Hard Disk Partitions

1
Hard Disk Partitions

• Disk divided into partitions
• Definition of partition group of adjacent
  cylinders
• 1 file system may reside in a single partition
• BIOS defines boot sector to be head 0, cylinder
  0, sector 1
• Master Boot Record (MBR) used to boot computer
• Partition table
• Start and end addresses of each partition
• One partition is marked as active
• When computer boots up
• BIOS executes MBR
• MBR locates active partition and executes boot
  block

2
Hard Disk Partitions

• Each partition
• Starts with boot block even if it does not
  contain bootable OS
• Superblock key parameters about file system in
  partition such as magic identifying file system
  type, number of blocks
• Bitmap or linked-list of free blocks free space
  management
• Inodes array of these (one per file)
• Root directory of partition
• Files and directories at the end

3
Hard Disk Partitions

• Why have partition hard disks?
• To subdivide disk into a number of intended uses
• Exp. May have data partition separate from
  application program partition.
• Which partition would need to be backed up often?
• Want to have different OS or file systems using
  incompatible data structures
• OS or file systems can be tuned separately
• Safety from physical separation
• Unmount file system when it is not needed
• Backup some partitions more frequently
• Use spare partitions for backup or for storing
  state when computer hibernates

4
Booting Up DOS

• BIOS
• Reads 446 byte program from partition sector into
  memory at 0x7C00
• Checks indicator at 0xAA55
• Executes program
• Program actions
• Recognize exactly 1 of 4 partition entries as
  active
• Move itself from 0x7C00 to 0x0600 to make room
  for partition boot sector
• Load active partitions boot sector into 0x7C00
  and jump to it

5
Unixs FSCK (File System Check)

• Check for file system consistency after a crash
• Causes of inconsistencies crash during
  multi-block transaction
• Phases of FSCK
• 1) Inode checks
• Out-of-range block numbers
• Blocks claimed by more than one file
• 2) Size checks
• Number of blocks claimed is inconsistent with
  size?
• Directory size not a proper multiple
• Directory checks
• Missing . or ..
• Entry points to out-of-range inode number
• Entries do not fit correctly overlap or not all
  space is accounted for

6
Unixs FSCK (File System Check)

• Phases of FSCK
• 4) Pathname checks
• File/directory not referenced by root-anchored
  pathname
• 5) Link count checks
• Inode link count does not equal sum of those in
  directory entries
• 6) Free map checks
• Blocks claimed by both free map and files
• Unclaimed blocks not in free map
• 7) Superblock checks
• Total free block and/or inode count incorrect
• More disk blocks allocated for inodes than max
  possible inodes

7
Unixs FSCK (File System Check)

• Non-empty unreferenced blocks placed in the
  /lostfound directory with inode number as the
  name
• If possible, implement by reading superblock and
  all cylinder group headers, directories, and
  inodes, into memory then apply tests
• Above data structures are a small fraction of
  disk space (most of file system occupied by data
  blocks)

8
Unixs FSCK (File System Check)

• Phase 5
• 1 counter per file
• Scan all directories increment count for inode
  when inode is seen in directory
• Compare count vs. link field in inode
• If (links ! count) and (count gt 0) set links
  count
• If (links ! count) and (count 0) link inode
  into /lostfound directory with inode number as
  name

9
Unixs FSCK (File System Check)

• Phase 1 6
• 2 counters per block number of times in a file
  and number of times in free map
• Check counter values as follows
• in file in free map consistent?
• 1 0 ok
• 0 1 ok
• 0 0 lost block, add to free map
• gt1 1 rebuild free map, then check
• if in-file gt 1
• gt1 0 copy block redefine files

10
File System Performance Improvements

• Caching
• Similar to memory pages
• If requested block is not in cache, load into
  cache first and then access
• Use hash table for quick search hash on device
  and disk address
• Block replacement algs. similar to memory page
  replacement algs. (FIFO, second chance, LRU)
• Goals are slightly different
• Blocks essential to consistency of file system
  should be written back to disk immediately
• Blocks not needed again should be written back to
  disk and released

11
File System Performance Improvements

• Caching continued
• Unix
• Sync command forces all modified blocks to be
  written to disk
• update program runs in background after system
  starts and calls sync every few seconds
• Windows
• Uses write-through caching
• Write every modified block to disk immediately
  most programs do internal buffering, so writes
  more than just one character at a time
• Block read ahead
• read more than requested block

12
File System Performance Improvements

• Reduce disk arm motion
• Block clustering attempt to place consecutive
  blocks in a file in the same cylinder
• Place inodes in the middle of disk rather than at
  the start after inode access, arm moves at most
  half disk width to seek to file
• Log-structured file system
• Rarely used
• Structure entire disk as a log
• Pending writes are buffered in memory, collected
  into a single segment, and written to disk as
  single contiguous segment at end of log
• Segment may contain inodes, directory blocks,
  data, etc.
• Need inode map indexed by inode number to find
  inodes since they are not scattered throughout
  disk
• Map is stored on disk and cached