Home > Articles > Software Development & Management

  • Print
  • + Share This
From the author of

6.4.4 The Windows 98 File System

The original release of Windows 95 used the MS-DOS file system, including the use of 8 + 3 character file names and the FAT-12 and FAT-16 file systems. Starting with the second release of Windows 95, file names longer than 8 + 3 characters were permitted. In addition, FAT-32 was introduced, primarily to allow larger disk partitions larger than 2 GB and disks larger than 8 GB, which were then available. Both the long file names and FAT-32 were used in Windows 98 in the same form as in the second release of Windows 95. Below we will describe these features of the Windows 98 file system, which have been carried forward into Windows Me as well.

Since long file names are more exciting for users than the FAT structure, let us look at them first. One way to introduce long file names would have been to just invent a new directory structure. The problem with this approach is that if Microsoft had done this, people who were still in the process of converting from Windows 3 to Windows 95 or Windows 98 could not have accessed their files from both systems. A political decision was made within Microsoft that files created using Windows 98 must be accessible from Windows 3 as well (for dual-boot machines). This constraint forced a design for handling long file names that was backward compatible with the old MS-DOS 8 + 3 naming system. Since such backward compatibility constraints are not unusual in the computer industry, it is worth looking in detail at how Microsoft accomplished this goal.

The effect of this decision to be backward compatible meant that the Windows 98 directory structure had to be compatible with the MS-DOS directory structure. As we saw, this structure is just a list of 32-byte entries as shown in Fig. 6-4. This format came directly from CP/M (which was written for the 8080), which goes to show how long (obsolete) structures can live in the computer world.

However, it was possible to now allocate the 10 unused bytes in the entries of Fig. 6-4, and that was done, as shown in Fig. 6-6. This change has nothing to do with long names, but it is used in Windows 98, so it is worth understanding.

Figure 6-6 The extended MOS-DOS directory entry used in Windows 98.

The changes consist of the addition of five new fields where the 10 unused bytes used to be. The NT field is mostly there for some compatibility with Windows NT in terms of displaying file names in the correct case (in MS-DOS, all file names are upper case). The Sec field solves the problem that it is not possible to store the time of day in a 16-bit field. It provides additional bits so that the new Creation time field is accurate to 10 msec. Another new field is Last access, which stores the date (but not time) of the last access to the file. Finally, going to the FAT-32 file system means that block numbers are now 32 bits, so an additional 16-bit field is needed to store the upper 16 bits of the starting block number.

Now we come to the heart of the Windows 98 file system: how long file names are represented in a way that is backward compatible with MS-DOS. The solution chosen was to assign two names to each file: a (potentially) long file name (in Unicode, for compatibility with Windows NT), and an 8 + 3 name for compatibility with MS-DOS. Files can be accessed by either name. When a file is created whose name does not obey the MS-DOS naming rules (8 + 3 length, no Unicode, limited character set, no spaces, etc.), Windows 98 invents an MS-DOS name for it according to a certain algorithm. The basic idea is to take the first six characters of the name, convert them to upper case, if need be, and append ~1 to form the base name. If this name already exists, then the suffix ~2 is used, and so on. In addition, spaces and extra periods are deleted and certain special characters are converted to underscores. As an example, a file named The time has come the walrus said is assigned the MS-DOS name THETIM~1. If a subsequent file is created with the name The time has come the rabbit said, it is assigned the MS-DOS name THETIM~2, and so on.

Every file has an MS-DOS file name stored using the directory format of Fig. 6-6. If a file also has a long name, that name is stored in one or more directory entries directly preceding the MS-DOS file name. Each long-name entry holds up to 13 (Unicode) characters. The entries are stored in reverse order, with the start of the file name just ahead of the MS-DOS entry and subsequent pieces before it. The format of each long-name entry is given in Fig. 6-7.

Figure 6-7 An entry for (part of) a long file name in Windows 98.

An obvious question is: ''How does Windows 98 know whether a directory entry contains an MS-DOS file name or a (piece of a) long file name?'' The answer lies in the Attributes field. For a long-name entry, this field has the value 0x0F, which represents an otherwise impossible combination of attributes. Old MS-DOS programs that read directories will just ignore it as invalid. Little do they know. The pieces of the name are sequenced using the first byte of the entry. The last part of the long name (the first entry in the sequence) is marked by adding 64 to the sequence number. Since only 6 bits are used for the sequence number, the theoretical maximum for file names is 63 ´ 13 or 819 characters. In fact they are limited to 260 characters for historical reasons.

Each long-name entry contains a Checksum field to avoid the following problem. First, a Windows 98 program creates a file with a long name. Second, the computer is rebooted to run MS-DOS or Windows 3. Third, an old program there then removes the MS-DOS file name from the directory but does not remove the long file name preceding it (because it does not know about it). Finally, some program creates a new file that reuses the newly-freed directory entry. At this point we have a valid sequence of long-name entries preceeding an MS-DOS file entry that has nothing to do with that file. The Checksum field allows Windows 98 to detect this situation by verifying that the MS-DOS file name following a long name does, in fact, belong to it. Of course, with only 1 byte being used, there is one chance in 256 that Windows 98 will not notice the file substitution.

To see an example of how long names work, consider the example of Fig. 6-8. Here we have a file called The quick brown fox jumps over the lazy dog. At 42-characters, it certainly qualifies as a long file name. The MS-DOS name constructed from it is THEQUI~1 and is stored in the last entry.

Figure 6-8 An example of how a long name is stored in Windows 98.

Some redundancy is built into the directory structure to help detect problems in the event that an old Windows 3 program has made a mess of the directory. The sequence number byte at the start of each entry is not strictly needed since the 0x40 bit marks the first one, but it provides additional redundancy, for example. Also, the Low field of Fig. 6-8 (the lower half of the starting cluster) is 0 in all entries but the last one, again to avoid having old programs misinterpret it and ruin the file system. The NT byte in Fig. 6-8 is used in NT and ignored in Windows 98. The A byte contains the attributes.

The implementation of the FAT-32 file system is conceptually similar to the implementation of the FAT-16 file system. However, instead of an array of 65,536 entries, there are as many entries as needed to cover the part of the disk with data on it. If the first million blocks are used, the table conceptually has 1 million entries. To avoid having all of them in memory at once, Windows 98 maintains a window into the table and keeps only in parts of it in memory at once.

  • + Share This
  • 🔖 Save To Your Account