Project 2: Command Line Shell (v 1.0) Solution

Description

Starter repository on GitHub: https://classroom.github.com/a/XS2Dqy2Y

The outermost layer of the operating system is called the shell. In Unix-based systems, the shell is generally a command line interface. Most Linux distributions ship with bash as the default (there are several others: csh, ksh, sh, tcsh, zsh). In this project, we’ll be implementing a shell of our own – see, I told you that you’d come to love command line interfaces in this class!

You will need to come up with a name for your shell first. The only requirement is that the name ends in ‘sh’, which is tradition in the computing world. In the following examples, my shell is named crash (Cool Really Awesome Shell) because of its tendency to crash.

​The Basics

Upon startup, your shell will print its prompt and wait for user input. Your shell should be able to run commands in both the current directory and those in the PATH environment variable. Use execvp to do this. To run a command in the current directory, you’ll need to prefix it with ./ as usual. If a command isn’t found, print an error message:

[?]─[1]─[mmalensek@glitter-sparkle:~/P2-malensek]$ ./hello Hello world! [?]─[2]─[mmalensek@glitter-sparkle:~/P2-malensek]$ ls /usr bin  include  lib  local  sbin  share  src [?]─[3]─[mmalensek@glitter-sparkle:~/P2-malensek]$ echo hello there! hello there! [?]─[4]─[mmalensek@glitter-sparkle:~/P2-malensek]$ ./blah crash: no such file or directory: ./blah [?]─[5]─[mmalensek@glitter-sparkle:~/P2-malensek]$ cd /this/does/not/exist chdir: no such file or directory: /this/does/not/exist [?]─[6]─[mmalensek@glitter-sparkle:~/P2-malensek]$

​Prompt

The shell prompt displays some helpful information. At a minimum, you must include:

• Command number (starting from 1)

• User name and host name: (username)@(hostname) followed by :

• The current working directory

• Process exit status

In the example above, these are separated by dashes and brackets to make it a little easier to read. The process exit status is shown as an emoji: a smiling face for success (exit code 0) and a sick face for failure (any nonzero exit code or failure to execute the child process). For this assignment, you are allowed to invent your own prompt format as long as it has the elements listed above. You can use colors, unicode characters, etc. if you’d like. For instance, some shells highlight the next command in red text after a nonzero exit code.

You will format the current working directory as follows: if the CWD is the user’s home directory, then the entire path is replaced with ~. Subdirectories under the home directory are prefixed with ~; if I am in /home/mmalensek/test, the prompt will show ~/test. Here’s a test to make sure you’ve handled ~ correctly:

[?]─[6]─[mmalensek@glitter-sparkle:~]$ whoami mmalensek [?]─[7]─[mmalensek@glitter-sparkle:~]$ cd P2-malensek # Create a directory with our full home directory in its path: # **Must use the username outputted above from whoami)** [?]─[8]─[mmalensek@glitter-sparkle:~/P2-malensek]$ mkdir -p /tmp/home/mmalensek/test [?]─[9]─[mmalensek@glitter-sparkle:~/P2-malensek]$ cd /tmp/home/mmalensek/test # Note that the FULL path is shown here (no ~): [?]─[10]─[mmalensek@glitter-sparkle:/tmp/home/mmalensek/test]$ pwd /tmp/home/mmalensek/test

​Scripting

Your shell must support scripting mode to run the test cases. Scripting mode reads commands from standard input and executes them without showing the prompt.

cat <<EOM | ./crash ls / echo "hi" exit EOM # Which outputs (note how the prompt is not displayed): bin  boot  dev  etc  home  lib  lost+found  mnt  opt  proc  root  run  sbin  srv  sys  tmp  usr  var hi # Another option (assuming commands.txt contains shell commands): ./crash < commands.txt (commands are executed line by line)

You should check and make sure you can run a large script with your shell. Note that the script should not have to end in exit.

To support scripting mode, you will need to determine whether stdin is connected to a terminal or not. If it’s not, then disable the prompt and proceed as usual. Here’s some sample code that does this with isatty:

#include <stdio.h> #include <unistd.h> int main(void) {     if (isatty(STDIN_FILENO)) {         printf("stdin is a TTY; entering interactive mode\n");     } else {         printf("data piped in on stdin; entering script mode\n");     }     return 0; }

Since the readline library we’re using for the shell UI is intended for interactive use, you will need to switch to a traditional input reading function such as getline when operating in scripting mode.

When implementing scripting mode, you will likely need to close stdin on the child process if your call to exec() fails. This prevents infinite loops.

​Built-In Commands

Most shells have built-in commands, including cd and exit. Your shell must support:

• cd to change the CWD. cd without arguments should return to the user’s home directory.

• # (comments): strings prefixed with # will be ignored by the shell

• history, which prints the last 100 commands entered with their command numbers

• ! (history execution): entering !39 will re-run command number 39, and !! re-runs the last command that was entered. !ls re-runs the last command that starts with ‘ls.’ Note that command numbers are NOT the same as the array positions; e.g., you may have 100 history elements, with command numbers 600 – 699.

• jobs to list currently-running background jobs.

• exit to exit the shell.

​Signal Handling

Your shell should gracefully handle the user pressing Ctrl+C:

[?]─[11]─[mmalensek@glitter-sparkle:~]$ hi there oh wait nevermind^C [?]─[11]─[mmalensek@glitter-sparkle:~]$ ^C [?]─[11]─[mmalensek@glitter-sparkle:~]$ ^C [?]─[11]─[mmalensek@glitter-sparkle:~]$ ^C [?]─[11]─[mmalensek@glitter-sparkle:~]$ sleep 100 ^C [?]─[12]─[mmalensek@glitter-sparkle:~]$ sleep 5

The most important aspect of this is making sure ^C doesn’t terminate your shell. To make the output look like the example above, in your signal handler you can (1) print a newline character, (2) print the prompt only if no command is currently executing, and (3) fflush(stdout).

​History

Here’s a demonstration of the history command:

[?]─[142]─[mmalensek@glitter-sparkle:~]$ history   43 ls -l   43 top   44 echo "hi" # This prints out 'hi' ... (commands removed for brevity) ...  140 ls /bin  141 gcc -g crash.c  142 history

In this demo, the user has entered 142 commands. Only the last 100 are kept, so the list starts at command 43. If the user enters a blank command, it should not be shown in the history or increment the command counter.

​I/O Redirection

Your shell must support file input/output redirection and pipe redirection:

# Create/overwrite 'my_file.txt' and redirect the output of echo there: [?]─[14]─[mmalensek@glitter-sparkle:~]$ echo "hello world!" > my_file.txt [?]─[15]─[mmalensek@glitter-sparkle:~]$ cat my_file.txt hello world! # Append text with '>>': [?]─[16]─[mmalensek@glitter-sparkle:~]$ echo "hello world!" >> my_file.txt [?]─[17]─[mmalensek@glitter-sparkle:~]$ cat my_file.txt hello world! hello world! # Pipe redirection: [?]─[18]─[mmalensek@glitter-sparkle:~]$ cat other_file.txt | sort (sorted contents shown) [?]─[19]─[mmalensek@glitter-sparkle:~]$ seq 100000 | wc -l 10000 [?]─[20]─[mmalensek@glitter-sparkle:~]$ cat /etc/passwd | sort > sorted_pwd.txt (sorted contents written to 'sorted_pwd.txt') # This is equivalent, but uses input redirection instead of cat: [?]─[21]─[mmalensek@glitter-sparkle:~]$ sort < /etc/passwd > sorted_pwd.txt # This is equivalent as well. Order of < and > don't matter: [?]─[22]─[mmalensek@glitter-sparkle:~]$ sort > sorted_pwd.txt < /etc/passwd # Here's input redirection by itself (not redirecting to a file): [?]─[23]─[mmalensek@glitter-sparkle:~]$ sort < sorted_pwd.txt (sorted contents shown)

Use pipe and dup2 to achieve this. Your implementation should support arbitrary numbers of pipes, but you only need to support one file redirection per command line (i.e., a > or >> in the middle of a pipeline is not something you need to consider).

​Background Jobs

If a command ends in &, then it should run in the background. In other words, don’t wait for the command to finish before prompting for the next command. If you enter jobs, your shell should print out a list of currently-running backgrounded processes (use the original command line as it was entered, including the & character). The status of background jobs is not shown in the prompt.

To implement this, you will need:

• A signal handler for SIGCHLD. This signal is sent to a process any time one of its children exit.

• A non-blocking call to waitpid in your signal handler. Pass in pid
= -1 and options
= WNOHANG.

The difference between a background job and a regular job is simply whether or not a blocking call to waitpid() is performed. If you do a standard waitpid() with options
= 0, then the job will run in the foreground and the shell won’t prompt for a new command until the child finishes (the usual case). Otherwise, the process will run and the shell will prompt for the next command without waiting.

Each background process should be stored in a job array, with a maximum of 10 background jobs. NOTE: your shell prompt output may print in the wrong place when using background jobs. This is completely normal.

​Incremental History Search

When the user presses the ‘up’ arrow key on the keyboard, display the most recent history entry. If the user continues to press ‘up’, they should be able to scroll through each history entry in reverse chronological order. Pressing ‘down’ navigates back toward the most recently-entered command in the history, ending back at the start (a blank line).

If the user enters a string prefix, e.g., ‘ec’ before pressing the ‘up’ arrow, only show history entries that begin with ‘ec.’ For instance, these might include various invocations of the ‘echo’ command. The ‘down’ key works similarly in this case. If the user edits any of the displayed history entries, start the process over again – in other words, if one of the history entries was echo
hello and the user changes the command line to read echo
helloworld then the history search is reset and pressing up will begin searching for ‘echo helloworld’ as the prefix instead.

You can use the rl_line_buffer variable to determine what the user has entered – this is a global variable defined in the readline library.

​Tab Completion

When the user presses the tab key, autocomplete their command. Search in this order:

1. Each executable entry in the user’s $PATH

2. The built-in commands your shell supports

3. Local files in the current working directory

Luckily the readline library we’re using provides support for local files as a fallback mechanism, so you are only responsible for builtins and items in the user’s $PATH. The command_generator function you will implement acts somewhat like an iterator; the readline library will call it continuously to receive each autocomplete suggestion (one per function call) until it encounters NULL.

​Helpful readline variables and functions

The readline documentation provides a good starting point for using the library. However, you can probably focus on the following to help with your implementation:

• rl_line_buffer – global variable that contains the current command line. You probably shouldn’t manipulate this directly, but you can use it to get a copy of the command line string.

• rl_point – cursor position (index) in rl_line_buffer

• rl_end – The number of characters present in rl_line_buffer

• rl_replace_line() – function that allows you to change the command line, completely replacing whatever the user has entered.

​Hints

Here’s some hints to guide your implementation:

• execvp will use the PATH environment variable (already set up by your default shell) to find executable programs. You don’t need to worry about setting up the PATH yourself.

• Check out the getlogin, gethostname, and getpwuid functions for constructing your prompt.

• Don’t use getwd to determine the CWD – it is deprecated on Linux. Use getcwd instead.

• For the cd command, use the chdir syscall.

• To replace the user home directory with ~, the strstr function may be useful. Some creative manipulation of character arrays and pointer arithmetic can save you some work.

​Testing Your Code

Check your code against the provided test cases. You should make sure your code runs on your Arch Linux VM. We’ll have interactive grading for projects, where you will demonstrate program functionality and walk through your logic.

Submission: submit via GitHub by checking in your code before the project deadline.

​Grading

• 26 pts – Passing the test cases

• 4 pts – Code review:

  • Prompt, UI, and general interactive functionality. We’ll run your shell to test this with a few commands.

  • Code quality and stylistic consistency

  • Functions, structs, etc. must have documentation in Doxygen format (similar to Javadoc). Describe inputs, outputs, and the purpose of each function. NOTE: this is included in the test cases, but we will also look through your documentation.

  • No dead, leftover, or unnecessary code.

  • You must include a README.md file that describes your program, how it works, how to build it, and any other relevant details. You’ll be happy you did this later if/when your revisit the codebase. Here is an example README.md file.

Restrictions: you may use any standard C library functionality. Other than readline, external libraries are not allowed unless permission is granted in advance (including the GNU history library). Your shell may not call another shell (e.g., running commands via the system function or executing bash, sh, etc.). Do not use strtok to tokenize input. Your code must compile and run on your VM set up with Arch Linux as described in class. Failure to follow these guidelines will will result in a grade of 0.

​Changelog

• Initial project specification posted (9/28)