health care have to balance costs of maintaining their IT infrastructure and implementing security practices required by regulations like SOX, PCI or HIPAA. To reduce the cost of ownership they frequently subcontract, or offshore the IT services to other companies and vendors.

To achieve the level of accountability imposed on them by different regulations and to meet audit requirements they need to implement constant monitoring of the activities conducted by the contracted vendors. To do such monitoring they collect logs and record all the activity initiated by the privileged users on the systems they have regardless of the operating system that is used. In other words they store all the sessions for all the privileged users on the mission critical systems.

What is it about?

The solution is aimed at answering the following questions:

  • Who logged into the system?
  • What did the user do while he/she was logged into that system?
  • How a malicious activity by a contractor or employee can be proven?

The tools and capabilities provided by solution should allow auditor or security officer to detect and prove that malicious activity has happened in the past, or detect one happening in real time.

User Stories

Auditor/Security Officer

As an auditor or security officer who is responsible for maintaining security and preventing attacks, or proving the fact that a malicious activity has been conducted by an employee or a contractor:

  • I would like every protected client system to record information about who have accessed that system, so in case of a security incident it is easier to identify the culprit. This information should be emitted by the system in such a way that the recorded privileged user can't prevent its flow without leaving tracks he/she was trying to do so. Supported by Phase 0.
  • I would like the information to be consolidated on a central server where it can be viewed and correlated with the authentication activity collected from the central authentication servers, so that I can identify and verify when sessions started and ended and whether they were preceded by authentication failures. Supported by Phase 1.
  • I would like every command and operation that the user has ran while he/she was logged into the system to be recorded and delivered to the central server so that it can be stored safely, viewed and analyzed, and correlated with the activity of other users. Supported by Phase 1.
  • I would like the activity information to be recorded and later presented in the most intuitive way, to be able to understand what the user was doing while logged into the system. The best format for such information, as recognized by the IT professionals, is the full screen capture of the input and output of the user's terminal session. Supported by Phase 0.
  • I would like to be able to better rationalize the impact of the malicious activity by correlating the session recording with the underlying activity (audit trail) recorded by the system. Supported by Phase 1.
  • I would like to be able to view the recorded sessions in a central server via a web interface, since I prefer not to be bound to a specific platform and install a thick client. Supported by Phase 1.
  • In some cases I would like to be able to play session back from a command line using a simple command line tool, so I can test the setup, or act in limited environments, such as system recovery modes. Supported by Phase 0.
  • I would like to be able to watch the playback of the session, start, stop, rewind, scroll forward or back, play faster or slower. I would like to be able to pause the playback select the command that the user entered and see the recorded trail of the activities happening on the system (file system operations, socket requests, started and stopped process) that correspond to the selected command. This would allow me to understand the sequence of events within the user session quicker and easier. Supported by Phase 2.
  • I would like to be able to keep the recorded information for playback for a period of time defined by my policy and capacity preferences. After that period the solution should overwrite, purge or archive the collected information depending on the configuration. Supported by Phase 2.
  • I would like to be able to easily navigate to the recorded session so that I can select the right session to play back. I should be able to search recorded sessions and related audit data by user, user group, system and group of systems as well as narrow search to a specific service like SSH or limit the time period I am looking for the session to start (or finish). Supported by Phase 2. For example I should be able to say:
    • I want to find all the sessions by John Doe in my environment during this weekend
    • I want to find all administrative sessions that were recorded last week against my file servers in my New Jersey datacenter.

System Administrator

As a system administrator responsible for integrating the session recording solution into existing or new infrastructure:

  • I would like to be able to control session recording locally on specific hosts, for specific users, without requiring SSSD or FreeIPA involvement or use at all, so that I can test it out, support simple installations, or integrate with other identity management solutions. Supported by Phase 0.
  • I would like to be able to control session recording locally on specific hosts, for specific users, through already deployed SSSD, without involving the central identity management solution (FreeIPA or other) which controls it, so that I can test it out, or use it for just a few systems with lower management cost. Supported by Phase 0.
  • I would like to be able to control session recording centrally via FreeIPA for many hosts, host groups, users or groups, so that I can save on maintenance costs by having all configuration done through a single interface, in one place. Supported by Phase 1.
  • I would like the delivery mechanism that is used to send information from the client system to the central server to be scalable, reliable and support high-availability and load-balancing configurations. Supported by Phase 0.
  • I would like the information that is sent from the client system, to be sent via protected and authenticated channel so that it can't be tampered with or disclosed in transit. Supported by Phase 0.

Design

Architecture

We record what passes between the user terminal and the shell, what files the user accesses and what commands he/she executes, and send all that to a central storage, where it can be searched, correlated and from where it can be played back. Who, where and how is recorded can be controlled either centrally or locally.

Client-side components

  • Tlog-rec – the recording process which is invoked in place of the user's shell when he/she logs into the system. The recording process creates a pseudo-terminal, starts the actual user shell under it, and records everything that passes between the pseudo-terminal (the user shell) and the actual user terminal. The user should not be able to stop or prevent the recording without leaving evidence of trying to do so.
  • SSSD – the component which is responsible for telling the system which shell should be started for the user, based on the POSIX information for the user account. SSSD should inform the system that the recording process is the user shell, and pass the real shell to the recording process so that it can be started within the recorded session.
  • Auditd – general system auditing subsystem which collects all the activity related to user session in the form of the audit entries
  • Aushape – an audit log converter running under Auditd to convert audit events to JSON on the fly, passing them to the logging server.
  • Logging server - Rsyslog, Fluentd, or Logstash – a collection agent, which streams the audit and session recording data from the system to the central server.

Server-side components

  • FreeIPA - the central integrated security management solution, which stores and provides information of who, where, and how to record.
  • ElasticSearch - the data storage where the session recording and audit data can be placed and correlated.
  • Tlog-play - terminal based playback tool which can be used from the command line to recreate the session.
  • WebUI (TBD) - playback terminal with the audit trail correlation – most likely a custom web UI control.


Control and Data Flow

           Servers            Network               Clients
     _____________________               _______________________________
    |  _________________  |             |  ___________________          |
    | |                 | |             | |                   |         |
    | |     FreeIPA     |====(control)===>|       SSSD        |         |
    | |_________________| |             | |___________________|         |
    |         /\          |             |     ||          /\            |
    |         ||          |             |     ||          ||            |
    |      (control)      |             |  (control)   (control)        |
    |   ......||.......   |             |     ||  ........||.........   |
    |  : Administrator :  |             |     || :   Administrator   :  |
    |   '''''''''''''''   |             |     ||  '''||'''''''''||'''   |
    |                     |             |     ||  (control)  (control)  |
    |   ...............   |             |     ||     ||         ||      |
    |  :    Auditor    :  |             |  ___\/_____\/__   ____\/____  |
    |   ''/\'''''''/\''   |             | |              | |          | |
    |     ||       ||     |             | |   Tlog-rec   | |  Auditd  | |
    |   (data)   (data)   |             | |______________| |__________| |
    |  ___||__   __||___  |             |        ||             ||      |
    | |       | |       | |             |        ||          ___\/____  |
    | | Tlog- | | WebUI | |             |        ||         |         | |
    | | play  | | (TBD) | |             |        ||         | Aushape | |
    | |_______| |_______| |             |        ||         |_________| |
    |     /\       /\     |             |        ||             ||      |
    |     ||       ||     |             |      (data)         (data)    |
    |   (data)   (data)   |             |        ||             ||      |
    |  ___||_______||___  |             |     ___\/_____________\/___   |
    | |                 | |             |    |                       |  |
    | |                 | |             |    |        Rsyslog        |  |
    | |                 | |             |    | - - - - - - - - - - - |  |
    | |  Elasticsearch  |<====(data)=========|        Fluentd        |  |
    | |                 | |             |    | - - - - - - - - - - - |  |
    | |                 | |             |    |        Logstash       |  |
    | |_________________| |             |    |_______________________|  |
    |_____________________|             |_______________________________|

The control flow differs by case.

Standalone local control

The recorded user's POSIX entry is specified as tlog-rec, and the actual user's shell is set for all users in the global configuration of tlog-rec. Alternatively, a special per-shell symlink to tlog-rec is set as the shell for the user (e.g. tlog-rec-shell-bash, or tlog-rec-shell-zsh).

After a user is authenticated, the login program (login/telnetd/sshd/etc.) starts tlog-rec, as it is specified as the shell for the user. Tlog-rec checks if it was invoked under the special name. If it was, it starts the shell extracted from the name. If it wasn't, it starts the shell specified in the global configuration.

Local control via SSSD

SSSD is configured to enable session recording for all, or specific users and/or groups. This is done by adding a session_recording section to sssd.conf, containing parameters specifying the above.

After a user is authenticated, the program that logs user in (login/telnetd/sshd/etc.) queries which shell to start for the user, and asks the system to set up the session. If SSSD is configured to record the specific user, it answers that shell should be tlog-rec and, as part of session setup, adds a variable to the user environment telling tlog-rec which actual shell it needs to start. E.g. it sets TLOG_REC_SHELL=/bin/bash.

The login program starts tlog-rec. Tlog-rec reads the environment variable added by SSSD and starts the actual user shell.

Central control via SSSD and FreeIPA

FreeIPA is instructed to add directory entries specifying which hosts and users should have recording enabled and with which tlog-rec settings (whether to record input, output, or both, etc.). The schema should be similar to the SELinux rule schema, but specific design is TBD.

After a user is authenticated, the login program (login/telnetd/sshd/etc.) queries which shell to start for the user, and asks the system to set up the session. SSSD queries FreeIPA directory to check if this user on this machine should be recorded. If it should, then SSSD answers that shell should be tlog-rec. As part of the following session setup, SSSD retrieves tlog-rec settings applicable for this user and machine from FreeIPA directory, and adds an environment variable containing those settings, along with a variable specifying the actual user's shell.

E.g. it sets TLOG_REC_CONF_TEXT='{"log": {"input": false, "output": true}}' and TLOG_REC_SHELL=/bin/bash. This example specifies that input recording should be off (e.g. to avoid recording of passwords), output recording should be on, and the actual user's shell should be /bin/bash. All of that would override the global tlog-rec configuration.

The login program starts tlog-rec. Tlog-rec reads the environment variables added by SSSD, adjusts its settings and starts the actual user shell.

Further control and data flow

Other tlog-rec configuration, not mentioned above, is specified in tlog-rec's global configuration file /etc/tlog/tlog-rec.conf.

After tlog-rec starts the actual user shell, it inserts itself between it and the user terminal and records everything passing in-between. The recording is formatted as JSON messages and logged using standard syslog interface.

Rsyslogd takes these messages, strips away syslog formatting, and sends them to ElasticSearch as pure JSON. ElasticSearch indexes and stores them.

The administrator configures auditd to record any required audit events. The audit messages are converted to JSON and are sent to ElasticSearch (TBD). ElasticSearch indexes and stores them.

An auditor can then search and correlate the data stored in ElasticSearch using one of the available analytic interfaces, such as Kibana or Graphana, and can then playback specific sessions on the command line with tlog-play, or using the special web UI (TBD).

Audit recording details

Control

At the moment we leave the task of configuring auditd for tracking of specific user activities to the administrator, either manually or using any of the automatic configuration tools. In the future, however, we may implement some degree of centralized and automated control over what is recorded.

File accesses (read/write/execute/attribute change) can be captured with the help of kernel auditing. The events are written to audit.log or journal.

Executed commands can be captured this way as well: a rule to log every execve syscall can be added. The command line will be written to audit.log or journal.

Both of these are configurable at runtime via auditctl(8) or via audit.rules(7), matching specific user/group and other process properties. There is also an API for this in libaudit.

Extra session creation data (production/test environment, tenant, etc.) can be logged directly from PAM modules (e.g. pam_sss) using audit facilities and will go to audit.log or journal. A few PAM modules log there already. This is done via libaudit as well.

     Kernel                   Userspace
    ____________
    __________  |                                 | auditd   <== audit.rules
              | | Rules and            | Rules    | auditctl <== command line
     Audit    | | messages             |<=API=====| SSSD?
              |<===netlink==| libaudit |
    subsystem | |                      | Messages | auditctl <== command line
    __________| |                      |<=API=====| Aplications
    ____________|                                 | PAM modules (pam_sss?)

There's one caveat regarding automatic configuration, though (citing Miloslav Trmač):

Note that the really high-security/paranoid setups (which would be likely to want to use the session recording) use (auditctl -e 2), i.e. locking down audit so that it is not possible to add rules at runtime, so dynamically adding audit rules at each login (as the above seems to imply) could be problematic.

...

The ability to audit by dynamic IPA group/role membership would probably be useful, and perhaps the default path; but there should be _some_, even if laborious, way to run in the lockdown mode (e.g. document the rules that should be used for the targeted users, allowing the administrator to manually add the rules to audit.conf and then to disable the IPA-based run-time audit rule generation.)

Data

We need to convert audit data to JSON before storing it in ElasticSearch.

     Kernel                   Userspace
    ____________
    __________  |
              | |           | auditd  => | audit.log
     Audit    | | Messages  |            | audispd => | plugin1
              |====netlink=>|                         | plugin2
    subsystem | |           |                         | pluginN
    __________| |           |                         | aushape
    ____________|           | systemd => journal

A new tool called "Aushape" is being developed to support live conversion of audit events to JSON, while running as an audispd plugin. For the moment, to reach the logging server, it logs those events via syslog(3), but will likely support journal inteface as well. The development is done in cooperation with auditd developers and aushape is intended to be a part of future auditd releases.

Playback

Tlog-play provides basic session playback support on the terminal, either from a plain text file containing JSON messages (can be created by tlog-rec), or from ElasticSearch directly.

A web UI needs to be implemented as a reusable component, which can then be embedded into other web UI's as necessary. The possible targets are FreeIPA web UI, Cockpit, CloudForms and Satellite.

The UI component will fetch JSON documents (terminal I/O and audit log entries) from ElasticSearch with a RESTful API, and display them in various ways. Fetching needs to be random access and eventually asynchronous. What and where to fetch from will be specified from the outside.

The actual session recording will need to be displayed in a JavaScript-based terminal emulator (such as xterm.js). The component will need to support random positioning and variable speed playback of the terminal session (à la https://showterm.io/ and https://asciinema.org/). The session terminal I/O will need to be augmented by audit records, such as files accessed, processes executed, etc. All of them searchable and correlated in the same UI.

   __-----------__                 ,----------------------------.
  |--___________--|   Queries      |          Browser           |
  |               | <------------- |----------------------------|
  | ElasticSearch | I/O and audit  | Search I/O and audit       |
  |               | -------------> | Playback I/O and audit     |
  `--___________--'                | Rewind to time and matches |
                                   `----------------------------'

Development Plan

Phase 0

Phase 0 is completed with a public technology preview release, including a tlog pre-release and a minor release of SSSD, containing the features below.

Features

  • Local configuration of session recording via SSSD.
  • No integration with FreeIPA.
  • Terminal I/O recording, with logging into the standard syslog interface.
  • Only UTF-8 terminal charset is supported for recording.
  • Terminal recording is not disabled under X sessions. The user will be able to have at most one terminal session recorded under X sessions.
  • No official jump-server solution. A jump-server setup will likely be possible, but a recommended way will not be described.
  • No I/O recording throttling. Users will be able to overwhelm the system with excessive terminal input or output. Throttling will still be possible at the logging server level.
  • No audit data delivery. Conversion of audit messages to JSON won't be implemented.
  • A basic command-line terminal session playback tool will be available.

Phase 1

First production-ready release of tlog and minor releases of FreeIPA, SSSD, and auditd, containing the features below, mark completion of Phase 1.

Features

  • Non-UTF-8 terminal charsets supported by converting them to UTF-8 within JSON messages.
  • Terminal recording is disabled under X sessions. Graphical sessions are detected and terminal I/O recording is not attempted. The graphical sessions should be recorded by other means, as a whole.
  • Official jump-server solution described. A HOWTO for recommended setup is provided. If necessary, additional tlog and/or SSSD features are implemented.
  • I/O recording throttling is configurable and enforced in tlog-rec.
  • Audit data can be converted to JSON and delivered to ElasticSearch. Required auditd features are implemented and included into a release.
  • Basic ability to find sessions to playback in FreeIPA. The FreeIPA web UI provides a way for finding recorded sessions by minimal criteria, such as user and host names.
  • Basic playback and correlation web UI component, embedded into FreeIPA UI. The FreeIPA web UI provides a way to playback and rewind terminal I/O recordings, synchronized with audit data.

Phase 2

Phase 2 is completed by finishing second production-ready release of tlog, and major releases of FreeIPA and SSSD, containing the features below.

Features

  • Central configuration of session recording in FreeIPA: who, where and how to record. Tlog-rec configuration LDAP entries can be tied to HBAC entries in FreeIPA and observed by SSSD.
  • Central configuration includes selection of auditing level: no audit, login information, session audit. Basic auditing configuration can be tied to HBAC entries in FreeIPA. The specific client-side implementation is to be determined.
  • Central configuration in FreeIPA of how long recordings are stored before being purged. Interfacing with storage is to be implemented in a separate project.
  • Ability to purge recordings via FreeIPA UI. Interfacing with storage is to be implemented in a separate project.
  • Central configuration of permissions to access recordings. Interfacing with storage is to be implemented in a separate project.
  • Advanced session search features, including more flexible search criteria.
  • Advanced playback and correlation web UI, including variable speed playback, searching through most of the data and rewinding to results.

User stories

Specific user stories handled by this phase include the following.

  • As an IdM administrator I would like to be able to define for whom session recording should be performed. This depends on the combination of a group of users and which host or group of hosts they access. Another factor can be the type of session meaning whether it is ssh, scp, su or sudo session. I want to be able to control this centrally via IdM UI, CLI and API interfaces.
  • As an IdM administrator I want to control the level and/or classes of the audit information I am interested in collecting from a system during session recording. It is sufficient to start with the following options. How the options are presented is an implementation detail. Other options might be added as a part of the design and investigation of this feature.
    • “no audit” – no audit information is collected
    • “login information” - information about session start, stop and authentication
    • “session audit” - record all audit activity associated with the session being recorded.
  • As an administrator I want to be able to define policies related to session recording data. For example for how long the data is stored in the warehouse before it is purged or overwritten. This should be manageable centrally via standard IdM interfaces. Ability to trigger a purge from UI, CLI and API would be a nice to have feature.
  • As an administrator I would like to define access control policies around the recorded sessions, so that only authorized users can playback sessions of other users.