Disk data layout

Before you install FreeBSD, you need to decide how you want to use the disk space available to you. If desired, FreeBSD can coexist with other operating systems on the Intel platform. In this section, we'll look at the way data is laid out on disk, and what we need to do to create FreeBSD file systems on disk.

PC BIOS and disks

The basics of disk drives are relatively straightforward: data is stored on one or more rotating disks with a magnetic coating similar in function to the coating on an audio tape. Unlike a tape, however, disk heads do not touch the surface: the rotating disk produces an air pressure against the head, which keeps it floating very close to the surface. The disk has (usually) one read/write head for each surface to transfer data to and from the system. People frequently talk about the number of heads, not the number of surfaces, though strictly speaking this is incorrect: if there are two heads per surface (to speed up access), you're still interested in the number of surfaces, not the number of heads.

While transferring data, the heads are stationary, so data is written on disks in a number of concentric circular tracks. Logically, each track is divided into a number of sectors, which nowadays almost invariably contain 512 bytes. A single positioning mechanism moves the heads from one track to another, so at any one time all the tracks under the current head position can be accessed without repositioning. This group of tracks is called a cylinder.

Since the diameter of the track differs from one track to the other, so does the storage capacity per track. Nevertheless, for the sake of simplicity, older drives, such as ST-506 (MFM and RLL) drives, had a fixed number of sectors per track. To perform a data transfer, you needed to tell the drive which cylinder, head and sector to address. This mode of addressing is thus called CHS addressing.

Modern disks have a varying number of sectors per track on different parts of the disk to optimize the storage space, and for the same reason they normally store data on the disk in much larger units than sectors. Externally, they translate the data into units of sectors, and they also optionally maintain the illusion of "tracks" and "heads," though the values have nothing to do with the internal organization of the disk. Nevertheless, BIOS setup routines still give you the option of specifying information about disk drives in terms of the numbers of cylinders, heads and sectors, and some insist on it. In reality, modern disk drives address sectors sequentially, so-called Logical Block Addressing or LBA. CHS addressing has an additional problem: various standards have limited the size of disks to 504 MB or 8 GB. We'll look at that in more detail on page 39.

SCSI drives are a different matter: the system BIOS normally doesn't know anything about them. They are always addressed in LBA mode. It's up to the host adapter to interrogate the drive and find out how much space is on it. Typically, the host adapter has a BIOS that interrogates the drive and finds its dimensions. The values it determines may not be correct: the PC BIOS 1 GB address limit (see page 39) might bite you. Check your host adapter documentation for details.

Disk partitioning

The PC BIOS divides the space on a disk into up to four partitions, headed by a partition table. For Microsoft systems, each partition may be either a primary partition that contains a file system (a "drive" in Microsoft terminology), or an extended partition that contains multiple file systems (or "logical partitions").

FreeBSD does not use the PC BIOS partition table directly. It maintains its own partitioning scheme with its own partition table. On the PC platform, it places this partition table in a single PC BIOS partition, rather in the same way that a PC BIOS extended partition contains multiple "logical partitions." It refers to PC BIOS partitions as "slices."

Partitioning offers the flexibility that other operating systems need, so it has been adopted by all operating systems that run on the PC platform. Figure 2-1 shows a disk with all four slices allocated. The Partition Table is the most important data structure. It contains information about the size, location and type of the slices (PC partitions). The PC BIOS allows one of these slices to be designated as active: at system startup time, its bootstrap record is used to start the system.

The partition table of a boot disk also contains a Master Boot Record (MBR), which is responsible for finding the correct slice and booting it. The MBR and the partition table take up the first sector on disk, and many people consider them to be the same thing. You only need an MBR on disks from which you boot the system.

Рис. 2.1. Partition table

PC usage designates at least one slice as the primary partition, the C: drive. Another slice may be designated as an extended partition that contains the other "drives" (all together in one slice).

UNIX systems have their own form of partitioning which predates the PC and is not compatible with the PC method. As a result, all versions of UNIX that can coexist with Microsoft implement their own partitioning within a single slice (PC BIOS partition). This is conceptually similar to an extended partition. FreeBSD systems define up to eight partitions per slice. They can be used for the following purposes:

• A partition can be a file system, a structure in which UNIX stores files.
• It can be used as a swap partition.FreeBSD uses virtual memory: the total addressed memory in the system can exceed the size of physical memory, so we need space on disk to store memory pages that don't fit into physical memory. Swap is a separate partition for performance reasons: you can use files for swap, like Microsoft does, but it is much less efficient.
• The partition may be used by other system components. For example, the Vinum volume manager uses special partitions as building blocks for volumes. We'll look at Vinum on page 221.
• The partition may not be a real partition at all. For example, partition c refers to the entire slice, so it overlaps all the rest. For obvious reasons, the partitions that represent file systems and swap space (a, b, and d through h) should not overlap.

Block and character devices

Traditional UNIX treats disk devices in two different ways. As we have seen, you can think of a disk as a large number of sequential blocks of data. Looking at it like this doesn't give you a file system—it's more like treating it as a tape. UNIX calls this kind of access raw access. You'll also hear the term character device.

Normally, of course, you want files on your disk: you don't care where they are, you just want to be able to open them and manipulate them. In addition, for performance reasons the system keeps recently accessed data in a buffer cache. This involves a whole lot more work than raw devices. These devices are called block devices.

By contrast with UNIX, Linux originally did not have character disk devices. Starting with Release 4.0, FreeBSD has taken the opposite approach: there are now no user-accessible block devices anymore. There are a number of reasons for this:

• Having two different names for devices is confusing. In older releases of FreeBSD, you could recognize block and character devices in an ls -l listing by the letters b and c at the beginning of the permissions. For example, in FreeBSD 3.1 you might have seen:

  $ ls -l /dev/rwd0s1a /dev/wd0s1a
  crw-r----  1  root  operator   3,  131072  Oct  31  19:59  /dev/rwd0s1a
  brw-r----  1  root  operator   0,  131072  Oct  31  19:59  /dev/wd0s1a
  

  wd is the old name for the current ad disks. The question is: when do you use which one? Even compared to UNIX System V, the rules were different.
• Nearly all access to disk goes via the file system, and user-accessible block devices add complication.
• If you write to a block device, you don't automatically write to the disk, only into buffer cache. The system decides when to write to disk. If there's a problem writing to disk, there's no way to notify the program that performed the write: it might even already have finished. You can demonstrate this very effectively by comparing the way FreeBSD and Linux write to a floppy disk. It takes 50 seconds to write a complete floppy disk—the speed is determined by the hardware, so the FreeBSD copy program finishes after 50 seconds. With Linux, though, the program runs only for a second or two, after which it finishes and you get your prompt back. In the meantime, the system flushes the data to floppy: you still need to wait a total of 50 seconds. If you remove the floppy in this time, you obviously lose data.

The removal of block devices caused significant changes to device naming. In older releases of FreeBSD, the device name was the name of the block device, and the raw (character) device had the letter r at the beginning of the name, as shown in the example above.

Let's look more carefully at how BSD names its partitions:

• Like all other devices, the device nodes, the entries that describe the devices, are stored in the directory /dev.Unlike traditional UNIX and older releases of FreeBSD, FreeBSD Release 5 includes the device file system or devfs, which creates the device nodes automatically, so you don't need to worry about creating them yourself.
• Next comes the name of the driver. As we have seen, FreeBSD has drivers for IDE and friends (ad), SCSI disks (da) and floppy disks (fd). For SCSI disks, we now have the name /dev/da.
  The original releases of FreeBSD had the abbreviation wd for IDE drives. This abbreviation arose because the most popular of the original MFM controllers were made by Western Digital. Others claim, however, that it's an abbreviation for & Winchester Disk.& SCSI disks were originally abbreviated sd The name da comes from the CAM standard and is short for direct access. BSD/OS, NetBSD and OpenBSD still use the old names.
• Next comes the unit number, generally a single digit. For example, the first SCSI disk on the system would normally be called /dev/da0.
  Generally, the numbers are assigned during the boot probes, but you can reserve numbers for SCSI disks if you want. This prevents the removal of a single disk from changing the numbers of all subsequent drives. See page 575 for more details.
• Next comes the partition information. The so-called strict slice name is specified by adding the letter s (for slice) and the slice number (1 to 4) to the disk name. BSD systems name partitions by appending the letters a to h to the disk name. Thus, the first partition of the first slice of our disk above (which would typically be a root file system) would be called /dev/da0s1a.

  Some other versions of BSD do not have the same support for slices, so they use a simpler terminology for the partition name. Instead of calling the root file system /dev/da0s1a, they refer to it as /dev/da0a. FreeBSD supports this method as well— it's called compatibility slice naming. The compatibility slice is simply the first FreeBSD slice found on the disk, and the partitions in this slice have two different names, for example /dev/ad0s1a and /dev/ad0a.
• Partition c is an exception: by convention, it represents the whole BSD disk (in this case, the slice in which FreeBSD resides).
• In addition, NetBSD reserves partition d for the entire disk, including other partitions. FreeBSD no longer assigns any special signifycance to partition d.

Figure 2-2 shows a typical layout on a system with a single SCSI disk, shared between Microsoft and FreeBSD. You'll note that partition /dev/da0s3c is missing from the FreeBSD slice, since it isn't a real partition. Like the PC BIOS partition table, the disk label contains information necessary for FreeBSD to manage the FreeBSD slice, such as the location and the lengths of the individual partitions. The bootstrap is used to load the kernel into memory. We’ll look at the boot process in more detail in "Starting and stopping the system" .

Рис. 2.2. Partition table with FreeBSD file system

Table 2-1 gives you an overview of the devices that FreeBSD defines for this disk.