Merge pull request #168 from kata-containers/release-0.5.1

release: v0.5.1

.github/workflows/bvt.yaml

@@ -1,7 +1,7 @@
 name: BVT
 on: [pull_request]
 env:
-  RUST_VERSION: 1.52
+  RUST_VERSION: 1.90.0
 jobs:
   build:
     name: Build
@@ -35,4 +35,3 @@ jobs:
       - run: rustup install ${{ env.RUST_VERSION }} && rustup default ${{ env.RUST_VERSION }}
       - run: make test
 
-

.gitignore

@@ -8,7 +8,3 @@ Cargo.lock
 
 # These are backup files generated by rustfmt
 **/*.rs.bk
-
-/target
-**/*.rs.bk
-Cargo.lock

Cargo.toml

@@ -5,7 +5,7 @@ repository = "https://github.com/kata-containers/cgroups-rs"
 keywords = ["linux", "cgroup", "containers", "isolation"]
 categories = ["os", "api-bindings", "os::unix-apis"]
 license = "MIT OR Apache-2.0"
-version = "0.2.11"
+version = "0.5.1"
 authors = ["The Kata Containers community <kata-dev@lists.katacontainers.io>", "Levente Kurusa <lkurusa@acm.org>", "Sam Wilson <tecywiz121@hotmail.com>"]
 edition = "2018"
 homepage = "https://github.com/kata-containers/cgroups-rs"
@@ -13,13 +13,19 @@ readme = "README.md"
 
 [dependencies]
 log = "0.4"
-regex = "1.1"
-nix = { version = "0.24", default-features = false, features = ["event", "fs", "process"] }
+nix = { version = "0.25.0", default-features = false, features = ["event", "fs", "process"] }
 libc = "0.2"
 serde = { version = "1.0", features = ["derive"], optional = true }
+thiserror = "1"
+oci-spec = { version = "0.10.0", optional = true }
+zbus = "5.8"
+bit-vec = "0.6"
 
 [dev-dependencies]
 libc = "0.2.76"
+rand = "0.8"
+nix = "0.25"
 
 [features]
 default = []
+oci = ["oci-spec"]

Makefile

@@ -19,9 +19,37 @@ build: debug
 # Tests and linters
 #
 
-.PHONY: test
-test:
-	cargo test -- --color always --nocapture
+# Tests that manipulate cgroups should run in sequence, so that
+# `--test-threads=1` is used.
+test: test-systemd test-fs-manager test-systemd-manager
+	cargo test --all-features -- --color always \
+	  --nocapture \
+	  --skip systemd::dbus::client::tests \
+	  --skip manager::fs::tests \
+	  --skip manager::systemd::tests
+
+.PHONY: test-systemd
+# Tests that manipulate cgroups should run in sequence, so that
+# `--test-threads=1` is used.
+test-systemd:
+	cargo test --package cgroups-rs --lib \
+	    -- systemd::dbus::client::tests \
+	    --color always --nocapture \
+		--test-threads=1
+
+.PHONY: test-fs-manager
+# See test-systemd
+test-fs-manager:
+	cargo test --all-features --package cgroups-rs \
+		--lib -- manager::fs::tests \
+		--color always --nocapture --test-threads=1
+
+.PHONY: test-systemd-manager
+# See test-systemd
+test-systemd-manager:
+	cargo test --all-features --package cgroups-rs \
+	  --lib -- manager::systemd::tests \
+	  --color always --nocapture --test-threads=1
 
 .PHONY: check
 check: fmt clippy

README.md

@@ -1,15 +1,14 @@
-# cgroups-rs ![Build](https://travis-ci.org/kata-containers/cgroups-rs.svg?branch=master)
-Native Rust library for managing control groups under Linux
+# cgroups-rs
 
-Both v1 and v2 of cgroups are supported.
+[![Crate](https://img.shields.io/crates/v/cgroups-rs.svg)](https://crates.io/crates/cgroups-rs)
 
-# Examples
+Native Rust library for managing Linux control groups. Supports both cgroups v1 and v2.
 
-## Create a control group using the builder pattern
-
-``` rust
+## Examples
 
+### Create a control group using the builder pattern
 
+```rust
 use cgroups_rs::*;
 use cgroups_rs::cgroup_builder::*;
 
@@ -44,7 +43,7 @@ cg.delete();
 // major version change.
 ```
 
-# Disclaimer
+## Disclaimer
 
 This crate is licensed under:
 

src/cgroup.rs

@@ -1,386 +0,0 @@
-// Copyright (c) 2018 Levente Kurusa
-// Copyright (c) 2020 Ant Group
-//
-// SPDX-License-Identifier: Apache-2.0 or MIT
-//
-
-//! This module handles cgroup operations. Start here!
-
-use crate::error::ErrorKind::*;
-use crate::error::*;
-
-use crate::{CgroupPid, ControllIdentifier, Controller, Hierarchy, Resources, Subsystem};
-
-use std::collections::HashMap;
-use std::convert::From;
-use std::fs;
-use std::path::{Path, PathBuf};
-
-/// A control group is the central structure to this crate.
-///
-///
-/// # What are control groups?
-///
-/// Lifting over from the Linux kernel sources:
-///
-/// > Control Groups provide a mechanism for aggregating/partitioning sets of
-/// > tasks, and all their future children, into hierarchical groups with
-/// > specialized behaviour.
-///
-/// This crate is an attempt at providing a Rust-native way of managing these cgroups.
-#[derive(Debug)]
-pub struct Cgroup {
-    /// The list of subsystems that control this cgroup
-    subsystems: Vec<Subsystem>,
-
-    /// The hierarchy.
-    hier: Box<dyn Hierarchy>,
-    path: String,
-}
-
-impl Clone for Cgroup {
-    fn clone(&self) -> Self {
-        Cgroup {
-            subsystems: self.subsystems.clone(),
-            path: self.path.clone(),
-            hier: crate::hierarchies::auto(),
-        }
-    }
-}
-
-impl Default for Cgroup {
-    fn default() -> Self {
-        Cgroup {
-            subsystems: Vec::new(),
-            hier: crate::hierarchies::auto(),
-            path: "".to_string(),
-        }
-    }
-}
-
-impl Cgroup {
-    /// Create this control group.
-    fn create(&self) {
-        if self.hier.v2() {
-            let _ret = create_v2_cgroup(self.hier.root(), &self.path);
-        } else {
-            for subsystem in &self.subsystems {
-                subsystem.to_controller().create();
-            }
-        }
-    }
-
-    pub fn v2(&self) -> bool {
-        self.hier.v2()
-    }
-
-    /// Create a new control group in the hierarchy `hier`, with name `path`.
-    ///
-    /// Returns a handle to the control group that can be used to manipulate it.
-    pub fn new<P: AsRef<Path>>(hier: Box<dyn Hierarchy>, path: P) -> Cgroup {
-        let cg = Cgroup::load(hier, path);
-        cg.create();
-        cg
-    }
-
-    /// Create a new control group in the hierarchy `hier`, with name `path` and `relative_paths`
-    ///
-    /// Returns a handle to the control group that can be used to manipulate it.
-    ///
-    /// Note that this method is only meaningful for cgroup v1, call it is equivalent to call `new` in the v2 mode
-    pub fn new_with_relative_paths<P: AsRef<Path>>(
-        hier: Box<dyn Hierarchy>,
-        path: P,
-        relative_paths: HashMap<String, String>,
-    ) -> Cgroup {
-        let cg = Cgroup::load_with_relative_paths(hier, path, relative_paths);
-        cg.create();
-        cg
-    }
-
-    /// Create a handle for a control group in the hierarchy `hier`, with name `path`.
-    ///
-    /// Returns a handle to the control group (that possibly does not exist until `create()` has
-    /// been called on the cgroup.
-    pub fn load<P: AsRef<Path>>(hier: Box<dyn Hierarchy>, path: P) -> Cgroup {
-        let path = path.as_ref();
-        let mut subsystems = hier.subsystems();
-        if path.as_os_str() != "" {
-            subsystems = subsystems
-                .into_iter()
-                .map(|x| x.enter(path))
-                .collect::<Vec<_>>();
-        }
-
-        Cgroup {
-            path: path.to_str().unwrap().to_string(),
-            subsystems,
-            hier,
-        }
-    }
-
-    /// Create a handle for a control group in the hierarchy `hier`, with name `path` and `relative_paths`
-    ///
-    /// Returns a handle to the control group (that possibly does not exist until `create()` has
-    /// been called on the cgroup.
-    ///
-    /// Note that this method is only meaningful for cgroup v1, call it is equivalent to call `load` in the v2 mode
-    pub fn load_with_relative_paths<P: AsRef<Path>>(
-        hier: Box<dyn Hierarchy>,
-        path: P,
-        relative_paths: HashMap<String, String>,
-    ) -> Cgroup {
-        // relative_paths only valid for cgroup v1
-        if hier.v2() {
-            return Self::load(hier, path);
-        }
-
-        let path = path.as_ref();
-        let mut subsystems = hier.subsystems();
-        if path.as_os_str() != "" {
-            subsystems = subsystems
-                .into_iter()
-                .map(|x| {
-                    let cn = x.controller_name();
-                    if relative_paths.contains_key(&cn) {
-                        let rp = relative_paths.get(&cn).unwrap();
-                        let valid_path = rp.trim_start_matches('/').to_string();
-                        let mut p = PathBuf::from(valid_path);
-                        p.push(path);
-                        x.enter(p.as_ref())
-                    } else {
-                        x.enter(path)
-                    }
-                })
-                .collect::<Vec<_>>();
-        }
-
-        Cgroup {
-            subsystems,
-            hier,
-            path: path.to_str().unwrap().to_string(),
-        }
-    }
-
-    /// The list of subsystems that this control group supports.
-    pub fn subsystems(&self) -> &Vec<Subsystem> {
-        &self.subsystems
-    }
-
-    /// Deletes the control group.
-    ///
-    /// Note that this function makes no effort in cleaning up the descendant and the underlying
-    /// system call will fail if there are any descendants. Thus, one should check whether it was
-    /// actually removed, and remove the descendants first if not. In the future, this behavior
-    /// will change.
-    pub fn delete(&self) -> Result<()> {
-        if self.v2() {
-            if !self.path.is_empty() {
-                let mut p = self.hier.root();
-                p.push(self.path.clone());
-                return fs::remove_dir(p).map_err(|e| Error::with_cause(RemoveFailed, e));
-            }
-            return Ok(());
-        }
-
-        self.subsystems.iter().try_for_each(|sub| match sub {
-            Subsystem::Pid(pidc) => pidc.delete(),
-            Subsystem::Mem(c) => c.delete(),
-            Subsystem::CpuSet(c) => c.delete(),
-            Subsystem::CpuAcct(c) => c.delete(),
-            Subsystem::Cpu(c) => c.delete(),
-            Subsystem::Devices(c) => c.delete(),
-            Subsystem::Freezer(c) => c.delete(),
-            Subsystem::NetCls(c) => c.delete(),
-            Subsystem::BlkIo(c) => c.delete(),
-            Subsystem::PerfEvent(c) => c.delete(),
-            Subsystem::NetPrio(c) => c.delete(),
-            Subsystem::HugeTlb(c) => c.delete(),
-            Subsystem::Rdma(c) => c.delete(),
-            Subsystem::Systemd(c) => c.delete(),
-        })
-    }
-
-    /// Apply a set of resource limits to the control group.
-    pub fn apply(&self, res: &Resources) -> Result<()> {
-        self.subsystems
-            .iter()
-            .try_fold((), |_, e| e.to_controller().apply(res))
-    }
-
-    /// Retrieve a container based on type inference.
-    ///
-    /// ## Example:
-    ///
-    /// ```text
-    /// let pids: &PidController = control_group.controller_of()
-    ///                             .expect("No pids controller attached!");
-    /// let cpu: &CpuController = control_group.controller_of()
-    ///                             .expect("No cpu controller attached!");
-    /// ```
-    pub fn controller_of<'a, T>(&'a self) -> Option<&'a T>
-    where
-        &'a T: From<&'a Subsystem>,
-        T: Controller + ControllIdentifier,
-    {
-        for i in &self.subsystems {
-            if i.to_controller().control_type() == T::controller_type() {
-                // N.B.:
-                // https://play.rust-lang.org/?gist=978b2846bacebdaa00be62374f4f4334&version=stable&mode=debug&edition=2015
-                return Some(i.into());
-            }
-        }
-        None
-    }
-
-    /// Removes a task from the control group.
-    ///
-    /// Note that this means that the task will be moved back to the root control group in the
-    /// hierarchy and any rules applied to that control group will _still_ apply to the task.
-    pub fn remove_task(&self, pid: CgroupPid) {
-        let _ = self.hier.root_control_group().add_task(pid);
-    }
-
-    /// Attach a task to the control group.
-    pub fn add_task(&self, pid: CgroupPid) -> Result<()> {
-        if self.v2() {
-            let subsystems = self.subsystems();
-            if !subsystems.is_empty() {
-                let c = subsystems[0].to_controller();
-                c.add_task(&pid)
-            } else {
-                Ok(())
-            }
-        } else {
-            self.subsystems()
-                .iter()
-                .try_for_each(|sub| sub.to_controller().add_task(&pid))
-        }
-    }
-
-    /// Attach a task to the control group by thread group id.
-    pub fn add_task_by_tgid(&self, pid: CgroupPid) -> Result<()> {
-        self.subsystems()
-            .iter()
-            .try_for_each(|sub| sub.to_controller().add_task_by_tgid(&pid))
-    }
-
-    /// Set notify_on_release to the control group.
-    pub fn set_notify_on_release(&self, enable: bool) -> Result<()> {
-        self.subsystems()
-            .iter()
-            .try_for_each(|sub| sub.to_controller().set_notify_on_release(enable))
-    }
-
-    /// Set release_agent
-    pub fn set_release_agent(&self, path: &str) -> Result<()> {
-        self.hier
-            .root_control_group()
-            .subsystems()
-            .iter()
-            .try_for_each(|sub| sub.to_controller().set_release_agent(path))
-    }
-
-    /// Returns an Iterator that can be used to iterate over the tasks that are currently in the
-    /// control group.
-    pub fn tasks(&self) -> Vec<CgroupPid> {
-        // Collect the tasks from all subsystems
-        let mut v = if self.v2() {
-            let subsystems = self.subsystems();
-            if !subsystems.is_empty() {
-                let c = subsystems[0].to_controller();
-                c.tasks()
-            } else {
-                vec![]
-            }
-        } else {
-            self.subsystems()
-                .iter()
-                .map(|x| x.to_controller().tasks())
-                .fold(vec![], |mut acc, mut x| {
-                    acc.append(&mut x);
-                    acc
-                })
-        };
-
-        v.sort();
-        v.dedup();
-        v
-    }
-}
-
-pub const UNIFIED_MOUNTPOINT: &str = "/sys/fs/cgroup";
-
-fn enable_controllers(controllers: &[String], path: &Path) {
-    let f = path.join("cgroup.subtree_control");
-    for c in controllers {
-        let body = format!("+{}", c);
-        let _rest = fs::write(f.as_path(), body.as_bytes());
-    }
-}
-
-fn supported_controllers() -> Vec<String> {
-    let p = format!("{}/{}", UNIFIED_MOUNTPOINT, "cgroup.controllers");
-    let ret = fs::read_to_string(p.as_str());
-    ret.unwrap_or_default()
-        .split(' ')
-        .map(|x| x.to_string())
-        .collect::<Vec<String>>()
-}
-
-fn create_v2_cgroup(root: PathBuf, path: &str) -> Result<()> {
-    // controler list ["memory", "cpu"]
-    let controllers = supported_controllers();
-    let mut fp = root;
-
-    // enable for root
-    enable_controllers(&controllers, &fp);
-
-    // path: "a/b/c"
-    let elements = path.split('/').collect::<Vec<&str>>();
-    let last_index = elements.len() - 1;
-    for (i, ele) in elements.iter().enumerate() {
-        // ROOT/a
-        fp.push(ele);
-        // create dir, need not check if is a file or directory
-        if !fp.exists() {
-            if let Err(e) = std::fs::create_dir(fp.clone()) {
-                return Err(Error::with_cause(ErrorKind::FsError, e));
-            }
-        }
-
-        if i < last_index {
-            // enable controllers for substree
-            enable_controllers(&controllers, &fp);
-        }
-    }
-
-    Ok(())
-}
-
-pub fn get_cgroups_relative_paths() -> Result<HashMap<String, String>> {
-    let path = "/proc/self/cgroup".to_string();
-    get_cgroups_relative_paths_by_path(path)
-}
-
-pub fn get_cgroups_relative_paths_by_pid(pid: u32) -> Result<HashMap<String, String>> {
-    let path = format!("/proc/{}/cgroup", pid);
-    get_cgroups_relative_paths_by_path(path)
-}
-
-fn get_cgroups_relative_paths_by_path(path: String) -> Result<HashMap<String, String>> {
-    let mut m = HashMap::new();
-    let content = fs::read_to_string(path).map_err(|e| Error::with_cause(ReadFailed, e))?;
-    for l in content.lines() {
-        let fl: Vec<&str> = l.split(':').collect();
-        if fl.len() != 3 {
-            continue;
-        }
-
-        let keys: Vec<&str> = fl[1].split(',').collect();
-        for key in &keys {
-            m.insert(key.to_string(), fl[2].to_string());
-        }
-    }
-    Ok(m)
-}

src/fs/blkio.rs

@@ -11,11 +11,11 @@
 use std::io::Write;
 use std::path::PathBuf;
 
-use crate::error::ErrorKind::*;
-use crate::error::*;
+use crate::fs::error::ErrorKind::*;
+use crate::fs::error::*;
 
-use crate::{read_string_from, read_u64_from};
-use crate::{
+use crate::fs::{read_string_from, read_u64_from};
+use crate::fs::{
     BlkIoResources, ControllIdentifier, ControllerInternal, Controllers, CustomizedAttribute,
     Resources, Subsystem,
 };
@@ -43,7 +43,7 @@ pub struct BlkIoData {
     pub data: u64,
 }
 
-#[derive(Eq, PartialEq, Debug)]
+#[derive(Eq, PartialEq, Debug, Default)]
 #[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
 /// Per-device activity from the control group.
 pub struct IoService {
@@ -59,6 +59,8 @@ pub struct IoService {
     pub sync: u64,
     /// How many items were asynchronously transferred.
     pub r#async: u64,
+    /// How many items were discarded.
+    pub discard: u64,
     /// Total number of items transferred.
     pub total: u64,
 }
@@ -87,44 +89,62 @@ pub struct IoStat {
 }
 
 fn parse_io_service(s: String) -> Result<Vec<IoService>> {
-    s.lines()
+    let mut io_services = Vec::<IoService>::new();
+    let mut io_service = IoService::default();
+
+    let lines = s
+        .lines()
         .filter(|x| x.split_whitespace().count() == 3)
         .map(|x| {
             let mut spl = x.split_whitespace();
-            (spl.next().unwrap(), spl.next().unwrap(), spl.next().unwrap())
+            (
+                spl.next().unwrap(),
+                spl.next().unwrap(),
+                spl.next().unwrap(),
+            )
         })
         .map(|(a, b, c)| {
             let mut spl = a.split(':');
-            (spl.next().unwrap(), spl.next().unwrap(), b, c)
-        })
-        .collect::<Vec<_>>()
-        .chunks(5)
-        .map(|x| {
-            match x {
-                [(major, minor, "Read", read_val), (_, _, "Write", write_val),
-                   (_, _, "Sync", sync_val), (_, _, "Async", async_val),
-                   (_, _, "Total", total_val)] =>
-                    Some(IoService {
-                        major: major.parse::<i16>().unwrap(),
-                        minor: minor.parse::<i16>().unwrap(),
-                        read: read_val.parse::<u64>().unwrap(),
-                        write: write_val.parse::<u64>().unwrap(),
-                        sync: sync_val.parse::<u64>().unwrap(),
-                        r#async: async_val.parse::<u64>().unwrap(),
-                        total: total_val.parse::<u64>().unwrap(),
-                    }),
-               _ => None,
-            }
-        })
-        .fold(Ok(Vec::new()), |acc, x| {
-            if acc.is_err() || x.is_none() {
-                Err(Error::new(ParseError))
-            } else {
-                let mut acc = acc.unwrap();
-                acc.push(x.unwrap());
-                Ok(acc)
-            }
+            (
+                spl.next().unwrap().parse::<i16>(),
+                spl.next().unwrap().parse::<i16>(),
+                b,
+                c,
+            )
         })
+        .collect::<Vec<_>>();
+
+    for (major_num, minor_num, op, val) in lines.iter() {
+        let major = *major_num.as_ref().map_err(|_| Error::new(ParseError))?;
+        let minor = *minor_num.as_ref().map_err(|_| Error::new(ParseError))?;
+
+        if (major != io_service.major || minor != io_service.minor) && io_service.major != 0 {
+            // new block device
+            io_services.push(io_service);
+            io_service = IoService::default();
+        }
+
+        io_service.major = major;
+        io_service.minor = minor;
+
+        let val = val.parse::<u64>().map_err(|_| Error::new(ParseError))?;
+
+        match *op {
+            "Read" => io_service.read = val,
+            "Write" => io_service.write = val,
+            "Sync" => io_service.sync = val,
+            "Async" => io_service.r#async = val,
+            "Discard" => io_service.discard = val,
+            "Total" => io_service.total = val,
+            _ => {}
+        }
+    }
+
+    if io_service.major != 0 {
+        io_services.push(io_service);
+    }
+
+    Ok(io_services)
 }
 
 fn get_value(s: &str) -> String {
@@ -161,13 +181,14 @@ fn parse_io_stat(s: String) -> Vec<IoStat> {
 
 fn parse_io_service_total(s: String) -> Result<u64> {
     s.lines()
-        .filter(|x| x.split_whitespace().count() == 2)
-        .fold(Err(Error::new(ParseError)), |_, x| {
-            match x.split_whitespace().collect::<Vec<_>>().as_slice() {
-                ["Total", val] => val.parse::<u64>().map_err(|_| Error::new(ParseError)),
-                _ => Err(Error::new(ParseError)),
+        .find_map(|line| {
+            let mut parts = line.split_whitespace();
+            match (parts.next(), parts.next(), parts.next()) {
+                (Some("Total"), Some(val), None) => val.parse::<u64>().ok(),
+                _ => None,
             }
         })
+        .ok_or_else(|| Error::new(ParseError))
 }
 
 fn parse_blkio_data(s: String) -> Result<Vec<BlkIoData>> {
@@ -410,10 +431,10 @@ impl<'a> From<&'a Subsystem> for &'a BlkIoController {
 
 impl BlkIoController {
     /// Constructs a new `BlkIoController` with `root` serving as the root of the control group.
-    pub fn new(root: PathBuf, v2: bool) -> Self {
+    pub fn new(point: PathBuf, root: PathBuf, v2: bool) -> Self {
         Self {
-            base: root.clone(),
-            path: root,
+            base: root,
+            path: point,
             v2,
         }
     }
@@ -665,8 +686,12 @@ impl BlkIoController {
     pub fn set_leaf_weight(&self, w: u64) -> Result<()> {
         self.open_path("blkio.leaf_weight", true)
             .and_then(|mut file| {
-                file.write_all(w.to_string().as_ref())
-                    .map_err(|e| Error::with_cause(WriteFailed, e))
+                file.write_all(w.to_string().as_ref()).map_err(|e| {
+                    Error::with_cause(
+                        WriteFailed("blkio.leaf_weight".to_string(), w.to_string()),
+                        e,
+                    )
+                })
             })
     }
 
@@ -675,7 +700,15 @@ impl BlkIoController {
         self.open_path("blkio.leaf_weight_device", true)
             .and_then(|mut file| {
                 file.write_all(format!("{}:{} {}", major, minor, weight).as_ref())
-                    .map_err(|e| Error::with_cause(WriteFailed, e))
+                    .map_err(|e| {
+                        Error::with_cause(
+                            WriteFailed(
+                                "blkio.leaf_weight_device".to_string(),
+                                format!("{}:{} {}", major, minor, weight),
+                            ),
+                            e,
+                        )
+                    })
             })
     }
 
@@ -683,95 +716,112 @@ impl BlkIoController {
     pub fn reset_stats(&self) -> Result<()> {
         self.open_path("blkio.reset_stats", true)
             .and_then(|mut file| {
-                file.write_all("1".to_string().as_ref())
-                    .map_err(|e| Error::with_cause(WriteFailed, e))
+                file.write_all("1".to_string().as_ref()).map_err(|e| {
+                    Error::with_cause(
+                        WriteFailed("blkio.reset_stats".to_string(), "1".to_string()),
+                        e,
+                    )
+                })
             })
     }
 
     /// Throttle the bytes per second rate of read operation affecting the block device
     /// `major:minor` to `bps`.
     pub fn throttle_read_bps_for_device(&self, major: u64, minor: u64, bps: u64) -> Result<()> {
-        let mut file = "blkio.throttle.read_bps_device";
+        let mut file_name = "blkio.throttle.read_bps_device";
         let mut content = format!("{}:{} {}", major, minor, bps);
         if self.v2 {
-            file = "io.max";
+            file_name = "io.max";
             content = format!("{}:{} rbps={}", major, minor, bps);
         }
-        self.open_path(file, true).and_then(|mut file| {
-            file.write_all(content.as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+        self.open_path(file_name, true).and_then(|mut file| {
+            file.write_all(content.as_ref()).map_err(|e| {
+                Error::with_cause(WriteFailed(file_name.to_string(), content.to_string()), e)
+            })
         })
     }
 
     /// Throttle the I/O operations per second rate of read operation affecting the block device
     /// `major:minor` to `bps`.
     pub fn throttle_read_iops_for_device(&self, major: u64, minor: u64, iops: u64) -> Result<()> {
-        let mut file = "blkio.throttle.read_iops_device";
+        let mut file_name = "blkio.throttle.read_iops_device";
         let mut content = format!("{}:{} {}", major, minor, iops);
         if self.v2 {
-            file = "io.max";
+            file_name = "io.max";
             content = format!("{}:{} riops={}", major, minor, iops);
         }
-        self.open_path(file, true).and_then(|mut file| {
-            file.write_all(content.as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+        self.open_path(file_name, true).and_then(|mut file| {
+            file.write_all(content.as_ref()).map_err(|e| {
+                Error::with_cause(WriteFailed(file_name.to_string(), content.to_string()), e)
+            })
         })
     }
     /// Throttle the bytes per second rate of write operation affecting the block device
     /// `major:minor` to `bps`.
     pub fn throttle_write_bps_for_device(&self, major: u64, minor: u64, bps: u64) -> Result<()> {
-        let mut file = "blkio.throttle.write_bps_device";
+        let mut file_name = "blkio.throttle.write_bps_device";
         let mut content = format!("{}:{} {}", major, minor, bps);
         if self.v2 {
-            file = "io.max";
+            file_name = "io.max";
             content = format!("{}:{} wbps={}", major, minor, bps);
         }
-        self.open_path(file, true).and_then(|mut file| {
-            file.write_all(content.as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+        self.open_path(file_name, true).and_then(|mut file| {
+            file.write_all(content.as_ref()).map_err(|e| {
+                Error::with_cause(WriteFailed(file_name.to_string(), content.to_string()), e)
+            })
         })
     }
 
     /// Throttle the I/O operations per second rate of write operation affecting the block device
     /// `major:minor` to `bps`.
     pub fn throttle_write_iops_for_device(&self, major: u64, minor: u64, iops: u64) -> Result<()> {
-        let mut file = "blkio.throttle.write_iops_device";
+        let mut file_name = "blkio.throttle.write_iops_device";
         let mut content = format!("{}:{} {}", major, minor, iops);
         if self.v2 {
-            file = "io.max";
+            file_name = "io.max";
             content = format!("{}:{} wiops={}", major, minor, iops);
         }
-        self.open_path(file, true).and_then(|mut file| {
-            file.write_all(content.as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+        self.open_path(file_name, true).and_then(|mut file| {
+            file.write_all(content.as_ref()).map_err(|e| {
+                Error::with_cause(WriteFailed(file_name.to_string(), content.to_string()), e)
+            })
         })
     }
 
     /// Set the weight of the control group's tasks.
     pub fn set_weight(&self, w: u64) -> Result<()> {
         // Attation: may not find in high kernel version.
-        let mut file = "blkio.weight";
+        let mut file_name = "blkio.weight";
         if self.v2 {
-            file = "io.bfq.weight";
+            file_name = "io.bfq.weight";
         }
-        self.open_path(file, true).and_then(|mut file| {
-            file.write_all(w.to_string().as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+        self.open_path(file_name, true).and_then(|mut file| {
+            file.write_all(w.to_string().as_ref()).map_err(|e| {
+                Error::with_cause(WriteFailed(file_name.to_string(), w.to_string()), e)
+            })
         })
     }
 
     /// Same as `set_weight()`, but settable per each block device.
     pub fn set_weight_for_device(&self, major: u64, minor: u64, weight: u64) -> Result<()> {
-        let mut file = "blkio.weight_device";
+        let mut file_name = "blkio.weight_device";
         if self.v2 {
             // Attation: there is no weight for device in runc
             // https://github.com/opencontainers/runc/blob/46be7b612e2533c494e6a251111de46d8e286ed5/libcontainer/cgroups/fs2/io.go#L30
             // may depends on IO schedulers https://wiki.ubuntu.com/Kernel/Reference/IOSchedulers
-            file = "io.bfq.weight";
+            file_name = "io.bfq.weight";
         }
-        self.open_path(file, true).and_then(|mut file| {
+        self.open_path(file_name, true).and_then(|mut file| {
             file.write_all(format!("{}:{} {}", major, minor, weight).as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+                .map_err(|e| {
+                    Error::with_cause(
+                        WriteFailed(
+                            file_name.to_string(),
+                            format!("{}:{} {}", major, minor, weight),
+                        ),
+                        e,
+                    )
+                })
         })
     }
 }
@@ -779,15 +829,16 @@ impl BlkIoController {
 impl CustomizedAttribute for BlkIoController {}
 #[cfg(test)]
 mod test {
-    use crate::blkio::{parse_blkio_data, BlkIoData};
-    use crate::blkio::{parse_io_service, parse_io_service_total, IoService};
-    use crate::error::*;
+    use crate::fs::blkio::{parse_blkio_data, BlkIoData};
+    use crate::fs::blkio::{parse_io_service, parse_io_service_total, IoService};
+    use crate::fs::error::*;
 
     static TEST_VALUE: &str = "\
 8:32 Read 4280320
 8:32 Write 0
 8:32 Sync 4280320
 8:32 Async 0
+8:32 Discard 1
 8:32 Total 4280320
 8:48 Read 5705479168
 8:48 Write 56096055296
@@ -804,28 +855,6 @@ mod test {
 8:0 Sync 7192576
 8:0 Async 0
 8:0 Total 7192576
-Total 61823067136
- ";
-
-    static TEST_WRONG_VALUE: &str = "\
-8:32 Read 4280320
-8:32 Write 0
-8:32 Async 0
-8:32 Total 4280320 8:48 Read 5705479168
-8:48 Write 56096055296
-8:48 Sync 11213923328
-8:48 Async 50587611136
-8:48 Total 61801534464
-8:16 Read 10059776
-8:16 Write 0
-8:16 Sync 10059776
-8:16 Async 0
-8:16 Total 10059776
-8:0 Read 7192576
-8:0 Write 0
-8:0 Sync 7192576
-8:0 Async 0
-8:0 Total 7192576
 Total 61823067136
  ";
 
@@ -855,6 +884,7 @@ Total 61823067136
                     write: 0,
                     sync: 4280320,
                     r#async: 0,
+                    discard: 1,
                     total: 4280320,
                 },
                 IoService {
@@ -864,6 +894,7 @@ Total 61823067136
                     write: 56096055296,
                     sync: 11213923328,
                     r#async: 50587611136,
+                    discard: 0,
                     total: 61801534464,
                 },
                 IoService {
@@ -873,6 +904,7 @@ Total 61823067136
                     write: 0,
                     sync: 10059776,
                     r#async: 0,
+                    discard: 0,
                     total: 10059776,
                 },
                 IoService {
@@ -882,12 +914,34 @@ Total 61823067136
                     write: 0,
                     sync: 7192576,
                     r#async: 0,
+                    discard: 0,
                     total: 7192576,
                 }
             ]
         );
-        let err = parse_io_service(TEST_WRONG_VALUE.to_string()).unwrap_err();
-        assert_eq!(err.kind(), &ErrorKind::ParseError,);
+
+        let invalid_values = vec![
+            "\
+8:32 Read 4280320
+8:32 Write a
+8:32 Async 1
+",
+            "\
+8:32 Read 4280320
+b:32 Write 1
+8:32 Async 1
+",
+            "\
+8:32 Read 4280320
+8:32 Write 1
+8:c Async 1
+",
+        ];
+
+        for value in invalid_values {
+            let err = parse_io_service(value.to_string()).unwrap_err();
+            assert_eq!(err.kind(), &ErrorKind::ParseError,);
+        }
     }
 
     #[test]

src/fs/cgroup.rs

@@ -0,0 +1,636 @@
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2020 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+//! This module handles cgroup operations. Start here!
+
+use crate::fs::error::ErrorKind::*;
+use crate::fs::error::*;
+
+use crate::fs::hierarchies::V1;
+use crate::fs::{CgroupPid, ControllIdentifier, Controller, Hierarchy, Resources, Subsystem};
+
+use std::collections::HashMap;
+use std::convert::From;
+use std::fs;
+use std::path::{Path, PathBuf};
+
+pub const CGROUP_MODE_DOMAIN: &str = "domain";
+pub const CGROUP_MODE_DOMAIN_THREADED: &str = "domain threaded";
+pub const CGROUP_MODE_DOMAIN_INVALID: &str = "domain invalid";
+pub const CGROUP_MODE_THREADED: &str = "threaded";
+
+/// A control group is the central structure to this crate.
+///
+///
+/// # What are control groups?
+///
+/// Lifting over from the Linux kernel sources:
+///
+/// > Control Groups provide a mechanism for aggregating/partitioning sets of
+/// > tasks, and all their future children, into hierarchical groups with
+/// > specialized behaviour.
+///
+/// This crate is an attempt at providing a Rust-native way of managing these cgroups.
+#[derive(Debug)]
+pub struct Cgroup {
+    /// The list of subsystems that control this cgroup
+    subsystems: Vec<Subsystem>,
+
+    /// The hierarchy.
+    hier: Box<dyn Hierarchy>,
+    path: String,
+
+    /// List of controllers specifically enabled in the control group.
+    specified_controllers: Option<Vec<String>>,
+}
+
+impl Clone for Cgroup {
+    fn clone(&self) -> Self {
+        Cgroup {
+            subsystems: self.subsystems.clone(),
+            hier: crate::fs::hierarchies::auto(),
+            path: self.path.clone(),
+            specified_controllers: None,
+        }
+    }
+}
+
+impl Default for Cgroup {
+    fn default() -> Self {
+        Cgroup {
+            subsystems: Vec::new(),
+            hier: crate::fs::hierarchies::auto(),
+            path: "".to_string(),
+            specified_controllers: None,
+        }
+    }
+}
+
+impl Cgroup {
+    pub fn v2(&self) -> bool {
+        self.hier.v2()
+    }
+
+    /// Return the path the cgroup is located at.
+    pub fn path(&self) -> &str {
+        &self.path
+    }
+
+    /// Create this control group.
+    pub fn create(&self) -> Result<()> {
+        if self.hier.v2() {
+            create_v2_cgroup(self.hier.root(), &self.path, &self.specified_controllers)
+        } else {
+            self.subsystems
+                .iter()
+                .try_for_each(|subsystem| subsystem.to_controller().create())?;
+            Ok(())
+        }
+    }
+
+    /// Create a new control group in the hierarchy `hier`, with name `path`.
+    ///
+    /// Returns a handle to the control group that can be used to manipulate it.
+    pub fn new<P: AsRef<Path>>(hier: Box<dyn Hierarchy>, path: P) -> Result<Cgroup> {
+        let cg = Cgroup::load(hier, path);
+        cg.create()?;
+        Ok(cg)
+    }
+
+    /// Create a new control group in the hierarchy `hier`, with name `path`.
+    ///
+    /// Returns a handle to the control group that can be used to manipulate it.
+    pub fn new_with_specified_controllers<P: AsRef<Path>>(
+        hier: Box<dyn Hierarchy>,
+        path: P,
+        specified_controllers: Option<Vec<String>>,
+    ) -> Result<Cgroup> {
+        let cg = if let Some(sc) = specified_controllers {
+            Cgroup::load_with_specified_controllers(hier, path, sc)
+        } else {
+            Cgroup::load(hier, path)
+        };
+        cg.create()?;
+        Ok(cg)
+    }
+
+    /// Create a new control group in the hierarchy `hier`, with name `path` and `relative_paths`
+    ///
+    /// Returns a handle to the control group that can be used to manipulate it.
+    ///
+    /// Note that this method is only meaningful for cgroup v1, call it is equivalent to call `new` in the v2 mode.
+    pub fn new_with_relative_paths<P: AsRef<Path>>(
+        hier: Box<dyn Hierarchy>,
+        path: P,
+        relative_paths: HashMap<String, String>,
+    ) -> Result<Cgroup> {
+        let cg = Cgroup::load_with_relative_paths(hier, path, relative_paths);
+        cg.create()?;
+        Ok(cg)
+    }
+
+    /// Create a handle for a control group in the hierarchy `hier`, with name `path`.
+    ///
+    /// Returns a handle to the control group (that possibly does not exist until `create()` has
+    /// been called on the cgroup.
+    pub fn load<P: AsRef<Path>>(hier: Box<dyn Hierarchy>, path: P) -> Cgroup {
+        let path = path.as_ref();
+        let mut subsystems = hier.subsystems();
+        if path.as_os_str() != "" {
+            subsystems = subsystems
+                .into_iter()
+                .map(|x| x.enter(path))
+                .collect::<Vec<_>>();
+        }
+
+        Cgroup {
+            path: path.to_str().unwrap().to_string(),
+            subsystems,
+            hier,
+            specified_controllers: None,
+        }
+    }
+
+    /// Create a handle for a specified control group in the hierarchy `hier`, with name `path`.
+    ///
+    /// Returns a handle to the control group (that possibly does not exist until `create()` has
+    /// been called on the cgroup.
+    pub fn load_with_specified_controllers<P: AsRef<Path>>(
+        hier: Box<dyn Hierarchy>,
+        path: P,
+        specified_controllers: Vec<String>,
+    ) -> Cgroup {
+        let path = path.as_ref();
+        let mut subsystems = hier.subsystems();
+        if path.as_os_str() != "" {
+            subsystems = subsystems
+                .into_iter()
+                .filter(|x| specified_controllers.contains(&x.controller_name()))
+                .map(|x| x.enter(path))
+                .collect::<Vec<_>>();
+        }
+
+        Cgroup {
+            path: path.to_str().unwrap().to_string(),
+            subsystems,
+            hier,
+            specified_controllers: Some(specified_controllers),
+        }
+    }
+
+    /// Create a handle for a control group in the hierarchy `hier`, with name `path` and `relative_paths`
+    ///
+    /// Returns a handle to the control group (that possibly does not exist until `create()` has
+    /// been called on the cgroup.
+    ///
+    /// Note that this method is only meaningful for cgroup v1, call it is equivalent to call `load` in the v2 mode
+    pub fn load_with_relative_paths<P: AsRef<Path>>(
+        hier: Box<dyn Hierarchy>,
+        path: P,
+        relative_paths: HashMap<String, String>,
+    ) -> Cgroup {
+        // relative_paths only valid for cgroup v1
+        if hier.v2() {
+            return Self::load(hier, path);
+        }
+
+        let path = path.as_ref();
+        let mut subsystems = hier.subsystems();
+        if path.as_os_str() != "" {
+            subsystems = subsystems
+                .into_iter()
+                .map(|x| {
+                    let cn = x.controller_name();
+                    if relative_paths.contains_key(&cn) {
+                        let rp = relative_paths.get(&cn).unwrap();
+                        let valid_path = rp.trim_start_matches('/').to_string();
+                        let mut p = PathBuf::from(valid_path);
+                        p.push(path);
+                        x.enter(p.as_ref())
+                    } else {
+                        x.enter(path)
+                    }
+                })
+                .collect::<Vec<_>>();
+        }
+
+        Cgroup {
+            subsystems,
+            hier,
+            path: path.to_str().unwrap().to_string(),
+            specified_controllers: None,
+        }
+    }
+
+    /// The list of subsystems that this control group supports.
+    pub fn subsystems(&self) -> &Vec<Subsystem> {
+        &self.subsystems
+    }
+
+    /// Deletes the control group.
+    ///
+    /// Note that this function makes no effort in cleaning up the descendant and the underlying
+    /// system call will fail if there are any descendants. Thus, one should check whether it was
+    /// actually removed, and remove the descendants first if not. In the future, this behavior
+    /// will change.
+    pub fn delete(&self) -> Result<()> {
+        if self.v2() {
+            if !self.path.is_empty() {
+                let mut p = self.hier.root();
+                p.push(self.path.clone());
+                return fs::remove_dir(p).map_err(|e| Error::with_cause(RemoveFailed, e));
+            }
+            return Ok(());
+        }
+
+        self.subsystems.iter().try_for_each(|sub| match sub {
+            Subsystem::Pid(pidc) => pidc.delete(),
+            Subsystem::Mem(c) => c.delete(),
+            Subsystem::CpuSet(c) => c.delete(),
+            Subsystem::CpuAcct(c) => c.delete(),
+            Subsystem::Cpu(c) => c.delete(),
+            Subsystem::Devices(c) => c.delete(),
+            Subsystem::Freezer(c) => c.delete(),
+            Subsystem::NetCls(c) => c.delete(),
+            Subsystem::BlkIo(c) => c.delete(),
+            Subsystem::PerfEvent(c) => c.delete(),
+            Subsystem::NetPrio(c) => c.delete(),
+            Subsystem::HugeTlb(c) => c.delete(),
+            Subsystem::Rdma(c) => c.delete(),
+            Subsystem::Systemd(c) => c.delete(),
+        })
+    }
+
+    /// Apply a set of resource limits to the control group.
+    pub fn apply(&self, res: &Resources) -> Result<()> {
+        self.subsystems
+            .iter()
+            .try_fold((), |_, e| e.to_controller().apply(res))
+    }
+
+    /// Retrieve a container based on type inference.
+    ///
+    /// ## Example:
+    ///
+    /// ```text
+    /// let pids: &PidController = control_group.controller_of()
+    ///                             .expect("No pids controller attached!");
+    /// let cpu: &CpuController = control_group.controller_of()
+    ///                             .expect("No cpu controller attached!");
+    /// ```
+    pub fn controller_of<'a, T>(&'a self) -> Option<&'a T>
+    where
+        &'a T: From<&'a Subsystem>,
+        T: Controller + ControllIdentifier,
+    {
+        for i in &self.subsystems {
+            if i.to_controller().control_type() == T::controller_type() {
+                // N.B.:
+                // https://play.rust-lang.org/?gist=978b2846bacebdaa00be62374f4f4334&version=stable&mode=debug&edition=2015
+                return Some(i.into());
+            }
+        }
+        None
+    }
+
+    /// Removes tasks from the control group by thread group id.
+    ///
+    /// Note that this means that the task will be moved back to the root control group in the
+    /// hierarchy and any rules applied to that control group will _still_ apply to the proc.
+    pub fn remove_task_by_tgid(&self, tgid: CgroupPid) -> Result<()> {
+        self.hier.root_control_group().add_task_by_tgid(tgid)
+    }
+
+    /// Removes a task from the control group.
+    ///
+    /// Note that this means that the task will be moved back to the root control group in the
+    /// hierarchy and any rules applied to that control group will _still_ apply to the task.
+    pub fn remove_task(&self, tid: CgroupPid) -> Result<()> {
+        self.hier.root_control_group().add_task(tid)
+    }
+
+    /// Moves tasks to the parent control group by thread group id.
+    pub fn move_task_to_parent_by_tgid(&self, tgid: CgroupPid) -> Result<()> {
+        self.hier
+            .parent_control_group(&self.path)
+            .add_task_by_tgid(tgid)
+    }
+
+    /// Moves a task to the parent control group.
+    pub fn move_task_to_parent(&self, tid: CgroupPid) -> Result<()> {
+        self.hier.parent_control_group(&self.path).add_task(tid)
+    }
+
+    /// Return a handle to the parent control group in the hierarchy.
+    pub fn parent_control_group(&self) -> Cgroup {
+        self.hier.parent_control_group(&self.path)
+    }
+
+    /// Kill every process in the control group. Only supported for v2 cgroups and on
+    /// kernels 5.14+. This will fail with InvalidOperation if the 'cgroup.kill' file does
+    /// not exist.
+    pub fn kill(&self) -> Result<()> {
+        if !self.v2() {
+            return Err(Error::new(CgroupVersion));
+        }
+
+        let val = "1";
+        let file_name = "cgroup.kill";
+        let p = self.hier.root().join(self.path.clone()).join(file_name);
+
+        // If cgroup.kill doesn't exist they're not on 5.14+ so lets
+        // surface some error the caller can check against.
+        if !p.exists() {
+            return Err(Error::new(InvalidOperation));
+        }
+
+        fs::write(p, val)
+            .map_err(|e| Error::with_cause(WriteFailed(file_name.to_string(), val.to_string()), e))
+    }
+
+    /// Attach a task to the control group.
+    pub fn add_task(&self, tid: CgroupPid) -> Result<()> {
+        if self.v2() {
+            let subsystems = self.subsystems();
+            if !subsystems.is_empty() {
+                let c = subsystems[0].to_controller();
+                let cgroup_type = self.get_cgroup_type()?;
+                // In cgroup v2, writing to the cgroup.threads file is only supported in thread mode.
+                if cgroup_type == *CGROUP_MODE_DOMAIN_THREADED
+                    || cgroup_type == *CGROUP_MODE_THREADED
+                {
+                    // It is used to move the threads of a process into a cgroup in thread mode.
+                    c.add_task(&tid)
+                } else {
+                    // When the cgroup type is domain or domain invalid,
+                    // cgroup.threads cannot be written.
+                    Err(Error::new(CgroupMode))
+                }
+            } else {
+                Err(Error::new(SubsystemsEmpty))
+            }
+        } else {
+            self.subsystems()
+                .iter()
+                .try_for_each(|sub| sub.to_controller().add_task(&tid))
+        }
+    }
+
+    /// Attach tasks to the control group by thread group id.
+    pub fn add_task_by_tgid(&self, tgid: CgroupPid) -> Result<()> {
+        if self.v2() {
+            let subsystems = self.subsystems();
+            if !subsystems.is_empty() {
+                let c = subsystems[0].to_controller();
+                // It is used to move a thread of the process to a cgroup,
+                // and other threads of the process will also move together.
+                c.add_task_by_tgid(&tgid)
+            } else {
+                Err(Error::new(SubsystemsEmpty))
+            }
+        } else {
+            self.subsystems()
+                .iter()
+                .try_for_each(|sub| sub.to_controller().add_task_by_tgid(&tgid))
+        }
+    }
+
+    /// set cgroup.type
+    pub fn set_cgroup_type(&self, cgroup_type: &str) -> Result<()> {
+        if self.v2() {
+            let subsystems = self.subsystems();
+            if !subsystems.is_empty() {
+                let c = subsystems[0].to_controller();
+                c.set_cgroup_type(cgroup_type)
+            } else {
+                Err(Error::new(SubsystemsEmpty))
+            }
+        } else {
+            Err(Error::new(CgroupVersion))
+        }
+    }
+
+    /// get cgroup.type
+    pub fn get_cgroup_type(&self) -> Result<String> {
+        if self.v2() {
+            let subsystems = self.subsystems();
+            if !subsystems.is_empty() {
+                let c = subsystems[0].to_controller();
+                let cgroup_type = c.get_cgroup_type()?;
+                Ok(cgroup_type)
+            } else {
+                Err(Error::new(SubsystemsEmpty))
+            }
+        } else {
+            Err(Error::new(CgroupVersion))
+        }
+    }
+
+    /// Set notify_on_release to the control group.
+    pub fn set_notify_on_release(&self, enable: bool) -> Result<()> {
+        self.subsystems()
+            .iter()
+            .try_for_each(|sub| sub.to_controller().set_notify_on_release(enable))
+    }
+
+    /// Set release_agent
+    pub fn set_release_agent(&self, path: &str) -> Result<()> {
+        self.hier
+            .root_control_group()
+            .subsystems()
+            .iter()
+            .try_for_each(|sub| sub.to_controller().set_release_agent(path))
+    }
+
+    /// Returns an Iterator that can be used to iterate over the procs that are currently in the
+    /// control group.
+    pub fn procs(&self) -> Vec<CgroupPid> {
+        // Collect the procs from all subsystems
+        let mut v = if self.v2() {
+            let subsystems = self.subsystems();
+            if !subsystems.is_empty() {
+                let c = subsystems[0].to_controller();
+                c.procs()
+            } else {
+                vec![]
+            }
+        } else {
+            self.subsystems()
+                .iter()
+                .map(|x| x.to_controller().procs())
+                .fold(vec![], |mut acc, mut x| {
+                    acc.append(&mut x);
+                    acc
+                })
+        };
+
+        v.sort();
+        v.dedup();
+        v
+    }
+
+    /// Returns an Iterator that can be used to iterate over the tasks that are currently in the
+    /// control group.
+    pub fn tasks(&self) -> Vec<CgroupPid> {
+        // Collect the tasks from all subsystems
+        let mut v = if self.v2() {
+            let subsystems = self.subsystems();
+            if !subsystems.is_empty() {
+                let c = subsystems[0].to_controller();
+                c.tasks()
+            } else {
+                vec![]
+            }
+        } else {
+            self.subsystems()
+                .iter()
+                .map(|x| x.to_controller().tasks())
+                .fold(vec![], |mut acc, mut x| {
+                    acc.append(&mut x);
+                    acc
+                })
+        };
+
+        v.sort();
+        v.dedup();
+        v
+    }
+
+    /// Checks if the cgroup exists.
+    ///
+    /// Returns true if at least one subsystem exists.
+    pub fn exists(&self) -> bool {
+        self.subsystems().iter().any(|e| e.to_controller().exists())
+    }
+}
+
+pub const UNIFIED_MOUNTPOINT: &str = "/sys/fs/cgroup";
+
+fn enable_controllers(controllers: &[String], path: &Path) {
+    let f = path.join("cgroup.subtree_control");
+    for c in controllers {
+        let body = format!("+{}", c);
+        let _rest = fs::write(f.as_path(), body.as_bytes());
+    }
+}
+
+fn supported_controllers() -> Vec<String> {
+    let p = format!("{}/{}", UNIFIED_MOUNTPOINT, "cgroup.controllers");
+    let ret = fs::read_to_string(p.as_str());
+    ret.unwrap_or_default()
+        .split(' ')
+        .map(|x| x.trim().to_string())
+        .collect::<Vec<String>>()
+}
+
+fn create_v2_cgroup(
+    root: PathBuf,
+    path: &str,
+    specified_controllers: &Option<Vec<String>>,
+) -> Result<()> {
+    // controler list ["memory", "cpu"]
+    let controllers = if let Some(s_controllers) = specified_controllers.clone() {
+        if verify_supported_controllers(s_controllers.as_ref()) {
+            s_controllers
+        } else {
+            return Err(Error::new(ErrorKind::SpecifiedControllers));
+        }
+    } else {
+        supported_controllers()
+    };
+
+    let mut fp = root;
+
+    // enable for root
+    enable_controllers(&controllers, &fp);
+
+    // path: "a/b/c"
+    let elements = path.split('/').collect::<Vec<&str>>();
+    let last_index = elements.len() - 1;
+    // Build up the directory hierarchy element by element, enabling the controllers for all
+    // parents along the way.
+    for (i, ele) in elements.iter().enumerate() {
+        // ROOT/a
+        fp.push(ele);
+        // create dir if necessary
+        if let Err(e) = std::fs::create_dir_all(fp.clone()) {
+            return Err(Error::with_cause(ErrorKind::FsError, e));
+        }
+
+        if i < last_index {
+            // enable controllers for subtree
+            enable_controllers(&controllers, &fp);
+        }
+    }
+
+    Ok(())
+}
+
+pub fn verify_supported_controllers(controllers: &[String]) -> bool {
+    let sc = supported_controllers();
+    for controller in controllers.iter() {
+        if !sc.contains(controller) {
+            return false;
+        }
+    }
+    true
+}
+
+pub fn get_cgroups_relative_paths() -> Result<HashMap<String, String>> {
+    let path = "/proc/self/cgroup".to_string();
+    get_cgroups_relative_paths_by_path(path)
+}
+
+pub fn get_cgroups_relative_paths_by_pid(pid: u32) -> Result<HashMap<String, String>> {
+    let path = format!("/proc/{}/cgroup", pid);
+    get_cgroups_relative_paths_by_path(path)
+}
+
+fn get_cgroup_destination(mut mount_root: String, pidpath: String) -> String {
+    if mount_root == "/" {
+        mount_root = String::from("");
+    }
+    pidpath.trim_start_matches(&mount_root).to_string()
+}
+
+pub fn existing_path(paths: HashMap<String, String>) -> Result<HashMap<String, String>> {
+    let mount_roots_v1 = V1::new();
+    let mut mount_roots_subsystems_map = HashMap::new();
+
+    for s in mount_roots_v1.subsystems().iter() {
+        let controller_name = s.controller_name();
+        let path_from_cgroup = paths
+            .get(&controller_name)
+            .ok_or(Error::new(Common(format!(
+                "controller {} found in mountinfo, but not found in cgroup.",
+                controller_name
+            ))))?;
+        let path_from_mountinfo = s.to_controller().base().to_string_lossy().to_string();
+
+        let des_path = get_cgroup_destination(path_from_mountinfo, path_from_cgroup.to_owned());
+        mount_roots_subsystems_map.insert(controller_name, des_path);
+    }
+    Ok(mount_roots_subsystems_map)
+}
+
+fn get_cgroups_relative_paths_by_path(path: String) -> Result<HashMap<String, String>> {
+    let mut m = HashMap::new();
+    let content =
+        fs::read_to_string(path.clone()).map_err(|e| Error::with_cause(ReadFailed(path), e))?;
+    // cgroup path may have ":" , likes
+    // "2:cpu,cpuacct:/system.slice/containerd.service/test.slice:cri-containerd:96b37a2edf84351487f42039e137427f1812f678850675fac214caf597ee5e4a"
+    for line in content.lines() {
+        if let Some((first_value_part, remaining_path)) =
+            line.split_once(':').unwrap_or_default().1.split_once(':')
+        {
+            let keys: Vec<&str> = first_value_part.split(',').collect();
+            keys.iter().for_each(|key| {
+                m.insert(key.to_string(), remaining_path.to_string());
+            });
+        }
+    }
+    Ok(m)
+}

src/fs/cgroup_builder.rs

@@ -16,10 +16,10 @@
 //! by a call to `build()`.
 //!
 //! ```rust,no_run
-//! # use cgroups_rs::*;
-//! # use cgroups_rs::devices::*;
-//! # use cgroups_rs::cgroup_builder::*;
-//! let h = cgroups_rs::hierarchies::auto();
+//! # use cgroups_rs::fs::*;
+//! # use cgroups_rs::fs::devices::*;
+//! # use cgroups_rs::fs::cgroup_builder::*;
+//! let h = cgroups_rs::fs::hierarchies::auto();
 //! let cgroup: Cgroup = CgroupBuilder::new("hello")
 //!      .memory()
 //!          .kernel_memory_limit(1024 * 1024)
@@ -57,11 +57,11 @@
 //!              .read(6, 1, 10)
 //!              .write(11, 1, 100)
 //!          .done()
-//!      .build(h);
+//!      .build(h).unwrap();
 //! ```
 
-use crate::{
-    BlkIoDeviceResource, BlkIoDeviceThrottleResource, Cgroup, DeviceResource, Hierarchy,
+use crate::fs::{
+    BlkIoDeviceResource, BlkIoDeviceThrottleResource, Cgroup, DeviceResource, Error, Hierarchy,
     HugePageResource, MaxValue, NetworkPriority, Resources,
 };
 
@@ -80,6 +80,8 @@ pub struct CgroupBuilder {
     name: String,
     /// Internal, unsupported field: use the associated builders instead.
     resources: Resources,
+    /// List of controllers specifically enabled in the control group.
+    specified_controllers: Option<Vec<String>>,
 }
 
 impl CgroupBuilder {
@@ -90,6 +92,7 @@ impl CgroupBuilder {
         CgroupBuilder {
             name: name.to_owned(),
             resources: Resources::default(),
+            specified_controllers: None,
         }
     }
 
@@ -134,10 +137,22 @@ impl CgroupBuilder {
     }
 
     /// Finalize the control group, consuming the builder and creating the control group.
-    pub fn build(self, hier: Box<dyn Hierarchy>) -> Cgroup {
-        let cg = Cgroup::new(hier, self.name);
-        let _ret = cg.apply(&self.resources);
-        cg
+    pub fn build(self, hier: Box<dyn Hierarchy>) -> Result<Cgroup, Error> {
+        if let Some(controllers) = self.specified_controllers {
+            let cg = Cgroup::new_with_specified_controllers(hier, self.name, Some(controllers))?;
+            cg.apply(&self.resources)?;
+            Ok(cg)
+        } else {
+            let cg = Cgroup::new(hier, self.name)?;
+            cg.apply(&self.resources)?;
+            Ok(cg)
+        }
+    }
+
+    /// Specifically enable some controllers in the control group.
+    pub fn set_specified_controllers(mut self, specified_controllers: Vec<String>) -> Self {
+        self.specified_controllers = Some(specified_controllers);
+        self
     }
 }
 
@@ -235,9 +250,9 @@ impl DeviceResourceBuilder {
         mut self,
         major: i64,
         minor: i64,
-        devtype: crate::devices::DeviceType,
+        devtype: crate::fs::devices::DeviceType,
         allow: bool,
-        access: Vec<crate::devices::DevicePermissions>,
+        access: Vec<crate::fs::devices::DevicePermissions>,
     ) -> DeviceResourceBuilder {
         self.cgroup.resources.devices.devices.push(DeviceResource {
             allow,

src/fs/cpu.rs

@@ -13,11 +13,11 @@ use std::fs::File;
 use std::io::{Read, Write};
 use std::path::PathBuf;
 
-use crate::error::ErrorKind::*;
-use crate::error::*;
-use crate::{parse_max_value, read_i64_from, read_u64_from};
+use crate::fs::error::ErrorKind::*;
+use crate::fs::error::*;
+use crate::fs::{parse_max_value, read_i64_from, read_u64_from};
 
-use crate::{
+use crate::fs::{
     ControllIdentifier, ControllerInternal, Controllers, CpuResources, CustomizedAttribute,
     MaxValue, Resources, Subsystem,
 };
@@ -59,7 +59,6 @@ impl ControllerInternal for CpuController {
     fn get_path(&self) -> &PathBuf {
         &self.path
     }
-
     fn get_path_mut(&mut self) -> &mut PathBuf {
         &mut self.path
     }
@@ -113,10 +112,10 @@ impl<'a> From<&'a Subsystem> for &'a CpuController {
 
 impl CpuController {
     /// Contructs a new `CpuController` with `root` serving as the root of the control group.
-    pub fn new(root: PathBuf, v2: bool) -> Self {
+    pub fn new(point: PathBuf, root: PathBuf, v2: bool) -> Self {
         Self {
-            base: root.clone(),
-            path: root,
+            base: root,
+            path: point,
             v2,
         }
     }
@@ -131,7 +130,7 @@ impl CpuController {
                     let res = file.read_to_string(&mut s);
                     match res {
                         Ok(_) => Ok(s),
-                        Err(e) => Err(Error::with_cause(ReadFailed, e)),
+                        Err(e) => Err(Error::with_cause(ReadFailed("cpu.stat".to_string()), e)),
                     }
                 })
                 .unwrap_or_default(),
@@ -145,14 +144,15 @@ impl CpuController {
     /// `shares` to `200` ensures that control group `B` receives twice as much as CPU bandwidth.
     /// (Assuming both `A` and `B` are of the same parent)
     pub fn set_shares(&self, shares: u64) -> Result<()> {
-        let mut file = "cpu.shares";
+        let mut file_name = "cpu.shares";
         if self.v2 {
-            file = "cpu.weight";
+            file_name = "cpu.weight";
         }
         // NOTE: .CpuShares is not used here. Conversion is the caller's responsibility.
-        self.open_path(file, true).and_then(|mut file| {
-            file.write_all(shares.to_string().as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+        self.open_path(file_name, true).and_then(|mut file| {
+            file.write_all(shares.to_string().as_ref()).map_err(|e| {
+                Error::with_cause(WriteFailed(file_name.to_string(), shares.to_string()), e)
+            })
         })
     }
 
@@ -174,8 +174,12 @@ impl CpuController {
         }
         self.open_path("cpu.cfs_period_us", true)
             .and_then(|mut file| {
-                file.write_all(us.to_string().as_ref())
-                    .map_err(|e| Error::with_cause(WriteFailed, e))
+                file.write_all(us.to_string().as_ref()).map_err(|e| {
+                    Error::with_cause(
+                        WriteFailed("cpu.cfs_period_us".to_string(), us.to_string()),
+                        e,
+                    )
+                })
             })
     }
 
@@ -200,8 +204,12 @@ impl CpuController {
         }
         self.open_path("cpu.cfs_quota_us", true)
             .and_then(|mut file| {
-                file.write_all(us.to_string().as_ref())
-                    .map_err(|e| Error::with_cause(WriteFailed, e))
+                file.write_all(us.to_string().as_ref()).map_err(|e| {
+                    Error::with_cause(
+                        WriteFailed("cpu.cfs_quota_us".to_string(), us.to_string()),
+                        e,
+                    )
+                })
             })
     }
 
@@ -262,23 +270,31 @@ impl CpuController {
         let line = format!("{} {}", new_quota, new_period);
         self.open_path("cpu.max", true).and_then(|mut file| {
             file.write_all(line.as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+                .map_err(|e| Error::with_cause(WriteFailed("cpu.max".to_string(), line), e))
         })
     }
 
     pub fn set_rt_runtime(&self, us: i64) -> Result<()> {
         self.open_path("cpu.rt_runtime_us", true)
             .and_then(|mut file| {
-                file.write_all(us.to_string().as_ref())
-                    .map_err(|e| Error::with_cause(WriteFailed, e))
+                file.write_all(us.to_string().as_ref()).map_err(|e| {
+                    Error::with_cause(
+                        WriteFailed("cpu.rt_runtime_us".to_string(), us.to_string()),
+                        e,
+                    )
+                })
             })
     }
 
     pub fn set_rt_period_us(&self, us: u64) -> Result<()> {
         self.open_path("cpu.rt_period_us", true)
             .and_then(|mut file| {
-                file.write_all(us.to_string().as_ref())
-                    .map_err(|e| Error::with_cause(WriteFailed, e))
+                file.write_all(us.to_string().as_ref()).map_err(|e| {
+                    Error::with_cause(
+                        WriteFailed("cpu.rt_period_us".to_string(), us.to_string()),
+                        e,
+                    )
+                })
             })
     }
 }
@@ -288,7 +304,7 @@ impl CustomizedAttribute for CpuController {}
 fn parse_cfs_quota_and_period(mut file: File) -> Result<CfsQuotaAndPeriod> {
     let mut content = String::new();
     file.read_to_string(&mut content)
-        .map_err(|e| Error::with_cause(ReadFailed, e))?;
+        .map_err(|e| Error::with_cause(ReadFailed("cpu.max".to_string()), e))?;
 
     let fields = content.trim().split(' ').collect::<Vec<&str>>();
     if fields.len() != 2 {

src/fs/cpuacct.rs

@@ -10,11 +10,11 @@
 use std::io::Write;
 use std::path::PathBuf;
 
-use crate::error::ErrorKind::*;
-use crate::error::*;
+use crate::fs::error::ErrorKind::*;
+use crate::fs::error::*;
 
-use crate::{read_string_from, read_u64_from};
-use crate::{ControllIdentifier, ControllerInternal, Controllers, Resources, Subsystem};
+use crate::fs::{read_string_from, read_u64_from};
+use crate::fs::{ControllIdentifier, ControllerInternal, Controllers, Resources, Subsystem};
 
 /// A controller that allows controlling the `cpuacct` subsystem of a Cgroup.
 ///
@@ -100,10 +100,10 @@ impl<'a> From<&'a Subsystem> for &'a CpuAcctController {
 
 impl CpuAcctController {
     /// Contructs a new `CpuAcctController` with `root` serving as the root of the control group.
-    pub fn new(root: PathBuf) -> Self {
+    pub fn new(point: PathBuf, root: PathBuf) -> Self {
         Self {
-            base: root.clone(),
-            path: root,
+            base: root,
+            path: point,
         }
     }
 
@@ -148,8 +148,9 @@ impl CpuAcctController {
     /// Reset the statistics the kernel has gathered about the control group.
     pub fn reset(&self) -> Result<()> {
         self.open_path("cpuacct.usage", true).and_then(|mut file| {
-            file.write_all(b"0")
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+            file.write_all(b"0").map_err(|e| {
+                Error::with_cause(WriteFailed("cpuacct.usage".to_string(), "0".to_string()), e)
+            })
         })
     }
 }

src/fs/cpuset.rs

@@ -13,11 +13,11 @@ use log::*;
 use std::io::Write;
 use std::path::PathBuf;
 
-use crate::error::ErrorKind::*;
-use crate::error::*;
+use crate::fs::error::ErrorKind::*;
+use crate::fs::error::*;
 
-use crate::{read_string_from, read_u64_from};
-use crate::{
+use crate::fs::{read_string_from, read_u64_from};
+use crate::fs::{
     ControllIdentifier, ControllerInternal, Controllers, CpuResources, Resources, Subsystem,
 };
 
@@ -144,7 +144,12 @@ fn find_no_empty_parent(from: &str, file: &str) -> Result<(String, Vec<PathBuf>)
         let current_value =
             match ::std::fs::read_to_string(current_path.clone().join(file).to_str().unwrap()) {
                 Ok(cpus) => String::from(cpus.trim()),
-                Err(e) => return Err(Error::with_cause(ReadFailed, e)),
+                Err(e) => {
+                    return Err(Error::with_cause(
+                        ReadFailed(current_path.display().to_string()),
+                        e,
+                    ))
+                }
             };
 
         if !current_value.is_empty() {
@@ -177,7 +182,12 @@ fn copy_from_parent(current: &str, file: &str) -> Result<()> {
         pb.push(file);
         match ::std::fs::write(pb.to_str().unwrap(), value.as_bytes()) {
             Ok(_) => (),
-            Err(e) => return Err(Error::with_cause(WriteFailed, e)),
+            Err(e) => {
+                return Err(Error::with_cause(
+                    WriteFailed(pb.display().to_string(), pb.display().to_string()),
+                    e,
+                ))
+            }
         }
     }
 
@@ -244,10 +254,10 @@ fn parse_range(s: String) -> Result<Vec<(u64, u64)>> {
 
 impl CpuSetController {
     /// Contructs a new `CpuSetController` with `root` serving as the root of the control group.
-    pub fn new(root: PathBuf, v2: bool) -> Self {
+    pub fn new(point: PathBuf, root: PathBuf, v2: bool) -> Self {
         Self {
-            base: root.clone(),
-            path: root,
+            base: root,
+            path: point,
             v2,
         }
     }
@@ -347,11 +357,19 @@ impl CpuSetController {
         self.open_path("cpuset.cpu_exclusive", true)
             .and_then(|mut file| {
                 if b {
-                    file.write_all(b"1")
-                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                    file.write_all(b"1").map_err(|e| {
+                        Error::with_cause(
+                            WriteFailed("cpuset.cpu_exclusive".to_string(), "1".to_string()),
+                            e,
+                        )
+                    })
                 } else {
-                    file.write_all(b"0")
-                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                    file.write_all(b"0").map_err(|e| {
+                        Error::with_cause(
+                            WriteFailed("cpuset.cpu_exclusive".to_string(), "0".to_string()),
+                            e,
+                        )
+                    })
                 }
             })
     }
@@ -362,11 +380,19 @@ impl CpuSetController {
         self.open_path("cpuset.mem_exclusive", true)
             .and_then(|mut file| {
                 if b {
-                    file.write_all(b"1")
-                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                    file.write_all(b"1").map_err(|e| {
+                        Error::with_cause(
+                            WriteFailed("cpuset.mem_exclusive".to_string(), "1".to_string()),
+                            e,
+                        )
+                    })
                 } else {
-                    file.write_all(b"0")
-                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                    file.write_all(b"0").map_err(|e| {
+                        Error::with_cause(
+                            WriteFailed("cpuset.mem_exclusive".to_string(), "0".to_string()),
+                            e,
+                        )
+                    })
                 }
             })
     }
@@ -377,8 +403,9 @@ impl CpuSetController {
     /// be represented via dashes.
     pub fn set_cpus(&self, cpus: &str) -> Result<()> {
         self.open_path("cpuset.cpus", true).and_then(|mut file| {
-            file.write_all(cpus.as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+            file.write_all(cpus.as_ref()).map_err(|e| {
+                Error::with_cause(WriteFailed("cpuset.cpus".to_string(), cpus.to_string()), e)
+            })
         })
     }
 
@@ -387,8 +414,9 @@ impl CpuSetController {
     /// Syntax is the same as with `set_cpus()`.
     pub fn set_mems(&self, mems: &str) -> Result<()> {
         self.open_path("cpuset.mems", true).and_then(|mut file| {
-            file.write_all(mems.as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+            file.write_all(mems.as_ref()).map_err(|e| {
+                Error::with_cause(WriteFailed("cpuset.mems".to_string(), mems.to_string()), e)
+            })
         })
     }
 
@@ -401,11 +429,19 @@ impl CpuSetController {
         self.open_path("cpuset.mem_hardwall", true)
             .and_then(|mut file| {
                 if b {
-                    file.write_all(b"1")
-                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                    file.write_all(b"1").map_err(|e| {
+                        Error::with_cause(
+                            WriteFailed("cpuset.mem_hardwall".to_string(), "1".to_string()),
+                            e,
+                        )
+                    })
                 } else {
-                    file.write_all(b"0")
-                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                    file.write_all(b"0").map_err(|e| {
+                        Error::with_cause(
+                            WriteFailed("cpuset.mem_hardwall".to_string(), "0".to_string()),
+                            e,
+                        )
+                    })
                 }
             })
     }
@@ -416,11 +452,19 @@ impl CpuSetController {
         self.open_path("cpuset.sched_load_balance", true)
             .and_then(|mut file| {
                 if b {
-                    file.write_all(b"1")
-                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                    file.write_all(b"1").map_err(|e| {
+                        Error::with_cause(
+                            WriteFailed("cpuset.sched_load_balance".to_string(), "1".to_string()),
+                            e,
+                        )
+                    })
                 } else {
-                    file.write_all(b"0")
-                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                    file.write_all(b"0").map_err(|e| {
+                        Error::with_cause(
+                            WriteFailed("cpuset.sched_load_balance".to_string(), "0".to_string()),
+                            e,
+                        )
+                    })
                 }
             })
     }
@@ -431,8 +475,12 @@ impl CpuSetController {
     pub fn set_rebalance_relax_domain_level(&self, i: i64) -> Result<()> {
         self.open_path("cpuset.sched_relax_domain_level", true)
             .and_then(|mut file| {
-                file.write_all(i.to_string().as_ref())
-                    .map_err(|e| Error::with_cause(WriteFailed, e))
+                file.write_all(i.to_string().as_ref()).map_err(|e| {
+                    Error::with_cause(
+                        WriteFailed("cpuset.sched_relax_domain_level".to_string(), i.to_string()),
+                        e,
+                    )
+                })
             })
     }
 
@@ -442,11 +490,19 @@ impl CpuSetController {
         self.open_path("cpuset.memory_migrate", true)
             .and_then(|mut file| {
                 if b {
-                    file.write_all(b"1")
-                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                    file.write_all(b"1").map_err(|e| {
+                        Error::with_cause(
+                            WriteFailed("cpuset.memory_migrate".to_string(), "1".to_string()),
+                            e,
+                        )
+                    })
                 } else {
-                    file.write_all(b"0")
-                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                    file.write_all(b"0").map_err(|e| {
+                        Error::with_cause(
+                            WriteFailed("cpuset.memory_migrate".to_string(), "0".to_string()),
+                            e,
+                        )
+                    })
                 }
             })
     }
@@ -457,11 +513,19 @@ impl CpuSetController {
         self.open_path("cpuset.memory_spread_page", true)
             .and_then(|mut file| {
                 if b {
-                    file.write_all(b"1")
-                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                    file.write_all(b"1").map_err(|e| {
+                        Error::with_cause(
+                            WriteFailed("cpuset.memory_spread_page".to_string(), "1".to_string()),
+                            e,
+                        )
+                    })
                 } else {
-                    file.write_all(b"0")
-                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                    file.write_all(b"0").map_err(|e| {
+                        Error::with_cause(
+                            WriteFailed("cpuset.memory_spread_page".to_string(), "0".to_string()),
+                            e,
+                        )
+                    })
                 }
             })
     }
@@ -472,11 +536,19 @@ impl CpuSetController {
         self.open_path("cpuset.memory_spread_slab", true)
             .and_then(|mut file| {
                 if b {
-                    file.write_all(b"1")
-                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                    file.write_all(b"1").map_err(|e| {
+                        Error::with_cause(
+                            WriteFailed("cpuset.memory_spread_slab".to_string(), "1".to_string()),
+                            e,
+                        )
+                    })
                 } else {
-                    file.write_all(b"0")
-                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                    file.write_all(b"0").map_err(|e| {
+                        Error::with_cause(
+                            WriteFailed("cpuset.memory_spread_slab".to_string(), "0".to_string()),
+                            e,
+                        )
+                    })
                 }
             })
     }
@@ -493,11 +565,25 @@ impl CpuSetController {
         self.open_path("cpuset.memory_pressure_enabled", true)
             .and_then(|mut file| {
                 if b {
-                    file.write_all(b"1")
-                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                    file.write_all(b"1").map_err(|e| {
+                        Error::with_cause(
+                            WriteFailed(
+                                "cpuset.memory_pressure_enabled".to_string(),
+                                "1".to_string(),
+                            ),
+                            e,
+                        )
+                    })
                 } else {
-                    file.write_all(b"0")
-                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                    file.write_all(b"0").map_err(|e| {
+                        Error::with_cause(
+                            WriteFailed(
+                                "cpuset.memory_pressure_enabled".to_string(),
+                                "0".to_string(),
+                            ),
+                            e,
+                        )
+                    })
                 }
             })
     }
@@ -505,7 +591,7 @@ impl CpuSetController {
 
 #[cfg(test)]
 mod tests {
-    use crate::cpuset;
+    use crate::fs::cpuset;
     #[test]
     fn test_parse_range() {
         let test_cases = vec![
@@ -516,7 +602,7 @@ mod tests {
             "1,2,3,4".to_string(),
             "1-5,6-7,8-9".to_string(),
         ];
-        let expecteds = vec![
+        let expecteds = [
             vec![(1, 1), (2, 2), (4, 6), (9, 9)],
             vec![],
             vec![(1, 1)],

src/fs/devices.rs

@@ -12,10 +12,10 @@ use std::path::PathBuf;
 
 use log::*;
 
-use crate::error::ErrorKind::*;
-use crate::error::*;
+use crate::fs::error::ErrorKind::*;
+use crate::fs::error::*;
 
-use crate::{
+use crate::fs::{
     ControllIdentifier, ControllerInternal, Controllers, DeviceResource, DeviceResources,
     Resources, Subsystem,
 };
@@ -46,6 +46,7 @@ pub enum DeviceType {
     Block,
 }
 
+#[allow(clippy::derivable_impls)]
 impl Default for DeviceType {
     fn default() -> Self {
         DeviceType::All
@@ -170,9 +171,9 @@ impl ControllerInternal for DevicesController {
 
         for i in &res.devices {
             if i.allow {
-                let _ = self.allow_device(i.devtype, i.major, i.minor, &i.access);
+                self.allow_device(i.devtype, i.major, i.minor, &i.access)?;
             } else {
-                let _ = self.deny_device(i.devtype, i.major, i.minor, &i.access);
+                self.deny_device(i.devtype, i.major, i.minor, &i.access)?;
             }
         }
 
@@ -203,10 +204,10 @@ impl<'a> From<&'a Subsystem> for &'a DevicesController {
 
 impl DevicesController {
     /// Constructs a new `DevicesController` with `root` serving as the root of the control group.
-    pub fn new(root: PathBuf) -> Self {
+    pub fn new(point: PathBuf, root: PathBuf) -> Self {
         Self {
-            base: root.clone(),
-            path: root,
+            base: root,
+            path: point,
         }
     }
 
@@ -237,8 +238,15 @@ impl DevicesController {
         };
         let final_str = format!("{} {}:{} {}", devtype.to_char(), major, minor, perms);
         self.open_path("devices.allow", true).and_then(|mut file| {
-            file.write_all(final_str.as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+            file.write_all(final_str.as_ref()).map_err(|e| {
+                Error::with_cause(
+                    WriteFailed(
+                        self.get_path().join("devices.allow").display().to_string(),
+                        final_str,
+                    ),
+                    e,
+                )
+            })
         })
     }
 
@@ -269,8 +277,15 @@ impl DevicesController {
         };
         let final_str = format!("{} {}:{} {}", devtype.to_char(), major, minor, perms);
         self.open_path("devices.deny", true).and_then(|mut file| {
-            file.write_all(final_str.as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+            file.write_all(final_str.as_ref()).map_err(|e| {
+                Error::with_cause(
+                    WriteFailed(
+                        self.get_path().join("devices.deny").display().to_string(),
+                        final_str,
+                    ),
+                    e,
+                )
+            })
         })
     }
 
@@ -280,43 +295,48 @@ impl DevicesController {
             let mut s = String::new();
             let res = file.read_to_string(&mut s);
             match res {
-                Ok(_) => {
-                    s.lines().fold(Ok(Vec::new()), |acc, line| {
-                        let ls = line.to_string().split(|c| c == ' ' || c == ':').map(|x| x.to_string()).collect::<Vec<String>>();
-                        if acc.is_err() || ls.len() != 4 {
-                            error!("allowed_devices: acc: {:?}, ls: {:?}", acc, ls);
-                            Err(Error::new(ParseError))
-                        } else {
-                            let devtype = DeviceType::from_char(ls[0].chars().next());
-                            let mut major = ls[1].parse::<i64>();
-                            let mut minor = ls[2].parse::<i64>();
-                            if major.is_err() && ls[1] == "*" {
-                                major = Ok(-1);
-                            }
-                            if minor.is_err() && ls[2] == "*" {
-                                minor = Ok(-1);
-                            }
-                            if devtype.is_none() || major.is_err() || minor.is_err() || !DevicePermissions::is_valid(&ls[3]) {
-                                error!("allowed_devices: acc: {:?}, ls: {:?}, devtype: {:?}, major {:?} minor {:?} ls3 {:?}",
-                                         acc, ls, devtype, major, minor, &ls[3]);
-                                Err(Error::new(ParseError))
-                            } else {
-                                let access = DevicePermissions::from_str(&ls[3])?;
-                                let mut acc = acc.unwrap();
-                                acc.push(DeviceResource {
-                                    allow: true,
-                                    devtype: devtype.unwrap(),
-                                    major: major.unwrap(),
-                                    minor: minor.unwrap(),
-                                    access,
-                                });
-                                Ok(acc)
-                            }
-                        }
-                    })
-                },
-                Err(e) => Err(Error::with_cause(ReadFailed, e)),
+                Ok(_) => s
+                    .lines()
+                    .map(|line| parse_device_line(line, true))
+                    .collect(),
+                Err(e) => Err(Error::with_cause(ReadFailed("devices.list".to_string()), e)),
             }
         })
     }
 }
+
+fn parse_device_number(s: &str) -> Result<i64> {
+    if s == "*" {
+        Ok(-1)
+    } else {
+        s.parse::<i64>().map_err(|_| Error::new(ParseError))
+    }
+}
+
+fn parse_device_line(line: &str, allow: bool) -> Result<DeviceResource> {
+    let parts: Vec<&str> = line.split([' ', ':']).collect();
+    if parts.len() != 4 {
+        error!("allowed_devices: invalid line format: {:?}", line);
+        return Err(Error::new(ParseError));
+    }
+
+    let devtype = DeviceType::from_char(parts[0].chars().next()).ok_or_else(|| {
+        error!("allowed_devices: invalid device type: {:?}", parts[0]);
+        Error::new(ParseError)
+    })?;
+    let major = parse_device_number(parts[1]).inspect_err(|_| {
+        error!("allowed_devices: invalid major number: {:?}", parts[1]);
+    })?;
+    let minor = parse_device_number(parts[2]).inspect_err(|_| {
+        error!("allowed_devices: invalid minor number: {:?}", parts[2]);
+    })?;
+    let access = DevicePermissions::from_str(parts[3])?;
+
+    Ok(DeviceResource {
+        allow,
+        devtype,
+        major,
+        minor,
+        access,
+    })
+}

src/fs/error.rs

@@ -8,40 +8,67 @@ use std::error::Error as StdError;
 use std::fmt;
 
 /// The different types of errors that can occur while manipulating control groups.
-#[derive(Debug, Eq, PartialEq)]
+#[derive(thiserror::Error, Debug, Eq, PartialEq)]
 pub enum ErrorKind {
+    #[error("fs error")]
     FsError,
+
+    #[error("common error: {0}")]
     Common(String),
 
     /// An error occured while writing to a control group file.
-    WriteFailed,
+    #[error("unable to write to a control group file {0}, value {1}")]
+    WriteFailed(String, String),
 
     /// An error occured while trying to read from a control group file.
-    ReadFailed,
+    #[error("unable to read a control group file {0}")]
+    ReadFailed(String),
 
     /// An error occured while trying to remove a control group.
+    #[error("unable to remove a control group")]
     RemoveFailed,
 
     /// An error occured while trying to parse a value from a control group file.
     ///
     /// In the future, there will be some information attached to this field.
+    #[error("unable to parse control group file")]
     ParseError,
 
     /// You tried to do something invalid.
     ///
     /// This could be because you tried to set a value in a control group that is not a root
     /// control group. Or, when using unified hierarchy, you tried to add a task in a leaf node.
+    #[error("the requested operation is invalid")]
     InvalidOperation,
 
     /// The path of the control group was invalid.
     ///
     /// This could be caused by trying to escape the control group filesystem via a string of "..".
     /// This crate checks against this and operations will fail with this error.
+    #[error("the given path is invalid")]
     InvalidPath,
 
+    #[error("invalid bytes size")]
     InvalidBytesSize,
 
+    /// The specified controller is not in the list of supported controllers.
+    #[error("specified controller is not in the list of supported controllers")]
+    SpecifiedControllers,
+
+    /// Using method in wrong cgroup version.
+    #[error("using method in wrong cgroup version")]
+    CgroupVersion,
+
+    /// Using method in wrong cgroup mode.
+    #[error("using method in wrong cgroup mode.")]
+    CgroupMode,
+
+    /// Subsystems is empty.
+    #[error("subsystems is empty")]
+    SubsystemsEmpty,
+
     /// An unknown error has occured.
+    #[error("an unknown error")]
     Other,
 }
 
@@ -53,29 +80,16 @@ pub struct Error {
 
 impl fmt::Display for Error {
     fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        let msg = match &self.kind {
-            ErrorKind::FsError => "fs error".to_string(),
-            ErrorKind::Common(s) => s.clone(),
-            ErrorKind::WriteFailed => "unable to write to a control group file".to_string(),
-            ErrorKind::ReadFailed => "unable to read a control group file".to_string(),
-            ErrorKind::RemoveFailed => "unable to remove a control group".to_string(),
-            ErrorKind::ParseError => "unable to parse control group file".to_string(),
-            ErrorKind::InvalidOperation => "the requested operation is invalid".to_string(),
-            ErrorKind::InvalidPath => "the given path is invalid".to_string(),
-            ErrorKind::InvalidBytesSize => "invalid bytes size".to_string(),
-            ErrorKind::Other => "an unknown error".to_string(),
-        };
-
         if let Some(cause) = &self.cause {
-            write!(f, "{} caused by: {:?}", msg, cause)
+            write!(f, "{} caused by: {:?}", &self.kind, cause)
         } else {
-            write!(f, "{}", msg)
+            write!(f, "{}", &self.kind)
         }
     }
 }
 
 impl StdError for Error {
-    fn cause(&self) -> Option<&dyn StdError> {
+    fn source(&self) -> Option<&(dyn StdError + 'static)> {
         #[allow(clippy::manual_map)]
         match self.cause {
             Some(ref x) => Some(&**x),

src/fs/events.rs

@@ -12,8 +12,8 @@ use std::path::Path;
 use std::sync::mpsc::{self, Receiver};
 use std::thread;
 
-use crate::error::ErrorKind::*;
-use crate::error::*;
+use crate::fs::error::ErrorKind::*;
+use crate::fs::error::*;
 
 // notify_on_oom returns channel on which you can expect event about OOM,
 // if process died without OOM this channel will be closed.
@@ -46,10 +46,11 @@ fn register_memory_event(
     arg: &str,
 ) -> Result<Receiver<String>> {
     let path = cg_dir.join(event_name);
-    let event_file = File::open(path).map_err(|e| Error::with_cause(ReadFailed, e))?;
+    let event_file = File::open(path.clone())
+        .map_err(|e| Error::with_cause(ReadFailed(path.display().to_string()), e))?;
 
-    let eventfd =
-        eventfd(0, EfdFlags::EFD_CLOEXEC).map_err(|e| Error::with_cause(ReadFailed, e))?;
+    let eventfd = eventfd(0, EfdFlags::EFD_CLOEXEC)
+        .map_err(|e| Error::with_cause(ReadFailed("eventfd".to_string()), e))?;
 
     let event_control_path = cg_dir.join("cgroup.event_control");
     let data = if arg.is_empty() {
@@ -59,7 +60,12 @@ fn register_memory_event(
     };
 
     // write to file and set mode to 0700(FIXME)
-    fs::write(&event_control_path, data).map_err(|e| Error::with_cause(WriteFailed, e))?;
+    fs::write(&event_control_path, data.clone()).map_err(|e| {
+        Error::with_cause(
+            WriteFailed(event_control_path.display().to_string(), data),
+            e,
+        )
+    })?;
 
     let mut eventfd_file = unsafe { File::from_raw_fd(eventfd) };
 

src/fs/freezer.rs

@@ -11,10 +11,10 @@
 use std::io::{Read, Write};
 use std::path::PathBuf;
 
-use crate::error::ErrorKind::*;
-use crate::error::*;
-
-use crate::{ControllIdentifier, ControllerInternal, Controllers, Resources, Subsystem};
+use crate::fs::error::ErrorKind::*;
+use crate::fs::error::*;
+use crate::fs::{ControllIdentifier, ControllerInternal, Controllers, Resources, Subsystem};
+use crate::FreezerState;
 
 /// A controller that allows controlling the `freezer` subsystem of a Cgroup.
 ///
@@ -31,17 +31,6 @@ pub struct FreezerController {
     v2: bool,
 }
 
-/// The current state of the control group
-#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
-pub enum FreezerState {
-    /// The processes in the control group are _not_ frozen.
-    Thawed,
-    /// The processes in the control group are in the processes of being frozen.
-    Freezing,
-    /// The processes in the control group are frozen.
-    Frozen,
-}
-
 impl ControllerInternal for FreezerController {
     fn control_type(&self) -> Controllers {
         Controllers::Freezer
@@ -84,50 +73,49 @@ impl<'a> From<&'a Subsystem> for &'a FreezerController {
 
 impl FreezerController {
     /// Contructs a new `FreezerController` with `root` serving as the root of the control group.
-    pub fn new(root: PathBuf, v2: bool) -> Self {
+    pub fn new(point: PathBuf, root: PathBuf, v2: bool) -> Self {
         Self {
-            base: root.clone(),
-            path: root,
+            base: root,
+            path: point,
             v2,
         }
     }
-
     /// Freezes the processes in the control group.
     pub fn freeze(&self) -> Result<()> {
-        let mut file = "freezer.state";
+        let mut file_name = "freezer.state";
         let mut content = "FROZEN".to_string();
         if self.v2 {
-            file = "cgroup.freeze";
+            file_name = "cgroup.freeze";
             content = "1".to_string();
         }
 
-        self.open_path(file, true).and_then(|mut file| {
+        self.open_path(file_name, true).and_then(|mut file| {
             file.write_all(content.as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+                .map_err(|e| Error::with_cause(WriteFailed(file_name.to_string(), content), e))
         })
     }
 
     /// Thaws, that is, unfreezes the processes in the control group.
     pub fn thaw(&self) -> Result<()> {
-        let mut file = "freezer.state";
+        let mut file_name = "freezer.state";
         let mut content = "THAWED".to_string();
         if self.v2 {
-            file = "cgroup.freeze";
+            file_name = "cgroup.freeze";
             content = "0".to_string();
         }
-        self.open_path(file, true).and_then(|mut file| {
+        self.open_path(file_name, true).and_then(|mut file| {
             file.write_all(content.as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+                .map_err(|e| Error::with_cause(WriteFailed(file_name.to_string(), content), e))
         })
     }
 
     /// Retrieve the state of processes in the control group.
     pub fn state(&self) -> Result<FreezerState> {
-        let mut file = "freezer.state";
+        let mut file_name = "freezer.state";
         if self.v2 {
-            file = "cgroup.freeze";
+            file_name = "cgroup.freeze";
         }
-        self.open_path(file, false).and_then(|mut file| {
+        self.open_path(file_name, false).and_then(|mut file| {
             let mut s = String::new();
             let res = file.read_to_string(&mut s);
             match res {
@@ -139,7 +127,7 @@ impl FreezerController {
                     "FREEZING" => Ok(FreezerState::Freezing),
                     _ => Err(Error::new(ParseError)),
                 },
-                Err(e) => Err(Error::with_cause(ReadFailed, e)),
+                Err(e) => Err(Error::with_cause(ReadFailed(file_name.to_string()), e)),
             }
         })
     }

src/fs/hierarchies.rs

@@ -5,38 +5,37 @@
 //
 
 //! This module represents the various control group hierarchies the Linux kernel supports.
-//!
-//! Currently, we only support the cgroupv1 hierarchy, but in the future we will add support for
-//! the Unified Hierarchy.
 
 use std::fs;
 use std::fs::File;
 use std::io::{BufRead, BufReader};
-use std::path::PathBuf;
+use std::path::{Path, PathBuf};
 
-use crate::blkio::BlkIoController;
-use crate::cpu::CpuController;
-use crate::cpuacct::CpuAcctController;
-use crate::cpuset::CpuSetController;
-use crate::devices::DevicesController;
-use crate::freezer::FreezerController;
-use crate::hugetlb::HugeTlbController;
-use crate::memory::MemController;
-use crate::net_cls::NetClsController;
-use crate::net_prio::NetPrioController;
-use crate::perf_event::PerfEventController;
-use crate::pid::PidController;
-use crate::rdma::RdmaController;
-use crate::systemd::SystemdController;
-use crate::{Controllers, Hierarchy, Subsystem};
+use crate::fs::blkio::BlkIoController;
+use crate::fs::cpu::CpuController;
+use crate::fs::cpuacct::CpuAcctController;
+use crate::fs::cpuset::CpuSetController;
+use crate::fs::devices::DevicesController;
+use crate::fs::freezer::FreezerController;
+use crate::fs::hugetlb::HugeTlbController;
+use crate::fs::memory::MemController;
+use crate::fs::net_cls::NetClsController;
+use crate::fs::net_prio::NetPrioController;
+use crate::fs::perf_event::PerfEventController;
+use crate::fs::pid::PidController;
+use crate::fs::rdma::RdmaController;
+use crate::fs::systemd::SystemdController;
+use crate::fs::{Controllers, Hierarchy, Subsystem};
 
-use crate::cgroup::Cgroup;
+use crate::fs::cgroup::Cgroup;
 
 /// Process mounts information.
 ///
 /// See `proc(5)` for format details.
 #[derive(Debug, PartialEq, Eq, Hash, Clone)]
 pub struct Mountinfo {
+    /// Mount root directory of the file system.
+    pub mount_root: PathBuf,
     /// Mount pathname relative to the process's root.
     pub mount_point: PathBuf,
     /// Filesystem type (main type with optional sub-type).
@@ -57,6 +56,7 @@ pub(crate) fn parse_mountinfo_for_line(line: &str) -> Option<Mountinfo> {
         return None;
     }
     let mount_point = PathBuf::from(s0_values[4]);
+    let mount_root = PathBuf::from(s0_values[3]);
     let fs_type_values: Vec<_> = s1_values[0].trim().split('.').collect();
     let fs_type = match fs_type_values.len() {
         1 => (fs_type_values[0].to_string(), None),
@@ -69,6 +69,7 @@ pub(crate) fn parse_mountinfo_for_line(line: &str) -> Option<Mountinfo> {
 
     let super_opts: Vec<String> = s1_values[2].trim().split(',').map(String::from).collect();
     Some(Mountinfo {
+        mount_root,
         mount_point,
         fs_type,
         super_opts,
@@ -123,47 +124,53 @@ impl Hierarchy for V1 {
         // The cgroup writeback feature requires cooperation between memcgs and blkcgs
         // To avoid exceptions, we should add_task for blkcg before memcg(push BlkIo before Mem)
         // For more Information: https://www.alibabacloud.com/help/doc-detail/155509.htm
-        if let Some(root) = self.get_mount_point(Controllers::BlkIo) {
-            subs.push(Subsystem::BlkIo(BlkIoController::new(root, false)));
+        if let Some((point, root)) = self.get_mount_point(Controllers::BlkIo) {
+            subs.push(Subsystem::BlkIo(BlkIoController::new(point, root, false)));
         }
-        if let Some(root) = self.get_mount_point(Controllers::Mem) {
-            subs.push(Subsystem::Mem(MemController::new(root, false)));
+        if let Some((point, root)) = self.get_mount_point(Controllers::Mem) {
+            subs.push(Subsystem::Mem(MemController::new(point, root, false)));
         }
-        if let Some(root) = self.get_mount_point(Controllers::Pids) {
-            subs.push(Subsystem::Pid(PidController::new(root, false)));
+        if let Some((point, root)) = self.get_mount_point(Controllers::Pids) {
+            subs.push(Subsystem::Pid(PidController::new(point, root, false)));
         }
-        if let Some(root) = self.get_mount_point(Controllers::CpuSet) {
-            subs.push(Subsystem::CpuSet(CpuSetController::new(root, false)));
+        if let Some((point, root)) = self.get_mount_point(Controllers::CpuSet) {
+            subs.push(Subsystem::CpuSet(CpuSetController::new(point, root, false)));
         }
-        if let Some(root) = self.get_mount_point(Controllers::CpuAcct) {
-            subs.push(Subsystem::CpuAcct(CpuAcctController::new(root)));
+        if let Some((point, root)) = self.get_mount_point(Controllers::CpuAcct) {
+            subs.push(Subsystem::CpuAcct(CpuAcctController::new(point, root)));
         }
-        if let Some(root) = self.get_mount_point(Controllers::Cpu) {
-            subs.push(Subsystem::Cpu(CpuController::new(root, false)));
+        if let Some((point, root)) = self.get_mount_point(Controllers::Cpu) {
+            subs.push(Subsystem::Cpu(CpuController::new(point, root, false)));
         }
-        if let Some(root) = self.get_mount_point(Controllers::Devices) {
-            subs.push(Subsystem::Devices(DevicesController::new(root)));
+        if let Some((point, root)) = self.get_mount_point(Controllers::Devices) {
+            subs.push(Subsystem::Devices(DevicesController::new(point, root)));
         }
-        if let Some(root) = self.get_mount_point(Controllers::Freezer) {
-            subs.push(Subsystem::Freezer(FreezerController::new(root, false)));
+        if let Some((point, root)) = self.get_mount_point(Controllers::Freezer) {
+            subs.push(Subsystem::Freezer(FreezerController::new(
+                point, root, false,
+            )));
         }
-        if let Some(root) = self.get_mount_point(Controllers::NetCls) {
-            subs.push(Subsystem::NetCls(NetClsController::new(root)));
+        if let Some((point, root)) = self.get_mount_point(Controllers::NetCls) {
+            subs.push(Subsystem::NetCls(NetClsController::new(point, root)));
         }
-        if let Some(root) = self.get_mount_point(Controllers::PerfEvent) {
-            subs.push(Subsystem::PerfEvent(PerfEventController::new(root)));
+        if let Some((point, root)) = self.get_mount_point(Controllers::PerfEvent) {
+            subs.push(Subsystem::PerfEvent(PerfEventController::new(point, root)));
         }
-        if let Some(root) = self.get_mount_point(Controllers::NetPrio) {
-            subs.push(Subsystem::NetPrio(NetPrioController::new(root)));
+        if let Some((point, root)) = self.get_mount_point(Controllers::NetPrio) {
+            subs.push(Subsystem::NetPrio(NetPrioController::new(point, root)));
         }
-        if let Some(root) = self.get_mount_point(Controllers::HugeTlb) {
-            subs.push(Subsystem::HugeTlb(HugeTlbController::new(root, false)));
+        if let Some((point, root)) = self.get_mount_point(Controllers::HugeTlb) {
+            subs.push(Subsystem::HugeTlb(HugeTlbController::new(
+                point, root, false,
+            )));
         }
-        if let Some(root) = self.get_mount_point(Controllers::Rdma) {
-            subs.push(Subsystem::Rdma(RdmaController::new(root)));
+        if let Some((point, root)) = self.get_mount_point(Controllers::Rdma) {
+            subs.push(Subsystem::Rdma(RdmaController::new(point, root)));
         }
-        if let Some(root) = self.get_mount_point(Controllers::Systemd) {
-            subs.push(Subsystem::Systemd(SystemdController::new(root, false)));
+        if let Some((point, root)) = self.get_mount_point(Controllers::Systemd) {
+            subs.push(Subsystem::Systemd(SystemdController::new(
+                point, root, false,
+            )));
         }
 
         subs
@@ -173,6 +180,12 @@ impl Hierarchy for V1 {
         Cgroup::load(auto(), "")
     }
 
+    fn parent_control_group(&self, path: &str) -> Cgroup {
+        let path = Path::new(path);
+        let parent_path = path.parent().unwrap().to_string_lossy().to_string();
+        Cgroup::load(auto(), parent_path)
+    }
+
     fn root(&self) -> PathBuf {
         self.mountinfo
             .iter()
@@ -212,29 +225,51 @@ impl Hierarchy for V2 {
         for s in controller_list {
             match s {
                 "cpu" => {
-                    subs.push(Subsystem::Cpu(CpuController::new(self.root(), true)));
+                    subs.push(Subsystem::Cpu(CpuController::new(
+                        self.root(),
+                        PathBuf::from(""),
+                        true,
+                    )));
                 }
                 "io" => {
-                    subs.push(Subsystem::BlkIo(BlkIoController::new(self.root(), true)));
+                    subs.push(Subsystem::BlkIo(BlkIoController::new(
+                        self.root(),
+                        PathBuf::from(""),
+                        true,
+                    )));
                 }
                 "cpuset" => {
-                    subs.push(Subsystem::CpuSet(CpuSetController::new(self.root(), true)));
+                    subs.push(Subsystem::CpuSet(CpuSetController::new(
+                        self.root(),
+                        PathBuf::from(""),
+                        true,
+                    )));
                 }
                 "memory" => {
-                    subs.push(Subsystem::Mem(MemController::new(self.root(), true)));
+                    subs.push(Subsystem::Mem(MemController::new(
+                        self.root(),
+                        PathBuf::from(""),
+                        true,
+                    )));
                 }
                 "pids" => {
-                    subs.push(Subsystem::Pid(PidController::new(self.root(), true)));
+                    subs.push(Subsystem::Pid(PidController::new(
+                        self.root(),
+                        PathBuf::from(""),
+                        true,
+                    )));
                 }
                 "freezer" => {
                     subs.push(Subsystem::Freezer(FreezerController::new(
                         self.root(),
+                        PathBuf::from(""),
                         true,
                     )));
                 }
                 "hugetlb" => {
                     subs.push(Subsystem::HugeTlb(HugeTlbController::new(
                         self.root(),
+                        PathBuf::from(""),
                         true,
                     )));
                 }
@@ -249,6 +284,12 @@ impl Hierarchy for V2 {
         Cgroup::load(auto(), "")
     }
 
+    fn parent_control_group(&self, path: &str) -> Cgroup {
+        let path = Path::new(path);
+        let parent_path = path.parent().unwrap().to_string_lossy().to_string();
+        Cgroup::load(auto(), parent_path)
+    }
+
     fn root(&self) -> PathBuf {
         PathBuf::from(self.root.clone())
     }
@@ -263,10 +304,10 @@ impl V1 {
         }
     }
 
-    pub fn get_mount_point(&self, controller: Controllers) -> Option<PathBuf> {
+    pub fn get_mount_point(&self, controller: Controllers) -> Option<(PathBuf, PathBuf)> {
         self.mountinfo.iter().find_map(|m| {
             if m.fs_type.0 == "cgroup" && m.super_opts.contains(&controller.to_string()) {
-                return Some(m.mount_point.clone());
+                return Some((m.mount_point.to_owned(), m.mount_root.to_owned()));
             }
             None
         })
@@ -297,43 +338,16 @@ impl Default for V2 {
 
 pub const UNIFIED_MOUNTPOINT: &str = "/sys/fs/cgroup";
 
-#[cfg(any(
-    all(target_os = "linux", not(target_env = "musl")),
-    target_os = "android"
-))]
 pub fn is_cgroup2_unified_mode() -> bool {
     use nix::sys::statfs;
 
     let path = std::path::Path::new(UNIFIED_MOUNTPOINT);
-    let fs_stat = statfs::statfs(path);
-    if fs_stat.is_err() {
-        return false;
-    }
+    let fs_stat = match statfs::statfs(path) {
+        Ok(fs_stat) => fs_stat,
+        Err(_) => return false,
+    };
 
-    // FIXME notwork, nix will not compile CGROUP2_SUPER_MAGIC because not(target_env = "musl")
-    fs_stat.unwrap().filesystem_type() == statfs::CGROUP2_SUPER_MAGIC
-}
-
-pub const INIT_CGROUP_PATHS: &str = "/proc/1/cgroup";
-
-#[cfg(all(target_os = "linux", target_env = "musl"))]
-pub fn is_cgroup2_unified_mode() -> bool {
-    let lines = fs::read_to_string(INIT_CGROUP_PATHS);
-    if lines.is_err() {
-        return false;
-    }
-
-    for line in lines.unwrap().lines() {
-        let fields: Vec<&str> = line.split(':').collect();
-        if fields.len() != 3 {
-            continue;
-        }
-        if fields[0] != "0" {
-            return false;
-        }
-    }
-
-    true
+    fs_stat.filesystem_type() == statfs::CGROUP2_SUPER_MAGIC
 }
 
 pub fn auto() -> Box<dyn Hierarchy> {
@@ -352,19 +366,19 @@ mod tests {
     fn test_parse_mount() {
         let mountinfo = vec![
             ("29 26 0:26 / /sys/fs/cgroup/cpuset,cpu,cpuacct rw,nosuid,nodev,noexec,relatime shared:10 - cgroup cgroup rw,cpuset,cpu,cpuacct",
-             Mountinfo{mount_point: PathBuf::from("/sys/fs/cgroup/cpuset,cpu,cpuacct"), fs_type: ("cgroup".to_string(), None), super_opts: vec![
+             Mountinfo{mount_root: PathBuf::from("/"), mount_point: PathBuf::from("/sys/fs/cgroup/cpuset,cpu,cpuacct"), fs_type: ("cgroup".to_string(), None), super_opts: vec![
                 "rw".to_string(),
                 "cpuset".to_string(),
                 "cpu".to_string(),
                 "cpuacct".to_string(),
              ]}),
             ("121 1731 0:42 / /shm rw,nosuid,nodev,noexec,relatime shared:68 master:66 - tmpfs shm rw,size=65536k",
-             Mountinfo{mount_point: PathBuf::from("/shm"), fs_type: ("tmpfs".to_string(), None), super_opts: vec![
+             Mountinfo{mount_root: PathBuf::from("/"), mount_point: PathBuf::from("/shm"), fs_type: ("tmpfs".to_string(), None), super_opts: vec![
                 "rw".to_string(),
                 "size=65536k".to_string(),
              ]}),
             ("121 1731 0:42 / /shm rw,nosuid,nodev,noexec,relatime shared:68 master:66 - tmpfs.123 shm rw,size=65536k",
-             Mountinfo{mount_point: PathBuf::from("/shm"), fs_type: ("tmpfs".to_string(), Some("123".to_string())), super_opts: vec![
+             Mountinfo{mount_root: PathBuf::from("/"), mount_point: PathBuf::from("/shm"), fs_type: ("tmpfs".to_string(), Some("123".to_string())), super_opts: vec![
                 "rw".to_string(),
                 "size=65536k".to_string(),
              ]}),

src/fs/hugetlb.rs

@@ -12,11 +12,11 @@ use log::warn;
 use std::io::Write;
 use std::path::PathBuf;
 
-use crate::error::ErrorKind::*;
-use crate::error::*;
-use crate::{flat_keyed_to_vec, read_u64_from};
+use crate::fs::error::ErrorKind::*;
+use crate::fs::error::*;
+use crate::fs::{flat_keyed_to_vec, read_u64_from};
 
-use crate::{
+use crate::fs::{
     ControllIdentifier, ControllerInternal, Controllers, HugePageResources, Resources, Subsystem,
 };
 
@@ -88,11 +88,11 @@ impl<'a> From<&'a Subsystem> for &'a HugeTlbController {
 
 impl HugeTlbController {
     /// Constructs a new `HugeTlbController` with `root` serving as the root of the control group.
-    pub fn new(root: PathBuf, v2: bool) -> Self {
+    pub fn new(point: PathBuf, root: PathBuf, v2: bool) -> Self {
         let sizes = get_hugepage_sizes();
         Self {
-            base: root.clone(),
-            path: root,
+            base: root,
+            path: point,
             sizes,
             v2,
         }
@@ -138,8 +138,11 @@ impl HugeTlbController {
     /// Get the limit (in bytes) of how much memory can be backed by hugepages of a certain size
     /// (`hugetlb_size`).
     pub fn limit_in_bytes(&self, hugetlb_size: &str) -> Result<u64> {
-        self.open_path(&format!("hugetlb.{}.limit_in_bytes", hugetlb_size), false)
-            .and_then(read_u64_from)
+        let mut file_name = format!("hugetlb.{}.limit_in_bytes", hugetlb_size);
+        if self.v2 {
+            file_name = format!("hugetlb.{}.max", hugetlb_size);
+        }
+        self.open_path(&file_name, false).and_then(read_u64_from)
     }
 
     /// Get the current usage of memory that is backed by hugepages of a certain size
@@ -165,19 +168,19 @@ impl HugeTlbController {
     /// Set the limit (in bytes) of how much memory can be backed by hugepages of a certain size
     /// (`hugetlb_size`).
     pub fn set_limit_in_bytes(&self, hugetlb_size: &str, limit: u64) -> Result<()> {
-        let mut file = format!("hugetlb.{}.limit_in_bytes", hugetlb_size);
+        let mut file_name = format!("hugetlb.{}.limit_in_bytes", hugetlb_size);
         if self.v2 {
-            file = format!("hugetlb.{}.max", hugetlb_size);
+            file_name = format!("hugetlb.{}.max", hugetlb_size);
         }
-        self.open_path(&file, true).and_then(|mut file| {
-            file.write_all(limit.to_string().as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+        self.open_path(&file_name, true).and_then(|mut file| {
+            file.write_all(limit.to_string().as_ref()).map_err(|e| {
+                Error::with_cause(WriteFailed(file_name.to_string(), limit.to_string()), e)
+            })
         })
     }
 }
 
 pub const HUGEPAGESIZE_DIR: &str = "/sys/kernel/mm/hugepages";
-use regex::Regex;
 use std::collections::HashMap;
 use std::fs;
 
@@ -259,37 +262,46 @@ pub fn get_decimal_abbrs() -> Vec<String> {
 }
 
 fn parse_size(s: &str, m: &HashMap<String, u128>) -> Result<u128> {
-    let re = Regex::new(r"(?P<num>\d+)(?P<mul>[kKmMgGtTpP]?)[bB]?$");
+    // Remove leading/trailing whitespace.
+    let s = s.trim();
 
-    if re.is_err() {
+    // Remove an optional trailing 'b' or 'B'
+    let s = if let Some(stripped) = s.strip_suffix('b').or_else(|| s.strip_suffix('B')) {
+        stripped
+    } else {
+        s
+    };
+
+    // Ensure that the string is not empty after stripping.
+    if s.is_empty() {
         return Err(Error::new(InvalidBytesSize));
     }
-    let caps = re.unwrap().captures(s).unwrap();
 
-    let num = caps.name("num");
-    let size: u128 = if let Some(num) = num {
-        let n = num.as_str().trim().parse::<u128>();
-        if n.is_err() {
-            return Err(Error::new(InvalidBytesSize));
-        }
-        n.unwrap()
-    } else {
+    // The last character should be the multiplier letter.
+    let last_char = s.chars().last().unwrap();
+    if !"kKmMgGtTpP".contains(last_char) {
         return Err(Error::new(InvalidBytesSize));
-    };
+    }
 
-    let q = caps.name("mul");
-    let mul: u128 = if let Some(q) = q {
-        let t = m.get(q.as_str());
-        if let Some(t) = t {
-            *t
-        } else {
-            return Err(Error::new(InvalidBytesSize));
-        }
-    } else {
+    // The numeric part is everything before the multiplier letter.
+    let num_part = &s[..s.len() - last_char.len_utf8()];
+    if num_part.trim().is_empty() {
         return Err(Error::new(InvalidBytesSize));
-    };
+    }
 
-    Ok(size * mul)
+    // Parse the numeric part into a u128.
+    let number: u128 = num_part
+        .trim()
+        .parse()
+        .map_err(|_| Error::new(InvalidBytesSize))?;
+
+    // Look up the multiplier in the provided HashMap.
+    let multiplier_key = last_char.to_string();
+    let multiplier = m
+        .get(&multiplier_key)
+        .ok_or_else(|| Error::new(InvalidBytesSize))?;
+
+    Ok(number * multiplier)
 }
 
 fn custom_size(mut size: f64, base: f64, m: &[String]) -> String {
@@ -301,3 +313,60 @@ fn custom_size(mut size: f64, base: f64, m: &[String]) -> String {
 
     format!("{}{}", size, m[i].as_str())
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_binary_size_valid() {
+        let m = get_binary_size_map();
+        // Valid inputs must include a multiplier letter.
+        assert_eq!(parse_size("1k", &m).unwrap(), KiB);
+        assert_eq!(parse_size("2m", &m).unwrap(), 2 * MiB);
+        assert_eq!(parse_size("3g", &m).unwrap(), 3 * GiB);
+        assert_eq!(parse_size("4t", &m).unwrap(), 4 * TiB);
+        assert_eq!(parse_size("5p", &m).unwrap(), 5 * PiB);
+    }
+
+    #[test]
+    fn test_decimal_size_valid() {
+        let m = get_decimal_size_map();
+        assert_eq!(parse_size("1k", &m).unwrap(), KB);
+        assert_eq!(parse_size("2m", &m).unwrap(), 2 * MB);
+        assert_eq!(parse_size("3g", &m).unwrap(), 3 * GB);
+        assert_eq!(parse_size("4t", &m).unwrap(), 4 * TB);
+        assert_eq!(parse_size("5p", &m).unwrap(), 5 * PB);
+    }
+
+    #[test]
+    fn test_trailing_b_suffix() {
+        let m = get_binary_size_map();
+        // Trailing 'b' or 'B' should be accepted.
+        assert_eq!(parse_size("1kb", &m).unwrap(), KiB);
+        assert_eq!(parse_size("2mB", &m).unwrap(), 2 * MiB);
+    }
+
+    #[test]
+    fn test_invalid_inputs() {
+        let m = get_binary_size_map();
+        // Missing multiplier letter results in error.
+        assert!(parse_size("1", &m).is_err());
+        // Invalid multiplier letter.
+        assert!(parse_size("10x", &m).is_err());
+        // Non-numeric input.
+        assert!(parse_size("abc", &m).is_err());
+        // Only multiplier letter with no number.
+        assert!(parse_size("k", &m).is_err());
+        // Number with an invalid trailing character.
+        assert!(parse_size("123z", &m).is_err());
+    }
+
+    #[test]
+    fn test_uppercase_multiplier_fails() {
+        let m = get_binary_size_map();
+        // Although the regex matches uppercase letters, the provided map only contains lowercase keys.
+        // Therefore, "1K" does not match any key and should produce an error.
+        assert!(parse_size("1K", &m).is_err());
+    }
+}

src/fs/memory.rs

@@ -14,14 +14,14 @@ use std::io::Write;
 use std::path::PathBuf;
 use std::sync::mpsc::Receiver;
 
-use crate::error::ErrorKind::*;
-use crate::error::*;
-use crate::events;
-use crate::{read_i64_from, read_string_from, read_u64_from};
+use crate::fs::error::ErrorKind::*;
+use crate::fs::error::*;
+use crate::fs::events;
+use crate::fs::{read_i64_from, read_string_from, read_u64_from};
 
-use crate::flat_keyed_to_hashmap;
+use crate::fs::flat_keyed_to_hashmap;
 
-use crate::{
+use crate::fs::{
     ControllIdentifier, ControllerInternal, Controllers, CustomizedAttribute, MaxValue,
     MemoryResources, Resources, Subsystem,
 };
@@ -142,9 +142,7 @@ fn parse_numa_stat(s: String) -> Result<NumaStat> {
     let hier_unevict_line = ls.next().unwrap_or_default();
 
     Ok(NumaStat {
-        total_pages: total_line
-            .split(|x| x == ' ' || x == '=')
-            .collect::<Vec<_>>()[1]
+        total_pages: total_line.split([' ', '=']).collect::<Vec<_>>()[1]
             .parse::<u64>()
             .unwrap_or(0),
         total_pages_per_node: {
@@ -157,9 +155,7 @@ fn parse_numa_stat(s: String) -> Result<NumaStat> {
                 })
                 .collect()
         },
-        file_pages: file_line
-            .split(|x| x == ' ' || x == '=')
-            .collect::<Vec<_>>()[1]
+        file_pages: file_line.split([' ', '=']).collect::<Vec<_>>()[1]
             .parse::<u64>()
             .unwrap_or(0),
         file_pages_per_node: {
@@ -172,9 +168,7 @@ fn parse_numa_stat(s: String) -> Result<NumaStat> {
                 })
                 .collect()
         },
-        anon_pages: anon_line
-            .split(|x| x == ' ' || x == '=')
-            .collect::<Vec<_>>()[1]
+        anon_pages: anon_line.split([' ', '=']).collect::<Vec<_>>()[1]
             .parse::<u64>()
             .unwrap_or(0),
         anon_pages_per_node: {
@@ -187,9 +181,7 @@ fn parse_numa_stat(s: String) -> Result<NumaStat> {
                 })
                 .collect()
         },
-        unevictable_pages: unevict_line
-            .split(|x| x == ' ' || x == '=')
-            .collect::<Vec<_>>()[1]
+        unevictable_pages: unevict_line.split([' ', '=']).collect::<Vec<_>>()[1]
             .parse::<u64>()
             .unwrap_or(0),
         unevictable_pages_per_node: {
@@ -204,9 +196,7 @@ fn parse_numa_stat(s: String) -> Result<NumaStat> {
         },
         hierarchical_total_pages: {
             if !hier_total_line.is_empty() {
-                hier_total_line
-                    .split(|x| x == ' ' || x == '=')
-                    .collect::<Vec<_>>()[1]
+                hier_total_line.split([' ', '=']).collect::<Vec<_>>()[1]
                     .parse::<u64>()
                     .unwrap_or(0)
             } else {
@@ -229,9 +219,7 @@ fn parse_numa_stat(s: String) -> Result<NumaStat> {
         },
         hierarchical_file_pages: {
             if !hier_file_line.is_empty() {
-                hier_file_line
-                    .split(|x| x == ' ' || x == '=')
-                    .collect::<Vec<_>>()[1]
+                hier_file_line.split([' ', '=']).collect::<Vec<_>>()[1]
                     .parse::<u64>()
                     .unwrap_or(0)
             } else {
@@ -254,9 +242,7 @@ fn parse_numa_stat(s: String) -> Result<NumaStat> {
         },
         hierarchical_anon_pages: {
             if !hier_anon_line.is_empty() {
-                hier_anon_line
-                    .split(|x| x == ' ' || x == '=')
-                    .collect::<Vec<_>>()[1]
+                hier_anon_line.split([' ', '=']).collect::<Vec<_>>()[1]
                     .parse::<u64>()
                     .unwrap_or(0)
             } else {
@@ -279,9 +265,7 @@ fn parse_numa_stat(s: String) -> Result<NumaStat> {
         },
         hierarchical_unevictable_pages: {
             if !hier_unevict_line.is_empty() {
-                hier_unevict_line
-                    .split(|x| x == ' ' || x == '=')
-                    .collect::<Vec<_>>()[1]
+                hier_unevict_line.split([' ', '=']).collect::<Vec<_>>()[1]
                     .parse::<u64>()
                     .unwrap_or(0)
             } else {
@@ -540,10 +524,10 @@ impl ControllerInternal for MemController {
 
 impl MemController {
     /// Contructs a new `MemController` with `root` serving as the root of the control group.
-    pub fn new(root: PathBuf, v2: bool) -> Self {
+    pub fn new(point: PathBuf, root: PathBuf, v2: bool) -> Self {
         Self {
-            base: root.clone(),
-            path: root,
+            base: root,
+            path: point,
             v2,
         }
     }
@@ -562,8 +546,9 @@ impl MemController {
             if let Some(v) = v {
                 let v = v.to_string();
                 self.open_path(f, true).and_then(|mut file| {
-                    file.write_all(v.as_ref())
-                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                    file.write_all(v.as_ref()).map_err(|e| {
+                        Error::with_cause(WriteFailed(f.to_string(), format!("{:?}", v)), e)
+                    })
                 })?;
             }
         }
@@ -572,18 +557,33 @@ impl MemController {
 
     // for v2
     pub fn get_mem(&self) -> Result<SetMemory> {
-        let mut m: SetMemory = Default::default();
-        self.get_max_value("memory.high")
-            .map(|x| m.high = Some(x))?;
-        self.get_max_value("memory.low").map(|x| m.low = Some(x))?;
-        self.get_max_value("memory.max").map(|x| m.max = Some(x))?;
-        self.get_max_value("memory.min").map(|x| m.min = Some(x))?;
+        let m = SetMemory {
+            high: self
+                .get_max_value("memory.high")
+                .map_or(Some(MaxValue::default()), Some),
+            low: self
+                .get_max_value("memory.low")
+                .map_or(Some(MaxValue::Value(0)), Some),
+            max: self
+                .get_max_value("memory.max")
+                .map_or(Some(MaxValue::default()), Some),
+            min: self
+                .get_max_value("memory.min")
+                .map_or(Some(MaxValue::Value(0)), Some),
+        };
 
         Ok(m)
     }
 
     fn memory_stat_v2(&self) -> Memory {
-        let set = self.get_mem().unwrap();
+        // NOTE: get_mem() always returns T, but let's
+        // still do `unwrap_or` for safety.
+        let set = self.get_mem().unwrap_or(SetMemory {
+            low: Some(MaxValue::Value(0)),
+            high: Some(MaxValue::default()),
+            max: Some(MaxValue::default()),
+            min: Some(MaxValue::Value(0)),
+        });
 
         Memory {
             fail_cnt: 0,
@@ -592,7 +592,10 @@ impl MemController {
                 .open_path("memory.current", false)
                 .and_then(read_u64_from)
                 .unwrap_or(0),
-            max_usage_in_bytes: 0,
+            max_usage_in_bytes: self
+                .open_path("memory.peak", false)
+                .and_then(read_u64_from)
+                .unwrap_or(0),
             move_charge_at_immigrate: 0,
             numa_stat: NumaStat::default(),
             oom_control: OomControl::default(),
@@ -726,7 +729,7 @@ impl MemController {
                 .open_path("memory.swap.events", false)
                 .and_then(flat_keyed_to_hashmap)
                 .map(|x| *x.get("fail").unwrap_or(&0) as u64)
-                .unwrap(),
+                .unwrap_or(0),
             limit_in_bytes: self
                 .open_path("memory.swap.max", false)
                 .and_then(read_i64_from)
@@ -735,7 +738,10 @@ impl MemController {
                 .open_path("memory.swap.current", false)
                 .and_then(read_u64_from)
                 .unwrap_or(0),
-            max_usage_in_bytes: 0,
+            max_usage_in_bytes: self
+                .open_path("memory.swap.peak", false)
+                .and_then(read_u64_from)
+                .unwrap_or(0),
         }
     }
 
@@ -769,8 +775,12 @@ impl MemController {
     /// Reset the fail counter
     pub fn reset_fail_count(&self) -> Result<()> {
         self.open_path("memory.failcnt", true).and_then(|mut file| {
-            file.write_all("0".to_string().as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+            file.write_all("0".to_string().as_ref()).map_err(|e| {
+                Error::with_cause(
+                    WriteFailed("memory.failcnt".to_string(), "0".to_string()),
+                    e,
+                )
+            })
         })
     }
 
@@ -783,8 +793,12 @@ impl MemController {
 
         self.open_path("memory.kmem.failcnt", true)
             .and_then(|mut file| {
-                file.write_all("0".to_string().as_ref())
-                    .map_err(|e| Error::with_cause(WriteFailed, e))
+                file.write_all("0".to_string().as_ref()).map_err(|e| {
+                    Error::with_cause(
+                        WriteFailed("memory.kmem.failcnt".to_string(), "0".to_string()),
+                        e,
+                    )
+                })
             })
     }
 
@@ -797,8 +811,12 @@ impl MemController {
 
         self.open_path("memory.kmem.tcp.failcnt", true)
             .and_then(|mut file| {
-                file.write_all("0".to_string().as_ref())
-                    .map_err(|e| Error::with_cause(WriteFailed, e))
+                file.write_all("0".to_string().as_ref()).map_err(|e| {
+                    Error::with_cause(
+                        WriteFailed("memory.kmem.tcp.failcnt".to_string(), "0".to_string()),
+                        e,
+                    )
+                })
             })
     }
 
@@ -806,8 +824,12 @@ impl MemController {
     pub fn reset_memswap_fail_count(&self) -> Result<()> {
         self.open_path("memory.memsw.failcnt", true)
             .and_then(|mut file| {
-                file.write_all("0".to_string().as_ref())
-                    .map_err(|e| Error::with_cause(WriteFailed, e))
+                file.write_all("0".to_string().as_ref()).map_err(|e| {
+                    Error::with_cause(
+                        WriteFailed("memory.memsw.failcnt".to_string(), "0".to_string()),
+                        e,
+                    )
+                })
             })
     }
 
@@ -815,20 +837,28 @@ impl MemController {
     pub fn reset_max_usage(&self) -> Result<()> {
         self.open_path("memory.max_usage_in_bytes", true)
             .and_then(|mut file| {
-                file.write_all("0".to_string().as_ref())
-                    .map_err(|e| Error::with_cause(WriteFailed, e))
+                file.write_all("0".to_string().as_ref()).map_err(|e| {
+                    Error::with_cause(
+                        WriteFailed("memory.max_usage_in_bytes".to_string(), "0".to_string()),
+                        e,
+                    )
+                })
             })
     }
 
     /// Set the memory usage limit of the control group, in bytes.
     pub fn set_limit(&self, limit: i64) -> Result<()> {
-        let mut file = "memory.limit_in_bytes";
+        let mut file_name = "memory.limit_in_bytes";
+        let mut limit_str = limit.to_string();
         if self.v2 {
-            file = "memory.max";
+            file_name = "memory.max";
+            if limit == -1 {
+                limit_str = "max".to_string();
+            }
         }
-        self.open_path(file, true).and_then(|mut file| {
-            file.write_all(limit.to_string().as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+        self.open_path(file_name, true).and_then(|mut file| {
+            file.write_all(limit_str.as_ref())
+                .map_err(|e| Error::with_cause(WriteFailed(file_name.to_string(), limit_str), e))
         })
     }
 
@@ -848,20 +878,27 @@ impl MemController {
                         warn!("memory.kmem.limit_in_bytes is unsupported by the kernel");
                         Ok(())
                     }
-                    Err(e) => Err(Error::with_cause(WriteFailed, e)),
+                    Err(e) => Err(Error::with_cause(
+                        WriteFailed("memory.kmem.limit_in_bytes".to_string(), limit.to_string()),
+                        e,
+                    )),
                 }
             })
     }
 
     /// Set the memory+swap limit of the control group, in bytes.
     pub fn set_memswap_limit(&self, limit: i64) -> Result<()> {
-        let mut file = "memory.memsw.limit_in_bytes";
+        let mut file_name = "memory.memsw.limit_in_bytes";
+        let mut limit_str = limit.to_string();
         if self.v2 {
-            file = "memory.swap.max";
+            file_name = "memory.swap.max";
+            if limit == -1 {
+                limit_str = "max".to_string();
+            }
         }
-        self.open_path(file, true).and_then(|mut file| {
-            file.write_all(limit.to_string().as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+        self.open_path(file_name, true).and_then(|mut file| {
+            file.write_all(limit_str.as_ref())
+                .map_err(|e| Error::with_cause(WriteFailed(file_name.to_string(), limit_str), e))
         })
     }
 
@@ -874,8 +911,15 @@ impl MemController {
 
         self.open_path("memory.kmem.tcp.limit_in_bytes", true)
             .and_then(|mut file| {
-                file.write_all(limit.to_string().as_ref())
-                    .map_err(|e| Error::with_cause(WriteFailed, e))
+                file.write_all(limit.to_string().as_ref()).map_err(|e| {
+                    Error::with_cause(
+                        WriteFailed(
+                            "memory.kmem.tcp.limit_in_bytes".to_string(),
+                            limit.to_string(),
+                        ),
+                        e,
+                    )
+                })
             })
     }
 
@@ -884,13 +928,14 @@ impl MemController {
     /// This limit is enforced when the system is nearing OOM conditions. Contrast this with the
     /// hard limit, which is _always_ enforced.
     pub fn set_soft_limit(&self, limit: i64) -> Result<()> {
-        let mut file = "memory.soft_limit_in_bytes";
+        let mut file_name = "memory.soft_limit_in_bytes";
         if self.v2 {
-            file = "memory.low"
+            file_name = "memory.low"
         }
-        self.open_path(file, true).and_then(|mut file| {
-            file.write_all(limit.to_string().as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+        self.open_path(file_name, true).and_then(|mut file| {
+            file.write_all(limit.to_string().as_ref()).map_err(|e| {
+                Error::with_cause(WriteFailed(file_name.to_string(), limit.to_string()), e)
+            })
         })
     }
 
@@ -899,22 +944,27 @@ impl MemController {
     ///
     /// Note that a value of zero does not imply that the process will not be swapped out.
     pub fn set_swappiness(&self, swp: u64) -> Result<()> {
-        let mut file = "memory.swappiness";
+        let mut file_name = "memory.swappiness";
         if self.v2 {
-            file = "memory.swap.max"
+            file_name = "memory.swap.max"
         }
 
-        self.open_path(file, true).and_then(|mut file| {
-            file.write_all(swp.to_string().as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+        self.open_path(file_name, true).and_then(|mut file| {
+            file.write_all(swp.to_string().as_ref()).map_err(|e| {
+                Error::with_cause(WriteFailed(file_name.to_string(), swp.to_string()), e)
+            })
         })
     }
 
     pub fn disable_oom_killer(&self) -> Result<()> {
         self.open_path("memory.oom_control", true)
             .and_then(|mut file| {
-                file.write_all("1".to_string().as_ref())
-                    .map_err(|e| Error::with_cause(WriteFailed, e))
+                file.write_all("1".to_string().as_ref()).map_err(|e| {
+                    Error::with_cause(
+                        WriteFailed("memory.oom_control".to_string(), "1".to_string()),
+                        e,
+                    )
+                })
             })
     }
 
@@ -952,7 +1002,7 @@ impl<'a> From<&'a Subsystem> for &'a MemController {
 
 #[cfg(test)]
 mod tests {
-    use crate::memory::{
+    use crate::fs::memory::{
         parse_memory_stat, parse_numa_stat, parse_oom_control, MemoryStat, NumaStat, OomControl,
     };
 

src/fs/mod.rs

@@ -0,0 +1,995 @@
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2020 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+#![allow(clippy::unnecessary_unwrap)]
+
+use std::collections::HashMap;
+use std::fmt;
+use std::fs::{self, File};
+use std::io::{BufRead, BufReader, Read, Write};
+use std::path::{Path, PathBuf};
+use std::str::FromStr;
+
+macro_rules! update_and_test {
+    ($self: ident, $set_func:ident, $value:expr, $get_func:ident) => {
+        if let Some(v) = $value {
+            $self.$set_func(v)?;
+            if $self.$get_func()? != v {
+                return Err(Error::new(Other));
+            }
+        }
+    };
+}
+
+macro_rules! update {
+    ($self: ident, $set_func:ident, $value:expr) => {
+        if let Some(v) = $value {
+            let _ = $self.$set_func(v);
+        }
+    };
+}
+
+pub mod blkio;
+pub mod cgroup;
+pub mod cgroup_builder;
+pub mod cpu;
+pub mod cpuacct;
+pub mod cpuset;
+pub mod devices;
+pub mod error;
+pub mod events;
+pub mod freezer;
+pub mod hierarchies;
+pub mod hugetlb;
+pub mod memory;
+pub mod net_cls;
+pub mod net_prio;
+pub mod perf_event;
+pub mod pid;
+pub mod rdma;
+pub mod systemd;
+
+use crate::fs::blkio::BlkIoController;
+use crate::fs::cpu::CpuController;
+use crate::fs::cpuacct::CpuAcctController;
+use crate::fs::cpuset::CpuSetController;
+use crate::fs::devices::DevicesController;
+use crate::fs::error::ErrorKind::*;
+use crate::fs::error::*;
+use crate::fs::freezer::FreezerController;
+use crate::fs::hugetlb::HugeTlbController;
+use crate::fs::memory::MemController;
+use crate::fs::net_cls::NetClsController;
+use crate::fs::net_prio::NetPrioController;
+use crate::fs::perf_event::PerfEventController;
+use crate::fs::pid::PidController;
+use crate::fs::rdma::RdmaController;
+use crate::fs::systemd::SystemdController;
+
+#[doc(inline)]
+pub use crate::fs::cgroup::Cgroup;
+
+/// Contains all the subsystems that are available in this crate.
+#[derive(Debug, Clone)]
+pub enum Subsystem {
+    /// Controller for the `Pid` subsystem, see `PidController` for more information.
+    Pid(PidController),
+    /// Controller for the `Mem` subsystem, see `MemController` for more information.
+    Mem(MemController),
+    /// Controller for the `CpuSet subsystem, see `CpuSetController` for more information.
+    CpuSet(CpuSetController),
+    /// Controller for the `CpuAcct` subsystem, see `CpuAcctController` for more information.
+    CpuAcct(CpuAcctController),
+    /// Controller for the `Cpu` subsystem, see `CpuController` for more information.
+    Cpu(CpuController),
+    /// Controller for the `Devices` subsystem, see `DevicesController` for more information.
+    Devices(DevicesController),
+    /// Controller for the `Freezer` subsystem, see `FreezerController` for more information.
+    Freezer(FreezerController),
+    /// Controller for the `NetCls` subsystem, see `NetClsController` for more information.
+    NetCls(NetClsController),
+    /// Controller for the `BlkIo` subsystem, see `BlkIoController` for more information.
+    BlkIo(BlkIoController),
+    /// Controller for the `PerfEvent` subsystem, see `PerfEventController` for more information.
+    PerfEvent(PerfEventController),
+    /// Controller for the `NetPrio` subsystem, see `NetPrioController` for more information.
+    NetPrio(NetPrioController),
+    /// Controller for the `HugeTlb` subsystem, see `HugeTlbController` for more information.
+    HugeTlb(HugeTlbController),
+    /// Controller for the `Rdma` subsystem, see `RdmaController` for more information.
+    Rdma(RdmaController),
+    /// Controller for the `Systemd` subsystem, see `SystemdController` for more information.
+    Systemd(SystemdController),
+}
+
+#[doc(hidden)]
+#[derive(Eq, PartialEq, Debug, Clone)]
+pub enum Controllers {
+    Pids,
+    Mem,
+    CpuSet,
+    CpuAcct,
+    Cpu,
+    Devices,
+    Freezer,
+    NetCls,
+    BlkIo,
+    PerfEvent,
+    NetPrio,
+    HugeTlb,
+    Rdma,
+    Systemd,
+}
+
+impl fmt::Display for Controllers {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            Controllers::Pids => write!(f, "pids"),
+            Controllers::Mem => write!(f, "memory"),
+            Controllers::CpuSet => write!(f, "cpuset"),
+            Controllers::CpuAcct => write!(f, "cpuacct"),
+            Controllers::Cpu => write!(f, "cpu"),
+            Controllers::Devices => write!(f, "devices"),
+            Controllers::Freezer => write!(f, "freezer"),
+            Controllers::NetCls => write!(f, "net_cls"),
+            Controllers::BlkIo => write!(f, "blkio"),
+            Controllers::PerfEvent => write!(f, "perf_event"),
+            Controllers::NetPrio => write!(f, "net_prio"),
+            Controllers::HugeTlb => write!(f, "hugetlb"),
+            Controllers::Rdma => write!(f, "rdma"),
+            Controllers::Systemd => write!(f, "name=systemd"),
+        }
+    }
+}
+
+mod sealed {
+    use super::*;
+
+    pub trait ControllerInternal {
+        fn apply(&self, res: &Resources) -> Result<()>;
+
+        // meta stuff
+        fn control_type(&self) -> Controllers;
+        fn get_path(&self) -> &PathBuf;
+        fn get_path_mut(&mut self) -> &mut PathBuf;
+        fn get_base(&self) -> &PathBuf;
+
+        /// Hooks running after controller crated, if have
+        fn post_create(&self) {}
+
+        fn is_v2(&self) -> bool {
+            false
+        }
+
+        fn open_path(&self, p: &str, w: bool) -> Result<File> {
+            let mut path = self.get_path().clone();
+            path.push(p);
+
+            if w {
+                match File::create(&path) {
+                    Err(e) => Err(Error::with_cause(
+                        ErrorKind::WriteFailed(
+                            path.display().to_string(),
+                            "[CREATE FILE]".to_string(),
+                        ),
+                        e,
+                    )),
+                    Ok(file) => Ok(file),
+                }
+            } else {
+                match File::open(&path) {
+                    Err(e) => Err(Error::with_cause(
+                        ErrorKind::ReadFailed(path.display().to_string()),
+                        e,
+                    )),
+                    Ok(file) => Ok(file),
+                }
+            }
+        }
+
+        fn get_max_value(&self, f: &str) -> Result<MaxValue> {
+            self.open_path(f, false).and_then(|mut file| {
+                let mut string = String::new();
+                let res = file.read_to_string(&mut string);
+                match res {
+                    Ok(_) => parse_max_value(&string),
+                    Err(e) => Err(Error::with_cause(ReadFailed(f.to_string()), e)),
+                }
+            })
+        }
+
+        #[doc(hidden)]
+        fn path_exists(&self, p: &str) -> bool {
+            std::path::Path::new(p).exists()
+        }
+    }
+
+    pub trait CustomizedAttribute: ControllerInternal {
+        fn set(&self, key: &str, value: &str) -> Result<()> {
+            self.open_path(key, true).and_then(|mut file| {
+                file.write_all(value.as_ref()).map_err(|e| {
+                    Error::with_cause(WriteFailed(key.to_string(), value.to_string()), e)
+                })
+            })
+        }
+
+        #[allow(dead_code)]
+        fn get(&self, key: &str) -> Result<String> {
+            self.open_path(key, false).and_then(|mut file: File| {
+                let mut string = String::new();
+                match file.read_to_string(&mut string) {
+                    Ok(_) => Ok(string.trim().to_owned()),
+                    Err(e) => Err(Error::with_cause(ReadFailed(key.to_string()), e)),
+                }
+            })
+        }
+    }
+}
+
+pub(crate) use crate::fs::sealed::{ControllerInternal, CustomizedAttribute};
+use crate::CgroupPid;
+
+/// A Controller is a subsystem attached to the control group.
+///
+/// Implementors are able to control certain aspects of a control group.
+pub trait Controller {
+    #[doc(hidden)]
+    fn control_type(&self) -> Controllers;
+
+    /// The file system path to the controller.
+    fn path(&self) -> &Path;
+
+    /// Root path of the file system to the controller.
+    fn base(&self) -> &Path;
+
+    /// Apply a set of resources to the Controller, invoking its internal functions to pass the
+    /// kernel the information.
+    fn apply(&self, res: &Resources) -> Result<()>;
+
+    /// Create this controller
+    fn create(&self) -> Result<()>;
+
+    /// Does this controller already exist?
+    fn exists(&self) -> bool;
+
+    /// Set notify_on_release
+    fn set_notify_on_release(&self, enable: bool) -> Result<()>;
+
+    /// Set release_agent
+    fn set_release_agent(&self, path: &str) -> Result<()>;
+
+    /// Delete the controller.
+    fn delete(&self) -> Result<()>;
+
+    /// Attach a task to this controller.
+    fn add_task(&self, pid: &CgroupPid) -> Result<()>;
+
+    /// Attach a task to this controller.
+    fn add_task_by_tgid(&self, pid: &CgroupPid) -> Result<()>;
+
+    /// set cgroup type.
+    fn set_cgroup_type(&self, cgroup_type: &str) -> Result<()>;
+
+    /// get cgroup type.
+    fn get_cgroup_type(&self) -> Result<String>;
+
+    /// Get the list of tasks that this controller has.
+    fn tasks(&self) -> Vec<CgroupPid>;
+
+    /// Get the list of procs that this controller has.
+    fn procs(&self) -> Vec<CgroupPid>;
+
+    fn v2(&self) -> bool;
+}
+
+impl<T> Controller for T
+where
+    T: ControllerInternal,
+{
+    fn control_type(&self) -> Controllers {
+        ControllerInternal::control_type(self)
+    }
+
+    fn path(&self) -> &Path {
+        self.get_path()
+    }
+
+    fn base(&self) -> &Path {
+        self.get_base()
+    }
+
+    /// Apply a set of resources to the Controller, invoking its internal functions to pass the
+    /// kernel the information.
+    fn apply(&self, res: &Resources) -> Result<()> {
+        ControllerInternal::apply(self, res)
+    }
+
+    /// Create this controller
+    fn create(&self) -> Result<()> {
+        std::fs::create_dir_all(self.get_path())
+            .map_err(|err| Error::with_cause(ErrorKind::FsError, err))?;
+        self.post_create();
+        Ok(())
+    }
+
+    /// Set notify_on_release
+    fn set_notify_on_release(&self, enable: bool) -> Result<()> {
+        if self.is_v2() {
+            return Err(Error::new(ErrorKind::CgroupVersion));
+        }
+        self.open_path("notify_on_release", true)
+            .and_then(|mut file| {
+                write!(file, "{}", enable as i32).map_err(|e| {
+                    Error::with_cause(
+                        ErrorKind::WriteFailed("notify_on_release".to_string(), enable.to_string()),
+                        e,
+                    )
+                })
+            })
+    }
+
+    /// Set release_agent
+    fn set_release_agent(&self, path: &str) -> Result<()> {
+        if self.is_v2() {
+            return Err(Error::new(ErrorKind::CgroupVersion));
+        }
+        self.open_path("release_agent", true).and_then(|mut file| {
+            file.write_all(path.as_bytes()).map_err(|e| {
+                Error::with_cause(
+                    ErrorKind::WriteFailed("release_agent".to_string(), path.to_string()),
+                    e,
+                )
+            })
+        })
+    }
+    /// Does this controller already exist?
+    fn exists(&self) -> bool {
+        self.get_path().exists()
+    }
+
+    /// Delete the controller.
+    fn delete(&self) -> Result<()> {
+        if !self.get_path().exists() {
+            return Ok(());
+        }
+
+        // Compatible with runC for remove dir operation
+        // https://github.com/opencontainers/runc/blob/main/libcontainer/cgroups/utils.go#L272
+        //
+        // We trying to remove all paths five times with increasing delay between tries.
+        // If after all there are not removed cgroups - appropriate error will be
+        // returned.
+        let mut delay = std::time::Duration::from_millis(10);
+        let cgroup_path = self.get_path();
+        for _i in 0..4 {
+            if let Ok(()) = remove_dir(cgroup_path) {
+                return Ok(());
+            }
+            std::thread::sleep(delay);
+            delay *= 2;
+        }
+
+        remove_dir(cgroup_path)
+    }
+
+    /// Attach a task to this controller.
+    fn add_task(&self, pid: &CgroupPid) -> Result<()> {
+        let mut file_name = "tasks";
+        if self.is_v2() {
+            file_name = "cgroup.threads";
+        }
+        self.open_path(file_name, true).and_then(|mut file| {
+            file.write_all(pid.pid.to_string().as_ref()).map_err(|e| {
+                Error::with_cause(
+                    ErrorKind::WriteFailed(file_name.to_string(), pid.pid.to_string()),
+                    e,
+                )
+            })
+        })
+    }
+
+    /// Attach a task to this controller by thread group id.
+    fn add_task_by_tgid(&self, pid: &CgroupPid) -> Result<()> {
+        let file_name = "cgroup.procs";
+        self.open_path(file_name, true).and_then(|mut file| {
+            file.write_all(pid.pid.to_string().as_ref()).map_err(|e| {
+                Error::with_cause(
+                    ErrorKind::WriteFailed(file_name.to_string(), pid.pid.to_string()),
+                    e,
+                )
+            })
+        })
+    }
+
+    /// Get the list of procs that this controller has.
+    fn procs(&self) -> Vec<CgroupPid> {
+        let file_name = "cgroup.procs";
+        self.open_path(file_name, false)
+            .map(|file| {
+                let bf = BufReader::new(file);
+                let mut v = Vec::new();
+                for line in bf.lines() {
+                    match line {
+                        Ok(line) => {
+                            let n = line.trim().parse().unwrap_or(0u64);
+                            v.push(n);
+                        }
+                        Err(_) => break,
+                    }
+                }
+                v.into_iter().map(CgroupPid::from).collect()
+            })
+            .unwrap_or_default()
+    }
+
+    /// Get the list of tasks that this controller has.
+    fn tasks(&self) -> Vec<CgroupPid> {
+        let mut file_name = "tasks";
+        if self.is_v2() {
+            file_name = "cgroup.threads";
+        }
+        self.open_path(file_name, false)
+            .map(|file| {
+                let bf = BufReader::new(file);
+                let mut v = Vec::new();
+                for line in bf.lines() {
+                    match line {
+                        Ok(line) => {
+                            let n = line.trim().parse().unwrap_or(0u64);
+                            v.push(n);
+                        }
+                        Err(_) => break,
+                    }
+                }
+                v.into_iter().map(CgroupPid::from).collect()
+            })
+            .unwrap_or_default()
+    }
+
+    /// set cgroup.type
+    fn set_cgroup_type(&self, cgroup_type: &str) -> Result<()> {
+        if !self.is_v2() {
+            return Err(Error::new(ErrorKind::CgroupVersion));
+        }
+        let file_name = "cgroup.type";
+        self.open_path(file_name, true).and_then(|mut file| {
+            file.write_all(cgroup_type.as_bytes()).map_err(|e| {
+                Error::with_cause(
+                    ErrorKind::WriteFailed(file_name.to_string(), cgroup_type.to_string()),
+                    e,
+                )
+            })
+        })
+    }
+
+    /// get cgroup.type
+    fn get_cgroup_type(&self) -> Result<String> {
+        if !self.is_v2() {
+            return Err(Error::new(ErrorKind::CgroupVersion));
+        }
+        let file_name = "cgroup.type";
+        self.open_path(file_name, false).and_then(|mut file: File| {
+            let mut string = String::new();
+            match file.read_to_string(&mut string) {
+                Ok(_) => Ok(string.trim().to_owned()),
+                Err(e) => Err(Error::with_cause(
+                    ErrorKind::ReadFailed(file_name.to_string()),
+                    e,
+                )),
+            }
+        })
+    }
+
+    fn v2(&self) -> bool {
+        self.is_v2()
+    }
+}
+
+// remove_dir aims to remove cgroup path. It does so recursively,
+// by removing any subdirectories (sub-cgroups) first.
+fn remove_dir(dir: &Path) -> Result<()> {
+    // try the fast path first.
+    if fs::remove_dir(dir).is_ok() {
+        return Ok(());
+    }
+
+    if dir.exists() && dir.is_dir() {
+        for entry in fs::read_dir(dir)
+            .map_err(|e| Error::with_cause(ReadFailed(dir.display().to_string()), e))?
+        {
+            let entry =
+                entry.map_err(|e| Error::with_cause(ReadFailed(dir.display().to_string()), e))?;
+            let path = entry.path();
+            if path.is_dir() {
+                remove_dir(&path)?;
+            }
+        }
+        fs::remove_dir(dir).map_err(|e| Error::with_cause(RemoveFailed, e))?;
+    }
+
+    Ok(())
+}
+
+#[doc(hidden)]
+pub trait ControllIdentifier {
+    fn controller_type() -> Controllers;
+}
+
+/// Control group hierarchy (right now, only V1 is supported, but in the future Unified will be
+/// implemented as well).
+pub trait Hierarchy: std::fmt::Debug + Send + Sync {
+    /// Returns what subsystems are supported by the hierarchy.
+    fn subsystems(&self) -> Vec<Subsystem>;
+
+    /// Returns the root directory of the hierarchy.
+    fn root(&self) -> PathBuf;
+
+    /// Return a handle to the root control group in the hierarchy.
+    fn root_control_group(&self) -> Cgroup;
+
+    /// Return a handle to the parent control group in the hierarchy.
+    fn parent_control_group(&self, path: &str) -> Cgroup;
+
+    fn v2(&self) -> bool;
+}
+
+/// Resource limits for the memory subsystem.
+#[derive(Debug, Clone, Eq, PartialEq, Default)]
+#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
+pub struct MemoryResources {
+    /// How much memory (in bytes) can the kernel consume.
+    pub kernel_memory_limit: Option<i64>,
+    /// Upper limit of memory usage of the control group's tasks.
+    pub memory_hard_limit: Option<i64>,
+    /// How much memory the tasks in the control group can use when the system is under memory
+    /// pressure.
+    pub memory_soft_limit: Option<i64>,
+    /// How much of the kernel's memory (in bytes) can be used for TCP-related buffers.
+    pub kernel_tcp_memory_limit: Option<i64>,
+    /// How much memory and swap together can the tasks in the control group use.
+    pub memory_swap_limit: Option<i64>,
+    /// Controls the tendency of the kernel to swap out parts of the address space of the tasks to
+    /// disk. Lower value implies less likely.
+    ///
+    /// Note, however, that a value of zero does not mean the process is never swapped out. Use the
+    /// traditional `mlock(2)` system call for that purpose.
+    pub swappiness: Option<u64>,
+    /// Customized key-value attributes
+    ///
+    /// # Usage:
+    /// ```
+    /// let resource = &mut cgroups_rs::fs::Resources::default();
+    /// resource.memory.attrs.insert("memory.numa_balancing".to_string(), "true".to_string());
+    /// // apply here
+    /// ```
+    pub attrs: HashMap<String, String>,
+}
+
+/// Resources limits on the number of processes.
+#[derive(Debug, Clone, Eq, PartialEq, Default)]
+#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
+pub struct PidResources {
+    /// The maximum number of processes that can exist in the control group.
+    ///
+    /// Note that attaching processes to the control group will still succeed _even_ if the limit
+    /// would be violated, however forks/clones inside the control group will have with `EAGAIN` if
+    /// they would violate the limit set here.
+    pub maximum_number_of_processes: Option<MaxValue>,
+}
+
+/// Resources limits about how the tasks can use the CPU.
+#[derive(Debug, Clone, Eq, PartialEq, Default)]
+#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
+pub struct CpuResources {
+    // cpuset
+    /// A comma-separated list of CPU IDs where the task in the control group can run. Dashes
+    /// between numbers indicate ranges.
+    pub cpus: Option<String>,
+    /// Same syntax as the `cpus` field of this structure, but applies to memory nodes instead of
+    /// processors.
+    pub mems: Option<String>,
+    // cpu
+    /// Weight of how much of the total CPU time should this control group get. Note that this is
+    /// hierarchical, so this is weighted against the siblings of this control group.
+    pub shares: Option<u64>,
+    /// In one `period`, how much can the tasks run in microseconds.
+    pub quota: Option<i64>,
+    /// Period of time in microseconds.
+    pub period: Option<u64>,
+    /// This is currently a no-operation.
+    pub realtime_runtime: Option<i64>,
+    /// This is currently a no-operation.
+    pub realtime_period: Option<u64>,
+    /// Customized key-value attributes
+    /// # Usage:
+    /// ```
+    /// let resource = &mut cgroups_rs::fs::Resources::default();
+    /// resource.cpu.attrs.insert("cpu.cfs_init_buffer_us".to_string(), "10".to_string());
+    /// // apply here
+    /// ```
+    pub attrs: HashMap<String, String>,
+}
+
+/// A device resource that can be allowed or denied access to.
+#[derive(Debug, Clone, Eq, PartialEq, Default)]
+#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
+pub struct DeviceResource {
+    /// If true, access to the device is allowed, otherwise it's denied.
+    pub allow: bool,
+    /// `'c'` for character device, `'b'` for block device; or `'a'` for all devices.
+    pub devtype: crate::fs::devices::DeviceType,
+    /// The major number of the device.
+    pub major: i64,
+    /// The minor number of the device.
+    pub minor: i64,
+    /// Sequence of `'r'`, `'w'` or `'m'`, each denoting read, write or mknod permissions.
+    pub access: Vec<crate::fs::devices::DevicePermissions>,
+}
+
+/// Limit the usage of devices for the control group's tasks.
+#[derive(Debug, Clone, Eq, PartialEq, Default)]
+#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
+pub struct DeviceResources {
+    /// For each device in the list, the limits in the structure are applied.
+    pub devices: Vec<DeviceResource>,
+}
+
+/// Assigned priority for a network device.
+#[derive(Debug, Clone, Eq, PartialEq, Default)]
+#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
+pub struct NetworkPriority {
+    /// The name (as visible in `ifconfig`) of the interface.
+    pub name: String,
+    /// Assigned priority.
+    pub priority: u64,
+}
+
+/// Collections of limits and tags that can be imposed on packets emitted by the tasks in the
+/// control group.
+#[derive(Debug, Clone, Eq, PartialEq, Default)]
+#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
+pub struct NetworkResources {
+    /// The networking class identifier to attach to the packets.
+    ///
+    /// This can then later be used in iptables and such to have special rules.
+    pub class_id: Option<u64>,
+    /// Priority of the egress traffic for each interface.
+    pub priorities: Vec<NetworkPriority>,
+}
+
+/// A hugepage type and its consumption limit for the control group.
+#[derive(Debug, Clone, Eq, PartialEq, Default)]
+#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
+pub struct HugePageResource {
+    /// The size of the hugepage, i.e. `2MB`, `1GB`, etc.
+    pub size: String,
+    /// The amount of bytes (of memory consumed by the tasks) that are allowed to be backed by
+    /// hugepages.
+    pub limit: u64,
+}
+
+/// Provides the ability to set consumption limit on each type of hugepages.
+#[derive(Debug, Clone, Eq, PartialEq, Default)]
+#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
+pub struct HugePageResources {
+    /// Set a limit of consumption for each hugepages type.
+    pub limits: Vec<HugePageResource>,
+}
+
+/// Weight for a particular block device.
+#[derive(Debug, Clone, Eq, PartialEq, Default)]
+#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
+pub struct BlkIoDeviceResource {
+    /// The major number of the device.
+    pub major: u64,
+    /// The minor number of the device.
+    pub minor: u64,
+    /// The weight of the device against the descendant nodes.
+    pub weight: Option<u16>,
+    /// The weight of the device against the sibling nodes.
+    pub leaf_weight: Option<u16>,
+}
+
+/// Provides the ability to throttle a device (both byte/sec, and IO op/s)
+#[derive(Debug, Clone, Eq, PartialEq, Default)]
+#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
+pub struct BlkIoDeviceThrottleResource {
+    /// The major number of the device.
+    pub major: u64,
+    /// The minor number of the device.
+    pub minor: u64,
+    /// The rate.
+    pub rate: u64,
+}
+
+/// General block I/O resource limits.
+#[derive(Debug, Clone, Eq, PartialEq, Default)]
+#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
+pub struct BlkIoResources {
+    /// The weight of the control group against descendant nodes.
+    pub weight: Option<u16>,
+    /// The weight of the control group against sibling nodes.
+    pub leaf_weight: Option<u16>,
+    /// For each device, a separate weight (both normal and leaf) can be provided.
+    pub weight_device: Vec<BlkIoDeviceResource>,
+    /// Throttled read bytes/second can be provided for each device.
+    pub throttle_read_bps_device: Vec<BlkIoDeviceThrottleResource>,
+    /// Throttled read IO operations per second can be provided for each device.
+    pub throttle_read_iops_device: Vec<BlkIoDeviceThrottleResource>,
+    /// Throttled written bytes/second can be provided for each device.
+    pub throttle_write_bps_device: Vec<BlkIoDeviceThrottleResource>,
+    /// Throttled write IO operations per second can be provided for each device.
+    pub throttle_write_iops_device: Vec<BlkIoDeviceThrottleResource>,
+
+    /// Customized key-value attributes
+    /// # Usage:
+    /// ```
+    /// let resource = &mut cgroups_rs::fs::Resources::default();
+    /// resource.blkio.attrs.insert("io.cost.weight".to_string(), "10".to_string());
+    /// // apply here
+    /// ```
+    pub attrs: HashMap<String, String>,
+}
+
+/// The resource limits and constraints that will be set on the control group.
+#[derive(Debug, Clone, Eq, PartialEq, Default)]
+#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
+pub struct Resources {
+    /// Memory usage related limits.
+    pub memory: MemoryResources,
+    /// Process identifier related limits.
+    pub pid: PidResources,
+    /// CPU related limits.
+    pub cpu: CpuResources,
+    /// Device related limits.
+    pub devices: DeviceResources,
+    /// Network related tags and limits.
+    pub network: NetworkResources,
+    /// Hugepages consumption related limits.
+    pub hugepages: HugePageResources,
+    /// Block device I/O related limits.
+    pub blkio: BlkIoResources,
+}
+
+impl Subsystem {
+    fn enter(self, path: &Path) -> Self {
+        match self {
+            Subsystem::Pid(mut cont) => Subsystem::Pid({
+                cont.get_path_mut().push(path);
+                cont
+            }),
+            Subsystem::Mem(mut cont) => Subsystem::Mem({
+                cont.get_path_mut().push(path);
+                cont
+            }),
+            Subsystem::CpuSet(mut cont) => Subsystem::CpuSet({
+                cont.get_path_mut().push(path);
+                cont
+            }),
+            Subsystem::CpuAcct(mut cont) => Subsystem::CpuAcct({
+                cont.get_path_mut().push(path);
+                cont
+            }),
+            Subsystem::Cpu(mut cont) => Subsystem::Cpu({
+                cont.get_path_mut().push(path);
+                cont
+            }),
+            Subsystem::Devices(mut cont) => Subsystem::Devices({
+                cont.get_path_mut().push(path);
+                cont
+            }),
+            Subsystem::Freezer(mut cont) => Subsystem::Freezer({
+                cont.get_path_mut().push(path);
+                cont
+            }),
+            Subsystem::NetCls(mut cont) => Subsystem::NetCls({
+                cont.get_path_mut().push(path);
+                cont
+            }),
+            Subsystem::BlkIo(mut cont) => Subsystem::BlkIo({
+                cont.get_path_mut().push(path);
+                cont
+            }),
+            Subsystem::PerfEvent(mut cont) => Subsystem::PerfEvent({
+                cont.get_path_mut().push(path);
+                cont
+            }),
+            Subsystem::NetPrio(mut cont) => Subsystem::NetPrio({
+                cont.get_path_mut().push(path);
+                cont
+            }),
+            Subsystem::HugeTlb(mut cont) => Subsystem::HugeTlb({
+                cont.get_path_mut().push(path);
+                cont
+            }),
+            Subsystem::Rdma(mut cont) => Subsystem::Rdma({
+                cont.get_path_mut().push(path);
+                cont
+            }),
+            Subsystem::Systemd(mut cont) => Subsystem::Systemd({
+                cont.get_path_mut().push(path);
+                cont
+            }),
+        }
+    }
+
+    pub fn to_controller(&self) -> &dyn Controller {
+        match self {
+            Subsystem::Pid(cont) => cont,
+            Subsystem::Mem(cont) => cont,
+            Subsystem::CpuSet(cont) => cont,
+            Subsystem::CpuAcct(cont) => cont,
+            Subsystem::Cpu(cont) => cont,
+            Subsystem::Devices(cont) => cont,
+            Subsystem::Freezer(cont) => cont,
+            Subsystem::NetCls(cont) => cont,
+            Subsystem::BlkIo(cont) => cont,
+            Subsystem::PerfEvent(cont) => cont,
+            Subsystem::NetPrio(cont) => cont,
+            Subsystem::HugeTlb(cont) => cont,
+            Subsystem::Rdma(cont) => cont,
+            Subsystem::Systemd(cont) => cont,
+        }
+    }
+
+    pub fn controller_name(&self) -> String {
+        self.to_controller().control_type().to_string()
+    }
+}
+
+/// The values for `memory.hight` or `pids.max`
+#[derive(Eq, PartialEq, Copy, Clone, Debug)]
+#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
+pub enum MaxValue {
+    /// This value is returned when the text is `"max"`.
+    Max,
+    /// When the value is a numerical value, they are returned via this enum field.
+    Value(i64),
+}
+
+#[allow(clippy::derivable_impls)]
+impl Default for MaxValue {
+    fn default() -> Self {
+        MaxValue::Max
+    }
+}
+
+impl MaxValue {
+    #[allow(clippy::should_implement_trait, clippy::wrong_self_convention)]
+    fn to_i64(&self) -> i64 {
+        match self {
+            MaxValue::Max => -1,
+            MaxValue::Value(num) => *num,
+        }
+    }
+}
+
+impl fmt::Display for MaxValue {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            MaxValue::Max => write!(f, "max"),
+            MaxValue::Value(num) => write!(f, "{}", num),
+        }
+    }
+}
+
+pub fn parse_max_value(s: &str) -> Result<MaxValue> {
+    if s.trim() == "max" {
+        return Ok(MaxValue::Max);
+    }
+    match s.trim().parse() {
+        Ok(val) => Ok(MaxValue::Value(val)),
+        Err(e) => Err(Error::with_cause(ParseError, e)),
+    }
+}
+
+// Flat keyed
+//  KEY0 VAL0\n
+//  KEY1 VAL1\n
+pub fn flat_keyed_to_vec(mut file: File) -> Result<Vec<(String, i64)>> {
+    let mut content = String::new();
+    file.read_to_string(&mut content)
+        .map_err(|e| Error::with_cause(ReadFailed("FIXME: read_string_from".to_string()), e))?;
+
+    let mut v = Vec::new();
+    for line in content.lines() {
+        let parts: Vec<&str> = line.split(' ').collect();
+        if parts.len() == 2 {
+            if let Ok(i) = parts[1].parse::<i64>() {
+                v.push((parts[0].to_string(), i));
+            }
+        }
+    }
+    Ok(v)
+}
+
+// Flat keyed
+//  KEY0 VAL0\n
+//  KEY1 VAL1\n
+pub fn flat_keyed_to_hashmap(mut file: File) -> Result<HashMap<String, i64>> {
+    let mut content = String::new();
+    file.read_to_string(&mut content)
+        .map_err(|e| Error::with_cause(ReadFailed("FIXME: read_string_from".to_string()), e))?;
+
+    let mut h = HashMap::new();
+    for line in content.lines() {
+        let parts: Vec<&str> = line.split(' ').collect();
+        if parts.len() == 2 {
+            if let Ok(i) = parts[1].parse::<i64>() {
+                h.insert(parts[0].to_string(), i);
+            }
+        }
+    }
+    Ok(h)
+}
+
+// Nested keyed
+//  KEY0 SUB_KEY0=VAL00 SUB_KEY1=VAL01...
+//  KEY1 SUB_KEY0=VAL10 SUB_KEY1=VAL11...
+pub fn nested_keyed_to_hashmap(mut file: File) -> Result<HashMap<String, HashMap<String, i64>>> {
+    let mut content = String::new();
+    file.read_to_string(&mut content)
+        .map_err(|e| Error::with_cause(ReadFailed("FIXME: read_string_from".to_string()), e))?;
+
+    let mut h = HashMap::new();
+    for line in content.lines() {
+        let parts: Vec<&str> = line.split(' ').collect();
+        if parts.is_empty() {
+            continue;
+        }
+        let mut th = HashMap::new();
+        for item in parts[1..].iter() {
+            let fields: Vec<&str> = item.split('=').collect();
+            if fields.len() == 2 {
+                if let Ok(i) = fields[1].parse::<i64>() {
+                    th.insert(fields[0].to_string(), i);
+                }
+            }
+        }
+        h.insert(parts[0].to_string(), th);
+    }
+
+    Ok(h)
+}
+
+fn read_from<T>(mut file: File) -> Result<T>
+where
+    T: FromStr,
+    <T as FromStr>::Err: 'static + Send + Sync + std::error::Error,
+{
+    let mut string = String::new();
+    match file.read_to_string(&mut string) {
+        Ok(_) => string
+            .trim()
+            .parse::<T>()
+            .map_err(|e| Error::with_cause(ParseError, e)),
+        Err(e) => Err(Error::with_cause(
+            ReadFailed("FIXME: can't get path in fn read_from".to_string()),
+            e,
+        )),
+    }
+}
+
+fn read_string_from(mut file: File) -> Result<String> {
+    let mut string = String::new();
+    match file.read_to_string(&mut string) {
+        Ok(_) => Ok(string.trim().to_string()),
+        Err(e) => Err(Error::with_cause(
+            ReadFailed("FIXME: can't get path in fn read_string_from".to_string()),
+            e,
+        )),
+    }
+}
+
+/// read and parse an u64 data
+fn read_u64_from(file: File) -> Result<u64> {
+    read_from::<u64>(file)
+}
+
+/// read and parse an i64 data
+fn read_i64_from(file: File) -> Result<i64> {
+    read_from::<i64>(file)
+}

src/fs/net_cls.rs

@@ -10,11 +10,11 @@
 use std::io::Write;
 use std::path::PathBuf;
 
-use crate::error::ErrorKind::*;
-use crate::error::*;
+use crate::fs::error::ErrorKind::*;
+use crate::fs::error::*;
 
-use crate::read_u64_from;
-use crate::{
+use crate::fs::read_u64_from;
+use crate::fs::{
     ControllIdentifier, ControllerInternal, Controllers, NetworkResources, Resources, Subsystem,
 };
 
@@ -76,10 +76,10 @@ impl<'a> From<&'a Subsystem> for &'a NetClsController {
 
 impl NetClsController {
     /// Constructs a new `NetClsController` with `root` serving as the root of the control group.
-    pub fn new(root: PathBuf) -> Self {
+    pub fn new(point: PathBuf, root: PathBuf) -> Self {
         Self {
-            base: root.clone(),
-            path: root,
+            base: root,
+            path: point,
         }
     }
 
@@ -88,8 +88,9 @@ impl NetClsController {
         self.open_path("net_cls.classid", true)
             .and_then(|mut file| {
                 let s = format!("{:#08X}", class);
-                file.write_all(s.as_ref())
-                    .map_err(|e| Error::with_cause(WriteFailed, e))
+                file.write_all(s.as_ref()).map_err(|e| {
+                    Error::with_cause(WriteFailed("net_cls.classid".to_string(), s), e)
+                })
             })
     }
 

src/fs/net_prio.rs

@@ -11,11 +11,11 @@ use std::collections::HashMap;
 use std::io::{BufRead, BufReader, Write};
 use std::path::PathBuf;
 
-use crate::error::ErrorKind::*;
-use crate::error::*;
+use crate::fs::error::ErrorKind::*;
+use crate::fs::error::*;
 
-use crate::read_u64_from;
-use crate::{
+use crate::fs::read_u64_from;
+use crate::fs::{
     ControllIdentifier, ControllerInternal, Controllers, NetworkResources, Resources, Subsystem,
 };
 
@@ -79,10 +79,10 @@ impl<'a> From<&'a Subsystem> for &'a NetPrioController {
 
 impl NetPrioController {
     /// Constructs a new `NetPrioController` with `root` serving as the root of the control group.
-    pub fn new(root: PathBuf) -> Self {
+    pub fn new(point: PathBuf, root: PathBuf) -> Self {
         Self {
-            base: root.clone(),
-            path: root,
+            base: root,
+            path: point,
         }
     }
 
@@ -94,36 +94,26 @@ impl NetPrioController {
     }
 
     /// A map of priorities for each network interface.
-    #[allow(clippy::iter_nth_zero, clippy::unnecessary_unwrap)]
     pub fn ifpriomap(&self) -> Result<HashMap<String, u64>> {
         self.open_path("net_prio.ifpriomap", false)
             .and_then(|file| {
                 let bf = BufReader::new(file);
-                bf.lines().fold(Ok(HashMap::new()), |acc, line| {
-                    if acc.is_err() {
-                        acc
-                    } else {
-                        let mut acc = acc.unwrap();
-                        let l = line.unwrap();
-                        let mut sp = l.split_whitespace();
+                bf.lines()
+                    .map(|line| {
+                        let line = line.map_err(|_| Error::new(ParseError))?;
+                        let mut parts = line.split_whitespace();
 
-                        let ifname = sp.nth(0);
-                        let ifprio = sp.nth(1);
-                        if ifname.is_none() || ifprio.is_none() {
-                            Err(Error::new(ParseError))
-                        } else {
-                            let ifname = ifname.unwrap();
-                            let ifprio = ifprio.unwrap().trim().parse();
-                            match ifprio {
-                                Err(e) => Err(Error::with_cause(ParseError, e)),
-                                Ok(_) => {
-                                    acc.insert(ifname.to_string(), ifprio.unwrap());
-                                    Ok(acc)
-                                }
-                            }
-                        }
-                    }
-                })
+                        let ifname = parts.next().ok_or(Error::new(ParseError))?;
+                        let ifprio_str = parts.next().ok_or(Error::new(ParseError))?;
+
+                        let ifprio = ifprio_str
+                            .trim()
+                            .parse()
+                            .map_err(|e| Error::with_cause(ParseError, e))?;
+
+                        Ok((ifname.to_string(), ifprio))
+                    })
+                    .collect::<Result<HashMap<String, _>>>()
             })
     }
 
@@ -132,7 +122,15 @@ impl NetPrioController {
         self.open_path("net_prio.ifpriomap", true)
             .and_then(|mut file| {
                 file.write_all(format!("{} {}", eif, prio).as_ref())
-                    .map_err(|e| Error::with_cause(WriteFailed, e))
+                    .map_err(|e| {
+                        Error::with_cause(
+                            WriteFailed(
+                                "net_prio.ifpriomap".to_string(),
+                                format!("{} {}", eif, prio),
+                            ),
+                            e,
+                        )
+                    })
             })
     }
 }

src/fs/perf_event.rs

@@ -9,9 +9,9 @@
 //!  [tools/perf/Documentation/perf-record.txt](https://raw.githubusercontent.com/torvalds/linux/master/tools/perf/Documentation/perf-record.txt)
 use std::path::PathBuf;
 
-use crate::error::*;
+use crate::fs::error::*;
 
-use crate::{ControllIdentifier, ControllerInternal, Controllers, Resources, Subsystem};
+use crate::fs::{ControllIdentifier, ControllerInternal, Controllers, Resources, Subsystem};
 
 /// A controller that allows controlling the `perf_event` subsystem of a Cgroup.
 ///
@@ -65,10 +65,10 @@ impl<'a> From<&'a Subsystem> for &'a PerfEventController {
 
 impl PerfEventController {
     /// Constructs a new `PerfEventController` with `root` serving as the root of the control group.
-    pub fn new(root: PathBuf) -> Self {
+    pub fn new(point: PathBuf, root: PathBuf) -> Self {
         Self {
-            base: root.clone(),
-            path: root,
+            base: root,
+            path: point,
         }
     }
 }

src/fs/pid.rs

@@ -11,11 +11,11 @@
 use std::io::{Read, Write};
 use std::path::PathBuf;
 
-use crate::error::ErrorKind::*;
-use crate::error::*;
+use crate::fs::error::ErrorKind::*;
+use crate::fs::error::*;
 
-use crate::read_u64_from;
-use crate::{
+use crate::fs::read_u64_from;
+use crate::fs::{
     parse_max_value, ControllIdentifier, ControllerInternal, Controllers, MaxValue, PidResources,
     Resources, Subsystem,
 };
@@ -92,10 +92,10 @@ impl<'a> From<&'a Subsystem> for &'a PidController {
 impl PidController {
     /// Constructors a new `PidController` instance, with `root` serving as the controller's root
     /// directory.
-    pub fn new(root: PathBuf, v2: bool) -> Self {
+    pub fn new(point: PathBuf, root: PathBuf, v2: bool) -> Self {
         Self {
-            base: root.clone(),
-            path: root,
+            base: root,
+            path: point,
             v2,
         }
     }
@@ -112,7 +112,7 @@ impl PidController {
                     },
                     None => Err(Error::new(ParseError)),
                 },
-                Err(e) => Err(Error::with_cause(ReadFailed, e)),
+                Err(e) => Err(Error::with_cause(ReadFailed("pids.events".to_string()), e)),
             }
         })
     }
@@ -130,7 +130,7 @@ impl PidController {
             let res = file.read_to_string(&mut string);
             match res {
                 Ok(_) => parse_max_value(&string),
-                Err(e) => Err(Error::with_cause(ReadFailed, e)),
+                Err(e) => Err(Error::with_cause(ReadFailed("pids.max".to_string()), e)),
             }
         })
     }
@@ -145,7 +145,10 @@ impl PidController {
             let string_to_write = max_pid.to_string();
             match file.write_all(string_to_write.as_ref()) {
                 Ok(_) => Ok(()),
-                Err(e) => Err(Error::with_cause(WriteFailed, e)),
+                Err(e) => Err(Error::with_cause(
+                    WriteFailed("pids.max".to_string(), format!("{:?}", max_pid)),
+                    e,
+                )),
             }
         })
     }

src/fs/rdma.rs

@@ -10,11 +10,11 @@
 use std::io::Write;
 use std::path::PathBuf;
 
-use crate::error::ErrorKind::*;
-use crate::error::*;
+use crate::fs::error::ErrorKind::*;
+use crate::fs::error::*;
 
-use crate::read_string_from;
-use crate::{ControllIdentifier, ControllerInternal, Controllers, Resources, Subsystem};
+use crate::fs::read_string_from;
+use crate::fs::{ControllIdentifier, ControllerInternal, Controllers, Resources, Subsystem};
 
 /// A controller that allows controlling the `rdma` subsystem of a Cgroup.
 ///
@@ -68,10 +68,10 @@ impl<'a> From<&'a Subsystem> for &'a RdmaController {
 
 impl RdmaController {
     /// Constructs a new `RdmaController` with `root` serving as the root of the control group.
-    pub fn new(root: PathBuf) -> Self {
+    pub fn new(point: PathBuf, root: PathBuf) -> Self {
         Self {
-            base: root.clone(),
-            path: root,
+            base: root,
+            path: point,
         }
     }
 
@@ -81,11 +81,17 @@ impl RdmaController {
             .and_then(read_string_from)
     }
 
+    /// Returns the max usage of RDMA/IB specific resources.
+    pub fn max(&self) -> Result<String> {
+        self.open_path("rdma.max", false).and_then(read_string_from)
+    }
+
     /// Set a maximum usage for each RDMA/IB resource.
     pub fn set_max(&self, max: &str) -> Result<()> {
         self.open_path("rdma.max", true).and_then(|mut file| {
-            file.write_all(max.as_ref())
-                .map_err(|e| Error::with_cause(WriteFailed, e))
+            file.write_all(max.as_ref()).map_err(|e| {
+                Error::with_cause(WriteFailed("rdma.max".to_string(), max.to_string()), e)
+            })
         })
     }
 }

src/fs/systemd.rs

@@ -7,9 +7,9 @@
 //!
 use std::path::PathBuf;
 
-use crate::error::*;
+use crate::fs::error::*;
 
-use crate::{ControllIdentifier, ControllerInternal, Controllers, Resources, Subsystem};
+use crate::fs::{ControllIdentifier, ControllerInternal, Controllers, Resources, Subsystem};
 
 /// A controller that allows controlling the `systemd` subsystem of a Cgroup.
 ///
@@ -62,10 +62,10 @@ impl<'a> From<&'a Subsystem> for &'a SystemdController {
 
 impl SystemdController {
     /// Constructs a new `SystemdController` with `root` serving as the root of the control group.
-    pub fn new(root: PathBuf, v2: bool) -> Self {
+    pub fn new(point: PathBuf, root: PathBuf, v2: bool) -> Self {
         Self {
-            base: root.clone(),
-            path: root,
+            base: root,
+            path: point,
             _v2: v2,
         }
     }

src/lib.rs

@@ -1,668 +1,37 @@
 // Copyright (c) 2018 Levente Kurusa
-// Copyright (c) 2020 Ant Group
+// Copyright (c) 2020-2025 Ant Group
 //
 // SPDX-License-Identifier: Apache-2.0 or MIT
 //
 
-#![allow(clippy::unnecessary_unwrap)]
-use log::*;
-
-use std::collections::HashMap;
-use std::fmt;
-use std::fs::{self, File};
-use std::io::{BufRead, BufReader, Read, Write};
-use std::path::{Path, PathBuf};
-use std::str::FromStr;
-
-macro_rules! update_and_test {
-    ($self: ident, $set_func:ident, $value:expr, $get_func:ident) => {
-        if let Some(v) = $value {
-            $self.$set_func(v)?;
-            if $self.$get_func()? != v {
-                return Err(Error::new(Other));
-            }
-        }
-    };
-}
-
-macro_rules! update {
-    ($self: ident, $set_func:ident, $value:expr) => {
-        if let Some(v) = $value {
-            let _ = $self.$set_func(v);
-        }
-    };
-}
-
-pub mod blkio;
-pub mod cgroup;
-pub mod cgroup_builder;
-pub mod cpu;
-pub mod cpuacct;
-pub mod cpuset;
-pub mod devices;
-pub mod error;
-pub mod events;
-pub mod freezer;
-pub mod hierarchies;
-pub mod hugetlb;
-pub mod memory;
-pub mod net_cls;
-pub mod net_prio;
-pub mod perf_event;
-pub mod pid;
-pub mod rdma;
+pub mod fs;
+#[cfg(feature = "oci")]
+pub mod manager;
+#[cfg(feature = "oci")]
+pub use manager::{FsManager, Manager, SystemdManager};
+pub mod stats;
+pub use stats::CgroupStats;
 pub mod systemd;
 
-use crate::blkio::BlkIoController;
-use crate::cpu::CpuController;
-use crate::cpuacct::CpuAcctController;
-use crate::cpuset::CpuSetController;
-use crate::devices::DevicesController;
-use crate::error::ErrorKind::*;
-use crate::error::*;
-use crate::freezer::FreezerController;
-use crate::hugetlb::HugeTlbController;
-use crate::memory::MemController;
-use crate::net_cls::NetClsController;
-use crate::net_prio::NetPrioController;
-use crate::perf_event::PerfEventController;
-use crate::pid::PidController;
-use crate::rdma::RdmaController;
-use crate::systemd::SystemdController;
+/// The maximum value for CPU shares in cgroups v1
+pub const CPU_SHARES_V1_MAX: u64 = 262144;
+/// The maximum value for CPU weight in cgroups v2
+pub const CPU_WEIGHT_V2_MAX: u64 = 10000;
 
-#[doc(inline)]
-pub use crate::cgroup::Cgroup;
-
-/// Contains all the subsystems that are available in this crate.
-#[derive(Debug, Clone)]
-pub enum Subsystem {
-    /// Controller for the `Pid` subsystem, see `PidController` for more information.
-    Pid(PidController),
-    /// Controller for the `Mem` subsystem, see `MemController` for more information.
-    Mem(MemController),
-    /// Controller for the `CpuSet subsystem, see `CpuSetController` for more information.
-    CpuSet(CpuSetController),
-    /// Controller for the `CpuAcct` subsystem, see `CpuAcctController` for more information.
-    CpuAcct(CpuAcctController),
-    /// Controller for the `Cpu` subsystem, see `CpuController` for more information.
-    Cpu(CpuController),
-    /// Controller for the `Devices` subsystem, see `DevicesController` for more information.
-    Devices(DevicesController),
-    /// Controller for the `Freezer` subsystem, see `FreezerController` for more information.
-    Freezer(FreezerController),
-    /// Controller for the `NetCls` subsystem, see `NetClsController` for more information.
-    NetCls(NetClsController),
-    /// Controller for the `BlkIo` subsystem, see `BlkIoController` for more information.
-    BlkIo(BlkIoController),
-    /// Controller for the `PerfEvent` subsystem, see `PerfEventController` for more information.
-    PerfEvent(PerfEventController),
-    /// Controller for the `NetPrio` subsystem, see `NetPrioController` for more information.
-    NetPrio(NetPrioController),
-    /// Controller for the `HugeTlb` subsystem, see `HugeTlbController` for more information.
-    HugeTlb(HugeTlbController),
-    /// Controller for the `Rdma` subsystem, see `RdmaController` for more information.
-    Rdma(RdmaController),
-    /// Controller for the `Systemd` subsystem, see `SystemdController` for more information.
-    Systemd(SystemdController),
-}
-
-#[doc(hidden)]
-#[derive(Eq, PartialEq, Debug, Clone)]
-pub enum Controllers {
-    Pids,
-    Mem,
-    CpuSet,
-    CpuAcct,
-    Cpu,
-    Devices,
-    Freezer,
-    NetCls,
-    BlkIo,
-    PerfEvent,
-    NetPrio,
-    HugeTlb,
-    Rdma,
-    Systemd,
-}
-
-impl fmt::Display for Controllers {
-    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
-        match self {
-            Controllers::Pids => write!(f, "pids"),
-            Controllers::Mem => write!(f, "memory"),
-            Controllers::CpuSet => write!(f, "cpuset"),
-            Controllers::CpuAcct => write!(f, "cpuacct"),
-            Controllers::Cpu => write!(f, "cpu"),
-            Controllers::Devices => write!(f, "devices"),
-            Controllers::Freezer => write!(f, "freezer"),
-            Controllers::NetCls => write!(f, "net_cls"),
-            Controllers::BlkIo => write!(f, "blkio"),
-            Controllers::PerfEvent => write!(f, "perf_event"),
-            Controllers::NetPrio => write!(f, "net_prio"),
-            Controllers::HugeTlb => write!(f, "hugetlb"),
-            Controllers::Rdma => write!(f, "rdma"),
-            Controllers::Systemd => write!(f, "name=systemd"),
-        }
-    }
-}
-
-mod sealed {
-    use super::*;
-
-    pub trait ControllerInternal {
-        fn apply(&self, res: &Resources) -> Result<()>;
-
-        // meta stuff
-        fn control_type(&self) -> Controllers;
-        fn get_path(&self) -> &PathBuf;
-        fn get_path_mut(&mut self) -> &mut PathBuf;
-        fn get_base(&self) -> &PathBuf;
-
-        /// Hooks running after controller crated, if have
-        fn post_create(&self) {}
-
-        fn is_v2(&self) -> bool {
-            false
-        }
-
-        fn verify_path(&self) -> Result<()> {
-            if self.get_path().starts_with(self.get_base()) {
-                Ok(())
-            } else {
-                Err(Error::new(ErrorKind::InvalidPath))
-            }
-        }
-
-        fn open_path(&self, p: &str, w: bool) -> Result<File> {
-            let mut path = self.get_path().clone();
-            path.push(p);
-
-            self.verify_path()?;
-
-            if w {
-                match File::create(&path) {
-                    Err(e) => Err(Error::with_cause(ErrorKind::WriteFailed, e)),
-                    Ok(file) => Ok(file),
-                }
-            } else {
-                match File::open(&path) {
-                    Err(e) => Err(Error::with_cause(ErrorKind::ReadFailed, e)),
-                    Ok(file) => Ok(file),
-                }
-            }
-        }
-
-        fn get_max_value(&self, f: &str) -> Result<MaxValue> {
-            self.open_path(f, false).and_then(|mut file| {
-                let mut string = String::new();
-                let res = file.read_to_string(&mut string);
-                match res {
-                    Ok(_) => parse_max_value(&string),
-                    Err(e) => Err(Error::with_cause(ReadFailed, e)),
-                }
-            })
-        }
-
-        #[doc(hidden)]
-        fn path_exists(&self, p: &str) -> bool {
-            if self.verify_path().is_err() {
-                return false;
-            }
-
-            std::path::Path::new(p).exists()
-        }
-    }
-
-    pub trait CustomizedAttribute: ControllerInternal {
-        fn set(&self, key: &str, value: &str) -> Result<()> {
-            self.open_path(key, true).and_then(|mut file| {
-                file.write_all(value.as_ref())
-                    .map_err(|e| Error::with_cause(WriteFailed, e))
-            })
-        }
-
-        fn get(&self, key: &str) -> Result<String> {
-            self.open_path(key, false).and_then(|mut file: File| {
-                let mut string = String::new();
-                match file.read_to_string(&mut string) {
-                    Ok(_) => Ok(string.trim().to_owned()),
-                    Err(e) => Err(Error::with_cause(ReadFailed, e)),
-                }
-            })
-        }
-    }
-}
-
-pub(crate) use crate::sealed::{ControllerInternal, CustomizedAttribute};
-
-/// A Controller is a subsystem attached to the control group.
-///
-/// Implementors are able to control certain aspects of a control group.
-pub trait Controller {
-    #[doc(hidden)]
-    fn control_type(&self) -> Controllers;
-
-    /// The file system path to the controller.
-    fn path(&self) -> &Path;
-
-    /// Apply a set of resources to the Controller, invoking its internal functions to pass the
-    /// kernel the information.
-    fn apply(&self, res: &Resources) -> Result<()>;
-
-    /// Create this controller
-    fn create(&self);
-
-    /// Does this controller already exist?
-    fn exists(&self) -> bool;
-
-    /// Set notify_on_release
-    fn set_notify_on_release(&self, enable: bool) -> Result<()>;
-
-    /// Set release_agent
-    fn set_release_agent(&self, path: &str) -> Result<()>;
-
-    /// Delete the controller.
-    fn delete(&self) -> Result<()>;
-
-    /// Attach a task to this controller.
-    fn add_task(&self, pid: &CgroupPid) -> Result<()>;
-
-    /// Attach a task to this controller.
-    fn add_task_by_tgid(&self, pid: &CgroupPid) -> Result<()>;
-
-    /// Get the list of tasks that this controller has.
-    fn tasks(&self) -> Vec<CgroupPid>;
-
-    fn v2(&self) -> bool;
-}
-
-impl<T> Controller for T
-where
-    T: ControllerInternal,
-{
-    fn control_type(&self) -> Controllers {
-        ControllerInternal::control_type(self)
-    }
-
-    fn path(&self) -> &Path {
-        self.get_path()
-    }
-
-    /// Apply a set of resources to the Controller, invoking its internal functions to pass the
-    /// kernel the information.
-    fn apply(&self, res: &Resources) -> Result<()> {
-        ControllerInternal::apply(self, res)
-    }
-
-    /// Create this controller
-    fn create(&self) {
-        self.verify_path()
-            .unwrap_or_else(|_| panic!("path should be valid: {:?}", self.path()));
-
-        match ::std::fs::create_dir_all(self.get_path()) {
-            Ok(_) => self.post_create(),
-            Err(e) => warn!("error create_dir: {:?} error: {:?}", self.get_path(), e),
-        }
-    }
-
-    /// Set notify_on_release
-    fn set_notify_on_release(&self, enable: bool) -> Result<()> {
-        self.open_path("notify_on_release", true)
-            .and_then(|mut file| {
-                write!(file, "{}", enable as i32)
-                    .map_err(|e| Error::with_cause(ErrorKind::WriteFailed, e))
-            })
-    }
-
-    /// Set release_agent
-    fn set_release_agent(&self, path: &str) -> Result<()> {
-        self.open_path("release_agent", true).and_then(|mut file| {
-            file.write_all(path.as_bytes())
-                .map_err(|e| Error::with_cause(ErrorKind::WriteFailed, e))
-        })
-    }
-    /// Does this controller already exist?
-    fn exists(&self) -> bool {
-        self.get_path().exists()
-    }
-
-    /// Delete the controller.
-    fn delete(&self) -> Result<()> {
-        if !self.get_path().exists() {
-            return Ok(());
-        }
-
-        // Compatible with runC for remove dir operation
-        // https://github.com/opencontainers/runc/blob/main/libcontainer/cgroups/utils.go#L272
-        //
-        // We trying to remove all paths five times with increasing delay between tries.
-        // If after all there are not removed cgroups - appropriate error will be
-        // returned.
-        let mut delay = std::time::Duration::from_millis(10);
-        let cgroup_path = self.get_path();
-        for _i in 0..4 {
-            if let Ok(()) = remove_dir(cgroup_path) {
-                return Ok(());
-            }
-            std::thread::sleep(delay);
-            delay *= 2;
-        }
-
-        remove_dir(cgroup_path)
-    }
-
-    /// Attach a task to this controller.
-    fn add_task(&self, pid: &CgroupPid) -> Result<()> {
-        let mut file = "tasks";
-        if self.is_v2() {
-            file = "cgroup.procs";
-        }
-        self.open_path(file, true).and_then(|mut file| {
-            file.write_all(pid.pid.to_string().as_ref())
-                .map_err(|e| Error::with_cause(ErrorKind::WriteFailed, e))
-        })
-    }
-
-    /// Attach a task to this controller by thread group id.
-    fn add_task_by_tgid(&self, pid: &CgroupPid) -> Result<()> {
-        self.open_path("cgroup.procs", true).and_then(|mut file| {
-            file.write_all(pid.pid.to_string().as_ref())
-                .map_err(|e| Error::with_cause(ErrorKind::WriteFailed, e))
-        })
-    }
-
-    /// Get the list of tasks that this controller has.
-    fn tasks(&self) -> Vec<CgroupPid> {
-        let mut file = "tasks";
-        if self.is_v2() {
-            file = "cgroup.procs";
-        }
-        self.open_path(file, false)
-            .map(|file| {
-                let bf = BufReader::new(file);
-                let mut v = Vec::new();
-                for line in bf.lines() {
-                    match line {
-                        Ok(line) => {
-                            let n = line.trim().parse().unwrap_or(0u64);
-                            v.push(n);
-                        }
-                        Err(_) => break,
-                    }
-                }
-                v.into_iter().map(CgroupPid::from).collect()
-            })
-            .unwrap_or_default()
-    }
-
-    fn v2(&self) -> bool {
-        self.is_v2()
-    }
-}
-
-// remove_dir aims to remove cgroup path. It does so recursively,
-// by removing any subdirectories (sub-cgroups) first.
-fn remove_dir(dir: &Path) -> Result<()> {
-    // try the fast path first.
-    if fs::remove_dir(dir).is_ok() {
-        return Ok(());
-    }
-
-    if dir.exists() && dir.is_dir() {
-        for entry in fs::read_dir(dir).map_err(|e| Error::with_cause(ReadFailed, e))? {
-            let entry = entry.map_err(|e| Error::with_cause(ReadFailed, e))?;
-            let path = entry.path();
-            if path.is_dir() {
-                remove_dir(&path)?;
-            }
-        }
-        fs::remove_dir(dir).map_err(|e| Error::with_cause(RemoveFailed, e))?;
-    }
-
-    Ok(())
-}
-
-#[doc(hidden)]
-pub trait ControllIdentifier {
-    fn controller_type() -> Controllers;
-}
-
-/// Control group hierarchy (right now, only V1 is supported, but in the future Unified will be
-/// implemented as well).
-pub trait Hierarchy: std::fmt::Debug + Send + Sync {
-    /// Returns what subsystems are supported by the hierarchy.
-    fn subsystems(&self) -> Vec<Subsystem>;
-
-    /// Returns the root directory of the hierarchy.
-    fn root(&self) -> PathBuf;
-
-    /// Return a handle to the root control group in the hierarchy.
-    fn root_control_group(&self) -> Cgroup;
-
-    fn v2(&self) -> bool;
-}
-
-/// Resource limits for the memory subsystem.
-#[derive(Debug, Clone, Eq, PartialEq, Default)]
+/// The current state of the control group
 #[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
-pub struct MemoryResources {
-    /// How much memory (in bytes) can the kernel consume.
-    pub kernel_memory_limit: Option<i64>,
-    /// Upper limit of memory usage of the control group's tasks.
-    pub memory_hard_limit: Option<i64>,
-    /// How much memory the tasks in the control group can use when the system is under memory
-    /// pressure.
-    pub memory_soft_limit: Option<i64>,
-    /// How much of the kernel's memory (in bytes) can be used for TCP-related buffers.
-    pub kernel_tcp_memory_limit: Option<i64>,
-    /// How much memory and swap together can the tasks in the control group use.
-    pub memory_swap_limit: Option<i64>,
-    /// Controls the tendency of the kernel to swap out parts of the address space of the tasks to
-    /// disk. Lower value implies less likely.
-    ///
-    /// Note, however, that a value of zero does not mean the process is never swapped out. Use the
-    /// traditional `mlock(2)` system call for that purpose.
-    pub swappiness: Option<u64>,
-    /// Customized key-value attributes
-    ///
-    /// # Usage:
-    /// ```
-    /// let resource = &mut cgroups_rs::Resources::default();
-    /// resource.memory.attrs.insert("memory.numa_balancing".to_string(), "true".to_string());
-    /// // apply here
-    /// ```
-    pub attrs: HashMap<String, String>,
-}
-
-/// Resources limits on the number of processes.
-#[derive(Debug, Clone, Eq, PartialEq, Default)]
-#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
-pub struct PidResources {
-    /// The maximum number of processes that can exist in the control group.
-    ///
-    /// Note that attaching processes to the control group will still succeed _even_ if the limit
-    /// would be violated, however forks/clones inside the control group will have with `EAGAIN` if
-    /// they would violate the limit set here.
-    pub maximum_number_of_processes: Option<MaxValue>,
-}
-
-/// Resources limits about how the tasks can use the CPU.
-#[derive(Debug, Clone, Eq, PartialEq, Default)]
-#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
-pub struct CpuResources {
-    // cpuset
-    /// A comma-separated list of CPU IDs where the task in the control group can run. Dashes
-    /// between numbers indicate ranges.
-    pub cpus: Option<String>,
-    /// Same syntax as the `cpus` field of this structure, but applies to memory nodes instead of
-    /// processors.
-    pub mems: Option<String>,
-    // cpu
-    /// Weight of how much of the total CPU time should this control group get. Note that this is
-    /// hierarchical, so this is weighted against the siblings of this control group.
-    pub shares: Option<u64>,
-    /// In one `period`, how much can the tasks run in nanoseconds.
-    pub quota: Option<i64>,
-    /// Period of time in nanoseconds.
-    pub period: Option<u64>,
-    /// This is currently a no-operation.
-    pub realtime_runtime: Option<i64>,
-    /// This is currently a no-operation.
-    pub realtime_period: Option<u64>,
-    /// Customized key-value attributes
-    /// # Usage:
-    /// ```
-    /// let resource = &mut cgroups_rs::Resources::default();
-    /// resource.cpu.attrs.insert("cpu.cfs_init_buffer_us".to_string(), "10".to_string());
-    /// // apply here
-    /// ```
-    pub attrs: HashMap<String, String>,
-}
-
-/// A device resource that can be allowed or denied access to.
-#[derive(Debug, Clone, Eq, PartialEq, Default)]
-#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
-pub struct DeviceResource {
-    /// If true, access to the device is allowed, otherwise it's denied.
-    pub allow: bool,
-    /// `'c'` for character device, `'b'` for block device; or `'a'` for all devices.
-    pub devtype: crate::devices::DeviceType,
-    /// The major number of the device.
-    pub major: i64,
-    /// The minor number of the device.
-    pub minor: i64,
-    /// Sequence of `'r'`, `'w'` or `'m'`, each denoting read, write or mknod permissions.
-    pub access: Vec<crate::devices::DevicePermissions>,
-}
-
-/// Limit the usage of devices for the control group's tasks.
-#[derive(Debug, Clone, Eq, PartialEq, Default)]
-#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
-pub struct DeviceResources {
-    /// For each device in the list, the limits in the structure are applied.
-    pub devices: Vec<DeviceResource>,
-}
-
-/// Assigned priority for a network device.
-#[derive(Debug, Clone, Eq, PartialEq, Default)]
-#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
-pub struct NetworkPriority {
-    /// The name (as visible in `ifconfig`) of the interface.
-    pub name: String,
-    /// Assigned priority.
-    pub priority: u64,
-}
-
-/// Collections of limits and tags that can be imposed on packets emitted by the tasks in the
-/// control group.
-#[derive(Debug, Clone, Eq, PartialEq, Default)]
-#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
-pub struct NetworkResources {
-    /// The networking class identifier to attach to the packets.
-    ///
-    /// This can then later be used in iptables and such to have special rules.
-    pub class_id: Option<u64>,
-    /// Priority of the egress traffic for each interface.
-    pub priorities: Vec<NetworkPriority>,
-}
-
-/// A hugepage type and its consumption limit for the control group.
-#[derive(Debug, Clone, Eq, PartialEq, Default)]
-#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
-pub struct HugePageResource {
-    /// The size of the hugepage, i.e. `2MB`, `1GB`, etc.
-    pub size: String,
-    /// The amount of bytes (of memory consumed by the tasks) that are allowed to be backed by
-    /// hugepages.
-    pub limit: u64,
-}
-
-/// Provides the ability to set consumption limit on each type of hugepages.
-#[derive(Debug, Clone, Eq, PartialEq, Default)]
-#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
-pub struct HugePageResources {
-    /// Set a limit of consumption for each hugepages type.
-    pub limits: Vec<HugePageResource>,
-}
-
-/// Weight for a particular block device.
-#[derive(Debug, Clone, Eq, PartialEq, Default)]
-#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
-pub struct BlkIoDeviceResource {
-    /// The major number of the device.
-    pub major: u64,
-    /// The minor number of the device.
-    pub minor: u64,
-    /// The weight of the device against the descendant nodes.
-    pub weight: Option<u16>,
-    /// The weight of the device against the sibling nodes.
-    pub leaf_weight: Option<u16>,
-}
-
-/// Provides the ability to throttle a device (both byte/sec, and IO op/s)
-#[derive(Debug, Clone, Eq, PartialEq, Default)]
-#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
-pub struct BlkIoDeviceThrottleResource {
-    /// The major number of the device.
-    pub major: u64,
-    /// The minor number of the device.
-    pub minor: u64,
-    /// The rate.
-    pub rate: u64,
-}
-
-/// General block I/O resource limits.
-#[derive(Debug, Clone, Eq, PartialEq, Default)]
-#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
-pub struct BlkIoResources {
-    /// The weight of the control group against descendant nodes.
-    pub weight: Option<u16>,
-    /// The weight of the control group against sibling nodes.
-    pub leaf_weight: Option<u16>,
-    /// For each device, a separate weight (both normal and leaf) can be provided.
-    pub weight_device: Vec<BlkIoDeviceResource>,
-    /// Throttled read bytes/second can be provided for each device.
-    pub throttle_read_bps_device: Vec<BlkIoDeviceThrottleResource>,
-    /// Throttled read IO operations per second can be provided for each device.
-    pub throttle_read_iops_device: Vec<BlkIoDeviceThrottleResource>,
-    /// Throttled written bytes/second can be provided for each device.
-    pub throttle_write_bps_device: Vec<BlkIoDeviceThrottleResource>,
-    /// Throttled write IO operations per second can be provided for each device.
-    pub throttle_write_iops_device: Vec<BlkIoDeviceThrottleResource>,
-
-    /// Customized key-value attributes
-    /// # Usage:
-    /// ```
-    /// let resource = &mut cgroups_rs::Resources::default();
-    /// resource.blkio.attrs.insert("io.cost.weight".to_string(), "10".to_string());
-    /// // apply here
-    /// ```
-    pub attrs: HashMap<String, String>,
-}
-
-/// The resource limits and constraints that will be set on the control group.
-#[derive(Debug, Clone, Eq, PartialEq, Default)]
-#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
-pub struct Resources {
-    /// Memory usage related limits.
-    pub memory: MemoryResources,
-    /// Process identifier related limits.
-    pub pid: PidResources,
-    /// CPU related limits.
-    pub cpu: CpuResources,
-    /// Device related limits.
-    pub devices: DeviceResources,
-    /// Network related tags and limits.
-    pub network: NetworkResources,
-    /// Hugepages consumption related limits.
-    pub hugepages: HugePageResources,
-    /// Block device I/O related limits.
-    pub blkio: BlkIoResources,
+pub enum FreezerState {
+    /// The processes in the control group are _not_ frozen.
+    Thawed,
+    /// The processes in the control group are in the processes of being frozen.
+    Freezing,
+    /// The processes in the control group are frozen.
+    Frozen,
 }
 
 /// A structure representing a `pid`. Currently implementations exist for `u64` and
 /// `std::process::Child`.
-#[derive(Debug, PartialEq, Eq, PartialOrd, Ord)]
+#[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Clone, Copy)]
 pub struct CgroupPid {
     /// The process identifier
     pub pid: u64,
@@ -674,241 +43,40 @@ impl From<u64> for CgroupPid {
     }
 }
 
-impl<'a> From<&'a std::process::Child> for CgroupPid {
+impl From<&std::process::Child> for CgroupPid {
     fn from(u: &std::process::Child) -> CgroupPid {
         CgroupPid { pid: u.id() as u64 }
     }
 }
 
-impl Subsystem {
-    fn enter(self, path: &Path) -> Self {
-        match self {
-            Subsystem::Pid(mut cont) => Subsystem::Pid({
-                cont.get_path_mut().push(path);
-                cont
-            }),
-            Subsystem::Mem(mut cont) => Subsystem::Mem({
-                cont.get_path_mut().push(path);
-                cont
-            }),
-            Subsystem::CpuSet(mut cont) => Subsystem::CpuSet({
-                cont.get_path_mut().push(path);
-                cont
-            }),
-            Subsystem::CpuAcct(mut cont) => Subsystem::CpuAcct({
-                cont.get_path_mut().push(path);
-                cont
-            }),
-            Subsystem::Cpu(mut cont) => Subsystem::Cpu({
-                cont.get_path_mut().push(path);
-                cont
-            }),
-            Subsystem::Devices(mut cont) => Subsystem::Devices({
-                cont.get_path_mut().push(path);
-                cont
-            }),
-            Subsystem::Freezer(mut cont) => Subsystem::Freezer({
-                cont.get_path_mut().push(path);
-                cont
-            }),
-            Subsystem::NetCls(mut cont) => Subsystem::NetCls({
-                cont.get_path_mut().push(path);
-                cont
-            }),
-            Subsystem::BlkIo(mut cont) => Subsystem::BlkIo({
-                cont.get_path_mut().push(path);
-                cont
-            }),
-            Subsystem::PerfEvent(mut cont) => Subsystem::PerfEvent({
-                cont.get_path_mut().push(path);
-                cont
-            }),
-            Subsystem::NetPrio(mut cont) => Subsystem::NetPrio({
-                cont.get_path_mut().push(path);
-                cont
-            }),
-            Subsystem::HugeTlb(mut cont) => Subsystem::HugeTlb({
-                cont.get_path_mut().push(path);
-                cont
-            }),
-            Subsystem::Rdma(mut cont) => Subsystem::Rdma({
-                cont.get_path_mut().push(path);
-                cont
-            }),
-            Subsystem::Systemd(mut cont) => Subsystem::Systemd({
-                cont.get_path_mut().push(path);
-                cont
-            }),
+#[cfg(test)]
+pub mod tests {
+    use std::fs;
+    use std::process::{Child, Command, Stdio};
+
+    /// Start a mock subprocess that will sleep forever
+    pub fn spawn_sleep_inf() -> Child {
+        let child = Command::new("sleep")
+            .arg("infinity")
+            .spawn()
+            .expect("Failed to start mock subprocess");
+        child
+    }
+
+    pub fn spawn_yes() -> Child {
+        let devnull = fs::File::create("/dev/null").expect("cannot open /dev/null");
+        let child = Command::new("yes")
+            .stdout(Stdio::from(devnull))
+            .spawn()
+            .expect("Failed to start mock subprocess");
+        child
+    }
+
+    pub fn systemd_version() -> Option<String> {
+        let output = Command::new("systemd").arg("--version").output().ok()?; // Return None if command execution fails
+        if !output.status.success() {
+            return None;
         }
-    }
-
-    pub fn to_controller(&self) -> &dyn Controller {
-        match self {
-            Subsystem::Pid(cont) => cont,
-            Subsystem::Mem(cont) => cont,
-            Subsystem::CpuSet(cont) => cont,
-            Subsystem::CpuAcct(cont) => cont,
-            Subsystem::Cpu(cont) => cont,
-            Subsystem::Devices(cont) => cont,
-            Subsystem::Freezer(cont) => cont,
-            Subsystem::NetCls(cont) => cont,
-            Subsystem::BlkIo(cont) => cont,
-            Subsystem::PerfEvent(cont) => cont,
-            Subsystem::NetPrio(cont) => cont,
-            Subsystem::HugeTlb(cont) => cont,
-            Subsystem::Rdma(cont) => cont,
-            Subsystem::Systemd(cont) => cont,
-        }
-    }
-
-    pub fn controller_name(&self) -> String {
-        self.to_controller().control_type().to_string()
+        Some(String::from_utf8_lossy(&output.stdout).to_string())
     }
 }
-
-/// The values for `memory.hight` or `pids.max`
-#[derive(Eq, PartialEq, Copy, Clone, Debug)]
-#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
-pub enum MaxValue {
-    /// This value is returned when the text is `"max"`.
-    Max,
-    /// When the value is a numerical value, they are returned via this enum field.
-    Value(i64),
-}
-
-impl Default for MaxValue {
-    fn default() -> Self {
-        MaxValue::Max
-    }
-}
-
-impl MaxValue {
-    #[allow(clippy::should_implement_trait, clippy::wrong_self_convention)]
-    fn to_i64(&self) -> i64 {
-        match self {
-            MaxValue::Max => -1,
-            MaxValue::Value(num) => *num,
-        }
-    }
-}
-
-impl fmt::Display for MaxValue {
-    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
-        match self {
-            MaxValue::Max => write!(f, "max"),
-            MaxValue::Value(num) => write!(f, "{}", num),
-        }
-    }
-}
-
-pub fn parse_max_value(s: &str) -> Result<MaxValue> {
-    if s.trim() == "max" {
-        return Ok(MaxValue::Max);
-    }
-    match s.trim().parse() {
-        Ok(val) => Ok(MaxValue::Value(val)),
-        Err(e) => Err(Error::with_cause(ParseError, e)),
-    }
-}
-
-// Flat keyed
-//  KEY0 VAL0\n
-//  KEY1 VAL1\n
-pub fn flat_keyed_to_vec(mut file: File) -> Result<Vec<(String, i64)>> {
-    let mut content = String::new();
-    file.read_to_string(&mut content)
-        .map_err(|e| Error::with_cause(ReadFailed, e))?;
-
-    let mut v = Vec::new();
-    for line in content.lines() {
-        let parts: Vec<&str> = line.split(' ').collect();
-        if parts.len() == 2 {
-            if let Ok(i) = parts[1].parse::<i64>() {
-                v.push((parts[0].to_string(), i));
-            }
-        }
-    }
-    Ok(v)
-}
-
-// Flat keyed
-//  KEY0 VAL0\n
-//  KEY1 VAL1\n
-pub fn flat_keyed_to_hashmap(mut file: File) -> Result<HashMap<String, i64>> {
-    let mut content = String::new();
-    file.read_to_string(&mut content)
-        .map_err(|e| Error::with_cause(ReadFailed, e))?;
-
-    let mut h = HashMap::new();
-    for line in content.lines() {
-        let parts: Vec<&str> = line.split(' ').collect();
-        if parts.len() == 2 {
-            if let Ok(i) = parts[1].parse::<i64>() {
-                h.insert(parts[0].to_string(), i);
-            }
-        }
-    }
-    Ok(h)
-}
-
-// Nested keyed
-//  KEY0 SUB_KEY0=VAL00 SUB_KEY1=VAL01...
-//  KEY1 SUB_KEY0=VAL10 SUB_KEY1=VAL11...
-pub fn nested_keyed_to_hashmap(mut file: File) -> Result<HashMap<String, HashMap<String, i64>>> {
-    let mut content = String::new();
-    file.read_to_string(&mut content)
-        .map_err(|e| Error::with_cause(ReadFailed, e))?;
-
-    let mut h = HashMap::new();
-    for line in content.lines() {
-        let parts: Vec<&str> = line.split(' ').collect();
-        if parts.is_empty() {
-            continue;
-        }
-        let mut th = HashMap::new();
-        for item in parts[1..].iter() {
-            let fields: Vec<&str> = item.split('=').collect();
-            if fields.len() == 2 {
-                if let Ok(i) = fields[1].parse::<i64>() {
-                    th.insert(fields[0].to_string(), i);
-                }
-            }
-        }
-        h.insert(parts[0].to_string(), th);
-    }
-
-    Ok(h)
-}
-
-fn read_from<T>(mut file: File) -> Result<T>
-where
-    T: FromStr,
-    <T as FromStr>::Err: 'static + Send + Sync + std::error::Error,
-{
-    let mut string = String::new();
-    match file.read_to_string(&mut string) {
-        Ok(_) => string
-            .trim()
-            .parse::<T>()
-            .map_err(|e| Error::with_cause(ParseError, e)),
-        Err(e) => Err(Error::with_cause(ReadFailed, e)),
-    }
-}
-
-fn read_string_from(mut file: File) -> Result<String> {
-    let mut string = String::new();
-    match file.read_to_string(&mut string) {
-        Ok(_) => Ok(string.trim().to_string()),
-        Err(e) => Err(Error::with_cause(ReadFailed, e)),
-    }
-}
-
-/// read and parse an u64 data
-fn read_u64_from(file: File) -> Result<u64> {
-    read_from::<u64>(file)
-}
-
-/// read and parse an i64 data
-fn read_i64_from(file: File) -> Result<i64> {
-    read_from::<i64>(file)
-}

src/manager/conv.rs

@@ -0,0 +1,113 @@
+// Copyright (c) 2025 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+use crate::manager::error::{Error, Result};
+use crate::{CPU_SHARES_V1_MAX, CPU_WEIGHT_V2_MAX};
+
+// Converts CPU shares, used by cgroup v1, to CPU weight, used by cgroup
+// v2.
+//
+// Cgroup v1 CPU shares has a range of [2^1...2^18], i.e. [2...262144],
+// and the default value is 1024.
+//
+// Cgroup v2 CPU weight has a range of [10^0...10^4], i.e. [1...10000],
+// and the default value is 100.
+pub(crate) fn cpu_shares_to_cgroup_v2(shares: u64) -> u64 {
+    if shares == 0 {
+        return 0;
+    }
+    if shares <= 2 {
+        return 1;
+    }
+    if shares >= CPU_SHARES_V1_MAX {
+        return CPU_WEIGHT_V2_MAX;
+    }
+
+    (((shares - 2) * 9999) / 262142) + 1
+}
+
+// ConvertMemorySwapToCgroupV2Value converts MemorySwap value from OCI spec
+// for use by cgroup v2 drivers. A conversion is needed since
+// Resources.MemorySwap is defined as memory+swap combined, while in cgroup
+// v2 swap is a separate value.
+pub(crate) fn memory_swap_to_cgroup_v2(memswap_limit: i64, mem_limit: i64) -> Result<i64> {
+    // For compatibility with cgroup1 controller, set swap to unlimited in
+    // case the memory is set to unlimited, and swap is not explicitly set,
+    // treating the request as "set both memory and swap to unlimited".
+    if mem_limit == -1 && memswap_limit == 0 {
+        return Ok(-1);
+    }
+
+    // -1 is "max", 0 is "unset", so treat as is
+    if memswap_limit == -1 || memswap_limit == 0 {
+        return Ok(memswap_limit);
+    }
+
+    // Unlimited memory, so treat swap as is.
+    if mem_limit == -1 {
+        return Ok(memswap_limit);
+    }
+
+    // Unset or unknown memory, can't calculate swap.
+    if mem_limit == 0 {
+        return Err(Error::InvalidLinuxResource);
+    }
+
+    // Does not make sense to subtract a negative value.
+    if mem_limit < 0 {
+        return Err(Error::InvalidLinuxResource);
+    }
+
+    // Sanity check.
+    if memswap_limit < mem_limit {
+        return Err(Error::InvalidLinuxResource);
+    }
+
+    Ok(memswap_limit - mem_limit)
+}
+
+#[cfg(test)]
+mod tests {
+    use crate::manager::conv::*;
+
+    #[test]
+    fn test_cpu_shares_to_cgroup_v2() {
+        assert_eq!(cpu_shares_to_cgroup_v2(0), 0);
+        assert_eq!(cpu_shares_to_cgroup_v2(1), 1);
+        assert_eq!(cpu_shares_to_cgroup_v2(2), 1);
+        assert_eq!(cpu_shares_to_cgroup_v2(100), 4);
+        assert_eq!(
+            cpu_shares_to_cgroup_v2(CPU_SHARES_V1_MAX),
+            CPU_WEIGHT_V2_MAX
+        );
+        assert_eq!(
+            cpu_shares_to_cgroup_v2(CPU_SHARES_V1_MAX - 1),
+            CPU_WEIGHT_V2_MAX - 1
+        );
+        assert_eq!(cpu_shares_to_cgroup_v2(u64::MAX), CPU_WEIGHT_V2_MAX);
+    }
+
+    #[test]
+    fn test_memory_swap_to_cgroup_v2() {
+        // memory no limit and swap is 0, treat it as no limit
+        assert_eq!(memory_swap_to_cgroup_v2(0, -1).unwrap(), -1);
+        // -1 is "max", 0 is "unset", so treat as is
+        assert_eq!(memory_swap_to_cgroup_v2(-1, 0).unwrap(), -1);
+        assert_eq!(memory_swap_to_cgroup_v2(0, 0).unwrap(), 0);
+
+        // Now swap cannot be 0 or -1
+
+        // Unlimited memory, so treat swap as is.
+        assert_eq!(memory_swap_to_cgroup_v2(100, -1).unwrap(), 100);
+        // Unset or unknown memory, can't calculate swap.
+        assert!(memory_swap_to_cgroup_v2(100, 0).is_err());
+        // Does not make sense to subtract a negative value.
+        assert!(memory_swap_to_cgroup_v2(100, -2).is_err());
+        // Swap + mem < mem
+        assert!(memory_swap_to_cgroup_v2(50, 100).is_err());
+        // Real swap
+        assert_eq!(memory_swap_to_cgroup_v2(200, 100).unwrap(), 100);
+    }
+}

src/manager/error.rs

@@ -0,0 +1,28 @@
+// Copyright (c) 2025 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+use crate::fs::error::Error as CgroupfsError;
+use crate::systemd::dbus::error::Error as SystemdDbusError;
+use crate::systemd::error::Error as SystemdCgroupError;
+
+pub type Result<T> = std::result::Result<T, Error>;
+
+#[derive(thiserror::Error, Debug)]
+pub enum Error {
+    #[error("invalid argument")]
+    InvalidArgument,
+
+    #[error("invalid linux resource")]
+    InvalidLinuxResource,
+
+    #[error("cgroupfs error: {0}")]
+    Cgroupfs(#[from] CgroupfsError),
+
+    #[error("systemd cgroup error: {0}")]
+    SystemdCgroup(#[from] SystemdCgroupError),
+
+    #[error("systemd dbus error: {0}")]
+    SystemdDbus(#[from] SystemdDbusError),
+}

src/manager/mod.rs

@@ -0,0 +1,113 @@
+// Copyright (c) 2025 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+mod error;
+use std::collections::HashMap;
+
+pub use error::{Error, Result};
+mod fs;
+pub use fs::FsManager;
+mod systemd;
+pub use systemd::SystemdManager;
+mod conv;
+
+use oci_spec::runtime::LinuxResources;
+
+use crate::systemd::SLICE_SUFFIX;
+use crate::{CgroupPid, CgroupStats, FreezerState};
+
+/// Check if the cgroups path is a systemd cgroup.
+pub fn is_systemd_cgroup(cgroups_path: &str) -> bool {
+    let parts: Vec<&str> = cgroups_path.split(':').collect();
+    parts.len() == 3 && parts[0].ends_with(SLICE_SUFFIX)
+}
+
+/// Manage cgroups designed for OCI containers.
+pub trait Manager: Send + Sync {
+    /// Add a process specified by its tgid.
+    fn add_proc(&mut self, tgid: CgroupPid) -> Result<()>;
+
+    /// Add a thread specified by its pid.
+    fn add_thread(&mut self, pid: CgroupPid) -> Result<()>;
+
+    /// Get the list of pids joint to the cgroups.
+    fn pids(&self) -> Result<Vec<CgroupPid>>;
+
+    /// Set the freezer cgroup to the specified state.
+    fn freeze(&self, state: FreezerState) -> Result<()>;
+
+    /// Remove the cgroups.
+    fn destroy(&mut self) -> Result<()>;
+
+    /// Set the resources to the cgroups.
+    fn set(&mut self, resources: &LinuxResources) -> Result<()>;
+
+    /// Get the cgroup path.
+    ///
+    /// # Arguments
+    ///
+    /// - `subsystem`: cgroup subsystem, for cgroup v1 the value should not
+    ///   be empty, while for cgroup v2 the only valid value is `None`.
+    fn cgroup_path(&self, subsystem: Option<&str>) -> Result<String>;
+
+    /// Enable CPUs, topdown from root in cgroup hierarchy, this would be
+    /// useful for CPU hotplug in the guest.
+    ///
+    /// The caller should update cgroup resources manually, in particular
+    /// cpuset, after this, in order to use the new CPUs (or avoid using
+    /// offline CPUs).
+    ///
+    /// # Arguments
+    ///
+    /// - `cpus`: online CPUs in the same format with `cat
+    ///   /sys/devices/system/cpu/online`, e.g. "0-3,6-7".
+    fn enable_cpus_topdown(&self, cpus: &str) -> Result<()>;
+
+    /// Get cgroup stats.
+    fn stats(&self) -> CgroupStats;
+
+    /// Get the mappings of subsystems to their relative path. The full
+    /// path would be something like "{mountpoint}/{relative_path}". The
+    /// mappings of mountpoints see "mounts()".
+    fn paths(&self) -> &HashMap<String, String>;
+
+    /// Get the mappings of subsystems to their mountpoints. The full
+    /// path would be something like "{mountpoint}/{relative_path}". The
+    /// mappings of relative paths see "paths()".
+    fn mounts(&self) -> &HashMap<String, String>;
+
+    /// Indicate whether the cgroup manager is using systemd.
+    fn systemd(&self) -> bool;
+
+    /// Indicate whether the cgroup manager is using cgroup v2.
+    fn v2(&self) -> bool;
+}
+
+#[cfg(test)]
+mod tests {
+    pub const MEMORY_512M: i64 = 512 * 1024 * 1024; // 512 MiB
+    pub const MEMORY_1G: i64 = 1024 * 1024 * 1024; // 1 GiB
+    pub const MEMORY_2G: i64 = 2 * 1024 * 1024 * 1024; // 2 GiB
+
+    #[macro_export]
+    macro_rules! skip_if_cgroups_v1 {
+        () => {
+            if !$crate::fs::hierarchies::is_cgroup2_unified_mode() {
+                eprintln!("Skipping test in cgroups v1 mode");
+                return;
+            }
+        };
+    }
+
+    #[macro_export]
+    macro_rules! skip_if_cgroups_v2 {
+        () => {
+            if $crate::fs::hierarchies::is_cgroup2_unified_mode() {
+                eprintln!("Skipping test in cgroups v2 mode");
+                return;
+            }
+        };
+    }
+}

src/manager/systemd.rs

@@ -0,0 +1,570 @@
+// Copyright (c) 2025 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+use std::collections::HashMap;
+
+use oci_spec::runtime::{LinuxCpu, LinuxMemory, LinuxPids, LinuxResources};
+use zbus::zvariant::Value as ZbusValue;
+
+use crate::manager::conv;
+use crate::manager::error::{Error, Result};
+use crate::manager::fs::{join_path, FsManager};
+use crate::systemd::props::PropertiesBuilder;
+use crate::systemd::utils::expand_slice;
+use crate::systemd::{
+    cpu, cpuset, memory, pids, Property, SystemdClient, DEFAULT_SLICE, SCOPE_SUFFIX, SLICE_SUFFIX,
+    TIMEOUT_STOP_USEC,
+};
+use crate::{CgroupPid, CgroupStats, FreezerState, Manager};
+
+/// Default kernel value for cpu quota period is 100000 us (100 ms), same
+/// for v1 [1] and v2 [2].
+///
+/// 1: https://www.kernel.org/doc/html/latest/scheduler/sched-bwc.html
+/// 2: https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html
+const DEFAULT_CPU_QUOTA_PERIOD: u64 = 100_000; // 100ms
+
+pub struct SystemdManager<'a> {
+    /// The name of slice
+    slice: String,
+    /// The name of unit
+    unit: String,
+    /// Systemd client
+    systemd_client: SystemdClient<'a>,
+    /// Cgroupfs manager
+    fs_manager: FsManager,
+}
+
+impl SystemdManager<'_> {
+    fn parse_slice_and_unit(path: &str) -> Result<(String, String)> {
+        let parts: Vec<&str> = path.split(':').collect();
+        if parts.len() != 3 {
+            return Err(Error::InvalidArgument);
+        }
+
+        let slice = if parts[0].is_empty() {
+            DEFAULT_SLICE.to_string()
+        } else {
+            parts[0].to_string()
+        };
+
+        let unit = new_unit_name(parts[1], parts[2]);
+
+        Ok((slice, unit))
+    }
+
+    /// Create a new `SystemdManager` from a cgroup path.
+    ///
+    /// # Arguments
+    ///
+    /// * `path` - A string slice that holds the cgroup path in the format
+    ///   "parent:scope_prefix:name".
+    pub fn new(path: &str) -> Result<Self> {
+        let (slice, unit) = Self::parse_slice_and_unit(path)?;
+        let props = PropertiesBuilder::default_cgroup(&slice, &unit).build();
+        let slice_base = expand_slice(&slice)?;
+
+        let fs_base = join_path(&slice_base, &unit);
+        let fs_manager = FsManager::new(&fs_base)?;
+
+        let cgroup = SystemdClient::new(&unit, props)?;
+
+        Ok(Self {
+            slice,
+            unit,
+            fs_manager,
+            systemd_client: cgroup,
+        })
+    }
+}
+
+impl SystemdManager<'_> {
+    /// Get the slice name.
+    pub fn slice(&self) -> &str {
+        &self.slice
+    }
+
+    /// Get the unit name.
+    pub fn unit(&self) -> &str {
+        &self.unit
+    }
+
+    fn set_cpuset(&self, props: &mut Vec<Property>, linux_cpu: &LinuxCpu) -> Result<()> {
+        if let Some(cpus) = linux_cpu.cpus().as_ref() {
+            let (id, value) = cpuset::cpus(cpus)?;
+            props.push((id, value.into()));
+        }
+
+        if let Some(mems) = linux_cpu.mems().as_ref() {
+            let (id, value) = cpuset::mems(mems)?;
+            props.push((id, value.into()));
+        }
+
+        Ok(())
+    }
+
+    fn set_cpu(&self, props: &mut Vec<Property>, linux_cpu: &LinuxCpu) -> Result<()> {
+        if let Some(shares) = linux_cpu.shares() {
+            let shares = if self.v2() {
+                conv::cpu_shares_to_cgroup_v2(shares)
+            } else {
+                shares
+            };
+            let (id, value) = cpu::shares(shares, self.v2())?;
+            props.push((id, value.into()));
+        }
+
+        let period = linux_cpu.period().unwrap_or(0);
+        let quota = linux_cpu.quota().unwrap_or(0);
+
+        if period != 0 {
+            let (id, value) = cpu::period(period)?;
+            props.push((id, value.into()));
+        }
+
+        if period != 0 || quota != 0 {
+            // Corresponds to USEC_INFINITY in systemd
+            let mut quota_systemd = u64::MAX;
+            let mut period = period;
+            if quota > 0 {
+                if period == 0 {
+                    period = DEFAULT_CPU_QUOTA_PERIOD;
+                }
+                // systemd converts CPUQuotaPerSecUSec (microseconds per
+                // CPU second) to CPUQuota (integer percentage of CPU)
+                // internally. This means that if a fractional percent of
+                // CPU is indicated by Resources.CpuQuota, we need to round
+                // up to the nearest 10ms (1% of a second) such that child
+                // cgroups can set the cpu.cfs_quota_us they expect.
+                quota_systemd = ((quota as u64) * s_to_us(1)) / period;
+                if quota_systemd % ms_to_us(10) != 0 {
+                    quota_systemd = (quota_systemd / ms_to_us(10) + 1) * ms_to_us(10);
+                }
+            }
+            let (id, value) = cpu::quota(quota_systemd)?;
+            props.push((id, value.into()));
+        }
+
+        Ok(())
+    }
+
+    fn set_memory(&self, props: &mut Vec<Property>, linux_memory: &LinuxMemory) -> Result<()> {
+        let v2 = self.v2();
+
+        let mem_limit = linux_memory.limit().unwrap_or(0);
+        if mem_limit != 0 {
+            let (id, value) = memory::limit(mem_limit, v2)?;
+            props.push((id, value.into()));
+        }
+
+        let reservation = linux_memory.reservation().unwrap_or(0);
+        if reservation != 0 && v2 {
+            let (id, value) = memory::low(reservation, v2)?;
+            props.push((id, value.into()));
+        }
+
+        let memswap_limit = linux_memory.swap().unwrap_or(0);
+        if memswap_limit != 0 && v2 {
+            let memswap_limit = conv::memory_swap_to_cgroup_v2(memswap_limit, mem_limit)?;
+            let (id, value) = memory::swap(memswap_limit, v2)?;
+            props.push((id, value.into()));
+        }
+
+        Ok(())
+    }
+
+    fn set_pids(&self, props: &mut Vec<Property>, linux_pids: &LinuxPids) -> Result<()> {
+        let limit = linux_pids.limit();
+        if limit == -1 || limit > 0 {
+            let (id, value) = pids::max(limit)?;
+            props.push((id, value.into()));
+        }
+
+        Ok(())
+    }
+
+    /// The systemd sends SIGTERM to processes in the unit on stop. Once a
+    /// timeout occurs, SIGKILL will be sent to the processes.
+    ///
+    /// The item could be retrieved by:
+    ///
+    /// ```bash
+    /// $ systemctl show <unit> -p TimeoutStopUSec
+    /// ```
+    pub fn set_term_timeout(&mut self, timeout_in_sec: u64) -> Result<()> {
+        let timeout_in_usec = timeout_in_sec * 1_000_000;
+        let prop = (TIMEOUT_STOP_USEC, ZbusValue::U64(timeout_in_usec));
+        self.systemd_client.set_properties(&[prop])?;
+        Ok(())
+    }
+}
+
+impl Manager for SystemdManager<'_> {
+    fn add_proc(&mut self, pid: CgroupPid) -> Result<()> {
+        if !self.systemd_client.exists() {
+            self.systemd_client.set_pid_prop(pid)?;
+            self.systemd_client.start()?;
+            // The fs_manager was created in load mode, which doesn't create
+            // the cgroups. So we create them here.
+            self.fs_manager.create_cgroups()?;
+            return Ok(());
+        }
+
+        let subcgroup = self.fs_manager.subcgroup();
+        self.systemd_client.add_process(pid, subcgroup)?;
+
+        Ok(())
+    }
+
+    /// `add_thread()` is the same as `add_proc()`, as systemd doesn't
+    /// expose an API to add a thread directly. As a result, the whole
+    /// threads belonging to one process will be added to this cgroup.
+    fn add_thread(&mut self, pid: CgroupPid) -> Result<()> {
+        self.add_proc(pid)
+    }
+
+    fn cgroup_path(&self, subsystem: Option<&str>) -> Result<String> {
+        self.fs_manager.cgroup_path(subsystem)
+    }
+
+    /// Destroy the cgroup and stop the transient unit.
+    ///
+    /// Please note that if the current manager is in the cgroup, the
+    /// manager will be killed with SIGTERM signal. If you do not intend
+    /// that, please ignore the signal and do cleanup things immediately.
+    /// Systemd will forcibly terminate the process with SIGKILL after a
+    /// while.
+    fn destroy(&mut self) -> Result<()> {
+        self.systemd_client.stop()?;
+        Ok(())
+    }
+
+    fn enable_cpus_topdown(&self, cpus: &str) -> Result<()> {
+        self.fs_manager.enable_cpus_topdown(cpus)
+    }
+
+    fn freeze(&self, state: FreezerState) -> Result<()> {
+        match state {
+            FreezerState::Thawed => self.systemd_client.thaw()?,
+            FreezerState::Frozen => self.systemd_client.freeze()?,
+            FreezerState::Freezing => return Err(Error::InvalidArgument),
+        }
+
+        Ok(())
+    }
+
+    fn pids(&self) -> Result<Vec<CgroupPid>> {
+        self.fs_manager.pids()
+    }
+
+    fn set(&mut self, resources: &LinuxResources) -> Result<()> {
+        let mut props = vec![];
+
+        if let Some(linux_cpu) = resources.cpu() {
+            self.set_cpuset(&mut props, linux_cpu)?;
+            self.set_cpu(&mut props, linux_cpu)?;
+        }
+
+        if let Some(linux_memory) = resources.memory() {
+            self.set_memory(&mut props, linux_memory)?;
+        }
+
+        if let Some(linux_pids) = resources.pids() {
+            self.set_pids(&mut props, linux_pids)?;
+        }
+
+        self.systemd_client.set_properties(&props)?;
+
+        Ok(())
+    }
+
+    fn stats(&self) -> CgroupStats {
+        self.fs_manager.stats()
+    }
+
+    fn paths(&self) -> &HashMap<String, String> {
+        self.fs_manager.paths()
+    }
+
+    fn mounts(&self) -> &HashMap<String, String> {
+        self.fs_manager.mounts()
+    }
+
+    fn systemd(&self) -> bool {
+        true
+    }
+
+    fn v2(&self) -> bool {
+        self.fs_manager.v2()
+    }
+}
+
+fn new_unit_name(scope_prefix: &str, name: &str) -> String {
+    // By default, we create a scope unless the user explicitly asks
+    // for a slice.
+    if !name.ends_with(SLICE_SUFFIX) {
+        if scope_prefix.is_empty() {
+            // {name}.scope
+            return format!("{}{}", name, SCOPE_SUFFIX);
+        }
+        // {scope_prefix}-{name}.scope
+        return format!("{}-{}{}", scope_prefix, name, SCOPE_SUFFIX);
+    }
+
+    name.to_string()
+}
+
+#[inline]
+/// Convert milliseconds to microseconds.
+fn ms_to_us(ms: u64) -> u64 {
+    ms * 1_000
+}
+
+#[inline]
+/// Convert seconds to microseconds.
+fn s_to_us(s: u64) -> u64 {
+    s * 1_000_000
+}
+
+#[cfg(test)]
+mod tests {
+    //! Tests for the `SystemdManager` implementation of the `Manager`
+    //! trait.
+    //!
+    //! Don't run tests in parallel, use `--test-threads=1`!
+    //!
+
+    use std::path::Path;
+    use std::thread::sleep;
+    use std::time::Duration;
+
+    use oci_spec::runtime::{LinuxCpuBuilder, LinuxMemoryBuilder, LinuxResourcesBuilder};
+    use rand::distributions::Alphanumeric;
+    use rand::Rng;
+
+    use crate::fs::cpu::CpuController;
+    use crate::fs::memory::MemController;
+    use crate::fs::{ControllIdentifier, Controller, Subsystem};
+    use crate::manager::systemd::*;
+    use crate::manager::tests::{MEMORY_1G, MEMORY_2G, MEMORY_512M};
+    use crate::tests::spawn_sleep_inf;
+    use crate::{skip_if_cgroups_v1, skip_if_cgroups_v2, skip_if_no_systemd};
+
+    fn new_cgroups_path() -> (String, String, String) {
+        let rand_string: String = rand::thread_rng()
+            .sample_iter(&Alphanumeric)
+            .take(5)
+            .map(char::from)
+            .collect();
+        (
+            "cgroupsrs-test.slice".to_string(),
+            "cri".to_string(),
+            format!("pod{}", rand_string),
+        )
+    }
+
+    fn new_systemd_manager<'a>() -> SystemdManager<'a> {
+        let (slice, scope_prefix, name) = new_cgroups_path();
+        SystemdManager::new(&format!("{}:{}:{}", slice, scope_prefix, name)).unwrap()
+    }
+
+    fn run_set_resources_failed(resources: LinuxResources) {
+        let mut child = spawn_sleep_inf();
+        let mut manager = new_systemd_manager();
+        manager
+            .add_proc(CgroupPid {
+                pid: child.id() as u64,
+            })
+            .unwrap();
+        assert!(manager.set(&resources).is_err());
+        manager.destroy().unwrap();
+        child.wait().unwrap();
+    }
+
+    fn run_set_resources<F>(linux_resources: LinuxResources, test_fn: F)
+    where
+        F: FnOnce(&mut SystemdManager),
+    {
+        let mut manager = new_systemd_manager();
+        let mut child = spawn_sleep_inf();
+
+        manager
+            .add_proc(CgroupPid {
+                pid: child.id() as u64,
+            })
+            .unwrap();
+        manager.set(&linux_resources).unwrap();
+
+        test_fn(&mut manager);
+
+        manager.destroy().unwrap();
+        child.wait().unwrap();
+    }
+
+    #[test]
+    fn test_new_unit_name() {
+        assert_eq!(new_unit_name("test", "unit"), "test-unit.scope");
+        assert_eq!(new_unit_name("test", "unit.slice"), "unit.slice");
+        assert_eq!(new_unit_name("", "unit"), "unit.scope");
+        assert_eq!(new_unit_name("", "unit.slice"), "unit.slice");
+        assert_eq!(new_unit_name("prefix", "unit"), "prefix-unit.scope");
+    }
+
+    #[test]
+    fn test_slice_and_unit() {
+        skip_if_no_systemd!();
+
+        let (slice, scope_prefix, name) = new_cgroups_path();
+        let manager = SystemdManager::new(&format!("{}:{}:{}", slice, scope_prefix, name)).unwrap();
+
+        assert_eq!(manager.slice(), "cgroupsrs-test.slice");
+        assert_eq!(manager.unit(), format!("{scope_prefix}-{name}.scope"));
+    }
+
+    #[test]
+    fn test_destory() {
+        skip_if_no_systemd!();
+
+        let (slice, scope_prefix, name) = new_cgroups_path();
+        let mut manager =
+            SystemdManager::new(&format!("{}:{}:{}", slice, scope_prefix, name)).unwrap();
+
+        let cgroup_path = manager.cgroup_path(Some("memory")).unwrap();
+        // Before starting the unit, no cgroup should exist.
+        assert!(!Path::new(&cgroup_path).exists());
+
+        let mut child = spawn_sleep_inf();
+        manager
+            .add_proc(CgroupPid {
+                pid: child.id() as u64,
+            })
+            .unwrap();
+
+        // Now cgroup should exist.
+        assert!(Path::new(&cgroup_path).exists());
+
+        manager.destroy().unwrap();
+
+        // This process should be killed.
+        child.wait().unwrap();
+
+        // No cgroup should exist after destroy, retry 5 times at 1-second
+        // intervals.
+        for _ in 0..5 {
+            if !Path::new(&cgroup_path).exists() {
+                break;
+            }
+            sleep(Duration::from_secs(1));
+        }
+        assert!(!Path::new(&cgroup_path).exists());
+        // Unit should be stopped.
+        assert!(!manager.systemd_client.exists());
+    }
+
+    fn controller<'a, T>(fs_manager: &'a FsManager) -> &'a T
+    where
+        &'a T: From<&'a Subsystem>,
+        T: Controller + ControllIdentifier,
+    {
+        let controller: &T = fs_manager.cgroup().controller_of().unwrap();
+
+        controller
+    }
+
+    #[test]
+    fn test_set_cpu() {
+        skip_if_no_systemd!();
+
+        // 1024 shares, every 100ms allows to use 1 CPU
+        let linux_cpu = LinuxCpuBuilder::default()
+            .shares(1024u64)
+            .quota(100000i64)
+            .period(100000u64)
+            .quota(100000i64)
+            .build()
+            .unwrap();
+
+        let linux_resources = LinuxResourcesBuilder::default()
+            .cpu(linux_cpu)
+            .build()
+            .unwrap();
+
+        run_set_resources(linux_resources, |manager| {
+            let controller: &CpuController = controller(&manager.fs_manager);
+            let shares = controller.shares().unwrap();
+            let period = controller.cfs_period().unwrap();
+            let quota = controller.cfs_quota().unwrap();
+
+            if manager.v2() {
+                assert_eq!(shares, conv::cpu_shares_to_cgroup_v2(1024));
+            } else {
+                assert_eq!(shares, 1024);
+            }
+
+            assert_eq!(period, 100000);
+            assert_eq!(quota, 100000);
+        })
+    }
+
+    #[test]
+    fn test_set_memory_v2() {
+        skip_if_no_systemd!();
+        skip_if_cgroups_v1!();
+
+        // Expected failure: swap < limit
+        let linux_memory = LinuxMemoryBuilder::default()
+            .limit(MEMORY_1G)
+            .swap(MEMORY_512M)
+            .build()
+            .unwrap();
+        let linux_resources = LinuxResourcesBuilder::default()
+            .memory(linux_memory)
+            .build()
+            .unwrap();
+        run_set_resources_failed(linux_resources);
+
+        // Expected success
+        let linux_memory = LinuxMemoryBuilder::default()
+            .limit(MEMORY_512M)
+            .swap(MEMORY_1G)
+            .reservation(MEMORY_2G)
+            .build()
+            .unwrap();
+        let linux_resources = LinuxResourcesBuilder::default()
+            .memory(linux_memory)
+            .build()
+            .unwrap();
+        run_set_resources(linux_resources, |manager| {
+            let controller: &MemController = controller(&manager.fs_manager);
+            let memory_stat = controller.memory_stat();
+            let memory_swap_stat = controller.memswap();
+
+            assert_eq!(memory_stat.limit_in_bytes, MEMORY_512M);
+            assert_eq!(memory_swap_stat.limit_in_bytes, MEMORY_512M);
+            assert_eq!(memory_stat.soft_limit_in_bytes, MEMORY_2G);
+        });
+    }
+
+    #[test]
+    fn test_set_memory_v1() {
+        skip_if_no_systemd!();
+        skip_if_cgroups_v2!();
+
+        // Expected success
+        let linux_memory = LinuxMemoryBuilder::default()
+            .limit(MEMORY_512M)
+            .build()
+            .unwrap();
+        let linux_resources = LinuxResourcesBuilder::default()
+            .memory(linux_memory)
+            .build()
+            .unwrap();
+        run_set_resources(linux_resources, |manager| {
+            let controller: &MemController = controller(&manager.fs_manager);
+            let memory_stat = controller.memory_stat();
+            assert_eq!(memory_stat.limit_in_bytes, MEMORY_512M);
+        });
+    }
+}

src/stats.rs

@@ -0,0 +1,156 @@
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2025 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+use std::collections::HashMap;
+
+#[derive(Debug, Default)]
+pub struct CgroupStats {
+    pub cpu: CpuCgroupStats,
+    pub memory: MemoryCgroupStats,
+    pub pids: PidsCgroupStats,
+    pub blkio: BlkioCgroupStats,
+    pub hugetlb: HugeTlbCgroupStats,
+}
+
+#[derive(Debug, Default)]
+pub struct CpuCgroupStats {
+    pub cpu_acct: Option<CpuAcctStats>,
+    pub cpu_throttling: Option<CpuThrottlingStats>,
+}
+
+#[derive(Debug, Default)]
+pub struct CpuAcctStats {
+    /// Usage in userspace, read from `cpuacct.stat` from the line starting
+    /// with `user`. Set 0 if no data.
+    pub user_usage: u64,
+    /// Usage in kernelspace, read from `cpuacct.stat` from the line
+    /// starting with `system`. Set 0 if no data.
+    pub system_usage: u64,
+    /// Total usage, read from `cpuacct.usage`. Set 0 if no data.
+    pub total_usage: u64,
+    /// Per-CPU usage, read from `cpuacct.usage_percpu`.
+    pub usage_percpu: Vec<u64>,
+}
+
+#[derive(Debug, Default)]
+pub struct CpuThrottlingStats {
+    /// Periods, read from `cpu.stat` from the line starting with
+    /// `nr_periods`. Set 0 if no data.
+    pub periods: u64,
+    /// Throttled periods, read from `cpu.stat` from the line starting with
+    /// `nr_throttled`. Set 0 if no data.
+    pub throttled_periods: u64,
+    /// Throttled time, read from `cpu.stat` from the line starting with
+    /// `throttled_time`. Set 0 if no data.
+    pub throttled_time: u64,
+}
+
+#[derive(Debug, Default)]
+pub struct MemoryCgroupStats {
+    pub memory: Option<MemoryStats>,
+    pub memory_swap: Option<MemoryStats>,
+    pub kernel_memory: Option<MemoryStats>,
+
+    /// Use hierarchy, read from `memory.use_hierarchy` in cgroups v1. Only
+    /// available in cgroups v1.
+    pub use_hierarchy: bool,
+
+    // The following data is read from `memory.stat`, see also
+    // `crate::fs::memory::MemoryStat::stat`.
+    pub cache: u64,
+    pub rss: u64,
+    pub rss_huge: u64,
+    pub shmem: u64,
+    pub mapped_file: u64,
+    pub dirty: u64,
+    pub writeback: u64,
+    pub swap: u64,
+    pub pgpgin: u64,
+    pub pgpgout: u64,
+    pub pgfault: u64,
+    pub pgmajfault: u64,
+    pub inactive_anon: u64,
+    pub active_anon: u64,
+    pub inactive_file: u64,
+    pub active_file: u64,
+    pub unevictable: u64,
+    pub hierarchical_memory_limit: i64,
+    pub hierarchical_memsw_limit: i64,
+    pub total_cache: u64,
+    pub total_rss: u64,
+    pub total_rss_huge: u64,
+    pub total_shmem: u64,
+    pub total_mapped_file: u64,
+    pub total_dirty: u64,
+    pub total_writeback: u64,
+    pub total_swap: u64,
+    pub total_pgpgin: u64,
+    pub total_pgpgout: u64,
+    pub total_pgfault: u64,
+    pub total_pgmajfault: u64,
+    pub total_inactive_anon: u64,
+    pub total_active_anon: u64,
+    pub total_inactive_file: u64,
+    pub total_active_file: u64,
+    pub total_unevictable: u64,
+}
+
+#[derive(Debug, Default)]
+pub struct MemoryStats {
+    /// Memory [swap] usage, read from `memory[.memsw].usage_in_bytes` in
+    /// cgroups v1 and `memory[.swap].current` in cgroups v2.
+    pub usage: u64,
+    /// Maximum memory [swap] usage observed by cgroups, read from
+    /// `memory[.memsw].max_usage_in_bytes` in cgroups v1 and
+    /// `memory[.swap].peak` in cgroups v2.
+    pub max_usage: u64,
+    /// Memory [swap] limit, read from `memory[.memsw].limit_in_bytes` in
+    /// cgroups v1 and `memory[.swap].max` in cgroups v2.
+    pub limit: i64,
+    /// Failure count, read from `memory[.memsw].failcnt`. Only available in
+    /// cgroups v1.
+    pub fail_cnt: u64,
+}
+
+#[derive(Debug, Default)]
+pub struct PidsCgroupStats {
+    /// Current number of processes in the cgroup, read from `pids.current`.
+    pub current: u64,
+    /// Maximum number of processes in the cgroup, read from `pids.limit`.
+    pub limit: i64,
+}
+
+#[derive(Debug, Default)]
+pub struct BlkioCgroupStats {
+    pub io_service_bytes_recursive: Vec<BlkioStat>,
+    pub io_serviced_recursive: Vec<BlkioStat>,
+    pub io_queued_recursive: Vec<BlkioStat>,
+    pub io_service_time_recursive: Vec<BlkioStat>,
+    pub io_wait_time_recursive: Vec<BlkioStat>,
+    pub io_merged_recursive: Vec<BlkioStat>,
+    pub io_time_recursive: Vec<BlkioStat>,
+    pub sectors_recursive: Vec<BlkioStat>,
+}
+
+#[derive(Debug, Default)]
+pub struct BlkioStat {
+    pub major: u64,
+    pub minor: u64,
+    pub op: String,
+    pub value: u64,
+}
+
+/// A structure representing the statistics of the `hugetlb` subsystem of a
+/// Cgroup. The key is the huge page size, and the value is the statistics
+/// for that size.
+pub type HugeTlbCgroupStats = HashMap<String, HugeTlbStat>;
+
+#[derive(Debug, Default)]
+pub struct HugeTlbStat {
+    pub usage: u64,
+    pub max_usage: u64,
+    pub fail_cnt: u64,
+}

src/systemd/consts.rs

@@ -0,0 +1,68 @@
+// Copyright (c) 2025 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+/// Who enum: all
+pub const WHO_ENUM_ALL: &str = "all";
+
+/// Unit mode: replace
+pub const UNIT_MODE_REPLACE: &str = "replace";
+
+/// No such unit error
+pub const NO_SUCH_UNIT: &str = "org.freedesktop.systemd1.NoSuchUnit";
+
+/// Default description for transient units.
+pub const DEFAULT_DESCRIPTION: &str = "cgroups-rs transient unit";
+
+/// Turn on CPU usage accounting for this unit.
+pub const CPU_ACCOUNTING: &str = "CPUAccounting";
+/// This setting controls the memory controller in the unified hierarchy.
+/// Added in version 208.
+pub const MEMORY_ACCOUNTING: &str = "MemoryAccounting";
+/// This setting controls the pids controller in the unified hierarchy.
+pub const TASKS_ACCOUNTING: &str = "TasksAccounting";
+/// This setting controls the io controller in the unified hierarchy.
+/// Added in version 230.
+pub const IO_ACCOUNTING: &str = "IOAccounting";
+/// This setting controls the block IO controller in the legacy hierarchy.
+/// Deprecated in version 252.
+pub const BLOCK_IO_ACCOUNTING: &str = "BlockIOAccounting";
+/// Description of the unit.
+pub const DESCRIPTION: &str = "Description";
+/// PIDs
+pub const PIDS: &str = "PIDs";
+/// Default dependencies for this unit.
+pub const DEFAULT_DEPENDENCIES: &str = "DefaultDependencies";
+/// Wants, expressing a weak dependency on other units.
+pub const WANTS: &str = "Wants";
+/// Slice, used to assign a unit to a specific slice.
+pub const SLICE: &str = "Slice";
+/// Turns on delegation of further resource control partitioning to
+/// processes of the unit.
+pub const DELEGATE: &str = "Delegate";
+/// Timeout for stopping the unit in microseconds.
+pub const TIMEOUT_STOP_USEC: &str = "TimeoutStopUSec";
+
+/// CPU shares in the legacy hierarchy.
+pub const CPU_SHARES: &str = "CPUShares";
+/// CPU shares in the unified hierarchy.
+pub const CPU_WEIGHT: &str = "CPUWeight";
+/// CPU quota period us.
+pub const CPU_QUOTA_PERIOD_US: &str = "CPUQuotaPeriodUSec";
+/// CPU quota us
+pub const CPU_QUOTA_PER_SEC_US: &str = "CPUQuotaPerSecUSec";
+/// Allowed CPUs
+pub const ALLOWED_CPUS: &str = "AllowedCPUs";
+/// Allowed memory nodes
+pub const ALLOWED_MEMORY_NODES: &str = "AllowedMemoryNodes";
+/// Memory limit in the legacy hierarchy.
+pub const MEMORY_LIMIT: &str = "MemoryLimit";
+/// Memory limit in the unified hierarchy.
+pub const MEMORY_MAX: &str = "MemoryMax";
+/// Memory low
+pub const MEMORY_LOW: &str = "MemoryLow";
+/// Memory swap max
+pub const MEMORY_SWAP_MAX: &str = "MemorySwapMax";
+/// Tasks max
+pub const TASKS_MAX: &str = "TasksMax";

src/systemd/cpu.rs

@@ -0,0 +1,29 @@
+// Copyright (c) 2025 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+use crate::systemd::error::Result;
+use crate::systemd::{CPU_QUOTA_PERIOD_US, CPU_QUOTA_PER_SEC_US, CPU_SHARES, CPU_WEIGHT};
+
+/// Returns the property for CPU shares.
+///
+/// Please note that if the shares is obtained from OCI runtime spec, it
+/// MUST be converted, see [1] and `convert_shares_to_v2()`.
+///
+/// 1: https://github.com/containers/crun/blob/main/crun.1.md#cgroup-v2
+pub fn shares(shares: u64, v2: bool) -> Result<(&'static str, u64)> {
+    let id = if v2 { CPU_WEIGHT } else { CPU_SHARES };
+
+    Ok((id, shares))
+}
+
+/// Returns the property for CPU period.
+pub fn period(period: u64) -> Result<(&'static str, u64)> {
+    Ok((CPU_QUOTA_PERIOD_US, period))
+}
+
+/// Return the property for CPU quota.
+pub fn quota(quota: u64) -> Result<(&'static str, u64)> {
+    Ok((CPU_QUOTA_PER_SEC_US, quota))
+}

src/systemd/cpuset.rs

@@ -0,0 +1,96 @@
+// Copyright (c) 2025 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+use bit_vec::BitVec;
+
+use crate::systemd::error::{Error, Result};
+use crate::systemd::{ALLOWED_CPUS, ALLOWED_MEMORY_NODES};
+
+const BYTE_IN_BITS: usize = 8;
+
+/// Returns the property for cpuset CPUs.
+pub fn cpus(cpus: &str) -> Result<(&'static str, Vec<u8>)> {
+    let mask = convert_list_to_mask(cpus)?;
+    Ok((ALLOWED_CPUS, mask))
+}
+
+/// Returns the property for cpuset memory nodes.
+pub fn mems(mems: &str) -> Result<(&'static str, Vec<u8>)> {
+    let mask = convert_list_to_mask(mems)?;
+    Ok((ALLOWED_MEMORY_NODES, mask))
+}
+
+/// Convert cpuset cpus/mems from the string in comma-separated list format
+/// to bitmask restored in `Vec<u8>`, see [1].
+///
+/// 1: https://man7.org/linux/man-pages/man7/cpuset.7.html
+///
+/// # Arguments
+///
+/// * `list` - A string slice that holds the list of CPUs in the format
+///   "0-3,5,7".
+fn convert_list_to_mask(list: &str) -> Result<Vec<u8>> {
+    let mut bit_vec = BitVec::from_elem(8, false);
+
+    let local_idx =
+        |index: usize| -> usize { index / BYTE_IN_BITS * BYTE_IN_BITS + 7 - index % BYTE_IN_BITS };
+
+    for part1 in list.split(',') {
+        let range: Vec<&str> = part1.split('-').collect();
+        match range.len() {
+            // x-
+            1 => {
+                let left: usize = range[0].parse().map_err(|_| Error::InvalidArgument)?;
+
+                while left >= bit_vec.len() {
+                    bit_vec.grow(BYTE_IN_BITS, false);
+                }
+                bit_vec.set(local_idx(left), true);
+            }
+            // x-y
+            2 => {
+                let left: usize = range[0].parse().map_err(|_| Error::InvalidArgument)?;
+                let right: usize = range[1].parse().map_err(|_| Error::InvalidArgument)?;
+
+                while right >= bit_vec.len() {
+                    bit_vec.grow(BYTE_IN_BITS, false);
+                }
+
+                for index in left..=right {
+                    bit_vec.set(local_idx(index), true);
+                }
+            }
+            _ => {
+                return Err(Error::InvalidArgument);
+            }
+        }
+    }
+
+    let mut mask = bit_vec.to_bytes();
+    mask.reverse();
+
+    Ok(mask)
+}
+
+#[cfg(test)]
+mod tests {
+    use crate::systemd::cpuset::convert_list_to_mask;
+
+    #[test]
+    fn test_convert_list_to_mask() {
+        let mask = convert_list_to_mask("2-4").unwrap();
+        assert_eq!(vec![0b00011100_u8], mask);
+
+        let mask = convert_list_to_mask("1,7").unwrap();
+        assert_eq!(vec![0b10000010_u8], mask);
+
+        let mask = convert_list_to_mask("0-4,9").unwrap();
+        assert_eq!(vec![0b00000010_u8, 0b00011111_u8], mask);
+
+        assert!(convert_list_to_mask("1-3-4").is_err());
+
+        assert!(convert_list_to_mask("1-3,,").is_err());
+    }
+}

src/systemd/dbus/README.md

@@ -0,0 +1,17 @@
+# Systemd Dbus
+
+How to generate `xxx_proxy.rs` files
+
+```shell
+# install zbus-xmlgen if not
+$ cargo install zbus-xmlgen
+# generate interface in XML format
+$ busctl introspect --xml-interface \
+    org.freedesktop.systemd1 \
+    /org/freedesktop/systemd1 \
+    org.freedesktop.systemd1.Manager > /tmp/systemd1-manager.xml
+# generate Rust code from XML
+$ zbus-xmlgen file /tmp/systemd1-manager.xml \
+    --output src/systemd/dbus/systemd_manager_proxy.rs
+$ rm -rf /tmp/systemd1-manager.xml
+```

src/systemd/dbus/client.rs

@@ -0,0 +1,544 @@
+// Copyright 2021-2023 Kata Contributors
+// Copyright (c) 2025 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+use zbus::zvariant::Value;
+use zbus::{Error as ZbusError, Result as ZbusResult};
+
+use crate::systemd::dbus::error::{Error, Result};
+use crate::systemd::dbus::proxy::systemd_manager_proxy;
+use crate::systemd::{Property, NO_SUCH_UNIT, PIDS, UNIT_MODE_REPLACE};
+use crate::CgroupPid;
+
+pub struct SystemdClient<'a> {
+    /// The name of the systemd unit (slice or scope)
+    unit: String,
+    props: Vec<Property<'a>>,
+}
+
+impl<'a> SystemdClient<'a> {
+    pub fn new(unit: &str, props: Vec<Property<'a>>) -> Result<Self> {
+        Ok(Self {
+            unit: unit.to_string(),
+            props,
+        })
+    }
+}
+
+impl SystemdClient<'_> {
+    /// Set the pid to the PIDs property of the unit.
+    ///
+    /// Append a process ID to the PIDs property of the unit. If not
+    /// exists, one property will be created.
+    pub fn set_pid_prop(&mut self, pid: CgroupPid) -> Result<()> {
+        if self.exists() {
+            return Ok(());
+        }
+
+        for prop in self.props.iter_mut() {
+            if prop.0 == PIDS {
+                // If PIDS is already set, we append the new pid to the existing list.
+                if let Value::Array(arr) = &mut prop.1 {
+                    arr.append(pid.pid.into())
+                        .map_err(|_| Error::InvalidProperties)?;
+                    return Ok(());
+                }
+                // Invalid type of PIDs
+                return Err(Error::InvalidProperties);
+            }
+        }
+        // If PIDS is not set, we create a new property.
+        self.props
+            .push((PIDS, Value::Array(vec![pid.pid as u32].into())));
+        Ok(())
+    }
+
+    /// Start a slice or a scope unit controlled and supervised by systemd.
+    ///
+    /// For more information, see:
+    /// https://www.freedesktop.org/software/systemd/man/latest/systemd.unit.html
+    /// https://www.freedesktop.org/software/systemd/man/latest/systemd.slice.html
+    /// https://www.freedesktop.org/software/systemd/man/latest/systemd.scope.html
+    pub fn start(&self) -> Result<()> {
+        // PIDs property must be present
+        if !self.props.iter().any(|(k, _)| k == &PIDS) {
+            return Err(Error::InvalidProperties);
+        }
+
+        let sys_proxy = systemd_manager_proxy()?;
+
+        let props_borrowed: Vec<(&str, &zbus::zvariant::Value)> =
+            self.props.iter().map(|(k, v)| (*k, v)).collect();
+        let props_borrowed: Vec<&(&str, &Value)> = props_borrowed.iter().collect();
+
+        sys_proxy.start_transient_unit(&self.unit, UNIT_MODE_REPLACE, &props_borrowed, &[])?;
+
+        Ok(())
+    }
+
+    /// Stop the current transient unit, the processes will be killed on
+    /// unit stop, see [1].
+    ///
+    /// 1. https://www.freedesktop.org/software/systemd/man/latest/systemd.kill.html#KillMode=
+    pub fn stop(&self) -> Result<()> {
+        let sys_proxy = systemd_manager_proxy()?;
+
+        let ret = sys_proxy.stop_unit(&self.unit, UNIT_MODE_REPLACE);
+        ignore_no_such_unit(ret)?;
+
+        // If we stop the unit and it still exists, it may be in a failed
+        // state, so we will try to reset it.
+        if self.exists() {
+            let ret = sys_proxy.reset_failed_unit(&self.unit);
+            ignore_no_such_unit(ret)?;
+        }
+
+        Ok(())
+    }
+
+    /// Set properties for the unit through dbus `SetUnitProperties`.
+    pub fn set_properties(&mut self, properties: &[Property<'static>]) -> Result<()> {
+        for prop in properties {
+            let new = prop.1.try_clone().map_err(|_| Error::InvalidProperties)?;
+            // Try to update the value first, if fails, append it.
+            if let Some(existing) = self.props.iter_mut().find(|p| p.0 == prop.0) {
+                existing.1 = new;
+            } else {
+                self.props.push((prop.0, new));
+            }
+        }
+
+        // The unit must exist before setting properties.
+        if !self.exists() {
+            return Ok(());
+        }
+
+        let sys_proxy = systemd_manager_proxy()?;
+
+        let props_borrowed: Vec<(&str, &Value)> = properties.iter().map(|(k, v)| (*k, v)).collect();
+        let props_borrowed: Vec<&(&str, &Value)> = props_borrowed.iter().collect();
+
+        sys_proxy.set_unit_properties(&self.unit, true, &props_borrowed)?;
+
+        Ok(())
+    }
+
+    /// Freeze the unit through dbus `FreezeUnit`.
+    pub fn freeze(&self) -> Result<()> {
+        let sys_proxy = systemd_manager_proxy()?;
+
+        sys_proxy.freeze_unit(&self.unit)?;
+
+        Ok(())
+    }
+
+    /// Thaw the frozen unit through dbus `ThawUnit`.
+    pub fn thaw(&self) -> Result<()> {
+        let sys_proxy = systemd_manager_proxy()?;
+
+        sys_proxy.thaw_unit(&self.unit)?;
+
+        Ok(())
+    }
+
+    /// Check if the unit exists.
+    pub fn exists(&self) -> bool {
+        let sys_proxy = match systemd_manager_proxy() {
+            Ok(proxy) => proxy,
+            _ => return false,
+        };
+
+        sys_proxy
+            .get_unit(&self.unit)
+            .map(|_| true)
+            .unwrap_or_default()
+    }
+
+    /// Add a process (tgid) to the unit through dbus
+    /// `AttachProcessesToUnit`.
+    pub fn add_process(&self, pid: CgroupPid, subcgroup: &str) -> Result<()> {
+        let sys_proxy = systemd_manager_proxy()?;
+
+        sys_proxy.attach_processes_to_unit(&self.unit, subcgroup, &[pid.pid as u32])?;
+
+        Ok(())
+    }
+}
+
+fn ignore_no_such_unit<T>(result: ZbusResult<T>) -> ZbusResult<bool> {
+    if let Err(ZbusError::MethodError(err_name, _, _)) = &result {
+        if err_name.as_str() == NO_SUCH_UNIT {
+            return Ok(true);
+        }
+    }
+    result.map(|_| false)
+}
+
+#[cfg(test)]
+pub mod tests {
+    //! Unit tests for the SystemdClient
+    //!
+    //! Not sure why the tests are going to fail if we run them in
+    //! parallel. Everything goes smoothly in serial.
+    //!
+    //! $ cargo test --package cgroups-rs --lib \
+    //!   -- systemd::dbus::client::tests \
+    //!   --show-output --test-threads=1
+
+    use std::fs;
+    use std::path::Path;
+    use std::process::Command;
+    use std::thread::sleep;
+    use std::time::Duration;
+
+    use rand::distributions::Alphanumeric;
+    use rand::Rng;
+
+    use crate::fs::hierarchies;
+    use crate::systemd::dbus::client::*;
+    use crate::systemd::props::PropertiesBuilder;
+    use crate::systemd::utils::expand_slice;
+    use crate::systemd::{DEFAULT_DESCRIPTION, DESCRIPTION, PIDS};
+    use crate::tests::{spawn_sleep_inf, spawn_yes};
+
+    const TEST_SLICE: &str = "cgroupsrs-test.slice";
+
+    fn test_unit() -> String {
+        let rand_string: String = rand::thread_rng()
+            .sample_iter(&Alphanumeric)
+            .take(5)
+            .map(char::from)
+            .collect();
+        format!("cri-pod{}.scope", rand_string)
+    }
+
+    #[macro_export]
+    macro_rules! skip_if_no_systemd {
+        () => {
+            if $crate::tests::systemd_version().is_none() {
+                eprintln!("Test skipped, no systemd?");
+                return;
+            }
+        };
+    }
+
+    fn systemd_show(unit: &str) -> String {
+        let output = Command::new("systemctl")
+            .arg("show")
+            .arg(unit)
+            .output()
+            .expect("Failed to execute systemctl show command");
+        String::from_utf8_lossy(&output.stdout).to_string()
+    }
+
+    fn start_default_cgroup(pid: CgroupPid, unit: &'_ str) -> SystemdClient<'_> {
+        let mut props = PropertiesBuilder::default_cgroup(TEST_SLICE, unit).build();
+        props.push((PIDS, Value::Array(vec![pid.pid as u32].into())));
+        let cgroup = SystemdClient::new(unit, props).unwrap();
+        // Stop the unit if it exists.
+        cgroup.stop().unwrap();
+
+        // Write the current process to the cgroup.
+        cgroup.start().unwrap();
+        cgroup.add_process(pid, "/").unwrap();
+        cgroup
+    }
+
+    fn stop_cgroup(cgroup: &SystemdClient) {
+        cgroup.stop().unwrap();
+    }
+
+    #[test]
+    fn test_start() {
+        skip_if_no_systemd!();
+
+        let v2 = hierarchies::is_cgroup2_unified_mode();
+        let unit = test_unit();
+        let mut child = spawn_sleep_inf();
+        let cgroup = start_default_cgroup(CgroupPid::from(child.id() as u64), &unit);
+
+        let base = expand_slice(TEST_SLICE).unwrap();
+
+        // Check if the cgroup exists in the filesystem
+        let full_base = if v2 {
+            format!("/sys/fs/cgroup/{}", base)
+        } else {
+            format!("/sys/fs/cgroup/memory/{}", base)
+        };
+        assert!(
+            Path::new(&full_base).exists(),
+            "Cgroup base path does not exist: {}",
+            full_base
+        );
+
+        // PIDs
+        let cgroup_procs_path = format!("{}/{}/cgroup.procs", full_base, &unit);
+        for i in 0..5 {
+            let content = fs::read_to_string(&cgroup_procs_path);
+            if let Ok(content) = &content {
+                if content.contains(&child.id().to_string()) {
+                    break;
+                }
+            }
+            // Retry attempts exhausted, resulting in failure
+            if i == 4 {
+                let content = content.as_ref().unwrap();
+                assert!(
+                    content.contains(&child.id().to_string()),
+                    "Cgroup procs does not contain the child process ID"
+                );
+            }
+            // Wait 500ms before next retrying
+            sleep(Duration::from_millis(500));
+        }
+
+        // Check the unit from "systemctl show <unit>"
+        let output = systemd_show(&cgroup.unit);
+
+        // Slice
+        assert!(
+            output
+                .lines()
+                .any(|line| line == format!("Slice={}", TEST_SLICE)),
+            "Slice not found"
+        );
+        // Delegate
+        assert!(
+            output.lines().any(|line| line == "Delegate=yes"),
+            "Delegate not set"
+        );
+        // DelegateControllers
+        // controllers: cpu cpuacct cpuset io blkio memory devices pids
+        let controllers = output
+            .lines()
+            .find(|line| line.starts_with("DelegateControllers="))
+            .map(|line| line.trim_start_matches("DelegateControllers="))
+            .unwrap();
+        let controllers = controllers.split(' ').collect::<Vec<&str>>();
+        assert!(
+            controllers.contains(&"cpu"),
+            "DelegateControllers cpu not set"
+        );
+        assert!(
+            controllers.contains(&"cpuset"),
+            "DelegateControllers cpuset not set"
+        );
+        if v2 {
+            assert!(
+                controllers.contains(&"io"),
+                "DelegateControllers io not set"
+            );
+        } else {
+            assert!(
+                controllers.contains(&"blkio"),
+                "DelegateControllers blkio not set"
+            );
+        }
+        assert!(
+            controllers.contains(&"memory"),
+            "DelegateControllers memory not set"
+        );
+        assert!(
+            controllers.contains(&"pids"),
+            "DelegateControllers pids not set"
+        );
+
+        // CPUAccounting
+        assert!(
+            output.lines().any(|line| line == "CPUAccounting=yes"),
+            "CPUAccounting not set"
+        );
+        // IOAccounting for v2, and BlockIOAccounting for v1
+        if v2 {
+            assert!(
+                output.lines().any(|line| line == "IOAccounting=yes"),
+                "IOAccounting not set"
+            );
+        } else {
+            assert!(
+                output.lines().any(|line| line == "BlockIOAccounting=yes"),
+                "BlockIOAccounting not set"
+            );
+        }
+        // MemoryAccounting
+        assert!(
+            output.lines().any(|line| line == "MemoryAccounting=yes"),
+            "MemoryAccounting not set"
+        );
+        // TasksAccounting
+        assert!(
+            output.lines().any(|line| line == "TasksAccounting=yes"),
+            "TasksAccounting not set"
+        );
+        // ActiveState
+        assert!(
+            output.lines().any(|line| line == "ActiveState=active"),
+            "Unit is not active"
+        );
+
+        stop_cgroup(&cgroup);
+        child.wait().unwrap();
+    }
+
+    #[test]
+    fn test_stop() {
+        skip_if_no_systemd!();
+
+        let unit = test_unit();
+        let mut child = spawn_sleep_inf();
+        let cgroup = start_default_cgroup(CgroupPid::from(child.id() as u64), &unit);
+
+        // Check ActiveState: expected to be "active"
+        let output = systemd_show(&cgroup.unit);
+        assert!(
+            output.lines().any(|line| line == "ActiveState=active"),
+            "Unit is not active"
+        );
+
+        stop_cgroup(&cgroup);
+
+        // Check ActiveState: expected to be "inactive"
+        let output = systemd_show(&cgroup.unit);
+        assert!(
+            output.lines().any(|line| line == "ActiveState=inactive"),
+            "Unit is not inactive"
+        );
+
+        child.wait().unwrap();
+    }
+
+    #[test]
+    fn test_set_properties() {
+        skip_if_no_systemd!();
+
+        let unit = test_unit();
+        let mut child = spawn_sleep_inf();
+        let mut cgroup = start_default_cgroup(CgroupPid::from(child.id() as u64), &unit);
+
+        let output = systemd_show(&cgroup.unit);
+        assert!(
+            output.lines().any(|line| line
+                == format!(
+                    "Description={} {}:{}",
+                    DEFAULT_DESCRIPTION, TEST_SLICE, unit
+                )),
+            "Initial description not set correctly"
+        );
+
+        let properties = [(
+            DESCRIPTION,
+            Value::Str("kata-container1 description".into()),
+        )];
+        cgroup.set_properties(&properties).unwrap();
+        assert!(cgroup.props.iter().any(|(k, v)| {
+            k == &DESCRIPTION && v == &Value::Str("kata-container1 description".into())
+        }));
+
+        let output = systemd_show(&cgroup.unit);
+        assert!(
+            output
+                .lines()
+                .any(|line| line == "Description=kata-container1 description"),
+            "Updated description not set correctly"
+        );
+
+        stop_cgroup(&cgroup);
+        child.wait().unwrap();
+    }
+
+    #[test]
+    fn test_freeze_and_thaw() {
+        skip_if_no_systemd!();
+
+        let unit = test_unit();
+        let mut child = spawn_yes();
+        let cgroup = start_default_cgroup(CgroupPid::from(child.id() as u64), &unit);
+
+        // Freeze the unit
+        cgroup.freeze().unwrap();
+
+        let pid = child.id() as u64;
+
+        let stat_path = format!("/proc/{}/stat", pid);
+        let content = fs::read_to_string(&stat_path).unwrap();
+        // The process state is the third field, e.g.:
+        // 1234 (bash) S 1233 ...
+        //             ^
+        let mut content_iter = content.split_whitespace();
+        assert_eq!(
+            content_iter.nth(2).unwrap(),
+            "S",
+            "Process should be in 'S' (sleeping) state after freezing"
+        );
+
+        // Thaw the unit
+        cgroup.thaw().unwrap();
+
+        // No more S now
+        let content = fs::read_to_string(&stat_path).unwrap();
+        let mut content_iter = content.split_whitespace();
+        assert_ne!(
+            content_iter.nth(2).unwrap(),
+            "S",
+            "Process should not be in 'S' (sleeping) state after thawing"
+        );
+
+        stop_cgroup(&cgroup);
+        child.wait().unwrap();
+    }
+
+    #[test]
+    fn test_exists() {
+        skip_if_no_systemd!();
+
+        let unit = test_unit();
+        let mut child = spawn_sleep_inf();
+        let cgroup = start_default_cgroup(CgroupPid::from(child.id() as u64), &unit);
+
+        assert!(cgroup.exists(), "Cgroup should exist after starting");
+
+        stop_cgroup(&cgroup);
+        child.wait().unwrap();
+    }
+
+    #[test]
+    fn test_add_process() {
+        skip_if_no_systemd!();
+
+        let unit = test_unit();
+        let mut child = spawn_sleep_inf();
+        let cgroup = start_default_cgroup(CgroupPid::from(child.id() as u64), &unit);
+
+        let mut child1 = spawn_sleep_inf();
+        let pid1 = CgroupPid::from(child1.id() as u64);
+        cgroup.add_process(pid1, "/").unwrap();
+
+        let cgroup_procs_path = format!(
+            "/sys/fs/cgroup/{}/{}/cgroup.procs",
+            expand_slice(TEST_SLICE).unwrap(),
+            unit
+        );
+        for i in 0..5 {
+            let content = fs::read_to_string(&cgroup_procs_path);
+            if let Ok(content) = content {
+                assert!(
+                    content.contains(&child1.id().to_string()),
+                    "Cgroup procs does not contain the child1 process ID"
+                );
+                break;
+            }
+            // Retry attempts exhausted, resulting in failure
+            if i == 4 {
+                content.unwrap();
+            }
+            // Wait 500ms before next retrying
+            sleep(Duration::from_millis(500));
+        }
+
+        stop_cgroup(&cgroup);
+        child.wait().unwrap();
+        child1.wait().unwrap();
+    }
+}

src/systemd/dbus/error.rs

@@ -0,0 +1,15 @@
+// Copyright (c) 2025 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0
+//
+
+pub type Result<T> = std::result::Result<T, Error>;
+
+#[derive(Debug, thiserror::Error)]
+pub enum Error {
+    #[error("invalid properties")]
+    InvalidProperties,
+
+    #[error("dbus error: {0}")]
+    Dbus(#[from] zbus::Error),
+}

src/systemd/dbus/mod.rs

@@ -0,0 +1,18 @@
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2020-2025 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+//! Systemd D-Bus interface for managing cgroups and units.
+//!
+//! References:
+//! https://www.freedesktop.org/software/systemd/man/latest/org.freedesktop.systemd1.html
+//! https://www.freedesktop.org/software/systemd/man/latest/systemd.service.html
+//! https://www.freedesktop.org/software/systemd/man/latest/systemd.resource-control.html
+
+mod client;
+pub mod error;
+mod systemd_manager_proxy;
+pub use client::SystemdClient;
+mod proxy;

src/systemd/dbus/proxy.rs

@@ -0,0 +1,16 @@
+// Copyright (c) 2025 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+use zbus::blocking::Connection;
+use zbus::Result;
+
+use crate::systemd::dbus::systemd_manager_proxy::ManagerProxyBlocking as SystemManager;
+
+pub(crate) fn systemd_manager_proxy<'a>() -> Result<SystemManager<'a>> {
+    let connection = Connection::system()?;
+    let proxy = SystemManager::new(&connection)?;
+
+    Ok(proxy)
+}

src/systemd/error.rs

@@ -0,0 +1,15 @@
+// Copyright (c) 2025 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+pub type Result<T> = std::result::Result<T, Error>;
+
+#[derive(Debug, thiserror::Error)]
+pub enum Error {
+    #[error("invalid argument")]
+    InvalidArgument,
+
+    #[error("resource not supported by cgroups v1")]
+    CgroupsV1NotSupported,
+}

src/systemd/memory.rs

@@ -0,0 +1,32 @@
+// Copyright (c) 2025 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+use crate::systemd::error::{Error, Result};
+use crate::systemd::{MEMORY_LIMIT, MEMORY_LOW, MEMORY_MAX, MEMORY_SWAP_MAX};
+
+/// Returns the property for memory limit.
+pub fn limit(limit: i64, v2: bool) -> Result<(&'static str, u64)> {
+    let id = if v2 { MEMORY_MAX } else { MEMORY_LIMIT };
+
+    Ok((id, limit as u64))
+}
+
+/// Returns the property for memory limit.
+pub fn low(low: i64, v2: bool) -> Result<(&'static str, u64)> {
+    if !v2 {
+        return Err(Error::CgroupsV1NotSupported);
+    }
+
+    Ok((MEMORY_LOW, low as u64))
+}
+
+/// Returns the property for memory swap.
+pub fn swap(swap: i64, v2: bool) -> Result<(&'static str, u64)> {
+    if !v2 {
+        return Err(Error::CgroupsV1NotSupported);
+    }
+
+    Ok((MEMORY_SWAP_MAX, swap as u64))
+}

src/systemd/mod.rs

@@ -0,0 +1,22 @@
+// Copyright (c) 2025 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+pub mod cpu;
+pub mod cpuset;
+pub mod dbus;
+pub use dbus::SystemdClient;
+mod consts;
+pub use consts::*;
+pub mod error;
+pub mod memory;
+pub mod pids;
+pub mod props;
+pub use props::Property;
+pub mod utils;
+
+pub const DEFAULT_SLICE: &str = "system.slice";
+
+pub const SLICE_SUFFIX: &str = ".slice";
+pub const SCOPE_SUFFIX: &str = ".scope";

src/systemd/pids.rs

@@ -0,0 +1,11 @@
+// Copyright (c) 2025 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+use crate::systemd::error::Result;
+use crate::systemd::TASKS_MAX;
+
+pub fn max(max: i64) -> Result<(&'static str, u64)> {
+    Ok((TASKS_MAX, max as u64))
+}

src/systemd/props.rs

@@ -0,0 +1,170 @@
+// Copyright (c) 2025 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+use zbus::zvariant::Value as ZbusValue;
+
+use crate::fs::hierarchies;
+use crate::systemd::utils::is_slice_unit;
+use crate::systemd::{
+    BLOCK_IO_ACCOUNTING, CPU_ACCOUNTING, DEFAULT_DEPENDENCIES, DEFAULT_DESCRIPTION, DELEGATE,
+    DESCRIPTION, IO_ACCOUNTING, MEMORY_ACCOUNTING, PIDS, SLICE, TASKS_ACCOUNTING,
+    TIMEOUT_STOP_USEC, WANTS,
+};
+
+pub type Property<'a> = (&'a str, ZbusValue<'a>);
+
+#[derive(Debug, Clone, Default)]
+pub struct PropertiesBuilder {
+    cpu_accounting: Option<bool>,
+    // MemoryAccount is for cgroup v2 as documented in dbus. However,
+    // "github.com/opencontainer/runc" uses it for all. Shall we follow the
+    // same way?
+    memory_accounting: Option<bool>,
+    task_accounting: Option<bool>,
+    // Use IO_ACCOUNTING for cgroup v2 and BLOCK_IO_ACCOUNTING for cgroup v1.
+    io_accounting: Option<bool>,
+    default_dependencies: Option<bool>,
+    description: Option<String>,
+    wants: Option<String>,
+    slice: Option<String>,
+    delegate: Option<bool>,
+    pids: Option<Vec<u32>>,
+    timeout_stop_usec: Option<u64>,
+}
+
+impl PropertiesBuilder {
+    pub fn default_cgroup(slice: &str, unit: &str) -> Self {
+        let mut builder = Self::default()
+            .cpu_accounting(true)
+            .memory_accounting(true)
+            .task_accounting(true)
+            .io_accounting(true)
+            .default_dependencies(false)
+            .description(format!("{} {}:{}", DEFAULT_DESCRIPTION, slice, unit));
+
+        if is_slice_unit(unit) {
+            // If we create a slice, the parent is defined via a Wants=.
+            builder = builder.wants(slice.to_string());
+        } else {
+            // Otherwise it's a scope, which we put into a Slice=.
+            builder = builder.slice(slice.to_string());
+            // Assume scopes always support delegation (supported since systemd v218).
+            builder = builder.delegate(true);
+        }
+
+        builder
+    }
+
+    pub fn cpu_accounting(mut self, enabled: bool) -> Self {
+        self.cpu_accounting = Some(enabled);
+        self
+    }
+
+    pub fn memory_accounting(mut self, enabled: bool) -> Self {
+        self.memory_accounting = Some(enabled);
+        self
+    }
+
+    pub fn task_accounting(mut self, enabled: bool) -> Self {
+        self.task_accounting = Some(enabled);
+        self
+    }
+
+    pub fn io_accounting(mut self, enabled: bool) -> Self {
+        self.io_accounting = Some(enabled);
+        self
+    }
+
+    pub fn default_dependencies(mut self, enabled: bool) -> Self {
+        self.default_dependencies = Some(enabled);
+        self
+    }
+
+    pub fn description(mut self, desc: String) -> Self {
+        self.description = Some(desc);
+        self
+    }
+
+    pub fn wants(mut self, wants: String) -> Self {
+        self.wants = Some(wants);
+        self
+    }
+
+    pub fn slice(mut self, slice: String) -> Self {
+        self.slice = Some(slice);
+        self
+    }
+
+    pub fn delegate(mut self, enabled: bool) -> Self {
+        self.delegate = Some(enabled);
+        self
+    }
+
+    pub fn pids(mut self, pids: Vec<u32>) -> Self {
+        self.pids = Some(pids);
+        self
+    }
+
+    pub fn timeout_stop_usec(mut self, timeout: u64) -> Self {
+        self.timeout_stop_usec = Some(timeout);
+        self
+    }
+
+    pub fn build(self) -> Vec<Property<'static>> {
+        let mut props = vec![];
+
+        if let Some(cpu_accounting) = self.cpu_accounting {
+            props.push((CPU_ACCOUNTING, ZbusValue::Bool(cpu_accounting)));
+        }
+
+        if let Some(memory_accounting) = self.memory_accounting {
+            props.push((MEMORY_ACCOUNTING, ZbusValue::Bool(memory_accounting)));
+        }
+
+        if let Some(task_accounting) = self.task_accounting {
+            props.push((TASKS_ACCOUNTING, ZbusValue::Bool(task_accounting)));
+        }
+
+        if let Some(io_accounting) = self.io_accounting {
+            if hierarchies::is_cgroup2_unified_mode() {
+                props.push((IO_ACCOUNTING, ZbusValue::Bool(io_accounting)));
+            } else {
+                props.push((BLOCK_IO_ACCOUNTING, ZbusValue::Bool(io_accounting)));
+            }
+        }
+
+        if let Some(default_dependencies) = self.default_dependencies {
+            props.push((DEFAULT_DEPENDENCIES, ZbusValue::Bool(default_dependencies)));
+        }
+
+        if let Some(description) = self.description {
+            props.push((DESCRIPTION, ZbusValue::Str(description.into())));
+        } else {
+            props.push((DESCRIPTION, ZbusValue::Str(DEFAULT_DESCRIPTION.into())));
+        }
+
+        if let Some(wants) = self.wants {
+            props.push((WANTS, ZbusValue::Str(wants.into())));
+        }
+
+        if let Some(slice) = self.slice {
+            props.push((SLICE, ZbusValue::Str(slice.into())));
+        }
+
+        if let Some(delegate) = self.delegate {
+            props.push((DELEGATE, ZbusValue::Bool(delegate)));
+        }
+
+        if let Some(pids) = self.pids {
+            props.push((PIDS, ZbusValue::Array(pids.into())));
+        }
+
+        if let Some(timeout) = self.timeout_stop_usec {
+            props.push((TIMEOUT_STOP_USEC, ZbusValue::U64(timeout)));
+        }
+
+        props
+    }
+}

src/systemd/utils.rs

@@ -0,0 +1,101 @@
+// Copyright (c) 2025 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+use crate::systemd::error::{Error, Result};
+use crate::systemd::{SCOPE_SUFFIX, SLICE_SUFFIX};
+
+/// Check if a systemd unit name is a slice unit.
+pub fn is_slice_unit(name: &str) -> bool {
+    name.ends_with(SLICE_SUFFIX)
+}
+
+/// Check if a systemd unit name is a scope unit.
+pub fn is_scope_unit(name: &str) -> bool {
+    name.ends_with(SCOPE_SUFFIX)
+}
+
+/// Expand a slice name to a full path in the filesystem.
+///
+/// # Arguments
+///
+/// * `slice` - A string slice that holds the slice name in the format
+///   "xxx-yyy-zzz.slice".
+///
+/// # Returns
+///
+/// A string that represents the full path of the slice in the filesystem.
+/// In the above case, the value would be
+/// "xxx.slice/xxx-yyy.slice/xxx-yyy-zzz.slice".
+pub fn expand_slice(slice: &str) -> Result<String> {
+    // Name has to end with ".slice", but can't be just ".slice".
+    if !slice.ends_with(SLICE_SUFFIX) || slice.len() < SLICE_SUFFIX.len() {
+        return Err(Error::InvalidArgument);
+    }
+
+    // Path-separators are not allowed.
+    if slice.contains('/') {
+        return Err(Error::InvalidArgument);
+    }
+
+    let name = slice.trim_end_matches(SLICE_SUFFIX);
+
+    // If input was -.slice, we should just return root now
+    if name == "-" {
+        return Ok("".to_string());
+    }
+
+    let mut slice_path = String::new();
+    let mut prefix = String::new();
+    for sub_slice in name.split('-') {
+        if sub_slice.is_empty() {
+            return Err(Error::InvalidArgument);
+        }
+
+        slice_path = format!("{}/{}{}{}", slice_path, prefix, sub_slice, SLICE_SUFFIX);
+        prefix = format!("{}{}-", prefix, sub_slice);
+    }
+
+    // We need a relative path, so remove the first slash.
+    slice_path.remove(0);
+
+    Ok(slice_path)
+}
+
+#[cfg(test)]
+mod tests {
+    use crate::systemd::utils::*;
+
+    #[test]
+    fn test_is_slice_unit() {
+        assert!(is_slice_unit("test.slice"));
+        assert!(!is_slice_unit("test.scope"));
+    }
+
+    #[test]
+    fn test_is_scope_unit() {
+        assert!(is_scope_unit("test.scope"));
+        assert!(!is_scope_unit("test.slice"));
+    }
+
+    #[test]
+    fn test_expand_slice() {
+        assert_eq!(expand_slice("test.slice").unwrap(), "test.slice");
+        assert_eq!(
+            expand_slice("test-1.slice").unwrap(),
+            "test.slice/test-1.slice"
+        );
+        assert_eq!(
+            expand_slice("test-1-test-2.slice").unwrap(),
+            "test.slice/test-1.slice/test-1-test.slice/test-1-test-2.slice"
+        );
+        assert_eq!(
+            expand_slice("slice-slice.slice").unwrap(),
+            "slice.slice/slice-slice.slice"
+        );
+        assert_eq!(expand_slice("-.slice").unwrap(), "");
+        assert!(expand_slice("invalid/slice").is_err());
+        assert!(expand_slice("invalid-slice").is_err());
+    }
+}

tests/builder.rs

@@ -5,24 +5,25 @@
 //
 
 //! Some simple tests covering the builder pattern for control groups.
-use cgroups_rs::blkio::*;
-use cgroups_rs::cgroup_builder::*;
-use cgroups_rs::cpu::*;
-use cgroups_rs::devices::*;
-use cgroups_rs::hugetlb::*;
-use cgroups_rs::memory::*;
-use cgroups_rs::net_cls::*;
-use cgroups_rs::pid::*;
-use cgroups_rs::*;
+use cgroups_rs::fs::blkio::*;
+use cgroups_rs::fs::cgroup_builder::*;
+use cgroups_rs::fs::cpu::*;
+use cgroups_rs::fs::devices::*;
+use cgroups_rs::fs::hugetlb::*;
+use cgroups_rs::fs::memory::*;
+use cgroups_rs::fs::net_cls::*;
+use cgroups_rs::fs::pid::*;
+use cgroups_rs::fs::*;
 
 #[test]
 pub fn test_cpu_res_build() {
-    let h = cgroups_rs::hierarchies::auto();
+    let h = cgroups_rs::fs::hierarchies::auto();
     let cg: Cgroup = CgroupBuilder::new("test_cpu_res_build")
         .cpu()
         .shares(85)
         .done()
-        .build(h);
+        .build(h)
+        .unwrap();
 
     {
         let cpu: &CpuController = cg.controller_of().unwrap();
@@ -35,14 +36,15 @@ pub fn test_cpu_res_build() {
 
 #[test]
 pub fn test_memory_res_build() {
-    let h = cgroups_rs::hierarchies::auto();
+    let h = cgroups_rs::fs::hierarchies::auto();
     let cg: Cgroup = CgroupBuilder::new("test_memory_res_build")
         .memory()
         .kernel_memory_limit(128 * 1024 * 1024)
         .swappiness(70)
         .memory_hard_limit(1024 * 1024 * 1024)
         .done()
-        .build(h);
+        .build(h)
+        .unwrap();
 
     {
         let c: &MemController = cg.controller_of().unwrap();
@@ -59,12 +61,13 @@ pub fn test_memory_res_build() {
 
 #[test]
 pub fn test_pid_res_build() {
-    let h = cgroups_rs::hierarchies::auto();
+    let h = cgroups_rs::fs::hierarchies::auto();
     let cg: Cgroup = CgroupBuilder::new("test_pid_res_build")
         .pid()
         .maximum_number_of_processes(MaxValue::Value(123))
         .done()
-        .build(h);
+        .build(h)
+        .unwrap();
 
     {
         let c: &PidController = cg.controller_of().unwrap();
@@ -78,12 +81,13 @@ pub fn test_pid_res_build() {
 #[test]
 #[ignore] // ignore this test for now, not sure why my kernel doesn't like it
 pub fn test_devices_res_build() {
-    let h = cgroups_rs::hierarchies::auto();
+    let h = cgroups_rs::fs::hierarchies::auto();
     let cg: Cgroup = CgroupBuilder::new("test_devices_res_build")
         .devices()
         .device(1, 6, DeviceType::Char, true, vec![DevicePermissions::Read])
         .done()
-        .build(h);
+        .build(h)
+        .unwrap();
 
     {
         let c: &DevicesController = cg.controller_of().unwrap();
@@ -104,7 +108,7 @@ pub fn test_devices_res_build() {
 
 #[test]
 pub fn test_network_res_build() {
-    let h = cgroups_rs::hierarchies::auto();
+    let h = cgroups_rs::fs::hierarchies::auto();
     if h.v2() {
         // FIXME add cases for v2
         return;
@@ -113,7 +117,8 @@ pub fn test_network_res_build() {
         .network()
         .class_id(1337)
         .done()
-        .build(h);
+        .build(h)
+        .unwrap();
 
     {
         let c: &NetClsController = cg.controller_of().unwrap();
@@ -125,7 +130,7 @@ pub fn test_network_res_build() {
 
 #[test]
 pub fn test_hugepages_res_build() {
-    let h = cgroups_rs::hierarchies::auto();
+    let h = cgroups_rs::fs::hierarchies::auto();
     if h.v2() {
         // FIXME add cases for v2
         return;
@@ -134,7 +139,8 @@ pub fn test_hugepages_res_build() {
         .hugepages()
         .limit("2MB".to_string(), 4 * 2 * 1024 * 1024)
         .done()
-        .build(h);
+        .build(h)
+        .unwrap();
 
     {
         let c: &HugeTlbController = cg.controller_of().unwrap();
@@ -147,12 +153,13 @@ pub fn test_hugepages_res_build() {
 #[test]
 #[ignore] // high version kernel not support `blkio.weight`
 pub fn test_blkio_res_build() {
-    let h = cgroups_rs::hierarchies::auto();
+    let h = cgroups_rs::fs::hierarchies::auto();
     let cg: Cgroup = CgroupBuilder::new("test_blkio_res_build")
         .blkio()
         .weight(100)
         .done()
-        .build(h);
+        .build(h)
+        .unwrap();
 
     {
         let c: &BlkIoController = cg.controller_of().unwrap();

tests/cgroup.rs

@@ -5,29 +5,62 @@
 //
 
 //! Simple unit tests about the control groups system.
-use cgroups_rs::memory::MemController;
-use cgroups_rs::Controller;
-use cgroups_rs::{Cgroup, CgroupPid, Subsystem};
+use std::process::Command;
+use std::thread::sleep;
+use std::time::Duration;
+
+use cgroups_rs::fs::cgroup::{
+    CGROUP_MODE_DOMAIN, CGROUP_MODE_DOMAIN_INVALID, CGROUP_MODE_DOMAIN_THREADED,
+    CGROUP_MODE_THREADED,
+};
+use cgroups_rs::fs::memory::MemController;
+use cgroups_rs::fs::Controller;
+use cgroups_rs::fs::{Cgroup, Subsystem};
+use cgroups_rs::CgroupPid;
 
 #[test]
-fn test_tasks_iterator() {
-    let h = cgroups_rs::hierarchies::auto();
+fn test_procs_iterator_cgroup() {
+    let h = cgroups_rs::fs::hierarchies::auto();
     let pid = libc::pid_t::from(nix::unistd::getpid()) as u64;
-    let cg = Cgroup::new(h, String::from("test_tasks_iterator"));
+    let cg = Cgroup::new(h, String::from("test_procs_iterator_cgroup")).unwrap();
+    {
+        // Add a task to the control group.
+        cg.add_task_by_tgid(CgroupPid::from(pid)).unwrap();
+
+        let mut procs = cg.procs().into_iter();
+        // Verify that the task is indeed in the xcontrol group
+        assert_eq!(procs.next(), Some(CgroupPid::from(pid)));
+        assert_eq!(procs.next(), None);
+
+        // Now, try removing it.
+        cg.remove_task_by_tgid(CgroupPid::from(pid)).unwrap();
+        procs = cg.procs().into_iter();
+
+        // Verify that it was indeed removed.
+        assert_eq!(procs.next(), None);
+    }
+    cg.delete().unwrap();
+}
+
+#[test]
+fn test_tasks_iterator_cgroup_v1() {
+    if cgroups_rs::fs::hierarchies::is_cgroup2_unified_mode() {
+        return;
+    }
+    let h = cgroups_rs::fs::hierarchies::auto();
+    let pid = libc::pid_t::from(nix::unistd::getpid()) as u64;
+    let cg = Cgroup::new(h, String::from("test_tasks_iterator_cgroup_v1")).unwrap();
     {
         // Add a task to the control group.
         cg.add_task(CgroupPid::from(pid)).unwrap();
 
-        use std::{thread, time};
-        thread::sleep(time::Duration::from_millis(100));
-
         let mut tasks = cg.tasks().into_iter();
-        // Verify that the task is indeed in the control group
+        // Verify that the task is indeed in the xcontrol group
         assert_eq!(tasks.next(), Some(CgroupPid::from(pid)));
         assert_eq!(tasks.next(), None);
 
         // Now, try removing it.
-        cg.remove_task(CgroupPid::from(pid));
+        cg.remove_task(CgroupPid::from(pid)).unwrap();
         tasks = cg.tasks().into_iter();
 
         // Verify that it was indeed removed.
@@ -37,11 +70,148 @@ fn test_tasks_iterator() {
 }
 
 #[test]
-fn test_cgroup_with_relative_paths() {
-    if cgroups_rs::hierarchies::is_cgroup2_unified_mode() {
+fn test_tasks_iterator_cgroup_threaded_mode() {
+    if !cgroups_rs::fs::hierarchies::is_cgroup2_unified_mode() {
         return;
     }
-    let h = cgroups_rs::hierarchies::auto();
+    let pid = libc::pid_t::from(nix::unistd::getpid()) as u64;
+    let cg = Cgroup::new(
+        cgroups_rs::fs::hierarchies::auto(),
+        String::from("test_tasks_iterator_cgroup_threaded_mode"),
+    )
+    .unwrap();
+    let cg_threaded_sub1 = Cgroup::new_with_specified_controllers(
+        cgroups_rs::fs::hierarchies::auto(),
+        String::from("test_tasks_iterator_cgroup_threaded_mode/threaded_sub1"),
+        Some(vec![String::from("cpuset"), String::from("cpu")]),
+    )
+    .unwrap();
+    let cg_threaded_sub2 = Cgroup::new_with_specified_controllers(
+        cgroups_rs::fs::hierarchies::auto(),
+        String::from("test_tasks_iterator_cgroup_threaded_mode/threaded_sub2"),
+        Some(vec![String::from("cpuset"), String::from("cpu")]),
+    )
+    .unwrap();
+    {
+        // Verify that cgroup type of the control group is domain mode.
+        assert_eq!(cg.get_cgroup_type().unwrap(), CGROUP_MODE_DOMAIN);
+
+        // Set cgroup type of the sub-control group is thread mode.
+        cg_threaded_sub1
+            .set_cgroup_type(CGROUP_MODE_THREADED)
+            .unwrap();
+        // Verify that cgroup type of the sub-control group is thread mode.
+        assert_eq!(
+            cg_threaded_sub1.get_cgroup_type().unwrap(),
+            CGROUP_MODE_THREADED
+        );
+        // Verify that the cgroup type of the sub-control group that does
+        // not set the cgroup type is domain invalid mode.
+        assert_eq!(
+            cg_threaded_sub2.get_cgroup_type().unwrap(),
+            CGROUP_MODE_DOMAIN_INVALID
+        );
+        // Verify whether the cgroup type of the parent control group of
+        // the control group whose cgroup type is set to thread mode is
+        // domain thread mode.
+        assert_eq!(cg.get_cgroup_type().unwrap(), CGROUP_MODE_DOMAIN_THREADED);
+
+        // Set cgroup type of the sub-control group is thread mode.
+        cg_threaded_sub2
+            .set_cgroup_type(CGROUP_MODE_THREADED)
+            .unwrap();
+        // Verify that cgroup type of the sub-control group is thread mode.
+        assert_eq!(
+            cg_threaded_sub2.get_cgroup_type().unwrap(),
+            CGROUP_MODE_THREADED
+        );
+
+        // Add a proc to the control group.
+        cg.add_task_by_tgid(CgroupPid::from(pid)).unwrap();
+
+        let mut procs = cg.procs().into_iter();
+        // Verify that the task is indeed in the x control group
+        assert_eq!(procs.next(), Some(CgroupPid::from(pid)));
+        assert_eq!(procs.next(), None);
+
+        // Add a task to the sub control group.
+        cg_threaded_sub1.add_task(CgroupPid::from(pid)).unwrap();
+
+        let mut tasks = cg_threaded_sub1.tasks().into_iter();
+        // Verify that the task is indeed in the xcontrol group
+        assert_eq!(tasks.next(), Some(CgroupPid::from(pid)));
+        assert_eq!(tasks.next(), None);
+
+        // Now, try move it to parent.
+        cg_threaded_sub1
+            .move_task_to_parent(CgroupPid::from(pid))
+            .unwrap();
+        tasks = cg_threaded_sub1.tasks().into_iter();
+
+        // Verify that it was indeed removed.
+        assert_eq!(tasks.next(), None);
+
+        // Now, try removing it.
+        cg.remove_task_by_tgid(CgroupPid::from(pid)).unwrap();
+        procs = cg.procs().into_iter();
+
+        // Verify that it was indeed removed.
+        assert_eq!(procs.next(), None);
+    }
+    cg_threaded_sub1.delete().unwrap();
+    cg_threaded_sub2.delete().unwrap();
+    cg.delete().unwrap();
+}
+
+#[test]
+fn test_kill_cgroup() {
+    if !cgroups_rs::fs::hierarchies::is_cgroup2_unified_mode() {
+        return;
+    }
+    let h = cgroups_rs::fs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("test_kill_cgroup")).unwrap();
+    {
+        // Spawn a proc, don't want to getpid(2) here.
+        let mut child = Command::new("sleep").arg("infinity").spawn().unwrap();
+        cg.add_task_by_tgid(CgroupPid::from(child.id() as u64))
+            .unwrap();
+
+        let cg_procs = cg.procs();
+        assert_eq!(cg_procs.len(), 1_usize);
+
+        // Now kill and wait on the proc.
+        cg.kill().unwrap();
+
+        let mut tries = 0;
+        let status: Option<std::process::ExitStatus> = loop {
+            match child.try_wait() {
+                Ok(Some(status)) => {
+                    break Some(status);
+                }
+                Ok(None) => {
+                    if tries > 3 {
+                        break None;
+                    }
+                    sleep(Duration::from_millis(100));
+                    tries += 1;
+                }
+                Err(e) => {
+                    child.kill().unwrap();
+                    panic!("error attempting to wait: {}", e);
+                }
+            }
+        };
+        assert!(status.is_some());
+    }
+    cg.delete().unwrap();
+}
+
+#[test]
+fn test_cgroup_with_relative_paths() {
+    if cgroups_rs::fs::hierarchies::is_cgroup2_unified_mode() {
+        return;
+    }
+    let h = cgroups_rs::fs::hierarchies::auto();
     let cgroup_root = h.root();
     let cgroup_name = "test_cgroup_with_relative_paths";
 
@@ -79,11 +249,11 @@ fn test_cgroup_with_relative_paths() {
 
 #[test]
 fn test_cgroup_v2() {
-    if !cgroups_rs::hierarchies::is_cgroup2_unified_mode() {
+    if !cgroups_rs::fs::hierarchies::is_cgroup2_unified_mode() {
         return;
     }
-    let h = cgroups_rs::hierarchies::auto();
-    let cg = Cgroup::new(h, String::from("test_v2"));
+    let h = cgroups_rs::fs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("test_v2")).unwrap();
 
     let mem_controller: &MemController = cg.controller_of().unwrap();
     let (mem, swp, rev) = (4 * 1024 * 1000, 2 * 1024 * 1000, 1024 * 1000);

tests/cpu.rs

@@ -4,13 +4,13 @@
 //
 
 //! Simple unit tests about the CPU control groups system.
-use cgroups_rs::cpu::CpuController;
-use cgroups_rs::Cgroup;
+use cgroups_rs::fs::cpu::CpuController;
+use cgroups_rs::fs::Cgroup;
 
 #[test]
 fn test_cfs_quota_and_periods() {
-    let h = cgroups_rs::hierarchies::auto();
-    let cg = Cgroup::new(h, String::from("test_cfs_quota_and_periods"));
+    let h = cgroups_rs::fs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("test_cfs_quota_and_periods")).unwrap();
 
     let cpu_controller: &CpuController = cg.controller_of().unwrap();
 

tests/cpuset.rs

@@ -3,17 +3,17 @@
 //
 // SPDX-License-Identifier: Apache-2.0 or MIT
 //
-
-use cgroups_rs::cpuset::CpuSetController;
-use cgroups_rs::error::ErrorKind;
-use cgroups_rs::{Cgroup, CgroupPid};
-
 use std::fs;
 
+use cgroups_rs::fs::cpuset::CpuSetController;
+use cgroups_rs::fs::error::ErrorKind;
+use cgroups_rs::fs::Cgroup;
+use cgroups_rs::CgroupPid;
+
 #[test]
 fn test_cpuset_memory_pressure_root_cg() {
-    let h = cgroups_rs::hierarchies::auto();
-    let cg = Cgroup::new(h, String::from("test_cpuset_memory_pressure_root_cg"));
+    let h = cgroups_rs::fs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("test_cpuset_memory_pressure_root_cg")).unwrap();
     {
         let cpuset: &CpuSetController = cg.controller_of().unwrap();
 
@@ -26,8 +26,8 @@ fn test_cpuset_memory_pressure_root_cg() {
 
 #[test]
 fn test_cpuset_set_cpus() {
-    let h = cgroups_rs::hierarchies::auto();
-    let cg = Cgroup::new(h, String::from("test_cpuset_set_cpus"));
+    let h = cgroups_rs::fs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("test_cpuset_set_cpus")).unwrap();
     {
         let cpuset: &CpuSetController = cg.controller_of().unwrap();
 
@@ -63,8 +63,8 @@ fn test_cpuset_set_cpus() {
 
 #[test]
 fn test_cpuset_set_cpus_add_task() {
-    let h = cgroups_rs::hierarchies::auto();
-    let cg = Cgroup::new(h, String::from("test_cpuset_set_cpus_add_task/sub-dir"));
+    let h = cgroups_rs::fs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("test_cpuset_set_cpus_add_task/sub-dir")).unwrap();
 
     let cpuset: &CpuSetController = cg.controller_of().unwrap();
     let set = cpuset.cpuset();
@@ -77,13 +77,13 @@ fn test_cpuset_set_cpus_add_task() {
 
     // Add a task to the control group.
     let pid_i = libc::pid_t::from(nix::unistd::getpid()) as u64;
-    let _ = cg.add_task(CgroupPid::from(pid_i));
+    let _ = cg.add_task_by_tgid(CgroupPid::from(pid_i));
     let tasks = cg.tasks();
     assert!(!tasks.is_empty());
     println!("tasks after added: {:?}", tasks);
 
     // remove task
-    let _ = cg.remove_task(CgroupPid::from(pid_i));
+    cg.remove_task_by_tgid(CgroupPid::from(pid_i)).unwrap();
     let tasks = cg.tasks();
     println!("tasks after deleted: {:?}", tasks);
     assert_eq!(0, tasks.len());

tests/devices.rs

@@ -6,18 +6,18 @@
 
 //! Integration tests about the devices subsystem
 
-use cgroups_rs::devices::{DevicePermissions, DeviceType, DevicesController};
-use cgroups_rs::{Cgroup, DeviceResource};
+use cgroups_rs::fs::devices::{DevicePermissions, DeviceType, DevicesController};
+use cgroups_rs::fs::{Cgroup, DeviceResource};
 
 #[test]
 fn test_devices_parsing() {
     // now only v2
-    if cgroups_rs::hierarchies::is_cgroup2_unified_mode() {
+    if cgroups_rs::fs::hierarchies::is_cgroup2_unified_mode() {
         return;
     }
 
-    let h = cgroups_rs::hierarchies::auto();
-    let cg = Cgroup::new(h, String::from("test_devices_parsing"));
+    let h = cgroups_rs::fs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("test_devices_parsing")).unwrap();
     {
         let devices: &DevicesController = cg.controller_of().unwrap();
 

tests/hugetlb.rs

@@ -4,20 +4,20 @@
 //
 
 //! Integration tests about the hugetlb subsystem
-use cgroups_rs::error::*;
-use cgroups_rs::hugetlb::{self, HugeTlbController};
-use cgroups_rs::Cgroup;
+use cgroups_rs::fs::error::*;
+use cgroups_rs::fs::hugetlb::{self, HugeTlbController};
+use cgroups_rs::fs::Cgroup;
 use std::fs;
 
 #[test]
 fn test_hugetlb_sizes() {
     // now only v2
-    if cgroups_rs::hierarchies::is_cgroup2_unified_mode() {
+    if cgroups_rs::fs::hierarchies::is_cgroup2_unified_mode() {
         return;
     }
 
-    let h = cgroups_rs::hierarchies::auto();
-    let cg = Cgroup::new(h, String::from("test_hugetlb_sizes"));
+    let h = cgroups_rs::fs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("test_hugetlb_sizes")).unwrap();
     {
         let hugetlb_controller: &HugeTlbController = cg.controller_of().unwrap();
         let _ = hugetlb_controller.get_sizes();

tests/memory.rs

@@ -4,14 +4,13 @@
 //
 
 //! Integration tests about the hugetlb subsystem
-use cgroups_rs::memory::{MemController, SetMemory};
-use cgroups_rs::Controller;
-use cgroups_rs::{Cgroup, MaxValue};
+use cgroups_rs::fs::memory::{MemController, SetMemory};
+use cgroups_rs::fs::{Cgroup, Controller, MaxValue};
 
 #[test]
 fn test_disable_oom_killer() {
-    let h = cgroups_rs::hierarchies::auto();
-    let cg = Cgroup::new(h, String::from("test_disable_oom_killer"));
+    let h = cgroups_rs::fs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("test_disable_oom_killer")).unwrap();
     {
         let mem_controller: &MemController = cg.controller_of().unwrap();
 
@@ -35,12 +34,12 @@ fn test_disable_oom_killer() {
 
 #[test]
 fn set_kmem_limit_v1() {
-    let h = cgroups_rs::hierarchies::auto();
+    let h = cgroups_rs::fs::hierarchies::auto();
     if h.v2() {
         return;
     }
 
-    let cg = Cgroup::new(h, String::from("set_kmem_limit_v1"));
+    let cg = Cgroup::new(h, String::from("set_kmem_limit_v1")).unwrap();
     {
         let mem_controller: &MemController = cg.controller_of().unwrap();
         mem_controller.set_kmem_limit(1).unwrap();
@@ -50,12 +49,12 @@ fn set_kmem_limit_v1() {
 
 #[test]
 fn set_mem_v2() {
-    let h = cgroups_rs::hierarchies::auto();
+    let h = cgroups_rs::fs::hierarchies::auto();
     if !h.v2() {
         return;
     }
 
-    let cg = Cgroup::new(h, String::from("set_mem_v2"));
+    let cg = Cgroup::new(h, String::from("set_mem_v2")).unwrap();
     {
         let mem_controller: &MemController = cg.controller_of().unwrap();
 

tests/pids.rs

@@ -5,19 +5,16 @@
 //
 
 //! Integration tests about the pids subsystem
-use cgroups_rs::pid::PidController;
-use cgroups_rs::Controller;
-use cgroups_rs::{Cgroup, MaxValue};
-
+use cgroups_rs::fs::pid::PidController;
+use cgroups_rs::fs::{Cgroup, Controller, MaxValue};
+use libc::pid_t;
 use nix::sys::wait::{waitpid, WaitStatus};
 use nix::unistd::{fork, ForkResult};
 
-use libc::pid_t;
-
 #[test]
 fn create_and_delete_cgroup() {
-    let h = cgroups_rs::hierarchies::auto();
-    let cg = Cgroup::new(h, String::from("create_and_delete_cgroup"));
+    let h = cgroups_rs::fs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("create_and_delete_cgroup")).unwrap();
     {
         let pidcontroller: &PidController = cg.controller_of().unwrap();
         pidcontroller.set_pid_max(MaxValue::Value(1337)).unwrap();
@@ -30,8 +27,8 @@ fn create_and_delete_cgroup() {
 
 #[test]
 fn test_pids_current_is_zero() {
-    let h = cgroups_rs::hierarchies::auto();
-    let cg = Cgroup::new(h, String::from("test_pids_current_is_zero"));
+    let h = cgroups_rs::fs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("test_pids_current_is_zero")).unwrap();
     {
         let pidcontroller: &PidController = cg.controller_of().unwrap();
         let current = pidcontroller.get_pid_current();
@@ -42,8 +39,8 @@ fn test_pids_current_is_zero() {
 
 #[test]
 fn test_pids_events_is_zero() {
-    let h = cgroups_rs::hierarchies::auto();
-    let cg = Cgroup::new(h, String::from("test_pids_events_is_zero"));
+    let h = cgroups_rs::fs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("test_pids_events_is_zero")).unwrap();
     {
         let pidcontroller: &PidController = cg.controller_of().unwrap();
         let events = pidcontroller.get_pid_events();
@@ -55,8 +52,8 @@ fn test_pids_events_is_zero() {
 
 #[test]
 fn test_pid_events_is_not_zero() {
-    let h = cgroups_rs::hierarchies::auto();
-    let cg = Cgroup::new(h, String::from("test_pid_events_is_not_zero"));
+    let h = cgroups_rs::fs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("test_pid_events_is_not_zero")).unwrap();
     {
         let pids: &PidController = cg.controller_of().unwrap();
         let before = pids.get_pid_events();
@@ -65,7 +62,7 @@ fn test_pid_events_is_not_zero() {
         match unsafe { fork() } {
             Ok(ForkResult::Parent { child, .. }) => {
                 // move the process into the control group
-                let _ = pids.add_task(&(pid_t::from(child) as u64).into());
+                let _ = pids.add_task_by_tgid(&(pid_t::from(child) as u64).into());
 
                 println!("added task to cg: {:?}", child);
 

tests/resources.rs

@@ -5,13 +5,13 @@
 //
 
 //! Integration test about setting resources using `apply()`
-use cgroups_rs::pid::PidController;
-use cgroups_rs::{Cgroup, MaxValue, PidResources, Resources};
+use cgroups_rs::fs::pid::PidController;
+use cgroups_rs::fs::{Cgroup, MaxValue, PidResources, Resources};
 
 #[test]
 fn pid_resources() {
-    let h = cgroups_rs::hierarchies::auto();
-    let cg = Cgroup::new(h, String::from("pid_resources"));
+    let h = cgroups_rs::fs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("pid_resources")).unwrap();
     {
         let res = Resources {
             pid: PidResources {