// A version of pstree targetting linux written in rust! // // This is based on the following exercise from the excellent // book "The Linux Programming Interface" by Michael Kerridsk. // //---------------------------------------------------------------------- // // Write a program that draws a tree showing the hierarchical // parent-child relationships of all processes on the system, going all // the way back to init. For each process, the program should display // the process ID and the command being executed. The output of the // program should be similar to that produced by pstree(1), although it // does need not to be as sophisticated. The parent of each process on // the system can be found by inspecing the PPid: line of all of the // /proc/PID/status files on the system. Be careful to handle the // possibilty that a process's parent (and thus its /proc/PID directory) // disappears during the scan of all /proc/PID directories. // Implementation Notes // -------------------- // The linux /proc filesystem is a virtual filesystem that provides information // about processes running on a linux system among other things. The /proc // filesystem contains a directory, /proc/, for each running process in // the system. // // Each process directory has a status file with contents including a bunch // of different items, notably the process name and its parent process id (ppid). // And with that information, we can build the process tree. use std::path::Path; use std::fs; use std::io::prelude::*; use std::fs::File; use std::collections::hash_map::Entry::{Occupied, Vacant}; use std::collections::HashMap; #[derive(Clone,Debug)] struct ProcessRecord { name: String, pid: i32, ppid: i32, } #[derive(Clone,Debug)] struct ProcessTreeNode { record: ProcessRecord, // the node owns the associated record children: Vec, // nodes own their children } #[derive(Clone,Debug)] struct ProcessTree { root: ProcessTreeNode, // tree owns ref to root node } impl ProcessTreeNode { // constructor fn new(record : &ProcessRecord) -> ProcessTreeNode { ProcessTreeNode { record: (*record).clone(), children: Vec::new() } } } // Given a status file path, return a hashmap with the following form: // pid -> ProcessRecord fn get_process_record(status_path: &Path) -> Option { let mut pid : Option = None; let mut ppid : Option = None; let mut name : Option = None; let mut reader = std::io::BufReader::new(File::open(status_path).unwrap()); loop { let mut linebuf = String::new(); match reader.read_line(&mut linebuf) { Ok(_) => { if linebuf.is_empty() { break; } let parts : Vec<&str> = linebuf[..].splitn(2, ':').collect(); if parts.len() == 2 { let key = parts[0].trim(); let value = parts[1].trim(); match key { "Name" => name = Some(value.to_string()), "Pid" => pid = value.parse().ok(), "PPid" => ppid = value.parse().ok(), _ => (), } } }, Err(_) => break, } } return if pid.is_some() && ppid.is_some() && name.is_some() { Some(ProcessRecord { name: name.unwrap(), pid: pid.unwrap(), ppid: ppid.unwrap() }) } else { None } } // build a simple struct (ProcessRecord) for each process fn get_process_records() -> Vec { let proc_directory = Path::new("/proc"); // find potential process directories under /proc let proc_directory_contents = fs::read_dir(&proc_directory).unwrap(); proc_directory_contents.filter_map(|entry| { let entry_path = entry.unwrap().path(); if fs::metadata(entry_path.as_path()).unwrap().is_dir() { let status_path = entry_path.join("status"); if let Ok(metadata) = fs::metadata(status_path.as_path()) { if metadata.is_file() { return get_process_record(status_path.as_path()); } } } None }).collect() } fn populate_node_helper(node: &mut ProcessTreeNode, pid_map: &HashMap, ppid_map: &HashMap>) { let pid = node.record.pid; // avoid binding node as immutable in closure let child_nodes = &mut node.children; match ppid_map.get(&pid) { Some(children) => { child_nodes.extend(children.iter().map(|child_pid| { let record = pid_map[child_pid]; let mut child = ProcessTreeNode::new(record); populate_node_helper(&mut child, pid_map, ppid_map); child })); }, None => {}, } } fn populate_node(node : &mut ProcessTreeNode, records: &Vec) { // O(n): build a mapping of pids to vectors of children. That is, each // key is a pid and its value is a vector of the whose parent pid is the key let mut ppid_map : HashMap> = HashMap::new(); let mut pid_map : HashMap = HashMap::new(); for record in records.iter() { // entry returns either a vacant or occupied entry. If vacant, // we insert a new vector with this records pid. If occupied, // we push this record's pid onto the vec pid_map.insert(record.pid, record); match ppid_map.entry(record.ppid) { Vacant(entry) => { entry.insert(vec![record.pid]); }, Occupied(mut entry) => { entry.get_mut().push(record.pid); }, }; } // With the data structures built, it is off to the races populate_node_helper(node, &pid_map, &ppid_map); } fn build_process_tree() -> ProcessTree { let records = get_process_records(); let mut tree = ProcessTree { root : ProcessTreeNode::new( &ProcessRecord { name: "/".to_string(), pid: 0, ppid: -1 }) }; // recursively populate all nodes in the tree starting from root (pid 0) { let root = &mut tree.root; populate_node(root, &records); } tree } fn print_node(node : &ProcessTreeNode, indent_level : i32) { // print indentation for _ in (0..indent_level) { print!(" "); } println!("- {} #{}", node.record.name, node.record.pid); for child in node.children.iter() { print_node(child, indent_level + 1); // recurse } } fn main() { let ptree = build_process_tree(); print_node(&(ptree.root), 0) }