Week 10

I am struggling with different formats for encoding the keys this week. Basically, Nick is using the PEM keys on the publisher side, while Google's e2e library can only import the PGP encoded keys. Although the body of both formats is in Base64, they have different key structure, and I cannot find a converter between the two formats at this moment.

Here is an example of a PEM encoded private key.

The PEM encoded keys adopt the ASN.1 (Abstract Syntax Notation One) notation, and the structure of the key is as follows:

It is simply an array of different parameters of the RSA algorithm.

And here comes a sample PGP encoded private key.

The key material is encoded in packets, which in turn consist of packet header, indicating the type and length of the data, and packet body, the actual data, encrypted or unencrypted. For more information, please refer to Section 5.5 in RFC 4880.

If we decide to transform a key in PEM format into one in PGP format, so that we can make use of the e2e library to import a user key, we may need to parse the PEM key and then generate the PGP formatted key according to the parameters obtained from the PEM key. I found a ASN1 decoder in javascript and modified it a little bit to print out the parameters in the PEM key shown above.

All of the parameters are in hexadecimal. For instance, the version number is "0x00", which is 0, and the public exponent is "0x010001", which is 65537 in decimal, a common number for the public exponent in RSA.

I will discuss with other teammates about the issue of the key format. If we decide to adopt the PEM encoded keys on the publisher side, probably I need to write a converter between the two key formats, PEM and PGP.

Week 9

Sorry for the late update. First is some clarification on some of the statements I made in my previous blog last week. I misunderstood the usage of the passphrase in the openpgp keyring of the google's e2e library. It served just as a key for the encryption or decryption of the keyring, rather than an identifier of different keyrings. Therefore, for the support of multiple users, only the passphrase sounds not enough, and the google's library stores only one keyring encrypted with the passphrase, given that it is not empty. When an user enters an incorrect passphrase, the library gives the following warning:

As a matter of fact, the localStorage is not modified. Instead, because the HMAC key is also generated from the passphrase, as long as an incorrect passphrase is given, the computed HMAC is not the same as the one stored in localStorage.

In one word, the e2e library stores only one keyring encrypted under the passphrase, and if an incorrect passphrase is entered to access an encrypted keyring, the attempt fails since the HMACs do not match.

A basic file parser was added to the extension. For now, the extension is able to handle AES decryption automatically. For an encrypted file like this,

The cipher for encrypting the message, "aes-256-cbc", is specified after the header. The group number, 22, the group state, 0, and the initialization vector for decrypting the encrypted file key and initialization vector, are shown after the cipher. And the encrypted file key and initialization vector come next, which is followed by the header and the body of the encrypted message. Once the extension knows the current group state, which is obtained from the key server, it unwinds it to previous state if necessary and takes the hash value of the correct state as the key. Using this key, the extension gets the file key and initialization vector, which is encrypted under "aes-256-cbc" by default (note that this cipher may be different from the one specified in the header, and the one in the header is only the cipher for encrypting the message). With file key and initialization vector, the decrypted message is retrieved from the body of the encrypted message.

Finally, there seems to be a bug in injecting codes to subframes. Google does not provide API for injecting code for a specific subframe, so I import a library implemented by a third-party developer to allow me to do so. However, this library, although works, is not stable and fails to inject codes to a subframe sometimes. I just came up with a new idea to inject codes using random number as an identifier for different subframes. I am going to implement this and talk about it if it works in my next blog.

Week 8

I implemented a user interface for entering passphrase which may allow the support of multiple users in the future. Besides, I took a look at the google's e2e library, especially the keyring implementation of the openpgp, and played around with it.

The extension is off by default, unless a user enter a non-empty passphrase to turn it on, and an empty passphrase is used to turn off or disable the extension. The user interface looks like this:

After entering the passphrase, the extension is on and gets the keyring specified by the passphrase, so multiple users are allowed to use the extension by providing different passphrase.

When the extension is on, the user can change the passphrase by entering a new and non-empty one, or turn it off by entering an empty passphrase.

As for the google's e2e library, I took a closer look at the "keyring.js" file under the openpgp library this week so as to learn how they handle the key management.

The keyring and all the associated keys are stored using the HTML5LocalStorage. To be more specific, the whole keyring is first serialized, and then encrypted using the passphrase. What is stored is an encrypted string recording the information about the keyring. The following is the function for encrypting the serialized keyring.

As shown in the figure, the serialized keyring is encrypted using AES128 under CFB mode with a key generated from the passphrase and a random initialization vector. Also, an HMAC (Hash-based message authentication code) is included for further verification of data integrity and authenticity. Therefore, even though the attackers may modify the HTML5LocalStorage, the library can detect this and avoid using the modified keyring.

However, it seems that the library is only able to import keys with the openpgp format in ASCII Armor. If we want to import keys in other format, probably we need to implement our own keyring.

Week 7

The chrome extension supports <iframe> tags with the 'src' attribute pointing to a gnocchi file on the server now. Also, I change the way the browser interprets a gnocchi file to prevent the possible code injection attack. Rather than "text/html", the modified content-type of a gnocchi file is "text/plain", so all the injected codes would be regarded as just plain text by the chrome browser.

To resolve the <iframe> reference, I add two listeners: one is added on the event "chrome.webRequest.onHeadersReceived", which is fired when the request header is received to check for gnocchi files, and the other one is added on the event "chrome.webNavigation.onCompleted", which is fired when the document resources are completely loaded and initialized.

For now, I'm using a global variable to record the urls of the gnocchi files. I know that using global variables is not a good practice in coding, but it seems to me that it is the best solution. Alternatives are message sending and local storage. Message sending is more tedious and turning the code even less readable than global variables. Besides, I have no idea how to embed a listener on the event "chrome.webNavigation.onCompleted" inside another listener on the event of receiving message at this point. Anyway, I will manage message sending to see how well it works if I have time. As for local storage, that is a inferior choice considering the performance of the extension. Also, local storage runs asynchronously, competition between different processes makes the extension less reliable.

Here is the result of the test case with 20 <iframe> tags in a webpage (some of them are not visible due to the limited size):

Key management on the client side is my next step. I will take a look at google's e2e library to learn more about how they manage it. Furthermore, message sending between different listeners of the extension is also in my plan so as to get rid of annoying global variables.

Week 6

I removed redirection from the chrome extension this week, and did some research on the chrome file system.

First thanks go to Olga for inspiring me to find a way to circumvent redirection. Here is how it goes. The chrome extension checks the response headers to detect gnocchi files, as I mentioned before. If a gnocchi encryption or signature file is to be received, the extension will modify the headers so that the browser would process the response body as with content-type "text/html", rather than "multipart/signed" or "multipart/encrypted". In this way, an associated DOM document is created, whose content is the data received from the server. After the content is loaded, the extension injects codes to replace the encrypted message obtained from the server with the corresponding decrypted message.

Here is a possible attack I come up with. Since the DOM document is first loaded with the data from the server as its content. Is it possible for a malicious attacker to inject codes by sending them as the encrypted data?

As for the file system, fortunately, google provides an API named "goog.fs" for temporarily storing some files locally. What's more, "goog.fs" tends to work asynchronously, which is a piece of good news for the extension's performance. With a workable file system, I am going to solve the <iframe> src reference as follows. The extension would detect the web requests. And if a request to a gnocchi file is detected, the extension would make a XMLHttpRequest to retrieve the encrypted message first, decrypt it, store the decrypted message locally using the file system, and finally modify the src attribute in the <iframe> tags to reference to the local decrypted files.

Next week, I mean, in the near future (why should I say "next week", if, as a matter of fact, I am not finishing the tasks next week?), I am going to resolve the <iframe> src reference and take a look at how google's e2e library manage their key storage. Hopefully, google's method is helpful for me to figure a way for key management in the chrome extension.

Week 5

The chrome extension performs basic AES decryption now. The data are in three different encoding formats:

• Base64, for retrieving from the server
• Byte arrays, for decryption
• Text string, for displaying

We adopt the AES algorithm in CBC mode for decryption and encryption. The following is the data retrieved from the server:

Since we do not have mechanism for key distribution now, the decryption key is included in the data, the shorter base64 string. The key length indicates that the encryption algorithm is AES128. Also, the initialization vector (iv) for CBC decryption is included in the ciphertext, the longer base64 string, as the first 16 bytes. Knowing the key and the iv, we can obtain the plaintext by decryption.

The plaintext includes the message, "Hello Secret world!" and the corresponding signature in RSA scheme.

I also managed to come up with a way for key management and storage on the client side. Because it is generally considered not possible to use file systems on the browser-side in JavaScript (see http://stackoverflow.com/questions/585234/how-to-read-and-write-into-file-using-javascript), I have to use the not-so-secure chrome.storage API.

To ensure security, the following is a preliminary idea to encrypt the keys stored on the extension. Every user creates a user name and the associated password. All the keys are stored encrypted using AES with the hash value of the password as the key. To verify the user, another pair of plaintext and ciphertext is stored with the plaintext randomly generated. Thus, only the encrypted keys and a pair of plaintext and cipher text are stored. In order to check whether a user enters the correct password, we only have to take the hash value of the input, use it as the key to decrypt the ciphertext, and check whether the resulted plaintext matches the one stored on the extension.

For next week, I will try to do more research on this problem: how to store keys securely with storage that the attackers may have access to. Also, I will implement the idea above or some other solutions to solve the problem.

Week 4

"Redirection" was implemented and the end-to-end (e2e) google crypto library in javascript was imported this week.

The quotation make in "Redirection" indicates that it is in fact not real redirection. It works as follows. Once the extension detects a gnocchi file to be received from the server, it opens a new tab in the browser, displays the content (encrypted message for now, since I don't have time to work more on the crypto part), and closes the original tab. I had tried to realize the real redirection, but there seemed to be a bug in chrome. When I tried to access a gnocchi file from a loaded tab, a tab loaded with contents from other links, it got redirected to "about:blank", although the extension worked well in the case of a blank tab, or a newly-opened tab. Please see https://code.google.com/p/chromium/issues/detail?id=300685 for more information related to this issue. (Update before Friday night: I have just fixed the issue. It is now real redirection to the file content, instead of about:blank)

I also included the end-to-end (e2e) google crypto library in the project. It is well-known that there is few reliable crypto library implemented in javascript. I took a look at crypto-js (a nice one with friendly interface, also easy to import), and SJCL (started by people at Stanford University). However, due to the inactivity of both libraries, I selected e2e at last. Though newly developed, e2e is maintained and updated by the developers at Google. Besides, it is relatively easy to import e2e using Google's Closure Library. Finally, Google knows chrome best, so e2e is expected to be compatible with chrome.

I will focus on the crypto part of the project next week, in order to have decryption work. Also, if I have time, I am going to do more research on file IO in javascript, too. I slow down my research on that because we are mainly handling text message and html document at this moment, and file IO will be an extra feature in the future, rather than the goal at present. Furthermore, file IO is very very tricky in javascript...