Week 3 Exercises: Error handling and I/O

Great job making it so far in the quarter! It has only been two weeks, but we’ve covered a lot of ground. You’ve now learned enough to build useful and practical programs! We hope you are staying healthy and enjoying yourself this quarter.


In these exercises, you’ll work through implementing a tool for inspecting file descriptors that you can use to debug your CS 110 assignments. You’ll get some practice with handling ownership/references and working with Option and Result, and you’ll also get some light exposure to object-oriented programming in Rust!

Due date: Thursday, April 22, 11:59pm (Pacific time)

Ping us on Slack if you are having difficulty with this assignment. We would love to help clarify any misunderstandings, and we want you to sleep!

Getting the code

You should have received an invite to join this week’s Github repository. If you didn’t get an email invite, try going to this link:


You can download the code using git as usual:

git clone https://github.com/cs110l/week3-YOURSUNETID.git week3

Inspecting File Descriptors

In this week’s assignment, you will build a tool to inspect the open files of processes. This is based on a tool that was the precursor to C Playground’s open files debugger; although it is a bit complicated to use and does not show a full representation of the open file table (the kernel doesn’t expose much information to userspace, and we had to modify the kernel for C Playground), it is still very useful for debugging file descriptor issues in CS 110’s assignments 3 (multiprocessing pset) and 4 (Stanford shell).

To use the tool to debug a file-descriptor-related problem, you would do the following:

inspect-fds then prints a representation of the file descriptor tables at that point in time. For example, here is the output where a parent process creates two pipes and forks, using the file descriptor 4 to write data to the child process’s stdin and file descriptor 5 to read data from the child process’s stdout:

Two pipes example output

The pipes are color coded so that it is easier to identify file descriptors that point to the same pipe (in this example, parent fd 4 writes to the pipe that child fd 0 reads from, and parent fd 5 reads from the pipe that child fd 1 writes to).

This tool is especially helpful for debugging mistakes in CS 110 assignment 4, in which you implement a shell. Shells do pretty complicated rewiring of the file descriptor table in order to create pipelines of processes of arbitrary length, and in order to read/write input from/to files on disk. Here is what the file descriptor tables should look like when the assignment 4 shell runs sleep 100 < /tmp/testinput | sort | wc > /tmp/testoutput. (This is a useless command, but we put sleep 100 there in order to keep the pipeline running while we go to run inspect-fds in the other terminal.)

stsh example output

Here, you can see sleep 100 is getting its input from /tmp/testinput and is writing output to a pipe, which is connected to stdin for sort, whose output goes to a second pipe, connected to stdin for wc, whose final output is written to /tmp/testoutput.

You can try this on bash for a similar result!

Implementing this tool involves a lot of file I/O, which will give you excellent practice with error handling. I think I/O is probably the absolute worst when it comes to error handling, so if you can put up with this, you can put up with anything. This exercise will also give you good practice working with structs. We hope that throughout the process, you have fun with it, learn a bit about how Linux works, and build yourself a tool that you can use in CS 110 for the next few weeks!

Running the code

Unfortunately, you need to run this tool on a Linux computer, because Mac and Windows don’t use the same API for providing information about processes.

If you’re on Linux, you can stop reading and just use cargo as you have in previous exercises. If you’re on Mac or Windows, you have two or three options:

Installing Docker

On Mac, you can download and install Docker here. Easy peasy.

On Windows, there are a few ways to install. Instructions are available here.

Building the Docker image

cd into the inspect-fds/ directory, and then run docker build:

docker build -t inspect-fds ./

This will build an inspect-fds image containing dependencies needed to run your program. This might take a while. (In our case, the dependencies are just a barebones version of Ubuntu, cargo, and make.)

Once you build this image, you won’t need to do it again!

Running cargo

Once the image is built, you can run your code. Normally, Docker is used with docker run commands, but there are a lot of arguments that you need to pass. Instead of asking you to write those out, we included a mini script that does the docker run part for you. You can run it like this:

./container make
./container cargo build
./container cargo run

You can edit code locally on your machine using whatever editor you like and run the ./container command to run your code. No need to upload or sync your files anywhere.

Before you start: Build the sample C programs

There are a few sample C programs included that are used as tests. Before you start working, you should run make (or, if you’re using Docker, run ./container make).

Milestone 1: Getting the target process

A user specifies the process to inspect using a command line argument. For example, if you are running bash, you can inspect it like so:

cargo run bash

Our first order of business is to get information about the process the user wants to inspect.

Open src/ps_utils.rs and quicky skim the code that is provided to you. We have written some functions to call ps and pgrep to get information about processes by PID or by command name. There is a lot of semi-complicated error handling in this file (as there often is when dealing with I/O), so it may be helpful to read as an example. You won’t need to modify anything in this file for this assignment, but you will need to call get_target.

Next, open src/main.rs. You’ll see that we have declared a target variable containing the first argument in argv. Use the ps_utils::get_target function from the previous file to search for the target process.

Here is our output (which you are not required to match). As a reminder, if you’re using Docker, remember to prefix these commands with ./container.

🍉  cargo run bash
    Finished dev [unoptimized + debuginfo] target(s) in 0.58s
     Running `target/debug/inspect-fds bash`
Found pid 18042
🍉  cargo run nonexistent
    Finished dev [unoptimized + debuginfo] target(s) in 0.72s
     Running `target/debug/inspect-fds nonexistent`
Target "nonexistent" did not match any running PIDs or executables

As a convenience, we have also provided tests that run your program, checking the exit code. (They don’t verify that your program prints anything in particular!) You can run cargo test exit_status -- --nocapture --test-threads=1:

🍉  cargo test exit_status -- --nocapture --test-threads=1
    Finished dev [unoptimized + debuginfo] target(s) in 0.53s
     Running target/debug/deps/inspect_fds-41ab7dbb115eafda

running 2 tests
test test::test_exit_status_invalid_target ... Target "./nonexistent" did not match any running PIDs or executables
test test::test_exit_status_valid_target ... Found pid 18306

test result: ok. 2 passed; 0 failed; 0 ignored; 0 measured; 3 filtered out

(Again, if using Docker, don’t forget to prefix with ./container.)

Milestone 2: Printing process info

Let’s start printing out some more information about these processes. First, in src/process.rs, let’s add a print() method to the Process struct.

Defining a method (as part of a struct) is a lot like defining a normal function. Because we want this method to be part of Process, we put it inside of the impl Process {} block:

impl Process {
    // Other methods that are already defined...

    fn print() {

There are two more things we need to do. First, we should add pub to the method definition to make it “public” (callable by other parts of our codebase). In Rust, methods are private by default, meaning they can only be called by other methods in the same class. Since we will eventually want to call this method from our main() function, we mark it as pub.

Second, this method should take &self as a parameter. In Rust, self refers to the current instance (e.g. if you write proc.print() somewhere, then print() will be invoked with the proc object as the self parameter). This is similar to this in Java/C++ and self in Python. When implementing the print method, you can get the process’s PID as self.pid, PPID as self.ppid, and so on.

Your function will print details about this process. For now, print out the command name, pid, and ppid in a format that is something like this:

========== "bash" (pid 18042, ppid 17996) ==========

You are welcome to format your output however you like.

Once you have implemented this method, go back to main and call your method on the Process object that you got in the previous milestone (e.g. proc.print()). Run your program (or use the cargo test command from the previous milestone) to ensure it is printing as you expect.

Milestone 3: Listing file descriptors

Let’s get into the meat of this program! The Linux operating system exposes some information from the kernel through the /proc filesystem. Files and directories under /proc do not actually exist on disk; instead, they are generated by the kernel on demand whenever you read them. You can see a “directory” for each pid in /proc, and you can list each process’s file descriptors by examining the directory /proc/{pid}/fd. For example, inspecting zsh's file descriptors (you would see a different result if you are running bash):

# Prefix "ls" with "./container" if you're using Docker
🍉  ls /proc/$$/fd     # $$ is a shell variable containing your current shell's pid
0  1  10  11  12  14  2

In this milestone, we will implement Process::list_fds, which lists the contents of /proc/{pid}/fd to get the file descriptor numbers that the process has open.

Open src/process.rs and have a look at the list_fds method. This function will return a list of file descriptor numbers (Vec<usize>) if the file descriptor table is available; otherwise, it will return None. It is important to handle the case of missing file descriptor tables for the purposes of handling zombie processes, which are processes that have exited but have not yet been reaped by their parents (i.e. their parents have not yet called waitpid on them). The processes still exist in the process table, but most of their resources have been freed, including the file descriptor table.

To implement this function, you should use fs::read_dir(path) to open the directory. Then, iterate over the directory entries, parse the file names as usize, and build a vector of file descriptor numbers. Here are some helpful hints:

When you have implemented this function, modify your print() function to call list_fds(), loop over each file descriptor, and print each one. Try running your program, and make sure it produces the output you expect. You can also run the provided tests:

cargo test list_fds

(If using Docker, don’t forget to prefix with ./container.)

Milestone 4: Printing additional open file information

Let’s take a closer look at /proc/{pid}/fd:

🍉  ls -l /proc/$$/fd
total 0
lrwx------ 1 rebs operator 64 Apr 22 15:13 0 -> /dev/pts/38
lrwx------ 1 rebs operator 64 Apr 22 15:13 1 -> /dev/pts/38
lrwx------ 1 rebs operator 64 Apr 22 15:13 10 -> /dev/pts/38
lr-x------ 1 rebs operator 64 Apr 22 15:13 11 -> /dev/urandom
lrwx------ 1 rebs operator 64 Apr 22 15:13 12 -> socket:[17099833]
lr-x------ 1 rebs operator 64 Apr 22 15:13 14 -> /usr/share/zsh/functions/Completion.zwc
lrwx------ 1 rebs operator 64 Apr 22 15:13 2 -> /dev/pts/38

(If you’re using Docker, prefix ls with ./container, or run ./container bash to get a Linux shell that you can run multiple commands in.)

Each file in this directory is a symbolic link pointing to whatever file the file descriptor points to in the vnode table. Here, you can see that file descriptors 0, 1, and 2 point to /dev/pts/38, which is the file that is mapped to the terminal I currently have open.

We can get additional information about each file descriptor from /proc/{pid}/fdinfo/{fd}:

🍉  cat /proc/$$/fdinfo/0
pos:	0
flags:	0100002
mnt_id:	22

This tells us the cursor, as well as flags set on the open file table entry (which includes flags like O_RDONLY, O_WRONLY, O_WRONLY).

Open src/open_file.rs and skim the code that is in this file. In this milestone, you will need to implement OpenFile::from_fd:

You can use the supplied tests to check your work:

cargo test openfile_from_fd

(If using Docker, don’t forget to prefix with ./container.)

Once you have implemented this function, open process.rs again and go to your Process::print function. Instead of iterating over self.list_fds(), use self.list_open_files() to get the file descriptors along with the corresponding OpenFiles. You can use the following code to print file descriptors, although you are welcome to write your own if you like:

match self.list_open_files() {
    None => println!(
        "Warning: could not inspect file descriptors for this process! \
            It might have exited just as we were about to look at its fd table, \
            or it might have exited a while ago and is waiting for the parent \
            to reap it."
    Some(open_files) => {
        for (fd, file) in open_files {
                "{:<4} {:<15} cursor: {:<4} {}",
                format!("({})", file.access_mode),

When this is done, your inspect-fds should be looking pretty good! Try it out:

🍉  cargo run bash
    Finished dev [unoptimized + debuginfo] target(s) in 1.16s
     Running `target/debug/inspect-fds bash`
========== "bash" (pid 19018, ppid 18803) ==========
0    (read/write)    cursor: 0    <terminal>
1    (read/write)    cursor: 0    <terminal>
2    (read/write)    cursor: 0    <terminal>
3    (read)          cursor: 0    /dev/urandom
4    (read/write)    cursor: 0    socket:[16103476]
255  (read/write)    cursor: 0    <terminal>

🍉  ./zombie_test & cargo run zombie_test
[1] 20630
    Finished dev [unoptimized + debuginfo] target(s) in 1.09s
     Running `target/debug/inspect-fds zombie_test`
========== "./zombie_test" (pid 20630, ppid 18509) ==========
0    (read/write)    cursor: 0    <terminal>
1    (read/write)    cursor: 0    <terminal>
2    (read/write)    cursor: 0    <terminal>
4    (write)         cursor: 0    <pipe #16102316>

If using Docker, don’t forget to prefix with ./container. For the second test listed above, you should run it like this:

./container bash -c "./zombie_test & cargo run zombie_test"

Milestone 5: Inspecting child processes

For this tool to be most useful in debugging file descriptor issues (such as how pipes are wired up), we don’t want to only show information about one process; we should also show information about other related processes. For our purposes, let’s print the user-specified process along with all its child processes.

This involves a simple modification to your code in main.rs. After printing information about the user-specified process, call ps_utils::get_child_processes to get a list of child processes (again, it’s acceptable to call expect() here). Iterate over these processes, and call your print function on each of them.

Your output should look something like this:

🍉  ./zombie_test & cargo run zombie_test
[1] 20630
    Finished dev [unoptimized + debuginfo] target(s) in 1.09s
     Running `target/debug/inspect-fds zombie_test`
========== "./zombie_test" (pid 20630, ppid 18509) ==========
0    (read/write)    cursor: 0    <terminal>
1    (read/write)    cursor: 0    <terminal>
2    (read/write)    cursor: 0    <terminal>
4    (write)         cursor: 0    <pipe #16102316>

========== "[zombie_test] <defunct>" (pid 20632, ppid 20630) ==========
Warning: could not inspect file descriptors for this process! It might have exited just as we were about to look at its fd table, or it might have exited a while ago and is waiting for the parent to reap it.

🍉  ./multi_pipe_test & cargo run multi_pipe_test
[1] 4060
    Finished dev [unoptimized + debuginfo] target(s) in 1.04s
     Running `target/debug/inspect-fds multi_pipe_test`
========== "./multi_pipe_test" (pid 4060, ppid 18509) ==========
0    (read/write)    cursor: 0    <terminal>
1    (read/write)    cursor: 0    <terminal>
2    (read/write)    cursor: 0    <terminal>
4    (write)         cursor: 0    <pipe #16301201>
5    (read)          cursor: 0    <pipe #16301202>

========== "./multi_pipe_test" (pid 4062, ppid 4060) ==========
0    (read)          cursor: 0    <pipe #16301201>
1    (write)         cursor: 0    <pipe #16301202>
2    (read/write)    cursor: 0    <terminal>

If you’re using Docker, you’ll need to run this instead:

./container bash -c "./zombie_test & cargo run zombie_test"
./container bash -c "./multi_pipe_test & cargo run multi_pipe_test"

That’s it! You’re done! You’ll be able to use this tool to debug your CS 110 assignment 3 and assignment 4 code.

Part 2: Weekly survey

Please let us know how you’re doing using this survey.

When you have submitted the survey, you should see a password. Put this code in survey.txt before submitting.

Submitting your work

As with last week, you can commit your progress using git:

git commit -am "Type some title here to identify this snapshot!"

In order to submit your work, commit it, then run git push. This will upload your commits (snapshots) to Github, where we can access them. You can verify that your code is submitted by visiting https://github.com/cs110l/week3-yourSunetid and browsing the code there. You can git push as many times as you’d like.


Part 1 (inspect-fds) will be worth 85% and Part 2 (the survey) will be worth 15%. Each milestone of Part 1 will be worth 17%. You’ll earn the full credit for each piece if we can see that you’ve made a good-faith effort to complete it.