padding-spec.txt

                         Tor Padding Specification

                        Mike Perry, George Kadianakis

Note: This is an attempt to specify Tor as currently implemented.  Future
versions of Tor will implement improved algorithms.

This document tries to cover how Tor chooses to use cover traffic to obscure
various traffic patterns from external and internal observers. Other
implementations MAY take other approaches, but implementors should be aware of
the anonymity and load-balancing implications of their choices.

      The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL
      NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED",  "MAY", and
      "OPTIONAL" in this document are to be interpreted as described in
      RFC 2119.

Table of Contents

    1. Overview
    2. Connection-level padding
        2.1. Background
        2.2. Implementation
        2.3. Padding Cell Timeout Distribution Statistics
        2.4. Maximum overhead bounds
        2.5. Reducing or Disabling Padding via Negotiation
        2.6. Consensus Parameters Governing Behavior
    3. Circuit-level padding
        3.1. Circuit Padding Negotiation
        3.2. Circuit Padding Machine Message Management
        3.3. Obfuscating client-side onion service circuit setup
            3.3.1. Common general circuit construction sequences
            3.3.2. Client-side onion service introduction circuit obfuscation
            3.3.3. Client-side rendezvous circuit hiding
            3.3.4. Circuit setup machine overhead
        3.4. Circuit padding consensus parameters
    A. Acknowledgments

1. Overview

  Tor supports two classes of cover traffic: connection-level padding, and
  circuit-level padding.

  Connection-level padding uses the CELL_PADDING cell command for cover
  traffic, where as circuit-level padding uses the RELAY_COMMAND_DROP relay
  command. CELL_PADDING is single-hop only and can be differentiated from
  normal traffic by Tor relays ("internal" observers), but not by entities
  monitoring Tor OR connections ("external" observers).

  RELAY_COMMAND_DROP is multi-hop, and is not visible to intermediate Tor
  relays, because the relay command field is covered by circuit layer
  encryption. Moreover, Tor's 'recognized' field allows RELAY_COMMAND_DROP
  padding to be sent to any intermediate node in a circuit (as per Section
  6.1 of tor-spec.txt).

  Tor uses both connection level and circuit level padding. Connection
  level padding is described in section 2. Circuit level padding is
  described in section 3.

  The circuit-level padding system is completely orthogonal to the
  connection-level padding. The connection-level padding system regards
  circuit-level padding as normal data traffic, and hence the connection-level
  padding system will not add any additional overhead while the circuit-level
  padding system is actively padding.


2. Connection-level padding

2.1. Background

  Tor clients and relays make use of CELL_PADDING to reduce the resolution of
  connection-level metadata retention by ISPs and surveillance infrastructure.

  Such metadata retention is implemented by Internet routers in the form of
  Netflow, jFlow, Netstream, or IPFIX records.  These records are emitted by
  gateway routers in a raw form and then exported (often over plaintext) to a
  "collector" that either records them verbatim, or reduces their granularity
  further[1].

  Netflow records and the associated data collection and retention tools are
  very configurable, and have many modes of operation, especially when
  configured to handle high throughput. However, at ISP scale, per-flow records
  are very likely to be employed, since they are the default, and also provide
  very high resolution in terms of endpoint activity, second only to full packet
  and/or header capture.

  Per-flow records record the endpoint connection 5-tuple, as well as the
  total number of bytes sent and received by that 5-tuple during a particular
  time period. They can store additional fields as well, but it is primarily
  timing and bytecount information that concern us.

  When configured to provide per-flow data, routers emit these raw flow
  records periodically for all active connections passing through them
  based on two parameters: the "active flow timeout" and the "inactive
  flow timeout".

  The "active flow timeout" causes the router to emit a new record
  periodically for every active TCP session that continuously sends data. The
  default active flow timeout for most routers is 30 minutes, meaning that a
  new record is created for every TCP session at least every 30 minutes, no
  matter what. This value can be configured from 1 minute to 60 minutes on
  major routers.

  The "inactive flow timeout" is used by routers to create a new record if a
  TCP session is inactive for some number of seconds. It allows routers to
  avoid the need to track a large number of idle connections in memory, and
  instead emit a separate record only when there is activity. This value
  ranges from 10 seconds to 600 seconds on common routers. It appears as
  though no routers support a value lower than 10 seconds.

  For reference, here are default values and ranges (in parenthesis when
  known) for common routers, along with citations to their manuals.

  Some routers speak other collection protocols than Netflow, and in the
  case of Juniper, use different timeouts for these protocols. Where this
  is known to happen, it has been noted.

                            Inactive Timeout              Active Timeout
    Cisco IOS[3]              15s (10-600s)               30min (1-60min)
    Cisco Catalyst[4]         5min                        32min
    Juniper (jFlow)[5]        15s (10-600s)               30min (1-60min)
    Juniper (Netflow)[6,7]    60s (10-600s)               30min (1-30min)
    H3C (Netstream)[8]        60s (60-600s)               30min (1-60min)
    Fortinet[9]               15s                         30min
    MicroTik[10]              15s                         30min
    nProbe[14]                30s                         120s
    Alcatel-Lucent[2]         15s (10-600s)               30min (1-600min)

  The combination of the active and inactive netflow record timeouts allow us
  to devise a low-cost padding defense that causes what would otherwise be
  split records to "collapse" at the router even before they are exported to
  the collector for storage. So long as a connection transmits data before the
  "inactive flow timeout" expires, then the router will continue to count the
  total bytes on that flow before finally emitting a record at the "active
  flow timeout".

  This means that for a minimal amount of padding that prevents the "inactive
  flow timeout" from expiring, it is possible to reduce the resolution of raw
  per-flow netflow data to the total amount of bytes send and received in a 30
  minute window. This is a vast reduction in resolution for HTTP, IRC, XMPP,
  SSH, and other intermittent interactive traffic, especially when all
  user traffic in that time period is multiplexed over a single connection
  (as it is with Tor).

2.2. Implementation

  Tor clients currently maintain one TLS connection to their Guard node to
  carry actual application traffic, and make up to 3 additional connections to
  other nodes to retrieve directory information.

  We pad only the client's connection to the Guard node, and not any other
  connection. We treat Bridge node connections to the Tor network as client
  connections, and pad them, but otherwise not pad between normal relays.

  Both clients and Guards will maintain a timer for all application (ie:
  non-directory) TLS connections. Every time a non-padding packet is sent or
  received by either end, that endpoint will sample a timeout value from
  between 1.5 seconds and 9.5 seconds using the max(X,X) distribution
  described in Section 2.3. The time range is subject to consensus
  parameters as specified in Section 2.6.

  If the connection becomes active for any reason before this timer
  expires, the timer is reset to a new random value between 1.5 and 9.5
  seconds. If the connection remains inactive until the timer expires, a
  single CELL_PADDING cell will be sent on that connection.

  In this way, the connection will only be padded in the event that it is
  idle, and will always transmit a packet before the minimum 10 second inactive
  timeout.

2.3. Padding Cell Timeout Distribution Statistics

  It turns out that because the padding is bidirectional, and because both
  endpoints are maintaining timers, this creates the situation where the time
  before sending a padding packet in either direction is actually
  min(client_timeout, server_timeout).

  If client_timeout and server_timeout are uniformly sampled, then the
  distribution of min(client_timeout,server_timeout) is no longer uniform, and
  the resulting average timeout (Exp[min(X,X)]) is much lower than the
  midpoint of the timeout range.

  To compensate for this, instead of sampling each endpoint timeout uniformly,
  we instead sample it from max(X,X), where X is uniformly distributed.

  If X is a random variable uniform from 0..R-1 (where R=high-low), then the
  random variable Y = max(X,X) has Prob(Y == i) = (2.0*i + 1)/(R*R).

  Then, when both sides apply timeouts sampled from Y, the resulting
  bidirectional padding packet rate is now a third random variable:
  Z = min(Y,Y).

  The distribution of Z is slightly bell-shaped, but mostly flat around the
  mean. It also turns out that Exp[Z] ~= Exp[X]. Here's a table of average
  values for each random variable:

     R       Exp[X]    Exp[Z]    Exp[min(X,X)]   Exp[Y=max(X,X)]
     2000     999.5    1066        666.2           1332.8
     3000    1499.5    1599.5      999.5           1999.5
     5000    2499.5    2666       1666.2           3332.8
     6000    2999.5    3199.5     1999.5           3999.5
     7000    3499.5    3732.8     2332.8           4666.2
     8000    3999.5    4266.2     2666.2           5332.8
     10000   4999.5    5328       3332.8           6666.2
     15000   7499.5    7995       4999.5           9999.5
     20000   9900.5    10661      6666.2           13332.8

  In this way, we maintain the property that the midpoint of the timeout range
  is the expected mean time before a padding packet is sent in either
  direction.

2.4. Maximum overhead bounds

  With the default parameters and the above distribution, we expect a
  padded connection to send one padding cell every 5.5 seconds. This
  averages to 103 bytes per second full duplex (~52 bytes/sec in each
  direction), assuming a 512 byte cell and 55 bytes of TLS+TCP+IP headers.
  For a client connection that remains otherwise idle for its expected
  ~50 minute lifespan (governed by the circuit available timeout plus a
  small additional connection timeout), this is about 154.5KB of overhead
  in each direction (309KB total).

  With 2.5M completely idle clients connected simultaneously, 52 bytes per
  second amounts to 130MB/second in each direction network-wide, which is
  roughly the current amount of Tor directory traffic[11]. Of course, our
  2.5M daily users will neither be connected simultaneously, nor entirely
  idle, so we expect the actual overhead to be much lower than this.

2.5. Reducing or Disabling Padding via Negotiation

  To allow mobile clients to either disable or reduce their padding overhead,
  the CELL_PADDING_NEGOTIATE cell (tor-spec.txt section 7.2) may be sent from
  clients to relays. This cell is used to instruct relays to cease sending
  padding.

  If the client has opted to use reduced padding, it continues to send
  padding cells sampled from the range [9000,14000] milliseconds (subject to
  consensus parameter alteration as per Section 2.6), still using the
  Y=max(X,X) distribution. Since the padding is now unidirectional, the
  expected frequency of padding cells is now governed by the Y distribution
  above as opposed to Z. For a range of 5000ms, we can see that we expect to
  send a padding packet every 9000+3332.8 = 12332.8ms.  We also half the
  circuit available timeout from ~50min down to ~25min, which causes the
  client's OR connections to be closed shortly there after when it is idle,
  thus reducing overhead.

  These two changes cause the padding overhead to go from 309KB per one-time-use
  Tor connection down to 69KB per one-time-use Tor connection. For continual
  usage, the maximum overhead goes from 103 bytes/sec down to 46 bytes/sec.

  If a client opts to completely disable padding, it sends a
  CELL_PADDING_NEGOTIATE to instruct the relay not to pad, and then does not
  send any further padding itself.

2.6. Consensus Parameters Governing Behavior

  Connection-level padding is controlled by the following consensus parameters:

    * nf_ito_low
      - The low end of the range to send padding when inactive, in ms.
      - Default: 1500

    * nf_ito_high
      - The high end of the range to send padding, in ms.
      - Default: 9500
      - If nf_ito_low == nf_ito_high == 0, padding will be disabled.

    * nf_ito_low_reduced
      - For reduced padding clients: the low end of the range to send padding
        when inactive, in ms.
      - Default: 9000

    * nf_ito_high_reduced
      - For reduced padding clients: the high end of the range to send padding,
        in ms.
      - Default: 14000

    * nf_conntimeout_clients
      - The number of seconds to keep circuits opened and available for
        clients to use. Note that the actual client timeout is randomized
        uniformly from this value to twice this value. This governs client
        OR conn lifespan. Reduced padding clients use half the consensus
        value.
      - Default: 1800

    * nf_pad_before_usage
      - If set to 1, OR connections are padded before the client uses them
        for any application traffic. If 0, OR connections are not padded
        until application data begins.
      - Default: 1

    * nf_pad_relays
      - If set to 1, we also pad inactive relay-to-relay connections
      - Default: 0

    * nf_conntimeout_relays
      - The number of seconds that idle relay-to-relay connections are kept
        open.
      - Default: 3600


3. Circuit-level padding

  The circuit padding system in Tor is an extension of the WTF-PAD
  event-driven state machine design[15]. At a high level, this design places
  one or more padding state machines at the client, and one or more padding
  state machines at a relay, on each circuit.

  State transition and histogram generation has been generalized to be fully
  programmable, and probability distribution support was added to support more
  compact representations like APE[16]. Additionally, packet count limits,
  rate limiting, and circuit application conditions have been added.

  At present, Tor uses this system to deploy two pairs of circuit padding
  machines, to obscure differences between the setup phase of client-side
  onion service circuits, up to the first 10 cells.

  This specification covers only the resulting behavior of these padding
  machines, and thus does not cover the state machine implementation details or
  operation. For full details on using the circuit padding system to develop
  future padding defenses, see the research developer documentation[17].

3.1. Circuit Padding Negotiation

  Circuit padding machines are advertised as "Padding" subprotocol versions
  (see tor-spec.txt Section 9). The onion service circuit padding machines are
  advertised as "Padding=2".

  Because circuit padding machines only become active at certain points in
  circuit lifetime, and because more than one padding machine may be active at
  any given point in circuit lifetime, there is also a padding negotiation
  cell and a negotiated response. These are relay commands 41 and 42, with
  relay headers as per section 6.1 of tor-spec.txt.

  The fields of the relay cell Data payload of a negotiate request are
  as follows:

     const CIRCPAD_COMMAND_STOP = 1;
     const CIRCPAD_COMMAND_START = 2;

     const CIRCPAD_RESPONSE_OK = 1;
     const CIRCPAD_RESPONSE_ERR = 2;

     const CIRCPAD_MACHINE_CIRC_SETUP = 1;

     struct circpad_negotiate {
       u8 version IN [0];
       u8 command IN [CIRCPAD_COMMAND_START, CIRCPAD_COMMAND_STOP];

       u8 machine_type IN [CIRCPAD_MACHINE_CIRC_SETUP];

       u8 unused; // Formerly echo_request

       u32 machine_ctr;
     };

  When a client wants to start a circuit padding machine, it first checks that
  the desired destination hop advertises the appropriate subprotocol version for
  that machine. It then sends a circpad_negotiate cell to that hop with
  command=CIRCPAD_COMMAND_START, and machine_type=CIRCPAD_MACHINE_CIRC_SETUP (for
  the circ setup machine, the destination hop is the second hop in the
  circuit). The machine_ctr is the count of which machine instance this is on
  the circuit. It is used to disambiguate shutdown requests.

  When a relay receives a circpad_negotiate cell, it checks that it supports
  the requested machine, and sends a circpad_negotiated cell, which is formatted
  in the data payload of a relay cell with command number 42 (see tor-spec.txt
  section 6.1), as follows:

     struct circpad_negotiated {
       u8 version IN [0];
       u8 command IN [CIRCPAD_COMMAND_START, CIRCPAD_COMMAND_STOP];
       u8 response IN [CIRCPAD_RESPONSE_OK, CIRCPAD_RESPONSE_ERR];

       u8 machine_type IN [CIRCPAD_MACHINE_CIRC_SETUP];

       u32 machine_ctr;
     };

  If the machine is supported, the response field will contain
  CIRCPAD_RESPONSE_OK. If it is not, it will contain CIRCPAD_RESPONSE_ERR.

  Either side may send a CIRCPAD_COMMAND_STOP to shut down the padding machines
  (clients MUST only send circpad_negotiate, and relays MUST only send
  circpad_negotiated for this purpose).

  If the machine_ctr does not match the current machine instance count
  on the circuit, the command is ignored.

3.2. Circuit Padding Machine Message Management

  Clients MAY send padding cells towards the relay before receiving the
  circpad_negotiated response, to allow for outbound cover traffic before
  negotiation completes.

  Clients MAY send another circpad_negotiate cell before receiving the
  circpad_negotiated response, to allow for rapid machine changes.

  Relays MUST NOT send padding cells or circpad_negotiated cells, unless a
  padding machine is active. Any padding-related cells that arrive at the client
  from unexpected relay sources are protocol violations, and clients MAY
  immediately tear down such circuits to avoid side channel risk.

3.3. Obfuscating client-side onion service circuit setup

  The circuit padding currently deployed in Tor attempts to hide client-side
  onion service circuit setup. Service-side setup is not covered, because doing
  so would involve significantly more overhead, and/or require interaction with
  the application layer.

  The approach taken aims to make client-side introduction and rendezvous
  circuits match the cell direction sequence and cell count of 3 hop general
  circuits used for normal web traffic, for the first 10 cells only. The
  lifespan of introduction circuits is also made to match the lifespan
  of general circuits.

  Note that inter-arrival timing is not obfuscated by this defense.

3.3.1. Common general circuit construction sequences

  Most general Tor circuits used to surf the web or download directory
  information start with the following 6-cell relay cell sequence (cells
  surrounded in [brackets] are outgoing, the others are incoming):

    [EXTEND2] -> EXTENDED2 -> [EXTEND2] -> EXTENDED2 -> [BEGIN] -> CONNECTED

  When this is done, the client has established a 3-hop circuit and also opened
  a stream to the other end. Usually after this comes a series of DATA cell that
  either fetches pages, establishes an SSL connection or fetches directory
  information:

    [DATA] -> [DATA] -> DATA -> DATA...(inbound cells continue)

  The above stream of 10 relay cells defines the grand majority of general
  circuits that come out of Tor browser during our testing, and it's what we use
  to make introduction and rendezvous circuits blend in.

  Please note that in this section we only investigate relay cells and not
  connection-level cells like CREATE/CREATED or AUTHENTICATE/etc. that are used
  during the link-layer handshake. The rationale is that connection-level cells
  depend on the type of guard used and are not an effective fingerprint for a
  network/guard-level adversary.

3.3.2. Client-side onion service introduction circuit obfuscation

  Two circuit padding machines work to hide client-side introduction circuits:
  one machine at the origin, and one machine at the second hop of the circuit.
  Each machine sends padding towards the other. The padding from the origin-side
  machine terminates at the second hop and does not get forwarded to the actual
  introduction point.

  From Section 3.3.1 above, most general circuits have the following initial
  relay cell sequence (outgoing cells marked in [brackets]):

  [EXTEND2] -> EXTENDED2 -> [EXTEND2] -> EXTENDED2 -> [BEGIN] -> CONNECTED
    -> [DATA] -> [DATA] -> DATA -> DATA...(inbound data cells continue)

  Whereas normal introduction circuits usually look like:

  [EXTEND2] -> EXTENDED2 -> [EXTEND2] -> EXTENDED2 -> [EXTEND2] -> EXTENDED2
    -> [INTRO1] -> INTRODUCE_ACK

  This means that up to the sixth cell (first line of each sequence above),
  both general and intro circuits have identical cell sequences. After that
  we want to mimic the second line sequence of

    -> [DATA] -> [DATA] -> DATA -> DATA...(inbound data cells continue)

  We achieve this by starting padding INTRODUCE1 has been sent. With padding
  negotiation cells, in the common case of the second line looks like:

    -> [INTRO1] -> [PADDING_NEGOTIATE] -> PADDING_NEGOTIATED -> INTRO_ACK

  Then, the middle node will send between INTRO_MACHINE_MINIMUM_PADDING (7) and
  INTRO_MACHINE_MAXIMUM_PADDING (10) cells, to match the "...(inbound data cells
  continue)" portion of the trace (aka the rest of an HTTPS response body).

  We also set a special flag which keeps the circuit open even after the
  introduction is performed. With this feature the circuit will stay alive for
  the same duration as normal web circuits before they expire (usually 10
  minutes).

3.3.3. Client-side rendezvous circuit hiding

  Following a similar argument as for intro circuits, we are aiming for padded
  rendezvous circuits to blend in with the initial cell sequence of general
  circuits which usually look like this:

  [EXTEND2] -> EXTENDED2 -> [EXTEND2] -> EXTENDED2 -> [BEGIN] -> CONNECTED
     -> [DATA] -> [DATA] -> DATA -> DATA...(incoming cells continue)

  Whereas normal rendezvous circuits usually look like:

  [EXTEND2] -> EXTENDED2 -> [EXTEND2] -> EXTENDED2 -> [EST_REND] -> REND_EST
     -> REND2 -> [BEGIN]

  This means that up to the sixth cell (the first line), both general and
  rend circuits have identical cell sequences.

  After that we want to mimic a [DATA] -> [DATA] -> DATA -> DATA sequence.

  With padding negotiation right after the REND_ESTABLISHED, the sequence
  becomes:

  [EXTEND2] -> EXTENDED2 -> [EXTEND2] -> EXTENDED2 -> [EST_REND] -> REND_EST
     -> [PADDING_NEGOTIATE] -> [DROP] -> PADDING_NEGOTIATED -> DROP...

  After which normal application DATA cells continue on the circuit.

  Hence this way we make rendezvous circuits look like general circuits up
  till the end of the circuit setup.

  After that our machine gets deactivated, and we let the actual rendezvous
  circuit shape the traffic flow. Since rendezvous circuits usually imitate
  general circuits (their purpose is to surf the web), we can expect that they
  will look alike.

3.3.4. Circuit setup machine overhead

  For the intro circuit case, we see that the origin-side machine just sends a
  single [PADDING_NEGOTIATE] cell, whereas the origin-side machine sends a
  PADDING_NEGOTIATED cell and between 7 to 10 DROP cells. This means that the
  average overhead of this machine is 11 padding cells per introduction circuit.

  For the rend circuit case, this machine is quite light. Both sides send 2
  padding cells, for a total of 4 padding cells.

3.4. Circuit padding consensus parameters

  The circuit padding system has a handful of consensus parameters that can
  either disable circuit padding entirely, or rate limit the total overhead
  at relays and clients.

  * circpad_padding_disabled
    - If set to 1, no circuit padding machines will negotiate, and all
      current padding machines will cease padding immediately.
    - Default: 0

  * circpad_padding_reduced
    - If set to 1, only circuit padding machines marked as "reduced"/"low
      overhead" will be used. (Currently no such machines are marked
      as "reduced overhead").
    - Default: 0

  * circpad_global_allowed_cells
    - This is the number of padding cells that must be sent before
      the 'circpad_global_max_padding_percent' parameter is applied.
    - Default: 0

  * circpad_global_max_padding_percent
    - This is the maximum ratio of padding cells to total cells, specified
      as a percent. If the global ratio of padding cells to total cells
      across all circuits exceeds this percent value, no more padding is sent
      until the ratio becomes lower. 0 means no limit.
    - Default: 0

  * circpad_max_circ_queued_cells
    - This is the maximum number of cells that can be in the circuitmux queue
      before padding stops being sent on that circuit.
    - Default: CIRCWINDOW_START_MAX (1000)


A. Acknowledgments

  This research was supported in part by NSF grants CNS-1111539,
  CNS-1314637, CNS-1526306, CNS-1619454, and CNS-1640548.

1. https://en.wikipedia.org/wiki/NetFlow
2. http://infodoc.alcatel-lucent.com/html/0_add-h-f/93-0073-10-01/7750_SR_OS_Router_Configuration_Guide/Cflowd-CLI.html
3. http://www.cisco.com/en/US/docs/ios/12_3t/netflow/command/reference/nfl_a1gt_ps5207_TSD_Products_Command_Reference_Chapter.html#wp1185203
4. http://www.cisco.com/c/en/us/support/docs/switches/catalyst-6500-series-switches/70974-netflow-catalyst6500.html#opconf
5. https://www.juniper.net/techpubs/software/erx/junose60/swconfig-routing-vol1/html/ip-jflow-stats-config4.html#560916
6. http://www.jnpr.net/techpubs/en_US/junos15.1/topics/reference/configuration-statement/flow-active-timeout-edit-forwarding-options-po.html
7. http://www.jnpr.net/techpubs/en_US/junos15.1/topics/reference/configuration-statement/flow-active-timeout-edit-forwarding-options-po.html
8. http://www.h3c.com/portal/Technical_Support___Documents/Technical_Documents/Switches/H3C_S9500_Series_Switches/Command/Command/H3C_S9500_CM-Release1648%5Bv1.24%5D-System_Volume/200901/624854_1285_0.htm#_Toc217704193
9. http://docs-legacy.fortinet.com/fgt/handbook/cli52_html/FortiOS%205.2%20CLI/config_system.23.046.html
10. http://wiki.mikrotik.com/wiki/Manual:IP/Traffic_Flow
11. https://metrics.torproject.org/dirbytes.html
12. http://freehaven.net/anonbib/cache/murdoch-pet2007.pdf
13. https://gitweb.torproject.org/torspec.git/tree/proposals/188-bridge-guards.txt
14. http://www.ntop.org/wp-content/uploads/2013/03/nProbe_UserGuide.pdf
15. http://arxiv.org/pdf/1512.00524
16. https://www.cs.kau.se/pulls/hot/thebasketcase-ape/
17. https://github.com/torproject/tor/tree/master/doc/HACKING/CircuitPaddingDevelopment.md
18. https://www.usenix.org/node/190967
    https://blog.torproject.org/technical-summary-usenix-fingerprinting-paper