The objectives of this assignment are:

1. Experience with fork() and exec() system calls.
2. Understand how a shell works.
3. Learning how to use pipes.

You are to write a simple Unix shell (the shell is simple, but this assignment is not!). Your program should read commands from the user (use readline!) one line at a time, and for each line the shell attempts to process the user command.

There are two basic kinds of commands that your shell must handle: built-in commands (that your shell handles itself) and external programs (the shell starts a new process which then executes the external program). The details of how to handle each of these types of commands is shown below. Your shell must support an unlimited number of pipes between external commands, automatically directing the output of one external command to the input of another. Your shell must also support I/O redirection (arranging it so that external commands will read from, or write to, a file). Finally, your shell will use the environment variable HWPATH as a list of directories in which to search for external commands.

There are a few commands that your shell must recognize and deal with directly - these commands are listed below. For each of these commands, the shell should not create any new processes! You do not need to be able to support using pipes with any shell internal command.

The actual commands you need to support are shown below, you must use the exact syntax described (otherwise our testing scripts won't work with your shell).

• cd:   The cd command allows the user to change the current working directory while using your shell. You need to support relative and absolute paths. If the user enters the cd command with no arguments, the shell should use the environment variable "HOME" to determine what directory to change to. If the user does not enter and arguments and there is no HOME environment variable (or it's value is empty), the cd command should not change the directory.

  If there is an argument to the cd command, the shell should attempt to switch to that directory (issue the "chdir()" system call). If this fails the user should be told so!

• pwd:   The pwd command prints out the current working directory. There are no arguments to pwd.

• set:   The set command is basically the same command you implemented in HW1. You need to set an environment variable specified by the user. The full syntax is just like in homework #1:

  set varname = somevalue

• print:   The print command is also just like in HW1:

  print varname

  The command should print out the value of an existing environment variable

• where:   The where command is used to find out what executable file corresponds to a specific command. For example, if you type "where ls", the shell should print out the full path to the ls command (something like "/usr/bin/ls"). This command should use the HWPATH environment variable as a list of directories in which too look for the command. Note that the shell needs to find a file with the right name, and the file must be executable (or the shell should ignore it).

  The HWPATH environment variable is modeled after the PATH variable that is actually used by real Unix shells. HWPATH is a single string that includes a list of directorys (relative or absolute) delimited by colons. For example, here is an HWPATH that tells the shell to look in /usr/bin, then in /usr/local/bin and finally in the current working directory:

  set HWPATH = /usr/bin:/usr/local/bin:.

  It is possible that the user types in the complete path to a file, for example the user could type "where /opt/foo/sillyface". In this case your shell should report that this is located at "/opt/foo/sillyface". Note that the user could also type "where ./foo" and you must handle this as well. In general if the command starts with either "/" or "." you need to treat the command as a complete path, otherwise treat is as a filename (that could be anywhere in the HWPATH).

  Note that when the shell encounters an external command (anything other than the 4 commands listed above), it will search for a file based on the HWPATH - so it makes sense to write a function that does this search - the same function can then be used by the where command and by the shell when executing an external command.

If the user input does not match any of the internal commands listed above, your shell must attempt to find an executable file with the same name as the command, and to run the command. If the command name starts with either "/" or "." - the user is specifying a complete path to the executable file. If the command name doesn't specify a complete path - your shell should assume it is a file name, and should use HWPATH to locate the actual executable file.

Your shell must support any number of command line arguments, for example the user could type "ls -al foo.c blah.h" and you need to make sure the ls command receives the command line arguments. Your shell does not need to support any special characters like "*" or "?", so if the user types in "ls *.c" you just pass along the "*.c" to the ls command (a real shell would replace this with the name of all files that end in .c).

Your shell must support an unlimited number of pipes - this means that each command line could include a number of external commands. Note that you do not need to support pipes with internal commands. For each pipe in the command line you need to take care of connecting stdout of the left command to stdin of the command following the "|". For example, if the user types "ls -al | sort", you need to arrange is so that the ls command is run with stdout directed to a Unix pipe, and that the sort command is run with stdin coming from that same pipe.

Your shell must support I/O redirection. If an external command is followed by "<", it must read input from the named file. For example, the command line "sort < foo", means that sort should be fed the file foo via stdin. If an external command is followed by a ">", out must arrange it so that stdout of the command is directed to a file. For example, "ls > save" means that the ls command should save it's output to the the file named "save". It is possible that you could have both on the same command, for example: "sort < infile > outfile".

You do not need to support redirection of standard error.

Your shell should be able to handle something like this:

ls -al | cut -c31- | sort -rn > foo

The following are the requirements for the project:

• Your program must compile and run on the CS department FreeBSD machines (freebsd.remote.cs.rpi.edu).

• Your shell must support the five internal commands chdir, pwd, set, print and where. Feel free to include other internal commands if you find this useful for your own testing.

• Your program must use the environment variable HWPATH as a list of directories to search. It is possible that this environment variable will exist when your shell is started (inherited from the real shell), it is also possible that it will be set using the set internal shell command.

• For command lines that do not involve an internal command, your shell must support multiple commands (including options) seperated by "|", and your shell must arrange the automatic creation of Unix pipes that facilitate the connection between individual commands. Your shell must also support I/O redirection for external commands.

• Your shell should print some kind of error message whenever it can't find a suitable file for an external command, or when anything unexpected happens (exec fails, can't create a pipe, etc). You must include code that checks the return value of every single ittsy-bitsy system call! All of them. Really!

• Your submission must include a Makefile we can use to build your shell on the CS FreeBSD machines. See the HW1 FAQ for information on building a simple Makefile

• Your submission must include a file named README that includes the following information:

  • Your name
  • A 1 line description of each file you are submitting
  • A description of any problems you had, known bugs or deficiencies.
  • Anything else you want us to know.

Grading: Your code will be graded according to the formula below. Note that to get full credit we must be able to understand your code (it must be commented!)

You can get partial credit for any part.

If you code does not compile and run under FreeBSD, you will lose at least 1/2 the relevant points ( partial credit will be awarded based on visual inspection of the code).

Log in to WebCT at webct.rpi.edu using your RCS id and password. Once you get to MyWebCT click on "Operating Systems", and from there go to the homework drop boxes. Submit your files (individually, zipped or tarred) to the drop box labeled HW3.