Booting Process

7.
8 THE SHELL 235
asynchronously, and the output of one process goes to the input of the other
process. The parent shell meanwhile waits for its child process (we) to exit, then
proceeds as usual: The entire command line completes when wc exits. The shell
loops and reads the next command.
7.9 SYSTEM BOOT AND THE INIT PROCESS

To initialize a system from an inactive state, an administrator goes through a
"bootstrap" sequence: The administrator "boots" the system. Boot procedures
vary according to machine type, but the goal is common to all machines: to get a
copy of the operating system into machine memory and to start executing it. This
is usually done in a series of stages; hence the name bootstrap. The administrator
may set switches on the computer console to specify the address of a special hard-
coded bootstrap program or just push a single button that instructs the machine to
load a bootstrap program from its microcode. This program may consist of only a
few instructions that instruct the machine to execute another program. On UNIX
systems, the bootstrap procedure eventually reads the boot block (block 0) of a
disk, and loads it into memory. The program contained in the boot block loads the
kernel from the file system (from the file "/unix", for example, or another name
specified by an administrator>. After the kernel is loaded in memory, the boot
program transfers control to the start address of the kernel, and the kernel starts
running (algorithm start, Figure 7.30).
The kernel initializes its internal data structures. For instance, it constructs the
linked lists of free buffers and inodes, constructs hash queues for buffers and inodes.
initializes region structures, page table entries, and so on. After completing the
initialization phase, it mounts the root file system onto root ("1") and fashions the
environment for process 0, creating a u area, initializing slot 0 in the process table
and making root the current directory of process 0, among other things.
When the environment of process 0 is set up, the system is running as process O.
Process 0 forks, invoking the fork algorithm directly from the kernel, because it is
executing in kernel mode. The new process, process 1, running in kernel mode,
creates its user-level context by allocating a data region and attaching it to its
address space. It grows the region to its proper size and copies code (described
shortly) from the kernel address space to the new region: This code now forms the
user-level context of process 1. Process 1 then sets up the saved user register
context, "returns" from kernel to user mode, and executes the code it had just
copied from the kernel. Process 1 is a user-level process as opposed to process 0,
which is a kernel-level process that executes in kernel mode. The text for process 1,
copied from the kernel, consists of a call to the exec system call to execute the
program "/etc/init". Process 1 calls exec and executes the program in the normal
fashion. Process 1 is commonly called init because it is responsible for initialization
of new processes.
Why does the kernel copy the code for the exec system call to the user address
space of process I? It could invoke an internal version of exec directly from the
136 PROCESS CONTROL
algorithm start ,. system startup procedure .,

input: none
output: none
{
initialize all kernel data structures;
pseudo-mount of root;
hand-craft environment of process 0;
fork process 1:
(
'* process 1 in here .,
allocate region;
attach region to init address space;
grow region to accommodate code about to copy in;
copy code from kernel space to init user space to exec init;
change mode: return from kernel to user mode;
,. init never gets here---as result of above change mode,
* init exec's 'etc'init and becomes a "normal" user process
.,
• with respect to invocation of system calls
,. proc 0 continues here .,

fork kernel processes;
,. process 0 invokes the swapper to manage the allocation of
• process address space to main memory and the swap devices.
.,
• This is an infinite loop; process 0 usually sleeps in the
• loop unless there is work for it to do.
execute code for swapper algorithm;
Figure 7.30. Algorithm for Booting the System
kernel, but that would be more complicated than the implementation just described.
To follow the latter procedure, exec would have to parse file names in kernel space,
not just in user space, as in the current implementation. Such generality, needed
only for init, would complicate the exec code and slow its performance in more
common cases.
The init process (Figure 7.31) is a process dispatcher, spawning processes that
allow users to log in to the system, among others. Init reads the file "/etc/inittab"
for instructions about which processes to spawn. The file "/etc/inittab" contains
lines that contain an "id," a state identifier (single user, multi-user, etc'), an
"action" (see exercise 7.43), and a program specification (see Figure 7.32). lnit
reads the file and, if the state in which it was invoked matches the state identifier
of a line, creates a process that executes the given program specification. For
example, when invoking init for the multi-user state (state 2), init typically spawns
7.9 SYSTEM BOOT AND THE (NIT PROCESS 237
algorithm init ,. init process, process 1 of the system .,

input: none
output: none
(
fd - open("'etc'inittab", O_RDONLY);
while (line read(fd, buffer»
{ -
,. read every line of file .,
if (invoked state !- buffer state)
continue;
,. state matched
if (forkO - - 0)
*' ,. loop back to while .,
{
execl ("process specified in buffer");
exitO;
*'
,. init process does not wait .,
,. loop back to while
while ((id - wait«int .) 0» !- -1)

{
,. check here if a spawned child died;
• consider respawning it .,
,. otherwise, just continue .,
Figure 7.31. Algorithm for Init
Format: identifier, state, action, process specification

Fields separated by colons.
Comment at end of line preceded by '#'
co::respawn:'etc'getty console console # Console in machine room

46:2:respawn:'etc'getty -t 60 tty46 4800H # comments here
Figure 7.32. Sample Inittab File

238 PROCESS CONTROL
getty processes to monitor the terminal lines configured on a system. When a user
successfully logs in, getty goes through a login procedure and execs a login shell,
described in Chapter 10. Meanwhile, init executes the wait system call, monitoring
the death of its child processes and the death of processes "orphaned" by exiting
parents.
Processes in the UNIX system are either user processes, daemon processes, or
kernel processes. Most processes on typical systems are user processes, associated
with users at a terminal. Daemon processes are not associated with any users but
do system-wide functions, such as administration and control of networks, execution
of time-dependent activities. line printer spooling, and so on. lnit may spawn
daemon processes that exist throughout the lifetime of the system or, on occasion,
users may spawn them. They are like user processes in that they run at user mode
and make system calls to access system services.
Kernel processes execute only in kernel mode. Process 0 spawns kernel
processes, such as the page-reclaiming process vhand, and then becomes the
swapper process. Kernel processes are similar to daemon processes in that they
provide system-wide services, but they have greater control over their execution
priorities since their code is part of the kernel. They can access kernel algorithms
and data structures directly without the use of system calls, so they are extremely
powerful. However, they are not as flexible as daemon processes, because the
kernel must be recompiled to change them.
7.10 SUMMARY
This chapter has discussed the system calls that manipulate the process context and
control its execution. The fork system call creates a new process by duplicating all
the regions attached to the parent process. The tricky part of the fork
implementation is to initialize the saved register context of the child process, so that
it starts executing inside the fork system call and recognizes that it is the child
process. All processes terminate in a call to the exit system call, which detaches
the regions of a process and sends a "death of child" signal to its parent. A parent
process can synchronize execution with the termination of a child process with the
wait system call. The exec system call allows a process to invoke other programs,
overlaying its address space with the contents of an executable file. The kernel
detaches the old process regions and allocates new regions, corresponding to the
executable file. Shared-text files and use of the sticky-bit mode improve memory
utilization and the startup time of execed programs. The system allows ordinary
users to execute with the privileges of other users, possibly superuser, with setuid
programs and use of the setuid system call. The brk system call allows a process to
change the size of its data region. Processes control their reaction to signals with
the signal system call. When they catch a signal, the kernel changes the user stack
and the user saved register context to set up the call to the signal handler.
Processes can send signals with the kill system call, and they can control receipt of
signals designated for particular process groups through the setpgrp system call.

Booting Process

Uploaded by

Copyright:

Available Formats

Booting Process

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Booting Process

Uploaded by

Copyright:

Available Formats

7.

8 THE SHELL 235

7.9 SYSTEM BOOT AND THE INIT PROCESS

algorithm start ,. system startup procedure .,

,. proc 0 continues here .,

execute code for swapper algorithm;

Figure 7.30. Algorithm for Booting the System

algorithm init ,. init process, process 1 of the system .,

while ((id - wait«int .) 0» !- -1)

Figure 7.31. Algorithm for Init

Format: identifier, state, action, process specification

co::respawn:'etc'getty console console # Console in machine room

Figure 7.32. Sample Inittab File

You might also like