Apple's Transition from Apple Partition Map to the GUID Partition Table
Volume Number: 23 (2007)
Issue Number: 03
Column Tag: System Technologies
Apple's Transition from Apple Partition Map to the GUID Partition Table
By Criss Myers
In January 2006, Apple moved from PowerPC processors to Intel processors. This brought with it a number of changes to the way that a Macintosh works. These changes involved the move from the IBM manufactured PowerPC RISC chip to the Intel Core Duo CISC chip, which meant a move for Apple to an X86 architecture. At the same time, Apple chose to make a few other changes. One was the move from Open Firmware, the open source version of Sun Microsystems boot firmware, to Intel's Exstensible Firmware Interface (EFI), previously used on Intels Itanium hardware. Another major change involved a change to the partitioning structure of the disk. Apple moved from its long standing Apple Partition Map (APM) to the Intel-created GUID Partition table (GPT) which is part of the EFI specification. This article will examine these changes to the partitioning structure and assess the impact that this will have on Apple.
APM vs GPT
Apple Partition Map (APM)
Apple's Partition Map was created in 1987 by Apple for the Macintosh II and has remained relatively unchanged since then. APM is a very robust and simple partition structure for partitioning a drive which can then be formated as either HFS or HFS+ filesystem. Unlike DOS partitions, APM does not contain any boot code; the computer's Open Firmware has the code to read this structure. The APM is read by Open Firmware, then the HFS+ file system is read and finally, the Operating System is booted. Open Firmware can only boot Operating Systems from HFS filesystems. Also unlike DOS partitioning, APM can describe as many partitions as required, whereas DOS is limited to 4 primary partitions. This type of partition structure is used on desktop Macs and portables as well as iPods and apples disk images, it is also used on dual formated CD's.
Hard disks are divided into blocks, or sectors, and each block typically consists of 512 bytes. On a Mac, hard disk block 0 is the driver descriptor block and contains the driver description record.
Block 1, actually the second block, contains the Apple Partition Map. Unlike other partition structures, it is defined as a partition in its own right. The APM describes the partition sturcture of the drive and contains partition entries for each partition including one for itself. These partition entries can be in any particular order and do not need to correspond to the physical organisation of the partitions. The number of partition entries is not restricted, but the map cannot be enlarged later. Once the drive has been partitioned and the APM created, the remaining space on the drive belongs to the rest of the partitions, and there is therefore no free space to enlarge the APM at a later time. All space on the drive must belong to a partition, as APM does not support free space. The APM is read by Open Firmware and all partition entries are read in order.
Partition Map Entries
The APM is layed out in 512 byte data structures, each 512 data structure is a map entry and defines each partition, these entries are laid out in consecutive sectors. Each entry describes how many partitions exist, the starting sector of the partition, its size and type, as well as volume name, the location of the data area and location of any boot code. Since each 512byte data structure lists the number of total partitions on the disk, from this the Open Firmware knows how many sectors to read in order for it to read the whole partition map.
Figure 1 shows the typical layout of an APM, the Partition Map being the first entry and also a partition, the second entry is partition 1, etc.
Fig 1. Typical Partition Map
Figure 2 shows the entry details for a single 512byte Partition Map Entry
Fig 2. Partition Map Entry
Fig 3. Status value for Apple partitions.
EFI, Extensible Firmware Inrerface, was created by Intel in an effort to introduce a modern firmware for generic PC hardware. GPT is the partitoning scheme included in this firmware. The traditional Master Boot Record used by most legacy PC BIOS systems has severe limitations and is outdated. Interestingly, due to the way the PC BIOS "grew up", it also has no official specifications. GPT was first used on Intel's Itanium hardware along with EFI.
GPT is defined by a formal standard as part of EFI. EFI is defined by Intel's "Extensible Firmware Interface Specification", version 1.1, or by UEFI, "Unified Extensible Firmware Interface Specification", version 2.0, to which many companies belong, called the Unifed EFI Forum. Apple is a member of this group.
It is very similar to Apple's APM scheme in that it also uses Logical Block Addressing (LBA) with each block being 512bytes. GPT is a partition scheme which defines the partitioning but unlike APM (which is defined as a partition itself in block 1) GPT encompasses the whole drive with a secondary partition table at the end of the drive. This acts as a backup for the primary data.
Like APM, GPT starts at block 1, but unlike APM which uses block 0 as the drive descriptor block, GPT uses block 0 as a Protective MBR (Master Boot Record). The purpose of the Protective MBR is protect the GPT drive from legacy PC BIOS Systems. EFI systems will read the GPT scheme on the drive, however, legacy MBR disk formatting tools cannot read GPT and would therefore potentionally write over it. The Protective MBR can be read by these disk utilites and indicates a single partition encompassing all of the GPT partitions on the drive. A system ID for this partition is set to 0xEE which indicates a GPT drive, EFI then reads this ID and thus ignores the protective MBR, however, 32bit Operating Systems will still read this and see it as a single inacessable disk.
Fig 4. Typical GPT Partition layout
The Fig 4 shows the Protective Master Boot Record (PMBR) in Logical Block Address 0 (LBA0), the Primary and Backup Partition Table at the beginning and end of the disk, consisting of the Table header and the table entries in purple.
Block 1 contains the Primary GPT Partition table header, this defines the usable blocks on the disk as well as the number and size of the partition entrys that make up the table. The header contains the disk GUID. It records its own size and location (always block1) and the secondary header, the last sector on the disk which can be used to recover a corrupt header. It also contains a CRC32 checksum for itself and the partition table (blocks 2-33) This checksum is read and verified by EFI and if found to be corrupt EFI will use the secondary header and write this header over the primary one. If both are corrupt then the drive is unreadable.
Figure 5 shows the entry details for a 512byte GPT Partition Table Header.
Fig 5. GUID Partition Table Header
Blocks 2-33 are the primary GPT entry array, these 32 sectors are reserved for GPT partitions. This equals 16,384 bytes and therefore a possible 128 partition entries of 128bytes. This GPT Entry Array is then mirrored by the secondary GPT Entry Array at the end of the disk, before the secondary partition header, which it uses to recover a damaged primary entry array. Since these blocks are reserved, this means that, unlike APM, the partition table can be altered at any time to create new partitions up to 128. An empty entry array is defined as 0, indicating that it is not in use. These arrays are continuous.
Each entry is simple and contains 128bytes, the first 16bytes designate the partition type with a partiton type GUID. The second 16bytes contains a GUID unique to the partition so that the partition can be identified. The rest of the bytes contain the partition names and attributes as well as the start and end blocks. These entries do not need to be sorted and just like the APM, they do not have to be in the same order as the partitions.
The remaining blocks on the disk are then the usable blocks for the partitions, but unlike APM, GPT does not require that all blocks must belong to a partition, therefore unpartitoned blocks are free space.
Figure 6 shows a typical Partition Enrty for a single partition.
Fig 6. GPT Partition Entry
Conclusions to Draw
As we can see, APM is restricted to 32 bits worth of blocks which will limit the total disk size to 2 terabytes. At the moment this is not really much of a problem but with disk sizes getting larger all the time this is a potential future limitation. Apple was given a choice: they could extend APM to support larger disk sizes, which would break all exisiting partitioning tools, or move to a completely different partition scheme. When Apple decided to move to an Intel based platform and switch from Open Firmware to an EFI, it gave them the perfect opportunity to use the GPT partition scheme. As we can see, this partition scheme is not unlike APM but offers more flexibility and backup recovery. However it does mean that there are compatabilty issues. An open firmware machine cannot boot from a GPT disk. Any Apple machine, be it PowerPC or Intel, running 10.4 and above can mount both a GPT disk and an APM disk. Any machine running 10.4.6 and above can also format GPT disks.
The GPT standard is very flexible and does not define exacty how to partition a disk. It merely describes the realm of what is possible. This means Apple can implement their own support for GPT, as can third party companies.
Apple's implementation of GPT
Apple has defined a set of Apple-specific partition types which are enterd in bytes 0-15 of the Partition Entry. Figure 7 describes these types.
Fig. 7 Apple Partition Types.
Apple also defines a partition policy for its disks and the way in which the partition map is layed out. Apple defines disks in 3 ways, tiny, small and big. Tiny being less than 1GB, small being between 1GB and 2GB and big being any disk larger than 2GB.
Tiny disks are created with no reserved free space and no extra partitions; the partitions are laid out as the user specifies.
Small disks are created with no extra partitions but have 128mb of free space at the end of each partition.
Big disks always have a 200mb EFI system partition called ESP as the first partition on the disk and also have the 128mb of free space after each partition (not including the ESP partition).
Each disk is then aligned to a 4KB boundary to accommodate the limitations of the HFS plus file system implimentation on Mac OS X.
Free space is left at the end of each partition to make it easier for future system software to manipulate the partition map in ways that Apple cannot anticipate at the moment.
ESP is a special partition from which EFI can load EFI boot-time device drivers. Apple's EFI supports ESP but does not use it at the moment but when you format a GPT disk this partition is created for future use.
We can see that Apple has done everything to support the compatibility of both APM and GPT and has considered future development when laying out the GPT partition on Apple disks. In general, there should be no issues with switching from APM to GPT: your new Intel Mac will be able to mount and write to any existing external drives you might have that are partitioned as APM, and as long as you have 10.4 on your PowerPC Mac you can mount and write to your GPT external drives. There should be no need to boot a GPT drive on a PowerPC.
There are, however, a few concerns when copying disks. These will mostly apply to software developers creating cloning and backup tools for GPT. In the GPT header is a GUID for the disk, and by its definition (globally unique ID) is a unique number. Apple system software relies on this to identify disks, and if a disk is cloned by a block-by-block tool, then both disks with have the exact same number. If the copy is to be a backup of the original and to be used to restore the original, then an identical GUID is required. However if the copy is to be used online as a duplicate, then the disk GUID will need to be changed so the system sees each drive as a different copy.
One last issue is with certain drive's firmware. Some drive firmware can be buggy and report errors when the very last block on a disk is accessed. For an APM disk this is not an issue as Apple's implementation of APM creates a small Apple_Free partition at the end of the disk for this reason. GPT requires the use of the last block for its backup partition table header and this cannot be avoided. Apples implementation ignores any errors given when reading the backup header. If an error occurs the disk has a valid primary but an invalid backup, such a disk is still considered readable. This disk will not be able to recover if the primary gets damaged. If the error occurs when the partition scheme is being created then the partitioning will fail and the drive cannot be partitioned by Apple software.
If you wish to run other operating systems on an Intel Mac such as Linux or Windows there are not many options, RedHat is developing a GPT based OS and Windows Vista should support GPT. Apple's move to GPT will make multi booting a Mac much easier in the future once GPT compatible operating systems are written and as more and more PC hardware manufacturers adopt EFI and GPT.
Bibliography and References
Apple Computers. Technical Note TN2166. Secrets of the GPT, Copyright 2006 Apple Computers Inc.
Brian Carrier. File System Forensic Analysis: PC Based Partitions, Chapter 3, Apple Partitions, Copyright 2005.
UEFI. UEFI Extensible Firmware Interface Specifications Version 2.0, Chapter 5 GUID Partition Table (GPT) Format, Copyright 2006.
Criss Myers is a Senior Mac IT Technician for the Faculty of Science and Technology, at the University of Central Lancashire, Preston, United Kingdom. He has been a Systems Server Administrator from the very first version of OS X Server. He Works with Macs as well as Linux, Unix and Windows.