Lecture 28

Filesystems and Disks

Mounting and unmounting

Mounting disk associates a disk's contents with a logical mount point in the logical filesystem.
Not only is data made available, but it is sewn into the logical filesystem.
The System Administrator can choose the appopriate hierarchical divisions in the filesystems mounted on a machine.
Special naming conventions are used for disks:
- AIX: /dev/hdiskn
- Linux: /dev/dsln
- Solaris: /dev/[r]dsk/cntndnsn
Mount disks using mount block-special-file mount-point
Mount point must be a directory that exists on your machine. It needn't contain anything.
mount /dev/dskc1d1s8 /users2
Unmount a filesystem using umount name
Name can be either the block special filename or the directory name.
An active filesystem (one for which some processes have open files cannot be unmounted. The program fuser can be used to find who (which user processes) have open files on a filesystem.
fuser -u /dev/dsk/c1t1d0
With the -k option, it will kill all processes using the filesystem. (This is not always a good idea.)
File System Configuration File
- Most Systems: /etc/fstab
  block-special-file mount-loc type opts dump-freq pass-number
  - block-special-file: name of special file for the filesystem.
  - mount-loc: logical mmount point
  - type: kind of partition (normal one for Linux is ext2 for SCSI, nfs for NFS mounted disks, swap for swap partitions.
  - opts: rw, ro, suid, nosuid, noauto, usrquota,grpquota of swap options if a swap fs
  - dump-freq: how many days between back ups
  - pass-number: what order to fsck.
- Solaris: /etc/vfstab blk-spfile char-spfile mount-loc type fsck-pass automount? opts
  - blk-spfile: samme
  - char-spfile: character special file to use
  - mount-loc: same
  - type: usual ufs or s5
  - fsck-pass
  - automount?
  - opts: rw, ro, rq (rw with quota), suid, nosuid
- AIX: /etc/filesystems and /etc/swapspaces
```
/:
  dev     = /dev/us00
  vfs     = jfs
  log     = /dev/loglv01
  mount   = true
  check   = 2
  options = rw
  quota   = userquota
```
  - dev: special file
  - mount: whether to mount automatically (true/false/automatic)
  - options: rw, ro
  - vfs: jfs = journalled file system, nfs, etc.
  - quota: specifies quotas in effect
  - log: where to put log info
fsck finds following sorts of things:
- block belonging to several files
- blocks marked free but in use
- blocks marked used but free
- incorrect link count in inodes
- inconsisten inode size vals
- illegal blocks (filesystem tables) in files
- inconsistent data in filesystem tables
- lost files
- illegal or unallocated inode numbers in dirs
Can't find corrupt data within a file.
If use fsck to fix a filesystem and then reboot, make sure you reboot -n! or do whatever is necessary to reboot without running fsck!

RAID: Redundant Arrays of Independent/Inexpensive Disks

RAID 0: Disk striping
- + Easy to do
- - Not really fault tolerant.
RAID 1: Disk mirroring
- + Redundancy
- - High ECC overhead
RAID 2: Disk striping, data split by single bytes, with separate disks for Hamming Code ECC data bits.
- + On the fly data error correction
- - Lots of disk space used for ECC
RAID 3: Disk striping with parity disk, data split across disks by single bytes
- + High read/write transfer rate or big files
- - Transaction rate equal to single drive (if synchronized) at best
RAID 4: Like RAID 3, except data is split across disks in larger blocks (2K is used by NetApp)
- + High read data transaction rate
- - Lowest write transaction rate
RAID 5: Like 3 except the parity disk is split across multiple component disks.
- + highest read transaction rate
- - difficult to rebuild in the event of single disk failure
RAID 6, RAID 7, RAID 53: find out on your own.