Slow Down Online Guessing Attacks with Device Cookies

Author: Anton Dedov
Contributor(s): Daniel Waller


Device cookies as additional authenticator for users devices have been discussed and used in practice for some time already. For example, it was discussed by Marc Heuse at PasswordsCon 14.

Marc speculates on various techniques for blocking online attacks and comes to notion of “device cookie” as good protection alternative (see the talk on Hydra at PasswordsCon 14, specifically from 32:50). As well as Alec Muffett at the same conference mentions “datr cookie” from Facebook (look from 10:15 of the talk).

The main idea behind the protocol is to issue a special “device” cookie to every client (browser) when it is used to successfully authenticate a user in a system. The device cookie can be used to:

  • Distinguish between known/trusted and unknown/untrusted clients
  • Establish universal temporary lockouts for all untrusted clients
  • Lockout trusted clients individually


There are few well-known ways to deal with online attacks:

  • Temporary account lockout
  • Use CAPTCHA to slow down attacker

Other tricks, like producing confusing answers for attacker are more like “security by obscurity” and cannot be used as first-class protection mechanism.

Temporary account lockout after several failed attempts is too simple of a target for DoS attacks against legitimate users. There is variation of this method that locks out pair of account/IP. It is better in regarding to DoS issues but have security downsides:

  • Attacks from botnets can be effective
  • Attacks through proxies can be effective

Moreover, it is not so trivial to implement account/IP lockout. Consider multiple proxies with chaining addresses in “X-Forwarded-For” header or IPv6.

The method described in this writing may be viewed as variant of account/IP blocking. But it proposes to use a browser cookie instead of an IP address. Thus it may be more predictable from security perspective and easier to implement.


Protocol parameters:

  • T – Time period for lockout duration/attempt counting
  • N – Max number of failed authentication attempts allowed during T

The sign ∎ hereafter states for end of algorithm.

Entry point for authentication request

  1. If the incoming request contains a device cookie:
    1. validate device cookie
    2. if device cookie is invalid, proceed to step 2.
    3. if the device cookie is in the lockout list -> reject authentication attempt ∎
    4. else -> authenticate user
  2. If authentication from untrusted clients is locked out for this user -> reject authentication attempt ∎
  3. Else -> authenticate user

Authenticate user

  1. Check user credentials
  2. If credentials are valid
    1. issue new device cookie to user’s client
    2. proceed with authenticated user
  3. Else
    1. register failed authentication attempt
    2. reject authentication attempt ∎

Register failed authentication attempt

  1. Register a failed authentication attempt with following information: { user, time, device cookie (if present) }
  2. If device cookie is present
    1. count number of failed authentication attempts within time period T for this specific cookie
    2. if number of failed attempts within T > N -> put device cookie in lockout list until now() + T
  3. Else (= no device cookie present)
    1. count number of all failed authentication attempts for this user during time period T
    2. if number of failed attempts within T > N -> lockout ALL authentication attempts from untrusted clients for this specific user until now() + T

Issue a browser cookie with a value like “LOGIN,NONCE,SIGNATURE”, where:

  • LOGIN – User’s login name (or internal ID) corresponding to an authenticated user
  • NONCE – Nonce of sufficient length or random value from CSRNG source
  • SIGNATURE – HMAC(secret-key, “LOGIN,NONCE”)
  • secret-key – server’s secret cryptographic key
  1. Validate that the device cookie is formatted as described above
  2. Validate that SIGNATURE == HMAC(secret-key, “LOGIN,NONCE”)
  3. Validate that LOGIN represents the user who is actually trying to authenticate


  • Putting the device cookie in a lockout list has the same effect as if the client had no device cookie. So clients with device cookies are potentially allowed to make N * 2 failed authentication attempts before actual lockout.
  • Issuing a new device cookie after each successful authentication allows us to avoid making decisions in situations like:

    “Should the system block a device cookie if there were registered N-1 failed attempts, then one successful authentication, and then again 1 failed attempt? All within period T.”

Implementation Tips

It is good idea to use a standard format for device cookies. So, if the size of a cookie is not an issue, it is recommended to use JWT.

The following standard claims can be used:

  • sub – LOGIN
  • jti – NONCE

Important: If you already use JWT for storing session tokens or other security stuff, make sure you cannot confuse device cookies with other types of tokens. There are two possible ways to mitigate this threat:

  1. Use different signature / encryption keys for different token types
  2. Add aud claim into device cookie token that unambiguously refers to brute force protection subsystem (e.g. “aud” : “brute-force-protection” or “device cookie”).

Threat Analysis


Threat details


Online attack against one user

Attacker performs a guessing attack against a specific account from an untrusted client(s).

As long as the lockout for untrusted clients blocks authN attempts for a specific user from all untrusted clients, the number of guesses is restricted with (N/T)*24h per day.

E.g. for N = 10 and T = 1h, any attacker will be limited with 240 attempts per day per user or 87600 attempts per user/year regardless of how large the botnet in their possession is.

If there is a good password policy in place that limits minimum number attempts to 106 then the targeted attack will last (on average) 5 years.

Online attack using stolen device cookies

Attacker performs long-running guessing attack against specific account(s) from known clients.

Using valid device cookies allows the attacker to scale attacks linearly, e.g. each device cookie adds extra N attempts per T for specific user.

There may be two ways to steal device cookies:

  • Using physical access to a known client
  • Active attack against a remote client

In both cases if the attacker can steal device cookies on a regular basis they may have enough power to steal not only device cookies but also session cookies or even passwords. However in our security model this attack is irrelevant if device cookies are adequately protected (use Secure and HttpOnly flags).

Proposal: Accidental access to one device cookie contributes little to attack speed. Applications may implement persistent lockout for certain device cookies to address such cases. E.g. permanently lockout a device cookie after N*10 failed attempts.

Online attack against multiple users

Attacker performs guessing attack against a specific number of accounts from untrusted clients.

Attacker may try the same password from a long list against many (all) accounts in a system without reaching N failed attempts per account during period T.

There is no specific mitigation for this threat in the protocol.

Good password policy may be a sufficient countermeasure for such attacks.

Additionally an application may temporarily require CAPTCHA solving for authentication from untrusted clients in case there are too many authentication attempts for different accounts during a specified period of time.

Spoof device cookie

Attacker may try to forge a valid device cookie for any user in a system.

Proper crypto implementation: HMAC and random secret key.

Tamper with existing device cookie

Attacker may try to tamper with a device cookie valid for one user to make it suitable for another.

Proper crypto implementation: HMAC and random secret key.

DoS for specific account

Attackers may cause permanent lockout for all untrusted devices for a specific user. Thus the user may be blocked from loggging into the system as they would need to login from a new device or to login after cleaning up their browser cache.

Issue a valid device cookie after visiting password reset link (an actual password reset is not necessary). Thus, if the user demonstrates their possession of a personal email account then the system may trust a client to try entering their credentials.

DoS for specific account when client is used by different accounts

There are situations when the same client is legitimately used to authenticate different users. For example, a family-shared PC or tablet. Thus, after successful authentication of one user another legitimate user will be treated as user of untrusted client and will be susceptible to a DoS attack.

Cases where the same client is shared between legitimate users may be considered out of scope by default.

If such cases are critical for application usability, it may deal with them by issuing device cookies with names that contain LOGIN and analyze them accordingly.