CPSC 457 - Principles of Operating Systems - Assignment 2

This assignment comes in two parts. Part A is a programming assignment to familiarize you with the methods for modifying Linux and adding new features to it. It should be completed on your virtual machines.

Part B includes questions to be answered which will encourage you to think about some of the work you did in Part A and some of the topics we have discussed in lecture.

This assignment will be marked out of 100 and will count for 20% of your total grade in the course. 60 marks will be dedicated to Part A and 40 marks will be dedicated to Part B (four marks for each question).

You will hand in your assignment by sharing your git repository with your TA through the CPSC gitlab and Tagging the branch which includes all of your files for Part A and Part B, "CPSC457-Assignment2-Complete".

This assignment is due at 2:00pm on Friday June 10.

This assignment was updated to fix some errors and add some clarifications. If you would like to see the original, look here.

Changes:

• Fixed number of marks for Part B from 400 to 40
• Removed "if the file is a directory, the sum of the size of all files and directories contained in the directory (to an arbitrary depth)." from the functional requirements.
• Changed "Include a usage statement" to "Include usage statements for each of the user program".
• Added requirement to demonstrate your assignment
• Added a quantum of 2ms to the round robin in question 7. If you worked out the solution to the question using average running times, this should give you the same answer.
• Added some hints to help things get going.
• Added the demo procedure

Part A - Coding

Your CEO has trouble remembering program names and process ID numbers. He recently left a large search engine company whose motto was "Don't sort -- search!" and encouraged tagging of everything. One day, he asks you to modify the Linux kernel to augment job control by tagging processes with an arbitrary list of keywords. Since he once took an OS course, he dimly remembers a discussion of extended attributes from the filesystem lecture. He asks you to "implement extended attributes for processes" (rather than files). He gives you a few weeks to develop a working prototype.

Your CTO takes pity on you, and gives you a few hints. She says:

• A) Create a user-level program called 'ptag' that takes a tag and a process ID as arguments and associates that tag with the process, e.g.,

  (eye@mordor hw2)$ ptag 1234 -a "yellow"

  ptag should also support removing a tag from a process.

  (eye@mordor hw2)$ ptag 1234 -r "yellow"

  (eye@mordor hw2)$ ptag 1234 -r "green"

  ptag should fail with an appropriate error message if the process ID is not valid or you don't own it. Use getopt to process the arguments (see `man 3 getopt').

• B) Implementing the ptag program program probably requires a new system call to associate and remove tags from processes. Be careful about copying arbitrary strings from userspace into kernel memory.
• C) You will probably need a data structure in the kernel to keep track of tags on a per-process basis. You may want to look into keeping an instance of the kernel's doubly-linked list around to track all the keywords (if any) associated with the process. It is possible to have a global list of keywords, but then you'll have to think carefully about how you'll associate a subset of them with each task. You should also use suitable concurrency primitives to protect access to this list. You should make sure the list is duplicated if the process forks (e.g., just like a parent and child share open file descriptors).

You do as your CTO recommends. The CEO is impressed, but complains that he cannot see the status of all the tagged processes. He leaves in a hurry to go golfing. Your CTO says:

• D) You should implement a utility called 'tagstat' that prints a table to stdout listing all processID-tag mappings, one per line, e.g.,

  22 : green

  22 : swordfish

  22 : hackers

  31 : yellow

  31 : sneakers

  56 : wargames

  133 : leet

  133 : lame

  By default it does not print entries for processes that do not have tags associated with them. By default it only prints entries for processes you own. Processes should appear in numerical order. You should also print the process state as a third column.

  You decide to have tagstat read this state out of a psuedo-device registered as /proc/ptags (so that tagstat is little more than a shell script that performs `cat /proc/ptags').

Your CEO is delighted, but laments the fact that he cannot do anything useful with these groups of processes, like kill all processes associated with the tag 'green'. After he goes to lunch, your CTO looks at you and says: "You're on your own."

• E) Implement a utility or shell script called tagkill. This utility takes one argument (a tag value) and kills each process (kill -9) associated with that tag.

void main(int argc, char *argv[]) {
	fork();
	fork();
	exit();
}