Take the 2-minute tour ×
Cryptography Stack Exchange is a question and answer site for software developers, mathematicians and others interested in cryptography. It's 100% free, no registration required.

Alice securely gives Bob a secret key so they can exchange 10 different messages encrypted with RC4. (Messages will include a counter for uniqueness and a MAC for authentication.) For each message exchange, Alice and Bob both start by permuting the standard byte array using the secret key (ie, the "KSA" key-scheduling algorithm, per Wikipedia):

for i from 0 to 255
    S[i] := i
endfor
j := 0
for i from 0 to 255
    j := (j + S[i] + key[i mod keylength]) mod 256
    swap values of S[i] and S[j]
endfor

At this point, Alice and Bob have each computed identical S[] arrays, and the secret key is no longer used by subsequent RC4 steps.

Shortcut 1: Alice decides that instead of securely sending Bob the secret key and making him do the KSA step for every message exchange, she'll compute the S[] array, securely send him the 256-byte array instead of the secret key to begin with, and from then on each time they need to encrypt/decrypt they'll both skip the KSA step to save time and just start with Alice's precomputed S[] array.

Shortcut 2: After using the above algorithm to compute the initial S[] array she had planned to send to Bob instead of the secret key, Alice realizes the KSA step seems to basically just want to scramble the S[] array. So instead of scrambling the array based on a key, she relies on the operating system random() function and just computes...

for i from 0 to 255
    S[i] := i
endfor
for i from 0 to 255
    swap(S[i], S[random(0..255)]) // swap S[i] with some arbitrary S[]
endfor

...and securely sends Bob her "randomly" scrambled array that doesn't even use a key.

Do those 2 shortcuts weaken RC4?

share|improve this question

2 Answers 2

up vote 1 down vote accepted

Regarding #1:

You don't lose anything with regards to the algorithm itself. The algorithm to go from the key to the key schedule (KS) is deterministic and well known, so anyone who gets the key can easily get the KS too. The actual algorithm itself isn't made less secure by communicating the KS instead of the key.

However, relying on the KS instead of the key makes things more practically difficult.

The scheme that you describe isn't secure because the encryption isn't randomized. All symmetric key encryption needs to be randomized. You are including a unique counter in the output, and some variations of that work for block ciphers, but that won't randomize the entire ciphertext for your stream cipher. For a stream cipher, the randomization occurs as a unique nonce that is combined with the key to generate a unique key stream that is in turn used to create a unique ciphertext.

Making your scheme secure could be done by adding nonces to the key for each encryption. So exchanging the KS itself actually makes using good encryption harder in this case because you can't simply exchange a new nonce for encryption, you have to exchange a whole new KS to include the change from the nonce. The nonce can be publicly exchanged and is short, but the KS must be kept secret and is longer. So then the question becomes, how are you securely exchanging the updated KSs? You might be able to securely agree upon the first key, but can you continuously agree upon new keys securely? Would you have to set all the KSs in advance when the original key is agreed upon? It gets messy.

If you are trying to optimize encryption, the standard way to do it is to exchange the key and then cache the KS on all the clients that use it. You would have to compute the KS once per client, but the RC4 KS is really short and shouldn't be a speed issue on any semi-modern platform. (Two loops of 256 count, an average of about three memory lookups and a few additions per loop. How bad could that be as a one-time performance penalty?) This way the optimization is abstracted within each client and you still have normal key exchange features.

So for your scheme, you shouldn't re-start the encryption for each message unless you want to include unique nonces for each reset. To avoid the nonces, you could just keep using the continued output stream of RC4 for all the messages (somehow ensuring that each client is synchronized as to how far into the stream they are). Either way would be better than the current scheme.

I know most of that isn't directly related to your question, but even though your proposal isn't inherently more insecure, it does make fixing your insecure protocol more difficult.

Regarding #2:

Note that, basically, you replaced

j := (j + S[i] + key[i mod keylength]) mod 256

with

j := random(0..255)

Assume for a moment that keylength is 256. Now consider the case that key and random() comes from a high quality entropy source, that is key is effectively just a random 256 byte string of the same quality that 256 bytes of random() would produce. The addition of constants and some slight "self-modification" from S[i] won't improve the quality of key. So for a good 256 byte random key, your scheme hasn't changed anything. In fact, you can calculate the effective key that your new scheme would create:

for i from 0 to 255
    S[i] := i
endfor
j := 0
for i from 0 to 255
    temp = random(0..255)
    k[i] := temp - j - S[i]
    j = temp
    swap(S[i], S[j])
endfor

That is to say, the key k here would have produced the same KS as your scheme (if given the same values from random()). This calculated key is still as secure as the output from random(). So instead of pre-generating a key to use, you're just generating it on the fly. There's really no difference.

(If the feedback-style role of j and S[i] concerns you, remember that we assumed that random() was a good entropy source. At best these variables are adding a random value to the output of random(), but that output was already securely random so no quality is added or removed. At worst they are adding constant values, and again it doesn't change anything.)

So your second scheme is the equivalent of just generating a 256 byte random key to begin with, and that's the best key that RC4 can use. Given that your scheme doesn't improve on the maximal security bounds of RC4, it doesn't really add anything. You can select the quality of the KS by choosing an appropriately strong key. If you want maximum security, just generate a high quality (256 byte) key. (Possibly via the call to random(), if you trust it to generate keys.)

Note:

Everything here assumes that you can distribute the initial key securely and that you trust the output of this random() function. It seems like you do assume this, but keep in mind that those are two very big assumptions. Ensure that your random() call is a cryptographically secure call, preferably to something like CryptGenRandom() on Windows or /dev/random on *nix.

share|improve this answer

The obvious problem with Shortcut 1 is that the same keystream will be used to encrypt all the messages. That is, an attacker can exclusive-or two ciphertexts together, and the result will be the exclusive-or of the two plaintexts; this may very well allow the attacker to recover both messages. How easy this is depends on the language the messages are in; if we assume they are in ASCII-English, this is fairly straight-forwarding as long as the messages aren't extremely brief.

There are two obvious problems with Shortcut 2:

  • How cryptographically secure is the operating system "random" function? Quite often, such operating system functions are only statistically random, this might not be good enough.

  • You say that Alice "securely" sends Bob the array; is she sending a separate array for each message? If so, why doesn't she use that secure method to send the message? If not (that is, she reused the same array for different messages), she runs into exactly the same problem as Shortcut 1.

Also, in shortcut 2, your pseudocode says "Assign a random S[] element to S[i]", while your comment says to swap them. Which do you mean? What you really want to do is swap the elements (it's important that S[] remain a permutation).

share|improve this answer
    
I think your concern with Shortcut 1 missed the point. Regardless of how many times they use the same key or switch to a new key, the question concerns sharing the precomputed array instead of sharing the key. Is it not true that giving Bob the array is equivalent to giving Bob the key, and Bob then doesn't have to compute the array? Any time you want to give Bob a key, how is it worse to give him the equivalent array instead? –  Witness Protection ID 44583292 Jun 5 '12 at 18:25
    
Re Shortcut 2, you ask "how cryptographically secure" the OS function is. The standard KSA swapping seems pseudorandom at best. If random() is "only statistically random", isn't that still at least as good as the standard KSA swapping? If a func like random() is used to scramble the array, what subsequent test could be used to determine the result is worse or better than the standard KSA swapping? Or, what's a known random() func that would be at least as good but possibly simpler than the standard KSA swapping? Is the standard swap trying to do something other than randomly scramble S[]? –  Witness Protection ID 44583292 Jun 5 '12 at 18:28
    
You ask if Alice is sending a separate array for each message. Just assume she's sending the array as frequently as she would be sending a new key. The point is not how or how often they are managing to securely exchange a key, the point is can't Alice simply send the array instead of the key so Bob can skip the KSA? As for the swapping, I've fixed the pseudocode. –  Witness Protection ID 44583292 Jun 5 '12 at 18:28
1  
@mike: Obviously, you never want to reuse the exact same RC4 key to encrypt two different messages. One thing you can do (if you have a counter) is compute Hash( actual_key, counter ), and use that as the RC4 key. However, if you precompute the permutation, you don't have the opportunity –  poncho Jun 5 '12 at 18:29
    
@mike It might miss the point of your question, but is an important consideration for using RC4 (and ciphers without initialization vectors in general): Never encrypt more than one message with the same key. –  Paŭlo Ebermann Jun 5 '12 at 18:30

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.