首页 > > 详细

CS 344留学生作业代做、C++编程作业调试、C++课程设计作业代写、代做command line作业代写留学生 Statistics统计、回归、迭代|调试

Program 3 – CS 344
Overview
In this assignment you will write your own shell in C, similar to bash. No other languages, including C++, are
allowed, though you may use any version of C you like, such as C99. The shell will run command line
instructions and return the results similar to other shells you have used, but without many of their fancier
features.
In this assignment you will write your own shell, called smallsh. This will work like the bash shell you are
used to using, prompting for a command line and running commands, but it will not have many of the special
features of the bash shell.
Your shell will allow for the redirection of standard input and standard output and it will support both
foreground and background processes (controllable by the command line and by receiving signals).
Your shell will support three built in commands: exit , cd , and status . It will also support comments, which
are lines beginning with the # character.
During the development of this program, take extra care to only do your work on our class server, as your
software will likely negatively impact whatever machine it runs on, especially before it's finished. If you cause
trouble on one of the non-class, public servers, it could hurt your grade! If you are having trouble logging in
to any of our EECS servers because of runaway processes, please use this page to kill off any programs
running on your account that might be blocking your access:
https://teach.engr.oregonstate.edu/teach.php?type=kill_runaway_processes
(https://teach.engr.oregonstate.edu/teach.php?type=kill_runaway_processes)
Specications
The Prompt
Use the colon : symbol as a prompt for each command line. Be sure you flush out the output buffers each
time you print, as the text that you're outputting may not reach the screen until you do in this kind of
interactive program. To do this, call fflush() immediately after each and every time you output text.
The general syntax of a command line is:
command [arg1 arg2 ...] [< input_file] [> output_file] [&]
2019/11/16 Program 3 - smallsh
https://oregonstate.instructure.com/courses/1738955/assignments/7656345 2/7
…where items in square brackets are optional. You can assume that a command is made up of words
separated by spaces. The special symbols <, >, and & are recognized, but they must be surrounded by
spaces like other words. If the command is to be executed in the background, the last word must be &. If the
& character appears anywhere else, just treat it as normal text. If standard input or output is to be redirected,
the > or < words followed by a filename word must appear after all the arguments. Input redirection can
appear before or after output redirection.
Your shell does not need to support any quoting; so arguments with spaces inside them are not possible.
We are also not implementing the pipe "|" operator.
Your shell must support command lines with a maximum length of 2048 characters, and a maximum of 512
arguments. You do not need to do any error checking on the syntax of the command line.
Finally, your shell should allow blank lines and comments. Any line that begins with the # character is a
comment line and should be ignored (mid-line comments, such as the C-style //, will not be supported). A
blank line (one without any commands) should also do nothing. Your shell should just re-prompt for another
command when it receives either a blank line or a comment line.
Command Execution
You will use fork(), exec(), and waitpid() to execute commands. From a conceptual perspective, consider
setting up your shell to run in this manner: let the parent process (your shell) continue running. Whenever a
non-built in command is received, have the parent fork() off a child. This child then does any needed
input/output redirection before running exec() on the command given. Note that when doing redirection, that
after using dup2() to set up the redirection, the redirection symbol and redirection destination/source are
NOT passed into the following exec command (i.e., if the command given is ls > junk , then you handle the
redirection to "junk" with dup2() and then simply pass ls into exec() ).
Note that exec() will fail, and return the reason why, if it is told to execute something that it cannot do, like
run a program that doesn't exist. In this case, your shell should indicate to the user that a command could
not be executed (which you know because exec() returned an error), and set the value retrieved by the builtin
status command to 1. Make sure that the child process that has had an exec() call fail terminates itself,
or else it often loops back up to the top and tries to become a parent shell. This is easy to spot: if the output
of the grading script seems to be repeating itself, then you've likely got a child process that didn't terminate
after a failed exec().
Your shell should use the PATH variable to look for non-built in commands, and it should allow shell scripts
to be executed. If a command fails because the shell could not find the command to run, then the shell will
print an error message and set the exit status to 1.
As above, after the fork() but before the exec() you must do any input and/or output redirection with dup2().
An input file redirected via stdin should be opened for reading only; if your shell cannot open the file for
reading, it should print an error message and set the exit status to 1 (but don't exit the shell). Similarly, an
output file redirected via stdout should be opened for writing only; it should be truncated if it already exists or
created if it does not exist. If your shell cannot open the output file it should print an error message and set
the exit status to 1 (but don't exit the shell).
Both stdin and stdout for a command can be redirected at the same time (see example below).
Your program must expand any instance of "$$" in a command into the process ID of the shell itself. Your
shell does not otherwise perform variable expansion. This feature makes it easier to create a grading script
2019/11/16 Program 3 - smallsh
https://oregonstate.instructure.com/courses/1738955/assignments/7656345 3/7
that keeps your work separate.
Background and Foreground
The shell should wait() for completion of foreground commands (commands without the &) before prompting
for the next command. If the command given was a foreground command, then the parent shell does NOT
return command line access and control to the user until the child terminates. It is recommend to have the
parent simply call waitpid() on the child, while it waits.
The shell will not wait for background commands to complete. If the command given was a background
process, then the parent returns command line access and control to the user immediately after forking off
the child. In this scenario, your parent shell will need to periodically check for the background child
processes to complete (with waitpid(...NOHANG...)), so that they can be cleaned up, as the shell continues
to run and process commands. Consider storing the PIDs of non-completed background processes in an
array, so that they can periodically be checked for. Alternatively, you may use a signal handler to
immediately wait() for child processes that terminate, as opposed to periodically checking a list of started
background processes. The time to print out when these background processes have completed is just
BEFORE command line access and control are returned to the user, every time that happens.
Background commands should have their standard input redirected from /dev/null if the user did not specify
some other file to take standard input from. What happens to background commands that read from
standard input if you forget this? Background commands should also not send their standard output to the
screen: redirect stdout to /dev/null if no other target is given.
The shell will print the process id of a background process when it begins. When a background process
terminates, a message showing the process id and exit status will be printed. You should check to see if any
background processes completed just before you prompt for a new command and print this message then.
In this way the messages about completed background processes will not appear during other running
commands, though the user will have to wait until they complete some other command to see these
messages (this is the way the C shell and Bourne shells work; see example below). You will probably
want to use waitpid() to check for completed background processes.
Signals
A CTRL-C command from the keyboard will send a SIGINT signal to your parent shell process and all
children at the same time (this is a built-in part of Linux). For this assignment, make sure that SIGINT does
not terminate your shell, but only terminates the foreground command if one is running. To do this, you'll
have to create the appropriate signal handlers with sigaction(). The parent should not attempt to terminate
the foreground child process when the parent receives a SIGINT signal: instead, the foreground child (if any)
must terminate itself on receipt of this signal.
If a child foreground process is killed by a signal, the parent must immediately print out the number of the
signal that killed it's foreground child process (see the example) before prompting the user for the next
command.
Background processes should also not be terminated by a SIGINT signal. They will terminate themselves,
continue running, or be terminated when the shell exits (see below).
A CTRL-Z command from the keyboard will send a SIGTSTP signal to your parent shell process and all
children at the same time (this is a built-in part of Linux). When this signal is received by your shell, your
shell must display an informative message (see below) immediately if it's sitting at the prompt, or
2019/11/16 Program 3 - smallsh
https://oregonstate.instructure.com/courses/1738955/assignments/7656345 4/7
immediately after any currently running foreground process has terminated, and then enter a state where
subsequent commands can no longer be run in the background. In this state, the & operator should simply
be ignored - run all such commands as if they were foreground processes. If the user sends SIGTSTP
again, display another informative message (see below) immediately after any currently running foreground
process terminates, and then return back to the normal condition where the & operator is once again
honored for subsequent commands, allowing them to be placed in the background. See the example below
for usage and the exact syntax which you must use for these two informative messages. Your foreground
and background child processes should all ignore a SIGTSTP signal: only your shell should react to it.
Built-in Commands
Your shell will support three built-in commands: exit , cd , and status . You do not have to support
input/output redirection for these built in commands and they do not have to set any exit status. These three
built-in commands are the only ones that your shell will handle itself - all others are simply passed on to a
member of the exec() family of functions (which member is up to you) as described above.
If the user tries to run one of these built-in commands in the background with the & option, ignore that option
and run it in the foreground anyway (i.e. don't display an error, just run the command in the foreground).
The exit command exits your shell. It takes no arguments. When this command is run, your shell must kill
any other processes or jobs that your shell has started before it terminates itself.
The cd command changes the working directory of your shell. By itself - with no arguments - it changes to
the directory specified in the HOME environment variable (not to the location where smallsh was executed
from, unless your shell executable is located in the HOME directory, in which case these are the same). This
command can also take one argument: the path of a directory to change to. Your cd command should
support both absolute and relative paths. When smallsh terminates, the original shell it was launched from
will still be in its original working directory, despite your use of chdir() in smallsh. Your shell's working
directory begins in whatever directory your shell's executible was launched from.
The status command prints out either the exit status or the terminating signal of the last foreground process
(not both, processes killed by signals do not have exit statuses!) ran by your shell. If this command is run
before any foreground command is run, then it should simply return the exit status 0. These three built-in
shell commands do not count as foreground processes for the purposes of this built-in command - i.e.,
status should ignore built-in commands.
Example
Here is an example run using smallsh. Note that CTRL-C has no effect towards the bottom of the example,
when it's used while sitting at the command prompt:
$ smallsh
: ls
junk smallsh smallsh.c
: ls > junk
: status
exit value 0
: cat junk
junk
smallsh
smallsh.c
: wc < junk > junk2
: wc < junk
3 3 23
2019/11/16 Program 3 - smallsh
https://oregonstate.instructure.com/courses/1738955/assignments/7656345 5/7
: test -f badfile
: status
exit value 1
: wc < badfile
cannot open badfile for input
: status
exit value 1
: badfile
badfile: no such file or directory
: sleep 5
^Cterminated by signal 2
: status &
terminated by signal 2
: sleep 15 &
background pid is 4923
: ps
PID TTY TIME CMD
4923 pts/0 00:00:00 sleep
4564 pts/0 00:00:03 bash
4867 pts/0 00:01:32 smallsh
4927 pts/0 00:00:00 ps
:
: # that was a blank command line, this is a comment line
:
background pid 4923 is done: exit value 0
: # the background sleep finally finished
: sleep 30 &
background pid is 4941
: kill -15 4941
background pid 4941 is done: terminated by signal 15
: pwd
/nfs/stak/faculty/b/brewsteb/CS344/prog3
: cd
: pwd
/nfs/stak/faculty/b/brewsteb
: cd CS344
: pwd
/nfs/stak/faculty/b/brewsteb/CS344
: echo 4867
4867
: echo $$
4867
: ^C^Z
Entering foreground-only mode (& is now ignored)
: date
Mon Jan 2 11:24:33 PST 2017
: sleep 5 &
: date
Mon Jan 2 11:24:38 PST 2017
: ^Z
Exiting foreground-only mode
: date
Mon Jan 2 11:24:39 PST 2017
: sleep 5 &
background pid is 4963
: date
Mon Jan 2 11:24:39 PST 2017
: exit $
Grading Method
2019/11/16 Program 3 - smallsh
https://oregonstate.instructure.com/courses/1738955/assignments/7656345 6/7
In addition to your shell needing to replicate the above example in functionality, this assignment is provided
with the actual grading test script that will be used to assign your program a grade. Your program must
function with this grading script, as follows. To run it, place it in the same directory as your compiled shell,
chmod it ( chmod +x ./p3testscript ) and run this command from a bash prompt:
$ p3testscript 2>&1
or
$ p3testscript 2>&1 | more
or
$ p3testscript > mytestresults 2>&1
Don’t worry if the spacing, indentation, or look of the output of the script is different than when you run it
interactively: that won’t affect your grade. The script may add extra colons at the beginning of lines or do
other weird things, like put output about terminating processes further down the script than you intended.
Use the script to prepare for your grade, as this is how it's being earned.
Note that as an extra challenge, no "clean run" script is provided for Program 3: you'll need to interpret the
results of your program yourself.
If your program does not work with the grading script, and you instead request that we grade your script by
hand, we will have to apply a 50% reduction to your final score. Make sure you work with the grading script
on our class server from the very beginning!
Hints
It is recommended that you program the built-in commands first, before tackling the fork(), exec(), waitpid()
specifications.
Don't forget to use fflush(stdout), as described above!
As stated above, make sure you work with the grading script on our class server from the very beginning -
don't leave this to the end!
Re-Entrancy
A topic we haven't covered much is the concept of re-entrancy
(https://en.wikipedia.org/wiki/Reentrancy_(computing)) . This is important when we consider that signal
handlers cause jumps in execution that cause problems with certain functions.
For our purposes, note that the printf() family of functions is NOT re-entrant. In your signal handlers, when
outputting text, you must use other output functions!
Where to Program
I HIGHLY recommend that you develop this program directly on our course server. Doing so will prevent you
from having problems transferring the program back and forth, and having compatibility problems. You have
been warned: it will not behave the same on your own computer!
If you do see ^M characters all over your files, try this command:
2019/11/16 Program 3 - smallsh
https://oregonstate.instructure.com/courses/1738955/assignments/7656345 7/7
$ dos2unix bustedFile
What to Turn In and When
Please submit a single zip file of your program code, which may be in as many different files as you want.
Also, inside that zip file, you must provide a file called "README" that contains instructions on HOW to
compile your code; you may compile your code however you wish. DO NOT include a copy of the testing
script. As our Syllabus says, please be aware that neither the Instructor nor the TA(s) are alerted to
comments added to the text boxes in Canvas that are alongside your assignment submissions, and they
may not be seen. No notifications (email or otherwise) are sent out when these comments are added, so we
aren't aware that you have added content! If you need to make a meta-comment about this assignment,
please include it in the README file in your .zip file, or email the person directly who will be grading it (see
the Home page for grading responsibilities).
The graders will compile your code according to your exact specifications. They will make a reasonable
effort to make it work, but if it doesn’t compile, you’ll receive a zero on this assignment.
The due date given below is the last minute that this can be turned in for full credit. The "available until" date
is NOT the due date, but instead closes off submissions for this assignment automatically once 48 hours
past the due date has been reached, in accordance with our Syllabus Grading policies.
Grading
Once the program is compiled, according to your specifications given in "readme.txt", your shell will be
executed to run a few sample commands against ( ls , status , exit , in that order). If the program does not
successfully work on those commands, it will receive a zero. If it works, it will have the p3testscript program
ran against it (as detailed above) for final grading. Points will be assigned according to the test script running
on our class server only, so make sure it runs there.
170 points are available in the test script, while the final 10 points will be based on your style, readability,
and commenting. Comment well, often, and verbosely: we want to see that you are telling us WHY you are
doing things, in addition to telling us WHAT you are doing.
The TAs will use this exact set of instructions: Program3 Grading.pdf to grade your submission.

联系我们
  • QQ:99515681
  • 邮箱:99515681@qq.com
  • 工作时间:8:00-23:00
  • 微信:codinghelp
热点文章
程序代写更多图片

联系我们 - QQ: 99515681 微信:codinghelp
© 2014 www.7daixie.com
程序代写网!