Friday, 19 April 2013

Linux : Hard disk layout

In this article, learn to design a disk partitioning layout for a Linux system. Learn to:

  • Allocate filesystems and swap space to separate partitions or disks
  • Tailor the design to the intended use of the system
  • Ensure the system can be booted
This article helps you prepare for Objective 102.1 in Topic 102 of the Linux Professional Institute's Junior Level Administration (LPIC-1) exam 101. The objective has a weight of 2.

Note: This article deals mostly with planning the layout.

Prerequisites

To get the most from the articles in this series, you should have a basic knowledge of Linux and a working Linux system on which you can practice the commands covered in this article. Sometimes different versions of a program will format output differently, so your results may not always look exactly like the listings and figures shown here.

Filesystem overview

A Linux filesystem contains files that are arranged on a disk or otherblock storage device in directories. As with many other systems, directories on a Linux system may contain other directories calledsubdirectories. Unlike a system such as Microsoft® Windows® with a concept of separate file systems on different drive letters (A:, C:, etc.), a Linux filesystem is a single tree with the / directory as its rootdirectory.

You might wonder why disk layout is important if the filesystem is just one big tree. Well, what really happens is that each block device, such as a hard drive partition, CD-ROM, or floppy disk, actually has a filesystem on it. You create the single tree view of the filesystem by mounting the filesystems on different devices at a point in the tree called a mount point.

Usually, the kernel starts this mount process by mounting the filesystem on some hard drive partition as /. You may mount other hard drive partitions as /boot, /tmp, or /home. You may mount the filesystem on a floppy drive as /mnt/floppy, and the filesystem on a CD-ROM as /media/cdrom1, for example. You may also mount files from other systems using a networked filesystem such as NFS. There are other types of file mounts, but this gives you an idea of the process. While the mount process actually mounts the filesystem on some device, it is common to simply say that you "mount the device," which is understood to mean "mount the filesystem on the device."

Now, suppose you have just mounted the root file system (/) and you want to mount a CD-ROM, /dev/sr0, at the mount point /media/cdrom. The mount point must exist before you mount the CD-ROM over it. When you mount the CD-ROM, the files and subdirectories on the CD-ROM become the files and subdirectories in and below /media/cdrom. Any files or subdirectories that were already in /media/cdrom are no longer visible, although they still exist on the block device that contained the mount point /media/cdrom. If the CD-ROM is unmounted, then the original files and subdirectories become visible again. You should avoid this problem by not placing other files in a directory intended for use as a mount point.

Table 1 shows the shows the directories required in / by the Filesystem Hierarchy Standard (for more detail on FHS.

Table 1. FHS directories in /
DirectoryDescription
binEssential command binaries
bootStatic files of the boot loader
devDevice files
etcHost-specific system configuration
libEssential shared libraries and kernel modules
mediaMount point for removable media
mntMount point for mounting a filesystem temporarily
optAdd-on application software packages
sbinEssential system binaries
srvData for services provided by this system
tmpTemporary files
usrSecondary hierarchy
varVariable data

Partitions

The first SCSI drive is usually /dev/sda. On an older Linux system, the first IDE hard drive is /dev/hda. With the advent of serially attached (SATA) IDE drives, a mixed PATA/SATA system would sometimes use /dev/hda for the first PATA drive and /devsda for the first SATA drive. On newer systems, all IDE drives are named /dev/sda, /dev/sdb, and so on. The change of name for IDE drives is a result of the hotplug system, which initially supported USB drives. Hotplug allows you to plug in new devices and use them immediately, and is now used for all devices whether they are built into the system or plugged later into a running system using USB or Firewire (IEEE 1394) or potentially other types of connection.

Traditionally, a hard drive is formatted into 512 byte sectors. All the sectors on a disk platter that can be read without moving the head constitute a track. Disks usually have more than one platter. The collection of tracks on the various platters that can be read without moving the head is called a cylinder. The geometry of a hard drive is expressed in cylinders, tracks (or heads)per cylinder, and sectors/track. At the time of this writing, drive manufacturers are starting to introduce disks with 4K sectors. If a filesystem still assumes 512-byte sectors, you may lose performance if a partition does not start at a sector that is on a 4K boundary.

Limitations on the possible sizes for cylinders, heads, and sectors used with DOS operating systems on PC systems resulted in BIOS translating geometry values so that larger hard drives could be supported. Eventually, even these methods were insufficient. More recent developments in disk drive technology have led to logical block addressing (LBA), so the CHS geometry measurements are less important, and the reported geometry on a modern disk may bear little or no relation to the real physical sector layout. The larger disks in use today have forced an extension to LBA known as LBA48, which reserves up to 48 bits for sector numbers.

The space on a hard drive is divided (or partitioned) into partitions. Partitions cannot overlap; space that is not allocated to a partition is called free space. The partitions have names like /dev/hda1, /dev/hda2, /dev/hda3, /dev/sda1, and so on. IDE drives are limited to 63 partitions on systems that do not use hotplug support for IDE drives. SCSI drives, USB drives, and IDE drives supported by hotplug are limited to 15 partitions. A partition is often allocated as an integral number of cylinders (based on the possibly inaccurate notion of a cylinder).

If two different partitioning programs have different understandings of the nominal disk geometry, it is possible for one partitioning program to report an error or possible problem with partitions created by another partitioning program. You may also see this kind of problem if a disk is moved from one system to another, particularly if the BIOS capabilities are different.

You can see the nominal geometry on a Linux system using either the parted or fdisk tools. Older Linux systems also reported geometry in the /proc filesystem, in a file such as /proc/ide/hda/geometry, a file that may not be present on newer systems. Listing 1 shows how to use the fdisk command to display the partitions and geometry of an IDE hard drive with SATA attachment. The -v parameter of fdisk shows the version. You will need to be root or have root authority via sudo, as shown here, to display or manipulate the partition table.

Listing 1. Hard disk geometry

ian@attic4:~$ fdisk -v
fdisk (util-linux-ng 2.16)
ian@attic4:~$ sudo fdisk /dev/sdb
[sudo] password for ian: 

The number of cylinders for this disk is set to 30401.
There is nothing wrong with that, but this is larger than 1024,
and could in certain setups cause problems with:
1) software that runs at boot time (e.g., old versions of LILO)
2) booting and partitioning software from other OSs
   (e.g., DOS FDISK, OS/2 FDISK)

Command (m for help): p

Disk /dev/sdb: 250.1 GB, 250059350016 bytes
255 heads, 63 sectors/track, 30401 cylinders
Units = cylinders of 16065 * 512 = 8225280 bytes
Disk identifier: 0x000404d6

   Device Boot      Start         End      Blocks   Id  System
/dev/sdb1               1          25      200781   83  Linux
/dev/sdb2              26       12965   103940550   83  Linux
/dev/sdb3           12966       30401   140054670   83  Linux
In Listing 1, note that fdisk prints a warning about the nominal position for the end of cylinder 1024. Cylinder 1024 is important in some older systems where the BIOS is only able to boot partitions that are completely located within the first 1024 cylinders of a disk. This is most likely to occur in a BIOS that does not have LBA support, or some older boot managers. It is not usually a problem in modern machines, although you should be aware that the limit may exist.

You can use fdisk to display units in sectors, using the -u option, or you can use the u subcommand in interactive mode to toggle between sectors and cylinders. The parted command supports several different units. Listing 2 illustrates the use of different units in parted for the same disk as used in Listing 1.

Listing 2. Using different units with parted

ian@attic4:~$ sudo parted /dev/sdb
[sudo] password for ian: 
GNU Parted 1.8.8.1.159-1e0e
Using /dev/sdb
Welcome to GNU Parted! Type 'help' to view a list of commands.
(parted) help u                                                          
  unit UNIT                                set the default unit to UNIT

 UNIT is one of: s, B, kB, MB, GB, TB, compact, cyl, chs, %, kiB, MiB,
        GiB, TiB
(parted) p                                                                
Model: ATA HDT722525DLA380 (scsi)
Disk /dev/sdb: 250GB
Sector size (logical/physical): 512B/512B
Partition Table: msdos

Number  Start   End    Size   Type     File system  Flags
 1      32.3kB  206MB  206MB  primary  ext3
 2      206MB   107GB  106GB  primary  ext4
 3      107GB   250GB  143GB  primary  ext3

(parted) u s                                                              
(parted) p                                                                
Model: ATA HDT722525DLA380 (scsi)
Disk /dev/sdb: 488397168s
Sector size (logical/physical): 512B/512B
Partition Table: msdos

Number  Start       End         Size        Type     File system  Flags
 1      63s         401624s     401562s     primary  ext3
 2      401625s     208282724s  207881100s  primary  ext4
 3      208282725s  488392064s  280109340s  primary  ext3

(parted) u chs                                                            
(parted) p                                                                
Model: ATA HDT722525DLA380 (scsi)
Disk /dev/sdb: 30401,80,62
Sector size (logical/physical): 512B/512B
BIOS cylinder,head,sector geometry: 30401,255,63.  Each cylinder is 8225kB.
Partition Table: msdos

Number  Start      End           Type     File system  Flags
 1      0,1,0      24,254,62     primary  ext3
 2      25,0,0     12964,254,62  primary  ext4
 3      12965,0,0  30400,254,62  primary  ext3

(parted)
Note that the apparent discrepancy between the starting cylinder and ending cylinders shown by parted and fdisk output is due to the fact that parted starts counting cylinders at 0, while fdisk starts counting them at 1. Listing 3 shows that fdiskdoes have the same starting and ending sector as parted.

Listing 3. Checking start and end sector numbers

ian@attic4:~$ sudo fdisk -ul /dev/sdb

Disk /dev/sdb: 250.1 GB, 250059350016 bytes
255 heads, 63 sectors/track, 30401 cylinders, total 488397168 sectors
Units = sectors of 1 * 512 = 512 bytes
Disk identifier: 0x000404d6

   Device Boot      Start         End      Blocks   Id  System
/dev/sdb1              63      401624      200781   83  Linux
/dev/sdb2          401625   208282724   103940550   83  Linux
/dev/sdb3       208282725   488392064   140054670   83  Linux
ian@attic4:~$ echo $(( 208282725 / 255 / 63 ))
12965
Partition

There are three types of partition on IDE drives: primarylogical, and extended. The partition table is located in the master boot record (MBR) of a disk. The MBR is the first sector on the disk, so the partition table is not a very large part of it. This limits the number of primary partitions on a disk to four. When more than four partitions are required, as is often the case, one of the primary partitions must instead become an extended partition.

An extended partition is simply a container for one, or usually more, logical partitions. This partitioning scheme was originally used with MS DOS and PC DOS and permits PC disks to be used by DOS, Windows, or Linux systems. A disk may contain only one extended partition. Data is stored in the logical partitions within the extended partition. You cannot store data in an extended partition without first creating a logical partition within it.

Linux numbers primary or extended partitions as 1 through 4, so dev/sda may have four primary partitions, /dev/sda1, /dev/sda2, /dev/sda3, and /dev/sda4. Or it may have a single primary partition /dev/sda1 and an extended partition /dev/sda2. If logical partitions are defined, they are numbered starting at 5, so the first logical partition on /dev/sda will be /dev/sda5, even if there are no primary partitions and one extended partition (/dev/sda1) on the disk. So if you want more than four partitions on an IDE drive, you will lose one partition number to the extended partition. Although the theoretical maximum number of partitions on an IDE drive is now limited to 15 for kernels with hotplug, you may or may not be able to create the last few. Be careful to check that everything can work if you plan on using more than 12 partitions on a drive.

The disk used in the earlier examples has three primary partitions, all formatted for Linux use. Two use the ext3 filesystem, while the other one uses ext4. Listing 4 shows the output from the parted command p for an internal drive with primary, extended, and logical partitions on a Ubuntu 9.10 system and for a USB drive attached to a Fedora 12 system. Note the different filesystem types. Note also that you can specify one or more parted commands on the command line to avoid interactive mode.

Listing 4. Displaying the partition table with parted

                    
ian@attic4:~$ sudo parted /dev/sda u s p
[sudo] password for ian: 
Model: ATA WDC WD6401AALS-0 (scsi)
Disk /dev/sda: 1250263728s
Sector size (logical/physical): 512B/512B
Partition Table: msdos

Number  Start       End          Size         Type      File system     Flags
 1      63s         2040254s     2040192s     primary   ext3
 2      2040255s    22523129s    20482875s    primary   linux-swap(v1)
 4      22523130s   1250258624s  1227735495s  extended                  boot
 5      22523193s   167397299s   144874107s   logical   ext3
 6      167397363s  310761359s   143363997s   logical   ext3
 7      310761423s  455442749s   144681327s   logical   ext3
 8      455442813s  600092009s   144649197s   logical   ext3
[root@echidna ~]# parted /dev/sdc p
Model: WD My Book (scsi)
Disk /dev/sdc: 750GB
Sector size (logical/physical): 512B/512B
Partition Table: msdos

Number  Start   End    Size    Type      File system  Flags
 1      32.3kB  135GB  135GB   primary   fat32        lba
 2      135GB   750GB  616GB   extended
 5      135GB   292GB  157GB   logical   ext3
 6      292GB   479GB  187GB   logical   ext3
 7      479GB   555GB  76.5GB  logical   ext3
 8      555GB   750GB  195GB   logical   ext3
Allocating disk space
As mentioned earlier, a Linux filesystem is a single large tree rooted at /. It is fairly obvious why data on floppy disks or CD-ROMs must be mounted, but perhaps less obvious why you should consider separating data that is stored on hard drives. Some good reasons for separating filesystems include:
  • Boot files. Some files must be accessible to the BIOS or boot loader at boot time.
  • Multiple hard drives. Typically each hard drive will be divided into one or more partitions, each with a filesystem that must be mounted somewhere in the filesystem tree.
  • Shareable files. Several system images may share static files such as executable program files. Dynamic files such as user home directories or mail spool files may also be shared, so that users can log in to any one of several machines on a network and still use the same home directory and mail system.
  • Potential overflow. If a filesystem might fill to 100 percent of its capacity, it is usually a good idea to separate this from files that are needed to run the system.
  • Quotas. Quotas limit the amount of space that users or groups can take on a filesystem.
  • Read-only mounting. Before the advent of journaling filesystems, recovery of a filesystem after a system crash often took a long time. Therefore, filesystems that seldom changed (such as a directory of executable programs) could be mounted read-only so as to not waste time for checking it after a system crash.
In addition to the filesystem use covered so far, you also need to consider allocating swap space on disk. For a Linux system, this is usually one, or possibly multiple, dedicated partitions.
Making choices
Let's assume you are setting up a system that has at least one hard drive, and you want to boot from the hard drive. (This article does not cover setup for a diskless workstation that is booted over a LAN or considerations for using a live CD or DVD Linux system.) Although it may be possible to change partition sizes later, this usually takes some effort, so making good choices up front is important. Let's get started.

Your first consideration is to ensure that your system will be bootable. Some older systems have a limitation that the BIOS can boot only from a partition that is wholly located within the first 1024 cylinders of disk. If you have such a system, then you mustcreate a partition that will eventually be mounted as /boot that will hold the key files needed to boot the system. Once these have been loaded, the Linux system will take over operation of the disk, and the 1024 cylinder limit will not affect further operation of the system. If you need to create a partition for /boot, approximately 100 megabytes (MB) is usually sufficient.

Your next consideration is likely to be the amount of required swap space. With current memory prices, swap space represents a very slow secondary memory. A once common rule of thumb was to create swap space equivalent to the amount of real RAM. Today, you might want to configure one or two times real RAM for a workstation so that you can use several large programs without running out of RAM. Even if it is slow to switch between them, you are probably working in only one or two at any given time.

A large swap space is also advisable for a system with very small memory. For a server, you might want to use a swap space of about half of your RAM, unless you are running an application that recommends a different value. In any event, you should monitor server memory usage so that you can add real RAM or distribute the workload across additional servers if needed. Too much swapping is seldom good on a server. It is possible to use a swap file, but a dedicated partition performs better.

Now we come to a point of divergence. Use of a personal workstation tends to be much less predictable than use of a server. My preference, particularly for new users, is to allocate most of the standard directories (/usr, /opt, /var, etc.) into a single large partition. This is especially useful for new users who may not have a clear idea of what will be installed down the line. A workstation running a graphical desktop and a reasonable number of development tools will likely require 5 or more gigabytes of disk space plus space for user needs. Some larger development tools may require several gigabytes each. I usually allocate somewhere between 40 GB and 60 GB per operating system, and I leave the rest of my disk free to load other distributions.

Server workloads will be more stable, and running out of space in a particular filesystem is likely to be more catastrophic. So, for them, you will generally create multiple partitions, spread across multiple disks, possibly using hardware or software RAID or logical volume groups.

You will also want to consider the workload on a particular filesystem and whether the filesystem is shared among several systems or used by just one system. You may use a combination of experience, capacity planning tools, and estimated growth to determine the best allocations for your system.

Regardless of whether you are configuring a workstation or a server, you will have certain files that are unique for each system located on the local drive. Typically, these include /etc for system parameters, /boot for files needed during boot, /sbin for files needed for booting or system recovery, /root for the root user's home directory, /var/lock for lock files, /var/run for running system information, and /var/log for log files for this system. Other filesystems, such as /home for user home directories, /usr, /opt, /var/mail, or /var/spool/news may be on separate partitions, or network mounted, according to your installation needs and preferences.

0 blogger-disqus:

Post a comment