Lecture 024

File System Layers

Layers:

• device drivers (disk sector)

• block I/O (partition, block)

• file I/O (file, block)

• file system (directory)

• namespace (mounting file systems)

Storage Organization

Multiple disks -glue-> volumes Volumes -split-> partitions

Partitions Record:

• Master Boot Record (MBR): 1st sector on disk, 512 bytes, contains partition table and boot code, only support at most 4 primary partitions

• GUID Partition Table (GPT): support up to 128 partitions, each partition entry is 128 bytes, with name support disks larger than 2TB, use CRC32 for integrity check

Partition: (start on byte 63 - because first 62 bytes are reserved for MBR for booting)

• Paging Area (swap space): fake memory managed by OS

• File System: persistent storage for files, managed by OS

Disk, Memory Structure

Disk Data structure: (persistent data structure)

• partition table

• boot area (first block/track/cylinder on disk)

• file system control block

  • block number, metadata
  • (or superblock in ext4)

• free space map

• inode (file control block)

  • a map from file block number to disk block number
  • ownership, permissions, timestamps

• directory: naming map from string to inode number

Memory Data structure: (non-persistent data structure)

• Cache: partition table, directory name lookup cache

• Open file table: file number, cursor position, allowed operations, etc. (both system-wide and per-process)

Directories

Directory: hash based map (name -> inode number)

Block Allocation

Methods:

• Contiguous: each file has pre-defined length (bad, not allow file growth)

• Linked List: each files are multiple blocks chained together, dynamically allocate blocks (bad, hard to access file blocks at random, need to seek for each block)

• FAT: linked list, but store next pointer in a separate linked list (so no link address in file block)

• Indexed: treat file like virtual address, and use page table (except we don't need full table, we make 1st layer dynamically sized using linked list)

Issues of FAT:

• Damage: if FAT is damaged, the whole file system is damaged (Solution: have a backup copy of FAT)

• Performance:

  • 2 seeks per block read/write, one data, one data structure, (Solution: cache FAT datastructure)
  • linear time travel in linked list
  • if a file is large and fragmented, need to seek back and forth

Cache Tricks

Tricks:

• you can predict what CPU request next (pre-fetching)

• only give a portion of the cache to large file reads (which most likely only read once)