UNIX Case Study

history and evolution -- quite the story

  UNIX mostly written in a high-level language (C), making it easier to port

  ran originally on popular hardware, PDP-11

  many universities became interested early in UNIX, came with source code

  UCB ported UNIX to 32-bit VAX-11 hardware with virtual memory

  UCB added TCP/IP networking protocols

  Sun Microsystems started selling workstations running UNIX

  POSIX standard

  Open Software Foundation vs. UNIX International

  AT&T sold UNIX Systems Laboratories to Novell (which will? donate UNIX
  trademark to the public domain)

user view

  goals

    interactive timesharing system for programmers for program development

  layers, Fig 7-2

    hardware, kernel (system call handler, resource manager), standard
    library, utilities, users

  user interface

    through standard utilities, standard library, or system calls

    UNIX has user IDs and keeps track of who owns what files

  command interpreter

    is a user-level program running outside the kernel

    wildcard characters, redirection of input and output, pipes, background
    processes, programmable shell scripts

  files and directories, Fig 7-3

    hierarchical file system of byte-stream files, protection bits

  utilities, Fig 7-4

fundamental concepts

  processes are single threaded, system is multiprogrammed

    creation with fork, parent and child, Fig 7-6

    pipes

    signals (software interrupts), Fig 7-7

    uid, superuser (root), setuid bit

  memory management

    text, initialized data, heap (uninitialized), stack segments, Fig 7-8

      brk system call to grow or shrink heap

    shared text segments

  file system

    file descriptor returned by open, each process starts with three (stdin,
    stdout, stderr)

    path names, absolute and relative, working directory

    links, Fig 7-9

    disk partitions and mounting file systems, Fig 7-10

    locking, from one byte to whole file, shared and exclusive, blocking or
    non-blocking

  input and output

    devices are block and character special files in /dev, can open, read,
    write, close them

    sockets, endpoints of communication e.g. over network, Fig 7-12

      TCP/IP, UDP, Internet domain naming, listen, connect, read, write

system call interface, Fig 7-13

    process management

      fork, waitpid, execve, exit, sigaction, kill, alarm, pause

      example in Fig 7-14

    memory management

      brk

    files and directories

      creat, open, close, read, write, lseek, stat (Fig 7-15), mkdir,
      rmdir, link, unlink, chdir, chmod

    input and output

      get and set input and output speed and terminal attributes

implementation

  kernel

    machine-dependent part: interrupt handlers, device drivers, memory
    management

    machine-independent part: system call handling, process management,
    scheduling, pipes, signals, paging, swapping, file system, high-level
    IO

  process table: info about all processes (info is always memory resident)

    scheduling parameters, memory image pointers, signal masks, process
    state, alarms, pid, parent pid, uid, gid

  user structure: other info not needed when process swapped out (so it
                  is swapped out too)

    registers, system call in progress, file descriptor table, accounting,
    kernel stack

  CPU scheduling, Fig 7-16

    swapper: high-level scheduler swaps processes in and out

      swap out a CPU hog, a long term memory resident, or a blocked process

      swap in those swapped out for a long time

    low-level scheduler gives high priority to processes that have just
    finished waiting for a system call to be completed, e.g. disk read

  paging and page replacement, Fig 7-17

    demand paging, grab page from free list

    two-handed global clock daemon keeps free list from getting too small

    scans page frames, first hand clears usage bit, second hand puts page
    frames with clear usage bit on free list (where they can be reclaimed
    before they get to the top of the free list)

    swapper invoked if the hands move too fast or can't keep free list
    big enough

  file systems, Fig 7-18

    block 0: boot block
    block 1: superblock (# i-nodes, # disk blocks, start of free list)
    blocks 2-x: i-nodes (owner, protection, data block addresses)
    blocks x+1-y: data blocks

    directories are (name, i-node #) pairs

    file descriptor table, open file description table, open i-node table,
    Fig 7-19

    cylinder groups

      replicate superblock for reliability

      group some i-nodes and cylinders together to reduce seek time

    fragments are the last block of a file and can be smaller than usual
    file block size to reduce internal fragmentation yet have a large
    block size

  input and output, Fig 7-20

    buffer cache for file IO

    C-lists for terminal IO containing 64 characters

    line discipline produces cooked character stream