• Document revision: 1.11
• Last updated: Nov 7, 2008
• Compiler/Editor: Peter Schepers
• Contributors/sources: Immers/Neufeld: “Inside Commodore DOS”, Wolfgang Moser
• Wiki rendition: Eek/Retrograde

Note: This document will try to explain the layout of the 1541 disks, as well as some of the files they contain. However, I do not explain GEOS or REL files here. See the file GEOS.TXT or REL.TXT for more information on those file types or disk structure.

Special thanks to Wolfgang Moser for his work on identifying the differences between the various 'Speeder' DOS's that exist. This information is included in a table later in this document.

In this wiki rendition, formatting - especially around tables - has been reworked to make the information more easy to consume.

First and foremost we have D64, which is basically a sector-for-sector copy of a 1540/1541 disk. There are several versions of these which I will cover shortly. The standard D64 is a 174848 byte file comprised of 256 byte sectors arranged in 35 tracks with a varying number of sectors per track for a total of 683 sectors. Track counting starts at 1, not 0, and goes up to 35. Sector counting starts at 0, not 1, for the first sector, therefore a track with 21 sectors will go from 0 to 20.

The original media (a 5.25“ disk) has the tracks laid out in circles, with track 1 on the very outside of the disk (closest to the sides) to track 35 being on the inside of the disk (closest to the inner hub ring). Commodore, in their infinite wisdom, varied the number of sectors per track and data densities across the disk to optimize available storage, resulting in the chart below. It shows the sectors/track for a standard D64. Since the outside diameter of a circle is the largest (versus closer to the center), the outside tracks have the largest amount of storage.

683 sectors total (for a 35 track image)

(*) Tracks 36-40 apply to 40-track images only

The directory track should be contained totally on track 18. Sectors 1-18 contain the entries and sector 0 contains the BAM (Block Availability Map) and disk name/ID. Since the directory is only 18 sectors large (19 less one for the BAM), and each sector can contain only 8 entries (32 bytes per entry), the maximum number of directory entries is 18 * 8 = 144. The first directory sector is always 18/1, even though the t/s pointer at 18/0 (first two bytes) might point somewhere else. It then follows the same chain structure as a normal file, using a sector interleave of 3. This makes the chain links go 18/1, 18/4, 18/7 etc.

Note that you can extend the directory off of track 18, but only when reading the disk or image. Attempting to write to a directory sector not on track 18 will cause directory corruption. See the section below called Non-Standard & Long Directories.

Each directory sector has the following layout (18/1 partial dump):

The first two bytes of the sector ($12/$04) indicate the location of the next track/sector of the directory (18/4). If the track is set to $00, then it is the last sector of the directory. It is possible, however unlikely, that the directory may not be competely on track 18 (some disks do exist like this). Just follow the chain anyhow.

When the directory is done, the track value will be $00. The sector link should contain a value of $FF, meaning the whole sector is allocated, but the actual value doesn't matter. The drive will return all the available entries anyways. This is a breakdown of a standard directory sector and entry:

This is a breakdown of the first directory entry:

This is a breakdown of the file type byte:

• File type
  • Typical values for this location are:
    • $00: Scratched (deleted file entry)
    • $80: DEL
    • $81: SEQ
    • $82: PRG
    • $83: USR
    • $84: REL
  • Bit 0-3: The actual filetype
    • %000 (0): DEL
    • %001 (1): SEQ
    • %010 (2): PRG
    • %011 (3): USR
    • %100 (4): REL
    • Values 5-15 are illegal, but if used will produce very strange results. The 1541 is inconsistent in how it treats these bits. Some routines use all 4 bits, others ignore bit 3, resulting in values from 0-7.
  • Bit 4: Not used
  • Bit 5: Used only during SAVE-@ replacement
  • Bit 6: Locked flag (Set produces ”>“ locked files)
  • Bit 7: Closed flag (Not set produces “*”, or “splat” files)

Files, on a standard 1541, are stored using an interleave of 10. Assuming a starting track/sector of 17/0, the chain would run 17/0, 17/10, 17/20, 17/8, 17/18, etc.

Note: No GEOS entries are listed in the above description. See the GEOS.TXT file for GEOS info.

Most Commdore floppy disk drives use a single dedicated directory track where all filenames are stored. This limits the number of files stored on a disk based on the number of sectors on the directory track. There are some disk images that contain more files than would normally be allowed. This requires extending the directory off the default directory track by changing the last directory sector pointer to a new track, allocating the new sectors in the BAM, and manually placing (or moving existing) file entries there. The directory of an extended disk can be read and the files that reside there can be loaded without problems on a real drive. However, this is still a very dangerous practice as writing to the extended portion of the directory will cause directory corruption in the non-extended part. Many of the floppy drives core ROM routines ignore the track value that the directory is on and assume the default directory track for operations.

To explain: assume that the directory has been extended from track 18 to track 19/6 and that the directory is full except for a few slots on 19/6. When saving a new file, the drive DOS will find an empty file slot at 19/6 offset $40 and correctly write the filename and a few other things into this slot. When the file is done being saved the final file information will be written to 18/6 offset $40 instead of 19/6 causing some directory corruption to the entry at 18/6. Also, the BAM entries for the sectors occupied by the new file will not be saved and the new file will be left as a SPLAT (*) file.

Attempts to validate the disk will result in those files residing off the directory track to not be allocated in the BAM, and could also send the drive into an endless loop. The default directory track is assumed for all sector reads when validating so if the directory goes to 19/6, then the validate code will read 18/6 instead. If 18/6 is part of the normal directory chain then the validate routine will loop endlessly.

The layout of the BAM area (sector 18/0) is a bit more complicated…

Note: The BAM entries for SPEED, DOLPHIN and ProLogic DOS use the same layout as standard BAM entries.

One of the interesting things from the BAM sector is the byte at offset $02, the DOS version byte. If it is set to anything other than $41 or $00, then we have what is called “soft write protection”. Any attempt to write to the disk will return the “DOS Version” error code 73 ,“CBM DOS V 2.6 1541”. The 1541 is simply telling you that it thinks the disk format version is incorrect. This message will normally come up when you first turn on the 1541 and read the error channel. If you write a $00 or a $41 into 1541 memory location $00FF (for device 0), then you can circumvent this type of write-protection, and change the DOS version back to what it should be.

The BAM entries require a bit (no pun intended) more of a breakdown. Take the first entry at bytes $04-$07 ($12 $FF $F9 $17). The first byte ($12) is the number of free sectors on that track. Since we are looking at the track 1 entry, this means it has 18 (decimal) free sectors. The next three bytes represent the bitmap of which sectors are used/free. Since it is 3 bytes (8 bits/byte) we have 24 bits of storage. Remember that at most, each track only has 21 sectors, so there are a few unused bits.

These entries must be viewed in binary to make any sense. We will use the first entry (track 1) at bytes 04-07:

$FF=%11111111, $F9=%11111001, $17=%00010111

In order to make any sense from the binary notation, flip the bits around.

               111111 11112222
    01234567 89012345 67890123
    --------------------------
    11111111 10011111 11101000
    ^                     ^
sector 0              sector 20

Since we are on the first track, we have 21 sectors, and only use up to the bit 20 position. If a bit is on (1), the sector is free. Therefore, track 1 has sectors 9,10 and 19 used, all the rest are free. Any leftover bits that refer to sectors that don't exist, like bits 21-23 in the above example, are set to allocated.

Each filetype has its own unique properties, but most follow one simple structure. The first file sector is pointed to by the directory and follows a t/s chain, until the track value reaches $00. When this happens, the value in the sector link location indicates how much of the sector is used. For example, the following chain indicates a file 6 sectors long, and ends when we encounter the $00/$34 chain. At this point the last sector occupies from bytes $02-$34.

These are some variations of the D64 layout:

1. Standard 35 track layout but with 683 error bytes added on to the end of the file. Each byte of the error info corresponds to a single sector stored in the D64, indicating if the sector on the original disk contained an error. The first byte is for track 1/0, and the last byte is for track 35/16.
2. A 40 track layout, following the same layout as a 35 track disk, but with 5 extra tracks. These contain 17 sectors each, like tracks 31-35. Some of the PC utilities do allow you to create and work with these files. This can also have error bytes attached like variant #1.
3. The Commodore 128 allowed for “auto-boot” disks. With this, t/s 1/0 holds a specific byte sequence which the computer recognizes as boot code. See the document C128BOOT.TXT for more info.

Below is a small chart detailing the standard file sizes of D64 images, 35 or 40 tracks, with or without error bytes.

The following table (provided by Wolfgang Moser) outlines the differences between the standard 1541 DOS and the various “speeder” DOS's that exist. The 'header 7/8' category is the 'fill bytes' as the end of the sector header of a real 1541 disk See the document G64.TXT for a better explanation of these bytes.

(*) There are also clones of SpeedDOS that exist, such as RoloDOS and DigiDOS. Both are just a change of the DOS startup string.

The location of the extra BAM information in sector 18/0, for 40 track images, will be different depending on what standard the disks have been formatted with. SPEED DOS stores them from $C0 to $D3, DOLPHIN DOS stores them from $AC to $BF and PrologicDOS stored them right after the existing BAM entries from $90-A3. PrologicDOS also moves the disk label and ID forward from the standard location of $90 to $A4. 64COPY and Star Commander let you select from several different types of extended disk formats you want to create/work with.

All three of the speeder DOS's mentioned above don't alter the standard sector interleave of 10 for files and 3 for directories. The reason is that they use a memory cache installed in the drive which reads the entire track in one pass. This alleviates the need for custom interleave values. They do seem to alter the algorithm that finds the next available free sector so that the interleave value can deviate from 10 under certain circumstances, but I don't know why they would bother.

Below is a HEX dump of a Speed DOS BAM sector. Note the location of the extra BAM info from $C0-D3.

Below is a HEX dump of a Dolphin DOS BAM sector. Note the location of the extra BAM info from $AC-BF.

Below is a HEX dump of a PrologicDOS BAM sector. Note that the disk name and ID are now located at $A4 instead of starting at $90.

Here is the meaning of the error bytes added onto the end of any extended D64. The CODE is the same as that generated by the 1541 drive controller… it reports these numbers, not the error code we usually see when an error occurs.

Some of what comes below is taken from Immers/Neufeld book “Inside Commodore DOS”. Note the descriptions are not completely accurate as to what the drive DOS is actually doing to seek/read/decode/write sectors, but serve as simple examples only. The “type” field is where the error usually occurs, whether it's searching for any SYNC mark, any header ID, any valid header, or reading a sector.

These first errors are “seek” errors, where the disk controller is simply reading headers and looking at descriptor bytes, checksums, format ID's and reporting what errors it sees. These errors do not necessarily apply to the exact sector being looked for. This fact makes duplication of these errors very unreliable.

Note that error 20 occurs twice during this phase. The first time is when a header ID is being searched for, the second is when the proper header pattern for the sector being searched for is not found.

From this point on, all the errors apply to the specific sector you are looking for. If a read passed all the previous checks, then we are at the sector being searched for.

Note that the entire sector is read before these errors are detected. Therefore the data, checksum and off bytes are available.

These errors only apply when writing to a disk. I don't see the usefulness of having these as they cannot be present when only reading a disk.

This is not an error at all, but when gets reported when the read of a sector is ok.

The advantage with using the 35 track D64 format, regardless of error bytes, is that it can be converted directly back to a 1541 disk by either using the proper cable and software on the PC, or send it down to the C64 and writing it back to a 1541. It is the best documented format since it is also native to the C64, with many books explaining the disk layout and the internals of the 1541.

The D64 layout is reasonably robust, being that it is an electronic representation of a physical 1541 disk. It shares most of the 1541 attributes and it supports all file formats, since all C64 files came from here. The only file I have found that can't be copied to a D64 is a T64 FRZ (FRoZen files), since you lose the extra file type attribute.

Since the D64 layout seems to be an exact byte copy of a 1541 floppy, it would appear to be the perfect format for any emulator. However, it does not contain certain vital bits of information that, as a user, you normally don't have access to.

Preceeding each sector on a real 1541 disk is a header block which contains the sector ID bytes and checksum. From the information contained in the header, the drive determines if there's an error on that header (27-checksum error, 29-disk ID mismatch). The sector itself also contains info (data block signature, checksum) that result in error detection (23 checksum, 22 data block not present, etc). The error bytes had to be added on to the D64 image, “extending” the format to take into account the missing info.

The disk ID is important in the copy protection of some programs. Some programs fail to work properly since the D64 doesn't contain these ID's. These bytes would be an addition to the format which has never been done and would be difficult to do. (As an aside, the 4-pack ZipCode files do contain the original master disk ID, but these are lost in the conversion of a ZipCode to a D64. Only storing one of the ID's is not enough, all the sector ID's should be kept.)

The extended track 1541 disks also presented a problem, as there are several different formats (and how/where to store the extra BAM entries in a sector that was not designed for them, yet still remain compatible). Because of the additions to the format (error bytes and 40 tracks) there exists 4 different types of D64's, all recognizeable by their size.

It is also the only format that uses the sector count for the file size rather than actual bytes used. This can present some problems when converting/copying the to another format because you may have to know the size before you begin (see LBR format).

It also contains no consistent signature, useful for recognizing if D64 is really what it claims to be. In order to determine if a file is a D64, you must check the file size.

• D64 files are the most widely supported and well-defined format, as it is simply an electronic version of a 1541 disk
• Supports all filenames, even those with $00's in them
• Filenames are padded with the standard $A0 character
• Supports REL and all GEOS files
• Allows complete directory customization
• Because it is a random-access device, it supports fast-loaders and random sector access
• PC Cluster slack-space loss is minimized since the file is a larger fixed size
• Has a label (description) field
• Format extensible to allow for 40-track disks
• With the inclusion of error bytes, you have support for basic copy-protection
• Files on a disk can easily be re-written, as long as there is free blocks

• The format doesn't contain all the info from the 1541 disk (no sector header info like ID bytes, checksums). This renders some of the original special-loaders and copy-protection useless.
• You don't really know the file size of the contained C64 files in bytes, only blocks
• It can't store C64s FRZ files due to FRZ files needing a special flag that a D64 can't store. This is by no means a big problem.
• It is not an expandable filetype, like LNX or T64
• Unless most of the space on a D64 disk is used, you do end up with wasted space
• Directory limited to 144 files maximum
• Cannot have loadable files with the same names
• Has no recognizeable file signature (unlike most other formats). The only reliable way to know if a file is a D64 is by its size
• It is too easy for people to muck up the standard layout
• It is much more difficult to support fully, as you really need to emulate the 1541 DOS (sector interleave, REL files, GEOS VLIR files)