CS 31 Lab 8: Shell Program

Lab 8 Goals:

• Demystify how a Unix shell program works by writing one, including the following features:
  • Parsing string command line arguments into an argv list in dynamic memory.
  • Executing commands that run in the foreground and the background.
  • Executing previous commands by interacting with the shell history (using the !num syntax.)
• Learn how to create and reap processes with the fork, execvp, waitpid system calls.
• Interact with signals and write signal handler code.
• Gain more expertise with gdb and valgrind for debugging C programs.

Lab Description

You will implement a shell, which is the program you interact with on the command line of a terminal window. A shell operates by doing the following things:

1. Print a prompt and wait for the user to type in a command.
2. Read in the command string entered by the user.
3. Parse the command line string into an argv list.
4. If the command (first item in the parsed argv list) is a built-in shell command, the shell will handle it on its own (without forking a child process).
5. Otherwise, if it's not a built-in command, fork a child process to execute the command and wait for it to finish (unless the command is to be run in the background, then the shell doesn't wait for the child to exit).
6. Repeat until the user enters the built-in command exit to exit the shell program.

Part 1: Command line parsing library.

For the first stage of your shell, you'll implement a parsecmd library that contains functions to parse a command line string into its individual command line arguments, and construct an argv list of strings from the command line args. Your library functions can then be used by other programs by #including your parsecmd.h file and linking in your parsecmd.o binary on the gcc command line:

$ gcc -g -o tester tester.c parsemd.o

Your library will export one function, parse_cmd_dynamic, though you're encouraged to create helper functions to aid your implementation of it. Helper functions should not be exported via the parsecmd.h header file.

The parse_cmd_dynamic function will take in a command line string and turn it into an argv list (an array of strings, one per command line argument) that's suitable for passing to a newly-starting program. The function will also test for the presence of an ampersand (&), which indicates that the command should be run in the background. For example, if the user enters the command line string:

$ cat foo.tex

The string "cat foo.tex" will be passed to parse_cmd_dynamic, which will then produce an argv array that looks like:

argv [0] ---->"cat"
argv [1] ---->"foo.tex"
argv [2] ----|  (NULL)

Note that there will be a newline character after the foo.tex and before the final null-terminator character as a result of the user hitting the [enter] key.

This operation is known as tokenizing the string into individual tokens, each of which is separated by white space. Given a command line string as input, your implementation of the parse_cmd_dynamic function should dynamically allocate and return the argv array of strings, one for each command line token. It takes the following parameters:

/*
 * parse_cmd_dynamic - Parse the passed command line into an argv array.
 *
 *    cmdline: The command line string entered at the shell prompt
 *             (const means that this function cannot modify cmdline).
 *             Your code should NOT attempt to modify these characters!
 *
 *         bg: A pointer to an integer, whose value your code should set.
 *             The value pointed to by bg should be 1 if the command is to be
 *             run in the background, otherwise set it to 0.
 *
 *    returns: A dynamically allocated array of strings, each element
 *             stores a string corresponding to a command line argument.
 *             (Note: the caller is responsible for freeing the returned
 *             argv list, not your parse_cmd_dynamic function).
 */
char **parse_cmd_dynamic(const char *cmdline, int *bg);

To produce a dynamic argv array, your implementation must first determine how many tokens are in the command line string. After that, it should malloc EXACTLY the right number of (char *) argv buckets, one for each of the tokens plus one extra bucket at the end for NULL, which indicates the end of the tokens. Then, for each token, it should malloc exactly enough space to store the string corresponding to the that token (don't forget one extra byte for the null-terminator '\0' character).

For example, if the command line string is:

"   ls   -l   -a   "

The function should allocate space for an array of four character pointers (char *'s). The first should have three bytes (characters) allocated to it for storing 'l', 's', and '\0'. The second should allocate three bytes to store "-l\0", and the third should hold "-a\0". The final pointer should be NULL.

   // Declare a local var to store dynamically allocated args array of strings.
   char **args;

   // After allocating memory and tokenizing, it should look like:

      args --------->[0]-----> "ls"
                     [1]-----> "-l"
                     [2]-----> "-a"
                     [3]-----|  (NULL)

To help test parse_cmd_dynamic's behavior and convince yourself of its correctness, you'll write a few test cases in the provided tester.c file.

Requirements

• Your implementation of parse_cmd_dynamic should behave according to the specifications described above.


• You may not change the prototype (argument types or return type) of parse_cmd_dynamic, and your solution should NOT use strtok.


• You may assume that if an ampersand (&) appears in the command line, indicating that the command is to be run in the background, that it will appear at the end of the line. For this lab, an ampersand is never considered to be a token or part of a token; it's just a way for the user to express that the program should be run in the background. Ampersands should NOT appear in the argv array that parse_cmd_dynamic produces.


• Your implementation should make no attempt to modify the input command line argument string. It should be treated as a constant. You ARE allowed to make a copy of the string (e.g. with strdup) and modify that however you'd like.


• Your function should work for command line strings entered with any amount of whitespace between command line tokens (assuming there's at least one whitespace character between them). For example, these inputs should yield identical argv lists returned by your function:

  cat foo.txt  blah.txt      &
  cat foo.txt  blah.txt&
               cat          foo.txt           blah.txt          &
  

  You need to TEST that your code works for command lines with any amount of whitespace between arguments!


• You should fill in the missing TODO's in tester.c such that, if we run your tester, it works correctly and there are no valgrind issues (like memory leaks, since there are no calls to free in the starter code).


• For full credit, your solution should be well-commented, it should not use global variables, it should demonstrate good design, and it should be free of valgrind errors.

Tips

• Test your code in small increments. It's much easier to localize a bug when you've only changed a few lines.


• In implementing parse_cmd_dynamic, you will likely want to make more than one pass through the characters in the input string.


• When working with pointers, and in particular double pointers, it's often very useful to draw some pictures about what's pointing to what, noting the types of variables. Don't shy away from using the whiteboard or scratch paper!

Part 2: Building a shell.

Now that we can tokenize command line strings, let's put together the rest of the pieces for executing user commands. Your shell should support the following features:

1. Running commands in the foreground.

   When a command is run in the foreground, for example:

   cs31shell> ./sleeper 2  
   

   Your shell program should fork() a child process to execute sleeper and then wait until the child process exits before proceeding. You can accomplish this by calling waitpid in the parent (your shell) by passing in the pid of the child process (the return value of fork()).

2. Running commands in the background.

   When a command is run in the background, for example:

   cs31shell> ./sleeper 3 & 
   

   Your shell program should fork() a child process to execute sleeper, but it should NOT wait for the child to exit. Instead, after forking the child process, it should immediately return to step 1 (print out the prompt and read in the next command line). The child process will execute the command concurrently while the parent shell handles other command(s).

   Your shell must still reap background processes after they exit, so you can't just forget about them! When a child that was run in the background exits, your shell program will receive a SIGCHLD signal. You should install a SIGCHLD handler that will call waitpid() to reap the exited child process(es).

   Your shell should be able to run any number of processes in the background, so if you type in quick succession:

   cs31shell> ./sleeper & 
   cs31shell> ./sleeper & 
   cs31shell> ./sleeper & 
   cs31shell> ps 
   

   The ps program output should list all three sleeper child processes.

3. Built-in commands.

   Your shell should respond to the following three built-in commands on its own. It should not fork off a child to handle these!

   1. exit: Terminate the shell program. You can print out a goodbye message, if you'd like.
   2. history: Print a list of the user's 10 more recently entered command lines. (Note: blank lines should not be added to the history.)
   3. !num (where num is an actual number, e.g., !5): Re-execute a previous command from the history. The previous command could be a run-in-the-foreground, run-in-the-background, or a built-in command that your shell should execute appropriately. The command line should be added into the history list (i.e. executing !5 should not put !5 in the history list, instead a copy of the command line associated with command ID 5 from the history list should be added to the history list). See the sample output below for some examples of history and !num commands.


4. History.

   Your shell program should keep a list of the 10 most recently entered command lines by the user (use a #define constant named MAXHIST whose value is 10). The built-in history command should print out the history list in order from first (oldest) to last (most recently entered) command. For each element in the list, it should print the command ID and its corresponding command line string. The command ID is an ever-increasing number, and each command should increment the command ID by one. I suggest using an unsigned int to store this value (don't worry, I'm not going to try executing 4 billion commands in an attempt to overflow this value).

   Your history list should be implemented as circular queue of history structs: a circular array of MAXHIST buckets where new commands are added to one end and old ones removed from the other end.

   Users can request the execution of a previous command, as stored in the history list, by executing the built-in !num, where "num" is a number corresponding to a history command ID. Upon receiving such a command:

   1. Search your history list for a command with a matching command ID. Remember that the command ID is not the position in the history list, it is the unique number of the command in your shell's execution history (i.e. !5 is the 5th command run by your shell, !34 is the 34th command run by your shell).
   2. If a matching command ID is not found, print out an error message.
   3. Otherwise, use the command line from the matching history command ID, and re-execute it (this command now also becomes the most recent command to add to your history list).

   If the command from the history list is not a built-in command, then your shell should run it just like it does any foreground or background program (parse its command line into argv list, fork-execvp and waitpid or not).

   If the command from the history list is a built-in command (which could only be the history command), then it should execute it directly just like any built-in command.

Requirements

• Your shell should support running programs in the foreground (e.g. ls -l)


• Your shell should support running programs in the background (e.g. ./sleeper &)


• Your shell should support the built-in command exit to terminate.


• Your shell should support the built-in command history that keeps a list of the MAXHIST most recently entered command lines entered by the user. Use a constant definition for MAXHIST, and submit your solution with it set to 10, but try your shell out with other sizes too.


• Your shell should support running commands from the history using !num syntax, where num is the command ID of a command from your command history (e.g. !33 should execute the command with command ID 33 from your history list). If a matching command num is not found the current history list, then your shell should print out and error message (command not found), otherwise the command line from the matching command on the history list should be executed (again).


• Use the execvp version of exec for this assignment and waitpid instead of wait. See the "Tips" section below for examples.


• You need to add a signal handler for SIGCHLD signals so that you can reap exited child processes that are run in the background. You should not leave any long-term zombies!


• Whenever your code calls a library or system call function that returns a value, you should have code that checks the return value and handles error values. You can call exit for unrecoverable errors, but print out an error message first (printf or perror for system call error return values).


• The only global variables allowed are those associated with the history list and its state. All other program variables should be declared locally and passed to functions that use them.


• For full credit, your shell should use good modular design, be well-commented and free of valgrind errors. The main function should probably not be much longer than that in the starting point code. Think about breaking your program's functionality into distinct functions.

Example Output

It may be helpful for you to take a look at Tia's sample output, particularly to see how the built-in history features work.

Tips

• Implement and test incrementally (and run valgrind as you go). Here is one suggestion for an order to implement:
  1. Add a call to your library from part 1 to parse the input line into argv strings.
  2. Add support for the built-in command exit.
  3. Add support for running commands in the foreground (the parent process, the shell, waits for the child pid that it forks off to exec the command).
  4. Add support for running commands in the background (the parent process, the shell, does NOT wait for the child pid that it forks off to run the command). After forking off a child to run the command, the shell program should go back to its main loop of printing out a prompt and waiting for the user to enter the next command. You will need to add a signal handler on SIGCHLD so that when the process that is running in the background terminates, the shell reaps it. Use waitpid to reap all child processes. Use the sleeper program to test:

       cs31shell> ./sleeper &
       cs31shell> ./sleeper 2 &
       cs31shell> ps w
       

  5. Add support for the history list (implemented as a circular queue). The operations on your circular queue are slightly different when the queue is not yet full from when it is full and you need to replace the oldest entry each time a new command is entered. Think about all state you will need to keep to keep track of the first element in the list (for printing out), the next insertion spot, and the end of the list.
  6. Add support for !num built-in command that will run command num from your history list. num is a command ID, which is increasing as your shell runs commands. It is NOT the bucket index into the history list.
  
  
• The maximum length of a command line is defined in parsecmd.h. You can use the MAXLINE constant in your program.


• Use the functions execvp(), waitpid(), and signal() for executing programs, waiting on children, and registering signal handlers. Note that the first argument to waitpid is the process id (PID) of the process you'd like to wait for, but if you pass in an argument of -1, it will wait for ANY reapable child process. This is useful inside your SIGCHLD handler, where you won't know which child (or children) exited. You can pass it WNOHANG as the third parameter to prevent it from blocking when there are no children to reap.


• Remember that if you dynamically allocate space for a string (using malloc), you need to allocate a space at the end for the terminating null character ('\0'), and that you need to explicitly free the space when you are done using it (call free). Since your parsing library is allocating memory for the argv list, it's up to your shell to free that memory when it's done with it.


• You can call fflush(stdout); after any calls to printf to ensure the printf output is written immediately to the terminal. Otherwise, the C standard I/O library might buffer it for a short while.


• When in doubt about what your shell should do, try running the command in the bash shell (a standard system terminal) and see what it does.

Submitting

Please remove any debugging output prior to submitting.

To submit your code, simply commit your changes locally using git add and git commit. Then run git push while in your lab directory. Only one partner needs to run the final push, but make sure both partners have pulled and merged each others changes. See the section on Using a shared repo on the git help page.