Max De.Zines...The Maxx Code Blog: 01/01/2013

Jan 30, 2013

Dynamic Dictionaries with C# 4.0

source: http://www.aaron-powell.com/dynamic-dictionaries-with-csharp-4

Have you ever been working with the Dictionary object in .NET and just wanted to find some way in which you can do this:

  var dictionary = new Dictionary { { "hello", "world!" } };

  ...

  var something = dictionary.hello;

It'd be sweet, but it's not possible. The dictionary is just a bucket and there isn't a way it can know at compile type about the objects which are within it. Damn, so you just have to go via the indexer of the dictionary.

But really, using dot-notation could be really cool!

Well with the .NET 4.0 framework we now have a built in DLR so can we use the dynamic features of the C# 4 to this?

Introducing the DynamicObject

Well the answer is yes, yes you can do this, and it's really bloody easy, in fact you can do it in about 10 lines of code (if you leave out error checking and don't count curly braces :P).

First off you need to have a look at the DynamicObject which is in System.Runtime. There's a lot of different things you can do with the DynamicObject class, and things which you can change. For this we are going to work with TryGetMember, with this we just need to override the base implementation so we can add our own dot-notation handler!

So lets start with a class:



using System;

using System.Collections.Generic;

using System.Dynamic;



namespace AaronPowell.Dynamics.Collections

{

  public class DynamicDictionary : DynamicObject

   {

     private IDictionary dictionary;


     public DynamicDictionary(IDictionary dictionary)

      {

         this.dictionary = dictionary;

       }

   }

}

Essentially this is just going to be a wrapper for our dynamic implementation of a dictionary. So we're actually making a class which has a private property which takes a dictionary instance into the constructor.

Now we've got our object we need do some work to get it handle our dot-notation interaction. First we'll override the base implementation:


public override bool TryGetMember(GetMemberBinder binder, out object result)

{

   var key = binder.Name;

   if (dictionary.ContainsKey(key))

   {

     result = dictionary[key];

     return true;

   }

   throw new KeyNotFoundException(string.Format("Key \"{0}\" was not found in the given dictionary", key));

}

And you know what, we're actually done! Now all you have to do:


var dictionary = new Dictionary {{ "hello", "world!" }};

dynamic dynamicDictionary = new DyanmicDictionary(dictionary);

Console.WriteLine(dynamicDictionary.hello); //prints 'world'

I'm going to be releasing the source for this shortly (well, an improved version), along with a few other nifty uses for dynamic. So keep watching this space for that ;).

----------------------------------------------------------

By overriding the TryGetMember and TrySetMember method you can dynamically add and remove properties.

dynamic person = new DynamicClass; person.Name = “John Smith”; person.Phone = “32345690″;

By overriding the TrySetIndex and TryGetIndex you can access properties you don’t know about at compile time. dynamic person = new DynamicClass; person["Name"] = “John Smith”;

and then both are interchangable


public class DynamicDictionary : DynamicObject 
{ 
 
  private Dictionary _dictionary = new Dictionary();

  public override bool TryGetMember(GetMemberBinder binder, out object result)

  {

    return _dictionary.TryGetValue(binder.Name, out result);

  }


  public override bool TrySetMember(SetMemberBinder binder, object value)
  {

    _dictionary[binder.Name] = value;

    return true;

  }


  public override bool TrySetIndex(SetIndexBinder binder, object[] indexes, object value)

  {

    if (dictionary.ContainsKey((string)indexes[0]))

       _dictionary[(string)indexes[0]] = value;

    else

      _dictionary.Add((string)indexes[0], value);

    return true;

  }


public override bool TryGetIndex(GetIndexBinder binder, object[] indexes, out object result)
  {

    return _dictionary.TryGetValue((string)indexes[0], out result);

  }

}

}

Umbraco

While we were working on some sexy features for Umbraco 5 over the CodeGarden 10 retreat we kept saying that we should look at using as many of the cool new .NET framework features which we can possibly get away with. To this extent we kept saying we need to work out how to implement the dynamic keyword in some way.

Well that's where the idea for the above code came from, in fact we've got a similar piece of code which will be usable within the framework of Umbraco 5 and entity design. But the full info on that will belong to another post ;).

Released!

I've rolled the above code (with some improvements mind you) into a new project that I've been working on for making working with dynamics in .NET a whole lot easier. You can check out my Dynamics Library and get dynamacising.

Jan 16, 2013

C# 3.0 : Automatic Properties

Being a C# programmer , you’ve used to leverage propcode snippet to create a property . C# 3.0 provides you with a nice feature called “Automatic Properties” which enables you to avoid having to declare private accessor field for properties and write get/set logic , this will make your code more concise and legible . However , you can still create regular properties and change get/set logic.
This is a property created by C# 2.0 :






 private string propertyName;


public string PropertyName


 {
 get { return propertyName; }


 set { age= propertyName; } 

 
}

And this one is re-writed using C# 3.0 :




public string PropertyName { get; set; }

Implementing this feature , the compiler automatically creates the get/set logics . If you need to make the above property readonly , add private keyword before set accessor :






public string PropertyName { get; private set; }

Jan 15, 2013

Keeping Passwords Secure

Keeping sensitive data in database has always been one of the most challenging tasks which needs deep understanding of system planning and security issues that might affect the reliability of the system. Mainly, you may implement one of the following ways to keep passwords in database:

1 – Storing Passwords as Clear-Text/Plain-Text: This option is the most insecure way of storing passwords in DB, because each password is kept as clear text without any kind of encryption/hashing algorithms used. The disadvantage of putting this into work is that anyone with access to your database will be able to read passwords and modify them. This imposes a great potential risk on your system and is not recommended in any scenario.

2 – Using Encryption Algorithms This is a more secure approach that may be taken into account. In this way, you may implement one of the Asymmetric/Symmetric algorithms to encrypt / decrypt the data. In asymmetric algorithms, the key which is used to encrypt data differs from that of decrypting; there is a public key which enables you to encrypt password and a private key allowing you to decrypt it. In contrast with Asymmetric algorithms, Symmetric algorithms use a single key for encrypting/decrypting data. Intending to implement this method, use a unique public/private key for each user. Consequently if attacker obtains the private/public key of a user, it will not be usable for other users. This algorithm is mostly common in scenarios which need password retrieval system, meaning that passwords can be recovered. So if password retrieval is not a part of your system planning, this method is not recommended.

3 – Hashing Passwords This technique is a one-way solution and makes it more secure. A “Hash Function” gets a value of variable length and produces an output with constant length. For instance, SHA256 algorithm gets the input value and generates a 256 bits output. Note that hashes of two sets of data are identical if and only if the corresponding data matches and any minor change in data causes the hash value to change dramatically, so this is a cool method when comparing large amount of data. Password recovery is also not possible when implementing this solution. When using hash functions to store scrambled data, keep in mind that if two users have the same password. the hash output will be identical and this is considered as a great vulnerability. To prevent this, you may use “Salted-Hash technique”, it means that you should add some additional data to your password and then compute the hash of new generated data. For example you may append username to the password and then hash the whole string. Consequently when authenticating users, you must first append username to entered password, hash it and then comparing it with the hash value stored in database. For more security you can generate a random salt for each user and store it in database.

.Net Framework System.Security.Cryptography namespace provides you with many cryptographic services such as Encryption/Decryption , Hashing , Random Number generation , etc.

4 – OTP – One-Time Password In spite of implementing salted hash, passwords are still stored on hard disk and are prone to be cracked. So a better approach is required here. In contrast with static passwords, one-time passwords are changed each time a user logs on to the system and usually users should carry a small hardware used for synchronizing with server.

Mainly there are two types of OTP :

Time-Synchronized: In this method, user should enter a password in a given period of time, otherwise it will be expired and a new password is generated. Of course this method may lead to clock-skew problem, it means that if the authentication server and the user token don’t keep the same time, authentication process will fail.
Counter-Synchronized: A counter is synchronized between server and user client and each time the device requests an OTP value, the counter is advanced. Like the previous solution, when user wants to logs on, he enters the password shown on the device.

For more information, visit Safer Authentication with a One-Time Password Solution.

Some Security Tips

Try implementing salted hash.
Enforce password policies when registering users and do not allow weak passwords.
Separate authentication data ( i.e.username, password, salt, etc) and user profile data (i.e. FirstName,BirthDate, etc) into different tables.
Implement strong hash algorithms like SHA256, SHA384, and SHA256 (known as SHA2).
When connecting to database, use an account with the minimum required privileges.
Prevent SQL Injection attacks by using stored procedures, parameterized queries and validating data before passing them to query.
Obtain information on different kinds of attacks (i.e. Dictionary,Brute-Force, SQL Injection, etc) and how to defend against them.
Keep security tasks and functions secure by encrypting them.
Use secure connections when sending/receiving data to/from server. (I.e. SSL).
…

The issues that mentioned in this article are just a few tips that should be considered in order to provide a secure system and the given recommendations are all dependent upon system requirements and scenario.

Using SecureString

I’m going to elaborate on a great feature called System.Security.SecureString class which was introduced in .Net Framework 2.0. This class provides you with a secure way to store sensitive data and prevent them from being revealed by hackers. Implementing standard System.String class is not a secure way for keeping sensitive information and also swap file is in danger of being disclosed. Let’s take a look at some disadvantages of putting System.String class into work :

As it’s not encrypted , anyone with access to swap file or process memory is able to read unencrypted data easily.
When modifying this class , old value is not removed from the memory , so both old and new versions are kept in memory.
There is not a certain way to dispose it from memory when finishing with it.

SecureString class uses DAPI to encrypt data. Information ecrypted in this way by CLR is only decrypted when accessing it and in contrast with standard System.String class , this class implements IDisposable interface so that it can be cleared out from memory and its allocated memory will be zeroed out when disposing it.

Now , let’s see an example :

using System.Security;

using System.Runtime.InteropServices; 

using System;

using System.Windows.Forms;

namespace SecureStringProject

{

public class SecureStringExample

{

public void ImplementSecureString()

{

SecureString secureString = new SecureString();

///Implementing AppendChar method to add

///characters to SecureString Object.

secureString.AppendChar(‘A’);

secureString.AppendChar(‘C’);

secureString.AppendChar(‘G’);

secureString.AppendChar(‘E’);

secureString.AppendChar(‘F’);

///Implementing InsertAt method to insert a character at specified index.

secureString.InsertAt(1, ‘B’);

///Implementing SetAt method to replace character

///at specified index with new character.

secureString.SetAt(3, ‘D’);

///Implementing RemoveAt method to

///remove a character at specified index.

secureString.RemoveAt(5);

///Reading SecureStrinng content.

IntPtr pointer = Marshal.SecureStringToBSTR(secureString);

MessageBox.Show(Marshal.PtrToStringUni(pointer));

///Clearing SecureString Object.

secureString.Clear();

///Disposing SecureString Object.

secureString.Dispose();

///Free BSTR pointer allocated using

///SecureStringToBSTR method.

Marshal.ZeroFreeBSTR(pointer);

}

}

}

Https explained

(http://www.moserware.com/2009/06/first-few-milliseconds-of-https.html)

The First Few Milliseconds of an HTTPS Connection

Convinced from spending hours reading rave reviews, Bob eagerly clicked "Proceed to Checkout" for his gallon of Tuscan Whole Milk and...
Whoa! What just happened?

In the 220 milliseconds that flew by, a lot of interesting stuff happened to make Firefox change the address bar color and put a lock in the lower right corner. With the help of Wireshark, my favorite network tool, and a slightly modified debug build of Firefox, we can see exactly what's going on.
By agreement of RFC 2818, Firefox knew that "https" meant it should connect to port 443 at Amazon.com:

Most people associate HTTPS with SSL (Secure Sockets Layer) which was created by Netscape in the mid 90's. This is becoming less true over time. As Netscape lost market share, SSL's maintenance moved to the Internet Engineering Task Force (IETF). The first post-Netscape version was re-branded as Transport Layer Security (TLS) 1.0 which was released in January 1999. It's rare to see true "SSL" traffic given that TLS has been around for 10 years.

Client Hello

TLS wraps all traffic in "records" of different types. We see that the first byte out of our browser is the hex byte 0x16 = 22 which means that this is a "handshake" record:

The next two bytes are 0x0301 which indicate that this is a version 3.1 record which shows that TLS 1.0 is essentially SSL 3.1.
The handshake record is broken out into several messages. The first is our "Client Hello" message (0x01). There are a few important things here:

Random:

There are four bytes representing the current Coordinated Universal Time (UTC) in the Unix epoch format, which is the number of seconds since January 1, 1970. In this case, 0x4a2f07ca. It's followed by 28 random bytes. This will be used later on.
Session ID:

Here it's empty/null. If we had previously connected to Amazon.com a few seconds ago, we could potentially resume a session and avoid a full handshake.
Cipher Suites:

This is a list of all of the encryption algorithms that the browser is willing to support. Its top pick is a very strong choice of "TLS_ECDHE_ECDSA_WITH_AES_256_CBC_SHA" followed by 33 others that it's willing to accept. Don't worry if none of that makes sense. We'll find out later that Amazon doesn't pick our first choice anyway.
server_name extension:

This is a way to tell Amazon.com that our browser is trying to reach https://www.amazon.com/. This is really convenient because our TLS handshake occurs long before any HTTP traffic. HTTP has a "Host" header which allows a cost-cutting Internet hosting companies to pile hundreds of websites onto a single IP address. SSL has traditionally required a different IP for each site, but this extension allows the server to respond with the appropriate certificate that the browser is looking for. If nothing else, this extension should allow an extra week or so of IPv4 addresses.

Server Hello

Amazon.com replies with a handshake record that's a massive two packets in size (2,551 bytes). The record has version bytes of 0x0301 meaning that Amazon agreed to our request to use TLS 1.0. This record has three sub-messages with some interesting data:

"Server Hello" Message (2):
- We get the server's four byte time Unix epoch time representation and its 28 random bytes that will be used later.
- A 32 byte session ID in case we want to reconnect without a big handshake.
- Of the 34 cipher suites we offered, Amazon picked "TLS_RSA_WITH_RC4_128_MD5" (0x0004). This means that it will use the "RSA" public key algorithm to verify certificate signatures and exchange keys, the RC4 encryption algorithm to encrypt data, and the MD5 hash function to verify the contents of messages. We'll cover these in depth later on. I personally think Amazon had selfish reasons for choosing this cipher suite. Of the ones on the list, it was the one that was least CPU intensive to use so that Amazon could crowd more connections onto each of their servers. A much less likely possibility is that they wanted to pay special tribute to Ron Rivest, who created all three of these algorithms.
Certificate Message (11):
- This message takes a whopping 2,464 bytes and is the certificate that the client can use to validate Amazon's. It isn't anything fancy. You can view most of its contents in your browser:
"Server Hello Done" Message (14):
- This is a zero byte message that tells the client that it's done with the "Hello" process and indicate that the server won't be asking the client for a certificate.

Checking out the Certificate

The browser has to figure out if it should trust Amazon.com. In this case, it's using certificates. It looks at Amazon's certificate and sees that the current time is between the "not before" time of August 26th, 2008 and before the "not after" time of August 27, 2009. It also checks to make sure that the certificate's public key is authorized for exchanging secret keys.
Why should we trust this certificate?
Attached to the certificate is a "signature" that is just a really long number in big-endian format:

Anyone could have sent us these bytes. Why should we trust this signature? To answer that question, need to make a speedy detour into mathemagic land:

Interlude: A Short, Not Too Scary, Guide to RSA

People sometimes wonder if math has any relevance to programming. Certificates give a very practical example of applied math. Amazon's certificate tells us that we should use the RSA algorithm to check the signature. RSA was created in the 1970's by MIT professors Ron *R*ivest, Adi *S*hamir, and Len *A*dleman who found a clever way to combine ideas spanning 2000 years of math development to come up with a beautifully simple algorithm:
You pick two huge prime numbers "p" and "q." Multiply them to get "n = p*q." Next, you pick a small public exponent "e" which is the "encryption exponent" and a specially crafted inverse of "e" called "d" as the "decryption exponent." You then make "n" and "e" public and keep "d" as secret as you possibly can and then throw away "p" and "q" (or keep them as secret as "d"). It's really important to remember that "e" and "d" are inverses of each other.
Now, if you have some message, you just need to interpret its bytes as a number "M." If you want to "encrypt" a message to create a "ciphertext", you'd calculate:
C ≡ M^e (mod n)
This means that you multiply "M" by itself "e" times. The "mod n" means that we only take the remainder (e.g. "modulus") when dividing by "n." For example, 11 AM + 3 hours ≡ 2 (PM) (mod 12 hours). The recipient knows "d" which allows them to invert the message to recover the original message:
C^d ≡ (M^e)^d ≡ M^e*d ≡ M¹ ≡ M (mod n)
Just as interesting is that the person with "d" can "sign" a document by raising a message "M" to the "d" exponent:
M^d ≡ S (mod n)
This works because "signer" makes public "S", "M", "e", and "n." Anyone can verify the signature "S" with a simple calculation:
S^e ≡ (M^d)^e ≡ M^d*e ≡ M^e*d ≡ M¹ ≡ M (mod n)
Public key cryptography algorithms like RSA are often called "asymmetric" algorithms because the encryption key (in our case, "e") is not equal to (e.g. "symmetric" with) the decryption key "d". Reducing everything "mod n" makes it impossible to use the easy techniques that we're used to such as normal logarithms. The magic of RSA works because you can calculate/encrypt C ≡ M^e (mod n) very quickly, but it is really hard to calculate/decrypt C^d ≡ M (mod n) without knowing "d." As we saw earlier, "d" is derived from factoring "n" back to its "p" and "q", which is a tough problem.

Verifying Signatures

The big thing to keep in mind with RSA in the real world is that all of the numbers involved have to be big to make things really hard to break using the best algorithms that we have. How big? Amazon.com's certificate was "signed" by "VeriSign Class 3 Secure Server CA." From the certificate, we see that this VeriSign modulus "n" is 2048 bits long which has this 617 digit base-10 representation:

1890572922 9464742433 9498401781 6528521078 8629616064 3051642608 4317020197 7241822595 6075980039 8371048211 4887504542 4200635317 0422636532 2091550579 0341204005 1169453804 7325464426 0479594122 4167270607 6731441028 3698615569 9947933786 3789783838 5829991518 1037601365 0218058341 7944190228 0926880299 3425241541 4300090021 1055372661 2125414429 9349272172 5333752665 6605550620 5558450610 3253786958 8361121949 2417723618 5199653627 5260212221 0847786057 9342235500 9443918198 9038906234 1550747726 8041766919 1500918876 1961879460 3091993360 6376719337 6644159792 1249204891 7079005527 7689341573 9395596650 5484628101 0469658502 1566385762 0175231997 6268718746 7514321

(Good luck trying to find "p" and "q" from this "n" - if you could, you could generate real-looking VeriSign certificates.)
VeriSign's "e" is 2^16 + 1 = 65537. Of course, they keep their "d" value secret, probably on a safe hardware device protected by retinal scanners and armed guards. Before signing, VeriSign checked the validity of the contents that Amazon.com claimed on its certificate using a real-world "handshake" that involved looking at several of their business documents. Once VeriSign was satisfied with the documents, they used the SHA-1 hash algorithm to get a hash value of the certificate that had all the claims. In Wireshark, the full certificate shows up as the "signedCertificate" part:

It's sort of a misnomer since it actually means that those are the bytes that the signer is going to sign and not the bytes that already include a signature.
The actual signature, "S", is simply called "encrypted" in Wireshark. If we raise "S" to VeriSign's public "e" exponent of 65537 and then take the remainder when divided by the modulus "n", we get this "decrypted" signature hex value:

0001FFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF FFFFFFFF00302130 0906052B0E03021A 05000414C19F8786 871775C60EFE0542 E4C2167C830539DB

Per the PKCS #1 v1.5 standard, the first byte is "00" and it "ensures that the encryption block, [when] converted to an integer, is less than the modulus." The second byte of "01" indicates that this is a private key operation (e.g. it's a signature). This is followed by a lot of "FF" bytes that are used to pad the result to make sure that it's big enough. The padding is terminated by a "00" byte. It's followed by "30 21 30 09 06 05 2B 0E 03 02 1A 05 00 04 14" which is the PKCS #1 v2.1 way of specifying the SHA-1 hash algorithm. The last 20 bytes are SHA-1 hash digest of the bytes in "signedCertificate."
Since the decrypted value is properly formatted and the last bytes are the same hash value that we can calculate independently, we can assume that whoever knew "VeriSign Class 3 Secure Server CA"'s private key "signed" it. We implicitly trust that only VeriSign knows the private key "d."
We can repeat the process to verify that "VeriSign Class 3 Secure Server CA"'s certificate was signed by VeriSign's "Class 3 Public Primary Certification Authority."
But why should we trust that? There are no more levels on the trust chain.

The top "VeriSign Class 3 Public Primary Certification Authority" was signed by itself. This certificate has been built into Mozilla products as an implicitly trusted good certificate since version 1.4 of certdata.txt in the Network Security Services (NSS) library. It was checked-in on September 6, 2000 by Netscape's Robert Relyea with the following comment:

"Make the framework compile with the rest of NSS. Include a 'live' certdata.txt with those certs we have permission to push to open source (additional certs will be added as we get permission from the owners)."

This decision has had a relatively long impact since the certificate has a validity range of January 28, 1996 - August 1, 2028.
As Ken Thompson explained so well in his "Reflections on Trusting Trust", you ultimately have to implicitly trust somebody. There is no way around this problem. In this case, we're implicitly trusting that Robert Relyea made a good choice. We also hope that Mozilla's built-in certificate policy is reasonable for the other built-in certificates.
One thing to keep in mind here is that all these certificates and signatures were simply used to form a trust chain. On the public Internet, VeriSign's root certificate is implicitly trusted by Firefox long before you go to any website. In a company, you can create your own root certificate authority (CA) that you can install on everyone's machine.
Alternatively, you can get around having to pay companies like VeriSign and avoid certificate trust chains altogether. Certificates are used to establish trust by using a trusted third-party (in this case, VeriSign). If you have a secure means of sharing a secret "key", such as whispering a long password into someone's ear, then you can use that pre-shared key (PSK) to establish trust. There are extensions to TLS to allow this, such as TLS-PSK, and my personal favorite, TLS with Secure Remote Password (SRP) extensions. Unfortunately, these extensions aren't nearly as widely deployed and supported, so they're usually not practical. Additionally, these alternatives impose a burden that we have to have some other secure means of communicating the secret that's more cumbersome than what we're trying to establish with TLS (otherwise, why wouldn't we use that for everything?).
One final check that we need to do is to verify that the host name on the certificate is what we expected. Nelson Bolyard's comment in the SSL_AuthCertificate function explains why:

/* cert is OK. This is the client side of an SSL connection.
 * Now check the name field in the cert against the desired hostname.
 * NB: This is our only defense against Man-In-The-Middle (MITM) attacks! */

This check helps prevent against a man-in-the-middle attack because we are implicitly trusting that the people on the certificate trust chain wouldn't do something bad, like sign a certificate claiming to be from Amazon.com unless it actually was Amazon.com. If an attacker is able to modify your DNS server by using a technique like DNS cache poisoning, you might be fooled into thinking you're at a trusted site (like Amazon.com) because the address bar will look normal. This last check implicitly trusts certificate authorities to stop these bad things from happening.

Pre-Master Secret

We've verified some claims about Amazon.com and know its public encryption exponent "e" and modulus "n." Anyone listening in on the traffic can know this as well (as evidenced because we are using Wireshark captures). Now we need to create a random secret key that an eavesdropper/attacker can't figure out. This isn't as easy as it sounds. In 1996, researchers figured out that Netscape Navigator 1.1 was using only three sources to seed their pseudo-random number generator (PRNG). The sources were: the time of day, the process id, and the parent process id. As the researchers showed, these "random" sources aren't that random and were relatively easy to figure out.
Since everything else was derived from these three "random" sources, it was possible to "break" the SSL "security" in 25 seconds on a 1996 era machine. If you still don't believe that finding randomness is hard, just ask the Debian OpenSSL maintainers. If you mess it up, all the security built on top of it is suspect.
On Windows, random numbers used for cryptographic purposes are generated by calling the CryptGenRandom function that hashes bits sampled from over 125 sources. Firefox uses this function along with some bits derived from its own function to seed its pseudo-random number generator.
The 48 byte "pre-master secret" random value that's generated isn't used directly, but it's very important to keep it secret since a lot of things are derived from it. Not surprisingly, Firefox makes it hard to find out this value. I had to compile a debug version and set the SSLDEBUGFILE and SSLTRACE environment variables to see it.
In this particular session, the pre-master secret showed up in the SSLDEBUGFILE as:

4456: SSL[131491792]: Pre-Master Secret [Len: 48]
03 01 bb 7b 08 98 a7 49 de e8 e9 b8 91 52 ec 81 ...{...I.....R..
4c c2 39 7b f6 ba 1c 0a b1 95 50 29 be 02 ad e6 L.9{......P)....
ad 6e 11 3f 20 c4 66 f0 64 22 57 7e e1 06 7a 3b .n.? .f.d"W~..z;

Note that it's not completely random. The first two bytes are, by convention, the TLS version (03 01).

Trading Secrets

We now need to get this secret value over to Amazon.com. By Amazon's wishes of "TLS_RSA_WITH_RC4_128_MD5", we will use RSA to do this. You could make your input message equal to just the 48 byte pre-master secret, but the Public Key Cryptography Standard (PKCS) #1, version 1.5 RFC tells us that we should pad these bytes with random data to make the input equal to exactly the size of the modulus (1024 bits/128 bytes). This makes it harder for an attacker to determine our pre-master secret. It also gives us one last chance to protect ourselves in case we did something really bone-headed, like reusing the same secret. If we reused the key, the eavesdropper would likely see a different value placed on the network due to the random padding.
Again, Firefox makes it hard to see these random values. I had to insert debugging statements into the padding function to see what was going on:

wrapperHandle = fopen("plaintextpadding.txt", "a");
fprintf(wrapperHandle, "PLAINTEXT = ");
for(i = 0; i < modulusLen; i++)
{
    fprintf(wrapperHandle, "%02X ", block[i]);
}
fprintf(wrapperHandle, "\r\n");
fclose(wrapperHandle);

In this session, the full padded value was:

00 02 12 A3 EA B1 65 D6 81 6C 13 14 13 62 10 53 23 B3 96 85 FF 24 FA CC 46 11 21 24 A4 81 EA 30 63 95 D4 DC BF 9C CC D0 2E DD 5A A6 41 6A 4E 82 65 7D 70 7D 50 09 17 CD 10 55 97 B9 C1 A1 84 F2 A9 AB EA 7D F4 CC 54 E4 64 6E 3A E5 91 A0 06 00 03 01 BB 7B 08 98 A7 49 DE E8 E9 B8 91 52 EC 81 4C C2 39 7B F6 BA 1C 0A B1 95 50 29 BE 02 AD E6 AD 6E 11 3F 20 C4 66 F0 64 22 57 7E E1 06 7A 3B

Firefox took this value and calculated "C ≡ M^e (mod n)" to get the value we see in the "Client Key Exchange" record:

Finally, Firefox sent out one last unencrypted message, a "Change Cipher Spec" record:

This is Firefox's way of telling Amazon that it's going to start using the agreed upon secret to encrypt its next message.

Deriving the Master Secret

If we've done everything correctly, both sides (and only those sides) now know the 48 byte (256 bit) pre-master secret. There's a slight trust issue here from Amazon's perspective: the pre-master secret just has bits that were generated by the client, they don't take anything into account from the server or anything we said earlier. We'll fix that be computing the "master secret." Per the spec, this is done by calculating:

master_secret = PRF(pre_master_secret, "master secret", ClientHello.random + ServerHello.random)

The "pre_master_secret" is the secret value we sent earlier. The "master secret" is simply a string whose ASCII bytes (e.g. "6d 61 73 74 65 72 ...") are used. We then concatenate the random values that were sent in the ClientHello and ServerHello (from Amazon) messages that we saw at the beginning.
The PRF is the "Pseudo-Random Function" that's also defined in the spec and is quite clever. It combines the secret, the ASCII label, and the seed data we give it by using the keyed-Hash Message Authentication Code (HMAC) versions of both MD5 and SHA-1 hash functions. Half of the input is sent to each hash function. It's clever because it is quite resistant to attack, even in the face of weaknesses in MD5 and SHA-1. This process can feedback on itself and iterate forever to generate as many bytes as we need.
Following this procedure, we obtain a 48 byte "master secret" of

4C AF 20 30 8F 4C AA C5 66 4A 02 90 F2 AC 10 00 39 DB 1D E0 1F CB E0 E0 9D D7 E6 BE 62 A4 6C 18 06 AD 79 21 DB 82 1D 53 84 DB 35 A7 1F C1 01 19

Generating Lots of Keys

Now that both sides have a "master secrets", the spec shows us how we can derive all the needed session keys we need using the PRF to create a "key block" where we will pull data from:

key_block = PRF(SecurityParameters.master_secret, "key expansion", SecurityParameters.server_random + SecurityParameters.client_random);

The bytes from "key_block" are used to populate the following:

client_write_MAC_secret[SecurityParameters.hash_size]
server_write_MAC_secret[SecurityParameters.hash_size]
client_write_key[SecurityParameters.key_material_length]
server_write_key[SecurityParameters.key_material_length]
client_write_IV[SecurityParameters.IV_size]
server_write_IV[SecurityParameters.IV_size]

Since we're using a stream cipher instead of a block cipher like the Advanced Encryption Standard (AES), we don't need the Initialization Vectors (IVs). Therefore, we just need two Message Authentication Code (MAC) keys for each side that are 16 bytes (128 bits) each since the specified MD5 hash digest size is 16 bytes. In addition, the RC4 cipher uses a 16 byte (128 bit) key that both sides will need as well. All told, we need 2*16 + 2*16 = 64 bytes from the key block.
Running the PRF, we get these values:

client_write_MAC_secret = 80 B8 F6 09 51 74 EA DB 29 28 EF 6F 9A B8 81 B0
server_write_MAC_secret = 67 7C 96 7B 70 C5 BC 62 9D 1D 1F 4A A6 79 81 61
client_write_key = 32 13 2C DD 1B 39 36 40 84 4A DE E5 6C 52 46 72
server_write_key = 58 36 C4 0D 8C 7C 74 DA 6D B7 34 0A 91 B6 8F A7

Prepare to be Encrypted!

The last handshake message the client sends out is the "Finished message." This is a clever message that proves that no one tampered with the handshake and it proves that we know the key. The client takes all bytes from all handshake messages and puts them into a "handshake_messages" buffer. We then calculate 12 bytes of "verify_data" using the pseudo-random function (PRF) with our master key, the label "client finished", and an MD5 and SHA-1 hash of "handshake_messages":

verify_data = PRF(master_secret, "client finished", MD5(handshake_messages) + SHA-1(handshake_messages)) [12]

We take the result and add a record header byte "0x14" to indicate "finished" and length bytes "00 00 0c" to indicate that we're sending 12 bytes of verify data. Then, like all future encrypted messages, we need to make sure the decrypted contents haven't been tampered with. Since our cipher suite in use is TLS_RSA_WITH_RC4_128_MD5, this means we use the MD5 hash function.
Some people get paranoid when they hear MD5 because it has some weaknesses. I certainly don't advocate using it as-is. However, TLS is smart in that it doesn't use MD5 directly, but rather the HMAC version of it. This means that instead of using MD5(m) directly, we calculate:

HMAC_MD5(Key, m) = MD5((Key ⊕ opad) ++ MD5((Key ⊕ ipad) ++ m)

(The ⊕ means XOR, ++ means concatenate, "opad" is the bytes "5c 5c ... 5c", and "ipad" is the bytes "36 36 ... 36").
In particular, we calculate:

HMAC_MD5(client_write_MAC_secret, seq_num + TLSCompressed.type + TLSCompressed.version + TLSCompressed.length + TLSCompressed.fragment));

As you can see, we include a sequence number ("seq_num") along with attributes of the plaintext message (here it's called "TLSCompressed"). The sequence number foils attackers who might try to take a previously encrypted message and insert it midstream. If this occurred, the sequence numbers would definitely be different than what we expected. This also protects us from an attacker dropping a message.
All that's left is to encrypt these bytes.

RC4 Encryption

Our negotiated cipher suite was TLS_RSA_WITH_RC4_128_MD5. This tells us that we need to use Ron's Code #4 (RC4) to encrypt the traffic. Ron Rivest developed the RC4 algorithm to generate random bytes based on a 256 byte key. The algorithm is so simple you can actually memorize it in a few minutes.
RC4 begins by creating a 256-byte "S" byte array and populating it with 0 to 255. You then iterate over the array by mixing in bytes from the key. You do this to create a state machine that is used to generate "random" bytes. To generate a random byte, we shuffle around the "S" array.
Put graphically, it looks like this:

To encrypt a byte, we xor this pseudo-random byte with the byte we want to encrypt. Remember that xor'ing a bit with 1 causes it to flip. Since we're generating random numbers, on average the xor will flip half of the bits. This random bit flipping is effectively how we encrypt data. As you can see, it's not very complicated and thus it runs quickly. I think that's why Amazon chose it.
Recall that we have a "client_write_key" and a "server_write_key." The means we need to create two RC4 instances: one to encrypt what our browser sends and the other to decrypt what the server sent us.
The first few random bytes out of the "client_write" RC4 instance are "7E 20 7A 4D FE FB 78 A7 33 ..." If we xor these bytes with the unencrypted header and verify message bytes of "14 00 00 0C 98 F0 AE CB C4 ...", we'll get what appears in the encrypted portion that we can see in Wireshark:

The server does almost the same thing. It sends out a "Change Cipher Spec" and then a "Finished Message" that includes all handshake messages, including the decrypted version of the client's "Finished Message." Consequently, this proves to the client that the server was able to successfully decrypt our message.

Welcome to the Application Layer!

Now, 220 milliseconds after we started, we're finally ready for the application layer. We can now send normal HTTP traffic that'll be encrypted by the TLS layer with the RC4 write instance and decrypt traffic with the server RC4 write instance. In addition, the TLS layer will check each record for tampering by computing the HMAC_MD5 hash of the contents.
At this point, the handshake is over. Our TLS record's content type is now 23 (0x17). Encrypted traffic begins with "17 03 01" which indicate the record type and TLS version. These bytes are followed by our encrypted size, which includes the HMAC hash.
Encrypting the plaintext of:

GET /gp/cart/view.html/ref=pd_luc_mri HTTP/1.1
Host: www.amazon.com
User-Agent: Mozilla/5.0 (Windows; U; Windows NT 6.0; en-US; rv:1.9.0.10) Gecko/2009060911 Minefield/3.0.10 (.NET CLR 3.5.30729)
Accept: text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8
Accept-Language: en-us,en;q=0.5
Accept-Encoding: gzip,deflate
Accept-Charset: ISO-8859-1,utf-8;q=0.7,*;q=0.7
Keep-Alive: 300
Connection: keep-alive
...

will give us the bytes we see on the wire:

The only other interesting fact is that the sequence number increases on each record, it's now 1 (and the next record will be 2, etc).
The server does the same type of thing on its side using the server_write_key. We see its response, including the tell-tale application data header:

Decrypting this gives us:

HTTP/1.1 200 OK
Date: Wed, 10 Jun 2009 01:09:30 GMT
Server: Server
...
Cneonction: close
Transfer-Encoding: chunked

which is a normal HTTP reply that includes a non-descriptive "Server: Server" header and a misspelled "Cneonction: close" header coming from Amazon's load balancers.
TLS is just below the application layer. The HTTP server software can act as if it's sending unencrypted traffic. The only change is that it writes to a library that does all the encryption. OpenSSL is a popular open-source library for TLS.
The connection will stay open while both sides send and receive encrypted data until either side sends out a "closure alert" message and then closes the connection. If we reconnect shortly after disconnecting, we can re-use the negotiated keys (if the server still has them cached) without using public key operations, otherwise we do a completely new full handshake.
It's important to realize that application data records can be anything. The only reason "HTTPS" is special is because the web is so popular. There are lots of other TCP/IP based protocols that ride on top of TLS. For example, TLS is used by FTPS and secure extensions to SMTP. It's certainly better to use TLS than inventing your own solution. Additionally, you'll benefit from a protocol that has withstood careful security analysis.

... And We're Done!

The very readable TLS RFC covers many more details that were missed here. We covered just one single path in our observation of the 220 millisecond dance between Firefox and Amazon's server. Quite a bit of the process was affected by the TLS_RSA_WITH_RC4_128_MD5 Cipher Suite selection that Amazon made with its ServerHello message. It's a reasonable choice that slightly favors speed over security.
As we saw, if someone could secretly factor Amazon's "n" modulus into its respective "p" and "q", they could effectively decrypt all "secure" traffic until Amazon changes their certificate. Amazon counter-balances this concern this with a short one year duration certificate:

One of the cipher suites that was offered was "TLS_DHE_RSA_WITH_AES_256_CBC_SHA" which uses the Diffie-Hellman key exchange that has a nice property of "forward secrecy." This means that if someone cracked the mathematics of the key exchange, they'd be no better off to decrypt another session. One downside to this algorithm is that it requires more math with big numbers, and thus is a little more computationally taxing on a busy server. The "Advanced Encryption Standard" (AES) algorithm was present in many of the suites that we offered. It's different than RC4 in that it works on 16 byte "blocks" at a time rather than a single byte. Since its key can be up to 256 bits, many consider this to be more secure than RC4.
In just 220 milliseconds, two endpoints on the Internet came together, provided enough credentials to trust each other, set up encryption algorithms, and started to send encrypted traffic.
And to think, all of this just so Bob can buy milk.

Application-wide action filters in ASP.NET MVC

(http://blog.maartenballiauw.be/post/2009/06/24/Application-wide-action-filters-in-ASPNET-MVC.aspx)

Ever had a team of developers using your ASP.NET MVC framework? Chances are you have implemented some action filters (i.e. for logging) which should be applied on all controllers in the application. Two ways to do this: kindly ask your developers to add a [Logging] attribute to the controllers they write, or kindly ask to inherit from SomeCustomControllerWithActionsInPlace.
If you have been in this situation, monday mornings, afternoons, tuesdays and other weekdays are in fact days where some developers will forget to do one of the above. This means no logging! Or any other action filters that are executed due to a developer that has not been fed with enough coffee… Wouldn’t it be nice to have a central repository where you can register application-wide action filters? That’s exactly what we are going to do in this blog post.
Note: you can in fact use a dependency injection strategy for this as well, see Jeremy Skinner’s blog.
Download the example code: MvcGlobalActionFilter.zip (24.38 kb)

The idea

Well, all things have to start with an idea, otherwise there’s nothing much left to do. What we’ll be doing in our solution to global action filters is the following:

Create a IGlobalFilter interface which global action filters have to implement. You can discuss about this, but I think it’s darn handy to add some convenience methods like ShouldBeInvoked() where you can abstract away some checks before the filter is actually invoked.
Create some IGlobalActionFilter, IGlobalResultFilter, IGlobalAuthorizationFilter, IGlobalExceptionFilter interfaces, just for convenience to the developer that is creating the global filters. You’ll see the use of this later on.
Create a GlobalFilterActionInvoker, a piece of logic that is set on each controller so the controller knows how to call its own action methods. We’ll use this one to inject our lists of global filters.
Create a GlobalFilterControllerFactory. I’m not happy with this, but we need it to set the GlobalFilterActionInvoker instance on each controller when it is created.

IGlobalFilter, IGlobalActionFilter, …

Not going to spend too much time on these. Actually, these interfaces are just descriptors for our implementation so it knows what type of filter is specified and if it should be invoked. Here’s a bunch of code. No comments.

public interface IGlobalFilter 
{ 
    bool ShouldBeInvoked(ControllerContext controllerContext); 
}


public interface IGlobalAuthorizationFilter : IGlobalFilter, IAuthorizationFilter
public interface IGlobalActionFilter : IGlobalFilter, IActionFilter { }
public interface IGlobalResultFilter : IGlobalFilter, IResultFilter { }
public interface IGlobalExceptionFilter : IGlobalFilter, IExceptionFilter { }

And yes, I did suppress some Static Code Analysis rules for this :-)

GlobalFilterActionInvoker

The GlobalFilterActionInvoker will take care of registering the global filters and making sure each filter is actually invoked on every controller and action method in our ASP.NET MVC application. Here’s a start for our class:

public class GlobalFilterActionInvoker : ControllerActionInvoker 
{ 
    protected FilterInfo globalFilters;
    public GlobalFilterActionInvoker() 
    { 
        globalFilters = new FilterInfo(); 
    }
    public GlobalFilterActionInvoker(FilterInfo filters) 
    { 
        globalFilters = filters; 
    }
    public GlobalFilterActionInvoker(List filters) 
        : this(new FilterInfo()) 
    { 
        foreach (var filter in filters) 
            RegisterGlobalFilter(filter); 
    }
    public FilterInfo Filters 
    { 
        get { return globalFilters; } 
    } 

    // - more code -
}

We’re providing some utility constructors that take a list of global filters and add it to the internal FilterInfo instance (which is an ASP.NET MVC class we can leverage in here). RegisterGlobalFilter() will do the magic of adding filters to the right collection in the FilterInfo instance.

public void RegisterGlobalFilter(IGlobalFilter filter) 
{ 
    if (filter is IGlobalAuthorizationFilter) 
        globalFilters.AuthorizationFilters.Add((IGlobalAuthorizationFilter)filter);
    if (filter is IGlobalActionFilter) 
        globalFilters.ActionFilters.Add((IGlobalActionFilter)filter);
    if (filter is IGlobalResultFilter) 
        globalFilters.ResultFilters.Add((IGlobalResultFilter)filter);
    if (filter is IGlobalExceptionFilter) 
        globalFilters.ExceptionFilters.Add((IGlobalExceptionFilter)filter); 
}

One override left in our implementation: ControllerActionInvoker, the class we are inheriting from, provides a method named GetFilters(), which is used to get the filters for a specific controller context. Ideal one to override:

protected override FilterInfo GetFilters(ControllerContext controllerContext, ActionDescriptor actionDescriptor) 
{ 
    FilterInfo definedFilters = base.GetFilters(controllerContext, actionDescriptor);
    foreach (var filter in Filters.AuthorizationFilters) 
    { 
        IGlobalFilter globalFilter = filter as IGlobalFilter; 
        if (globalFilter == null || 
            (globalFilter != null && globalFilter.ShouldBeInvoked(controllerContext))) 
        { 
            definedFilters.AuthorizationFilters.Add(filter); 
        } 
    }
    // - same for action filters -
    // - same for result filters -
    // - same for exception filters -
    return definedFilters; 
}

Basically, we are querying our IGlobalFilter if it should be invoked for the given controller context. If so, we add it to the FilterInfo object that is required by the ControllerActionInvoker base class. Piece of cake!

GlobalFilterControllerFactory

I’m not happy having to create this one, but we need it to set the GlobalFilterActionInvoker instance on each controller that is created. Otherwise, there is no way to specify our global filters on a controller or action method… Here’s the class:

public class GlobalFilterControllerFactory : DefaultControllerFactory 
{ 
    protected GlobalFilterActionInvoker actionInvoker;
    public GlobalFilterControllerFactory(GlobalFilterActionInvoker invoker) 
    { 
        actionInvoker = invoker; 
    }
    public override IController CreateController(System.Web.Routing.RequestContext requestContext, string controllerName) 
    { 
        IController controller = base.CreateController(requestContext, controllerName); 
        Controller controllerInstance = controller as Controller; 
        if (controllerInstance != null) 
        { 
            controllerInstance.ActionInvoker = actionInvoker; 
        } 
        return controller; 
    } 
}

What we do here is let the DefaultControllerFactory create a controller. Next, we simply set the controller’s ActionInvoker property to our GlobalFilterActionInvoker .

Plumbing it all together!

To plumb things together, add some code in your Global.asax.cs class, under Application_Start:

protected void Application_Start() 
{ 
    RegisterRoutes(RouteTable.Routes);
    ControllerBuilder.Current.SetControllerFactory( 
        new GlobalFilterControllerFactory( 
            new GlobalFilterActionInvoker( 
                new List 
                { 
                    new SampleGlobalTitleFilter() 
                } 
            ) 
        ) 
    ); 
}

We are now setting the controller factory for our application to GlobalFilterControllerFactory, handing it a GlobalFilterActionInvoker which specifies one global action filter: SampleGlobalTitleFilter.

Sidenote: SampleGlobalTitleFilter

As a sidenote, I created a sample result filter named SampleGlobalTitleFilter, which is defined as a global filter that always appends a string (“ – Sample Application”) to the page title. Here’s the code for that one:

public class SampleGlobalTitleFilter : IGlobalResultFilter 
{ 
    public bool ShouldBeInvoked(System.Web.Mvc.ControllerContext controllerContext) 
    { 
        return true; 
    }
    public void OnResultExecuted(System.Web.Mvc.ResultExecutedContext filterContext) 
    { 
        return; 
    }
    public void OnResultExecuting(System.Web.Mvc.ResultExecutingContext filterContext) 
    { 
        if (filterContext.Controller.ViewData["PageTitle"] == null) 
            filterContext.Controller.ViewData["PageTitle"] = "";
        string pageTitle = filterContext.Controller.ViewData["PageTitle"].ToString();
        if (!string.IsNullOrEmpty(pageTitle)) 
            pageTitle += " - ";
        pageTitle += "Sample Application";
        filterContext.Controller.ViewData["PageTitle"] = pageTitle; 
    } 
}

Conclusion

Download the sample code: MvcGlobalActionFilter.zip (24.38 kb)
There is no need for my developers to specify SampleGlobalTitleFilter on each controller they write. There is no need for my developers to use the ControllerWithTitleFilter base class. People can come in and even develop software without drinking 2 liters of coffee! Really, development should not be hard for your developers. Make sure all application-wide infrastructure is there and our people are ready to go. And I’m really loving ASP.NET MVC’s extensibility on that part!

Security transparency in CLR 2.0 (.NET 2.0, .NET3.0, .NET 3.5) and in CLR 4.0 (.NET 4.0)

David DeWinter has an excellent series of post on different topics security in both CLR 2.0 and CLR 4.0

Transparency Can Reduce Your Security Footprint – #1
Using Transparency in CLR 2.0 – #2
Transparent Code Behavior in CLR 2.0 – #3
What’s New With Security in .NET 4? – #4
Opting Out of Security Changes in .NET 4 – #5
The AllowPartiallyTrustedCallersAttribute (APTCA) – #6

How to Host a Partial Trust Sandbox – #7
Conditional APTCA in .NET 4 – #8
Hosting Conditional APTCA Assemblies – #9
An Introduction to Security Transparency in .NET 4 – #10
Type Transparency in .NET 4 – #11
Partial Trust, APTCA, and Security Transparency – #12
Transparency and Implicit Static Constructors – #13
Determining the Security Rules for Your Assemblies – #14
The SecurityRulesAttribute – #15
Mixing Level 1 and Level 2 Transparency Rules – #16
How to Build APIs with Transparency in Mind – #17
Asserting for Permissions in .NET 4 – #18

A nice and informative JQuery tutorial

Learning resources for Entity Framework 4.0 new features

(http://bogdanbrinzarea.wordpress.com/2009/08/05/learning-resources-for-entity-framework-4-0-new-features/)

Jan 30, 2013

Jan 16, 2013

Jan 15, 2013

Using SecureString

The First Few Milliseconds of an HTTPS Connection

Client Hello

Server Hello

Checking out the Certificate

Interlude: A Short, Not Too Scary, Guide to RSA

Verifying Signatures

Pre-Master Secret

Trading Secrets

Deriving the Master Secret

Generating Lots of Keys

Prepare to be Encrypted!

RC4 Encryption

Welcome to the Application Layer!

... And We're Done!

The idea

IGlobalFilter, IGlobalActionFilter, …

GlobalFilterActionInvoker

GlobalFilterControllerFactory

Plumbing it all together!

Sidenote: SampleGlobalTitleFilter

Conclusion

New features

SQL improvements

POCO Objects

Deferred Loading and Explicit Loading

Functions in LINQ to Entities

Model Defined Functions

EDM Designer

Model-First Support

Code Only Support

Complex Type Support

Table splitting

Testability

Design patterns

Events

Architecture

Self tracking entities

POCO T4 templates

Validation

EDM designer

Foreign keys

Pluralization

Tips