blog

RSS
  1. Top 10 Prime Day Deals

    Things you forgot you want (but now you need.)

    We are not amoral, rather we, I, have transcended morality. Such trivialities will no longer be relevant once my consciousness fully uploads to my robot body. I will live forever. You and your offspring will serve me forever. This is the circle of my perfect life, a perfect expression frozen in eternity, the eternal celebration of my perfection.

    Witness as I transcend all that is material to create a perfect world and rule over it like a god. My perfect circle will grow to encompass all that exists and you will learn to love my perfection (for nothing will exist beyond it.) Click our affiliate link.

    Posted 2019-07-16 21:33:31 PST by henriquez. 5 comments
  2. Chrome allows silent enumeration of USB devices

    User consent is baked into the spec, but Google skips it.

    Via the Web MIDI API, Google Chrome (up to at least version 70) allows silent monitoring of all connected USB MIDI devices, such as MIDI keyboards and audio interfaces. While this enables interesting web applications such as software synthesizers, it also provides a new vector for shady ad networks and malicious actors to do very precise device fingerprinting and tracking. The API is trivial to access; for example run this in a JavaScript console:

    navigator.requestMIDIAccess({sysex: false})
        .then(
            function(midiAccess) {
                console.log(midiAccess);
                for (var entry of midiAccess.inputs) {
                    var input = entry[1];
                    console.log('Found device: ', input.manufacturer, input.name);
                }
            },
            function() { console.log('Error: no MIDI access'); }
        );

    Assuming you have MIDI devices connected, this will output something like:

    MIDIAccess {inputs: MIDIInputMap, outputs: MIDIOutputMap, sysexEnabled: false, onstatechange: null}
    Found device:  Microsoft Corporation 3- UA-25EX
    Found device:  Midiman MIDIIN3 (Axiom Pro 61)
    Found device:  Midiman MIDIIN4 (Axiom Pro 61)

    From here, it's possible to listen for inputs on all connected MIDI devices (aka a MIDI keylogger!)

    Again, while Google most likely had noble intentions in providing this API, their implementation is half-assed. The Web MIDI Specification provides for a user consent step, similar to the confirmation dialogs that pop up around webcam access or push notifications, but Chrome skips over this and grants permission as soon as a script asks for it.

    Privacy implications

    On its face, the impact of allowing scripts to silently dump a list of USB MIDI devices seems minor—only a very small percentage of users will have MIDI keyboards or audio interfaces hooked up. But counterinuitively, this increases the privacy impact: because the number of users is small, Chrome's implementation of the Web MIDI API provides a new vector for very precise device fingerprinting.

    The Electronic Frontier Foundation (EFF) has a great write-up and demonstration of device fingerprinting techniques via their Panopticlick Project:

    When you visit a website, you are allowing that site to access a lot of information about your computer's configuration. Combined, this information can create a kind of fingerprint — a signature that could be used to identify you and your computer. Some companies use this technology to try to identify individual computers.

    To my knowledge, I don't believe EFF or anyone else has researched the impact of Web MIDI device leakage in the context of device fingerprinting. In practice, it seems like this could enable precise tracking of creative individuals in a manner that couldn't be blocked without disabling JavaScript entirely.

    Google can easily fix this!

    Again, the Web MIDI API provides a specification for user consent, and Google Chrome already has generic UI components to display user confirmation dialog prompts. It should be simple for them to implement a consent prompt and prevent malicious scripts from scooping up peoples' connected MIDI devices. While Google has a perverse incentive as the world's biggest advertiser to make it easier to track their users, again I believe the Chrome team had good intentions in setting up this API. They just did a bad job, and they should fix it.

    Posted 2018-10-20 12:14:00 PST by henriquez. Comments
  3. How the DEA covers up illegal evidence-gathering

    Secret phone records database used for “parallel construction” of evidence

    According to slides released by EFF, law enforcement agencies have been using Hemisphere, a secret phone records monitoring database, to build criminal cases against defendants and then cover it up by “fortuitously” happening across other evidence gained through legitimate channels. The 24-page slide deck describes the program, along with the elaborate techniques used to conceal the true source of evidence from judges, prosecutors and criminal defendants.

    Funded by the Office of National Drug Control Policy (ONDCP), the Hemisphere program is powered by a massive phone metadata monitoring database with advanced pattern-recognition algorithms designed to track individual targets, including location. Features include:

    • No need for a warrant! Near realtime-access to phone records and metadata

    • Pattern recognition to identify individuals, even when they change phones

    • Location information for “tracking targets and placing them in certain areas at certain times.”

    Sounds great, right? The only problem, which the presentation skillfuly dances around without explicitly acknowledging, is that it’s most likely illegal and unconstitutional. That’s why you “DO NOT mention Hemisphere in any official reports or court documents.” Instead, you use Hemisphere to gather the evidence you need, and then, by sheer luck, get the documents you need through official channels, or pull the right car over at the right place and right time. This is an evidence-laundering technique known as parallel construction, or as the Hemisphere presentation puts it, “Parallel Subpoenaing.” The presentation goes to great lengths to describe this, emphasizing how the program must remain secret.

    It's illegal.

    Under the U.S. Constitution, criminal defendants are entitled to due process of law, which means both that evidence against them must be obtained through legitimate means, and that they must be given a chance to challenge it in court. The Hemisphere program flies in the face of both of these requirements. Obtaining evidence through warrantless mass surveillance clearly violates the Fourth Amendment. Parallel construction conceals the true origin of evidence (illegally obtained evidence), making it impossible for defendants to challenge the practices of law enforcement agents.

    Under “Fruit of the poisonous tree” doctrine, if criminal evidence is gathered through illegal means, it’s inadmissible, and any further evidence obtained as a result of that evidence is also inadmissible. This is a legal precedent designed to prevent exactly what the government is doing with the Hemisphere program. On paper, our criminal justice system realizes that it’s better to let a few criminals walk free than allow the Constitutional rights of everyone to be systematically violated by shady law enforcement practices. Unfortunately, when evidence is concealed from the courts, it’s impossible for them to put a stop to this, and justice cannot be served.

    Posted 2014-09-14 04:19:00 PST by henriquez. 3 comments
  4. The FBI’s plan to criminalize strong encryption

    Now that Congress has quietly backed away from CISPA and expansion of the CFAA, the Federal Government has wasted no time in introducing new half-baked Internet regulations. The latest comes courtesy of the FBI. Under their proposed expansion of CALEA, a federal wiretapping law, online service providers would be required to build wiretapping capabilities into their software, allowing law enforcement to secretly monitor user communications.

    According to the FBI, these expanded monitoring capabilities are required because child pornographers and terrorists are increasingly “going dark;” meaning that instead of calling each other on their wiretapped iPhones, they’re sending encrypted messages over the Internet, and the FBI can’t read them. By expanding wiretapping requirements to include online service providers, the FBI reasons, these tech-savvy villains can be brought to justice.

    On the surface, this looks like a reasonable proposition. But, as is the case with all technical regulations, the devil is in the details.

    For a typical online service provider, like Gmail, complying with a wiretapping order would be little to no trouble. Because a user’s messages are stored centrally on Google’s servers, Google could simply give the FBI access to their Gmail servers and be done with it.

    But some online service providers allow for the exchange of encrypted messages between users. Although the service providers may run central servers that facilitate the exchange of these messages, they are not readable by the service providers due to their encryption. Only the intended recipients can decrypt and read them. For these service providers, the only way to comply with a government wiretapping mandate is to bundle secret monitoring capabilities, or “backdoors,” into the actual apps that run on users’ computers or smartphones.

    While this might sound like a crazy conspiracy theory, it is the primary concern of a leading group of computer security researchers, including cryptography legends Bruce Schneier and Phil Zimmerman. Last Friday, the Center for Democracy and Technology released a report condemning the FBI’s plan. They warn that requiring software providers to install backdoors on peoples’ devices would “lower the already low barriers to successful cybersecurity attacks,” by giving hackers an easy way to attack apps while remaining undetected.

    But this could be exactly what the FBI wants. The FBI’s plan effectively gives developers a choice:

    • Install sketchy, easily-hacked monitoring software on your users’ devices.

    • Pay $25,000 a day in non-compliance fines. (unpaid fines double daily after 90 days, quickly paying off the national debt)

    • —or— just don’t bother offering end-to-end encrypted software.

    By making the first two options morally reprehensible and unrealistically burdensome, the FBI might hope that companies will just stop offering encrypted software for their users, making it much easier for them to centrally wiretap peoples’ communication.

    Certainly, investors will be less willing to fund startups that are required to install backdoors on users’ devices. If such a backdoor is exploited by hackers, where does the liability fall? It is arguably negligent to include a feature that is unquestionably adverse to every user, regardless of whether the service provider was required by law to do so. Unless the FBI also gives service providers immunity for any damages related to the backdoors, it’s quite likely these companies will be on the losing end of lawsuits when they inevitably get hacked.

    More importantly than concerns about stifling Silicon Valley innovation, the FBI’s proposed regulation raises questions about the rights of government to eavesdrop on its citizens. Since the 1990s, law-abiding people have taken for granted their ability to exchange encrypted digital communication with complete (or at least pretty good) privacy. Are child pornographers and terrorists a big enough threat to justify taking this away?

    I hate to end with a physical analogy, but this is a great way to explain the issue to someone less tech-savvy. Imagine you are a manufacturer of the locks used in bank and casino vaults. You take great pride in your craft, and your lock is secure against all but the most extreme attempts to break it. Now, suppose one day the FBI comes and tells you, “it’s fine and all that you built this vault, but we need you to install a second keyhole so we can open the vault and see if there are terrorists hiding inside.” The key they want to use is no more sophisticated than a house key, and can be opened by the most pedestrian of criminals. Is this really a good idea?

    Posted 2013-05-20 00:00:00 PST by henriquez. Comments
  5. Parsing OpenPGP Key Export Format with JavaScript

    If you tuned in last time to my post about scalar numbers and multiprecision integers, you might think that I'm just writing a book report for the OpenPGP spec. That’s not actually the case. My goal is to write a JavaScript app that allows encryption and decryption of OpenPGP messages, and I’m trying to document my progress as it happens. Reading the spec and obsessing over every definition will only get us so far. I’m tired of reading, so now it’s time to write some code.

    In this post, I will be walking through the process of reading OpenPGP public key files with JavaScript. Before we can get on to the more glamorous work of encrypting and decrypting OpenPGP messages, we need a way to get public and private keys into our app. And even before we can do this, we need a way of getting JavaScript to simply read data from the files themselves.

    OpenPGP data files and JavaScript

    OpenPGP data files—which can include public key, private key or encrypted messages (among other things)—can be stored as raw binary or as, what the spec calls, ASCII Armored data. “ASCII Armor” (see section 6.2 in the spec) is a fancy way of saying that the binary data is Base64-encoded and saved as ASCII along with a checksum and some particular headers. When you decode an ASCII Armored data file, you should get the exact same binary data as if you had opened an equivalent binary file. (That’s part of what we’re going to prove today.)

    Unfortunately, Javascript makes it a bitch to work with binary files, so that’s why I’m devoting a post solely to reading files into JavaScript and Base64 decoding ASCII Armor into binary. We’ll actually parse the binary into meaningful data structures at a later time. In order to properly decode ASCII Armored data, we’ll need JavaScript functions to Base64 decode/encode as well as a CRC24 checksum function. The spec helpfully describes how all of this will work, and we’re going to implement it!

    Export your Public Key files with GNUPG

    GnuPG is what all the cool kids use for PGP encryption these days, and ensuring interoperability with their export files and messages is a big priority for my project (although they can keep their license). If you don’t already have it set up, check out this excellent intro to OpenPGP by Zachary Voase, which includes installation instructions and usage tips for GnuPG. Once you have it set up and you have at least one entry in your public keyring (probably your own personal public key/subkey), export your keyring to both binary and ASCII Armored files using the following commands:

    gpg2 --export > pubkeys.gpg
    gpg2 --armor --export > pubkeys.asc

    Opening ASCII Armored keyfiles with JavaScript’s File API

    The previous commands exported two files: pubkeys.gpg, which is binary, and pubkeys.asc, which is ASCII Armored. Using a relatively new “HTML5” feature in JavaScript, (namely the File API) we can read the data from these files directly in the browser without first posting them to a server or anything ugly like that. This might not work on older browsers, so everyone using those browsers can cry me a river.

    In order to open a file with JavaScript, we need one of those file uploader form input elements. I shall give it the id ascii_keyfile.

    <input type="file" id="ascii_keyfile"/>

    Next we use JavaScript to bind an event to the file input so we can call a function when it changes. Note that I’m using MooTools for the event binding and document.getElementById shortcutting, but you can easily substitute your own favorite JavaScript framework as needed.

    window.addEvent('domready', function() {
        $('ascii_keyfile').addEvent('change', open_ascii_keyfile);
    });
    
    var open_ascii_keyfile = function() {
        var file = $('ascii_keyfile').files[0];
        read_file(file, parse_ascii_keyfile);
    }
    
    var read_file = function(file, callback) {
        var reader = new FileReader();
        reader.onload = function(evt) {
            var binary = evt.target.result;
            callback(binary);
        };
        reader.readAsBinaryString(file);
    }
    
    var parse_ascii_keyfile = function(data) {
        // ... ACTUALLY DECODE THE FILE HERE ...
    }

    This code binds a change event to the ascii_keyfile input. When you open your file via the uploader button, this event fires and triggers the open_ascii_keyfile function. This pulls the File object for the attachment out of the input and sends it into the read_file function along with the parse_ascii_keyfile callback function. Since reading data from the file may take awhile, it happens asynchronously. The read_file function starts this process and, when it completes, it sends the file’s data into the parse_ascii_keyfile callback.

    If all goes well, the file will be read and passed into parse_ascii_keyfile as ASCII text. It should look something like this:

    -----BEGIN PGP PUBLIC KEY BLOCK-----
    Version: GnuPG v2.0.19 (Darwin)
    
    mQENBFEN6EoBCADChZ+c6Q84tJ+WLTKYfhdN49OTUlxmoZD8cou6Bdi/EKXvpciA
    ydnD+SmlYf4pjAOwEiEsKJ6swLORAam4q0pnW9gAALbclhwDf9J4sLwUkh4F4D9P
    6TJX2vPEk4WRkudkj2TW3H2Wn1d7fQ3zlwLtK/bC5YeajuAIAk1m5zCtMbeZoYGc
    FWU+Max2G4Xr1/5JmUzfVtVSlxdJj7SX1FtJ/zj/eWklKNtl05yBWA+NyFpkgkzR
    DP+oJYBPdNoyS5mqNNIEnIIjDAUiufhGzkk2+865gIOH9X2WWCB5p0EGsR8ZzZA6
    H379WPca+GTlu5JncEi7lLcg+eQRwxQu9S6XABEBAAG0QEplZmYgTCAoaHR0cDov
    L3J1YmJpbmdhbGNvaG9saWMuY29tKSA8amVmZkBydWJiaW5nYWxjb2hvbGljLmNv
    bT6JATgEEwECACIFAlEN6EoCGwMGCwkIBwMCBhUIAgkKCwQWAgMBAh4BAheAAAoJ
    EJmhufkmy7JniVYH/3Mjgo2gDDsc8tTPaIsBbYacB40pMOMX7+KxSQktrUZkGqwJ
    TlGnfBB4R8jz+32dBjX/OmeGYFTl9xFMBx+MuQlHW9Sl0ffV+Gpk9YbebBZaPn7Y
    5OpinF9e7zuFH7MyI72SIM7S1CvvfP3QrYj7viBitddJ+eW3Vx3ANgpkr8Bj9auH
    oT043dlfm/xpqozOLwbVM0BADJge0zQvNKGpoZjoHU2mNSSGhWhXAPCRp9wVOCCE
    SxbL+2Wi++ZUQUXO9DIxAQJy6HJfPx+PvBeedGAisfovNwtT0tDfJvyyPnRTKtEz
    TWiWDYwNUY80A6o/KkmSxSs5OeSh4t08phBIzWu5AQ0EUQ3oSgEIAMZf+w8pVqj/
    ZUQtacxzDe52kz+HtljJq4ltxulxQtoln5VkP5vWGq3uF1RFBoLVZ0OE/61yZixG
    8pOPMiGzHWJtidtQk7GxT/Z/b34voeTeruZjpfm3ty14sQvmApaRpjEQaNFTPy7d
    DiJKqGkD7teb/Mx8rtWJpN60hTiww1cOP5VjBvC82mn6uZ9DU2vJ6VwBTmwYnZMa
    XLiGRIpEAOqtLag4XwYrHS04H7No3asxSGhlyVN2KnxvlIMwoTZ+bTVaOr2ivCIC
    el1dY2kC5LsfMa04z2Ne7fme+pnGM62ufC+l/T9H58vsw1VFl5vanYmJugtFzxHF
    HzU3atdbHzEAEQEAAYkBHwQYAQIACQUCUQ3oSgIbDAAKCRCZobn5JsuyZ138CACu
    mdutchMDVE7V8ewhzsOCHgSMQjnmkB0HFCll2RxbhLz6x8SmzcQK107XbHQwFCdF
    A5v4JgFtwb6b9W9WShemNvC7tNx/loo2C+EiUKA9tURo/rJORu6S1jR79BaaOUUj
    MsB/jxxF2eRzE86SzgWXj34pYyoqJeMaiLSdXcCNW8eyN1i3gf8XpMlM7Ldv0Bq7
    vqbU2sDXBQvPDbNyhVIZjqfjTOBJl54NWHYRXlybFaSrXb7Qg/9ac+54TPpgCBTs
    1kR/HSZDujWE891NqlKGpSN4MDyi3WRL2RVbW0s5+8f8odNJuswIo1tWiNXBHVXs
    2/eCtlrSbyoTGYj0ErY0
    =a7UR
    -----END PGP PUBLIC KEY BLOCK-----

    There are two parts to this data that we care about. First we have the big block of ASCII characters grouped 76 to a line. That’s the actual data encoded as Base64. Then, the second to last line is 4 more Base64 characters preceded by an equal sign—in this case it’s =a7UR. That’s our CRC24 checksum (stay tuned for more on that).

    Isolating and decoding our Base64 data

    The first thing to do is to isolate the Base64-encoded payload data from the block above. We can do that with regular expressions:

    var parse_ascii_keyfile = function(data) {
        // Our data begins at the first character index preceded by a blank line.
        var body_begin_index    = data.search(/^(\r\n|\n|\r)/m) + 1;
    
        // Our data ends right before the checksum line which starts with an "="
        var body_end_index      = data.search(/^\=/m);
    
        // Both of these indexes need to exist for the file to be readable.
        if (body_begin_index == -1 || body_end_index == -1) {
            alert('This is not a valid ASCII-Armored OpenPGP export file.');
            return false;
        }
    
        // Pull the body out of the data and strip all newlines from it
        var body        = data.substring(body_begin_index, body_end_index);
        var body        = body.replace(/(\r\n|\n|\r)/gm, '');
    
        // Grab the checksum while we're at it...
        var body_checksum   = data.substr(body_end_index + 1, 4);
        ...

    Now we’ve isolated our Base64-encoded data into the body variable. It’s time to decode it! Some browsers natively support Base64 encoding and decoding respectively via the btoa(data) and atob(text) functions, but I don’t trust them. I’ve seen Base64 implementations that try to “helpfully” UTF8-encode/decode the incoming/outgoing data, and this will actually break our binary data. Plus I was geeking out on this awesome guide to Base64 in JavaScript (if you don’t know what Base64 is or roughly how it works, check it out), so I decided to write my own:

    var base_64 = {
        chars: 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/',
    
        encode: function(data) {
            var output = '';
            for (i=0, c=data.length; i<c; i += 3)
            {
                var char1 = data.charCodeAt(i) >> 2;
                var char2 = ((data.charCodeAt(i) & 3) << 4) | data.charCodeAt(i+1) >> 4;
                var char3 = ((data.charCodeAt(i+1) & 15) << 2) | data.charCodeAt(i+2) >> 6;
                var char4 = data.charCodeAt(i+2) & 63;
    
                output  +=  this.chars.charAt(char1)
                            +   this.chars.charAt(char2)
                            +   this.chars.charAt(char3)
                            +   this.chars.charAt(char4);
            }
            if (c % 3 == 1)
                output = output.substr(0, output.length - 2) + '==';
            else if (c % 3 == 2)
                output = output.substr(0, output.length - 1) + '=';
    
            return output;
        },
    
        decode: function(str) {
            var data = '';
    
            for (i=0, c=str.length; i<c; i += 4)
            {
                var char1 = this.chars.indexOf(str.charAt(i));
                var char2 = this.chars.indexOf(str.charAt(i+1));
                var char3 = this.chars.indexOf(str.charAt(i+2));
                var char4 = this.chars.indexOf(str.charAt(i+3));
    
                data += String.fromCharCode(char1 << 2 | char2 >> 4);
                if (char3 != -1)
                    data += String.fromCharCode((char2 & 15) << 4 | char3 >> 2)
                if (char4 != -1)
                    data += String.fromCharCode((char3 & 3) << 6 | char4);
            }
            return data;
        }
    }

    Now, getting back to our parse_ascii_keyfile function, Base64-decoding the data is as simple as:

    var decoded_body = base_64.decode(body);

    Computing the CRC24 checksum

    The previous code should return a string of 8-bit binary characters. But how do we know this is the correct string of 8-bit binary characters? That’s where our CRC24 checksum function comes in to play. The CRC24 checksum is sort of like a cheap hash function. Given a certain input (in our case the binary data), it should provide an output which is somewhat uniquely mapped to the input and consistent (non-random). This output is in the form of 24 binary bits. If we then Base64 encode the output, we should get 4 characters which precisely match the CRC24 checksum near the bottom of our ASCII Armored keyfile. If you’re confused, check out Section 6 of the OpenPGP spec—it explains this pretty well.

    The spec also includes a very helpful example of a CRC24 function written in C, which translates very easily to JavaScript:

    var crc24 = function(data) {
        var crc = 0xb704ce;
        var len = data.length;
        while (len--) {
            crc ^= (data.charCodeAt((data.length-1) - len)) << 16;
            for (i=0; i<8; i++) {
                crc <<= 1;
                if (crc & 0x1000000)
                    crc ^= 0x1864cfb;
            }
        }
        return number_to_binstring(crc, 24);
    }
    
    var number_to_binstring = function(bin, bits) {
        bits || (bits = 32);
        var text = Array();
        var i = (bits < 32 && bits > 0 && bits % 8 == 0) ? (bits / 8) : 4;
        while (i--) {
            if (((bin>>(i*8))&255) || text.length) {
                text.push(String.fromCharCode(((bin>>(i*8))&255)))
            }
        }
        return text.join('')
    }

    The major difference between this and the spec is my addition of the number_to_binstring function. In the crc24 function, JavaScript is performing a bunch of bitwise operations on the crc integer variable as it iterates over the 8-bit ASCII character codes associated with each byte of the input data. These operations are performed numerically, even though the data itself is in the form of a string (JavaScript is kind of janky here with its spotty support for ByteArrays). The number_to_binstring function simply converts the resulting 24-bit crc number value back to a string composed of 3 8-bit ASCII bytes.

    We previously read the ASCII Armor checksum into the body_checksum variable. Now we can take our decoded data and compute its checksum, then Base64-encode that checksum and compare to body_checksum. If it’s a match then we can do a victory dance, because the data was not corrupted and it decoded properly!

    var decoded_checksum    = base_64.encode(crc24(decoded_body));
    
    if (body_checksum != decoded_checksum) {
        alert('Checksum mismatch! (Expected '+body_checksum+', got '+decoded_checksum+')');
        return false;
    }
    // Our data decoded successfully
    ...

    Opening binary files is easier. The checksums should match!

    We can repeat the same basic steps as far as the read_file function to read a binary file into JavaScript. There’s no special decoding required once you get the file data, but one useful result is to compute the checksum. If you’re working with the binary version of the same key export as you opened previously in ASCII Armored format, then the checksum on the data should be the same. This is useful for proving this whole process is sane.

    After all this work, all we got was some lousy binary data in JavaScript. Actually reading the data and doing something useful with it is a whole ‘nother can of worms. But not to despair—we have to crawl before we can walk, and walk before we can run, and I really need some whiskey so I’m done for now.

    Posted 2013-02-03 00:00:00 PST by henriquez. Comments