File

1
File

• File
• named unit of storage defined by OS (storage unit
  is the same regardless of underlying hardware)
• Defined as a sequence of bits, bytes, or records
• Sequential or random access to file contents
• Operations create, write-next, read-next,
  reposition next pointer, delete, truncate
• Provides access on permanent media to named files
• OS keeps track of which sectors comprise the file
• Storage media
• Hard disks
• Removable floppy, zip
• Read-only CDROM
• Linear Tape

2
File Structure

• Unstructured sequence of bytes
• Most flexible
• Any meaning can be imposed by user program
• Used by Unix and Windows
• Sequence of fixed-length records with internal
  structure
• One entire record is read/written at a time
• This structure is not currently used
• Tree of records
• Records may be of varying lengths
• Each record has key in fixed position in record
• Records sorted by key for easy search and
  retrieval
• Used with large mainframe systems

3
Aspects of a File System

• Comprised of 3 parts
• Model of operation naming, access, type system,
  API, attributes
• Storage subsystem how bits are stored on medium
  (file structure)
• Distributed access definition of consistency
  and how caching/replication work to satisfy this
  definition (handle sharing of files)

4
File System Model

• Model is what user sees in the file system
• Naming
• Syntax, such as case sensitivity
• Restrictions on name length
• Types of links
• Attributes metadata characterizing files data
• Typed vs. untyped entities
• Access mounting, sharing, memory-mapped access
• Entities in file system name space devices
  (special files), sockets, pipes, other hosts and
  domains
• System calls and semantics
• Is data really on disk when write() call returns?
• What is atomic?
• Which data are read during concurrent operations?
• Meaning of ambiguous operations seek beyond end
  of file unlinking an open file cd .. after
  following a symbolic link

5
Model Naming

• Individual names
• Length restrictions (exp. MS DOS required 8.3
  file names)
• Relative importance of extension for typing
• Windows use extensions to associate files with
  programs
• Unix use extensions just as convention and not
  enforced by OS
• Directory
• Name space organizing concept
• hard vs. symbolic link
• No links -gt tree hard -gt DAG symbolic -gt
  arbitrary graph
• In Unix, directory concept is built into process
  concept current root vs. current directory

6
Model Naming

• Symbolic links
• Indicated by file attribute
• During name interpretation, symbolic link is read
  and interposed into name
• File listing with symbolic links can cause
  infinite loops (ls lR vs ls LR)
• OS quits name interpretation after seeing more
  than a certain number of symbolic links in a
  single name interpretation operation or if it
  sees the same inode twice
• Why have symbolic links if they cause so much
  trouble?
• Its the only way to have links across file
  systems

7
Model File Attributes

• Attributes are metadata data describing the
  file data and/or how to use it
• Possible attributes type, size, access control,
  dates, lock bit, version number
• Attributes vary across file systems
• Type attribute
• Special types system file, hidden
• Information relevant to type program that
  created data record size location of key field
  length of key
• Size current, maximum allowed
• Dates creation, last read, last modified,
  version number
• Access control owner (user or program), group
  access, password, lock indicator

8
Model File Types

• General types
• Regular files contain user information
• Directories system files for maintaining
  structure of file system
• Character special files I/O for character
  devices
• Block special files I/O for block devices

9
Model Regular File Typing

• Strong typing
• OS is aware of certain types
• Optimizes access to these types
• At system call level, prevents all but certain
  uses of these types and/or certain programs from
  using the files
• Disadvantages
• There will always be new types
• Different types important to different users, so
  never have all the types everyone wants
• N types means N modules of code in the OS
  bloats OS code
• Harder to change type code inside OS
• Need escape mechanism to use typed file in a
  different way, but lose benefits of strong typing

10
Model Regular File Typing

• Unix untyped files were considered an advance
  over IBM record-oriented file types
• Still, often useful to know the type
• Prevent mistakes by naïve users
• Mac PC convention is to map one file type to
  one program
• Levels of typing
• Strong typing enforced by OS at system call level
  (all applications must obey)
• Weaker typing enforced at level of user
  visibility by launching applications based on
  type
• Weakest extension is just a convention, no OS or
  shell enforcement, though user applications may
  choose to enforce

11
Unix Executable Type

• Unix executables identified via magic number that
  is part of the data (in first few bytes of the
  file)
• Interpreted ASCII scripts have interpreter
  explicitly named at beginning of file
• The chararcter on 1st line indicates a script
• Interpreter also named on 1st line
  !/usr/bin/perl
• Some programs have their own first-byte
  convention (exp. Postscript !PS-Adobe-2.0)
• Example of Unix executable binary file