The current version of the MS-PST open specification document can be found here: http://msdn.microsoft.com/en-us/library/ff385210(office.12).aspx
The PST is a structured binary file format that contains messages, folders, and attachments. The file structure can be logically separated into 3 layers: the Node Database (NBD) layer, the Lists, Tables, and Properties (LTP) layer, and the Messaging layer. Each layer builds on the one before and because of this it is necessary to completely understand the NDB layer before the LTP layer, and to understand the LTP layer before the Messaging layer. Like building a house, you cannot start framing the walls until you pour the foundation. The Node BTree, which is an important part of the NBD layer is part of that foundation. Without it, you could not hope to be able to understand how the PST file format works much less be able to find what you are looking for.
We will be focusing on the Unicode version of the PST file format that was introduced in Outlook 2003 since most of the PST files you are likely to be working with will be in this format. Information on the differences between the Unicode and ANSI version can be found in section 188.8.131.52.
Structures that you should be familiar with
HEADER: The structure at the beginning of the file that contains meta-data about the PST file.
ROOT: A structure in the HEADER that contains file state information.
BREF: A structure that contains a BID/IB pair.
BID: Block ID, a unique identifier.
NID: Node ID, a unique identifier
IB: Byte Index, an unsigned 64-bit value that represents an absolute file position.
BTree: Generic binary tree structure.
NBT: The Node BTree.
Page: A generic 512 byte section of the file.
PageTrailer: A 16 byte structure at the end of each Page that contains meta-data about the page.
BTPage: A 512 byte page that is part of the Node or Block BTree.
BTEntry: Generic BTree entry.
NBTEntry: Leaf node entries in the Node BTree.
The ROOT structure in the HEADER contains the BREFNBT structure. This is a BREF structure, which contains a BID structure and an IB. The IB is the second 64 bits that represents an unsigned 64-bit integer and the absolute offset into the file where the first NBT page exists. In a Unicode PST file, we can find the IB at offset 0xE0.
Figure 1: How to find the location of the NBT Root page.
Remember that the value is stored in Little Endian format on disk, so what you see in the file laid out as "00 90 00 00 00 00 00 00" becomes "00 00 00 00 00 00 90 00" when read into memory. Using this we can see that the first NBT page is located at offset 0x9000.
The NBT Root Page
The following is the root NBT page for a new PST file. Yours might look slightly different.
Figure 2: The NBT Root page.
The first 488 bytes is an array of rgentries. By referring to the table in section 184.108.40.206.1 we can verify that the pType property 0x81 is pTypeNBT. We can also see that the BID in the PageTrailer matches the BID from the BREFNBT structure in the ROOT. It's good to verify that the data we are looking at is the data structure we expect it to be.
Looking at the section right before the PageTrailer, there are 3 important properties that will determine how you read the data in the rgentries array...
Figure 3: How to decipher the rgentries block.
The number of entries in the rgentries array is discovered by reading the cEnt byte. The cbEnt byte tells us the size of each entry in bytes. The entries will be of type BTENTRY if the cLevel byte is greater than 0 (see section 220.127.116.11.7.1). So, in Figure 3 we can see that the rgentries array contains 3 entries, each one is 0x18 (24) bytes in length, and they are going to be BTENTRY structures.
So, lets take a look at one of these entries.
Figure 4: A sample BTENTRY.
The first 8 bytes are the btKey and the 16 bytes that follow is the BREF structure. The important part of the BREF is the IB which is the last 8 bytes which tells us where the next page is in the file. So, this entry has a key of 0x21 (33), the Block ID is 0x130 and the Page that it points to is at offset 0x8800.
After reading the contents of the 3 records in the rgentries array, our Node BTree looks like this:
Figure 5: The Node BTree structure up to this point.
Traversing the Node BTree
Lets look at the first NBTPage that the Root page points to at offset 0x8800
Figure 6: A sample NBT page.
This one has the familiar page type of 0x81 (pTypeNBT) and we can verify that the BID matches (0x130). Other than that it looks a lot different. Following the same logic as before we can see that this page contains 0x0E (14) entries and that each one is 0x20 (32) bytes in length. cLevel is 0 this time which means the entries are NBTENTRY types (see section 18.104.22.168.7.1). This tells us that we are at the leaf node of the tree. If cLevel was greater than 0 we would follow the same logic as we did with the entries on the NBTRoot page.
Lets take a look at the first NBTENTRY…
Figure 7: A sample NBTENTRY.
The first 8 bytes is the NID which is 0x21. The next 8 bytes is the BID of the data block for this node (0xAC). That tells us the node ID of the block in the Block Btree where the actual data is for this node. The next 8 bytes is the BID of the subnode block for this node. In this case, there isn't one. The next 4 bytes is the parent NID. The parent NID contains the ID of the folder object that the data that belongs to this node belongs to. In other words, if the data that this node points to represents is an e-mail, the parent NID is the ID of the folder that the e-mail resides in. The last 4 bytes is for alignment purposes and should be ignored.
Following this logic you should have no problem building the complete Node BTree.