The .ADF (Amiga Disk File) format FAQ

Laurent Clévy, laurent.clevy@alcatel.fr

v1.07, 27. May 1999

This document describes the .ADF file format. An Amiga Disk File is a sector per sector dump of an Amiga formatted disk. The intent is to explain in detail how the Amiga stores files and directories on floppy and hard disks.
A set of C routines (ADFlib) will be supplied to manage the ADF format.

0. Changes

1. Introduction

2. How bytes are physically read from and written to a disk ?

3. What is the Amiga floppy disk geometry ?

4. What is the logical organisation of an Amiga volume ?

5. How does a blank disk look like ?

6. The structure of a hard disks ?

7. The Hard file : a big floppy dump file

8. Advanced information

9. References and links

10. C Routines : the ADF Library

11. Other Amiga file systems


0. Changes

Since version 1.06 (2. May 1999), by Heiko Rath (hr@brewhr.swb.de) : Since version 1.04 (16. January 1999) : Since version 0.9 (28. May 1997) :

1. Introduction

In this document, we will describe how the AmigaDOS is (was?) managing storage media, from the magnetic layer to the files and directories layer.

With physical layer, I'm talking about the way bytes are physically stored on a magnetic surface, with the RLL or MFM encoding.
The next layer, according to the 'most physical' to 'most conceptual' order, is the partitions layer : this is how the AmigaDOS is managing media with more then one partition, like Zip disks or hard disks.
The next and last layer is the volume layer : where the files and directories are stored.

The physical layer is described in the 2nd chapter,
The volume layer is the biggest part of the document (4th and 5th chapters), since it's the most interesting,
The partitions layer is explained in the 6th chapter.

Let's continue with more conventional things in an introduction.


1.1 Disclaimer and copyright

This document is Copyright (C) 1997-1998 by Laurent Clévy, but may be freely distributed, provided the author name and addresses are included and no money is charged for this document.

This document is provided "as is". No warranties are made as to its correctness.

Amiga and AmigaDOS are registered Trademarks of Gateway 2000.
Macintosh is a registered Trademark of Apple.


1.2 Feedback, updates

If you find any mistakes in this document, have any comments about its content, feel free to send me an e-mail.
Corrections are very welcome.

You can find new versions of this document at :


1.3 Conventions

In this document, hexadecimal values use the C syntax : for example 0x0c is the decimal value 12.

Byte ordering

Since the Amiga is a 680x0 based computer, integers that require more than one byte are stored on disk in 'Motorola order' : the most significant byte comes first, then the less significant bytes in descending order of significance (MSB LSB for two-byte integers, B3 B2 B1 B0 for four-byte integers). This is usually called big endian byte ordering.
The Intel based PCs are using the little endian byte ordering.

Vocabulary

A 'word' or 'short' is a 2-byte (16 bits) integer, a 'long' a 4-byte (32 bits) integer. Values are unsigned unless otherwise noted.

A 'block' in this document will be 512 consecutive bytes on disk, unless noted otherwise, the variable 'BSIZE' will denote the blocksize.
The word 'sector' and 'block' will be used as synonyms here, even if 'sector' is usually related to the physical side, and the 'block' to the logical side. This is because the AmigaDOS can only handle one sector per block. Some other Unix filesystems can have more then one sector per block.

A block pointer is the number of this block on the disk. The first one is the #0 block.
There are 'logical' and 'physical' block pointers. 'Logical' ones are related to the start of one volume, 'physical' one are related to the start of a physical media. If a volume starts at the #0 physical sector, a physical pointer and a logical pointer is the same thing, like with floppies.

A simple definition of 'Hashing' could be : "a method to access tables : given a number or a string, a hash function returns an index into an array". This definition is correct for this document, but there is a lot of other hashing methods, that might be far more complex.

Linked lists are cell-oriented data structures. Each cell contains a pointer to the next or previous cell or both, the last cell pointer is null.

C example :

struct lcell {
	char name[10];
	/* contains next cell adress, or NULL if this cell is the last */
	struct lcell *next_cell;
	};

Block names begin with a capital (Rootblock). Field names are noted between quotes ('field_name').

All formats are described as tables, one row per field. Here is an example with the beginning of the well known GIF format :

offset  type   length  name          comments
----------------------------------------------------------
0       char    3      signature     'GIF'
3       char    3      version       '87a' or '89a'
6       short   1      screen width  (little endian)
8       short   1      screen height (little endian)

The .ADF format is the format created and used by the -incredible- UNIX Amiga Emulator (UAE), written by Berndt Schmitt. The home page is here : http://www.freiburg.linux.de/~uae/

The .ADF files can be created with the program transdisk.


1.4 Acknowledgements


2. How are bytes physically read from / written to a disk ?

The following part deals with the way the Amiga disk controller accesses the magnetic medium. If you only want to understand the .ADF format, you don't need to read this part.

Information is written on disk with magnetic fields. Magnetic fields can be made 'on' or 'off'. But the read/write heads are not capable of detecting directly if a field is on or off. An encoding is used to store memory bits on the medium. The CHANGE of fields polarisation will indicate if the bit is 1 or 0. For Amiga floppy disks (and PC floppies), the encoding scheme is MFM (Modified frequency modulation).

Notes on the Amiga floppy disk controller :

The Amiga floppy disk controller (FDC) which is called 'Paula' is very flexible. It is capable of reading/writting Amiga/PC/Macintosh floppy disks.

Paula can read a variable number of bytes from disk, the PC FDC can't. The PC FDC uses the index hole to find the beginning of a track, Paula uses a synchronization word. The Macintosh uses GCR encoding instead of MFM.
In fact, Paula is simpler than the PC FDC because it does not perform automatically the decoding just after the read operation, and the encoding just before the write operation : it must be done by software. The MFM decoding/encoding is done by hardware with the PC FDC, the Amiga can do GCR or MFM decoding/encoding because it's done with the CPU.

Classic PC FDCs can't read Amiga floppy disks even if they are MFM encoded on a 3.5 inch floppy, because they can not find the beginning of a track. This is why the .ADF format has been created.

However, a custom FDC available on PC machines is capable of reading/writing Amiga, PC, Macintosh, Atari and C64 floppies !!! This is CatWeasel : http://catweasel.vme.de/index.html and http://www.jschoenfeld.com/eindex.html

There's also the Amiga Floppy Reader Project at http://metamonkt.com/afr/ a project describing a small piece of hardware to connect to the parallel port of a standard PC, enabling it to read Amiga disks.

Paula parametrization for Amiga disks :

Paula is able to put the read/write heads on a cylinder, and is able to read with the lower or upper side head. A track of 0x1900 words is usually read.


2.1 What is MFM encoding/decoding ?

The MFM decoding is made by the Amiga CPU, not by Paula. This allows custom encoding, to protect floppies against copying for example.

Here follows the MFM encoding scheme :

    user's data bit      MFM coded bits
    ---------------      --------------
        1                   01
        0                   10 if following a 0 data bit
        0                   00 if following a 1 data bit

User data long words are split in two parts, a part with even bits part first, followed by a part with odd bits. Once encoded, the amount of data stored doubles.
The MFM decoding will transform magnetic fields into computer usuable bits.

The encoding process will take one long (user's data), and produces two longs (MFM coded longs): one for the even bits of the user long, a second for the odd bits of the user long.
Vice versa, the decoding process will take the half of two MFM longs to produce one user's long.


2.2 What is the MFM track format ?

Paula will search two synchronization words, and then read 0x1900 words of data. We will call those 0x1900 words a 'MFM track'.
There are 80 cylinders on a Amiga floppy disk. Each cylinder has 2 MFM tracks, 1 on each side of the disk.

Double density (DD) disks have 11 sectors per MFM track, High density (HD) disks have 22 sectors.

So a MFM track consists of 11/22 MFM encoded sectors, plus inter-track-gap. Note that sectors are not written from #0 to #10/21, you must use the 'info' field to restore the correct order when you read the tracks. Each MFM track begins with the first sector, and ends with the end of the last sector.
Each sector starts with 2 synchronization words. The synchronization value is 0x4489.


2.3 What is the MFM sector format ?

There are brut data and encoded data.
Brut data (also called MFM data) doesn't need to be decoded, this is the synchronization data, the header checksum and data checksum.

The encoded parts are 'header' and 'data'.

Here it comes :

00/0x00	word	2	MFM value 0xAAAA AAAA (used for timing ?)

	SYNCHRONIZATION
04/0x04	word	1	MFM value 0x4489
06/0x06	word	1	MFM value 0x4489

	HEADER
08/0x08	long	1	info (even bits)
12/0x0c	long	1	info (odd bits)
			decoded long is : 0xff TT SS SG
				TT = track number ( 3 means cylinder 1, head 1)
				SS = sector number ( 0 upto 10/21 )
					sectors are not ordered !!!
				SG = number of sectors before gap (including
					current one)
			Example for cylinder 0, head 1 of a DD disk :
				0xff010009
				0xff010108
				0xff010207
				0xff010306
				0xff010405
				0xff010504
				0xff010603
				0xff010702
				0xff010801
					-- inter-sector-gap here !
				0xff01090b (0xb means -1 ?)
				0xff010a0a (0xa means -2 ?)

16/0x10	long	4	sector label (even)
32/0x20	long	4	sector label (odd)
			decoded value seems to be always 0
	END OF HEADER

48/0x30	long	1	header checksum (even)
52/0x34	long	1	header checksum (odd)
			(computed on mfm longs,
			longs between offsets 8 and 44 
			== 2*(1+4) longs)

56/0x38	long	1	data checksum (even)
60/0x3c	long	1	data checksum (odd)
			(from 64 to 1088 == 2*512 longs)

	DATA
64/0x40	long	512	coded data (even)
576/0x240 long	512	coded data (odd)
1088/0x440
	END OF DATA

2.4 How to decode MFM data ?

C algorithm :


#define MASK 0x55555555	/* 01010101 ... 01010101 */
unsigned long *p1;	/* MFM coded data buffer (size == 2*data_size) */
unsigned long *q;	/* decoded data buffer (size == data_size) */
unsigned long a,b;
unsigned long chksum;
int data_size;		/* size in long, 1 for header's info, 4 for header's sector label */
int count;

chksum=0L;
/* the decoding is made here long by long : with data_size/4 iterations */
for (count=0; count<data_size/4; count++) {
	a = *p1;                /* longs with even bits */
	b = *(p1+data_size);    /* longs with odd bits : located 'data_size' bytes farther */
	chksum^=a;              /* eor */
	chksum^=b;
        /*
         * MFM decoding, explained on one byte here (x and y will produce t) :
         * the MFM bytes 'abcdefgh' == x and 'ijklmnop' == y will become
         * y & 0x55U = '0j0l0n0p'
         * ( x & 0x55U) << 1 = 'b0d0f0h0'
         * '0j0l0n0p' | 'b0d0f0h0' = 'bjdlfnhp' == t
         */ 
	/* on one long here : */
	*q = ( b & MASK ) | ( ( a & MASK ) << 1 );
	p1++;    /* next 'even bits' long and 'odd bits' long  */
	q++;     /* next location of the future decoded long */
	}
chksum&=MASK;	/* must be 0 after decoding */
For example, to decode the DATA field of a MFM sector :

3. What is the Amiga floppy disk geometry ?

After MFM decoding, you have usuable 'sectors' or 'blocks' into memory.

Here we remind the disk geometries for Double Density disks (DD) and High Density disks (HD) :
		bytes/sector	sector/track	track/cyl	cyl/disk
------------------------------------------------------------------------
DD disks	512		11		2		80
HD disks	512		22		2		80
The relations between sectors, sides and cylinders are for a DD disk :
Block	sector	side	cylinder
--------------------------------
0	0	0	0
1	1	0	0
2	2	0	0
...
10	10	0	0
11	0	1	0
...
21	10	1	0
22	0	0	1
..
1759	10	1	79
Order = increasing sectors, then increasing sides, then increasing cylinders.

A DD disk has 11*2*80=1760 (0 to 1759) blocks, a HD disk has 22*2*80=3520 blocks.

The length of .ADF files for a DD disk is therefore 512*11*2*80 = 901120 bytes.

Those 'raw' blocks, 512 consecutive bytes, store different 'logical' blocks to manage files and directories.

The classic Amiga filesystem has a internal command with one 32 bits wide offset parameter (unsigned). It tells where to start the read/write operation. The biggest size for an Amiga disk is therefore 2^32 = 4 GB.
Anyway, there exists a 3rd party patch which changes the 32 bits limit to 64 bits (on Aminet, disk/misc/ffstd64.lha).


4. What is the logical organisation of an Amiga volume ?

A volume is a floppy disk or a hard disk partition.

The first file system for the Amiga was embedded in the version 1.2 of AmigaDOS.
With version 2.xx of AmigaDOS the Fast File System (FFS) was introduced, an improved version of the 1.2, also called old file system (OFS).
The version 3.0 of AmigaDOS added an international characters mode (INTL) and a directory cache mode (DIRC).

Links are only supported under FFS.

The start of a floppy volume contains space for sectors which may contain boot code.
The middle of the volume contains information about the root (upper most) directory contents and information about free and used blocks.
Other blocks are of course used to store files and directories.

The file length, the directory tree depth, the number of entries per directory are only limited by disk size. (Actually the maximum filesize is limited to 4 Gbyte sizeof(ulong) which should normally be more than sufficient).

Let's introduce the logical structures used by the Amiga file system in a table (for floppies) :

Object		Related logical blocks
------------+----------------------------------------------------------------
Volume		Rootblock, Bitmap block
File		File Header block, Extension block, Data block, Link block
Directory	Rootblock, Directory block, Directory Cache block, Link block
The main data types are a trees and linked lists.


4.1 What is a Bootblock ?

Prior to Kickstart 2.0 the bootblock was hardcoded to consist of the first two sectors of the floppy disks (sector #0 and #1). As of Kick 2.0, booting via the boot-block could be done with any device driver and the number of blocks could be changed independantly of the number of reserved blocks by using BOOTBLOCKS in the DOS environment vector (DosEnvVec).
* BootBlock
-------------------------------------------------------------------------------
offset	size    number	name		meaning
-------------------------------------------------------------------------------
0/0x00  char    4       DiskType	'D''O''S' + flags
                                        flags = 3 least signifiant bits
                                               set         clr
					  0    FFS         OFS
                                          1    INTL ONLY   NO INTL ONLY
                                          2    DIRC&INTL   NO DIRC&INTL
4/0x04  ulong   1       Chksum          special block checksum
8/0x08  ulong   1       Rootblock       Value is 880 for DD and HD
12/0x0c char    *       Bootblock code  (see 5.2 'Bootable disk' for more info)
                                        The size for a floppy disk is 1012,
                                        for a harddisk it is
                                        (DosEnvVec->Bootblocks * BSIZE) - 12
-------------------------------------------------------------------------------
The DiskType flag informs of the disk format. The Old filesystem may have the international and dircache mode enabled. If the international mode is enabled, the bit #1 is set. If the dircache is enabled, its flag is set (bit #2), and the international mode is also enabled, but the related flag (bit #1) will stay cleared. The correct values for flag are therefore : 0 (OFS), 1 (FFS), 2 (OFS/INTL), 3 (FFS/INTL), 4 (OFS/DIRC&INTL), 5 (FFS/DIRC&INTL).

There are few differences between the two file systems :

If the Bootblock starts with the three characters 'PFS', another filesystem is used in place of AmigaDOS : the Professional File System.

If the checksum and the DiskType are correct, the system will execute the bootblock code, at boot time, of course :-).

The Bootblock code is optional, see 5.2 section.

The Bootblock checksum algorithm follows :

* in 68000 assembler :

	lea 	bootbuffer,a0
        move.l  a0,a1
        clr.l   4(a1)			;clear the checksum
        move.w  #(BOOTBLOCKSIZE/4)-1,d1	;for floppy disks = 1024
                                        ;for hd = (DosEnvVec->Bootblocks * BSIZE)
        moveq   #0,d0
lpchk:  add.l   (a0)+,d0		;accumulation
        bcc.s   jump                    ;if carry set, add 1 to checksum
        add.l   #1,d0
jump:   dbf     d1,lpchk		;next long word

        not.l   d0
        move.l  d0,4(a1)		;new checksum


* in C :

#include<limits.h>
#define Short(p) ((p)[0]<<8 | (p)[1])
#define Long(p) (Short(p)<<16 | Short(p+2))

unsigned long newsum,d;
unsigned char buf[BOOTBLOCKSIZE];	/* contains bootblock */ 
                                        /* for floppy disks = 1024,
                                         * for hd = (DosEnvVec->Bootblocks * BSIZE) */
int i;

memset(buf+4,0,4);			/* clear old checksum */
newsum=0L;
for(i=0; i<BOOTBLOCKSIZE/4; i++) {
	d=Long(buf+i*4);
	if ( (ULONG_MAX-newsum) < d )	/* overflow */
		newsum++; 
	newsum+=d; 
        } 

newsum=~newsum;		/* not */


4.2 What is a Rootblock ?

The Rootblock is located at the physical middle of the media : block number 880 for DD disks, block 1760 for HDs. The exact calculation where it is stored is as follows:

numCyls = highCyl - lowCyl + 1

highKey = numCyls * numSurfaces * numBlocksPerTrack - 1

rootKey = INT (numReserved + highKey) / 2

The Rootblock contains information about the disk : its name, its formatting date, etc ...

It also contains information to access the files/directories/links located at the uppermost (root) directory.

* Root block (BSIZE bytes) sector 880 for a DD disk, 1760 for a HD disk
------------------------------------------------------------------------------------------------
        0/ 0x00	ulong	1	type		block primary type = T_HEADER (value 2)
        4/ 0x04	ulong	1	header_key	unused in rootblock (value 0)
		ulong 	1 	high_seq	unused (value 0)
       12/ 0x0c	ulong	1	ht_size		Hash table size in long (= BSIZE/4 - 56)
                	                        For floppy disk value 0x48
       16/ 0x10	ulong	1	first_data	unused (value 0)
       20/ 0x14	ulong	1	chksum		Rootblock checksum
       24/ 0x18	ulong	*	ht[]		hash table (entry block number)
        	                                * = (BSIZE/4) - 56
                	                        for floppy disk: size= 72 longwords
BSIZE-200/-0xc8	ulong	1	bm_flag		bitmap flag, -1 means VALID
BSIZE-196/-0xc4	ulong	25	bm_pages[]	bitmap blocks pointers (first at 0)
BSIZE- 96/-0x60	ulong	1	bm_ext		first bitmap extension block
						(Hard disks only)
BSIZE- 92/-0x5c	ulong 	1 	r_days		last root alteration date : days since 1 jan 78
BSIZE- 88/-0x58	ulong 	1 	r_mins 		minutes past midnight
BSIZE- 84/-0x54	ulong 	1 	r_ticks 	ticks (1/50 sec) past last minute
BSIZE- 80/-0x50	char	1	name_len	volume name length
BSIZE- 79/-0x4f	char	30	diskname[]	volume name
BSIZE- 49/-0x31	char	1	UNUSED		set to 0
BSIZE- 48/-0x30	ulong	2	UNUSED		set to 0
BSIZE- 40/-0x28	ulong	1	v_days		last disk alteration date : days since 1 jan 78
BSIZE- 36/-0x24	ulong	1	v_mins		minutes past midnight
BSIZE- 32/-0x20	ulong	1	v_ticks		ticks (1/50 sec) past last minute
BSIZE- 28/-0x1c	ulong	1	c_days		filesystem creation date
BSIZE- 24/-0x18	ulong	1	c_mins 		
BSIZE- 20/-0x14	ulong	1	c_ticks
		ulong	1	next_hash	unused (value = 0)
		ulong	1	parent_dir	unused (value = 0)
BSIZE-  8/-0x08	ulong	1	extension	FFS: first directory cache block,
						0 otherwise
BSIZE-  4/-0x04	ulong	1	sec_type	block secondary type = ST_ROOT 
						(value 1)
------------------------------------------------------------------------------------------------

The '/' and ':' characters are forbidden in the volume name, but *!@#$%|^+&_()=\-[]{}';",<>.? and international like âè are allowed.

The date fields in the root block (and other blocks) are structured in the form of DAYS, MINS and TICKS. The DAYS field contains the number of days since January 1. 1978. MINS is the number of minutes that have passed since midnight and TICKS are expressed in 1/50s of a second. A day value of zero is considered illegal by most programs.

The r_date / r_min / r_ticks fields are updated to the last recent change of the root directory of this volume.

The v_date / v_min / v_ticks fields are updated whenever any change was made to this volume, not just the root directory.

The c_date / c_min / c_ticks fields contain the date and time when this volume was initialized (i.e. formatted) and is not changed during its lifetime.

Some date constraints : 0 <= Mins < 60*24, 0 <= Ticks < 50*60
There should be no 2000 year real problem with the Amiga filesystem. Anyway, there is a page on 2YK problem with the Amiga at : http://www.amiga.com.

4.2.1 How to find the first sector of a directory entry ?

Given the name of a file/directory/link you first have to compute its hash value with this algorithm :

* The hash function :

#include<ctype.h>

int HashName(char *name)
{
int hash;				/* sizeof(int)>=2 */
int i,l;

l=hash=strlen(name);
for(i=0; i<l; i++) {
        hash=hash*13;
        hash=hash + toupper(name[i]);	/* not case sensitive */
        hash=hash & 0x7ff;
        }
hash=hash % ((BSIZE/4)-56);		/* 0 < hash < 71
                                         * in the case of 512 byte blocks */

return(hash);
}

// this code only works with non international mode disks
// see section 5.4

The hash value is then used to access HashTable ('ht' field in Rootblock/Directory block).

HashTable[ HashValue ] contains the number of the first block of your object (File header block, Directory block or Link block).

But different names can result in the same HashValue. If more then one name has the same HashValue, the other blocks (for files and directory only) are stored in a chained list. This linked list starts at the 'next_hash' field of the File header or Directory block.

For example : 'file_1a', 'file_24' and 'file_5u' have the same hash value.

Here follows the method to find the requested block :

HashValue = HashName( name );
name=uppercase(name);
nsector = Hashtable[ HashValue ];
if (nsector != 0) {
	sector=Load(nsector);		/* reads the 'nsector' sector */
        sector.name = uppercase(sector.name);
        /*
         *  follows the 'same HashValue' chained list if needed
         */
	while ( sector.name != name and sector.Next_hash != 0) {
		sector = Load(nsector);
       	        sector.name = uppercase(sector.name);
	}
	if (sector.name != name)
		puts("File/Dir not found");
}
else
	puts("File/Dir not found");


// this code only works with non international mode disks
// see section 5.4
Figure : HashTable and Directory content

Filenames characters can be lowercase and uppercase, but as shown in the Hash function, are not case sensitive.

If, for a new entry, the value at hashTable[hashvalue] is different than 0, the new sector pointer will be stored in the last entry of the same-hashvalue-linked-list. It is necessary to check if the entry name already exists in this directory. In one word, in the same-hashValue list, the addition is made at the tail, not the head.

4.2.2 How to list all the directory entries ?

Look through the whole HashTable and follow the same 'HashValue' linked lists if they exist.

4.2.3 How to compute the checksum ?

#define Short(p) ((p)[0]<<8 | (p)[1])
#define Long(p) (Short(p)<<16 | Short(p+2))

unsigned long newsum;
unsigned char buf[BSIZE];	/* contains rootblock */
int i;

memset(buf+20,0,4);		/* clear old checksum */
newsum=0L;
for(i=0; i<(BSIZE/4); i++)
	newsum+=Long(buf+i*4);
newsum=-newsum;			/* negation */

This checksum algorithm works for most blocks type except for Bootblock.

The bitmap table ('bm_pages[]') stores one or several pointers to Bitmap blocks. The first pointer is at index 0.


4.3 How are the free and used block lists managed?

Bitmap blocks contain information about free and allocated sectors. One bit is used per sector. If the bit is set, the sector is free, a cleared bit means an allocated sector.

Boot blocks allocation (2 for floppy, for hard disks the value can be found at DOSEnvVec->Bootblocks) is not stored in bitmap. Bitmap is stored in longs, the first sector of a long is the bit 0.

* Bitmap block (BSIZE bytes), often at rootblock+1
-------------------------------------------------------------------------------
0/0x00	long	1		checksum	normal algorithm
4/0x04	long	(BSIZE/4)-1	map
-------------------------------------------------------------------------------

Here follows for a DD disk the relationship between bitmap and sector number :

block #		long #	bit #
-------------------------------
2		0	0
3		0	1
4		0	2
...
33		0	31
34		1	0
35		1	1
...
880		27	14
881		27	15
...
1759		54	28
1760		54	29
This map is 1758 bits long (1760-2) and is stored on 54 full filled long and the first 30th bits of the 55th long.

* What is the 'bm_ext' field in Rootblock ?

If 25 bitmap blocks (which pointers are stored on Rootblock) are not enough (for Hard Disks > ca. 50 Mbyte), extra needed bitmap blocks pointers are stored in bitmap extension blocks.

* Bitmap extension block (BSIZE bytes) (Hard disk only)
-------------------------------------------------------------------------------
       0/0x00	ulong	(BSIZE/4)-1	bitmap block pointers
BSIZE- 4/0x04	ulong	1		next (0 for last)
-------------------------------------------------------------------------------
The Bitmap extension linked list start at Rootblock with the 'bm_ext'.


4.4 How are files stored ?

The first block of a file is the File header block, which contains information about the file (size, last access date, ...) and pointers to the first ((BSIZE/4)-56) Data blocks (i.e. for the often found case of BSIZE=512 it contains 72 pointers).
Data blocks store the actual file data.

If more than ((BSIZE/4)-56) Data blocks are needed to store a file, the other Data block pointers are stored in File extension blocks.

File extension blocks are organised in a linked list, which starts in File header block ('extension' field).

Figure : Chained lists of the blocks which store files

* File header block (BSIZE bytes) 
------------------------------------------------------------------------------------------------
        0/ 0x00 ulong	1	type		block primary type T_HEADER (==2)
        4/ 0x04 ulong	1	header_key	self pointer (to this block)
        8/ 0x08	ulong	1	high_seq	number of data block ptr stored here
       12/ 0x0c ulong	1	data_size	unused (==0)
       16/ 0x10	ulong	1	first_data	first data block ptr
       20/ 0x14	ulong	1	chksum		same algorithm as rootblock
       24/ 0x18 ulong	*	data_blocks[]	data blk ptr (first at BSIZE-204 )
        	                                * = (BSIZE/4) - 56
BSIZE-200/-0xc8	ulong	2	UNUSED		set to 0
BSIZE-192/-0xc0	ulong	1	protect		protection flags (bit set)
                                                (default set to 0)
					0	delete forbidden (D)
					1	modification forbidden (E)
					2	write/update forbidden (W)
					3	read forbidden (R)
					4	file is archived (A)
					5	pure, can be made resident (P)
					6	file is a script (S)
					7	hidden file with DIR (H)
                                     23-8	reserved
				    31-24	user
BSIZE-188/-0xbc	ulong	1	byte_size	file size in bytes
BSIZE-184/-0xb8	char	1	comm_len	file comment length
BSIZE-183/-0xb7	char	79	comment[]	comment (max. 79 chars permitted)
BSIZE-104/-0x69	char	12	UNUSED		set to 0
BSIZE- 92/-0x5c	ulong	1	days		last access date (days since 1 jan 78)
BSIZE- 88/-0x58	ulong	1	mins		last access time
BSIZE- 84/-0x54	ulong	1	ticks		in 1/50s of a seconds
BSIZE- 80/-0x50	char	1	name_len	filename length
BSIZE- 79/-0x4f char	30	filename[]	filename (max. 30 chars permitted)	
BSIZE- 49/-0x31 char	1	UNUSED		set to 0
BSIZE- 48/-0x30 ulong	1	UNUSED		set to 0
BSIZE- 44/-0x2a	ulong	1	real_entry	FFS : unused (== 0)
BSIZE- 40/-0x28	ulong	1	next_link	FFS : hardlinks chained list (first=newest)
BSIZE- 36/-0x24	ulong	5	UNUSED		set to 0
BSIZE- 16/-0x10	ulong	1	hash_chain	next entry ptr with same hash
BSIZE- 12/-0x0c	ulong	1	parent		parent directory
BSIZE-  8/-0x08	ulong	1	extension	pointer to 1st file extension block
BSIZE-  4/-0x04	ulong	1	sec_type	secondary type : ST_FILE (== -3)
------------------------------------------------------------------------------------------------

As within volume names ':' and '/' are forbidden within file names.

The number of blocks used to store one file depends on the filesystem used, OFS or FFS. If one file has 7 datablocks, the first is at datablock[71-0], the last at datablocks[71-6], and highseq is equals to 7.

For the OFS there are two ways of reading the contents of a file. First by traversing the linked list of data blocks that is pointed to in first_data (offset 16) and then following the pointers in each file data block. The other way of accessing the file data is by using the data_blocks[] table and going backwards through the data blocks listed there and then the File extension blocks.

As the FFS doesn't contain extra information in the data blocks (no pointer list, no checksum) the only way of accessing the file contents is by going through the data_blocks[] table and the File extension blocks.

For an empty file, a Header file block is written, with 'byte_size' equal to 0, and no Data block pointers in 'data_blocks[]'.

* File extension block (BSIZE bytes) (first pointer in File header)
------------------------------------------------------------------------------------------------
        0/ 0x00	ulong	1	type		primary type : T_LIST (== 16)
        4/ 0x04	ulong	1	header_key	self pointer
        8/ 0x08	ulong	1	high_seq	number of data blk ptr stored
       12/ 0x0c	ulong	1	UNUSED		unused (== 0)
       16/ 0x10	ulong	1	UNUSED		unused (== 0)
       20/ 0x14	ulong	1	chksum		rootblock algorithm
       24/ 0x18	ulong	*	data_blocks[]	data blk ptr (first at BSIZE-204)
        	                                * = (BSIZE/4) - 56
BSIZE-200/-0xc8	ulong	46	info		unused (== 0)
BSIZE- 16/-0x10	ulong	1	UNUSED		unused (== 0)
BSIZE- 12/-0x0c	ulong	1	parent		ptr to parent directory
BSIZE-  8/-0x08	ulong	1	extension	next file header extension block, 
	                                        0 for the last
BSIZE-  4/-0x04	ulong	1	sec_type	secondary type : ST_FILE (== -3)
------------------------------------------------------------------------------------------------
* Data blocks (BSIZE bytes) (first pointer in File header 'first_data' and 'data_blocks[((BSIZE/4)-57)]')
Old File System data block (BSIZE bytes)
-------------------------------------------------------------------------------
0/0	ulong	1	type		primary type : T_DATA (== 8)
4/4	ulong	1	header_key	pointer to file header block
8/8	ulong	1	seq_num		file data block number (first is #1) 
12/c	ulong	1	data_size	data size <= (BSIZE-24)
16/10	ulong	1	next_data	next data block ptr (0 for last)
20/14	ulong	1	chksum		rootblock algorithm
24/18	UCHAR	*	data[]		file data size <= (BSIZE-24)
-------------------------------------------------------------------------------
In OFS, there is a second way to read a file : using the Data block chained list. The list starts in File header ('first_data') and goes on with 'next_data' in each Data block.

Fast File System (BSIZE bytes)
-------------------------------------------------------------------------------
0/0	UCHAR	BSIZE	data[]		file data
-------------------------------------------------------------------------------
In FFS, the only way to read or recover a file is to use data_blocks[] in the file header block and the File extension blocks. If a File header or File extension block is unreadable, there is no way to find the corresponding Data blocks.

The OFS is more robust than FFS, but slower and can store less data on disk. As you see, disk salvaging is easier with OFS.

When a file is deleted, only its File header block number is cleared from the Directory block (or from the same-hash-value list) and the bitmap is updated. File header block, Data blocks and File extension blocks are not cleared ! but the bitmap blocks are updated. Nevertheless, the undelete operation is easy .


4.5 How are directories stored?

Directory blocks are very similar to Rootblock, except they don't need information about bitmap and disk, but have comments.
* User directory block (BSIZE bytes)
------------------------------------------------------------------------------------------------
        0/ 0x00	ulong	1	type		block primary type = T_HEADER (value 2)
        4/ 0x04	ulong	1	header_key	self pointer
	8/ 0x08	ulong 	3 	UNUSED		unused (== 0)
       20/ 0x14	ulong	1	chksum		normal checksum algorithm
       24/ 0x18	ulong	*	ht[]		hash table (entry block number)
        	                                * = (BSIZE/4) - 56
                	                        for floppy disk: size= 72 longwords
BSIZE-200/-0xc8	ulong	2	UNUSED		unused (== 0)
BSIZE-192/-0xc0	ulong	1	protect		protection flags (bit set)
                                                (default set to 0)
					0	delete forbidden (D)
					1	modification forbidden (E)
					2	write/update forbidden (W)
					3	read forbidden (R)
					4	directory is archived (A)
					5	pure, can be made resident (P)
					6	file is a script (S)
					7	hidden file with DIR (H)
                                     23-8	reserved
				    31-24	user
BSIZE-188/-0xbc	ulong	1	UNUSED		unused (== 0)
BSIZE-184/-0xb8	char	1	comm_len	directory comment length
BSIZE-183/-0xb7	char	79	comment[]	comment (max. 79 chars permitted)
BSIZE-104/-0x69	char	12	UNUSED		set to 0
BSIZE- 92/-0x5c	ulong	1	days		last access date (days since 1 jan 78)
BSIZE- 88/-0x58	ulong	1	mins		last access time
BSIZE- 84/-0x54	ulong	1	ticks		in 1/50s of a seconds
BSIZE- 80/-0x50	char	1	name_len	directory name length
BSIZE- 79/-0x4f char	30	dirname[]	directory (max. 30 chars permitted)	
BSIZE- 49/-0x31 char	1	UNUSED		set to 0
BSIZE- 48/-0x30 ulong	2	UNUSED		set to 0
BSIZE- 40/-0x28	ulong	1	next_link	FFS : hardlinks chained list (first=newest)
BSIZE- 36/-0x24	ulong	5	UNUSED		set to 0
BSIZE- 16/-0x10	ulong	1	hash_chain	next entry ptr with same hash
BSIZE- 12/-0x0c	ulong	1	parent		parent directory
BSIZE-  8/-0x08	ulong	1	extension	FFS : first directory cache block
BSIZE-  4/-0x04	ulong	1	sec_type	secondary type : ST_USERDIR (== 2)
------------------------------------------------------------------------------------------------
You can obtain directory listing exactly like with the root directory.


4.6 How are links implemented in AmigaDOS ?

Only FFS handles links and directory caching.
Soft links are not supported by AmigaDOS 3.0. Linked files are seen as directories...
However, some shells (like Csh 5.37) support them, so I'm supplying the structure.

4.6.1 Hard links

* Hard link (BSIZE bytes)
------------------------------------------------------------------------------------------------
        0/ 0x00	ulong	1	type		block primary type = T_HEADER (value 2)
        4/ 0x04	ulong	1	header_key	self pointer
	8/ 0x08	ulong 	3 	UNUSED		unused (== 0)
       20/ 0x14	ulong	1	chksum		normal checksum algorithm
       24/ 0x18	ulong	*	UNUSED		set to 0
        	                                * = (BSIZE/4) - 54
                	                        for floppy disk: size= 74 longwords
BSIZE-192/-0xc0	ulong	1	protect		protection flags (bit set)
                                                (default set to 0)
					0	delete forbidden (D)
					1	modification forbidden (E)
					2	write/update forbidden (W)
					3	read forbidden (R)
					4	directory is archived (A)
					5	pure, can be made resident (P)
					6	file is a script (S)
					7	hidden file with DIR (H)
                                     23-8	reserved
				    31-24	user
BSIZE-188/-0xbc	ulong	1	UNUSED		unused (== 0)
BSIZE-184/-0xb8	char	1	comm_len	comment length
BSIZE-183/-0xb7	char	79	comment[]	comment (max. 79 chars permitted)
BSIZE-104/-0x69	char	12	UNUSED		set to 0
BSIZE- 92/-0x5c	ulong	1	days		last access date (days since 1 jan 78)
BSIZE- 88/-0x58	ulong	1	mins		last access time
BSIZE- 84/-0x54	ulong	1	ticks		in 1/50s of a seconds
BSIZE- 80/-0x50	char	1	name_len	hard link name length
BSIZE- 79/-0x4f char	30	hlname[]	hardlink name (max. 30 chars permitted)	
BSIZE- 49/-0x31 char	1	UNUSED		set to 0
BSIZE- 48/-0x30 ulong	1	UNUSED		set to 0
BSIZE- 44/-0x2c	ulong	1	real_entry	FFS : pointer to "real" file or directory
BSIZE- 40/-0x28	ulong	1	next_link	FFS : hardlinks chained list (first=newest)
BSIZE- 36/-0x24	ulong	5	UNUSED		set to 0
BSIZE- 16/-0x10	ulong	1	hash_chain	next entry ptr with same hash
BSIZE- 12/-0x0c	ulong	1	parent		parent directory
BSIZE-  8/-0x08	ulong	1	UNUSED		set to 0
BSIZE-  4/-0x04	ulong	1	sec_type	secondary type : ST_LINKFILE = -4
						ST_LINKDIR = 4
------------------------------------------------------------------------------------------------
A 'real' entry is a file or directory entry, opposed to link entries.

A hard link can only be created on the same disk as the real entry disk. Several links can be made on the same real entry. This is stored with a chained list.
'real entry' always contains the real entry block pointer.
'next_link' stores the links linked list.

New entries are added at the head:

>ls
  ------rw-d     1912  15-May-96 22:28:08  real
Chained list state :
block# real	next	name
----------------------------
484	0	0	real


>ln real link1
>ls
  ------rw-d     1912  15-May-96 22:28:08  real
  -H----rw-d     1912  15-May-96 22:28:10  link1 -> Empty:real

block# real	next	name
----------------------------
484	0	104	real
104	484	0	link1


>ln link1 link2
>ls
  ------rw-d     1912  15-May-96 22:28:08  real
  -H----rw-d     1912  15-May-96 22:28:10  link1 -> Empty:real
  -H----rw-d     1912  15-May-96 22:28:12  link2 -> Empty:real

block# real	next	name
----------------------------
484	0	107	real
104	484	0	link1
107	484	104	link2
The links are stored 'newest first', due to the adding at head.

real -> newest link -> ... -> oldest link -> 0

-> means "points to"

4.6.2 Soft links

* Soft link (BSIZE bytes)
------------------------------------------------------------------------------------------------
        0/ 0x00	ulong	1	type		block primary type = T_HEADER (value 2)
        4/ 0x04	ulong	1	header_key	self pointer
	8/ 0x08	ulong 	3 	UNUSED		unused (== 0)
       20/ 0x14	ulong	1	chksum		normal checksum algorithm
       24/ 0x18	ulong	*	symbolic_name	path name to referenced object, Cstring
        	                                * = ((BSIZE - 224) - 1)
                	                        for floppy disk: size= 288 - 1 chars
BSIZE-200/-0xc8	ulong	2	UNUSED		unused (== 0)
BSIZE-192/-0xc0	ulong	1	protect		protection flags (bit set)
                                                (default set to 0)
					0	delete forbidden (D)
					1	modification forbidden (E)
					2	write/update forbidden (W)
					3	read forbidden (R)
					4	directory is archived (A)
					5	pure, can be made resident (P)
					6	file is a script (S)
					7	hidden file with DIR (H)
                                     23-8	reserved
				    31-24	user
BSIZE-188/-0xbc	ulong	1	UNUSED		unused (== 0)
BSIZE-184/-0xb8	char	1	comm_len	comment length
BSIZE-183/-0xb7	char	79	comment[]	comment (max. 79 chars permitted)
BSIZE-104/-0x69	char	12	UNUSED		set to 0
BSIZE- 92/-0x5c	ulong	1	days		last access date (days since 1 jan 78)
BSIZE- 88/-0x58	ulong	1	mins		last access time
BSIZE- 84/-0x54	ulong	1	ticks		in 1/50s of a seconds
BSIZE- 80/-0x50	char	1	name_len	soft link name length
BSIZE- 79/-0x4f char	30	slname[]	softlink name (max. 30 chars permitted)	
BSIZE- 49/-0x31 char	1	UNUSED		set to 0
BSIZE- 48/-0x30 ulong	8	UNUSED		set to 0
BSIZE- 16/-0x10	ulong	1	hash_chain	next entry ptr with same hash
BSIZE- 12/-0x0c	ulong	1	parent		parent directory
BSIZE-  8/-0x08	ulong	1	UNUSED		set to 0
BSIZE-  4/-0x04	ulong	1	sec_type	secondary type : ST_SOFTLINK = 3
------------------------------------------------------------------------------------------------

4.7 How are the blocks associated with the directory cache mode ?

To speed up directory listing, Directory cache blocks have been created.
Directory cache blocks are also organised in chained lists.
The list starts at the directory block (root or normal directory) with the 'extension' field.
* Directory cache block (BSIZE bytes)
-------------------------------------------------------------------------------
0/0	ulong	1	type		DIRCACHE == 33 (0x21)
4/4	ulong	1	header_key	self pointer
8/8	ulong	1	parent		parent directory
12/c	ulong	1	records_nb	directory entry records in this block
16/10	ulong	1	next_dirc	dir cache chained list
20/14	ulong	1	chksum		normal checksum
24/18	UCHAR	*	records[]	entries list (size = BSIZE-24)
-------------------------------------------------------------------------------
The directory entries are stored this way :
* Directory cache block entry record (26 <= size (in bytes) <= 77)
-------------------------------------------------------------------------------
0	ulong	1	header		entry block pointer
                                        (the link block for a link)
4	ulong	1	size		file size (0 for a directory or a link)
8	ulong	1	protect		protection flags (0 for a link ?)
12	ulong	1	UNUSED		( == 0 )
16	short	1	days		date (always filled)
18	short	1	mins		time (always filled)
20	short	1	ticks
22	char	1	type		secondary type
23	char	1	name_len	1 <= len <= 30 (nl)
24	char	?	name		name
24+nl	char	1	comm_len	0 <= len <= 22 (cl)
25+nl	char	?	comment		comment
25+nl+cl char	1	OPTIONAL padding byte(680x0 longs must be word aligned)
-------------------------------------------------------------------------------

5. How does a blank disk look like ?

A minimal blank disk has a Bootblock, a Rootblock and a Bitmap block.

5.1 a Minimal blank floppy disk

* The Bootblock (0 and 1)

0	char	4	ID		'D''O''S' + flags
4	long	1023	full of zeros


* The Rootblock (880)

0	long	1	type		2
12/c	long	1	ht_size		0x48
20/14	long	1	checksum	computed
312/138	long	1	bm_flag		-1 (valid bitmap)
316/13c	long	1	bm_pages[0]	bitmap sector #
420/1a4	long	1	last access date
424/1a8	long	1	last access time
428/1ac	long	1	last access time
432/1b0	char	1	disk_name size
433/1b1	char	?	disk_name
472/1d8	long	1	last access date
476/1dc	long	1	last access time
480/1e0	long	1	last access time
484/1e4 long    1       creation date
488/1e8 long    1       creation time
492/1ec long    1       creation time
504/1f8	long	1	FFS : first dir cache sector  or 0
508/1fc	long	1	sub_type	1

No specified fields are set to 0.


* The Bitmap block (here 881) for a DD disk

0	long	1	checksum
4	long	27	free sectors	0xffffffff
112/70	long	1	root+bitmap	0xffff3fff
116/74	long	27	free sectors	0xffffffff
120/78	long	72	unused		!=0

5.2 A 'Bootable' floppy disk

* The Bootblock becomes :
0/0x00	long	1	ID		'D''O''S' + flags
4/0x04	long	1	checksum	computed
8/0x08	long	1	rootblock ?	880
12/0x0c	byte	81	bootcode	AmigaDOS 3.0 version

	values			disassembled
	--------------+---------------------
	43FA003E		lea	exp(pc),a1	;Lib name
	7025			moveq	#37,d0		;Lib version
	4EAEFDD8		jsr	-552(a6)	;OpenLibrary()
	4A80			tst.l	d0		;error == 0
	670C			beq.b	error1
	2240			move.l	d0,a1		;lib pointer
	08E90006 0022		bset	#6,34(a1)	;(*)
	4EAEFE62		jsr	-414(a6)	;CloseLibrary()
	43FA0018	error1:	lea	dos(PC),a1	;name
	4EAEFFA0		jsr	-96(a6)		;FindResident()
	4A80			tst.l	d0
	670A			beq.b	error2		;not found
	2040			move.l	d0,a0
	20680016		move.l	22(a0),a0	;DosInit sub
	7000			moveq	#0,d0
	4E75			rts
	70FF		error2:	moveq	#-1,d0
	4E75			rts
	646F732E 6C696272 617279
			dos:	"dos.library"
	00						;padding byte
	65787061 6E73696F 6E2E6C69 62726172 79
			exp:	"expansion.library"

93/0x5d	byte	931	full of zeros
(*) from Thomas Kessler (tkessler@ra.abo.fi), may 1997 :
This bit tells the shell (which opens its shell-window when booting the startup-sequence) not to open window unless needed, so a black screen stays there during boot instead of an empty shell-windows (it's a os2.x feature).


5.3 A Directory cache mode floppy disk

* A directory cache block (here 882)

0	long	1	type		0x21
4	long	1	self pointer	882
8	long	1	cached dir	880 (root)
12/c	long	1	entries number	0
16/10	long	1	next dir cache	0 (last)
20/14	long	1	checksum	computed
24	long	122	full of zeros

5.4 International Mode

The toupper() function in the HashName() function (3.2.1 paragraph) is replaced by the following function with the aim of better handling international characters :
int intl_toupper(int c)
{
   return (c>='a' && c<='z') || (c>=224 && c<=254 && c!=247) ? c - ('a'-'A') : c ;
}
In the Amiga ASCII table, the international character codes are between 192 and 254. Uppercase caracters are between 192 and 222, the lowercase versions of them are between 224 and 254. The only exception are the codes 215 and 247, which are respectively the multiply sign and the divide sign.

The Amiga character set is the same as ISO 8879 Latin-1 character set, used in HTML standard. This character set is described here : http://www.w3c.org/


6. The structure of a hard disk

The following structures is mainly extracted from the 'devices/hardblocks.h' and 'dos/filehandler.h' files delivered in Commodore developer kits.

The hard disk specific structures mainly store the drive geometry, the written partitions sizes and the filesystem bootcode.

The five kind of blocks are in a reserved area, at the beginning of the surface. The first of them, Rigid Disk block (RDSK), must be found within the first 16 blocks of BSIZE lenght. But it can be written inside the data area, which is dangerous.

6.1 What is the Rigid Disk Block ?

* Rigid Disk block (256 bytes) must exist within the first 16 blocks
-------------------------------------------------------------------------------
0/0	char	4	id		'RDSK'
4/4	ulong	1	size in longs 	== 64
8/8	long	1	checksum	classic Rootblock algorithm
12/c	ulong	1	hostID		SCSI Target ID of host
					(== 7 for IDE and ZIP disks)
16/10	ulong	1 	block size 	typically 512 bytes, but can
					be other powers of 2
20/14	ulong	1	flags 		typically 0x17
				Bits	If set means :
				0 	No disks exists to be configured 
					after this one on this controller
				1 	No LUNs exists to be configured greater
					than this one at this SCSI Target ID
				2 	No target IDs exists to be configured
					greater than this one on this SCSI bus
				3 	Don't bother trying to perform
					reselection when talking to this drive
				4 	Disk indentification valid
				5 	Controller indentification valid
				6 	Drive supports SCSI synchronous mode
					(can be dangerous if it doesn't)
24/18 	ulong 	1 	Bad blockList 	block pointer (-1 means last block)
28/1c 	ulong 	1 	PartitionList	block pointer (-1 means last)
32/20 	ulong 	1 	FileSysHdrList 	block pointer (-1 means last)
36/24 	ulong 	1 	DriveInit code 	optional drive-specific init code
					DriveInit(lun,rdb,ior) : 
					"C" stack and d0/a0/a1
40/28 	ulong 	6 	RESERVED 	== -1

	Physical drive caracteristics
64/40	ulong 	1 	cylinders 	number of drive cylinder
68/44 	ulong 	1 	sectors 	sectors per track
72/48	ulong 	1 	heads 		number of drive heads
76/4c 	ulong 	1 	interleave
80/50 	ulong 	1 	parking zone 	landing zone cylinders
					soon after the last cylinder
84/54 	ulong	3 	RESERVED 	== 0
96/60 	ulong 	1 	WritePreComp 	starting cyl : write precompensation
100/64	ulong 	1 	ReducedWrite 	starting cyl : reduced write current
104/68 	ulong 	1 	StepRate 	drive step rate
108/6c 	ulong 	5 	RESERVED 	== 0

	Logical drive caracteristics
128/80 	ulong 	1 	RDB_BlockLo 	low block of range reserved for hardblk
132/84 	ulong 	1 	RDB_BlockHi 	high block of range for this hardblocks
136/88 	ulong 	1 	LoCylinder 	low cylinder of partitionable disk area
140/8c 	ulong 	1 	HiCylinder 	high cylinder of partitionable data area
144/90 	ulong 	1 	CylBlocks 	number of blocks available per cylinder
148/94 	ulong 	1 	AutoParkSeconds zero for no autopark
152/98 	ulong 	1 	HighRSDKBlock 	highest block used by RDSK 
					(not including replacement bad blocks)
156/9c 	ulong 	1 	RESERVED 	== 0

	Drive identification
160/a0 	char 	8 	DiskVendor 	ie 'IOMEGA'
168/a8	char 	16 	DiskProduct 	ie 'ZIP 100'
184/b8	char 	4 	DiskRevision 	ie 'R.41'
188/bc 	char 	8 	ControllerVendor
196/c4 	char 	16 	ControllerProduct
212/d4 	char 	4 	ControllerRevision
216/d8 	ulong 	10 	RESERVED 	== 0
256/100
-------------------------------------------------------------------------------

* How to find the physical geometry of the disk ?

A hard disk is made of several physical disks. They have one head for each writable side. Each physical disk consists of several tracks, which consist of several sectors. One cylinder is the set of the tracks which have the same number on each disk.

The total size of the hard disk is expressed in cylinders ('cylinders').
The size of a cylinder is :
the number of heads per cylinder ('heads')
x the number of sectors per track ('sectors')
x the size of a block ('block size').

The 'CylBlocks' field equals to 'heads' x 'sectors'.

The reserved area is often the 2 first cylinders, between the 'RDB_BlockLo' block and the 'RDB_BlockHi' block, included. The partitionable area, starts at the 'LoCylinder' cylinder until the 'HiCylinder' cylinder, included.

The really last used sector in the reserved area is the sector numbered 'HighRSDKBlock', the first is numbered 0. The SCSI 'hostID' is set to the id of the SCSI host controller, which is typically 7. Real SCSI drives ID must be between 0 and 6.

The RDSK block is the "root" of the reserved area. It also contains the first blocks of three linked lists : one the bad blocks replacement, one for the partition definitions and one last for the filesystem information.

Some geometry examples :


6.2 How are bad blocks managed ?

* Bad Block block (BSIZE bytes) first in RDSK 'BadBlockList' field
-------------------------------------------------------------------------------
0/0 	ulong 	1 	id 		'BADB'
4/4 	ulong 	1 	size in longs 	== 128 for BSIZE = 512
8/8 	long 	1 	checksum
12/c 	ulong 	1 	HostID 		== 7 ?
16/10 	ulong 	1 	next 		next BadBlock block
20/14 	ulong 	1 	RESERVED
24/18 	 	* 	BlockPairs[]	bad block entries table
					* size = ((BSIZE/4)-6)/2
					(for BSIZE=512 = 61*8 byte entries)
-------------------------------------------------------------------------------
* Bad Block entry (8 bytes) stored in BadBlock 'BlockPairs[]' field
-------------------------------------------------------------------------------
0/0 	ulong 	1 	BadBlock 	block number of bad block
4/4 	ulong 	1 	GoodBlock 	block number of replacement block
-------------------------------------------------------------------------------

6.3 How are partitions stored?

* Partition block (256 bytes) first in RDSK 'PartitionList' field
-------------------------------------------------------------------------------
0/0 	char 	4 	ID 		'PART'
4/4 	ulong 	1 	size in long 	of checksummed structure (== 64)
8/8 	ulong 	1 	checksum        classic algorithm
12/c 	ulong 	1 	hostID 		SCSI Target ID of host (== 7)
16/10 	ulong 	1 	next 		block number of the next Partitionblock
20/14 	ulong 	1 	Flags
				Bits 	If set means
				0 	This partition is bootable
				1 	No automount
24/18 	ulong 	2 	RESERVED
32/20 	ulong 	1 	DevFlags 	preferred flags for OpenDevice
36/24 	char 	1 	DriveName len 	length of Drive name (e.g. 3)
37/25	char 	31 	DriveName 	e.g. 'DH0'
68/44 	ulong 	15 	RESERVED

	DOS Environment vector (DOSEnvVec) (often defined in MountLists)
128/80 	ulong 	1 	size of vector 	== 16 (longs), 11 is the minimal value
132/84 	ulong 	1 	SizeBlock	size of the blocks in longs ==
					128 for BSIZE = 512
136/88 	ulong 	1 	SecOrg 		== 0
140/8c 	ulong 	1 	Surfaces 	number of heads (surfaces) of drive
144/90 	ulong 	1 	sectors/block 	sectors per block == 1
148/94 	ulong 	1 	blocks/track 	blocks per track
152/98 	ulong 	1 	Reserved 	DOS reserved blocks at start of partition
                                        usually = 2 (minimum 1)
156/9c 	ulong 	1 	PreAlloc 	DOS reserved blocks at end of partition
					(no impact on Root block allocation)
					normally set to == 0
160/a0 	ulong 	1 	Interleave 	== 0
164/a4 	ulong 	1 	LowCyl		first cylinder of a partition (inclusive)
168/a8 	ulong 	1 	HighCyl		last cylinder of a partition (inclusive)
172/ac 	ulong 	1 	NumBuffer 	often 30 (used for buffering)
176/b0 	ulong 	1 	BufMemType 	type of mem to allocate for buffers ==0
180/b4 	ulong 	1 	MaxTransfer 	max number of type to transfer at a type
					often 0x7fff ffff
184/b8 	ulong 	1 	Mask 		Address mask to block out certain memory
					often 0xffff fffe
188/bc 	ulong	1 	BootPri 	boot priority for autoboot
192/c0 	char	4	DosType 	'DOS' and the FFS/OFS flag only
					also 'UNI'\0 = AT&T SysV filesystem
					'UNI'\1 = UNIX boot filesystem
					'UNI'\2 = BSD filesystem for SysV
					'resv' = reserved (swap space)
196/c4  ulong	1	Baud 		Define default baud rate for Commodore's
					SER and AUX handlers, originally
					used with the A2232 multiserial board
200/c8  ulong	1	Control		used by Commodore's AUX handler
204/cc  ulong	1	Bootblocks	Kickstart 2.0: number of blocks
					containing boot code to be
					loaded at startup
208/d0	ulong	12 	RESERVED
-------------------------------------------------------------------------------
There exists one 'PART' block per partition.

The block pointers in the reserved area are relative to the beginning of the media. The block pointers in a partition are relative to the first block of the partition.

The Rootblock of a partition is normally located in the middle of an AmigaDOS filesystem. Please see 4.2 What is a Rootblock? for the exact calculation of it's location.

The first two blocks of a partition contain a Bootblock. You have to use it to determine the correct file system, and if the international or dircache modes are used. Don't rely only on the PART and FSHD 'DosType' field.


6.4 What are FSHD blocks ?

* Filesystem header block (256 bytes) first in RSDK 'FileSysHeaderList'
-------------------------------------------------------------------------------
0/0 	char 	4 	id 		'FSHD'
4/4 	ulong 	1 	size in longs 	== 64
8/8 	long 	1 	checksum        classic algorithm
12/c 	ulong 	1 	hostID 		SCSI Target ID of host (often 7)
16/10 	ulong 	1 	next 	 	block number of next FileSysHeaderBlock
20/14 	ulong 	1 	flags
24/18 	ulong 	2 	RESERVED
32/20 	char 	4 	DosType 	'DOS' and OFS/FFS DIRCACHE INTL bits
36/24 	ulong 	1 	Version 	filesystem version 0x0027001b == 39.27
40/28 	ulong 	1 	PatchFlags 	bits set for those of the following
					that need to be substituted into a
 					standard device node for this 
					filesystem : e.g. 0x180 to substitute
					SegList and GlobalVec
	Device node
44/2c 	ulong 	1 	Type 		device node type == 0
48/30 	ulong 	1 	Task 		standard DOS "task" field == 0
52/34	ulong 	1 	Lock 		not used == 0
56/38 	ulong 	1 	Handler 	filename to loadseg == 0
60/3c 	ulong 	1 	StackSize 	stacksize to use when starting task ==0
64/40 	ulong 	1 	Priority 	task priority when starting task == 0
68/44 	ulong 	1 	Startup 	startup msg == 0
72/48 	ulong 	1 	SegListBlock 	first of linked list of LoadSegBlocks :
					note that this entry requires some
 					processing before substitution
76/4c 	ulong 	1 	GlobalVec 	BCPL global vector when starting task =-1
80/50 	ulong 	23 	RESERVED 	by PatchFlags
172/ac 	ulong 	21 	RESERVED
-------------------------------------------------------------------------------
This block contains information on how to lauch the task which will manage the filesystem. You don't need it to reach partitions.

6.5 What are LSEG blocks ?

* LoadSeg block (BSIZE bytes) first in FileSysHeaderBlock 'SegListBlocks' field
-------------------------------------------------------------------------------
0/0 	char 	4 	id 		'LSEG'
4/4 	long 	* 	size in longs 	size of this checksummed structure
					* size = BSIZE/4
8/8 	long 	1 	checksum 	classic checksum
12/c 	long 	1 	hostID 		SCSI Target ID of host (often 7)
16/10 	long 	1 	next 		block number of the next LoadSegBlock
                                        (-1 for the last)
20/14 	uchar 	* 	LoadData[] 	code stored like an executable, with
					relocation hunks
					* size = ((BSIZE/4) - 5)
-------------------------------------------------------------------------------
This block contains the code of the filesystem. It isn't needed to reach partitions.

7. The Hard file : a big floppy dump file

A hardfile is a file which contains an Amiga volume.

It is created with WinUAE (http://www.codepoet.com/UAE/), and not the Amiga and the AmigaDOS. WinUAE is able to produce an empty file with random contents of a choosen size, often several megabytes long.
Under WinUAE, a AmigaDOS device appears, associated with the uaehf.device (UAE hardfile). You have to format it with the Workbench, and you obtain an 'hardfile'. This volume is then usable inside the emulator by AmigaDOS (it should also be mountable under Linux with the AFFS filesystem).

An hardfile is like a floppy disk dump, but bigger : it has a bootblock, a rootblock, a bitmap and perhaps dircache blocks.
The first three bytes of a hardfile is then 'D' 'O' 'S'.


8. Advanced information

Bitmap related

* Bitmap allocation starts at root block, upto highest block. The next allocated blocks are located just after boot blocks and finally the last allocated block is the sector before root block.

root -> max -> boot+1 -> root-1

-> means "followed on disk by"

If you free some blocks by deleting a file, for example, the first next used block will be the first free block encountered starting from the Rootblock. The just freed blocks will be reused. It means that when you delete a file and you want to recover it, don't write anything else on the disk.
This strategy must have been chosen to minimize fragmentation

Files related

* The order in which data and file extension blocks for a given file are written on disk differs with OFS and FFS.


OFS:
header -> data blocks -> ext. block -> data blocks -> ext. block -> data blocks

FFS:
header -> data blocks -> all ext. block -> all remaining data blocks

-> means "followed on disk by"

This difference is probably the main reason why FFS is faster then OFS.

Compare the ADF logical blocks

			    root  dir 	fileh 	hlink 	slink 	fext	data 	dirc
----------------------------------------------------------------------------------------
        0/ 0x00 1st_type    2 	  2 	2 	2 	2 	16	8 	33
        4/ 0x04 header_key  / 	  x 	x 	x 	x 	x 	x 	x
        8/ 0x08  	    / 	  / 	nb_blo	/ 	/ 	nb_blo 	block# 	PARENT
       12/ 0x0c table_size  72 	  / 	/ 	/ 	/ 	/ 	nb_data nb_rec
       16/ 0x10 list 	    / 	  / 	data#1 	/ 	/ 	/ 	next 	next
       20/ 0x14 chksum 	    x 	  x 	x 	x 	x 	x 	x 	x
       24/ 0x18 table 	    ht 	  ht 	blocks 	/ 	/ 	blocks  data 	records
    ...
BSIZE-184/-0xb8	comment_len /	  x 	x 	/ 	/ 	/ 	/ 	/
BSIZE-183/-0xb7 comment     /	  x 	x 	/ 	/ 	/ 	/ 	/
    ...
BSIZE- 92/-0x5c	days 	    x	  x	x 	x 	x 	/ 	/ 	/
BSIZE- 88/-0x58 mins 	    x	  x	x 	x 	x 	/ 	/ 	/
BSIZE- 84/-0x54 ticks 	    x	  x	x 	x 	x 	/ 	/ 	/
BSIZE- 80/-0x50	name_len    x 	  x 	x 	x 	x 	/ 	/ 	/
BSIZE- 79/-0x4f name 	    x 	  x 	x 	x 	x 	/ 	/ 	/
    ...
BSIZE- 16/-0x10	hash_chain  / 	  x 	x 			/ 	/ 	/
BSIZE- 12/-0x0c	parent	    / 	  x 	x 	x 	x 	fhdr 	/ 	/
BSIZE-  8/-0x08	extension   cache cache	fext    / 	/ 	next 	/ 	/
BSIZE-  4/-0x04	2nd_type    1 	  2 	-3	-4/4 	3 	-3 	/ 	/
----------------------------------------------------------------------------------------

type of blocks :
 root=rootblock,  dir=directory,  fileh=file header,  fext=file extension,
 hlink=hard link,  slink=soft link,  dirc=directory cache,  data=OFS data.

special values :
 /=unused
 x=used
 next=next block of same type

How to rename an entry ?

  1. Compute the new hashvalue
  2. Move the first sector pointer from the old hashvalue index to the new one
  3. Change the name in the directory or file header block

9. References and links

* ASM Sources:

* On-Line material :
* Books :

The AmigaDOS reference manual probably contains a lot of information about Amiga file systems, but i don't own it (Addison Wesley). The most detailed information about AmigaDOS can be found in Ralph Babel's "Amiga Guru Book".


10. C routines : the ADF library

The ADFlib is a portable C library designed to manage Amiga formatted devices like harddisks and ZIP disks, or dump files of this kind of media via the .ADF format.

The API permits you to :

A callbacks system makes it easy to write a real device driver for any platform. The ADFOpus ( http://www.chromerhino.demon.co.uk/adfopus) application (a useful Windows Explorer like for ADF files and devices), written by Dan Sutherland is able to access from Win95/98/NT an 2.5 inches harddisk formatted under AmigaDOS.

See the 1.2 section to see how to obtain the package.


11. Other Amiga FileSystems


The .ADF format FAQ ends here !