File systems

Background

A filesystem is the street grid of your hard drive. It’s a map of addresses to where data is located on your drive. Your operating system uses the filesystem to store data on the drive.

There are a number of different types of filesystems. Some are better at handling many small files (ReiserFS), some are much better at large files and deleting files quickly (XFS, EXT4).

The version of Unix you use will have picked a filesystem which is used by default, on Linux this is often EXT4.

Understanding the way filesystems work is important when you have to fix issues related to disk space, performance issues with reading and writing to disk, and a host of other issues.

In this section we will discuss creating partitions, file systems on those partitions, and then mounting those file systems so your operating system can use them.

Working with disks in Linux

Disks in Linux are normally named /dev/sda, /dev/sdb, etc. If you are in a VM, they may be named /dev/xvda, /dev/xvdb, etc. The last letter (“a”, “b”, “c”..) relates to the physical hard drive in your computer. “a” is the first drive, “b” is the second.

If you have an already configured system, you will likely see entries like this:

-bash-4.1$ ls -la /dev/sd*
brw-rw---- 1 root disk 8, 0 Jul  6 16:51 /dev/sda
brw-rw---- 1 root disk 8, 1 Sep 18  2011 /dev/sda1
brw-rw---- 1 root disk 8, 2 Sep 18  2011 /dev/sda2
brw-rw---- 1 root disk 8, 3 Sep 18  2011 /dev/sda3

The number at the end of each drive maps to the partition on the drive. A partition refers to a fixed amount of space on the physical drive. Drives must have at least one partition. Depending on your specific needs, you might want more than one partition, but to start with, we’ll assume you just need one big partition.

Configuring your drive with partitions

The parted tool is for modifying and creating disk partitions and disk labels. Disk labels describe the partitioning scheme. Legacy Linux systems will have the msdos partitioning table, although newer systems with EFI use the gpt partitioning table. Drives with msdos disk labels will have a Master Boot Record, or MBR, at the beginning of the drive. This is where the bootloader is installed. GPT-labeled drives will usually have a FAT-formatted partition at the beginning of the disk for EFI programs and the bootloader.

parted has a subshell interface. It takes as an argument a device name.

root@opsschool ~# parted /dev/sda
GNU Parted 2.3
Using /dev/sda
Welcome to GNU Parted! Type 'help' to view a list of commands.
(parted) print
Model: ATA VBOX HARDDISK (scsi)
Disk /dev/sda: 42.9GB
Sector size (logical/physical): 512B/512B
Partition Table: msdos

Number  Start   End     Size    Type     File system  Flags
 1      8225kB  42.9GB  42.9GB  primary  ext4         boot

 (parted)

In this example parted ran against /dev/sda. The user then used the print command to print out information about the disk and the current partitioning scheme. The user found a 43 GB disk using the msdos partition table format. The disk had one partition which was flagged as bootable. Looking at a second example:

root@opsschool ~# parted /dev/sdb
GNU Parted 2.3
Using /dev/sdb
Welcome to GNU Parted! Type 'help' to view a list of commands.
(parted) print
Error: /dev/sdb: unrecognised disk label
(parted) mklabel msdos
(parted) print
Model: ATA VBOX HARDDISK (scsi)
Disk /dev/sdb: 8590MB
Sector size (logical/physical): 512B/512B
Partition Table: msdos

Number  Start  End  Size  Type  File system  Flags

(parted) mkpart primary 1 1G
(parted) set 1 boot on
(parted) mkpart primary 1G 5G
(parted) mkpart primary 5G 7G
(parted) mkpart primary 7G 8G
(parted) p
Model: ATA VBOX HARDDISK (scsi)
Disk /dev/sdb: 8590MB
Sector size (logical/physical): 512B/512B
Partition Table: msdos

Number  Start   End     Size    Type     File system  Flags
 1      1049kB  1000MB  999MB   primary               boot
 2      1000MB  5000MB  3999MB  primary
 3      5000MB  7000MB  2001MB  primary
 4      7000MB  8590MB  1590MB  primary

(parted)

parted failed to read the label, so the user created a new msdos disk label on the disk. After that parted was able to see that the disk was 8GB. We created a primary 1GB partition at the beginning of the disk for /boot and set the bootable flag on that partition. We created 4GB, 2GB, and 1GB partitions for root, var, and swap, respectively.

Formatting partitions with new file systems

New filesystems are created with the mkfs family of commands. There are a variety of file systems to choose from, man fs has a list of filesystems with short descriptions of each. Choosing a filesystem involves characterizing the workload of the filesystem and weighing engineering tradeoffs. On Linux systems, ext4 is a good general purpose choice. Following from the example above, we will create filesystems on each of the four partitions we created.

fdisk is another, older, tool to view and modify partitions on disk. It is limited to the msdos disk label.

root@opsschool ~# fdisk -l  /dev/sdb

Disk /dev/sdb: 8589 MB, 8589934592 bytes
255 heads, 63 sectors/track, 1044 cylinders, total 16777216 sectors
Units = sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disk identifier: 0x0004815e

   Device Boot      Start         End      Blocks   Id  System
   /dev/sdb1   *        2048     1953791      975872   83  Linux
   /dev/sdb2         1953792     9764863     3905536   83  Linux
   /dev/sdb3         9764864    13672447     1953792   83  Linux
   /dev/sdb4        13672448    16777215     1552384   83  Linux

The first partition, to contain /boot, will be ext2. Create this by running:

root@opsschool ~# mkfs.ext2 /dev/sdb1
mke2fs 1.42 (29-Nov-2011)
Filesystem label=
OS type: Linux
Block size=4096 (log=2)
Fragment size=4096 (log=2)
Stride=0 blocks, Stripe width=0 blocks
61056 inodes, 243968 blocks
12198 blocks (5.00%) reserved for the super user
First data block=0
Maximum filesystem blocks=251658240
8 block groups
32768 blocks per group, 32768 fragments per group
7632 inodes per group
Superblock backups stored on blocks:
  32768, 98304, 163840, 229376

Allocating group tables: done
Writing inode tables: done
Writing superblocks and filesystem accounting information: done

The second and third partitions, to contain / and /var, will be ext4. Create these by running:

root@opsschool:~# mkfs.ext4 /dev/sdb2
root@opsschool:~# mkfs.ext4 /dev/sdb3

The output of mkfs.ext4 is very close to the output of mkfs.ext2 and so it is omitted.

Finally, /dev/sdb4 is set aside for swap space with the command:

root@opsschool:~# mkswap /dev/sdb4
Setting up swapspace version 1, size = 1552380 KiB
no label, UUID=cc9ba6e5-372f-48f6-a4bf-83296e5c7ebe

Mounting a filesystem

Mounting a filesystem is the act of placing the root of one filesystem on a directory, or mount point, of a currently mounted filesystem. The mount command allows the user to do this manually. Typically, only the superuser can perform mounts. The root filesystem, mounted on /, is unique and it is mounted at boot. See The Boot Process.

In the following example, the filesystem of /dev/sdb1 will be mounted to the folder /mnt. That means anything inside /dev/sdb1 will become accessible under /mnt/.... (Don’t worry about the options -t and -o just yet; those will be described in more detail below).

root@opsschool # mount -t ext4 -o noatime /dev/sdb1 /mnt

It is common to specify which filesystem type is present on /dev/sdb1 and which mounting options you would like to use, but if that information is not specified then the Linux mount command is pretty good at picking sane defaults. Most administrators would have typed the following instead of the above:

root@opsschool # mount /dev/sdb1 /mnt

The filesystem type refers to the format of the data structure that is used as the filesystem on disk. Files (generally) do not care what kind of filesystem they are on. You have to concern yourself with the filesystem type only in the initial filesystem creation, automatic mounting, and performance tuning.

Example filesystem types are ext2, ext3, ext4, FAT, NTFS, HFS, JFS, XFS, ZFS, Btrfs. On Linux hosts, ext4 is a good default. For maximum compatibility with Windows and Macintosh, use FAT. See the following Wikipedia page for a more thorough comparison of file systems.

Automatic mounting using fstab

The fstab, or file system table, is the file that configures automatic mounting at boot. It tabulates block devices, mount points, type and options for each mount. The dump and pass fields control booting behavior. Dumping is the act of creating a backup of the filesystem (often to tape), and is not in common use. Pass is much more important. When the pass value is nonzero, the filesystem is analyzed early in the boot process by fsck, the file system checker, for errors. The number, fs_passno, indicates priority. The root filesystem should always be 1, other filesystems should be 2 or more. A zero value causes checks to be skipped, an option often used to accelerate the boot process. In /etc/fstab, there are a number of ways to specify the block device containing the filesystem . UUID, or universally unique identifier, is one common way in modern Linux based systems to specify a filesystem.

root@opsschool # cat /etc/fstab

# <file system> <mount point>  <type>  <options>         <dump>  <pass>
/dev/sda5         /            ext4    errors=remount-ro 0       1
/dev/sda6         none         swap    sw                0       0
/dev/sda1         /boot/efi    auto    auto              0       0

This /etc/fstab file mounts /dev/sda5 on / using the ext4 filesystem . If it encounters a filesystem corruption it will use the fsck utility early in the boot process to try to clear the problem. If the physical disk reports errors in writing while the filesystem is mounted, the OS will remount / readonly. The /dev/sda6 partition will be used as swap. The /dev/sda1 partition will be mounted on /boot/efi using autodetection, the partition will not be scanned for filesystem errors.

Automatic mounting using autofs

autofs is another way to tabulate filesystems for mounting. It is different from the /etc/fstab because the filesystems listed in auto.master are not mounted at boot. The automounter allows the system to mount filesystems on demand, then clean up those filesystems when they are no longer being used.

The auto.master file controls the autofs service:

root@opsschool # cat /etc/auto.master

/home -rw,hard,intr,nosuid,nobrowse bigserver:/exports/home/&
/stash ldap:ou=auto_stash,ou=Autofs,dc=example,dc=com -rw,hard,intr,nobrowse

In the above example, the system mounts home directories for each user from a remote NFS server. The filesystem remains unmounted until the user logs in, and is unmounted a short time after the user logs out. The automounter is triggered by an attempt to cd into /home/<key>. It will then attempt to find an NFS share on /exports/home/<key> and mount it on /home/key, then allow the cd command to return successfully. The /home example above is using the & expansion syntax.

The second line is using the LDAP syntax to look up a key under /stash/<key> in LDAP. LDAP will be covered later in the curriculum.

The auto.master file is known as auto_master on FreeBSD, Solaris, and Mac OS X.

Filesystem options

Passing options to the mount command, or inserting them into the /etc/fstab file, control how the filesystem behaves. Different filesystems at different versions support different options, but some options are ubiquitous.

async

The async option sets writing operations to the filesystem to be asynchronous. This means that the cp command will exit normally before the entire copy is done, and that the system will persist the files to the disk at some point later on. You don’t have a lot of guarantees here about when that will happen, though for a generally unloaded system you won’t have to wait long. It is also hard to tell when the filesystem is actually done with the copy. The sync utility can be used to force a filesystem sync to immediately persist to disk. The opposite of this option, the sync option is the default on most filesystems.

noatime

The noatime option tells the filesystem not to keep track of atime or access time. If you recall your inode lessons, you’ll remember that the inode keeps track of three dates: ctime (change time), mtime (modification time), and atime (access time). Under normal circumstances, whenever a user reads from a file, the operating system will write a new atime to the inode. For large groups of small files, read by a number of people, or by automated processes, this final write operation can hurt disk performance. As a result, many admins will turn off atime on filesystems to increase performance.

Note that atime is not really a security/auditing feature. Any regular user can use the touch utility on a file to set the atime to some point in the past, if they have the appropriate permissions for that file.

Also note that contrary to popular belief, ctime does not store a file’s creation time. The change time stored in ctime is when the file’s attributes or contents were changed, whereas the modification time stored in mtime only changes if the file’s contents are modified.

ro

The ro option, called ‘read-only’, tells the filesystem to not allow writes to any of the files on the filesystem. This is useful for a legitimately read-only backing store such as a CD-ROM. It is also useful for protecting the filesystem from yourself or others, such as when you are mounting a drive you pulled from a different machine. If the filesystem is mounted read-only, you can’t accidentally delete any data off of it. It is common for network shares to be mounted read-only, this way client machines can copy files from the network share, but can’t corrupt the share with write locks or deletes. Sometimes, when the operating system detects that the block device (such as a hard drive) beneath the filesystem is beginning to fail, the operating system will remount the filesystem read-only. This will cause lots of problems for your running system, but it is intended to give you the maximum amount of time to copy your important data off of the filesystem before the disks beneath it fail completely. If this happens you can usually see it by checking mount for filesystems that should be read-write mounted as read-only. You can also check dmesg for messages like:

root@opsschool # dmesg

[26729.124569] Write(10): 2a 00 03 96 5a b0 00 00 08 00
[26729.124576] end_request: I/O error, dev sda, sector 60185264
[26729.125298] Buffer I/O error on device sda2, logical block 4593494
[26729.125986] lost page write due to I/O error on sda2

These messages strongly indicate that the disk /dev/sda is dying. In some cases you can recover the filesystem with the file system checker fsck. You may also be able to force remount the filesystem read-write, as shown in the remount section below.

rw

The rw option, called ‘read-write’, tells the filesystem to allow reads and writes to all files on the filesystem.

Mounting a filesystem read-write is usually the default. There are times where you will not be able to make a read-write mount, such as when the backing physical media is fundamentally read-only, like a CD-ROM.

remount

Sometimes a filesystem will be mounted read-only, either by default, or because the operating system has remounted it ro because of disk failures. It is possible to remount the filesystem read-write with the following command:

root@opsschool ~# mount | grep boot
/dev/sda1 on /boot type ext3 (ro)
root@opsschool ~# mount -o remount,rw /boot
root@opsschool ~# mount | grep boot
/dev/sda1 on /boot type ext3 (rw)

The syntax of the remount option is -o remount,<option>.

noexec

The noexec option tells the filesystem to ignore the execute bit on a filesystem. This would never work for the root filesystem, but it is often used as a security measure on world-writeable filesystems such as /tmp . Note that there are many ways to get around this restriction.

nosuid

Like the noexec option, the nosuid option ignores any setuserid bits set on files in the filesystem. This is also used for security purposes. It is generally recommended for removable devices such as CD-ROMs, USB sticks, and network filesystems.

nobarrier

rbind

How filesystems work

Files, directories, inodes

Inodes

What they contain, how they work

The POSIX standard dictates files must have the following attributes:

  • File size in bytes.

  • A device ID.

  • User ID of file’s owner.

  • Group ID of file.

  • The file’s mode (permissions).

  • Additional system and user flags (e.g. append only or ACLs).

  • Timestamps when the inode was last modified (ctime), file content last modified/accessed (mtime/atime).

  • Link count of how many hard links point to the inode.

  • Pointers to the file’s contents.

http://en.wikipedia.org/wiki/Inode

File system layout

File system hierarchy standard is a reference on managing a Unix filesystem or directory structure.

http://www.pathname.com/fhs/

Fragmentation in unix filesystems

Filesystem objects

Filesystem contain more than just files and directories. Talk about devices (mknod), pipes (mkfifo), sockets, etc.