Apache POI™ - HPSF Internals

Introduction

A Microsoft Office document is internally organized like a filesystem with directory and files. Microsoft calls these files streams. A document can have properties attached to it, like author, title, number of words etc. These metadata are not stored in the main stream of, say, a Word document, but instead in a dedicated stream with a special format. Usually this stream's name is \005SummaryInformation, where \005 represents the character with a decimal value of 5.

A single piece of information in the stream is called a property, for example the document title. Each property has an integral ID (e.g. 2 for title), a type (telling that the title is a string of bytes) and a value (what this is should be obvious). A stream containing properties is called a property set stream.

This document describes the internal structure of a property set stream, i.e. the HPSF. It does not describe how a Microsoft Office document is organized internally and how to retrieve a stream from it. See the POIFS documentation for that kind of stuff.

The HPSF is not only used in the Summary Information stream in the top-level document of a Microsoft Office document. Often there is also a property set stream named \005DocumentSummaryInformation with additional properties. Embedded documents may have their own property set streams. You cannot tell by a stream's name whether it is a property set stream or not. Instead you have to open the stream and look at its bytes.

Data Types

Before delving into the details of the property set stream format we have to have a short look at data types. Integral values are stored in the so-called little endian format. In this format the bytes that make out an integral value are stored in the "wrong" order. For example, the decimal value 4660 is 0x1234 in the hexadecimal notation. If you think this should be represented by a byte 0x12 followed by another byte 0x34, you are right. This is called the big endian format. In the little endian format, however, this order is reversed and the low-value byte comes first: 0x3412.

The following table gives an overview about some important data types:

HPSF Overview

A property set stream consists of three main parts:

1. The header and
2. the section(s) containing the properties.

The Header

The first bytes in a property set stream is the header. It has a fixed length and looks like this:

Section List

Following the header is the section list. This is an array of pairs each consisting of a section format ID and an offset. This array has as many pairs of ClassID and and DWord fields as the section count field in the header says. The Summary Information stream contains a single section, the Document Summary Information stream contains two.

Section

A section is divided into three parts: the section header (with the section length and the number of properties in the section), the properties list (with type and offset of each property), and the properties themselves. Here are the details:

Property IDs

As seen above, a section holds a property list: an array with property IDs and offsets. The property ID gives each property a meaning. For example, in the Summary Information stream the property ID 2 says that this property is the document's title.

If you want to know a property ID's meaning, it is not sufficient to know the ID itself. You must also know the section format ID. For example, in the Document Summary Information stream the property ID 2 means not the document's title but its category. Due to Microsoft's infinite wisdom the section format ID is not part of the section. Thus if you have only a section without the stream it is in, you cannot make any sense of the properties because you do not know what they mean.

So each section format ID has its own name space of property IDs. Microsoft defined some "well-known" property IDs for the Summary Information and the Document Summary Information streams. You can extend them by your own additional IDs. This will be described below.

Property IDs in The Summary Information Stream

The Summary Information stream has a single section with a section format ID of 0xF29F85E04FF91068AB9108002B27B3D9. The following table defines the meaning of its property IDs. Each row associates a property ID with a name and an ID string. (The property type is just for informational purposes given here. As we have seen above, the type is always given along with the value.)

The property name is a readable string which could be displayed to the user. However, this string is useful only for users who understand English. The property name does not help with other languages.

The property ID string is about the same but looks more technically and is nothing a user should bother with. You could the ID string and map it to an appropriate display string in a particular language. Of course you could do that with the property ID as well and with less overhead, but people (including software developers) tend to be better in remembering symbolic constants than remembering numbers.

Property IDs in The Document Summary Information Stream

The Document Summary Information stream has two sections with a section format ID of 0xD5CDD5022E9C101B939708002B2CF9AE for the first one. The following table defines the meaning of the property IDs in the first section. See the preceding section for interpreting the table.

Property Types

A property consists of a DWord type field followed by the property value. The property type is an integer value and tells how the data byte following it are to be interpreted. In the Microsoft world it is also known as the variant.

The Usage column says where a variant type may occur. Not all of them are allowed in a property set but just those marked with a [P]. [V] - may appear in a VARIANT, [T] - may appear in a TYPEDESC, [P] - may appear in an OLE property set, [S] - may appear in a Safe Array.

The Dictionary

What a dictionary is good for is explained in the HPSF HOW-TO. This chapter explains how it is organized internally.

The dictionary has a simple header consisting of a single UInt value. It tells how many entries the dictionary comprises:

The dictionary entries follow the header. Each one looks like this:

The entries are not aligned, i.e. each one follows its predecessor without any gap or fill characters.

References

In order to assemble the HPSF description I used information publically available on the Internet only. The references given below have been very helpful. If you have any amendments or corrections, please let us know! Thank you!

1. In Understanding OLE documents, Ken Kyler gives an introduction to OLE2 documents and especially to property sets. He names the property names, types, and IDs of the Summary Information and Document Summary Information stream.
2. The ActiveX Programmer's Reference at https://www.dwam.net/docs/oleref/ seems a little outdated, but that's what I have found.
3. An overview of the VT_ types is in Variant Type Definitions.
4. What is a FILETIME? The answer can be found under , https://www.vbapi.com/ref/f/filetime.html or https://www.cs.rpi.edu/courses/fall01/os/FILETIME.html. In short: The FILETIME structure holds a date and time associated with a file. The structure identifies a 64-bit integer specifying the number of 100-nanosecond intervals which have passed since January 1, 1601. This 64-bit value is split into the two dwords stored in the structure.
5. Microsoft provides some public information in the MSDN Library. Use the search function to try to find what you are looking for, e.g. "codepage" or "document summary information" etc.