Chapter 12: FileSystem Implementation

1
Chapter 12 File-System Implementation

• Translation of logical to physical addresses
• Logical file blocks 0, 1, , N-1
• Physical disk addresses device, cylinder,
  sector
• File control block, Directory implementation
• Allocation methods
• Free space management
• Log-structured file systems

2
File Control Blocks (FCB)

• Files are represented on disk by file control
  blocks
• FCBs contain
• Permissions
• Dates
• Owner, group, ACL
• Size
• Data blocks (e.g., pointers to these blocks)
• FCB may actually be multiple blocks on disk
• E.g., if pointers to data blocks require more
  space than can fit in a single disk block

3
Directory Representation

• Directories contain
• File/directory names pointers to FCBs
• On UNIX, directories represented just like files,
  except data blocks of a directory file contain a
  series of name/FCB pointer entries

On Linux, directory entries form a linked list.
Why?

4
Some Relevant Methods for a Directory Class

• class Directory
• Directory()
• Directory ()
• void Add(char name, Entry e)
• void Remove(char name)
• void Entry Find(char name)

5
Directory Find Method

• Want to implement method
• // retrieve directory entry for a given file name
• // e.g., method of Directory class/type
• Entry Find(char fileName)
• Linear List
• List of names of files pointers to data blocks
• Find/insert/delete Need to sequentially search
  list
• O(n) time complexity to search list n
  directory size
• Hash Table
• Given a file name, hash function gives index into
  table
• Need collision resolution upon insert/delete/find

6
Hierarchical directories

• How do you represent a hierarchical directory
  structure?
• E.g., for system call creat(/home/users/data.txt
  , 0777)
• How do you locate appropriate directory?

7
Pseudo-code

• // assume we have Directory root
• // giving us a pointer to / directory
• Entry e root-gtFind(home)
• If (e NULL) (! e-gtIsDirectory()) then
  error
• Directory d e-gtConvertToDirectory()
• e d-gtFind(users)
• If (e1 NULL) (! e-gtIsDirectory()) then
  error
• d e-gtConvertToDirectory()
• e d-gtFind(data.txt)
• If (e NULL) then
• // No file exists, so create new file
• If (e-gtIsDirectory() then error
• // otherwise, file exists, so make zero length

Note Complete implementation would iterate
through path name
8
FCB (File Control Block)Data Block Organization

• FCB representation includes data blocks of a file
• File contains blocks 0, 1, , N-1, and we need to
  map those blocks to the physical blocks of a disk
• How should this representation be organized?
• Constraints on disk block allocation
• Effectively use disk space
• Provide fast file access

9
Common methods of allocation

• Contiguous blocks for a specific file allocated
  together, in a sequence on the physical device
• FCB has pointer to first block number blocks in
  file
• Linked Use space in each data block for a
  pointer to next block indicate end block with
  special value pointer (e.g., -1)
• FCB contains pointer to first block in linked
  list
• Indexed Use separate (index) blocks of disk
  space for pointers to data blocks of the file
• FCB contains pointer(s) to index block(s)

10
Contiguous allocation

• Efficient method for direct (relative) access
• In direct access we need to be able to compute
  the disk address relative to the start of a file
• Easy here, because the blocks are in a linear
  sequence from some starting address on device
• Issues
• Internal fragmentation we may have to specify
  the maximum extent that the file will grow to
• External fragmentation
• We may have sufficient number of disk blocks to
  create a new file, but they may not be in a
  contiguous sequence

11
Linked Allocation

• Efficient method for sequential access
• File pointer is represented by disk block and
  offset
• No external fragmentation
• Issues
• Direct access is slow!
• Access to ith block requires i disk accesses
• In each block, use some space for disk block
  pointer (e.g., 32, 64 bits)
• On average ½ block internal fragmentation
• Reliability If we one block pointer is
  corrupted, we loose file after that point

12
FAT Variation on Linked Allocation

• Additional information in disk partition
• File allocation table (FAT)
• Array of blocks on a disk
• Each array entry is
• 0 (unused), or
• index of next block in particular file, or
• EOF indication for particular file
• If FAT is cached
• speeds direct access to file blocks

13
Indexed Allocation

• Positive aspects
• More efficient method for direct access
• Also good for sequential access
• No external fragmentation (all disk blocks can be
  used)
• Issues
• Overhead for pointers increases
• On average ½ of a block internal fragmentation
• What if a large file requires more space than a
  single index block?
• Linked index blocks
• Multilevel indexing

14
UNIX inode structure

• inode is the UNIX FCB (file control block)
• Variable level indexed allocation method
• Direct pointers contained in FCB
• Multi-level indirect index blocks
• Question Is the Unix inode suitable for
  sequential or direct (relative) file access?

15
Free-Space Management

• Need to keep track of disk blocks (sectors) not
  allocated to a file or directory
• Creating a file
• Allocate blocks from free space
• Deleting a file
• Return blocks to the free space
• Free space data structures
• Bit vector each bit indicates if a block is free
• Linked list link each free block together on the
  disk
• Grouping index blocks of free blocks
• Counting count of contiguous blocks available

16
Disk Device Access

• Disk devices are not truly random access
• Head(s) on disk move across disk surface
• To tracks, cylinders of disk
• Faster access to data block currently under head
• Rotational latency
• Time to rotate to a sector on current track
• Seek latency
• Time to move head from current track to another
  track
• Contiguous blocks on a track will be accessed
  faster
• Rotational seek latencies are amortized
  (divided) across the number of contiguous blocks
• UNIX makes use of this by attempting to keep
  data blocks for a file near the inode for a file

17
File System Consistency

• Metadata file system structure, not user file
  contents
• File system operations typically involve several
  metadata changes
• E.g., file create involves
• Directory modification, FCB allocation, Data
  blocks allocated, Free counts decremented
• If system crashes part way through file creation
  sequence, then
• File system metadata () will be inconsistent
• Want sequences of metadata changes to be atomic
• Sequence of changes should be all completed or
  not done at all
• E.g., want all of file create changes to occur or
  none

18
Log-Based Recovery

• A solution to this problem is to apply log-based
  techniques to file-system meta-data updates
• Write meta-data changes to a log file,
  sequentially
• Log file is a disk structure, separate from file
  system
• Each series of changes that performs some task
  (e.g., create a file) is a transaction
• Once series of changes written to log, they are
  committed
• File system is now updated from the log
• Entries for the transaction removed when update
  to file system is completed
• Upon a system crash
• Check if there unprocessed transactions in log
• Reverse (rollback) partially completed
  transactions

19
Network File Systems - 1

• NFS
• Generic abbreviation for network file systems
• Also, a specific network file system, NFS
• Textbook discusses specific NFS system
• NFS is a Unix network file system
• Goals
• Allow some degree of file sharing among a set of
  independent computers
• Should provide transparent access
• Provide sharing between any two computers
• Heterogeneous set of computers, O/S, networks

20
Network File Systems - 2

• Client-server organization
• Uses RPC based services based on XDR (external
  data representation)
• XDR converts between data representation of
  client/server computer
• Each computer runs a server
• Client operation
• Mount remote directory mounted over directory of
  local file system
• 1) name of remote (server) directory sent to
  server computer via RPC
• 2) server checks export list of export-allowable
  file systems
• 3) server returns to client a file handle
• Similar to a file descriptor, but works across
  network and allows access to file system and not
  just a single file
• File access
• Files are opened, read, written, closed etc.
  using same system calls as with non-NFS files
• File system uses client NFS file handle to
  access remote file system